TMG-F 92

UTOMATIC
YNAMIC
NCREMENTAL
ONLINEAR
NALYSIS
Theory and Modeling Guide

Volume III: CFD & FSI
ADINA 9.2 July 2016
ADINA R & D, Inc.

ADINA
Theory and Modeling Guide
Volume III:
CFD & FSI
July 2016
ADINA R&D, Inc.

71 Elton Avenue
Watertown, MA 02472 USA
tel. (617) 926-5199

telefax (617) 926-0238
www.adina.com
Notices
ADINA R & D, Inc. owns both this software program system and its documentation. Both the
program system and the documentation are copyrighted with all rights reserved by ADINA R & D,
Inc.
The information contained in this document is subject to change without notice.
ADINA R & D, Inc. makes no warranty whatsoever, expressed or implied that the Program and its
documentation including any modifications and updates are free from errors and defects. In no
event shall ADINA R & D, Inc. become liable to the User or any party for any loss, including but
not limited to, loss of time, money or goodwill, which may arise from the use of the Program and its
documentation including any modifications and updates.
Trademarks
ADINA is a registered trademark of K.J. Bathe /ADINA R & D, Inc.
All other product names are trademarks or registered trademarks of their respective owners.
Copyright Notice
ADINA R & D, Inc. 1987-2016

July 2016 Printing
Printed in the USA
Table of contents
Table of contents
Nomenclature............................................................................................................11
Chapter 1 Introduction .........................................................................................23
Chapter 2 Various forms of fluid equations ........................................................27

2.1 General conservative Navier-Stokes equations ...............................................27
2.2 Nonconservative Navier-Stokes equations......................................................29
2.3 Rotating frame of reference.............................................................................30
2.4 Planar flow ......................................................................................................33
2.5 Axisymmetric flow..........................................................................................34
2.6 Incompressible flow ........................................................................................36
2.7 Fluid flow through porous media ....................................................................41
2.8 Slightly compressible flow..............................................................................45
2.9 Compressible flows .........................................................................................47
2.10 High-speed compressible flow ......................................................................48
2.11 Turbulence models ........................................................................................50
2.12 Mass transfer .................................................................................................62
2.13 Steady-state equations ...................................................................................68
2.14 Arbitrary-Lagrangian-Eulerian formulations ................................................69
2.15 System units and nondimensional forms.......................................................72
Chapter 3 Formulation of incompressible, slightly compressible and low-speed

compressible flows ...............................................................................77
3.1 Governing equations .......................................................................................80
3.2 Numerical method ...........................................................................................80
3.2.1 Time integration ......................................................................................80
3.2.2 Discretized equations ..............................................................................82
3.2.3 Upwinding techniques.............................................................................83
3.2.4 Solution method for nonlinear finite element equations .........................85
3.3 Elements ..........................................................................................................87
3.4 Boundary conditions .......................................................................................88
ADINA R & D, Inc. 3

Table of contents
3.4.1 General descriptions................................................................................88

3.4.2 Usual boundary conditions for fluid .......................................................91
3.4.3 Special boundary conditions for fluid .....................................................96
3.4.4 Usual boundary conditions for temperature ..........................................120
3.4.5 Special boundary conditions for temperature........................................122
3.4.6 Thermal resistance condition ................................................................128
3.4.7 User-supplied boundary condition ........................................................128
3.4.8 Shell-thermal condition.........................................................................136
3.5 Initial conditions............................................................................................137
3.6 Material models.............................................................................................138
3.6.1 Constant material model .......................................................................138
3.6.2 Time-dependent model..........................................................................139
3.6.3 Power-law model ..................................................................................139
3.6.4 Carreau model .......................................................................................140
3.6.5 Temperature-dependent model..............................................................140
3.6.6 Temperature-dependent power-law model............................................140
3.6.7 Second order model ..............................................................................141
3.6.8 Large-eddy-simulation model ...............................................................141
3.6.9 User-supplied materials.........................................................................143
3.6.10 ASME steam table ..............................................................................149
3.6.11 P-T-dependent material.......................................................................150
3.6.12 Field friction........................................................................................151
3.6.13 User-modified materials......................................................................151
3.6.14 Material curves....................................................................................152
3.6.15 Gas models..........................................................................................154
Chapter 4 Formulation of fluid flows through porous media..........................158

4.1 Governing equations .....................................................................................159
4.2 Numerical method .........................................................................................160
4.3 Elements ........................................................................................................161
4.4 Boundary conditions .....................................................................................161
4.6 Material model ..............................................................................................163
Chapter 5 Formulation of turbulence in incompressible, slightly compressible

and low-speed compressible flows....................................................165
4 ADINA CFD & FSI Theory and Modeling Guide

Table of contents
5.3 Elements ........................................................................................................169

5.4.1 Prescribed turbulence K - ..................................................................170
5.4.2 Prescribed turbulence K - ..................................................................171
5.4.3 Zero flux of turbulence variables ..........................................................172
5.4.4 Solid walls.............................................................................................172
5.6 Material models.............................................................................................175
5.6.1 K - turbulence model...........................................................................175
5.6.2 RNG K - turbulence model ................................................................177
5.6.3 High Reynolds K - turbulence model ................................................177
5.6.4 Low Reynolds K - turbulence model.................................................177
5.6.5 Shear Stress Transport model................................................................178
5.6.6 Spalart-Allmaras and Detached Eddy Simulation material models ......178
5.6.7 Two-layer zonal material model ...........................................................179
5.6.8 Material curves......................................................................................179
5.6.9 Gas models............................................................................................179
Chapter 6 Formulation of high-speed compressible flows ...............................180

6.2.1 Geometric conservation law in moving meshes....................................183
6.2.2 Flux-Splitting method ...........................................................................185
6.2.3 Finite element method...........................................................................187
6.2.4 Solution procedure ................................................................................188
6.2.5 Explicit time integration scheme...........................................................188
6.3 Control volumes and finite elements.............................................................190
6.3.1 Control volume in triangular elements (3-node) ...................................190
6.3.2 Control volume in tetrahedral elements (4-node) .................................192
6.4.1 General descriptions..............................................................................193
6.4.2 Usual boundary conditions for fluid .....................................................201
6.4.3 Special boundary conditions .................................................................205
6.6 Material models.............................................................................................219
ADINA R & D, Inc. 5

Table of contents
6.6.1 Constant material model .......................................................................219

6.6.2 Power-law model ..................................................................................220
6.6.3 Temperature-dependent model..............................................................220
6.6.4 Pressure-dependent model ....................................................................221
6.6.5 Pressure-temperature-dependent model ................................................221
Chapter 7 Formulation of turbulence in high-speed compressible flows .......224

7.3 Elements ........................................................................................................227
7.4.1 Prescribed turbulence K - ...................................................................227
7.4.2 Zero flux of turbulence variables ..........................................................228
7.4.3 Solid walls.............................................................................................228
7.6 Material models.............................................................................................230
Chapter 8 Formulation of mass transfer in fluid flows ....................................232

8.3 Elements ........................................................................................................235
8.4.1 Prescribed mass-ratio ............................................................................235
8.4.2 Distributed mass flux ............................................................................236
8.4.3 Mass Convection...................................................................................236
8.6 Material models.............................................................................................238
8.6.1 Constant material data model................................................................238
8.6.2 Velocity-dependent model ....................................................................238
8.6.3 Pressure-temperature-dependent model ................................................239
Chapter 9 Formulation of fluid-structure interactions ....................................246

9.1 Introduction ...................................................................................................246

Table of contents
9.2 Theory ...........................................................................................................247

9.2.1 Kinematic and dynamic conditions .......................................................248
9.2.2 Separate meshes of fluid and solid models ...........................................249
9.2.3 Consistent time integration for fluid and solid models .........................252
9.2.4 Finite element equations of the coupled system....................................253
9.3 Iterative computing of two-way coupling .....................................................255
9.4 Direct computing of two-way coupling.........................................................256
9.5 Direct computing of one-way coupling.........................................................258
9.6 Indirect computing of one-way coupling ......................................................259
9.7 Elements on fluid and solid interfaces ..........................................................260
9.8 Material models.............................................................................................267
9.9 Procedure of model preparation and testing..................................................268
9.10 Other types of fluid-structure interactions...................................................270
9.10.1 Fluid-solid interaction in porous media ..............................................272
9.10.2 Coupled thermal fluid-structural interaction .......................................273
9.10.3 Model one-way thermally coupled problem using mapping...............274
Chapter 10 Elements .............................................................................................276

10.1 Boundary line elements ...............................................................................276
10.2 2D triangular element (3-node) ...................................................................277
10.3 2D quadrilateral element (9-node) ..............................................................279
10.4 2D triangular element (6-node) ...................................................................280
10.5 3D tetrahedral element (4-node)..................................................................282
10.6 3D brick element (27-node) ........................................................................283
10.7 2D FCBI elements (4- and 3-node) .............................................................285
10.8 3D FCBI elements (8-, 4-, 6- and 5-node)...................................................287
10.9 FCBI-C elements.........................................................................................289
Chapter 11 Solutions of nonlinear equations ......................................................292

11.1 Newton-Raphson method ............................................................................292
11.2 Segregated method ......................................................................................293
11.2.1 Basic parameters .................................................................................295
11.2.2 Outer iteration control .........................................................................297
11.2.3 Inner iteration control..........................................................................303
11.2.4 Comments and tips on convergence....................................................305
11.2.5 Limitations ..........................................................................................309
ADINA R & D, Inc. 7

Table of contents
11.3 Solvers of algebraic equations.....................................................................310

11.3.1 Gauss elimination method...................................................................310
11.3.2 Sparse solver .......................................................................................311
11.3.3 Iterative methods.................................................................................314
Chapter 12 Other capabilities...............................................................................318

12.1 Automatic time-stepping CFL option..........................................................318
12.2 Automatic time-stepping ATS option .........................................................319
12.3 Restart analysis............................................................................................320
12.4 Mapping solution.........................................................................................321
12.5 Skew system................................................................................................322
12.6 Constraint conditions...................................................................................323
12.7 Conjugate heat transfer problems................................................................324
12.8 Element birth-death option..........................................................................325
12.9 Pressure datum ............................................................................................326
12.10 ALE formulation and leader-follower option............................................328
12.10.1 ALE formulation ...............................................................................328
12.10.2 Leader-follower option......................................................................330
12.11 Electro-static and steady current conduction analyses ..............................333
12.11.1 Discretization and method of solution ..............................................334
12.11.2 Boundary conditions .........................................................................334
12.12 Phase change between liquid and vapor....................................................336
12.12.1 General phase change model.............................................................336
12.12.2 Cavitation model ...............................................................................338
12.12.3 Phase-change model setup ................................................................341
12.13 VOF method..............................................................................................342
12.13.1 Transport equations...........................................................................342
12.13.2 Material .............................................................................................342
12.13.3 Surface tension ..................................................................................343
12.13.4 VOF-wall angle condition.................................................................343
12.13.5 Liquid-solid phase change for primary species.................................344
12.13.6 Loading of the species.......................................................................345
12.14 Rigid motion of element group .................................................................345
12.15 Steered adaptive meshing..........................................................................346
12.15.1 Introduction.......................................................................................346
12.15.2 General procedure .............................................................................349

Table of contents
12.15.3 Criteria for selection of preferable element sizes..............................353

12.15.4 Combination of criteria .....................................................................355
12.15.5 Smoothing technique.........................................................................355
12.15.6 Control on the maximum number of elements..................................356
12.15.7 Comments and limitations.................................................................356
12.15.8 Adaptive Meshing Modeling Tips ....................................................358
Chapter 13 Other topics ........................................................................................360

13.1 Model preparation and testing.....................................................................360
13.1.1 Choosing the fluid model ....................................................................360
13.1.2 Choosing the analysis type..................................................................364
13.1.3 Choosing the time/load step size.........................................................367
13.1.4 Choosing the computational domain...................................................369
13.1.5 Choosing proper boundary conditions ................................................372
13.1.6 Choosing the proper element ..............................................................376
13.1.7 Choosing the material model ..............................................................377
13.1.8 Choosing the solver.............................................................................377
13.1.9 Model testing.......................................................................................378
13.2 Moving mesh control ..................................................................................381
13.3 Strategies toward obtaining converged solutions ........................................384
13.3.1 Incremental solution procedure...........................................................387
13.3.2 Improving initial conditions using restart runs ...................................390
13.3.3 Improving the initial conditions using the mapping file .....................392
13.3.4 Improving the matrix conditioning using the CFL option ..................392
13.3.5 Improving the matrix conditioning using proper units........................393
13.4 Consistent units, SI units and USCS units ..................................................396
13.5 Use of memory and disk..............................................................................398
13.5.1 Memory usage.....................................................................................398
13.5.2 Disk usage ...........................................................................................403
13.6 Brief summary of recommendations for solutions .....................................404
For low Re incompressible flows...................................................................405
For high Re incompressible flows..................................................................406
Slightly compressible flows ...........................................................................406
High-speed and low-speed compressible flows .............................................406
Ref..................................................................................................................406
Appendix-1 List of figures .....................................................................................408
ADINA R & D, Inc. 9

Table of contents
Appendix-2 List of tables .......................................................................................413
Appendix-3 List of error/warning messages ........................................................415
Index 514

Nomenclature
Nomenclature
Typical units used here are:
Time = s , second
Length = m , meter
Mass = kg , kilogram
Temperature = K , Kelvin
Velocity = m s , meters per second
Density = kg m3 , kilogram per cubic meter
Pressure = Pa = N m 2 , Pascal
Force = N = kg-m s 2 , Newton
Energy = J = N-m, Joule
Power = W = N-m s , Watt
Electric potential = V, Volt
[.] = dimensionless
Scales and reference datums used here in nondimensional quantities are:
L* = length scale
V = velocity scale
D = density scale
C = specific heat scale
T = temperature scale
* = mass-ratio scale
xr = ( xr , yr , zr ) , reference vector of coordinates
r = reference temperature
*
= electric potential scale ( 1 here )
ADINA R & D, Inc. 11

Nomenclature
Nondimensional
Notation Explanation Typical unit
form
1
vi v , specific kinetic
b 2 m 2 s 2 b V
2
energy
2
D2 g L3 T C k 2 ,
Bo [.] Bo
Boussinesq number
2
Br
V kT , Brinkman [.] Br
number
c p , sound speed [ m s] c V*
empirical constant in
c1 [.] c1
K - turbulence model
parameters for
temperature units such
c1 , c2 that c1 + c2 is in an
absolute temperature
scale
c2 [.] c2
c3 [.] c3
c empirical constant in c
[.]
V , Capillary
Ca [.] Ca
number
Cp specific heat at constant m 2 s 2 -K Cp C
pressure
specific heat at constant m 2 s 2 -K
Cv Cv C
volume
C specific heat scale m 2 s 2 -K 1
diffuse reflectivity of
d [.] d
specular radiation
d w+ [.] d w+
turbulence models

Nomenclature
e e eij eij ,
D deformation rate (or shear [1 s] DL V
rate)
diffusion coefficient
D tensor in mass transfer [ kg m-s ] D ( DV L )
equations
D density scale kg m3 1
e specific internal energy m 2 s 2 e V
2
emittance of specular
e [.] e
radiation
2 ( v + v ) , velocity
1 T
e [1 s] eL* V
strain tensor
axis vector of moving
ei [.] ei
Cartesian frame
2 v i v + e , specific total
1
E m 2 s 2 E V
2
energy
axis vector of fixed
Ei [.] Ei
Cartesian frame
2 v r i v r + e , variable
1
Er used as specific total m 2 s 2 Er V

2
energy in axisymmetric
planes
2
V C T , Eckert
Ec [.] Ec
number
view factor in radiation
f [.] f
heat transfer
f vapor fraction [.] f
centrifugal force or
fc Coriolis force per unit m3 fc L ( DV )
2
volume
body force per unit m3 2
fB f B L ( DV )
volume
buoyancy force per unit m3 2
fbB fbB L ( DV )
volume

Nomenclature
f gB gravitational force per m3

2
f gB L ( DV )
unit volume
2 2
F fluid force [N] F ( DV L )
inviscid flux term in
F compact form of Navier-
Stokes equations
gravitational acceleration m s 2 2
g gL V
vector
viscid flux term in
G compact form of Navier-
Stokes equations
D2 g L3 T 2 ,
Gr [.] Gr
Grashof number
heat convection W m 2 -K
h h ( DC V )
coefficient
shape functions of
hi , H i [.] hi , H i
element variables
virtual quantity of the
hf variable f
[.] hf
H E + p , enthalpy m 2 s 2 H V
2
k heat conductivity [ W m-K ] k ( DV LC )

kinetic energy of m 2 s 2 2
K K V
turbulence
L* length scale [m] 1
L latent heat m 2 s 2 L V
2
Le D C k , Lewis number [.] Le

average bandwidth of the
mb [.] mb
stiffness matrix
M v c , Mach number [.] M
number of matrix blocks
nb [.] nb
in out-of-core solution
number of blocks while
nbc the specular view factor [.] nbc
matrix is computed

Nomenclature

nbi the specular view factor [.] nbi
matrix is read in
nbr the specular radiosity [.] nbr
matrix is computed
number of solution
nd variables (1 to 7 for [.] nd
u , v, w, p, , K , )
number of degrees of
Nd [.] Nd
freedoms
Ne number of elements [.] Ne
Nn number of nodes [.] Nn
Nt number of time steps [.] Nt
number of time steps for
N ts [.] N ts
which solutions are saved
maximum number of
N es specular elements in one [.] N es
group
maximum number of
N ns specular nodes in one [.] N ns
group
unit boundary normal
n vector outwards from [.] n
computational domain
n normalized flow direction [i ] n
2
p fluid pressure [Pa] p ( DV )
2
p0 pressure datum [Pa] p0 ( DV )
2
pk p ps , kinetic pressure [Pa] pk ( DV )
gi(x x0 ) + p0 , 2
ps [Pa] ps ( DV )
hydrostatic pressure
2
plv evaporation pressure [Pa] plv ( DV )

Nomenclature
2
pvl condensation pressure [Pa] pvl ( DV )
DV LC k , Peclet
Pe [.] Pe
number
Pr C k , Prandtl number [.] Pr
q k , heat flux vector W/m
2
q ( DV T C )
rate of heat generated per W/m3
qB q B L ( DV T C )
unit volume
Q heat [W] Q ( DV T C L )
2
first local coordinate in

r [.] r
element
relative distance from
rf fluid FSI nodes to solid [.] rf
FSI interface
relative distance from
rs solid FSI nodes to fluid [.] rs
FSI interface
R heat resistance m 2 -K W RD*C *V *
DV L , Reynolds
Re [.] Re
number
D2 g L3 T C k ,
Ra [.] Ra
Rayleigh number
specular reflectivity of
s [.] s
radiation
second local coordinate in
s [.] s
element
free surface position
S normal to original free [m] S L
surface
Sc D , Schmidt number [.] Sc
t time [s] tV L
third local coordinate in
t [.] t
element
t transmittance of specular [.] t

Nomenclature
radiation
t time step size [s] tV L
tc critical time step size [s] tc V L
T temperature scale [K] 1 r T
x -component velocity, or
u
normal velocity
[ m s] u V
y -component velocity,
v [ m s] v V
or tangential velocity
v velocity vector [ m s] v V
dimensionless tangential
v+ velocity in turbulent wall [.] v+
functions
shear velocity in near
v
wall regions
[ m s] v V
V velocity scale [ m s] 1
z -velocity, or second
w tangential velocity, or [ m s] w V
wave speed
w mesh velocity vector [ m s] w V*
averaged mesh velocity
wn on the interface of control [ m s] wn V
volumes
first coordinate of
x [m] ( x xr ) L
position vector
x position vector [m] (x xr ) / L
center of rotational
x0 coordinate frame, or [m] (x0 xr ) / L
reference datum location
coordinates of position
xi [m] ( xi xr ) L
vector
reference datum of
xr [m] 0
x-coordinate
reference datum of
xr [m] 0
coordinate vector
y second coordinate of [m] ( y yr ) / L

Nomenclature
position vector
dimensionless normal
y+ [.] y+
distance near walls
dimensionless normal
y + [.] y +
size of near-wall element
dimensionless near-wall
yw+ location used in turbulent [.] yw+
wall boundary conditions
reference datum of
yr y -coordinate [m] 0
third coordinate of
z [m] ( z zr* ) L
position vector
reference datum of
zr [m] 0
z -coordinate
1 , = ratio of
[.]
specific heats
parameter in time
[.]
integration -method
[.]
number of neighboring
nodes, typically 8, 25, 30
m and 125 for elements of [.] m
node=3, 9, 4 and 27,
respectively
[.]
K - turbulence models
( d d ) , thermal
expansion coefficient in [1 K ] T
Boussinesq
approximation
( d di ) , mass
i expansion coefficient in i
buoyancy force
K [.] K
empirical constant in [.]

Nomenclature
[.]
ij Kronecker delta [.] ij
dissipation rate of m 2 s3

3
turbulence
L V
fluid potential m 2 s (V L )
mass-ratio of the i-th
i [.] i *
species
Euler angle in cylindrical
[.]
coordinate system
porosity of porous media [.]
electric potential [V] *
* electric potential scale [V] 1
* mass-ratio scale 1
C p Cv , ratio of specific
[.]
heats
fluid viscosity [ kg m-s ] ( DV L )
, ratio of the fluid
2* [.] 2*
viscosities
dp d , fluid bulk 2
[Pa] ( DV )
modulus of elasticity
von Karman constant in
0 [.] 0
turbulence modelings
permeability tensor of m 2 2
L
porous media
second fluid viscosity [ kg/m-s] ( DV L )
stress relaxation factor in
fluid-structure interaction [.]
problems
displacement relaxation
d factor in fluid-structure [.] d
interaction problems

Nomenclature
CFL CFL number [.] CFL

temperature [K] ( r ) T
reference temperature in
0 [K] ( 0 r ) T
buoyancy force
environmental
e [K] ( e r ) T
temperature
lv evaporation temperature [K] (lv r ) T
vl condensation temperature [K] ( vl r ) T
reference datum of
r [K] 0
temperature
fluid density kg m3 D
s solid density kg m3 s D
coefficient of surface

tension
[ Pa-m] 2
( DV L )
5.6696 108 W/m 2 -K 4 ,
Stefan-Boltzmann W/m 2 -K 4 3
T ( DV C )
constant
electrical conductivity W m-V 2 2
* ( DV L )
3
K [.] K
turbulence models
[.]
[.]
turbulence models
[.]
( p + i v ) + 2 e , 2
[Pa] ( DV )
fluid stress
ps I , kinetic stress, 2
k [Pa] k ( DV )
ps = hydrostatic pressure
p + v ,
2
m mathematical formulation [Pa] m ( DV )
of fluid stress

Nomenclature
specific dissipation rate

of turbulence
[1 s] L V
magnitude of angular

velocity vector
[1 s] L V
angular velocity vector [1 s] L V
Other notations
Notation Explanation
used as subscripts indicating directions of Cartesian coordinates;
i, j, k
repeated indices also mean summations unless noted otherwise.
used as a subscript to indicate a fluid variable when it becomes
f
necessary
used as a subscript to indicate a solid variable when it becomes
s
necessary
r used as a subscript to indicate a relative variable
a used as a subscript to indicate a variable in axisymmetric cases
a nondimensional representation of a variable a
indicates that the value of a is specified as an input parameter in
a
boundary conditions
a indicates that the value of a is defined along fluid-structure
interfaces only
t used as a left superscript to indicate a solution time
X
unless specified otherwise, it is defined as x
i
2
i , Euclidean
vector norm, where xi is the element of X.

Nomenclature
This page intentionally left blank

Chapter 1: Introduction
Chapter 1 Introduction
In this Theory and Modeling Guide, the theoretical bases and guidelines
for the use of the ADINA CFD and FSI capabilities are presented. These
capabilities are implemented in ADINA-F program code. Hence, we use
the term ADINA-F throughout this Guide to refer to CFD and FSI
capabilities.
The intent with this Theory and Modeling Guide is
To provide a document that summarizes the methods and assumptions

used in the computer program ADINA-F.
To provide specific references that describe the features of ADINA-F
in more detail.
Hence, this manual has been compiled to provide a bridge between the
actual practical use of the ADINA-F program and the theory documented in
various publications. Much reference is made to the book Finite Element
Procedures (ref. KJB) and to other publications but we endeavored to be
specific when referencing these publications so as to help you to find the
relevant information.
ref. K.J. Bathe, Finite Element Procedures, Cambridge, MA:

Klaus-Jrgen Bathe, 2006
ADINA-F is a general finite-element/finite-volume code that can be

used for analyzing fluid-flow related problems. They can be
incompressible, slightly compressible, compressible flows and flows in
porous media. Most importantly, solid models created in ADINA can be
coupled with any fluid model for analyzing fluid-structure interaction
problems.
This manual is organized as follows.

We briefly describe the capabilities of ADINA-F and the structure of
this manual in the current chapter.

In Chapter 2, we present theoretical bases of various forms of fluid

related equations and their variations that have been implemented in
ADINA-F.
In Chapters 3 through 7, we present separately our formulations of the
fluid models, including the governing equations, numerical methods,
available elements or control volumes, boundary conditions, initial
conditions and material models. As an exception, we present the
incompressible, slightly compressible and low-speed compressible fluid
models altogether in Chapter 3, since many similarities exist in these
models.
The formulation of mass transfer is presented in Chapter 8, again
including the governing equations, numerical method, available elements,
boundary conditions, initial conditions and material models. The mass
transfer can be coupled with any fluid model described in Chapters 3-7.
The fundamental theory in fluid-structure interactions is presented in
Chapter 9. The physical coupling of the fluid and structure models is
described. The iterative (partitioned) and direct (simultaneous) numerical
procedures are presented.
In Chapter 10, all elements that have been implemented in ADINA-F
are detailed for further reference.
The methods for solving nonlinear equations as well as the associated
linear equation solvers are described in Chapter 11.
The rest of the capabilities or options are described in Chapter 12,
including the CFL option, ATS option, skew systems, constraint equations,
conjugate heat transfer option, element birth-death option, pressure datum,
ALE formulation, etc.
In Chapter 13, some guidelines and experiences for using ADINA-F are
described. They include the most frequently asked questions, the possible
modeling errors that may often occur, strategies on choosing and testing
models and on obtaining converged solutions, etc.
In the appendices, the figures and tables are listed. All possible error
and/or warning messages that ADINA-F may print are also listed in the
order of the message code numbers together with explanations.
Finally, the index can be useful in finding topics quickly.
The following table presents a quick overview of the capabilities that
are available in ADINA-F. Some combinations may not be available in
certain cases. For more details, check the related chapters.

Table 1-1 Overview of capabilities of ADINA CFD & FSI
Category Capability
incompressible, slightly compressible and low-speed compressible
flow models
high-speed compressible flow model
Fluid models K - and K - turbulence models
Note: Heat transfer flow through porous media
must be included in solid element groups
compressible flow heat transfer
models. mass transfer
electro-static and steady current conduction analyses
volume of fraction
liquid-vapor phase change
ADINA solid model: mechanical interaction, porous media
interaction, and thermal interaction
porous medium element groups can be coupled with
incompressible, slightly compressible and low-speed compressible
Coupled models flow models
heat transfer can be included in fluid models
mass transfer can be included in fluid models
fluid and heat transfer are coupled with electro-static and steady
current conduction
Computational domains 2D planar, 2D axisymmetric and 3D
Analyses steady-state and transient
Galerkin 2D 3/6/9-node elements
Elements or control Galerkin 3D 4/29-node elements
volumes FCBI and FCBI-C 2D 3/4-node elements
FCBI and FCBI-C 3D 4/5/6/8-node elements
Constant, time-dependent, temperature-dependent, pressure-
Material models dependent, pressure-temperature-dependent, non-Newtonian,
velocity-dependent, user-supplied, etc.
prescribed solution variables, zero solution variables, zero flux of
solution variables, prescribed rotational velocity, concentrated
force load, distributed normal-traction load, field centrifugal load,
concentrated heat flow load, distributed heat flux load, heat and
mass convections, radiation.
Boundary conditions wall, free surface, fluid-fluid interface, phase-change, fluid-
structure interface, gap, specular radiation, thermal resistance,
uniform flow, angular velocity, fan, vent, friction, , user-supplied,
etc.
external flow, supersonic at inlet, subsonic at inlet, supersonic at
outlet, subsonic at outlet, symmetric.
Gauss elimination method (COLSOL)
Solvers for linearized
sparse solver
equations
iterative methods (RPBCG, RPGMRES, and AMG solvers)

automatically nondimensional procedure

automatic time-stepping CFL option
automatic time-stepping ATS option
skew system
mapping solution
restart analysis
Other capabilities constraint condition
conjugate heat transfer
element birth-death option
pressure datum
include/exclude hydrostatic pressure
physical or mathematical formulation
with or without dissipation

2.1 General conservative Navier-Stokes equations
Chapter 2 Various forms of fluid equations
In this chapter, we summarize the governing equations used in various

fluid flow models in ADINA-F.
The motion of a continuous fluid medium is governed by the principles

of classical mechanics and thermodynamics. In a fixed Cartesian
coordinate frame of reference, they can be expressed in conservative forms
for mass, momentums and energy, respectively

+ i( v ) = 0
t
v
+ i( vv ) = f B
t
E
+ i( vE i v + q) = f B i v + q B (2.1)
t
where, t is the time, is the density, v is the velocity vector, f B is the

body force vector of the fluid medium, is the stress tensor, E is the
specific total energy, q is the heat flux and q B is the specific rate of heat
generation. The specific total energy and the stress are defined as
1
E= vi v + e b + e
2
= ( p + i v ) + 2e

Chapter 2: Various forms of fluid equations
where e is the specific internal energy, b is the specific kinetic energy, p

is the pressure, and are the two coefficients of fluid viscosity and e
is the velocity strain tensor
e = 12 (v + v )
The heat flux is assumed to obey Fourier's law of heat conduction
q = k
where is the temperature and k is the heat conductivity coefficient.

One of the most important body forces included in f B is the
gravitational force
f gB = g
where g is the gravitational acceleration vector.

The above equations are based on the Eulerian approach for the
description of the continuum motion: the characteristic properties of the
medium are considered as functions of time and space in the frame of
reference.
The basic or the independent solution variables of the governing
equations are either the primitive variables ( p, v, ) or the conservative
variables ( , v, ) . In ADINA-F, the conservative variables are used
as solution variables in high-speed compressible flow models and the
primitive variables are used in other flow models.
In order to obtain a closed system for solution variables, additional
equations called state equations must be provided to correlate the variables,
p, , , and e . They are usually provided in the form
= ( p, ), e = e ( p , )
which can be obtained either from fundamental thermodynamics or directly

from the fluid properties.

Based on the types of state equations, various forms of equations can be

derived for describing incompressible, slightly compressible and
compressible flows.
When different coordinate frames of reference are applied, the
governing equations are different too. These frames can be fixed, rotational
or in arbitrary Lagrangian-Eulerian coordinate systems.
The governing systems that are applied to geometries of two-
dimensional planes, two-dimensional axisymmetric planes and three-
dimensional spaces have various representations as well.
When the turbulence of flows is required, additional turbulence
equations must be solved. These equations can have various forms too,
depending on the flow types, coordinate frames, geometries of
computational domains, etc.
The equations for the transportation of neutral species must be
presented as well to be associated with the fluid equations.
All these equations are related to each other and discussed in the
following sections.
2.2 Nonconservative Navier-Stokes equations
The nonconservative forms for mass, momentums and energy,

respectively, can be expressed in a fixed Cartesian coordinate frame of
reference as

+ vi + i v = 0
t
v
+ viv i = f B
t

Cv + Cv vi + iq = 2 D 2 + Sc + q B (2.2)
t
where D is the deformation rate (shear rate) and Sc is the heat

source due to the fluid compressibility:

D = e e eij eij
Sc = i v ( p + i v )
The nonconservative form of the momentum equation is obtained by

subtracting v i (mass equation) from the conservative momentum equation,
and the nonconservative form of the energy equation is obtained by
subtracting (e b) (mass equation) and v i (momentum equation) from
the conservative energy equation. In this procedure, the variation of
internal energy from thermo-dynamics de = Cv d is used, where Cv is
the specific heat under the constant volume condition. In incompressible
flows, Cv is equal to the specific heat C p that is obtained at constant
pressure.
2.3 Rotating frame of reference
Consider the rotating frame defined in the figure below, where

(E1 , E2 , E3 ) are the base vectors of the fixed Cartesian coordinate system
and ( X 1 , X 2 , X 3 ) are the coordinates used therein, (e1 , e 2 , e3 ) are the base
vectors of the Cartesian coordinate system of the rotating frame,
( x1 , x2 , x3 ) are the coordinates used therein, x0 = x0 (t ) is the origin of the
rotating frame, and = (t ) is the angular velocity vector that, without
losing generality, is assumed to be in the fixed direction e3 . In order to
have a clearer presentation, we further use to represent the time in the
rotating system. The task here is to transform the Navier-Stokes equations
from the fixed Cartesian coordinate system to the rotating Cartesian
coordinate system.

2.3: Rotating frame of reference
e3
e2

3
e1
x0
Dx = x x 0
x
2
1
Figure 2.1 Rotating coordinate frame
First, we obtain from the definitions,
e1 (t ) = (t )e 2 (t )
e2 (t ) = (t )e1 (t )
e3 (t ) = 0
Second, consider an arbitrary fluid particle trajectory expressed in the

rotating coordinate system
x = x0 + x = X 0i Ei + xiei
Its velocity is, by taking the time derivative,
v = x = X 0i Ei + xiei + xiei
(2.3)
= v 0 + v r + x
where v 0 is the velocity of the frame center, v r is the velocity with respect
to the rotating frame and x is the Coriolis velocity resulting from
the rotation of the frame. It can be further shown that

t = ( v v r )i
i v = iv r
v r iv = v r iv r + v r
e = 12 (v r + v r )
Using these observations, the nonconservative governing equations in the

rotating coordinate system can be expressed as

+ v r i + i v r = 0

v
r + v r iv r i = f B + fc

Cv + Cv v r i + iq = 2 D 2 + Sc + q B

We can see that these equations are the same as the nonconservative forms
of the Navier-Stokes equations by simply replacing the velocity v by the
relative velocity v r and adding the additional force term
fc = 0 x 2 v r x
It is important to observe that these equations are independent of the

coordinate system. We conclude that solving a problem for ( p, v, ) is
equivalent to solving a problem for ( p, v r , ) with a centrifugal force. The
advantage of solving for relative velocity is that we may only deal with a
small relative velocity, particularly when the rotational velocity is very
large.
It must be remembered, however, that the relative velocity is with
respect to the rotating frame. The conditions imposed on relative velocity
may be different from those on the velocity. Whenever the velocity is
required, it must be calculated according to Eq.(2.3), using the computed
relative velocity.

2.3: Rotating frame of reference
It is further noted that the stress tensor is the same as calculated either
using the velocity or the relative velocity. This observation is in fact the
theoretical basis that these analysis procedures can directly be applied to
fluid domains in fluid-structure interaction analyses.
The equivalent conservative forms of the equations in the rotating frame
are

+ i( v r ) = 0

v r
+ i( v r v r ) = f B + fc

Er
+ i( v r Er i v r + q) = (f B + fc )i v r + q B

where, Er = 12 v r i v r + e , is the representation of the specific total energy

with respect to the rotating frame. Notice that the specific total energy is
different from Er
E = Er + 12 ( v 0 + x)i( v 0 + x + 2 v r )
This equation indicates that the conditions imposed on the total energy
must be modified if the frame is rotating.
2.4 Planar flow
A planar flow is also referred to as a two-dimensional plane flow. In

this type of flow, the solution variables do not change on planes parallel to
the flow direction. The governing equations for planar flows can be
directly deduced from the general forms of the Navier-Stokes equations by
choosing a plane that contains the flow trajectories as a coordinate surface.
In ADINA-F, the surface is always defined as x1 =const., so that
x1 = v1 = 0 . The resultant equations, in conservative or
nonconservative forms, in stationary or rotating frames, are the same as the

equations described in the earlier sections. Of course, the operator in

planar flows must be interpreted as 2 x2 + 3 x3 .
2.5 Axisymmetric flow
An axisymmetric flow is also referred to as a two-dimensional

axisymmetric flow. In this type of flow, the solution variables are the same
on each plane defined by an arbitrary angle in a cylindrical coordinate
system (see the figure below). Solutions can therefore be defined in a
domain on the ( y , z ), with y corresponding to the r -coordinate in the
cylindrical coordinate system.
z, x3
Computational domain
for axisymmetric flows
y, x2
Figure 2.2 Coordinate system for axisymmetric flows
In order to obtain the governing equations for axisymmetric flows, the

Navier-Stokes equations are rewritten in the cylindrical coordinate system
( y, , z ) = ( x2,, x3 ) and then simplified by using the assumptions =
const., = 0 and a zero angular velocity in -direction. As in the

2.5: Axisymmetric flows
planar flow case, we use the notational definition

2 x2 + 3 x3 .
First, the shear rate in the axisymmetric plane can be expressed as
e = e a + e a
where
e a = 12 (v + v )
e a = ee v2 y1
and the deformation rate is then
D = e e = e a e a + (v2 y1 )
2
Similarly, the stress tensor can be expressed as
= a + a
with
a = p + y1i( yv ) I + 2e a
a = 2e a
Since the normal boundary direction of any domain on the ( y, z ) plane is

perpendicular to the angular direction e , it follows
a in = in
This equation indicates that the stress computed on the ( y, z ) plane can be
applied, as in three-dimensional space, to structural boundaries in fluid-
structure interaction analyses.

Finally, we present the axisymmetric flow equations, in a conservative

form,
( y )
+ i( y v ) = 0
t

( y v ) + i y ( vv a ) = y f B + fa
t

( y E ) + i y ( vE a i v + q) = y (f B i v + q B )
t
and, in a nonconservative form,

y + i( yv ) + yvi = 0
t
v
y + y viv i( y a ) = yf B + f a
t

yCv + yCv vi + i( yq) = y (2 D 2 + q B ) + Sca
t
where
Sca = i( yv ) p + y 1i( yv )

fa =

( ie ) = e 2 p + y 1i( yv ) + 2v2 y1
2.6 Incompressible flow
The compressibility of a fluid, measured as the change of fluid volume

or, equivalently speaking, as the change of density under a variation of
external pressure, can be represented by the fluid bulk modulus of elasticity
( )
(= p ) . The parameter w = represents the wave speed
of a small perturbation traveling in the fluid medium. When Bernoullis

2.6: Incompressible flows
equation p + 12 v i v = const. is applied, the density variation can be

approximated by
2
p v
=
2 w2
It is therefore clear that a flow can be recognized as incompressible if the

fluid speed is much slower than the wave speed. Strictly speaking, the
process conditions for which the bulk modulus is defined whether
isothermal, isentropic, etc. must be designated.
For liquids, these distinctions are not particularly important. Taking
water as an example, 2 109 Pa , and 103 kg m3 under standard
conditions. This means that the wave speed in water is of the order
1400m/s or 5000km/h . In most water flows, fluid velocities are far less
than the wave speed.
For air under isentropic conditions, p = const. and
w = p/ = c , where c is the sound speed and ( C p / Cv 1.4) is
the ratio of specific heats. Therefore, is approximately equal to
M 2 , where M is called the Mach number. For most applications of air
flows, M < 0.3 can be used as a critical value to define incompressible
flow conditions.
In some applications, the fluid is confined and subjected to an incoming
flow rate or deformations of boundaries. The flow must then be treated as
compressible even though the deformations are small. In these flows, the
compressibility of fluids is a physical phenomenon of interest and required
to be modeled. These flows are discussed in another section as slightly
compressible flows.
In general, incompressible flows are characterized by the state
equations
= const., e = Cv
Applying these conditions to the nonconservative forms of the Navier-

Stokes equations, the governing equations for incompressible flows and
heat transfers are obtained

i v = 0
v
+ viv i = f B
t

Cv + Cv vi + i q = 2 D 2 + q B
t
where the stress tensor is also simplified to
= pI + 2e
Let us discuss in more detail the following cases.
Mathematical formulation
Assuming a constant viscosity and applying the mass conservation

i v = 0 to the stress tensor, we have
i = i m
where m = pI + v . Using m instead of as stress here is

purely mathematical since m differs from the physical stress. In
particular, the difference of their normal components on a boundary
( m )in ( v in)
may not be small along open boundaries. However, it can be small enough
to be negligible near solid walls. Fluid boundary conditions along most
open boundaries comprise either prescribed velocities or applied distributed
normal-traction loads. Therefore, in general, the mathematical formulation
can be used if no distributed normal-traction conditions are applied. For
most practical analyses, mathematical formulation is not recommended.

Boussinesq approximation
Consider a flow field with temperature variations, where the

gravitational force is important. In the Boussinesq approximation, the
density is assumed constant except for the gravitational force, which is
expanded to the first order with respect to the temperature variation

= 0 + ( 0 )

where 0 is the density at a reference temperature 0 . Introducing the

thermal expansion coefficient
1
=
0
the gravitational force becomes
f gB = 0 g 1 ( 0 )
For many flows, can be approximated by an average value near the

reference temperature 0 . When the temperature variation is large in the
simulation range, must be treated as a function of temperature.
Because the values of density that appear in other places in the
equations are always assumed constant in the Boussinesq approximation,
we will not distinguish the notation from 0 in the Boussinesq
approximation.
Decoupling of fluid flows from the temperature field
In the case where the body force f B and the fluid constitutive relations
do not depend on the temperature, or the temperature variation is small
enough to be negligible in the flow field, the continuity equation and
momentum equations can be solved without simultaneously solving the
energy equation. The temperature field, however, must then be solved with

the calculated fluid velocities. In this case, the fluid flow is decoupled from
the temperature field.
Effect of dissipation
The importance of the dissipation term, 2 D 2 in the energy

conservation equation depends on the fluid properties and the flows. Let us
compare the major terms in the energy equation using some characteristic
scales, L* , V * , T * and D for length, velocity, temperature and density
respectively,
dissipation V * Ec
* *
=
convection Cv D T L Re
2
dissipation V *
= PrEc
diffusion kT *
The effect of the dissipation term can then be neglected if both terms are
small.
For air where Pr 0.7 and Ec M 2 , the dissipation term can be
neglected if the Mach number is very small since the Reynolds number is
usually not small.
For oil, on the other hand, the Prandtl number could be as high as 104 ,
then whether to drop the dissipation term depends on the Eckert number
and the Reynolds number.
Finally, whenever the dissipation is small enough to be negligible, the
Eckert number is generally small and the temperature is weakly coupled to
the flow field.
Hydrostatic pressure
In a stationary fluid domain, the pressure becomes the hydrostatic

pressure that is balanced by the gravitational force
p = ps = g i(x x 0 ) + p0

where p0 is a pressure datum at x 0 . In general, the hydrostatic pressure

ps is always part of the absolute pressure solution p. The hydrostatic
pressure is not coupled with the velocity field and thus hidden in the
solution procedure. In many flows, the hydrostatic pressure is much larger
in magnitude than the kinetic pressure
pk = p ps
In this case, the presence of the hydrostatic pressure may cause instability
in numerical computations. It is then recommended to use the kinetic
pressure instead of the absolute pressure as the solution variable. This
procedure can be enforced by means of simply removing the hydrostatic
pressure term from the gravitational force
fbB = f gB ps = g ( 0 )
The force fbB is called the buoyancy force. Thus, the kinetic pressure
replaces the pressure and the buoyancy force replaces the gravitational
force. Keep in mind that the stress used in the new formulation is not the
physical stress. It is called the kinetic stress and does not contain the
hydrostatic pressure effect
k = + ps I
Hence, if this formulation is applied to fluid-structure interaction problems,

the hydrostatic pressure must be applied separately as an additional force to
structures.
2.7 Fluid flow through porous media
A porous medium saturated with a fluid has very different properties

than a pure fluid medium (see the following figure for a typical saturated
porous medium). The porous domain consists of both fluid and solid. In
the case where the solid is rigid, fluid flows depend only on the properties

of the fluid region that is formed by many small holes of possibly different
shapes and sizes. Although the Navier-Stokes equations are valid for flows
through porous media, the simulation of fluid flows through these regions
of small scales is impractical and far beyond the capacity of present day
computers if a large domain is considered.
solid
mean
flow v
fluid
p, vm 2R p+dp
Figure 2.3 Illustration of porous media models
In order to determine the flow properties in porous media, the mean (or
averaged) velocity must be used. During the averaging procedure, an
important property of the porous medium, called permeability, is
introduced.
Let us consider a simple example as shown in the figure above, where
viscous fluid flows through a micro-circular hole are subjected to a constant
pressure drop. The equation that governs the flow can be obtained from
momentum equation
dp m
= y
dx y y y
where vm is the velocity of fluid in the hole. By imposing the no-slip

condition on the wall, the exact solution can be obtained

2.7: Fluid flow through porous media
1 p 2
vm =
4 x
( y R2 )
Integrating the solution along the cross section of the hole, we obtain the
averaged velocity through the hole
1 2 R R 2 p
v =
R2
0 0
vm ydyd =
8 x
Considering that such holes are sparsely located in a porous medium that
has a porosity (the ratio of the volume occupied by the fluid to the
volume of the mixed medium), the averaged velocity in that medium is
R 2 p
v = v =
8 x
The coefficient R 2 8 here is called the permeability of the porous

medium discussed in this example. It is clear then that the permeability has
units of square length.
Of course, the actual permeability of a porous medium depends on the
shapes of the saturated region, that could differ from each other even within
the same porous medium. Furthermore, the permeability may be
anisotropic and flow dependent. Nevertheless, a more general form that
governs flows through porous media can be expressed as
F v v + 1 i v = p + f B (2.4)
where
= ij Ei E j
is the permeability tensor and F is the inertia factor. Eq.(2.4) is called

Darcys law or the Darcy-Forchheimer equation. It governs the momentum
conservation of fluid flows through porous media.

The continuity equations obtained in all sections are valid for the
averaged velocity, which controls the mass conservation of the fluid in the
porous medium considered. In other words, the porous media flows can be
assumed to be incompressible, slightly compressible or compressible.
The energy equations obtained in the previous and later sections are all
valid as well, in both fluid and solid regions in porous media. In order to
obtain an averaged temperature solution, the material properties are
averaged using fluid and solid properties
Cv = ( f Cv f ) + (1 )(sCv s )
k = k f + (1 ) ks
The continuity equation, Darcys law (replacing the momentum

equations) and the energy equation (using averaged material data) form the
governing equation system for fluid flows through porous media.
The Darcys law equation is different from the Navier-Stokes equations.
For instance, there is no derivative of the velocity. Some conventional no-
slip wall boundary conditions may not be appropriate for porous media
flows. For example, consider a saturated porous medium in a pipe
subjected again to a constant pressure drop as shown in Fig.2.3. The size of
the pipe here of course is much larger than that of the micro-holes
discussed earlier. There would be no solution if a no-slip condition were
applied on the wall. Instead, Darcys law gives the exact solution if a slip
wall condition is applied.
Also note that Darcys law has no time derivatives. Therefore, if the
body force f B is constant, it is unnecessary to perform transient analyses.
Many fluid flow problems in porous media are in fact coupled with
temperature using the Boussinesq approximation to the gravitational force.
Finally, for a better understanding of Darcys equation, let us apply the
mass conservation condition to it,
i ip = i if B
This is a type of Poisson equation for pressure. The pressure is then very
much like a velocity potential.
The pressure in porous media sometimes is also called pore pressure.
When the deformability of the solid in porous media is considered, the

2.7: Fluid flow through porous media
velocity in Darcys equation becomes the relative velocity v w , where w

is the velocity of the structure.
2.8 Slightly compressible flow
Any fluid exhibits some compressibility. Even some fluids

conventionally classified as incompressible, like water and oil, exhibit the
compressibility that cannot be ignored, for example, in confined flows. The
compressibility of a fluid is characterized by its bulk modulus of elasticity
. As discussed in Section 2.6, when the fluid velocity is much smaller
than the wave velocity, the flow can generally be treated as incompressible.
However, the compressibility can be a key factor in so-called
"incompressible" fluids. For example, when a fluid is subjected to a sudden
load of disturbance, e.g., a load of pressure, a very high fluid pressure can
quickly build up. Whenever the propagation of a pressure wave is to be
calculated, or the fluid domain is enclosed by deformable boundaries, the
compressibility must be included. This fluid response to such a system is
considered in many applications of fluid and fluid-structure interaction
problems, such as shock absorbers and air bags in the automobile industry,
traveling of pressure waves in explosions, simulation of noise, etc.
A slightly compressible flow model is suitable for such problems. It
can be identified by the state equations
p
= 0 1 + , e = Cv (2.5)

where is the fluid density with the compressibility and 0 is the density
at p = 0 . The conservative form of continuity equation remains the same
as that in the general case. The nonconservative form of the continuity
equation in slightly compressible flows can be written as
0 p
+ vip + iv = 0
t
In the axisymmetric case, this equation becomes

y0 p
+ v ip + i( yv ) = 0
t
The momentum and energy equations for incompressible flows are still
valid.
To be able to further understand the behavior of the slightly
compressible model, let us consider the special case where the velocity is
negligibly small, f B = 0 and is a large constant. The continuity and
momentum equations can be simplified to
0 p
+ 0i v = 0
t
v
0 + p = 0
t
Fluid flows governed by these equations are called acoustic flows. They
can be further simplified by introducing the fluid potential ,

v = , p = 0
t
to
2
= w22 (2.6)
t 2
where w is the wave speed. Eq.(2.6) is called the acoustic equation. This
equation can be solved in general using the ADINA program (see Theory
and Modeling Guide, Volume I: ADINA Solids & Structures).
We can see that the slightly compressible flow model is a more general
version of acoustic flows.

2.9: Compressible flows
2.9 Compressible flows
A typical fluid medium of compressible flows is air. The ideal gas law
p = (C Cv ) , e = Cv (2.7)
with constants specific heats C p and Cv is usually used in the simulation

of gas flows. Under most conditions, the specific heats depend only on
temperature although they may depend on pressure as well in rare
occasions.
The physical phenomenon of compressible flows differs considerably
from incompressible flows. A most important difference is the
interdependence of pressure, temperature and density. This dependency
results in a strong coupling of the momentum and energy equations, and
presents a highly nonlinear behavior in solutions.
The most important parameter that characterizes the difference is the
Mach number M. The flow is classified as subsonic when M < 1,
supersonic when M > 1 and transonic if the flow varies between the
subsonic and supersonic regimes. As soon as the Mach number exceeds the
magnitude one, shock waves, strong rarefactions, sonic surfaces, and
contact discontinuities begin to appear. These phenomena present serious
difficulties in computational fluid dynamics. Therefore, in general, the
methods that are used to analyze incompressible flows are not effective for
high-speed compressible flows. Furthermore, some special boundary
conditions that deal with these nonlinear behaviors and coupling of
variables are required uniquely for each flow regime along the boundaries.
On the other hand, when the Mach number is very small, M 1 , the
flows behave much like incompressible flows and yet the methods that are
effective for high-speed compressible flows may not be appropriate or
applicable to flows in this regime. The methods that are suitable for
incompressible flows may then be applied to this type of flow even if the
compressibility is included.
We will, therefore, distinguish compressible flows for high- and low-
speed regimes. In our approach, the state equation for compressible flows
is given in Eq.(2.7). The conservative form of the continuity equation is
used for all compressible flows. For high-speed flows, the conservative
equations are used, and the solution variables are conservative variables

( , v , ) . For low-speed flows, conservative equations are used for

FCBI and FCBI-C elements while nonconservative equations are used other
types of elements. The solution variables are primitive variables ( p, v, ) .
If FCBI-C element is used, there is an option for using either the total
energy equation (2.1) or the heat transfer equation (2.2). In addition to a
general temperature dependent ideal gas law, some real gas models are also
available for FCBI-C element with the total energy equation: the standard
Redlich Kwong model, the Aungier Redlich Kwong model, the Soave
Redlich Kwong model and the Peng Robinson model.
2.10 High-speed compressible flow
The conservative forms of compressible flow equations can be written

in a compact form
U
+ i F (U ) + G (U) = C
t
where
0 0

U = v , G = , C = fB

E i v + q f i v + q
B B
(2.8)
0

F(U) F( v, U) = vU + pI

vp

and and H are, respectively, the shear stress and the enthalpy

2.10: High-speed compressible flow
= + pI = i vI + 2e
p
H =E+

Note that we have expressed F as a function of v and U . The argument

v here represents only the convective velocity defined in the first term on
the right-hand side of Eq.(2.8). This notation will be more convenient later
in the discussion on the ALE formulation.
When viscosity and heat conductivity are neglected (G = 0), the flow is
called inviscid and the governing equations are the Euler equations.
Consider the flux vector in a direction n, Fn = niF , subject to a variation
of the variable U ,
Fn = AU
where A = Fn U is the Jacobian of the flux function Fn . The matrix

A can be further written in terms of its eigenvectors and eigenvalues
A = D 1
where the column vectors of the matrix P are right eigenvectors, the row
vectors of the matrix P1 are left eigenvectors and D is the diagonal
matrix which consists of the eigenvalues
1 1 1 0 0

= v + cn v cn v 1 2 (2.9)

H + uc H uc H c 2 v v
1 2
uc + b cn v

uc + b cn v
1
P1 = 2 2c 2 2b 2 v 2 (2.10)

2c 2
2c v1 2 c 2 1 0
2c 2 v 2c 2 2 0
2

D(u , c) = diag {u + c, u c, u , u , u } (2.11)
where, (n, 1 , 2 ) form a locally orthogonal coordinate system and

(u , v1 , v2 ) are the corresponding velocity components.
When the flow is steady, t = 0 , we can see that the Euler equations
are singular at locations where v = 0 . Such a position is called a
stagnation point. Hence, if the Mach number becomes very small, the
system becomes sensitive and eventually becomes singular when the Mach
number approaches zero.
Furthermore, we can see that waves propagate at different speeds
(represented by the eigenvalues defined in Eq.(2.11)) in their corresponding
directions (represented by the eigenvectors defined in Eq. (2.9)). Special
boundary conditions are therefore required to deal with such flow
complexities. These conditions are discussed in Chapter 6.
2.11 Turbulence models
In fluid mechanics, the flows of practical relevance are almost always

turbulent; this means that the fluid motion is highly random, unsteady and
three-dimensional. Due to these complexities, the turbulent motion and the
associated heat and mass transfer phenomena are extremely difficult to
describe and to present theoretically and numerically. Theoretically
speaking, the Navier-Stokes equations are still valid for describing
turbulent flows. However, the storage capacity and speed of present day
computers are far from sufficient to allow a solution for any practically
relevant turbulent flow. The reason is that the turbulent motion contains
scales that are much smaller than the extent of the flow pattern described by
the averaged variables. On the other hand, when the speed of fluid flow is
not very high, the random motion of fluid does not contribute much energy
to the mainstream fluid flow compared with that of averaged variables.
Only when such random motions of fluid become too strong to be ignored,
they must be modeled into the stream of averaged variables. Generally, a
high Reynolds number characterizes turbulent flows.

2.11: Turbulence models
Many turbulence models have been developed to predict the effect of

these random flow motions to the flow patterns of regularly large scales. In
these models, the viscosity and heat conductivity coefficient are modified
to
= 0 + t , k = k0 + kt
where 0 and k0 are, respectively, the laminar viscosity and the laminar
heat conductivity, while t and kt are their turbulence counterparts,
which are computed as described in the following models.
Large-eddy-simulation model for incompressible flows
The large-eddy-simulation model is an algebraic model, where t and

kt are computed by directly defined functions
t C p
t = 2k D2 2 D, kt =
Pr
in which D is the effective deformation rate, k D is a dimensionless model

constant, is the element size and Pr is the turbulent Prandtl number.
K - turbulence model in incompressible flows

turbulence model: K - model
In this model, t and kt are formed as
K2 t C p
t = c , kt =

where K and , being called kinetic energy and rate of dissipation of the
turbulence respectively, are defined as
1
K = v i v , = 0 (v ) (v )
2

Here v is the fluctuating velocity or the velocity difference between the

turbulent velocity and the averaged velocity. K and are additional
variables governed by two additional equations
( yaK )
+ i y a ( vK q K ) = y a S K
t
( y a )
+ i y a ( v q ) = y a S
t
where a is zero for two- and three-dimensional flows and one for
axisymmetric flows. Other corresponding terms are defined as

q K = 0 + t K
K

q = 0 + t

S K = 2t D 2 + B

S =
2c1t D 2 c2 + c1 (1 c3 ) B
K

B = 0 + t gi

Here, c , c1 , c2 , c3 , K , , and are model constants.

The K - turbulence model turbulence model: K - modelis
applicable to certain high Reynolds number flows.
K - RNG turbulence model in incompressible flows

turbulence model: K - RNG model
The K - RNG model uses the same formulae as the turbulence K -
model except the coefficient c1 in the -equation is modified to
A (1 A / 4.38 )
c1 c1
1 + 0.015 A3

where A = 2 KD . The K - RNG turbulence model is also applicable

to certain high Reynolds number flows.
- high Reynolds number turbulence model in incompressible

flows
turbulence model: K - high-Reynolds number model
The variable , initially introduced purely from a mathematical point
of view, is called the specific dissipation rate of turbulence. The variable
is related to K and as
Then t and kt are computed by
K t C p
t = , kt =

The governing equations for K and are
( yaK )
+ i y a ( vK q K ) = y a GK
t
( y a )
+ i y a ( v q ) = y a G
t
where a is zero for two- and three-dimensional flows and one is for
axisymmetric flows. Other corresponding terms are defined as

q = 0 + t

GK = 2t D 2 K K + B

G =
K
( 2 t D 2 K + B )

Here, , , K , , K , , and are model constants.
- low Reynolds number turbulence model in incompressible

flows
turbulence model: K - low-Reynolds number model
The basic equations for the low Reynolds number model are the same
as those for the high Reynolds number model. However, the following
parameters are used
1 40 + RK 1 10 + R 1 5 18 + R
=h , = h , K = Kh
1 + RK 1 + R 1 + R
where the superscript h indicates the values defined in the high Reynolds
number model and
4
R R R K
RK = t , R = t , R = t , Rt =
6 2.7 8 0
- turbulence model in compressible flows

turbulence model: K - model
In this model, the effects of compressibility have been taken into
account. The specific total energy is modified to include the kinetic energy
of turbulence
1
E = vi v + e + K
2
and the stress tensor is also modified to
2
= p + i v K I + 2e
3
With these modifications, the - equations of the - model in

incompressible flows are used.

Turbulent wall function

turbulence model: K - modelturbulence model: K - model
An important step towards the successful use of the two-equation
models, the - model and the - model, is a proper treatment of the
fluid-solid boundary conditions. In particular, the dissipation rate can
become infinite and therefore requires special attention. When boundary
conditions are applied directly at the wall, the equations must be integrated
through the viscous sub-layer, which is undesirable, since many elements
have to be placed in this sub-layer for proper resolution of the high
Reynolds number turbulence phenomenon. Instead, extensive research has
been conducted and has provided empirical laws of sufficient generality
that connect the wall conditions to dependent variables just outside the
viscous sub-layer. The universal law of the near-wall velocity profile, also
called the turbulent wall function, is the most frequently used law. The so-
called Reichardt law is used in ADINA-F

+ +
y y
1
v W ( y ) = ln (1 + 0 y ) + 7.8 1 e 11 11

+ + + y+
e 3
0

where
v + = ( v - w ) / v
y + = v y 0
v is the velocity component tangential to the wall, w is the tangential

velocity of the solid, v is the shear velocity, y is the distance from the
wall, 0 is the von Karman constant, is the fluid density and 0 is the
laminar viscosity.
Shear Stress Transport (SST) model in incompressible flows
The main idea behind the SST turbulence model is to combine the
accuracy of the - model in the near wall region and the free stream
independence of the - model. As a result, the SST model delivers an
improved separation prediction and no free stream dependency.

The governing equations for the - model apply to the SST model
without modifications. However, the eddy viscosity is redefined in the
following way:
K
t = s a1
max(a1 , 2 DF2 )
Additionally, the source term in the equation of the SST model, Gs , has
an additional contribution called the cross-diffusion term, C:
Gs = G + C = G + 2 (1 F1 )2 (K ) ( ln )
The functions F1 and F2 are designed to be one near the wall and zero for
the free shear layers
(
( F1 , F2 ) = tanh ( 14 ) , tanh ( 22 ) )
where

K 500
4 K
1 = min
max
,
2
, + 22

K y y C y

2 K 500

2 = max
,

0.09 y y
2

C + = max {C ,1010 }
The constants used in the SST model = ( s , s , s , Ks , s ,...) are

calculated using the function F1 in the following way:
= F11 + (1 F1 )2

where 1 and 2 are the constants for the K - and the K - model
respectively.
Spalart-Allmaras model
The Spalart-Allmaras (SA) is a one-equation model. The new

variable , obtained through the SA transport equation is called the
modified eddy viscosity that has the same unit as the kinematic
viscosity, namely, L2 / t .
The eddy viscosity in the SA model is defined as,
t = f v1
where f v1 is a non-dimensional function of and will be given

later.
The SA transport equation for is,
( y a )
+ i y a (v q ) = y aG
t
where
1
q = ( + )
s
2
c
G = cb1S cw1 f w + b 2 ( )i( )
d s
and d is the distance to the closest wall. The related functions are
defined as

3
f v1 = 3
+ c31

S=S+ 2
1
d 1 + f v1
2
1/ 6
1 + cw6 3
f w = g 6
g +c
6
w3
where
=
S = + C prod min (0, e )
e = 2D
= 1
2
( = v (v) ) T
g = r + cw 2 ( r 6 r ) (r = (S d ))
2 2
and cw1 , cw 2 , cw3 , , cb1 , cb 2 , cv1 , s , C prod , are model constants.
The wall boundary condition for the SA transport equation is

= 0 . For the momentum equations, the wall function approach or
near wall treatment is used to obtain the wall shear stress in the same
as in the two-equation models.
The typical inlet or free stream boundary condition is prescribed

. The desired values should be about (0.033 i Re) , where
is the laminar viscosity of the fluid, Re is the inlet Reynolds
number, and i (= 0.01 0.1) is turbulence intensity. For most free
stream conditions, = 0.1 / can be used.

Detached Eddy Simulation (DES) model
DES is initially developed to solve problems with large regions

of separation and high Reynolds numbers. For such type of flows
LES is very expensive and Reynolds-Averaged Navier-Stokes
models (RANS) are not accurate enough.
DES uses a one-equation RANS model (in our case the Spalart-
Allmaras model) which functions as a subgrid scale model (SGS) in
regions where the grid density is fine enough and as a RANS model
in other regions. This is accomplished by replacing the distance to
the wall, d, in the SA model by
d = min(d , CDES )
where 0.61 < CDES < 0.78 (typically 0.65) and is an element
length scale,
(VOL 1/ 3
element ) 3-D
=
( AREA 1/ 2
element ) 2-D
Since the value of d controls the switching between SGS-LES

and RANS, the user actually controls the behavior of the model
through the size of meshes. Note that in DES the filter is of order
.
Two-layer zonal model
In the two-layer zonal model, the whole domain is subdivided into a

near-wall viscous region and a fully turbulent region. This division
is determined by a wall-distance based Reynolds number defined by
Ky
Ry =
0

where y is the shortest distance from the location to the wall. In the
fully turbulent region ( Ry > Rc ) , the standard K - model is
employed. However, in the near-wall viscous region, a general one-
equation model is employed. In this model, the turbulent viscosity
T and are computed from
1
T = C K 2 l
3
=K 2
l
where the length scales are computed from
y A
R

l = cl y 1 e

( )
Ry

l = cl y 1 e A
Note that , in the near-wall viscous region, is not obtained by

solving the - equation; it is instead obtained algebraically from the
above formula.
An indicator of a good mesh for the two-layer zonal model is that the
nondimensional wall distance, y + 1 . Another indication of an
ideal mesh quality is that at least 10 layers of elements are within the
viscous layer. Since the parameter Rc plays a role in switching
between the two models, its values are somewhat flexible. It is more
convenient to specify the number of viscous layers, rather than the
value of Rc . Therefore, we allow the user to input a range of
Rc [ Rc min , Rc max ] , as well as a preferred number of near-wall
layers ( Lw ) . The program adjusts Rc in the given range such that
the number of viscosity-affected element layers is as close to Lw as
possible. It is understandable that, even with the automatic
adjustment, a sufficient number of elements must be placed in near-
wall regions.

The two-layer zonal model is available for the FCBI-C and FCBI
elements and is always associated with the turbulence K - model.
In practice, the empirical constants A and A are found to be not

only physically dependent, but also scheme-dependent. They are
constants that should be calibrated and embedded into the program.
In our study, the preferable values of A and A are found to be the
same and shown in the table below.
Values of A and A
Cases
FCBI-C elements FCBI elements
3D 105.5 100
2D Axisymmetry 52 19
2D plane 50 70
The default values of the other empirical constants are
cl = 2.43, Rc min = 50, Rc max = 400, Lw = 10
In order to obtain converged solutions at each time step, the near-

wall elements are fixed after a certain number of iterations. The
convergence criteria are
either i I upper
or i I lower and re min {ee , e -lower }
where i is the iteration number, and re and e are, respectively, the

effective residual and effective tolerance. The values of the
constants I upper , I lower , e and e -lower are currently fixed and shown
in the table
I lower I upper e e -lower

FCBI-C 30 100 10 0.01

FCBI 7 20 10 0.01
2.12 Mass transfer
Mass averages
Consider n neutral species of density i , ( i = 1, 2,..., n ) that are

transported in a fluid medium of density 0 . The mass density is defined
as usual as the mass of species per unit volume. It is sometimes also called
mass concentration or partial density of the mixture.
The mass of each of these species must be conserved if there is no
exchange between these species and the environment. Allowing a mass
creation rate mi generated by other species, the mass conservation of each
of the species reads
i
+ i(i v i ) = mi
t
where v i is the velocity of the species i. Summing up the mass

conservations of all species, we obtain
n n n
i
+ i i i = mi
v
t i=0 i=0 i=0
When there is no mass exchange between the considered media and their
environment, the right-hand side of this equation becomes zero. A
comparison of this equation with the general fluid conservative equation
reveals two key concepts in mass transfers, namely, the bulk density and
the bulk velocity of the media. They are defined respectively as
n n
1
= i , v = v

i i
i =0 i =0

2.12: Mass transfer
According to Ficks law, the deviation of the mass flow rate is proportional
to the corresponding mass concentration gradient
i
+ i( vi Di ii ) = mi
t
where di is the mass diffusivity, sometimes called mass diffusion

coefficient. In general, di is a tensor of the second order. Combining
Ficks law and the species conservation equation, we obtain the mass
transfer equation
i
+ i(i v di ii ) = mi
t
We may multiply this equation by an arbitrary constant, say , to

represent the same equation as
i
+ i(i v di ii ) = mi
t
In this case, the solution variable and the mass creation rate become * i
and *mi respectively. This fact is important for us later to generalize our
approach to mass transfer problems.
In many applications, it is more convenient to use the mass-ratios
i
i =

as solution variables. If the bulk density can be well approximated by a

constant, as it is in incompressible flows, the mass transfer equation can be
represented by
i
+ i( vi di ii ) = mi
t

Molar averages
Another important approach to mass transfer problems is the so-called

molar concentration, particularly in gaseous mixtures. In this approach, the
molar concentration Ci = N i V is used as the solution variable, where N i
is the number of moles of species i in the mixture and V is the volume of
the mixture. Since the molecular weight M i is constant and
i = Ni M i V = Ci M i , the mass conservation of the species i can be written
as
Ci m
+ i(Ci v i ) = i
t Mi
Similarly, we can introduce the concepts of the bulk molar density and the
molar averaged velocity respectively by
n
1 n
C = Ci , v = Ci vi
i =0 C i =0
Applying Ficks law to the molar flux Ci ( v i v ) = di iCi results in the

molar transfer equation
Ci m
+ i(Ci v di iCi ) = i
t Mi
Again, introducing an arbitrary constant, we obtain
Ci m
+ i(Ci v di iCi ) = i
t Mi
in which we treat *Ci as the solution variable and * mi M i as the

creation rate.
Similarly, the molar-ratio i = Ci C can also be used as a solution
(
variable. When the molar mass of the mixture M = C = i = 0 i M i
n
)

2.12: Mass transfer
and the bulk molar density C are considered constant, the molar transfer
equation can be represented by
i Mmi
+ i(i v di ii ) =
t Mi
A single approach to the mass transfer equations
We have so far introduced the mass density i , the mass-ratio i , the

molar density Ci and the molar-ratio i . Their properties are governed by
the corresponding equations introduced earlier in this section. These
equations fundamentally behave in the same manner and can be solved in a
single formulation as follows.
Let the solution variable, diffusion coefficient and source term be
denoted by i , Di and mi , respectively. Let the transport equation that
governs the solution variable be represented by
i
+ i( vi Di ii ) = mi
t
together with a general definition of the bulk density through a general

formula
n n
= i i i ( i ) (2.12)
i =0 i =0
Based on the choice of solution variables, every term in the equation has a
specific meaning. In particular, irrespective of the choice of the solution
variable, the bulk density has the specific expression given above.
The most direct choice of a solution variable is the mass-ratio. In this

case, i is the mass-ratio, Di = d i , mi = mi''' and the bulk density is
the physically meaningful bulk density. The coefficients of the bulk
density are then i = 1, ( i = 1, 2,..., n ) . The bulk density is then
expressed in terms of the mass-ratio

n
0
= 0 + i = n
i =1
1 i
i =1
In the event that the mass-ratios are small, the problem can be well
approximated by choosing i = 0, ( i = 1, 2,..., n ) . In this case, the
fluid equation is separated from the mass transfer equation. If the
mass creations are further zeros, the mass-ratios are decoupled from
each other as well. The resulting equation represents the simplest yet
most popular formulation of mass transfer in incompressible flows.
When the molar-ratio is used for the solution variable, i represents the
molar-ratio i , Di = d i , mi = Mmi M i and the bulk density is also
the physical bulk density. In order to write the bulk density in terms of
molar-ratios, we follow the definitions of molar averages
n
= 0 + i
i =1
Ci M i
n
= 0 + CM
i =1 CM
n
M
= 0 + i i
i =1 M
which shows that 0 = 1 and i = M i M ( i = 1, 2,..., n ) . The bulk

density can then be similarly expressed in terms of the molar-ratio
0
= n
1 ii
i =1
As before, if the molar-ratios are small, the problem can be well

approximated by choosing i = 0, ( i = 1, 2,..., n ) .

2.12: Mass transfer
When the solution variable is the mass density, i represents the mass
density i , Di = *di , mi = * mi and bulk density represents any
arbitrary constant * . A reasonable choice is the fluid density 0 .
This formulation can therefore only be applied to incompressible flows.
The obvious choice of the coefficients in the bulk density is 0 = 1
and i = 0, (i = 1, 2,..., n) .
When the solution variable is the molar density, i represents the molar
density Ci , Di = *di , mi = * mi M i and the bulk density
represents any arbitrary constant * . A reasonable choice is the fluid
density 0 . This formulation can therefore only be applied to
incompressible flows. The obvious choice of the coefficients in the
bulk density is 0 = 1 and i = 0, (i = 1, 2,..., n) .
For simplicity of description of the mass transfer solutions, we will

always call i the mass-ratio, is the bulk density, Di the mass
diffusion coefficient and mi the mass creation rate.
The governing equations for mass transfer in axisymmetric case are
yi
+ i y ( vi Di ii ) = ymi
t
The general procedure to couple the mass transfer solutions with fluid
flow solutions is as follows. We solve the mass transfer equation for
i = 1, 2,..., n and compute the bulk density defined in Eq.(2.12). This
density will be used in the mass transfer equation. Except for the high-
speed compressible flow case, the bulk density will also be used in fluid
equations. In this case, the density 0 represents the original fluid density
introduced in the previous sections. The bulk density introduced here is a
correction of the fluid density due to the non-uniform distribution of
additional species.

Correction to buoyancy force
Similar to the Boussinesq approximation to the temperature-dependent

density in incompressible flows, the mass-dependent density can also be
simplified in many practical problems. In this case, we assume that the
bulk density is independent of the mass-ratios except in the buoyancy force
term, where it is extended to include the mass-ratios
n

f gB = g = 0 g 1 ( 0 ) + i (i i 0 )
i =1
where i is a mass expansion coefficient and i 0 is a reference mass-ratio

in the expansion. In this approach, the coefficients of the bulk density must
be 0 = 1 and i = 0, (i = 1, 2,..., n) , based on the assumption made here.
Corresponding to the mass solution variables, the mass-ratios, mass
densities, molar-ratios and molar-densities, i =1, 1 0 , M i M and
M i 0 can be used, respectively.
2.13 Steady-state equations
The steady-state equations are obtained from the governing equations

by removing all dynamic terms, i.e., we assume t = 0 . The steady-state
analysis is to solve the steady-state equations. The solutions obtained in
steady-state analyses are therefore in steady state.
It is noteworthy that a steady-state solution can be obtained by
performing either a steady-state analysis or a transient analysis. For
problems that have multi-solutions (for example, bifurcation problems), in
general, transient analyses must be performed.

2.14: Arbitrary-Lagrangian-Eulerian formulations
2.14 Arbitrary-Lagrangian-Eulerian formulations
In fluid dynamics, moving boundaries are frequently encountered, for

example, prescribed moving walls, deformable structures, free surfaces, etc.
(see the following figure for a typical example). When any part of the
computational domain is deformable, the Eulerian description of fluid
flows is no longer applicable and the Lagrangian description must be used.
Anywhere else in the fluid domain, the fluid flow can be described in an
arbitrary coordinate system as long as it meets the coordinate requirements
along the physical boundaries. Such a description is called an Arbitrary-
Lagrangian-Eulerian (ALE) formulation. Obviously, the Navier-Stokes
equations discussed so far need to be rewritten in the ALE system.
fixed wall, moving wall,

Eulerian Lagrangian
coordinate is used coordinate is used
w(t)
ALE coordinate is used
Figure 2.4 A typical example of using ALE coordinates
The purpose of choosing an ALE system is to freeze the

computational domain. Let us consider a transformation such that the new
coordinate is the sum of the initial coordinate and its displacement:
x = + d ( , ) x ( , )
t =
where the moving coordinate system (x, t ) has been transformed into a
new coordinate system ( , ) . Note that, to give a more precise
explanation, we have used to represent the time in the new coordinate

system. The vector d ( , ) is used to deal with the arbitrarily moving

coordinates.
3
2
3
1
2
1
Figure 2.5 Definition of the ALE coordinate system
The time derivative of an arbitrary, f ( x, t ) = f ( + d(, ), ) , is
f f x f
= + i
t x
Using w ( x = d ) to represent the moving coordinate velocity,

we have
f f x f f
= i = w i f (2.13)
t x
When this equation is applied to time derivatives in the Navier-Stokes
equations that are expressed in a purely Eulerian coordinate system, the
differential forms of ALE equations are obtained. For example, in the ALE
system, the nonconservative continuity equation becomes,

+ ( v w )i + i v = 0


2.14: Arbitrary-Lagrangian-Eulerian formulations
Hence, the only difference between an ALE formulation and the Eulerian
formulation is that the relative velocity replaces the convective velocity.
Furthermore, the case w = 0 corresponds to a purely Eulerian description,
while w = v corresponds to a purely Lagrangian description.
When Eq.(2.13) is directly used, however, only the nonconservative
forms of Navier-Stokes equations can be obtained. The conservative ALE
Navier-Stokes equations can only be expressed in their integral forms. In
general, a conservative form of the differential equation
f
+ iB = R
t
in a fixed coordinate system will be transformed into an integral form in an
ALE coordinate system

fdV + (wf + B)i dS = RdV
V S V
\where V ( t ) is an arbitrary material volume enclosed by its surface S (t )
and S is the surface vector of S (t ) in the outward direction. The

quantities f , B and R are tensors such that B is one order higher than f
and R . In particular, the volume can be any control volume if finite
volume methods are used. We will frequently refer to this equation when
finite volume methods are discussed.
Using this equation, all the conservative forms of equations introduced

in the previous sections can be directly transformed into their
corresponding integral forms of conservation. For example, the compact
form of the conservative equation can be written in the following integral
form, in an arbitrarily moving system,

UdV + [F (v, U) + G ]i dS = CdV
V S V
where v = v w is the relative velocity. Furthermore, the Jacobian

of Fn ( v, U ) has the same eigenvectors as defined in Eqs.(2.9) and
(2.10). Its eigenvalues, however, have been modified to D (u , c ) ,
where u = u niw u wn .

2.15 System units and nondimensional forms
The Navier-Stokes equations and associated conditions and material

data can be used in any system of units (S.I. units, USCS units, etc.) as long
as all units are consistent. In particular, dimensionless units can be used.
To define nondimensional units, characteristic values of all the variables
entering the Navier-Stokes equations must be constructed from a
predetermined set of independent reference quantities and scales. They are
listed in the Nomenclature.
All other variables can then be derived from these basic quantities and
scales. For example, using the tilde to indicate dimensionless quantities,
we have
x - xr tV
x= , t= , = ,
L L D
v p
v= , p= , = ,
V DV 2 D V L

f B L
= , fB =
D V L

DV
2
The nonconservative momentum equations become, using nondimensional

quantities,
v
+ viv i = f B
t
where
= ( p + i v ) I + 2e
1
e=
2
( v + v T )
It is noted that material data now are , , etc.

2.15: System units and nondimensional forms
For compressible flows, since the temperature is strongly coupled to the

other fluid variables, we use
V 2
T =
, r = 0
C
In natural convections, since there is no free stream, a proper velocity

scale can be

V =
D L
where is a characteristic viscosity.

All parameters, in dimensional and nondimensional forms, are listed in
the Nomenclature section of this manual. The following well-known
dimensionless parameters are listed next since they are useful indicators of
flow patterns.
Reynolds number, defined as
DV L 1
Re = =

is the ratio of the inertia force to the friction force, which measures the
intensity of turbulence and the boundary layer thickness. Typical transition
Reynolds numbers to turbulent flows, for internal and external flows
respectively, are of the order 103 and 106 . The typical fluid boundary layer
thickness is of the order Re 2 L .
1
Prandtl number, defined as
C
Pr = =
k k
is the ratio of the diffusivity of momentum to the diffusivity of heat. Note

that the Prandtl number depends only on fluid properties. Typical Pr values

are, under normal conditions, 0.01, 0.7, 10 and 1000 for mercury, gas,
water and oil, respectively.
Rayleigh number, defined as
2 3
D g L T *C g
Ra = = = Pr Gr
k k
is the ratio of the energy generated by the buoyancy force to the energy
dissipated by thermal diffusion and fluid viscosity. Ra is an important
parameter in natural convection problems to measure the intensity of flows
and the thickness of the thermal boundary layers. When the inertia force is
negligible in the momentum equation, the typical thermal boundary layer
thickness is of the order Ra 4 L . In another limit, when the viscous force
1
is negligible, the typical thermal boundary layer thickness is of the order

Bo 4 L , where Bo = RaPr is the Boussinesq number.
1
Mach number, defined as
v
M=
c
is the ratio of the fluid velocity to the speed of sound . The Mach number
is an important parameter in determining the regime of fluid flows. It is
usually acceptable that flows can be treated as incompressible when the
Mach number is smaller than 0.3. Once the Mach number exceeds one,
flows become supersonic and many kinds of phenomena related to shocks
may occur.
Boussinesq number, defined as
2 3 2
D g L T *C g
Bo = 2
= = RaPr
k k2

2.15: System units and nondimensional forms
is the ratio of the energy generated by the buoyancy force to the energy
dissipated by thermal diffusion. As mentioned earlier, when the
contribution of the flow viscosity is negligible, the Boussinesq number
measures the thermal boundary layer thickness.
Grashof number, defined as
2 3
D g L T * g Ra
Gr = = =
2
2 Pr
is the ratio of the buoyancy force to the viscous force. When Pr < 1, the
Grashof number is an indicator of the thickness of fluid shear viscous layer
within the thermal boundary layer. The typical thickness of the shear layer
is of the order Gr 4 L .
1
Peclet number, defined as
DV LC 1
Pe = =
k k
is the ratio of the heat transferred by thermal convection to the heat

transferred by conduction. The Peclet number is an important parameter in
forced convection problems. The thickness of the thermal boundary layer
is of the order Pe 2 L .
1
Eckert number, defined as
2
V
Ec =
C T
is the ratio of the kinetic energy to the internal energy. In compressible

flows, Eckert number is of the same order as the Mach number. For a
perfect gas,
Ec = ( 1) M 2C p

Schmidt number, defined as

Sc = =
D D
is the ratio of the diffusivity of fluids to the diffusivity of mass-ratios.

Schmidt number is the analog of the Prandtl number between the fluid and
the mass transfer. For most common liquids, Sc is two or more orders of
magnitude greater than Pr. For gases, Sc is typically of the same order of
magnitude as Pr.
Lewis number, defined as
D C D Pr
Le = = =
k k Sc
is the ratio of the diffusivity of the mass-ratio to the diffusivity of heat. It is

typically of the order 10-2 or 10-3 for most common liquids and order one
for gases.
Capillary number, defined as
V
Ca = =

is the ratio of the viscous forces to the surface tension forces. The smaller
the Capillary number, the more important the effect of surface tension.
Brinkman number, defined as
2
V
Br = = Pr Ec
kT
is the ratio of the energy dissipated by the fluid viscosity to the energy
dissipated by thermal conduction.

3.1: Governing equations
Chapter 3 Formulation of incompressible, slightly

compressible and low-speed compressible
flows
In this chapter, we introduce our formulation for incompressible,

slightly compressible and low-speed compressible flows. Heat transfer can
be included in incompressible and slightly compressible flows. In low-
speed compressible flows, the temperature must be included.
The fluid flow models introduced here can be coupled with ADINA
solid models in fluid-structure interaction analyses. They can also be
associated with mass transfer analyses. Porous media models can be
employed as additional element groups.
Solving a fluid flow problem involves solving the governing equations
associated with the input of physical fluid material data, well-posed
boundary conditions, initial conditions and the use of appropriate numerical
methods. These issues are discussed in this chapter.
ref. Bathe, K.J. and Zhang, H., A flow-condition-based

interpolation finite element procedure for incompressible
fluid flows, Computers & Structures, Vol. 80, pp. 1267-
1277, 2002.
ref. Bathe, K.J., Zhang, H., and Ji, S., Finite Element Analysis
of Fluid Flows fully Coupled with Structural Interactions,
Computers & Structures, Vol. 72, pp. 1-16, 1999.
ref. Bathe, K.J., Zhang, H., and Zhang, X., Some Advances in
the Analysis of fluid flows, Computers & Structures, Vol.
64, pp. 909-930, 1997.
ref. Bathe, K.J., Zhang, H., and Wang, M.H., Finite Element
Analysis of Incompressible and Compressible fluid flows
with Free Surfaces and Structural Interactions, Computers
& Structures, Vol. 56, pp. 193-214, 1995.

Chapter 3: Formulations of incompressible, slightly compressible and
ref. Bathe, K.J., Walczak, J., and Zhang, H., Some Recent
Advances for Practical Finite Element Analysis, Computers
& Structures, Vol. 47, pp. 511-521, 1993.

are available for incompressible, slightly compressible and low-speed
compressible flows.
Table 3-1 Functionalities of incompressible, slightly

compressible and low-speed compressible flow
models
Category Functionality
ADINA solid model
porous medium element groups
solid element groups
heat transfer (always coupled in compressible flows)
Coupled models
mass transfer
volume of fraction
2D: 3-node linear element and 6/9-node bilinear elements
Galerkin Elements
3D: 4-node linear element and 27-node bilinear elements
FCBI and FCBI-C 2D: 3/4-node linear elements
elements 3D: 4/5/6/8-node linear elements
constant
time-dependent
power-law
Carreau
temperature-dependent
Material models temperature-dependent power-law
second order
large-eddy-simulation
ASME steam table
Pressure-temperature-dependent
user-supplied

prescribed velocities
v
zero velocities
prescribed pressure p
zero pressure
prescribed rotational velocity
concentrated force load
distributed normal-traction load
v
Boundary conditions field centrifugal load
fixed wall
Note: The column uniform flow
furthest to the right angular velocity
represents the solution moving wall
v, d
variables or equations fluid-structure interface
that are immediately free surface
affected by the fluid-fluid interface D
specified condition. phase-change
The solution variable gap all
d indicates a moving prescribed temperature
boundary condition. zero temperature
concentrated heat flow load
distributed heat flux load
convection
radiation
specular radiation
thermal resistance
user-supplied all
Initial conditions zero (default conditions)
Note: Zero pressure or specified
temperature cannot be mapped from other solutions
specified in
compressible models. restart run
sparse solver
equations
automatic nondimensionalization procedure
skew systems
constraint condition
Other capabilities conjugate heat transfer
pressure datum
physical or mathematical formulations for incompressible flows
with or without dissipations

3.1 Governing equations
The governing equations used in our formulations are in general given

in Chapter 2, in various forms, corresponding to different cases.
Irrespective of the solution variables used, the final solutions for
visualization purpose are always primitive variables ( p, v,,...) . The ALE
formulation described in Section 2.14 is used for moving mesh problems.
It is worth noting that the heat transfer equation is used as default. For
FCBIC element and low-speed compressible fluid, user can choose either
the heat transfer equation (as default) or the conservative form of the
energy equation that includes the total energy and the state equation of gas.
In practice, heat transfer equation is robust, but is not always accurate. An
example where the heat transfer equation is not accurate is quasi-static
polytropic compression and expansion. On the other hand, total energy
equation is more accurate, but could be numerically less stable at higher
flow velocities.
It is worth noting that the heat transfer equation is used as default. For
FCBIC element and low-speed compressible fluid, user can choose either
the heat transfer equation (as default) or the conservative form of the
energy equation that includes the total energy and the state equation of gas.
In practice, heat transfer equation is robust, but is not always accurate. An
example where the heat transfer equation is not accurate is quasi-static
polytropic compression and expansion. On the other hand, total energy
equation is more accurate, but could be numerically less stable at higher
flow velocities.
3.2 Numerical method
3.2.1 Time integration
In steady-state analyses, the transient terms in the governing equations

are removed and therefore there is no time integration involved. In this
case, time will only be used in time-dependent materials, loads and
boundary conditions. The time integration method discussed here is only
for transient analyses.

3.2: Numerical methods
Assume that the solution has been obtained at time t and the next
solution is to be calculated at time t + t , where t is the time step size.
The initial condition defines the solution at time 0. The algorithm used to
obtain the solution at time t + t is the basic procedure to obtain,
successively, the solutions at all required times.
We use two implicit time integration methods: the Euler - method of
the first order Euler -method and the ADINA composite scheme of the
second order.
With an Euler method, the equation u t = f ( u ) is computed by
t +t
u = t u + tf ( t +t
u)
where t +t
u = (1 ) t u + t +t u . The Euler method is a scheme of the first
order accuracy and is unconditional L- stable if 12 < 1 . Note that
although it gives an accuracy of the second order in time, the trapezoidal
rule ( = 12 ) is numerically unstable unless the velocity is extremely small.
The default first order scheme is the Euler backward method ( = 1 ).
In the ADINA composite method, the solution at time t + t is
obtained in two consecutive sub-time-steps
t + t
u = t u + tf ( t + 12 t
u )
t +t
u= t + t
u + (1 ) tf ( t +t
u)
where t + t
u = (1 ) t u + t +t u , = 2 1 and = 2 ( 2 1) .
With the choice of 1
2 < < 1 , the method has second order accuracy and is
unconditionally L- stable. The default is = 1 2 , which was proven to
give minimum truncation error for linear systems. Although the
computational cost is doubled per time step, the composite scheme provides
solutions of better accuracy and may have less CPU time overall since
fewer number of time steps can be used.

3.2.2 Discretized equations
The finite element method is used to discretize the governing equations.

The finite element equations are obtained by establishing a weak form of
the governing equations using the Galerkin procedure. The continuity
equation, the momentum equation and the energy equation are weighted
with the virtual quantities of pressure, velocities and temperature. The
governing equations are integrated over the computational domain V. The
divergence theorem is used to lower the order of the derivatives of the
stress and the heat flux, resulting in the expressions of natural boundary
conditions. These conditions are part of the usual boundary conditions
discussed later in this chapter.
Our variational forms of governing equations depend on the flow
models, frames of coordinate systems and geometry dimensions. As an
example, the conservative forms of the three-dimensional equations can be
integrated and expressed in a compact form as
(h G f + Q f ih f ) dV = h f Q f i dS
f
(3.1)
V
where f represents p, v and for the continuity equation, momentum

equations and energy equation, respectively; h p , h v and h are virtual
quantities of pressure, velocity and temperature, and

Gp =+ i( v )
t
v
Gv = + i( vv ) f B
t
E
G = + i( vE i v ) f B i v q B
t
Qp = 0
Qv =
Q = q
In particular, if h f are step functions, the Galerkin method becomes the

finite volume method (according to conventional definition). For these
elements, a set of Flow-Condition-Based-Interpolation (FCBI) functions

has been used (see Chapter 10 for details). When a degree of freedom is
defined at the element center, the FCBI element becomes FCBI-C element.
The FCBI and FCBI-C elements can be 2D 4-node quadrilateral and 3-node
triangle, 3D 8-node brick, 4-node tetrahedron, 5-node pyramid and 6-node
prism. All FCBI and FCBI-C elements of the same dimensions can be
mixed in one model problem.
3.2.3 Upwinding techniques
The finite element formulation of incompressible fluid flows

ref. KJB follows the standard Galerkin procedure. Since the elements
Sections 4.4 associated with this procedure satisfy the inf-sup condition, excellent
and 4.5
results may be obtained for low Reynolds/Peclet number flows.
When Reynolds/Peclet numbers become larger, the scheme becomes
less stable and eventually unstable when certain critical conditions
ref. KJB are violated. These conditions may be met far before the flows
Section 7.4.3
become physically turbulent. Therefore, such instability originates
solely from the numerical methods that are employed. The
numerical instability cannot be overcome by simply employing a
turbulence model, since a numerically unstable method obviously
cannot be applied to turbulence models as well.
In order to understand the difficulties of high Reynolds/Peclet number

flows, let us consider a simple example of the one-dimensional heat transfer
equation expressed in nondimensional form:
Pe ' ( x ) '' ( x ) = 0
( 0 ) = 0, (1) = 1
where Pe ( = C p v L k ) is the Peclet number and L is the length of the

model problem. The exact solution of the problem is
e xPe 1
= (3.2)
e Pe 1
Let us use N linear elements of equal size x = 1/ N and apply the

standard Galerkin method to the model equation. The finite element

equations that govern the variables defined at nodal points

xi = ix, ( i = 0,1, 2,...N ) are
P P
1 + i 1 + 2i 1 i +1 = 0
2 2
where P (= Pex) is called the cell Peclet number.

The exact solution (which is the numerical solution) of the discretized
equation can be obtained
ai 1
i =
aN 1
where a = (1 + P 2 ) (1 P 2 ) . It is noted that 2 is a critical value of P, at

which a becomes infinity. In case P > 2 ( a < 0 ), a i is positive for even
numbers of point indices and negative for the odd numbers. This indicates
that the numerical solution oscillates when the cell Peclet number is larger
than 2. It is also noted that when P > 2, the diagonal dominance of the
matrix is no longer seen.
In the standard finite element procedure, the shape functions that are
employed are isotropic, i.e., they have equal gradients in all directions.
They are good for low Reynolds/Peclet number flows, in which the
equation system is close to being symmetric. For high Reynolds/Peclet
number flows, the nonsymmetrical property of the underlying continuum
mechanics in the equation system is strong. Such a non-uniform property
cannot be well-represented by elements of uniform shape functions.
Upwinding techniques are introduced to overcome the above
difficulties. The fundamental principle of upwindings is to modify the
solution procedure such that the final discretized equations can more
accurately represent the physical situation.
An obvious approach to a good upwinding scheme is to modify the
value of a such that the numerical solution exactly equals the analytical
solution e P . We can achieve this result by modifying the heat conductivity
coefficient such that
k (1 + ) k

where is an expected correction factor as a result of the upwinding.

Substituting it into Eq.(3.2) and solving for
P P
= + 1 (3.3)
e 1 2
P
Note that is zero when P is zero, and approaches P 2 when P

becomes large.
According to this approach, the best solution (the analytical solution)
can be obtained numerically, if the standard Galerkin method is applied to
the governing equation with the modified k.
In general, fluid flows and governing equations are nonlinear and the
sizes of elements employed are non-uniform. The modification is then
based on the cell Peclet number evaluated in each element.
In multidimensional domains, the construction of a good upwinding
scheme is more challenging. In finite element formulations, various
formulations have been proposed. In our approach to triangular and
tetrahedral elements, all the integrations are evaluated analytically.
Upwinding is then added based on the analytical solutions in the flow
direction.
In the finite volume approach of upwinding, on the other hand, the cell
Reynolds/Peclet numbers are evaluated based on the definitions of control
volumes and used to determine the upwinding.
Both approaches are used to modify our standard Galerkin finite
element methods. We remark that the two approaches are identical in one-
dimensional solutions with constant Reynolds/Peclet numbers.
3.2.4 Solution method for nonlinear finite element equations
The finite element or finite volume equations form a coupled nonlinear

system of all solution variables. The methods used for solving nonlinear
equations and the linear equation solvers are described in Chapter 11.
The global matrix and the right-hand side vector are assembled using
the matrices and the right-hand sides computed within each element. They
also include modifications and/or additions from boundary elements that
describe the specified boundary conditions.

When a zero variable condition is applied, the solution corresponding to

that equation is removed from the governing equations and this procedure
reduces degrees of freedom. For example, if a zero pressure is applied to
node number 3, the continuity equation defined at node 3 is removed and,
consequently, the number of equations is reduced.
When a prescribed variable is applied, the equation corresponding to
that variable is modified as follows. If variable xi is prescribed as xi ( t ) ,
the i -th equation in the linearized system is modified to
Aij = ij , Bi = xi (t ) xi
which is equivalent to the equation
xi = xi (t ) xi
Concentrated force terms are directly added as additional terms to the

right-hand side of the equations. For example, if a force F (t ) is applied to
a velocity component corresponding to equation i , the i-th component of
the right-hand side vector is modified as
Bi + F (t ) Bi
Similarly, a concentrated heat flow load will be directly added to the right-
hand side of the corresponding energy equation.
When a distributed normal-traction load is applied, the nodal force is
computed by integrating the distributed normal-traction load, and then is
added to the equation as for a concentrated force. A distributed heat flux is
similarly integrated to a nodal heat flow and then added to the
corresponding energy equation.
Centrifugal loads are directly formulated in the governing equation and
assembled.
Unlike the treatment of usual boundary conditions, special boundary
conditions are discretized within boundary elements. The boundary sides
of the field elements form the boundary elements. Similar to regular
elements, the coefficient matrices and the right-hand sides of boundary
elements are formulated from the applied boundary condition and
assembled into the global system. A boundary element is treated after the

attached field element equations are formulated and before they are
assembled into the global system. The field element equations may or may
not be then modified depending on the type of the special boundary
condition.
For example, a radiation boundary condition results in a boundary
element equation that is directly formulated in the boundary element and
assembled into the global energy conservation equation. It is clear that
both the matrix and the right-hand side will receive contributions from this
condition.
As another example, when a slip wall condition is applied, the normal
component of the momentum field element equation is replaced by the
boundary condition equation. The tangential component of the finite
element equation will not be changed by this condition.
Special boundary conditions can be more complicated than the two
examples here. In specular radiation boundary conditions, for example, the
concentrated heat flow includes the variable G that cannot be computed
until the radiosity solution R is obtained. Therefore, prior to the
computation of the heat flux, the radiosity equations must be solved. In
order to form this equation, a ray tracing procedure is performed to setup
the view factor matrix. Furthermore, these equations also depend on the
boundary temperature. Thus, the two systems are actually coupled. The
global iteration procedure must then be applied to the radiosity equations.
Unlike the sparse matrix of the fluid equation system, the matrix of the
radiosity equation is full. For most problems associated with specular
boundary conditions, solving the radiosity equations is far more expensive,
in terms of computational speed and memory usage, than solving the fluid
equation system.
3.3 Elements
In Galerkin formulation, 2D 3-node and 3D 4-node linear elements and

2D 9/6-node and 3D 27-node bilinear elements are applicable. The
upwinding technique is used in linear elements and, therefore, they are
suitable for high and low Reynolds/Peclet number flows. Bilinear elements
are suitable for low Reynolds/Peclet number flows.

All FCBI and FCBI-C elements are linear elements. They include 3-
node triangle and 4-node quadrilateral for 2D domains; 4-node tetrahedron,
5-node pyramid, 6-node prism and 8-node brick for 3D domains. Elements
of the same dimensions can be mixed in one problem.
FCBI-C elements are only associated with the Segregated method and
other elements are only associated with the Newton-Raphson method. An
option of the second-order in space is available for FCBI-C elements. Note
that solution accuracy does depend on the element quality. In general,
taking 2D 4-node element as an example, the element should be aligned
along the flow direction as much as possible, and the element size variation
should be as smooth as possible.
3.4 Boundary conditions
3.4.1 General descriptions
Using proper boundary conditions is essential to successfully solving

fluid flow problems. The required boundary conditions depend on the type
of flows or, equivalently, on the governing equations used and the
associated computational domain.
Fluid velocity and pressure are always strongly coupled and must be
treated like one variable as far as boundary conditions are concerned. On
the other hand, the fluid flow might not be directly coupled to the
temperature (although the flow may be coupled to the temperature field
through temperature-dependent materials and the buoyancy force in the
Boussinesq approximation). Therefore, as a general rule, temperature must
be considered separately when applying boundary conditions.
A typical fluid domain and associated boundary conditions are shown
in the following figure, where the whole boundary S of the computational
domain has been divided into S1 (where velocities are prescribed) and S 2
(where a distributed normal-traction load is applied). The Si here
represents the interior part of that part of the boundary, while the over bar
indicates that it contains the boundary of Si as well. Therefore, the
common part of S1 and S 2 is the empty geometric set, while S1 and S 2

3.4: Boundary conditions
share their common frontier where both conditions are applied. This
boundary partition can be expressed as
S1 S2 = S = S , S1 S2 =
prescribed velocities on S 1
applied normal-traction on S2
v=0
v=0 nn
v
Figure 3.1 A typical boundary condition set for incompressible,

slightly compressible and low-speed compressible
flows
From this set of conditions, we observe that:
All parts of the boundary have been assigned one and only one fluid
boundary condition (except on the interface of the sub-boundaries,
where two conditions meet).
For steady-state analyses, at least one velocity condition and one
pressure condition must be specified (the distributed normal-traction
load acts as a prescribed pressure condition).
These observations are fundamental guidelines for specifying fluid

boundary conditions in most fluid flow problems.
A similar boundary condition set for temperature is shown in the
following figure, where the whole boundary S of the computational
domain has been divided into S3 (where a temperature is prescribed) and
S4 (where heat fluxes are applied). This boundary partition can be
expressed as:

S3 S 4 = S = S , S3 S 4 =
applied heat-fluxes on S4
prescribed temperature on S 3
qn =0
qn

Figure 3.2 A typical boundary condition set for temperature
From this set of conditions, we have similar observations:
All parts of the boundary have been assigned one and only one
temperature boundary condition (except on the interface of the sub-
boundaries, where two conditions meet).
For steady-state analyses, at least one prescribed temperature condition
must be specified. Notice that a convection condition or a radiation
condition acts as a prescribed temperature condition.
Variables directly imposed on the solution are referred to as usual

boundary conditions. They are applied to the governing equations in a
simple manner. These conditions are either prescribed conditions or natural
boundary conditions. A prescribed condition forces the solution variable to
be equal to the prescribed value and the corresponding equation is replaced.
A natural boundary condition acts as a force (to the momentum equation) or
a heat flux (to the energy equation), and is added to the right-hand side of
the corresponding equations.
Other boundary conditions may be more complicated and strongly
depend on the solution variables. We call them special boundary
conditions. Unlike usual boundary conditions, the special boundary
conditions containing unknown variables are discretized and assembled

into the global system of equations. The conditions are usually, although
not necessarily, nonlinear and complex.
Unless noted otherwise, special boundary conditions can only be
applied to geometries of one dimension lower than the computational
domain. More specifically, in 2D models, the geometries are lines or
nodes that can be connected to lines; and in 3D models the geometries
are surfaces or nodes that can be connected to surfaces. These
geometries may be located on the boundary or inside the domain,
depending on the type of boundary conditions.
Most boundary conditions impose only constraints on the fluid
variables (velocity, pressure, temperature, etc.). Some other conditions
impose additional constraints on the boundary positions or boundary
displacements. These conditions are called kinematic conditions or moving
boundary conditions. When such a condition is applied, the boundary
nodal displacements are either computed or specified. The nodal positions
inside the fluid domain are adjusted in an arbitrary way to ensure a good
mesh quality. In this case, an arbitrary Lagrangian-Eulerian (ALE)
coordinate system is used in the governing equations. It is understood then
that additional variables (displacements) add additional difficulties and
computer costs in the solution procedure.
Unless explained otherwise, we have used the following notation in the
discussion of boundary conditions. A time-dependent value, say b (t ) is
defined as
b (t ) = b p fb (t )
by a prescribed constant value b p (which has a unit that corresponds to the

prescribed variable) and an associated time function fb (t ) (which has no
units).
3.4.2 Usual boundary conditions for fluid
3.4.2.1 Prescribed velocity
In this condition, a time-dependent velocity is directly prescribed
vi = vi ( t )

and applied to boundaries. The xi -momentum equations at the boundary

nodes are then replaced by this condition.
The prescribed velocity condition is usually applied to inlet boundaries
where velocities are known. It can also be applied to fixed solid walls.
When the Eulerian coordinate system is used on a moving wall, this
condition can also be applied. In this case, the prescribed fluid velocity
equals the velocity of the moving wall, while the displacement is zero. If
non-zero displacements of the moving wall must be modeled, the moving
wall conditions must be applied.
There is also a time birth and death option associated with the
prescribed velocity. If only a birth option is used, the prescribed condition
is inactive until the solution time is larger than the birth time. If only the
death option is used, the prescribed condition is active until the solution
time is larger than the death time. If both birth and death options are used,
the prescribed condition is only active between the birth time and the death
time. It is then understood that the death time must be larger than the birth
time. While the prescribed condition is inactive, the velocity degree of
freedom is free.
3.4.2.2 Zero velocity
Physically, this condition is equivalent to a prescribed zero velocity.

However, the zero velocity condition removes the velocity degree of
freedom. Hence the birth and death options are not applicable.
3.4.2.3 Prescribed pressure
In this condition, a time-dependent pressure is directly prescribed
p = p (t )
and applied to boundaries. The continuity equations at the boundary nodes

are replaced by this condition.
Prescribed pressure condition is usually applied to confined flow
ref. KJB problems to ensure a mathematically well-defined problem. Along
Section 4.5 open boundaries where the pressure is known, a boundary condition
of a distributed normal-traction load is more appropriate.

Similar to prescribed velocity conditions, the time birth and death

options are available in prescribed pressure condition.
3.4.2.4. Zero pressure
Physically, this condition is equivalent to a prescribed zero pressure.

However, a zero pressure condition removes the pressure degree of
freedom and, therefore, the time birth and death options cannot be used.
This condition cannot be used in low-speed compressible flow models,
since a zero pressure is not allowed.
3.4.2.5. Prescribed rotational velocity
In this condition, the velocity is prescribed by means of a time-

dependent angular velocity ( t ) (see also the figure below)
v(t ) = ( t ) ( x x0 )
where, x0 is the center of the rotation and x denotes the coordinates of the
boundaries.
_ _
x
boundary is fixed
_
x
_
(t)
Figure 3.3 Prescribed rotational velocity

This condition can be applied to a wall where the solid boundary is

rotating. It is important to understand that, since the Eulerian coordinate
system is used for the fluid domain, the boundary nodes are fixed while the
fluid velocity equals the solid velocity. Therefore, from a fluid point of
view, the boundary is fixed. Note that a moving boundary or a non-zero
displacement condition cannot be modeled using this condition.
3.4.2.6 Concentrated force load
In this condition, a prescribed time-dependent force F ( t ) is directly

applied to boundaries. The force is added to the right-hand side of the
momentum equations of the boundary nodes. Note that this condition can
directly be applied to nodes as well. This force is also called external nodal
force.
This condition has no effect on the nodes where a prescribed velocity
condition is used since the momentum equation has been replaced by the
velocity condition.
3.4.2.7 Distributed normal-traction load
This is one of the most important usual boundary conditions. It can

only be applied to boundary lines and surfaces of two-dimensional and
three-dimensional computational domains, respectively.
In this condition, a time-dependent normal stress (called normal
traction) nn (t ) = ni in is prescribed. The stress is integrated to an
equivalent nodal force condition
F ( t ) = hv nn ( t ) dS
and then added to the right-hand side of the momentum equations as the
concentrated force load. Here hv is the virtual quantity of velocity on the
boundary.
Note that the normal stress consists of the pressure and the normal shear
stress. Along open boundaries, the normal shear stress is usually negligible
compared with the pressure. Therefore, a normal-traction is usually applied

to open boundaries where the pressure is known. In particular, when

nn ( t ) = 0 , the application of the normal-traction loads is trivial because it
is equivalent to no normal-traction loads.
This condition has no effect on the nodes where a normal velocity
condition is prescribed, since the normal momentum equation has been
replaced by the normal velocity condition.
3.4.2.8 Field centrifugal load
As described in Section 2.3, when the whole field is rotating, the flow
equations can be formulated in a rotational reference coordinate system.
The problem is then equivalent to a problem defined in a fixed coordinate
system subjected to additional (centrifugal) forces. These forces can be
written as
F ( t ) = hv fc dV
where hv is the virtual quantity of velocity and fc is defined in Chapter 2

but rewritten here, for the purpose of indicating the input parameters in the
force definition
fc = x0 (t ) (t ) (x x0 (t )) 2 (t ) v
(t ) (t ) (x x0 (t ))
The center position x0 (t ) and the angular velocity (t ) are required to

define this condition. In principle, the acceleration of the center position,
x0 (t ) , could be evaluated by the program. However, the numerical error
could be severe for many applications. Therefore x0 (t ) is also required in
the input.
Keep in mind that the velocity solved in this coordinate system is the
relative velocity with respect to the rotational frame (which is denoted as
v r in Section 2.3).
Strictly speaking, this condition is not a boundary condition since the
force is applied to the whole domain.
It is noteworthy to make a few remarks.

Consider the case of zero angular velocity. The only term left in the
load is the acceleration of the center position of the frame. This term
can be used to model any field force load. For example, a time-
dependent gravitational force can be modeled. In the meantime, the
gravity must be specified as zero in the material data set to prevent a
duplicated gravity load definition.
In steady-state analyses, the first two terms vanish. No input of the
center acceleration is required.
3.4.3 Special boundary conditions for fluid
3.4.3.1 Fixed wall
At the interface of a fluid and a fixed solid, no-slip or slip conditions

are usually applied. On the fixed wall condition, the boundary is fixed. In
other words, the boundary displacement is zero. Fixed wall conditions can
only be applied to boundary lines and surfaces, respectively, of two-
dimensional and three-dimensional computational domains.
No-slip condition on fixed walls
When a no-slip condition on fixed walls is applied, the fluid velocity

vector on that wall is prescribed to be zero
v=0
This condition is usually applied to wall boundaries in viscous flows. It

is clear that this condition is equivalent to applying a zero velocity or
prescribed zero velocity to all components of the velocity.

zero velocity is
prescribed: v=0
boundary is fixed
Figure 3.4 No-slip condition on fixed walls for incompressible,

flows
Slip condition on fixed walls
On the other hand, when a slip wall condition is applied, the normal
component of the velocity vector is prescribed to be zero
v in = 0
while the tangential components are free or computed as unknown variables

from the governing equations.
In principle, this condition can be replaced by the condition of
prescribed zero normal velocity. However, in the case of irregular
boundaries, the procedure of defining the normal directions is tedious. It is
much more convenient to use this slip condition.
This condition is usually applied to symmetric boundaries and to the
wall boundaries where viscous effects are negligible. In certain
applications, the computer capacity is limited such that large elements have
to be used and, therefore, the boundary layers cannot be modeled. In this
case, only because of the computer limitation, a slip condition may be used.

zero normal velocity is

prescribed: v n=0
boundary is fixed
Figure 3.5 Slip condition on fixed walls for incompressible,

flows
3.4.3.2 Uniform flow
The uniform flow condition is based on the assumption that the flow is
uniform along the normal direction of a boundary, or
v
=0
n
There are no parameters required in the input.

This condition can be applied to the boundaries at which the flow is
almost uniform, for example, on a boundary representing infinity.
Uniform flow can only be applied to boundary lines and surfaces of two-
dimensional and three-dimensional computational domains, respectively.
3.4.3.3 Angular velocity
This condition is similar to the prescribed rotational velocity, but can be

applied to moving boundaries, where the coordinates of the boundary nodes
vary with time (see figure below)
v(t ) = ( t ) ( x(t ) x0 )

Here, we do not consider how the boundary nodes move. In a special case
where the boundary is fixed, this condition is the same as a prescribed
rotational velocity. Angular velocity can only be applied to boundary lines
and surfaces of two-dimensional and three-dimensional computational
domains, respectively.
_ _
x moving or fixed
boundary nodes
_
x
_
(t)
Figure 3.6 Angular velocity boundary condition for

incompressible, slightly compressible and low-speed
compressible flows
3.4.3.4 Moving wall
This condition broadens applications of the fixed wall condition to

moving walls. Moving wall conditions can only be applied to boundary
lines and surfaces of two-dimensional and three-dimensional computational
No-slip condition on moving walls
The no-slip condition on a moving wall is the consistency between the

fluid velocity and the solid velocity
v d(t ) = 0

This implies that the Lagrangian formulation is applied to all directions of

the fluid boundary nodes. The boundary displacement d(t ) is required in
the input of the condition, while the fluid velocity is computed based on the
displacement.
_
velocity is prescribed: v=d(t)
boundary displacement
_
is prescribed: d(t)
Figure 3.7 No-slip condition on moving walls for

compressible flows
This condition is usually applied to moving wall boundaries in viscous

flows. The dominant motion of the moving walls must be in the normal
direction of the boundary. If the tangential movement is dominant, the
moving wall condition with specified tangential velocity should be
considered. They are introduced later in this section.
Slip condition on moving walls
In slip wall conditions, only the normal component of the velocity is

enforced to be consistent with the solid counterpart
(v d(t ))in = 0

while the tangential components are free and computed as unknown

variables from the governing equations. This implies that the Lagrangian
formulation is used for only the normal direction of the fluid boundary
nodes.
This condition is usually applied to the moving wall boundaries where
viscous effects are negligible.
normal velocity is prescribed:

_
v n =d(t) n
_
is prescribed: d(t)
Figure 3.8 Slip condition on moving walls for incompressible,

flows
In the no-slip moving wall boundary condition, the Lagrangian

coordinate system is applied to all directions on the boundary. The
condition sometimes presents difficulty in dealing with moving meshes if
the tangential displacement is large. The use of a Lagrangian coordinate
system in the tangential direction can result in large mesh distortions.
However, the use of the Lagrangian formulation to that direction is not
absolutely necessary. Most tangential movements do not change the shape
of the computational fluid domain. Therefore, an ALE coordinate system
can be applied to the moving wall. In particular, the displacement in the
tangential direction can be completely avoided if the Eulerian formulation
is applied to that direction.
In this ALE formulation, both the tangential velocity and the
displacement vector are prescribed. They may or may not be consistent in
the tangential direction. When they are consistent, the Lagrangian

formulation is implicitly used. When the tangential displacement is zero,

the Eulerian formulation is then specified.
The new method is not only advantageous regarding the mesh quality,
but also necessary for cases in which the validity of the mesh cannot be
maintained.
Two such formulations are presented here.
No-slip condition on moving walls (type=tangential)
In this type of condition, the displacement is prescribed, denoted here as

d(t ) . This displacement causes the boundary to move. The normal fluid
velocity is then computed based on the normal component of the
displacement. The tangential velocity is prescribed separately (see figure
below). This condition is
v = v (t ) + d (t )i(I )
where v (t ) is the prescribed tangential velocity and is the tangential

direction that is computed by means of a specified direction a and the
current normal direction of the boundary n
= an
Note that the tangential direction is not directly required, since it is, in
general, not a constant vector.
velocity is prescribed:
_ _
v=v(t)+d(t) (I)
_
boundarydisplacement
_
is prescribed: d(t)

Figure 3.9 Moving wall condition (type=tangential) for

compressible flows
No-slip condition on moving walls (type=rotational)
When the dominant boundary motion is determined by a rotation (see

the next figure), we can apply a Eulerian formulation to that direction while
leave the Lagrangian coordinate system in the other directions. In this case,
the displacement and the angular velocity are prescribed. The velocity is
computed as
v(t ) = (t ) (x(t ) x0 ) + d (t )i(I )
where x0 is the center coordinates of the rotation, is the angular

velocity vector of the solid wall and the tangential direction is computed
as
(t ) (x(t ) x0 )
=
(t ) (x(t ) x0 )
velocity is prescribed:
_ _ _
v=(t) (x(t)x0)+d(t) (I)
x
_
is prescribed: d(t)
_
x0
_
(t)
Figure 3.10 Moving wall condition (type=rotational) for


compressible flows
3.4.3.5 Fluid-structure interface
A fluid-structure interface is a moving wall for which the interface

displacement is the solution of a solid model. However, fluid-structure-
interaction means much more than just specifying an interface.
First, a solid model must have been created to which the fluid is
coupled. In this solid model, fluid-structure interfaces must be specified
corresponding to the interfaces specified in the fluid model, so that the
program knows which parts of the fluid and solid models are interacting.
Second, since the Lagrangian formulation is used along the interface,
the displacement as well as the fluid velocity are determined by the solid
solution on the interface. This condition is called the kinematic condition
of the fluid model. On the other hand, the fluid force must be applied to the
solid interface to ensure the force balance on the interface. This condition
is called the dynamic condition of the solid model.
Third, the nonlinear-coupled system must be solved to ensure that the
kinematic and dynamic conditions are satisfied. The methods can be either
iterative (between the fluid model and the solid model), or direct (a
combined matrix system is solved).
In this section, we only present the definition of the fluid kinematic
conditions and leave the details of the complete solution procedure of fluid-
structure-interaction to Chapter 9.
The kinematic condition of the fluid-structure interface is exactly
parallel to the moving wall boundary condition, except that the
displacement d(t ) is part of the solution from the solid model, rather than
being prescribed. The condition is also classified as a no-slip condition, a
slip condition and no-slip conditions with prescribed tangential or rotational
velocities. Of course, in the case where a tangential or a rotational velocity
is prescribed, the displacement of the fluid nodes in that direction is
generally (but not necessarily) zero.
This condition can only be applied to boundary lines and surfaces of
two-dimensional and three-dimensional computational domains,
respectively. The boundary geometries must coincide with their
counterparts that are defined in the solid model.

3.4.3.6 Free surface
A free surface boundary is an interface between a liquid and a gas that

has a negligible mass density and is thus treated as a vacuum. The free
surface can be subjected to an ambient pressure and a surface tension. This
condition can only be applied to boundary lines and surfaces, respectively,
of two-dimensional and three-dimensional computational domain.
A free surface is a moving boundary. Two conditions are then applied:
the kinematic and dynamic conditions.
_
pa n +
_
pa n
fluid domain
Figure 3.11 Sketch of a free surface condition
The kinematic condition of a free surface is that the fluid particle cannot
move out of the surface or the normal velocity of the free surface must be
the same as the normal velocity of the fluid
( v d)in = 0
where d is the displacement of the free surface. However, this equation

alone cannot define the displacement of the free surface. Assume that the
tangential displacement is in the Eulerian coordinate system while we leave
the normal displacement in the Lagrangian coordinate system and let it be
denoted as S . In the originally fixed coordinate system (Eulerian), the
kinematic condition becomes
S
+ viS = 0
t

This equation, in a moving coordinate system (here it is the free surface), is

simply a coordinate transformation (see Section 2.14 for details),
S
+ ( v w )iS = 0 (3.4)

We remark that the free surface position is steady here with respect to
the new time frame and S n .
The kinematic condition states that the normal velocity of a fluid
particle on the free surface is equal to the normal velocity of the free
surface itself at the point where the particle is located.
The dynamic boundary condition on the free surface (applied to the
momentum equations of the fluid) is the ambient pressure together with the
surface tension
n = pa n + H
where pa is the ambient pressure, is the surface tension and H is the

curvature vector of the free surface. Let the free surface be r = r ( s1 , s2 ) ,
expressed in a local coordinate system si , ( i = 1, 2 ) , we have
1
H = 2r = i ( Jg ij j r )
J
where
1
r r r r
i i
s s s1 s2
( g ) = ( gij ) = r1 r1
1
ij

i r r
i
s1 s2 s2 s2
J = g11 g 22 g12 g 21
In particular, in a two dimensional space, r = r ( s1 ) ,

m
H=
1 r r m r e
r m r r 2

where m = 0 and 1 in planar and axisymmetric cases, respectively.
It is important to note that, in order to obtain accurate solution of free

surface problems, time step length must be sufficiently small (usually
Courant number vt x 0.1 ).
3.4.3.7 Fluid-fluid interface
A fluid-fluid interface is defined between two fluids. It is a moving

boundary condition. Two boundary conditions are then applied: the
kinematic and dynamic conditions. This condition can only be applied to
internal lines and surfaces, respectively, of two-dimensional and three-
dimensional computational domain.
The kinematic boundary condition is the same as for a free surface.
The dynamic boundary condition on the interface is
(n ) (n ) = H
1 2
where is the surface tension, (n ) and (2)

1
n are the normal components of
the stress tensors in fluids 1 and 2 respectively, and H is the curvature
vector of the interface. The dynamic condition indicates that a surface
tension force is applied to the interface between two fluids. The pressure
and velocity are assumed to be continuous across the interface.
3.4.3.8 Phase-change boundary
This condition is applied to problems involving changes of phase

between liquid and solid, such as those occurring in crystal growths, metal
castings, etc. This condition can only be applied to internal lines and
surfaces of two and dimensional computational domains, respectively.
These internal geometries must be shared with the (ADINA-F) solid and

fluid element groups. The solidification temperature must be prescribed as

an additional usual boundary condition on the interface.
The phase-change boundary is also a moving boundary condition. The
kinematic condition is that the liquid obeys a no-slip wall condition. The
dynamic boundary condition on the interface is obtained by using the mass
flux conservation and energy conservation (including latent heat release
across the interface). The resultant condition becomes
q f qs
din = 0
L s
where d is the interface position vector, q f and q s are the heat fluxes in
the fluid and the solid domains, respectively, s is the solid density, L is
the latent heat per unit volume and n is the normal direction of the
interface.
ADINA-F ADINA-F
fluid element solid element
group group
_
prescribed f
Figure 3.12 A typical phase-change boundary condition
3.4.3.9 Gap condition
The gap is an interface between two fluid domains. This condition can
only be applied to internal lines and surfaces of two-dimensional and three-
dimensional computational domains, respectively.
The gaps status may change due to specified physical conditions,
indicating the connection or disconnection of the two domains. When they

are connected (the gap status is called open), the fluid can flow across the
interface without involving any treatment of boundary conditions. In this
case, the fluid variables are continuous across the interface. When the two
domains are disconnected (the gap status is called closed), the fluid cannot
flow across the interface. The gap condition functions as a no-slip wall
condition to the fluid equations and as a zero heat flux condition to the
temperature equation for both sides of the fluid domains. In this case,
therefore, some solution variables are generally discontinuous (of course,
the velocities are zero and continuous).
A typical application of the gap condition is shown in the next figure.
The enlarged figures indicate the element connections when the gap is open
and closed. When the gap is closed, the elements are disconnected,
representing the disconnection of the two fluid domains. To achieve this
condition, additional nodes are automatically generated at the same
locations as the originally generated nodes (the figure shows here that they
are shifted slightly purely for the purpose of clarification of disconnection).
When the gap is open, the solution variables on these additional nodes are
constrained to their original counterparts. Hence, continuous solution
variables are modeled.
upstream gap is
fluid domain closed if
G<Gclose
downstream
gap is fluid domain
open if
G>Gopen
gap boundary
Figure 3.13 A typical gap boundary condition
The open/close status is controlled by certain criteria. Three physical

conditions are currently available to control the gap status:

1) Gap status is controlled by the gap size G . If the gap is initially

closed, it remains closed until the gap size exceeds a specified
value Gopen . Conversely, if the gap is initially open, it remains
open until the gap size is less than a specified value Gclose . In a
two dimensional domain (including planar and axisymmetric
cases), the size is the length of the interface, while in a three
dimensional domain, the size is the area of the interface. Hence,
the gap condition of this type must always be associated with
some moving mesh boundary conditions.
2) Gap status is controlled by a time function G . If the gap is
initially closed, it remains closed until the value of the time
function value exceeds a specified value Gopen . Conversely, if
the gap is initially open, it remains open until the value of the
time function is less than a specified value Gclose .
3) Gap status is controlled by the averaged pressure difference
G = pu pd , where pu and pd are the pressures upstream and
downstream of the gap, respectively. If the gap is initially
closed, it remains closed until the pressure difference exceeds a
specified open value Gopen . However, once it is open, the gap
cannot be closed. Recall that the solution variables are
continuous across the gap, the pressure difference remains
zero for the rest of solution times.
G gap is open
_
Gopen
_
Gclose
t
gap is closed
Figure 3.14 Diagram of gap open-close conditions

In order to prevent numerical oscillation in changing the gap status, the

open value Gopen must be slightly larger than the close value Gclose , usually
a small percentage of Gclose .
3.4.3.10 Sliding-mesh interface
General description
Sliding-mesh capability is particularly useful for both fluid-flow and

fluid-structure-interaction analyses that involve one or more moving parts.
The sliding mesh boundary condition is designed to allow meshes in
different regions to move relative to each other while the physical variables
across the sliding interface remain continuous. There is a wide variety of
fluid and FSI problems that can be solved using sliding mesh boundary
conditions. Some typical applications include the flow in turbomachinery,
mixing tanks and other devices that have one or more rotating parts as one
example shown below:
sliding mesh
interface stationary
rotating
Figure 3.15 2D Rotor-Stator Interaction
Other problems that do not have rotating parts but translating moving parts
may also require a sliding mesh boundary condition. One example, the two
passing vehicles problem, is depicted in the following figure

sliding mesh
interface
Figure 3.16 Two Passing Trains/Cars
Basic strategy and theory
It is assumed that in general the nodal points on both sides of the sliding
interface are independent of each other. The numerical scheme at the
interface should satisfy the conservation laws of mass, momentum and
energy. For simplicity, the strategy used in a 2-D case depicted in the
following figure is described. The nodal points M, S1 and S2 are sitting at
different sides of the mesh interface. It is required to calculate the
flux/force over the control volume represented by the node M. The control
volume of M is composed of upper and lower sub-volumes. For the lower
half, there is no difficulty to get the flux using the information from node
M. However, for the upper half control volume, there is no direct
information since the mesh is not connected across the interface.
Therefore, the flux for the upper half control volume is calculated using the
neighboring nodes S1 and S2.

control volume of M
(upper half)
sliding mesh
interface S1 S2
M
control volume of M
(lower half)
Figure 3.17 Sliding mesh interface coupling
Setting up a Sliding Mesh Model
The sliding mesh boundary condition needs to be applied on both sides

of the zone interface. Two sets of geometry entities are required at the
interface, having a common spatial location. The geometry entities at the
interface can also have a small gap as shown in the figure
sliding mesh geometry entity 1

boundary
geometry entity 2
Figure 3.18 2-D geometry for sliding mesh interface
A sliding mesh boundary condition must be applied to both sides of the

interface. The boundary condition applied to each side must be different,
that is, they must have different numbers. After applying the sliding mesh
boundary condition, a boundary condition pair has to be defined containing
the sliding mesh boundary conditions previously defined. Since the sliding

mesh interface is only a mesh interface, physical variables are expected to

be continuous across it. Therefore, the physical conditions at the edge of the
sliding mesh interface must also be continuous. In other words, the
physical condition at the edge of the two sides must be the same, as shown
in the figure
sliding mesh condition 1
boundary
condition pair same physical
condition
sliding mesh condition 2
Figure 3.19 Physical conditions at edge of sliding mesh

interface
In principle, the elements at two sides of the interface are independent,

so different mesh subdivisions can be applied on each side. However, to
obtain better convergence, it is recommended to use a similar element size
at the interface. There are several ways to ensure that the nodal points are
separate at the interface. For example, one can use different element
groups for mesh generation and prevent same element group nodal
coincidence checking at the interface.
Limitations for sliding mesh feature in current version
There are some situations where sliding mesh boundary conditions

cannot be used. They are:
It is available for FCBI and FCBI-C elements.

For FCBI elements, it is only associated with the sparse solver
It is not available for high-speed compressible flows.
It cannot be used with the direct FSI coupling method.

3.4.3.11 Boundary friction condition
A general friction model has the form,
2
f = 12 CD ( vie) e
where f is the nodal force per unit of area or length, and CD is a drag
coefficient and e is a direction vector. The nodal force is obtained by
integration
F = h vf idS
where h v is the virtual quantity of the velocity. The drag coefficient is

assumed to be a function of the local Reynolds number, CD = a ( Re L ) ,
b
where a, b user-supplied constants, Re L = v ie L , is the fluid

viscosity, and L is a user-defined length scale. The direction of the
friction force can be set to be the boundary normal, the velocity tangent or a
user-supplied vector. Therefore, depending upon the choice of direction,
the vector e becomes:
1. normal direction: e = n ;
2. velocity tangential direction: e = v i(I nn ) v i(I nn ) ;
3. user-defined vector: e = specified.
3.4.3.12 Vent boundary condition
Vents are regions where a flow resistance is imposed by the presence of

a slot or other type of obstruction. The resistance to the flow is modeled by
adding a force, Fvent , to the right-hand side of the momentum equations,
2
fvent = 12 Closs ( vin) n
Fvent = hv f idS
vent

where is the density of the fluid and Closs is the loss coefficient
associated with the vent, h v is the virtual quantity of the velocity, and n is
the surface normal.
The vent boundary condition is a special case of boundary friction,

where the drag coefficient is a constant.
The user can also specify an arbitrary direction, e , instead of the

boundary normal n . This is for cases where angled slots are present.
3.4.3.13 Fan boundary condition
The objective of the fan boundary condition is to provide a lumped-

parameter model to predict the amount of flow through a fan, given its
characteristic curve. A fan characteristic curve has the form,
ps = f ( Q ) 0
where Q = fan
v idS is the flow rate (L3 / t), ps is the fan static pressure,
ps = pd pi , pd is the discharge pressure, and pi is the inlet pressure.
The fan curve model in ADINA F is assumed to be a polynomial of the

form,
f ( Q ) = C0 + C1Q M1 C2Q M 2
where C0 , C1 , C2 , M 1 and M 2 are user-specified constants.
The fan is modeled as an additional body force in the direction normal

to the fan plane. Such force is proportional to the static pressure or the
pressure rise due to the presence of the fan,
F fan = hv f ( Q ) e dS

The direction e depends on the type of fan,
(1) for intake fan: e = n ;

(2) for exhaust fan: e = n ;
(3) for internal fan: e = specified
3.4.3.14 Multiple reference frames
Many fluid flow problems involve multiple rotating parts. For

examples, rotor/stator interactions, multiple rotors (see figures below), etc.,
those widely exist in turbomachinery flows, mixing tanks and other related
device. ADINA-F/FSI provides the Multiple Reference Frame (MRF)
method for such problems. In the MRF approach, the interactions between
the rotating parts are assumed weak and uniform, and therefore a steady-
state interaction condition can be applied to the interface between different
reference frames.
stationary frame
rotating frame
Figure 3.20 A weak rotor-stator interaction

rotating frames
stationary frame
Figure 3.21 Two rotors in a chamber
With the MRF method, the computational domain is divided into a few
sub-domains, each of them having different reference frames applied while
the meshes are maintained stationary. Different frames are modeled using
corresponding centrifugal forces as used in the single rotating model. The
physical quantities on each interface are properly computed to ensure the
conservations of mass, momentum, energy, etc.
In order to set up a MRF model, one has to first apply centrifugal

loading (related to a specific reference frame) to a certain element group(s).
Secondly, one has to apply a sliding-mesh condition on both interfaces that
are connected (see the following figure). The sub-domains are completely
separated and it is not required to have the interface mesh-nodes matched.
One can also refer to the sliding-mesh condition for more explanations.
Please note, the boundaries separating different reference frames should
have zero normal frame velocity. For most cases, the boundary should be
circular/cylindrical in the 2D/3D domain.

reference frame 1
sliding mesh
boundary
reference frame 2
Figure 3.22 Model setup of a multiple-reference frame
The MRF feature is currently only available for FCBI-C elements.
3.4.3.15 Periodic boundary condition
The periodic boundary condition is for cases where the physical

geometry and the expecting flow pattern have a periodicity in space. Each
condition is specified as a pair in which one boundary is transformed from
the other. For example, as shown in the two figures in this section, the
boundary of condition B is obtained by translating/rotating the boundary of
condition A. Currently the types of transformation can be either translation
or rotation. Meshes of the pair may be different.
This is the procedure for creating a periodic boundary condition: (1)
define periodic boundary condition A, in which the transformation from A
to B is specified; (2) define periodic boundary condition B (in which the
transformation is not required); (3) define a BCD-PAIR that is formed by
A and B.
The periodic boundary condition is only available for FCBI-C elements.

Illustration of translating periodic boundary pair
Illustration of rotating periodic boundary pair
3.4.4 Usual boundary conditions for temperature
3.4.4.1 Prescribed temperature
In this condition, a time-dependent temperature can be directly

prescribed and applied to boundaries:
= (t )

The energy equation at the boundary nodes is then replaced by this

equation.
This condition is usually applied to boundaries where the temperature is
known.
There is a time birth and death option associated with the prescribed
temperature. If only the birth option is used, the prescribed condition is
inactive until the solution time is larger than the birth time. If only the
death option is used, the prescribed condition is active until the time is
larger than the death time. If both the birth and death options are used, the
prescribed condition is only active between the birth time and death time.
Hence, the death time must be larger than the birth time. While the
prescribed condition is inactive, the temperature degree of freedom is free.
3.4.4.2 Zero temperature
Physically, this condition is equivalent to a prescribed zero temperature.

However, the zero temperature condition removes the temperature degree
of freedom from the governing system of discretized equations. There are,
therefore, no time birth and death options for this condition.
3.4.4.3 Concentrated heat flow load
In this condition, a prescribed time-dependent heat flow load Q ( t ) is

directly applied to boundaries. The load is directly added to the right-hand
side of the energy equations of boundary nodes. Note that heat flow loads
can directly be applied to nodes as well.
This condition has no effect if applied to the boundary where a
temperature condition is prescribed, since the energy equation has been
replaced by the prescribed temperature condition.
3.4.4.4 Distributed heat flux load
This is one of the most important boundary conditions for the

temperature solution. This condition can only be applied to boundary lines
and surfaces of two-dimensional and three-dimensional computational

This condition results in concentrated heat flow loads by specifying the

normal component of the external heat flux qn ( t ) . The applied heat flux is
then integrated to a heat load
Q ( t ) = h qn ( t ) dS
and added to the right-hand side of the energy equations as in the case of
concentrated heat flow loads. Here h is the virtual quantity of
temperature on the boundary. Note that when qn ( t ) = 0 , the application of
the heat flux is trivial since it is equivalent to no heat flux loads.
This condition is usually applied to boundaries where the heat flux is
known. The adiabatic boundary has a zero heat flux condition.
This condition has no effect if applied to the boundary where a
temperature condition is prescribed since the energy equation has been
replaced by the prescribed temperature condition.
3.4.5 Special boundary conditions for temperature
3.4.5.1 Convection
The heat flux from the external convection condition is defined as
qn = h ( e )
where h is a heat convection coefficient and e is an environmental

temperature, both can be constant, time-dependent or temperature-
dependent. The applied heat flow is then computed by
Q ( t ) = h qn dS (3.5)
where h is the virtual quantity of temperature on the boundary. Noting

that the temperature is the unknown solution variable, it has therefore
contributed terms to both the right-hand side and the coefficient matrix.
Convection condition is usually applied to a boundary that is exposed to
an open environment, typically as shown in the following figure. This

condition can only be applied to boundary lines and surfaces of two-

ADINA-F ADINA-F environment

group group convection
boundary
condition
Figure 3.23 A typical convection boundary condition
3.4.5.2 Radiation
In this condition, the heat flux from an external radiation source is

specified as
qn = f e ( Re Rt )
where Re is the energy from the radiation source and Rt is the energy
transmitted into the environment
Re = (c1 e + c2 ) 4 , Rt = (c1 + c2 ) 4 (3.6)
in which, is the Stefan-Boltzmann constant, f is the shape factor, e is

the emissivity, e is the temperature of the radiative source (or sink) , and
c1 and c2 are two constants such that c1 + c2 has an absolute temperature
scale. e and e can be constant, time-dependent or temperature-
dependent. The conventional temperature unit is Kelvin in radiation, which
corresponds to c1 = 1 and c2 = 0 if the temperature variable has the unit
Kelvin. Some frequently used temperature units, when transformed to
Kelvin or Rankine, are shown in the following table:

Table 3-2 Transformations of temperature units to absolute

temperature units

c1 + c2 K o
R o
C o
F

c1 = 1 c1 = 5 9 c1 = 1 c1 = 5 9
K
c2 = 0 c2 = 0 c2 = 273.15 c2 = 255.37
c1 = 9 5 c1 = 1 c1 = 9 5 c1 = 1
o
R
c2 = 0 c2 = 0 c2 = 491.67 c2 = 459.67
The heat flow is computed using Eq.(3.5) and discretized and

assembled into the global matrix system.
The radiation condition is applied to the boundary that is exposed to a
radiation source or sink. This condition can only be applied to boundary
lines and surfaces of two-dimensional and three-dimensional computational
3.4.5.3 Specular radiation
In some problems, certain solid boundaries have a high reflectivity in

responding to radiative energies. Also, they may have certain percentages
of energy emitted, diffused and transmitted. The specular radiation
boundary condition is designed for these problems.
The following figure shows a portion of a surface surrounded by a fluid
medium. The surface is assumed gray, diffused and specularly reflective,
and may or may not be transparent. Consider an incident unit ray of
energy, which is decomposed into four portions:
a diffusely reflected portion d ;

a transmitted portion t ;
an absorbed portion which, by Kirchhoffs law, could be equal to the
emittance e;
a specularly reflected portion s .

_
diffusely reflected energy, d
incident unit
of energy specularly
_ reflected
energy, s
i r
absorbed energy, e solid
transmitted
_
energy, t
Figure 3.24 Schematic representation of radiation energies

corresponding to an incident ray
The sum of the four portions satisfies
e + s + d + t =1 (3.7)
The parameters, s , d and t , required in the input of this condition, can

be constant, time-dependent or temperature-dependent. The parameter e is
computed from Eq.(3.7). The transmitted energy could be lost to the
environment and not traced in the solution, or could arrive at the next
specular boundary and be reflected again. In addition, the specularly
reflected energy is calculated in the solution algorithm by tracing the
reflected ray of energy. The incident ray angle i and the reflected ray
angle r are assumed equal.
Let R be the rate of outgoing radiant energy per unit area. It consists
of the absorbed or emitted energy from the environment, transmitted energy
and diffusely reflected energy,
R = eRe + t f Rt + dG (3.8)
where Re and Rt are defined in Eq.(3.6), G is the rate of all incoming

radiant energies per unit area, is the Stefan-Boltzmann constant, f is

the shape factor, e is the temperature of the radiative source (or sink) , and
c1 and c2 are two constants such that c1 + c2 has an absolute temperature
scale (see previous section for details).
The rate of incoming radiant energy per unit area at any specular
boundary element k is the sum of all rates of outgoing radiant energies
from all specular radiation boundary elements
Gk = Flk Rl dS (3.9)
Sl
l
where Sl is the area of the boundary element l and Flk are the view factor
matrices which are calculated using a ray tracing technique based on
Lamberts law. It can be explained as the ratio of the energy that is emitted
from element l and arrives at element k , directly and indirectly, to the
energy emitted by point l . It is clear that the view factor matrix has a
dimension N es N es , where N es is the number of specular boundary
elements.
Using Eq.(3.9) and applying the Galerkin method to Eq.(3.8), we obtain
the finite element equation governing R
S
( )
h R eRe t f Rt dG dS = 0 (3.10)
where h is the virtual quantity of temperature on the boundary. This

equation, called a radiosity equation, is linear and governs N ns unknown
variables R defined at specular boundary nodes.
In certain cases, not all specular boundary elements need to be coupled.
For example, a domain is divided into two chambers that are separated by a
solid wall as illustrated in the figure below. The specular boundaries in
different chambers cannot see each other since no rays can penetrate
through the solid wall (unless the ray is specified as being continuously
transmitted from one side of the wall to the other side). This indicates that
the view factor matrix has the form
F 0
F= 1
0 F2

with the two block diagonal matrices corresponding to the two chambers,
respectively. The computation can then be carried out separately for the
chambers. Therefore, we allow the specular boundary conditions to be
divided into groups. Only specular boundary conditions within the same
group specularly transmit energies to each other. Of course, energy is also
transmitted by means of conduction and convection between the chambers.
specular specular
group 1 group 2
convection convection
conduction
Figure 3.25 A typical usage of specular groups
With the variables R solved from Eq.(3.10) and G computed from

Eq.(3.9), we can finally define the heat flux load that consists of absorbed
incoming radiant energy and lost transmitted energy
qn = e ( G Rt )
and the heat flow is then computed using Eq.(3.5).

It is clear then the application of the specular radiation boundary means
much more than just specifying a regular boundary condition. Computing
the heat flow load requires the solution of the variables R from a large
linear system, which can be costly. Computational issues will be addressed
in Section 13.5.
This condition can only be applied to lines and surfaces of two-
These lines/surfaces must either be on the boundary or be interfaces
between (ADINA-F) solid and fluid element groups.

3.4.6 Thermal resistance condition
In certain models where there is a very thin layer that has different heat
conduction, this condition can be used to model the heat transfer through
that layer. In this condition, a parameter R called thermal resistance must
be specified. It is defined as

R=
k
where k and are , respectively, the heat conductivity and thickness of

the thin layer. A local analytical solution will then be applied to that layer.
Thermal resistance condition can be applied along lines (in 2D models)
or along surfaces (in 3D models). These lines/surfaces can be located at
boundaries or inside the computational domain.
3.4.7 User-supplied boundary condition
This is a specially defined user-supplied boundary condition. Similar to

user-supplied material data, the boundary condition is programmed in
Fortran by the user in a specially designed subroutine. This subroutine is
invoked whenever the boundary condition is required.
In this subroutine, the solution variable identification number, the
coordinates of the current spatial point, the element group, the element and
the independent solutions, mass-ratios, etc., are provided as input
arguments. In return, the subroutine will output the discretized boundary
condition. This condition is represented in the form:
S = A + B i
which is treated as a source of the corresponding equation from the

boundary. Here, A is a scalar coefficient and B is a coefficient vector and
is the solution vector defined at the boundary node. The solution vector
is defined as (based on the identification number I var denoted as IVARID
in the subroutine)

u , v, w, p, if I var = 1

= K , if I var = 2
, ,... if I var = 3
1 2
The values of A and B will be assembled into the right-hand side of the
discretized equation and the global matrix respectively as follows.
Corresponding to the current node, let ik be the equation number of the
current variable, i1 , i2 ,... be the equation numbers of the solution variables
, and Aij X j = Bi be the original equation system. The assemblage
procedure in the Newton-Raphson method is
Aik i j B j Aik i j j = 1, 2,...

Bik + A + B j j Bik
As an example, if an external force, F (t ) = a + bp , is added to the y-

momentum equation at boundary nodes where the user-supplied boundary
condition l is applied, the coding would have the following structure in the
Newton-Raphson iteration:
for boundary condition l

for var iable identification I var = 1
for var iable v { A = a + bp, B = 0}
In cases where can be no confusion in the specified user-supplied

boundary conditions, the condition checking can be omitted. For example,
if only one condition is specified, in which only the force is specified, the
above code can be simplified to
A = a + bp, B = 0
It can be seen that advanced users can achieve much more with this
capability. They can invoke their own programs to compute, possibly in a

complicated way, the value of the force and split it in any way they prefer.
A different version of the program would be
i compute a and b by any means possible

i A = a + bp
b if i = k
i Bi =
0 else
where, k is the index of the variable corresponding to the y-velocity (in

this case, k = 2 ). This version would be more stable in general cases.
As another example, consider a pressure-dependent temperature is
prescribed, ( p ) = a + bp + cp 2 , at boundary nodes where the user-supplied
boundary condition l is applied. Since d dp = b + 2cp , the coding
would have the following structure:
i compute a, b and c by any means possible

i A = 1021 (a + bp + cp 2 )
b + 2cp if i = 4 (coefficient of p)

i Bi = 1021 1 if i = 5 (coefficient of )

0 else
Where the large number 1021 is the pivot number added to ensure the
specified condition is so dominant that the original terms in that equation
become negligible.
The following is a sample example for 3 different user-supplied

boundary conditions for mass ratios:
(1) Boundary condition label 1: prescribed (1 ,2 ) = (0.2,0.1)

(2) Boundary condition label 2: prescribed (1 ,2 ) = (0.1,0.2)
(3) Boundary condition label 4: apply the load that is defined as

qi = 0.002M i j + i
j i
Here, the mass-ratio has been assumed to be the mass density i , so that
Ci i M i
i = =
C j M j
j
SUBROUTINE ADFUBC(L,NL,R,NR,VAR,NVAR,A,B)
C======================================================================
C
C
C USER-SUPPLIED BOUNDARY CONDITIONS
C
C THIS ROUTINE WILL BE CALLED AT EACH INTEGRATION POINT OF EVERY
C BOUNDARY ELEMENT OF BOUNDARY CONDITION LABEL "IBCL" AT EACH
C NONLINEAR EQUILIBRIUM ITERATION AT EACH TIME STEP.
C
C HERE WE EXPRESS THE BOUNDARY CONDITION AS A SOURCE S, IMPOSED
C INTO THE COMPUTATIONAL DOMAIN.
C
C -----------------------------------------------------------------
C
C PASSED IN: L(NL),R(NR),VAR(NVAR)
C
C PASSED OUT: A,B(NVAR)
C
C -----------------------------------------------------------------
C WHERE
C
C L( 1) = IBCL = BOUNDARY CONDITION LABEL (NAME IN THE
C BOUNDARY-CONDITION COMMAND)
C L( 2) = IVARID = ID OF THE VARIABLE (=1,2 AND 3 FOR FLUID,
C TURBULENCE AND MASS (VARIABLE IN THE
C BOUNDARY-CONDITION COMMAND)
C L( 3) = ICOMID = ID OF THE CURRENT EQUATION:
C = (1,2,3,4,5) FOR (U,V,W,P,T ), IF IVARID=1
C = (1,2 ) FOR (K,E/W) , IF IVARID=2
C = (1,2,... ) FOR (S_1,S_2,...), IF IVARID=3
C L( 4) = METHOD = ID OF THE ITERATION METHOD:
C = 0, IF SUCCESIVE SUBSTITUTION METHOD
C = 1, IF NEWTON METHOD
C L( 5) = IELG = ELEMENT GROUP NUMBER
C L( 6) = IELM = ELEMENT NUMBER
C L( 7) = IU = ENTRY OF THE X-VELOCITY IN VAR(*)
C L( 8) = IV = ENTRY OF THE Y-VELOCITY IN VAR(*)
C L( 9) = IW = ENTRY OF THE Z-VELOCITY IN VAR(*)
C L(10) = IP = ENTRY OF THE PRESSURE IN VAR(*)
C L(11) = IT = ENTRY OF THE TEMPERATURE IN VAR(*)
C L(12) = IK = ENTRY OF THE TURBULENCE-K IN VAR(*)
C L(13) = IE = ENTRY OF THE TURBULENCE-E/W IN VAR(*)
C L(14) = IX = ENTRY OF THE X-COORDINATE IN VAR(*)
C L(15) = IY = ENTRY OF THE Y-COORDINATE IN VAR(*)

C L(16) = IZ = ENTRY OF THE Z-COORDINATE IN VAR(*)

C L(17) = IM = ENTRY OF THE MASS SPECIES IN VAR(*)
C L(18) = IQ = ENTRY OF THE CURRENT VARIABLE IN VAR(*)
C NOTE THAT IQ DEPENDS ON IVARID AND ICOMID:
C IQ = IU IF(IVARID=1 AND ICOMID=1)
C = IV IF(IVARID=1 AND ICOMID=2)
C = IW IF(IVARID=1 AND ICOMID=3)
C = IP IF(IVARID=1 AND ICOMID=4)
C = IT IF(IVARID=1 AND ICOMID=5)
C = IK IF(IVARID=2 AND ICOMID=1)
C = IE IF(IVARID=2 AND ICOMID=2)
C = IM+ICOMID, IF(IVARID=3)
C L(20) = JX = ENTRY OF THE X-OUT-NORMAL IN R(*)
C L(21) = JY = ENTRY OF THE Y-OUT-NORMAL IN R(*)
C L(22) = JZ = ENTRY OF THE Z-OUT-NORMAL IN R(*)
C
C VAR(IU) = U = X-VELOCITY
C VAR(IV) = V = Y-VELOCITY
C VAR(IW) = W = Z-VELOCITY
C VAR(IP) = P = PRESSURE
C VAR(IT) = T = TEMPERATURE
C VAR(IK) = TK = TURBULENCE_K
C VAR(IE) = TE = TURBULENCE_E OR TURBULENCE-W
C VAR(IX) = X = X-COORDINATE
C VAR(IY) = Y = Y-COORDINATE
C VAR(IZ) = Z = Z-COORDINATE
C VAR(IM) = S_1= MASS-RATIO_0 (FLUID ITSELF)
C VAR(IM+1) = S_2= MASS-RATIO_1
C ......
C VAR(IQ) = VALUE OF THE CURRENT VARIABLE
C
C R(JX) = SX = X-OUT-NORMAL OF THE BOUNDARY
C R(JY) = SY = Y-OUT-NORMAL OF THE BOUNDARY
C R(JZ) = SZ = Z-OUT-NORMAL OF THE BOUNDARY
C
C -----------------------------------------------------------------
C THE DEFINITION OF THE PASSED OUT VARIABLES A AND B(*) DEPENDS ON
C THE FLAGS "IVARID", "METHOD" AND THE FLOW TYPE (INCOMPRESSIBLE
C OR COMPRESSIBLE).
C
C +------------------------------------------------------+
C | |
C | CASE: FLUID OR TURBULENCE (IVARID<=2) |
C | SUCCESSIVE SUBSTITUTION METHOD (METHOD=0 ) |
C | INCOMPRESSIBLE FLOWS |
C | |
C +------------------------------------------------------+
C
C THE SOURCE S IS SPLIT INTO THE FORM
C
C S = A + B(IU)*U + B(IV)*V + B(IW)*W
C + B(IP)*P + B(IT)*T
C + B(IK)*TK + B(IE)*TE
C
C CAUTION: THERE ARE MANY WAYS TO SPLIT S. SOME OF
C THEM MAY CAUSE NUMERICAL INSTABILITY. IT
C IS RECOMMENDED TO ALWAYS KEEP B(IQ)<=0.
C
C +------------------------------------------------------+
C | |
C | NEWTON METHOD (METHOD=1 ) |
C | INCOMPRESSIBLE FLOWS |

C | |
C +------------------------------------------------------+
C
C A = THE SOURCE S
C B(IU) = DERIVATIVE OF S WITH RESPECT TO U
C B(IV) = DERIVATIVE OF S WITH RESPECT TO V
C B(IW) = DERIVATIVE OF S WITH RESPECT TO W
C B(IP) = DERIVATIVE OF S WITH RESPECT TO P
C B(IT) = DERIVATIVE OF S WITH RESPECT TO T
C B(IK) = DERIVATIVE OF S WITH RESPECT TO TK
C B(IE) = DERIVATIVE OF S WITH RESPECT TO TE
C
C +------------------------------------------------------+
C | |
C | SUCCESSIVE SUBSTITUTION METHOD (METHOD=0 ) |
C | COMPRESSIBLE FLOWS |
C | |
C +------------------------------------------------------+
C
C
C S = A + B(IU)*DU + B(IV)*DV + B(IW)*DW
C + B(IP)*D + B(IT)*E
C + B(IK)*TK + B(IE)*TE
C
C WHERE, D=DENSITY, E=TOTAL ENERGY, DU=D*U, DV=D*V
C AND DW=D*W.
C
C
C +------------------------------------------------------+
C | |
C | NEWTON METHOD (METHOD=1 ) |
C | COMPRESSIBLE FLOWS |
C | |
C +------------------------------------------------------+
C
C A = THE SOURCE S
C B(IU) = DERIVATIVE OF S WITH RESPECT TO DU
C B(IV) = DERIVATIVE OF S WITH RESPECT TO DV
C B(IW) = DERIVATIVE OF S WITH RESPECT TO DW
C B(IP) = DERIVATIVE OF S WITH RESPECT TO D
C B(IT) = DERIVATIVE OF S WITH RESPECT TO E
C B(IK) = DERIVATIVE OF S WITH RESPECT TO TK
C B(IE) = DERIVATIVE OF S WITH RESPECT TO TE
C
C WHERE, D=DENSITY, E=TOTAL ENERGY, DU=D*U, DV=D*V
C AND DW=D*W.
C
C +------------------------------------------------------+
C | |
C | CASE: MASS TRANSFERS (IVARID=3) |
C | SUCCESSIVE SUBSTITUTION AND NEWTON METHODS |
C | INCOMPRESSIBLE AND COMPRESSIBLE FLOWS |
C | |
C +------------------------------------------------------+
C
C

C S = A + B(IQ)*VAR(IQ)
C
C
C======================================================================
IMPLICIT DOUBLE PRECISION (A-H,O-Z)
DIMENSION L(NL),R(NR),VAR(NVAR),B(NVAR)
C----------------------------------------------------------------------
C START TO CODE HERE
C----------------------------------------------------------------------
PARAMETER (NMASS=2)
DIMENSION XM(NMASS),X(NMASS)
C-----------------------------------------------------------------------
R(1) = R(1)
C-----------------------------------------------------------------------
IBCL = L( 1)
IVARID = L( 2)
ICOMID = L( 3)
C METHOD = L( 4)
IM = L(17)
IQ = L(18)
UNDER = 1.D-20
C
C======================================================================
C
IF(IVARID.EQ.1) THEN
C
C +-------------------------------------------------------------+
C | |
C | FLUID BOUNDARY CONDITIONS ARE CODED HERE (U,V,W,P,T) |
C | |
C +-------------------------------------------------------------+
C
C
C +----------------------------------------------------------+
C | END OF FLUID BOUNDARY CONDITIONS HERE |
C +----------------------------------------------------------+
C
GO TO 9999
ENDIF
C
C======================================================================
C
C
C +-------------------------------------------------------------+
C | |
C | TURBULENCE BOUNDARY CONDITIONS ARE CODED HERE (K,E/W) |
C | |
C +-------------------------------------------------------------+
C
C
C +----------------------------------------------------------+
C | END OF RTURBULENCE BOUNDARY CONDITIONS HERE |
C +----------------------------------------------------------+
C
GO TO 9999
ENDIF
C
C======================================================================
C

C
C +-------------------------------------------------------------+
C | |
C | MASS TRANSFER BOUNDARY CONDITIONS ARE CODED HERE |
C | |
C | THE FOLLOWING EXAMPLE IS FOR 3 BOUNDARY CONDITIONS: |
C | |
C | IBCL = 1: SPECIFY THE MASS RATIO 0.2 |
C | IBCL = 2: SPECIFY THE MASS RATIO 0.1 |
C | IBCL = 4: CVD CONDITION |
C | |
C | NMASS = NUMBER OF REACTING SPECIES (=2 HERE) |
C | XM(*) = MOLECULAR WEIGHT |
C | X(*) = MOLE FRACTION |
C | XNU = STOICHIOMETRIC COEFFICIENT |
C | XKA = KENETIC REACTION RATE CONSTANT |
C | NOTE: THE LARGE NUMBER 1.D+20 USED HERE IS TO FORCE THE |
C | SPECIFIED VALUE TO BE RESPECTED IN THE FINAL EQUATION|
C | |
C +-------------------------------------------------------------+
C
IF(IBCL.EQ.1) THEN
IF(ICOMID.EQ.1) THEN
SS = 0.2D0
ELSE
SS = 0.1D0
ENDIF
B(IQ) = - 1.D+20
A = 1.D+20*SS
GO TO 9999
ENDIF
C
IF(IBCL.EQ.2) THEN
IF(ICOMID.EQ.1) THEN
SS = 0.1D0
ELSE
SS = 0.2D0
ENDIF
B(IQ) = - 1.D+20
A = 1.D+20*SS
GO TO 9999
ENDIF
C
IF(IBCL.EQ.4) THEN
XM(1) = 2.D0
XM(2) = 1.D0
XNU = 1.0D0
XKA = 0.002D0
C
C COMPUTE X(*)
C
SUM = 0.D0
DO 1 I = 1, NMASS
X(I) = VAR(IM+I)/XM(I)
SUM = SUM + X(I)
1 CONTINUE
SUM = MAX(SUM,UNDER)
DO 2 I = 1, NMASS
X(I) = X(I)/SUM
2 CONTINUE
C
C COMPUTE A AND B

C
XX = 1.D0
DO 3 I = 1, NMASS
IF(I.EQ.ICOMID) GO TO 3
XX = XX*X(I)
3 CONTINUE
FAC = XNU*XKA
B(IQ) = - FAC/SUM
A = - FAC*XM(ICOMID)*XX
GO TO 9999
ENDIF
C
C +----------------------------------------------------------+
C | END OF MASS BOUNDARY CONDITIONS HERE |
C +----------------------------------------------------------+
C
ENDIF
C----------------------------------------------------------------------
9999 CONTINUE
RETURN
END
3.4.8 Shell-thermal condition
In certain models where a thin shell structure is present, this condition

can be used to model the heat transfer through that layer. The shell
deformation may be neglected or computed (as an TFSI problem). This
condition is classified into three types, namely boundary-convection, heat-
flux and internal-convection. It can be applied to model shell structure
inside fluid domain or on fluid boundary.
The boundary-convection condition is defined as
qb = k s d s ( e )
where, k s is shell thermal conduction, d s is shell thickness and e is the

environmental temperature. When applied to a boundary of a pure fluid
problem, this condition is equivalent to a convection condition (observing
that hs = k s d s is convection coefficient). When this condition is applied
to a fluid-shell interface in TFSI problems, the thermal results are
automatically used in determining the shell deformation.
The heat-flux condition is the same as the boundary-convection

condition, except that the heat-flux qb = qb is directly specified.

The internal-convection condition is to link heat transfers between two

fluid domains separated by a thin shell. For convenience of explanation,
we will refer to the two domains as left and right respectively. The
condition that is applied to the left fluid boundary is defined as
qi = k s d s ( R L )
where, L and R are fluid boundary temperatures on the left and right of
the shell respectively. Similarly, qi is applied to the fluid boundary on
the right of the shell. We note that the left and right fluid boundaries must
be separated (different nodes, elements, etc.), but share the same geometric
location (where the shell is). The shell deformation may or may not be
modeled. If the shell is modeled, the thermal results are automatically used
in determining the shell deformation. For more information on modeling
TFSI problems, refer to the fluid-structure interaction chapter.
3.5 Initial conditions
In transient analyses, all solution variables must be specified. The

default initial conditions to all variables are zero values. Although initial
conditions are not required in steady-state analyses, they are used as a
guessed solution at the start of the equilibrium iterations. A good initial
condition, as will be seen in Chapter 13, may not only accelerate the
convergence during equilibrium iterations, but also be a key factor in
obtaining solutions in certain cases.
In low-speed compressible flows, recalling that pressure, temperature
and density must satisfy the state equation of the ideal gas law, zero
pressure or zero temperature must not be specified. Since the default initial
conditions to all variables are zeros, initial conditions are always required
in both transient and steady-state analyses in compressible flow models.
Initial conditions can be improved in restart analyses or using the
solutions mapped from a similar solution.

3.6 Material models
3.6.1 Constant material model
This is the simplest yet most frequently used material model. In this
model, all fluid properties are assumed to be constant. They are
= fluid density
= fluid viscosity
g = gravitational acceleration vector
Cp = specific heat at constant pressure
Cv = specific heat at constant volume
k = thermal conductivity coefficient
qB = rate of heat generated per unit volume
= thermal expansion coefficient
0 = reference temperature in buoyancy force
= coefficient of surface tension
= bulk modulus of elasticity
The default values of these parameters are all zeros except the that is
20
10 . Note that not all these parameters are required for a given problem.
The parameters that must be input depend on the problem to be solved.
In incompressible flows, the bulk modulus of elasticity and the
specific heat at constant volume Cv are not required. In this case, is
assumed to be infinite and Cv is forced to be equal to C p . If heat transfer
is not considered, C p , k , q B , , and 0 can be further ignored.
In slightly compressible flows, Cv is forced to be equal to C p and,
therefore, can be ignored in the input. C p , k , q B , and 0 can also be
omitted if heat transfer is not included.
In low-speed compressible flows, is not used and therefore is
ignored. The density is determined through the state equation as a function
of pressure and temperature and thus not required. All other material data
must be input.

3.6: Material models
The above notation will be used throughout the chapter in other

material models.
3.6.2 Time-dependent model
This model contains the same parameters as those in the constant

model. However, the following parameters in this model depend on time
, C p , Cv , k , , q B , , = f (t )
Besides its application in truly time-dependent materials, this type of

material is also very useful in dealing with complex flows with constant
properties. For example, with a time-dependent viscosity of decreasing-
value in a steady-state analysis, the solution obtained at an earlier time step
is the solution of a lower Reynolds number. It provides automatically a
good initial condition for the next time step, which is a problem of a higher
Reynolds number.
3.6.3 Power-law model
The viscosity of the power-law model is defined as
max { AD n , 0 } if n 0

=
min { AD , 0 } if n < 0
n
where A, 0 and n are constants and D is the effective deformation rate.

All other parameters are the same as those in the constant model.
The power-law model is a typical non-Newtonian fluid model. The
definitions of A and a can be determined from the fundamental
constitutive relation, in terms of shear stress and deformation rate,
= D = AD n +1
The parameter 0 serves as a lower bound (if n is non-negative) or an

upper bound (if n is negative) of the viscosity.

3.6.4 Carreau model
This model is an extension of the power-law model. It sets both upper

and lower bounds of the viscosity as defined in
= + ( 0 ) (1 + AD 2 )
n
where A, 0 and n are constants and D is the effective deformation rate.

Similar to the power-law model, the Carreau model is also a non-
Newtonian fluid model.
3.6.5 Temperature-dependent model
This model contains the same parameters as those in the constant

material condition model. However, the following parameters in this
model depend on temperature
, C p , Cv , k , , q B , , = f ( )
This material model can only be applied to problems with heat transfer.
3.6.6 Temperature-dependent power-law model
This model is another extension of the power-law model. It includes

the thermal effect in the viscosity as defined in
max { AD n eC , 0 } if n 0

=
min { AD e , 0 }
n C
if n < 0
where A, 0 and C are constants and D is the effective deformation rate.

Similar to the power-law model, this model is also a non-Newtonian
fluid model. However, it can only be applied to problems with heat
transfer.

3.6.7 Second order model
This model is an extension of the temperature-dependent power-law

model. In this model, the viscosity is defined as

=
{
max AD n eC + E 2 ,
0 } if n 0
{ 2
min AD n eC + E , 0
} if n < 0
with
n = F + G ln D + H
where A, 0 , C , E , F , G and H are constants and D is the effective

deformation rate. All other parameters are the same as those in the constant
model.
Similar to the temperature-dependent power-law model, this model is
also a non-Newtonian fluid model and can only be applied to problems
with heat transfer.
3.6.8 Large-eddy-simulation model
The large-eddy-simulation (LES) model is a zeroth order turbulence

model. Since it has no additional solution variables as introduced in
the K - model, it is still treated as a laminar material model. The
viscosity and the heat conductivity of the LES model are defined as
= 0 + t 0 + 2 k D2 2 D
t C p
k = k 0 + kt k 0 +
Pr
where D is the effective deformation rate, k D is a dimensionless constant,

is the element size and Pr is the turbulent Prandtl number. Besides the
data defined in the constant material model, the constants k D and Pr are
required.

Smagorinsky-Lilly Subgrid-Scale model
The effective viscosity and the heat conductivity are defined as,
= 0 + t = 0 + 2 2s D
t C p
k = k 0 + kt = k 0 +
Pr
s = min (d , k DV 1/ 3
)
where s is the mixing length for sub-grid scales, is the Von Karman
constant (default is = 0.41 ), d is the distance to the closest wall and V is
the element volume.
The Smagorinsky-Lilly model can be used in conjunction with wall

functions for near-wall modeling.
RNG-based Subgrid-Scale Model
The RNG-based effective subgrid viscosity is,
1/ 3
= 0 + t 0 [1 + H ( x) ]
where
x, x > 0
H ( x) =
0, x 0
x = s2 02 100
s = 2 D ( CRNG )
2
in which, CRNG = 0.09348 and is the element size. The heat

conductivity is defined in the same way as in the Smagorinsky-Lilly model.

The RNG-based Subgrid-Scale model is only available for FCBI-C

elements.
When using the Smagorinsky-Lilly or the RNG-based subgrid-scale

models, it is possible to superimpose time-dependent random fluctuations
to the mean velocity value at the inlets.
3.6.9 User-supplied materials
In this model, the user can specify the material data through a user-
supplied subroutine written in Fortran. It is designed only for
incompressible, slightly compressible and low-speed compressible flows.
The subroutine is invoked whenever material data are required in the
computation. All necessary data is passed in as arguments, and the
program returns the required material data. The arguments are the
coordinates of the current spatial point, the time, the element group and
element labels, current solution variables and their derivatives at the point,
etc. The returned data are the same as in the constant model.
Using this model, the user can define many material data, for example,
the mixing-length zeroth order turbulence model, where the viscosity is
computed as
= 0 + 2 l 2 D
with 0 the laminar viscosity, the density, l the mixing length and D
the effective deformation rate.
As another example, the user can invoke their own subroutines or
libraries to compute the material data, possibly in a very complicated way,
say solving some partial differential equations.
The following sample example defines the material data for four
different problems. In order to shorten the descriptions, we define a
piecewise linear function as shown in the following figure.

f (x | x1,x2 ,y1 ,y2 )
y1
x1 x2 x
Figure 3.26 A piecewise linear function
(1) For the problem with the heading CATERP: The material is
dependent of the mass-ratio 1 . The original fluid is assumed to be air.
The mass-ratio is used in simulating the ratio of the oil and the mixture,
so that the fluid medium varies as air, air-oil mixture and oil, depending
on the value of the mass-ratio.
(
= f 1 0, 1, 2.082 105 , 3.56 102 )
= f (1 0, 1, 0.995, 853.9 )
C p = f (1 0, 1, 1009, 2118 )
k = f (1 0, 1, 0.0304, 0.138 )
, 0 , g1 , g 2 , g3 , , , Cv = 0
(2) For the problem with the heading F7518, the material is of varying
density, which is assumed to be a function of temperature:
= f ( 100, 1000, 0.086, 0.053)

= 8 106
C p = 14030
k = 0.156
, 0 , g1 , g 2 , g3 , , , Cv = 0

(3) For the problem with the heading F7519, the material density is
assumed to be a function of the z-coordinate:
2
=1+ z
5
and
= f ( z 0, Z , 1, 1 + 52 Z ) ( Z = const. > z )
= 0.2
C p , k , , 0 , g1 , g 2 , g3 , , , Cv = 0
(4) For the problem with the heading F74007, the viscosity and
density are defined as a function of the mass-ratio 1 :
= f (1 0.1, 0.9, 0.1, 0.01)

= f (1 0.1, 0.9, 10, 1)
C p , k , , 0 , g1 , g 2 , g3 , , , Cv = 0
SUBROUTINE ADFUSR(CHEAD,L,NL,R,NR,CMAT,NCMAT)
C=======================================================================
C
C
C USER-SUPPLIED MATERIAL DATA FOR INCOMPRESSIBLE FLOWS WITHOUT
C TURBULENT-KE OR TURBULENT-KW MODELS
C
C THIS ROUTINE WILL BE CALLED AT EACH INTEGRATION POINT OF EVERY
C ELEMENT OF EVERY ELEMENT GROUP WHERE THE MATERIAL TYPE
C "TURBULENT-MIXL" WAS SPECIFIED, AT EACH NONLINEAR EQUILIBRIUM
C ITERATION AT EACH TIME STEP.
C
C IN ORDER TO ACCESS THIS ROUTINE, USER MUST CHOOSE THE MATERIAL
C TYPE "TURBULENT-MIXL".
C
C USER MUST SPECIFY ALL MATERIAL DATA EXPLAINED HERE INTO THE ARRAY
C CMAT(NCMAT) IN THIS ROUTINE. DEFINE ZERO VALUES IF THEY ARE
C NOT USED.
C
C WARNING: USER IS SUPPOSED TO BE AWARE OF THE AVAILABILITY OF THE
C DATA STORED IN THE ARRAYS L(*) AND R(*) FOR THE
C PROBLEM BEING SOLVED. IF NOT AVAILABLE, ARRAYS COULD
C BE RANDOM.
C
C -----------------------------------------------------------------
C
C PASSED IN: CHEAD, L(NL),R(NR)

C
C PASSED OUT: CMAT(NCMAT)
C
C -----------------------------------------------------------------
C WHERE
C
C CHEAD = THE STRING OF THE PROBLEM HEADING
C
C L( 5) = IELG = ELEMENT GROUP NUMBER
C L( 6) = IELM = ELEMENT NUMBER
C L( 7) = IU = ENTRY OF THE X-VELOCITY IN R(*)
C L( 8) = IV = ENTRY OF THE Y-VELOCITY IN R(*)
C L( 9) = IW = ENTRY OF THE Z-VELOCITY IN R(*)
C L(10) = IP = ENTRY OF THE PRESSURE IN R(*)
C L(11) = IT = ENTRY OF THE TEMPERATURE IN R(*)
C L(12) = IK = ENTRY OF THE TURBULENCE-K IN R(*)
C L(13) = IE = ENTRY OF THE TURBULENCE-E/W IN R(*)
C L(14) = IX = ENTRY OF THE X-COORDINATE IN R(*)
C L(15) = IY = ENTRY OF THE Y-COORDINATE IN R(*)
C L(16) = IZ = ENTRY OF THE Z-COORDINATE IN R(*)
C L(19) = ITIM = ENTRY OF THE TIME IN R(*)
C L(20) = IM = ENTRY OF THE MASS-RATIOS IN R(*)
C
C R(IU) = U = X-VELOCITY
C R(IU+1) = DXU = DERIVATIVE OF U WITH RESPECT TO X
C R(IU+2) = DYU = DERIVATIVE OF U WITH RESPECT TO Y
C R(IU+3) = DZU = DERIVATIVE OF U WITH RESPECT TO Z
C R(IV) = V = Y-VELOCITY
C R(IV+1) = DXV = DERIVATIVE OF V WITH RESPECT TO X
C R(IV+2) = DYV = DERIVATIVE OF V WITH RESPECT TO Y
C R(IV+3) = DZV = DERIVATIVE OF V WITH RESPECT TO Z
C R(IW) = W = Z-VELOCITY
C R(IW+1) = DXW = DERIVATIVE OF W WITH RESPECT TO X
C R(IW+2) = DYW = DERIVATIVE OF W WITH RESPECT TO Y
C R(IW+3) = DZW = DERIVATIVE OF W WITH RESPECT TO Z
C R(IP) = P = PRESSURE
C R(IP+1) = DXP = DERIVATIVE OF P WITH RESPECT TO X
C R(IP+2) = DYP = DERIVATIVE OF P WITH RESPECT TO Y
C R(IP+3) = DZP = DERIVATIVE OF P WITH RESPECT TO Z
C R(IT) = T = TEMPERATURE
C R(IT+1) = DXT = DERIVATIVE OF T WITH RESPECT TO X
C R(IT+2) = DYT = DERIVATIVE OF T WITH RESPECT TO Y
C R(IT+3) = DZT = DERIVATIVE OF T WITH RESPECT TO Z
C R(IK) = TK = TURBULENCE_K
C R(IK+1) = DXK = DERIVATIVE OF TK WITH RESPECT TO X
C R(IK+2) = DYK = DERIVATIVE OF TK WITH RESPECT TO Y
C R(IK+3) = DZK = DERIVATIVE OF TK WITH RESPECT TO Z
C R(IE) = TE = TURBULENCE_E OR TURBULENCE-W
C R(IE+1) = DXE = DERIVATIVE OF TE WITH RESPECT TO X
C R(IE+2) = DYE = DERIVATIVE OF TE WITH RESPECT TO Y
C R(IE+3) = DZE = DERIVATIVE OF TE WITH RESPECT TO Z
C R(IX) = X = X-COORDINATE
C R(IY) = Y = Y-COORDINATE
C R(IZ) = Z = Z-COORDINATE
C R(ITIM) = TIME = TIME
C R(IM) = S = MASS-RATIO-1
C R(IM+1) = DXS = DERIVATIVE OF S WITH RESPECT TO X
C R(IM+2) = DYS = DERIVATIVE OF S WITH RESPECT TO Y
C R(IM+3) = DZS = DERIVATIVE OF S WITH RESPECT TO Z
C ...... CONTINUE FOR MASS-RATIO-2,...,UNTIL THE MASS-RATIO-10
C
C -----------------------------------------------------------------
C

C CMAT(1 ) = XMU = VISCOSITY

C CMAT(2 ) = CP = SPECIFIC HEAT AT CONSTANT PRESSURE
C CMAT(3 ) = XKCON = THERMAL CONDUCTIVITY
C CMAT(4 ) = BETA = COEFFICIENT OF VOLUME EXPANSION
C CMAT(5 ) = RO = DENSITY
C CMAT(6 ) = TC = REFERENCE TEMPERATURE
C CMAT(7 ) = GRAVX = X-GRAVITATION
C CMAT(8 ) = GRAVY = Y-GRAVITATION
C CMAT(9 ) = GRAVZ = Z-GRAVITATION
C CMAT(10) = QB = RATE OF HEAT GENERATED PER VOLUME
C CMAT(11) = SIGMA = COEFFICIENT OF SURFACE TENSION
C CMAT(12) = XKAPA = BULK MODULUS
C CMAT(13) = CV = SPECIFIC HEAT AT CONSTANT VOLUME
C
C======================================================================
CHARACTER*80 CHEAD
DIMENSION L(NL),R(NR),CMAT(NCMAT)
C----------------------------------------------------------------------
C WRITE YOUR OWN CODE HERE AND GO TO 9999
C----------------------------------------------------------------------
IF(CHEAD(1:6).EQ.'CATERP') THEN
C
C GET THE CURRENT VALUE OF MASS-RATIO
C
IM = L(20)
S = R(IM)
C
C SET UP THE AIR MATERIALS
C
DENAIR = 0.995D0
VISAIR = 2.082D-5
CPVAIR = 1009.D0
XKCAIR = 0.0304D0
C
C SET UP THE OIL MATERIALS
C
DENOIL = 853.9D0
VISOIL = 3.56D-2
CPVOIL = 2118.D0
XKCOIL = 0.138D0
C
C COMPUTE THE VISCOSITY AND DENSITY OF THE MIXTURE
C
IF(S.GT.1.D0) THEN
VIS = VISOIL
DEN = DENOIL
CPV = CPVOIL
XKC = XKCOIL
ELSE IF(S.LE.0.D0) THEN
VIS = VISAIR
DEN = DENAIR
CPV = CPVAIR
XKC = XKCAIR
ELSE
VIS = VISAIR + S*(VISOIL-VISAIR)
DEN = DENAIR + S*(DENOIL-DENAIR)
CPV = CPVAIR + S*(CPVOIL-CPVAIR)
XKC = XKCAIR + S*(XKCOIL-XKCAIR)
ENDIF
C
C PASS OUT ALL MATERIAL DATA REQUIRED
C

CMAT(1 ) = VIS
CMAT(2 ) = CPV
CMAT(3 ) = XKC
CMAT(4 ) = 0.D0
CMAT(5) = DEN
CMAT(6 ) = 0.D0
CMAT(7 ) = 0.D0
CMAT(8 ) = 0.D0
CMAT(9 ) = 0.D0
CMAT(10) = 0.D0
CMAT(11) = 0.D0
CMAT(12) = 0.D0
CMAT(13) = 0.D0
C
GO TO 9999
ENDIF
C
C----------------------------------------------------------------------
IF(CHEAD(1:6).EQ.'F71001' .OR. CHEAD(1:5).EQ.'F7518') THEN
C +---------------------------------------------------+
C | THIS FOR THE INTERNAL VERIFICATION PROBLEM F71001 |
C | WHICH HAS A TEMPERATURE DEPENDENT DENSITY |
C +---------------------------------------------------+
CMAT(1 ) = 8.D-6
CMAT(2 ) = 14030.D0
CMAT(3 ) = 0.156D0
CMAT(4 ) = 0.D0
CMAT(5 ) = 0.086D0
CMAT(6 ) = 0.D0
CMAT(7 ) = 0.D0
CMAT(8 ) = 0.D0
CMAT(9 ) = 0.D0
CMAT(10) = 0.D0
CMAT(11) = 0.D0
CMAT(12) = 0.D0
CMAT(13) = 0.D0
C
IT = L(11)
T = R(IT)
TP1 = 100.D0
TP2 = 1000.D0
DP1 = 0.086D0
DP2 = 0.053D0
IF(T .LT. TP1) THEN
CMAT(5) = DP1
ELSE IF(T .LT. TP2) THEN
CMAT(5) = DP1 + (T-TP1)*(DP2-DP1)/(TP2-TP1)
ELSE
CMAT(5) = DP2
ENDIF
ELSE IF(CHEAD(1:5).EQ.'F7100' .OR. CHEAD(1:5).EQ.'F7519') THEN
C +---------------------------------------------------+
C | THIS FOR THE INTERNAL VERIFICATION PROBLEM F7100X |
C | WHICH HAS A Z-COODINATE DEPENDENT DENSITY (X=3-8) |
C +---------------------------------------------------+
CMAT(1 ) = 0.2D0
CMAT(2 ) = 0.D0
CMAT(3 ) = 0.D0
CMAT(4 ) = 0.D0
CMAT(5 ) = 0.D0
CMAT(6 ) = 0.D0
CMAT(7 ) = 0.D0
CMAT(8 ) = 0.D0

CMAT(9 ) = 0.D0
CMAT(10) = 0.D0
CMAT(11) = 0.D0
CMAT(12) = 0.D0
CMAT(13) = 0.D0
C
IZ = L(16)
Z = R(IZ)
CMAT(5) = 1.D0 + Z/10.D0 * 4.0D0
ELSE IF(CHEAD(1:6).EQ.'F74007') THEN
C +-----------------------------------------------------+
C | THIS FOR THE INTERNAL VERIFICATION PROBLEM F74007 |
C | WHICH HAS A MASS DEPENDENT DENSITY |
C +-----------------------------------------------------+
IM = L(20)
S = R(IM)
CMAT(1 ) = 0.01D0
CMAT(2 ) = 0.D0
CMAT(3 ) = 0.D0
CMAT(4 ) = 0.D0
CMAT(5 ) = 0.D0
CMAT(6 ) = 0.D0
CMAT(7 ) = 0.D0
CMAT(8 ) = 0.D0
CMAT(9 ) = 0.D0
CMAT(10) = 0.D0
CMAT(11) = 0.D0
CMAT(12) = 0.D0
CMAT(5) = 1.D0
C
IF(S.GT.0.9D0) THEN
CMAT(1) = 0.1D0
CMAT(5) = 10.D0
ELSE IF(S.LT.0.1D0) THEN
CMAT(1) = 0.01D0
CMAT(5) = 1.D0
ELSE
RRR = (S-0.1D0)/0.8D0
CMAT(1) = 0.01D0 + RRR*(0.1D0-0.01D0)
CMAT(5) = 1.D0 + RRR*(10.D0-1.D0 )
ENDIF
ENDIF
C----------------------------------------------------------------------
9999 CONTINUE
RETURN
END
3.6.10 ASME steam table
This model automatically calculates water and steam properties based

on the ASME steam table (ASME International Steam Tables for Industrial
Use, ASME Press, 2000). The user has the choice of whether to consider
or not meta-stable vapor status. The following parameters and conversion
factors are input by the user:


0 = reference temperature
p0 = reference pressure
p , p = conversion factors for pressure, p m = ( p s p0c ) p c
c
0
c
0c , c = conversion factors for temperature, m = ( s 0c ) c

c = conversion factor for density, m = s c
Lc = conversion factor for length, Lm = Ls Lc
where p m , m , m and Lm are the pressure, temperature, density and

length specified in the model. Additionally, p s , s , s and Lm are the
pressure (Pascal), temperature (K), density (kg/m3) and length (m) in SI
units as defined in the ASME stem tables.
In the incompressible flow model, is the value of density at pressure

p0 and temperature 0 . In the slightly compressible flow model, m is
the value of density at pressure p and temperature . Furthermore, in
low-speed compressible flow model, both m and are the values of
density at pressure p and temperature .
3.6.11 P-T-dependent material
When compared with the constant material model, this model has one
additional parameter, namely, the reference pressure, p0 . In addition, the
following parameters are a function of both pressure and temperature
, C p , Cv , k , , q B , , , = f ( p, )
In the incompressible flow model, is the density at pressure p0 and

temperature 0 . In the slightly compressible flow model,
m = (1 + ( p p0 ) ) and is the density at pressure p0 and

temperature 0 . Furthermore, in low-speed compressible flow model, both

m and are the density at pressure p and temperature .
3.6.12 Field friction
This model adds a field friction force to the right-hand side of the
momentum equations. The force has the following form,
2f
fB = v C v
D
where f = a ( Re D ) , Re D = v D , C = Ci ei ei , ei , ( i = 1, 2,3) are

b
normalized independent direction vectors ( (e1 e 2 ) e3 0 ) , , v, are

local density, velocity, and viscosity respectively, D is characteristic
length, and a, b are two dimensionless constants.
The parameters D, a, b, C1 , C2 , C3 , e1 , e 2 and e3 are required.
Note: In 2D models, e1 and the x- components of e 2 and e3 are ignored.
3.6.13 User-modified materials
For incompressible, slightly compressible and low-speed compressible

flows, the user can modify the material properties in the user-supplied
subroutine USRMAT. This option covers all type of materials described
above except the user-supplied materials. In particular, it covers turbulence
and porous media materials. The table at the beginning of the user
subroutine lists all material properties that can be modified.
This subroutine is invoked whenever material data is required during

computation. All necessary data is passed in through arguments and the
program returns the material data that has been modified. User may modify
part or all of the material data. For unchanged data, the values defined in
the model material data are used.

If a field friction loading is applied to the element group, user may also
modify it in this subroutine. Please note, once changed, the complete field
friction data associated with this element group must be re-defined. They
replace original friction data.
To activate this option, choose "Modify Material Properties During

Solution with User Subroutine" in element group.
3.6.14 Material curves
A more general material set can be defined with the selected materials
associated with up to three material curves of type tabulated,
polynomial and 2-argument respectively. Those material curves replace
the default material data in the selected materials.
The following types of material model can be associated with the material
curves:
Constant material model

Power-law model
Temperature-dependent power-law model
Carreau model
Second-order model
Large-eddy-simulation model
User-suuplied materials
Turbulent K - model
Turbulent K - model
Turbulent SA model
Porous material model (for flow through porous media)
The three material curves are introduces here after.
A tabulated material curve defines tabulated material data in the form
xi : Pi1 ( xi ) , Pi 2 ( xi ) ,..., Pi , J ( x j ) ( i = 1, 2,..., I )

where x can be time, temperature or pressure, and P are the values of the
parameters to be modified. The possible parameters that can be selected are

= density
= viscosity
k = heat conductivity
c = the empirical constant in K - models
= the empirical constant in K - models
= fluid bulk modulus of elasticity
A polynomial material curve defines material data in a polynomial
Ii
Pj = ai x ei ( j = 1, 2,..., J )
i =1
where x and P are the same as defined for tabulated material curve.
A 2-argunment material curve defines material data varying as function of

two arguments (any set of two arguments of time, temperature and
pressure). The curve is input in the form
( xi , MCVi ) ( i = 1, 2,..., I )
where x can be time, temperature or pressure, and MCVi are either
tabulated or polynomial material curves. Thus, at xi , the material values
are determined by a material curve that is a function of another varying
argument different from x . For example, if temperature represents x in
the 2-argument material curve and, in which, MCVi are pressure-
dependent material curves of C p and Cv , the curve defines temperature-
pressure-dependent material parameters C p ( , p ) and Cv ( , p ) .

3.6.15 Gas models
The following state equations are available for low-speed compressible

flows with FCBI-C elements associated with the total energy equation:
ideal gas, standard Redlich-Kwong, Aungier-Redlich-Kwong, Soave-
Redlich-Kwong, and Peng-Robinson. These are defined using the STATE-
EQUATION command, and are applied using the STATE-EQ parameter of
the MATERIAL commands. Different materials in the model can use
different equations of state.
In case that Cv is temperature-dependent (it is in most cases), define it

by appllying a material curve in the selected material.
The following state equations are available for low-speed compressible

flows with FCBI-C elements: ideal gas, standard Redlich-Kwong, Aungier-
Redlich-Kwong, Soave-Redlich-Kwong, and Peng-Robinson. These are
defined using the STATE-EQUATION command, and are applied using
the STATE-EQ parameter of the MATERIAL commands. Different
materials in the model can use different equations of state.
The state equations implicitly express the specific volume v ( = 1 ) as

a function of pressure, p , temperature, , and the specific gas constant,
Rsp . Depending on the model, the critical pressure, pC , critical
temperature, C , critical specific volume, vC , and an acentric factor, ,
may also be required.
The state equation affects the internal energy of the fluid. This effect
can be accounted in the energy equation using the concept of an effective
( )
specific heat at constant volume Cv* , where de = Cv* d and e is the
specific internal energy. The effect can be switched on or off using the CV-
EFFECTIVE parameter of the STATE-EQUATION command. If the effect
is accounted for, Cv* is different from the user-input value of Cv . If it is not

accounted for, the user-input value of Cv is directly used in the energy

equation (i.e., Cv* = Cv is used in this case).
Ideal Gas law (IDEAL)
The ideal gas law is represented as
pv = Rsp
where the specific gas constant, Rsp , is a user-defined input.
This gas model is used when Cv should vary in the energy equation,
but the specific gas constant, Rsp , should not vary in the equation of state.
In this model, the user-input Cv defined in the MATERIAL command is
only used in the energy equation, and the user-input C p is ignored.
Note that if STATE-EQ=0 in the MATERIAL command, the ideal gas

law is used, where the specific gas constant is computed from the specific
heats, Rsp = C p Cv .
Redlich Kwong models (RK, ARK, and SRK)
All Redlich-Kwong models can be expressed in the following general

form
Rsp a
p=
v b+c v (v + b)
where the expressions for a , b , and c depend on the type of Redlich-

Kwong model used.

Standard Redlich Kwong model (RK)
In the standard Redlich-Kwong model, the expressions for a , b , and c

are given by
n

a = a0
C
Rsp C
b = 0.08664
pC
Rsp2 C2
a0 = 0.42747
pC
c=0
n= 1
2
where the required input parameters are Rsp , pC , C .
Aungier Redlich Kwong model (ARK)
The Aungier-Redlich-Kwong model is a modified version of the

standard Redlich-Kwong model with
Rsp C
c= + b vC
a0
pC +
vC ( vC + b )
n = 0.4986 + 1.1735 + 0.4754 2
The required input parameters are Rsp , pC , C , , vC .
Soave Redlich Kwong model (SRK)
The Soave-Redlich-Kwong model is another modified version of the

standard Redlich-Kwong model with

2

a = a0 1 + n 1

C
n = 0.480 + 1.574 0.176 2
The required input parameters are Rsp , pC , C , .
Peng Robison model (PR)
The Peng-Robison model is expressed as
RspT a
p=
vb v + 2vb b 2
2
where a has the same expression as Soave-Redlich-Kwong model and
Rsp C
b = 0.0778
pC
Rsp2 C2
a0 = 0.45724
pC
n = 0.37464 + 1.54226 0.26993 2
and the required input parameters are Rsp , pC , C , .

Chapter 4: Formulation of fluid flows through porous media
Chapter 4 Formulation of fluid flows through porous

media
In this chapter, we introduce our formulation for fluid flows and heat
transfer in porous media.
The model introduced here can be coupled with ADINA solid models in
fluid-structure interaction analyses. It can also be associated with mass
transfer analyses. Used as additional element groups, it can also be
employed in incompressible, slightly compressible and low-speed
compressible flow solutions.

are available for porous media flows.
Table 4-1 Functionalities in the porous media flow model
ADINA solid model
incompressible, slightly compressible and low-speed compressible flows
heat transfer
Coupled models
mass transfer
volume of fraction
Galerkin Elements
Material models constant
Boundary conditions prescribed velocities
v
zero velocities
Note: The column prescribed pressure
furthest to the right p
zero pressure
represents the solution prescribed rotational velocity v
variables or equations concentrated force load
that are immediately
affected by the

affected by the field centrifugal load

specified condition. fixed wall
The solution variable uniform flow
d indicates a moving angular velocity
boundary condition. moving wall
v, d
fluid-structure interface
free surface
fluid-fluid interface d
phase-change
gap all
prescribed temperature
zero temperature
concentrated heat flow load

convection
radiation
specular radiation
thermal resistance
user-supplied all
zero (default conditions)
specified
Initial conditions
mapped from other solutions
restart run
sparse solver
equations
skew system
Other capabilities conjugate heat transfer
pressure datum
physical or mathematical formulations for incompressible flows
The governing equations used in our formulations are the continuity

equation, energy equation and Darcys equation (momentum equation).
They are listed in Table 2-1. The solution variables are the primitive

variables ( p, v, ) . In case of moving mesh problems, the ALE

formulation described in Section 2.14 is used.
The numerical method for continuity and energy equations in porous

media is the same as for incompressible flow analyses.
The momentum equation is weighted with the virtual quantity of
velocity and is integrated over the computational domain V . The
divergence theorem is used to lower the order of the derivatives of the
pressure, resulting in the expression of natural boundary conditions. This
natural condition is part of usual boundary conditions. The variational
form is written as
h v ( 1 i v f B ) ph v dV = h v pdS
V (4.1)
where h v the is virtual quantity of velocity. The associated elements can

be any one that is applicable to porous media flows. Upwinding techniques
are not applied to Eq.(4.1).
In 2D 4-node and 3D 8-node elements, a finite volume method is used.

Using the related interpolation functions described in Chapter 10, the
discretized equations are
1 B
V
i( v w ) f dV + S pdS = 0
Similar to incompressible flows, all variables at every node will form

the solution vector that is governed by the finite element equation. Both
the Newton-Raphson and successive substitution methods are available to
solving the nonlinear equations. All solvers introduced in Chapter 11 are
applicable to solving the linear algebraic equations.
The same solution procedure as for solving incompressible flows is
carried out for solving porous media flow problems.

4.3: Elements
4.3 Elements
The elements that can be used for porous media are the same as those
for incompressible flows.
Using proper boundary conditions is essential to successfully solving

fluid flow problems. The required boundary conditions strongly depend on
the type of flows or, equivalently, on the governing equations used and the
associated computational domain.
Fluid velocity and pressure are always strongly coupled and must be
treated like one variable as far as boundary conditions are concerned. On
the other hand, the fluid flow might not be directly coupled to the
temperature (although the flow may be coupled to the temperature field
through temperature-dependent materials and the buoyancy force in the
Boussinesq approximation). Therefore, as a general rule, they must be
considered separately when applying boundary conditions.
A typical fluid domain and associated boundary conditions are shown
in the following figure, where the whole boundary S of the computational
domain has been divided into S1 (where velocities are prescribed) and S 2
(where a normal-traction load is applied). The Si here represents the
interior part of that part of the boundary, while the over bar indicates that it
contains the boundary of Si as well. Therefore, the common part of S1
and S 2 is the empty geometric set, while S1 and S 2 share their common
frontier where both conditions are applied. This boundary partition can be
expressed as
S1 S2 = S = S , S1 S2 =

prescribed velocities on S 1
applied normal-traction on S2
v n=0
v n=0 nn
v
Figure 4.1 A typical boundary condition set for porous media

flows
For steady-state analyses, at least one velocity condition and one
pressure condition must be specified (the distributed normal- traction
load acts as a prescribed pressure condition).
A slip condition is usually applied on walls.
These observations are fundamental guidelines for specifying boundary

conditions in most porous media flow problems.
The discussion of boundary conditions for heat transfer in Section 3.4 is
also valid for heat transfer in porous media.
All the boundary conditions available in incompressible flows are also
applicable to porous media flows. These conditions have been introduced
in Section 3.4.

In transient analyses, all solution variables must be specified. The

default initial conditions to all variables are zeros. Although initial
conditions are not required in steady-state analyses, they are used as a
guessed solution at the start of the equilibrium iterations. A good initial
condition may accelerate the convergence during equilibrium iterations. In
certain cases, the initial condition may become a key factor in obtaining
converged solutions.
4.6 Material model
The following data are required in porous media flow models:
= fluid density
s = solid density
= fluid viscosity
= porosity
= permeability tensor
Cp f = specific heat of fluid
Cps = specific heat of solid
kf = thermal conductivity of fluid
ks = thermal conductivity of solid
B
q = rate of heat generated per unit volume

Note: A more general time, temperature or pressure dependent porous

material set can be defined using material curves. See corresponding
sections in Chapter 3 for more details.

Chapter 5 Formulation of turbulence in

incompressible, slightly compressible and
low-speed compressible flows
In this chapter, we introduce our two-equation turbulence models for

incompressible and slightly compressible flows with heat transfer. These
models are also applicable to low-speed compressible flows.
The models introduced here can be coupled with ADINA solid models
in fluid-structure interaction analyses. They can also be used in mass
transfer analyses.
are available for turbulence models for incompressible, slightly
compressible and low-speed compressible flows.
Table 5-1 Functionalities in K - and K - turbulence models

for incompressible, slightly compressible and low-
speed compressible flows
ADINA solid model
Coupled models solid element groups
Note: Heat transfer
heat transfer
must be coupled in
low-speed compressible mass transfer
flow models. Electro-static and steady current conduction analyses
2D: 3-node linear element
Galerkin Elements
3D: 4-node linear element
sandard turbulent K - and K - RNG
Material models high and low Reynolds turbulent K -
Shear Stress Transport (SST) model

Chapter 5: Formulations of turbulence in incompressible,
Prescribed velocities
v
zero velocities
prescribed pressure
p
zero pressure
prescribed rotational velocity
v
field centrifugal load
fixed wall v, K ,
Boundary conditions uniform flow
v
angular velocity
Note: The column
moving wall
furthest to the right v, K , , d
represents the solution fluid-structure interface
variables or equations free surface
that are immediately fluid-fluid interface d
affected by the phase-change
specified condition. Gap all
The solution variable prescribed temperature
d indicates a moving zero temperature
boundary condition. concentrated heat flow load

convection
radiation
specular radiation
thermal resistance
user-supplied all
prescribed turbulence K - K,
zero flux of turbulence variables K,
specified
Initial conditions
restart run
sparse solver
equations
skew system
Other capabilities
pressure datum

The governing equations for the fluid flow are those of incompressible,
slightly compressible or low-speed compressible flows. However, the
viscosity and heat conductivity are modified to include the turbulence effect
and the K - and K - models are used to provide the required variables.
These equations are listed in Table 2-1. The solution variables are
( p, v,,, ) in the K - models and ( p, v,,, ) in the K - models. In
the case of moving mesh problems, the ALE formulation described in
Section 2.14 is used.
In the K - models, both the standard and the RNG models can be used.
In the K - models, both the high and low Reynolds models can be used.
The time integration is the same as for the fluid flow equations. The
fluid equations and their discretization are described in Section 3.2.
The finite element method is used to discretize the governing equations

of the turbulence. As for the fluid flow equations, the finite element
equations are obtained by establishing a weak form of the turbulence
equations using the Galerkin procedure. The K, and equations are
weighted with the virtual quantities of K, and , respectively. The
equations are integrated over the computational domain V . The divergence
theorem is used to lower the order of the derivatives of turbulence fluxes.
The variational forms for the K- equations are
(h G f + Q f ih f ) dV = 0
f
V
where f represents K and , h K and h are virtual quantities of K and

respectively, and

K
G K = y a + viK S K
t

G = y a + vi S
t
Q K = y aq K
Q = y aq
Similarly, finite element equations for the K - models are obtained

with f representing K and , and h K and h are virtual quantities of K
and , respectively, and
K
G K = y a + v i K GK
t

G = y a + v i G
t
Q K = y aq K
Q = y aq
The upwinding techniques are also used in constructing the finite

element equations governing the turbulence variables.
In FCBI and FCBI-C elements, conservative forms associated with a

finite volume method is used, in which the corresponding terms are defined
as
K vK q K GK
f = y , B = y v q , R = y G

a a a
v q G

5.3: Elements
5.3 Elements
The elements that can be used for turbulence flows here are the same as
those used for incompressible flows except that bilinear elements are not
applicable.
The boundary conditions for fluid variables, ( v, p, ) , are usually the

same as for laminar models, although the treatment of the wall condition or
its analogue is modified as explained later in this section. The conditions
introduced here are only for the turbulence variables K , and .
A typical fluid domain and its associated boundary conditions for
turbulence variables, K, and , are shown in the following figure, where
the whole boundary S of the computational domain has been divided into
S5 (where the turbulence variables are prescribed), S6 (where zero
turbulence fluxes are applied) and S7 (where wall conditions have been
applied). The Si here represents the interior part of that part of the
boundary, while the over bar indicates that it contains the boundary of Si
as well. Therefore, the common part of Si and S j is the empty geometric
set when i j , while Si and S j share their common frontier where both
conditions are applied. This boundary partition can be expressed as
S5 S6 S7 = S = S
Si S j = i j , i, j = 5,6,7

prescribed zero fluxes of turbulence

turbulence on S 5 varaibles on S 6
wall conditions on S 7
Figure 5.1 A typical boundary condition set for turbulence

variables
All parts of the boundary have been assigned one and only one
turbulence boundary condition (except on the interface of the
sub-boundaries, where two conditions meet).
There is at least one wall condition or a prescribed turbulence variable
condition.
These observations are fundamental guidelines for specifying boundary

conditions for turbulence variables. The available conditions are
introduced here.
5.4.1 Prescribed turbulence K -

prescribed turbulence K -
This condition can only be applied to a K- model. Time-dependent
turbulence variables are directly prescribed
K = K (t )
= (t )
and applied to boundaries. The K and equations at boundary nodes are

then replaced by these equations.

In addition to directly given values, they can also be prescribed

indirectly via
2
K= 3
2 (i v )
= K 2 (0.3L )
3
where, v , L and i (= 0.01 ~ 0.1) are velocity scale, length scale and
turbulence intensity respectively.
Note that, in SA and DES models, through all input and output, K
represents (modified eddy viscosity) and can be prescribed as
= 0.033 i v L .
This condition is usually applied to inlet boundaries.

This condition can only be applied to a K- model. Time-dependent
turbulence variables are directly prescribed
K = K (t )
= (t )
and applied to boundaries. The K and equations at boundary nodes are

then replaced by these conditions.
indirectly via
2
K= 3
2 (i v )
= K 2 (0.27 L )
1
where, v , L and i (= 0.01 ~ 0.1) are velocity scale, length scale and
turbulence intensity respectively.

5.4.3 Zero flux of turbulence variables
The natural boundary condition of any equation requires a specified

normal flux of that variable. The turbulence flux condition is currently
unavailable in ADINA-F (practically such conditions are not available).
Therefore a zero flux is implicitly assumed along the boundary where no
turbulence condition is explicitly specified.
5.4.4 Solid walls
This condition is probably the most important one in modeling

turbulence. Let us take the K- model as an example. At a solid wall, the
no-slip condition applies so that both the mean and fluctuating velocities
are zero. In the near wall region where 0< y + <100, the wall function is
applied. In this region, the Reynolds stresses v ' v ' are nearly constant, so
is the kinetic energy K. The convection and diffusion in the K-equation are
negligible so that local equilibrium prevails. When buoyancy effects are
absent, this implies that all source terms in the K-equation are zero. We
further recall the definition of the shear stress and obtain
t (v ') 2 = , t v ' = v*2
The solutions of the above equations determine the variables K and in

the near-wall region
v*2 v*4
Kw = , w = W '( yw+ )
c 0
Similarly, the boundary values of turbulence variables in K - models

are obtained
v*2 v*2
Kw = , w = W '( yw+ )
K 0 K

The condition for momentum equations is the friction force computed

by integrating the shear stress v*2
F = hi v*2 dS
and added to the right hand side of the momentum equations, where is
the tangential direction of the wall.
Wall temperature can be specified in two ways. The first one is to

directly apply usual or special thermal conditions as done in laminar
models. In this method, temperature condition must not be specified in
wall boundary condition. The second method is to specify a thermal
condition in the wall boundary condition, such that the wall function shall
be incorporated into the temperature condition. This method is described
below.
Reynoldss analogy between momentum and energy transport gives a
similar law for mean temperature. The law-of-the-wall for temperature
implemented in ADINA-F is
Pr y + y + < y+
T ( y , Pr, ) =
+ +
(5.1)
W ( y ) + P
+
y + y+
where P is computed by using the formula given by Jayatilleke
(
P = 9.24 ( Pr ) 4 1 1 + 0.28e 0.007Pr )
3
Given the molecular Prandtl number, the thermal sublayer thickness y+ is

computed as the y + value at which the two curves in Eq. (5.1) intersect.

The formula
v*Cv ( w )
T+ =
qw
is then used to determine the wall heat flux or wall temperature, depending
on the type of the thermal wall boundary condition ( specified temperature
w or heat flux qw ).
The near wall treatment of the fluid and the turbulence variables, as
discussed above, is automatically enforced whenever a wall condition or its
analogue is applied. These conditions are
No-slip fixed/moving wall conditions

Moving wall condition (type=tangential)
Moving wall condition (type=rotational)
No-slip fluid-structure interface
Fluid-structure interface (type=tangential)
Fluid-structure interface (type=rotational)
These conditions are defined in the same user-supplied subroutines and

applied in the same manner as for fluid flows and temperature conditions.
The detailed description is given in Section 3.4.

In transient analyses, all solution variables ( v, p, ,, ) or

( v, p, ,, ) must be specified, depending on the turbulence model
selected. The default initial conditions to all variables are zero conditions.
Although initial conditions are not required in steady-state analyses, they
are used as a guessed solution in the equilibrium iterations. A good
initial condition, as will be seen in Chapter 13, may accelerate the
convergence during equilibrium iterations. In certain cases, the initial
condition may become a key factor in obtaining converged solutions.
In low-speed compressible flows, recalling that pressure, temperature
and density must satisfy the state equation of the ideal gas law, zero
pressure or zero temperature must not be specified. Since the default initial
conditions to all variables are zero, actual initial conditions are always
required in this case in both transient and steady-state analyses.
In turbulent flows, initial conditions can be selected as the same as the
value of turbulent variables at inlet boundary conditions.
5.6 Material models
5.6.1 K - turbulence model

material model: K - turbulence
This material model can only be applied to the turbulence K - model.
In this model, both the fluid properties and the empirically determined
model constants must be input.
The fluid properties are
= fluid density
= fluid viscosity


k = thermal conductivity
As mentioned in Section 3.6, not all these parameters are required in all
problems. The parameters that must be input depend on the problem to be
solved.
In incompressible flows considered here, the bulk modulus of elasticity
and the specific heat at constant volume Cv are not required. In this
case, is assumed to be infinity and Cv is forced to be equal C p . If heat
transfer is not considered, C p , k , q B , , and 0 can be ignored.
In slightly compressible flows, Cv is forced to be equal to C p and,
therefore, can be ignored. C p , k , q B , and 0 can also be omitted if heat
transfer is not included.
In low-speed compressible flows, is not used in the computations
and therefore ignored. The density is also predetermined through the state
equation and thus not required. All other material data must be input.
The above convention is used throughout the chapter.
Besides the above fluid properties, the additional model constants
c , c1 , c2 , c3 , K , , , d w+ and 0 must also be specified. Their
default values are
c = 0.09, c1 = 1.44, c2 = 1.92,

c3 = 0.8, K = 1, = 1.3,
= 0.9, d w+ = 70, 0 = 0.4

5.6.2 RNG K - turbulence model

material model:RNG K - turbulence
The data required in this model is the same as that in the K - turbulence
K - model. The default values of the empirical constants are the same as
for the standard K - model. Another conventional set of these data are
c = 0.085, c1 = 1.42, c2 = 1.68,

c3 = 0.8, k = 0.7179, = 0.7179,
= 0.9, d w+ = 70, 0 = 0.4
5.6.3 High Reynolds K - turbulence model

material model: K - high-Reynolds turbulence
Besides the required fluid properties as defined in the K - material model,
the additional model constants for turbulence variables,
, , K , , K , , , , d w+ and 0 must also be specified. Their
default values are
= 1, = 5/9, = 0.09,
= 0.075, K = 2, = 2,
= 0.712, = 1, d w+ = 70,
0 = 0.41
5.6.4 Low Reynolds K - turbulence model

material model: K - low-Reynolds turbulence
This material model contains the same input as the high Reynolds K -
model and can only be applied to the turbulence K - model.

5.6.5 Shear Stress Transport model
This material model can only be applied to the K - turbulence model.

The constants for turbulence variables are set automatically and their values
are:
K1 = 1.176, K2 = 1.0, 1 = 2.0,

2 = 1.168, = 1.0, 1 = 0.553,
2 = 0.553, 1 = 0.075, 2 = 0.0828,
a1 = 0.31 d w+ = 70 0 = 0.41
The Low-Reynolds and High Reynolds variants are available in the SST
material model. In addition, the model is available for FCBI and FCBI-C
elements.
5.6.6 Spalart-Allmaras and Detached Eddy Simulation material models
This material model can only applied to SA and DES turbulence

models. Besides the fluid properties, there are additional constants that
must be specified. These constants and their default values are:
cw 2 = 0.3, cw3 = 2.0, = 0.41,

cb1 = 0.1355, cb 2 = 0.622, cv1 = 7.1,
s = 2 3, C prod = 2, = 1,
CDES = 6.4, cw1 = cb1 2 + (1 + cb 2 ) s
The SA and DES models are available for FCBI and FCBI-C elements.
In both pre- and post-processing the turbulent kinetic energy K (as in the
K - model), through all input and output, represents . Therefore, when
prescribing , specifying its initial value, relaxation factors, and other
input parameters, K as displayed on AUI and in its commands becomes .

5.6.7 Two-layer zonal material model
This model is always associated with K - model. Its material includes all
data for K - model plus some additional ones, names
A , A , cl , Rc min , Rc max , Lw . Their default values are determined by the
program and explained in Chapter 2.
5.6.8 Material curves
See corresponding section in Chapter 3 for details.
5.6.9 Gas models
See corresponding section in Chapter 3 for details.

Chapter 6: Formulations of high-speed compressible flows
Chapter 6 Formulation of high-speed compressible

flows
In this chapter, we introduce our formulations for high-speed

compressible flows. This model can be coupled with ADINA solid models
and be associated with mass transfer solutions.
ref. Bathe, K.J., Zhang, H., and Ji, S., Finite Element Analysis
of Fluid Flows fully Coupled with Structural Interactions,
Computers & Structures, Vol.72, pp.1-16, 1999.
ref. Bathe, K.J., Zhang, H., and Zhang, X., Some Advances in
the Analysis of fluid flows, Computers & Structures,
Vol.64, pp.909-930, 1997.
ref. Bathe, K.J., Zhang, H., and Wang, M.H., Finite Element
Analysis of Incompressible and Compressible fluid flows
with Free Surfaces and Structural Interactions, Computers
& Structures, Vol.56, pp.193-214, 1995.
ref. Bathe, K.J., Walczak, J., and Zhang, H., Some Recent
Advances for Practical Finite Element Analysis, Computers
& Structures, Vol.47, pp.511-521, 1993.

are available for high-speed compressible flows.
Table 6-1 Functionalities in the high-speed compressible flow

model
Coupled models ADINA solid model
Note: Heat transfer solid element groups
must be coupled. mass transfer

6.1: Governing Equations

3-node triangle
Elements
4-node tetrahedron
Constant
temperature-dependent
Material models pressure-dependent
pressure-temperature-dependent
user-supplied
prescribed specific discharges
zero specific discharges
prescribed density v
Boundary conditions field centrifugal load
prescribed total energy
Note: The column concentrated heat flow load E
furthest to the right distributed heat flux load
represents the solution fixed wall
variables or equations v,
angular velocity
that are immediately
moving wall v, , d
affected by the
specified condition. external flow
The solution variable supersonic at inlet
d indicates a moving subsonic at inlet
, v, E
boundary condition. supersonic at outlet
subsonic at outlet
symmetric
fluid-structure interface v, , d
Gap
all
User-supplied
Initial conditions Zero (default conditions)
specified. restart run
sparse solver
equations
iterative methods (RPBCG, RPGMRES, and multi-grid)
Other capabilities skew system
pressure datum

The governing equations used in our formulations are the conservative

forms of the Navier-Stokes equations that are listed in Table 2-1. For
moving mesh problems, the ALE formulation described in Section 2.14 is
used. The solution variables are the conservative variables ( , v, E )
(although the primitive variables are still used in the input phase for the
purpose of convenience).
The numerical method used here in solving the governing equations

combines a finite volume approach to the inviscid terms and a finite
element approach to the viscous terms.
The basic step of a finite volume procedure is the calculation of the flux
at the interface of a control volume. The total flux consists of the inviscid
Eulerian flux term, Fn , that includes the convective term and pressure, and
the viscous term, G n , that consists of viscous and conductive terms. We
calculate the flux Fn using a finite volume-based flux-splitting method
while we calculate the viscous term G n using a finite element-based
method.
In two-dimensional geometries, the control volume is based on the use
of the triangular element. The faces of a control volume consist of straight
lines. In three-dimensional geometries, the control volume is based on the
uses of the tetrahedral element. The faces of a control volume consist of
rectangular planes and each of the rectangular consists of two triangles.
Each control volume is divided into a few sub-control volumes that are
located in the neighboring elements. They are detailed in Section 6.3.
Let us integrate the governing equation, in an exact manner on a general
moving control volume V enclosed by its faces S
(VU ) t (VU ) +
t +t
(F + G )dS CdV dt = 0 (6.1)
n V
t +t
n
t

The surface and volume integrations are then approximated by some

averaged values denoted by double-over bars
t +t
(VU ) t (VU ) + t (Fn + G n ) S CV = 0

The above terms without superscripts are calculated at time t + t if an

implicit method is used, or at time t if an explicit method used. We will
introduce our approaches to these averaged terms in the following sections.
6.2.1 Geometric conservation law in moving meshes
ref. Zhang, H., Reggio, M., Trpanier, J.Y., and Camarero, R.,
Discrete Form of the GCL for Moving Meshes and Its
Implementation in CFD schemes, Computers Fluids,
Vol.22, No.1, pp.9-23, 1993.
Consider a special flow pattern: stationary flow with constant density,

pressure and temperature. In this trivial case, Eq.(6.1) becomes
t +t
V tV =
t +t
wn dSdt V (6.2)
t S
This equation states that during a time interval the increased volume equals
the volumetric increase as a consequence of the motion of the faces.
Eq.(6.2) is not an additional equation that needs to be solved, but must be
satisfied implicitly as a fundamental geometric rule. It is so called the
geometric conservation law (GCL). Any violation of GCL will produce
errors to the flow field calculation, resulting in numerical oscillations or
unsuccessful solution procedure.
The key point here is to define some averaged mesh velocity wn on the
faces and embed it into the numerical scheme such that the GCL is exactly
satisfied. Our task here is to evaluate the volumetric increments in an exact
manner in different flow fields. In order to simplify the notation, we first
consider the volumetric increments in a time step t .
First we define the nodal mesh velocity as

t +t
ri t ri
wi =
t
Thus, the moving nodal coordinates can be expressed as
ri = t ri + t w i , [0,1]
In two-dimensional cases, two points define a face, say 1 and 2, and we

use the local coordinate to represent the line
w = w1 + ( w 2 w1 )
r = r1 + (r2 r1 ), [0,1]
Using these transformations we can obtain the volumetric increment
1 1
V (t ) = t w iSd d = tw 0 i t+t 2 S
0 0
where w 0 = (w1 + w 2 ) 2 .
In the axisymmetric case, similarly, we can obtain
1 1
V (t ) = t yw iSd d
0 0
t t+t 2
= (2 y1 + t+t 2 y2 )w1 i t S
6
t t+t 2
+ (2 y2 + t+t 2 y1 )w 2 i t S
6
t 2 t+2t 3
+ ( y1 + t+2t 3 y2 )w1 w 2
4
In the three-dimensional case, we only need to consider triangular

surfaces. Let points 1, 2 and 3 define the vertexes of the triangle, the local
transformations are
w = w1 + ( w 2 w1 ) + ( w 3 w 1 )
r = r1 + (r2 r1 ) + (r3 r1 ), , , 1 [0,1]

We can then obtain
1 1 1
V (t ) = t w iS d d d
0 0 0
t 3
= tw 0 i t+t 2 S + ( w1 w 2 )i w 3
24
where w 0 = (w1 + w 2 + w 3 ) 3 . Note that the volumetric increment of a

rectangular surface is the sum of the volumetric increments of the two
triangles that form the rectangular.
Using the computed volumetric increment, we can then define our
averaged mesh velocity from time t to time t + Dt on the interface
V (t )
w n =
t t +tS
This averaged mesh velocity computed on each face will be used

everywhere on that face regardless of how other variables are computed.
The GCL is therefore always satisfied.
6.2.2 Flux-Splitting method
The inviscid flux term Fn deserves special attention, not only because it
dominates the total flux in the outflow region, but also because wave
motions and discontinuities in the flow field are caused by this term. When
the viscous term is negligible, the governing equations form a hyperbolic
system.
Consider the following one-dimensional hyperbolic equation
U f (U ) f i+1 2 f i1 2
=
t x x
The problem is to evaluate the numerical fluxes f at the positions i + 1 2

and i 1 2 . An upwinding scheme can be constructed as

1
fi+1 2 = ( f i + f i+1 ) f i+ 1 2 (U i+1 U i ) (6.3)
2
If the wave speed, represented here by f ' , is properly evaluated such that
fi+1 f i fi+ 1 2 (U i+1 U i ) (6.4)
Eq.(6.3) becomes
fi if f i+1 2 > 0
fi+1 2 =
fi+1 if f i+ 1 2 < 0
This gives the solution of the so-called Riemann problem and a program
using the above scheme is called a Riemann solver.
For a full Eulerian equation system, approximate solutions of the
Riemann scheme are widely used. The method used here is based on Roes
scheme. In this scheme, a set of averaged variables is used to approximate
the Jacobian matrix A (U) = Fn (U ) U , such that Eq.(6.4) is exactly
satisfied.
Considering any face shared by two control volumes, denoted as states
L and R that refer to the left and right volumes respectively, the Roe
average variables are given by
= L R
= L L + R R
, = ( v, H )
L + R
Other variables are computed as functions of the Roe averaged variables.

For example, the conservative variables and flux increments are computed
by

= U = PDP
1U
U v , F = A

( 2 )
1 v i v + H

where
U = U R U L
F = Fn (u R w n , U R ) Fn (uL w n , U L ) Fn R Fn L
f = f (U
), f = A, P, D and P 1
It is important to note that the mesh velocity is always computed

according to GCL along the interface. The averaged inviscid flux on an
interface is therefore computed by
1 1 1
Fn = Fn L + Fn R P D
P U (6.5)
2 2
where
D = diag { u + c , u c , u , u , u }
u = u wn
6.2.3 Finite element method
We use the finite element method to compute the viscous terms. For
both the triangular element and the tetrahedral element, the viscous terms
can be calculated exactly along the control volume faces. Thus we have
G dS G S
S
n n
For example, using the element shape functions hi , the heat flux is
computed as
S
qi dS = k nihi i dS =S k ni hi i
S

Here, of course, the temperature must be evaluated using conservative

variables while the global matrix system is assembled. The interpolation
functions and their computations are detailed in Chapter 10.
6.2.4 Solution procedure
The procedure that we have developed for the calculation of high-speed

compressible flows is basically the same as that for incompressible flows
when Newton-Raphson method is used. In forming the linear equations for
high-speed compressible, the following steps are performed, where N e is
the number of elements, and n f and nc are the numbers of sub-volumes
and sub-faces, respectively, within the element,
for ie = 1,2,..., N e

V , hi = ...

for i f = 1, 2,..., n f

S , w n = ...

f = S ( Fn + G n ), f U = ...

( f , f U ) left control volume

( f , f U ) right control volume

for ic = 1, 2,..., nc

{ ( }
t ) t +t (V U ) t (V U ) VC ...
1

6.2.5 Explicit time integration scheme
Explicit time integration of the method described in the previous

sections is available for high-speed compressible flows. However, it cannot
be used with any moving boundary conditions.
The time step size is restricted by the well-known Courant-Friedrichs-
Lewy (CFL) stability condition

x
t min t c
u + c + 2 x
where x is the element size. The explicit method should always be

associated with the CFL automatic time stepping option (see Chapter 12 for
details), which allows the program to determine the sub-time-step size for a
specified factor CFL
ti = CFL tc
In order to keep the scheme stable, the factor CFL must be smaller than 1,
usually about 0.8.
The explicit procedure is summarized here. We start with the initial
guess of the solution U, which is the solution at time t . In the first time
step, the initial condition is used. The following procedure is used to obtain
the solution at t + t : Start with the current solution time ti = t ,
until (ti t + t )
U 0 = ti U , V 0 = ti V and compute ti

for ie = 1, 2,..., N e

V , hi = ...

for i f = 1, 2,..., n f

S , w = ..., f = S (F + G )

n n n

( f , f ) (left , right ) control volumes

for ic = 1, 2,..., nc

{

U = V 1 V 0U 0 t

( i (Fn + G n )S CV

)}

ti = ti + ti , U = U , V = V
ti ti

6.3 Control volumes and finite elements
6.3.1 Control volume in triangular elements (3-node)
A control volume for two-dimensional planar and axisymmetric flows

is obtained based on the triangular elements as defined in the figure below.
The 3-node triangular element is detailed in Chapter 10.
control volume
3
1
triangular elements
2
Figure 6.1 Definition of a two-dimensional control volume
Each triangular element is divided into three sub-control volumes with

nodes 1, 2 and 3 respectively, by connecting the center point to the
midpoints of the sides. These are three internal faces that are shared by the
three pairs of sub-control volumes. The six other element faces, every two
of them forming a side of the triangle, may be internal faces shared with
another part of the same control volume, or physical boundary sides that are
on the boundary of the computational domain.

6.3: Control volumes and finite elements
3
interface of sub-control volumes
sub-control volumes
L and R
S
L R
1
2
Figure 6.2 Sub-control volumes and their interfaces within one

triangular element
All conservative variables are defined at the corner nodes, representing

averaged quantities over control volumes. A straightforward computation
of flux terms is carried out on the internal faces and assembled in the two
control volumes that share the face.
On a physical boundary face, the flux term will either be computed in
conjunction with the applied boundary condition, or the boundary condition
is used to replace the corresponding equation defined at that node,
depending on the type of condition.
Non-physical boundary faces are skipped in the computation loop since
the flux is cancelled out.
In computing viscous terms, the coordinates and all primitive variables

are interpolated by
3
f = hi f i f = x, v, p and
i =1
where, the interpolation functions are the same as used for incompressible
flows and defined in Chapter 10.

6.3.2 Control volume in tetrahedral elements (4-node)
A similar control volume for three-dimensional flows is obtained based

on tetrahedral elements as defined in the figure below. The 4-node
tetrahedral element is detailed in Chapter 10.
interface of
sub-control volumes
L and R
3
L
S
1
R
Figure 6.3 Sub-control volumes and their interfaces within one

tetrahedral element
Each tetrahedron is divided into 4 sub-control volumes containing the

nodes 1, 2, 3 and 4, respectively, by connecting the center point of the
tetrahedron and the center points of the four triangular surfaces. The
triangular surfaces are subdivided by connecting their center points to their
corresponding midpoints of the sides. There are a total of six internal faces
shared by the six pairs of sub-control volumes. The twelve other element
faces, every three of them forming a triangular surface of the tetrahedron,
may be internal faces that are shared with another part of the same control
volume, or physical boundary faces that are part of the boundary of the
computational domain.
All conservative variables are defined at corner nodes, representing
averaged quantities over control volumes. Similar to the two-dimensional

6.3: Control volumes and finite elements
case, the computation of flux terms is only carried out on the internal faces
and the physical boundary faces.
Regarding the computation of viscous terms, the coordinates and all
primitive variables are interpolated using
4
f = hi f i f = x, v, p and
i =1
where, the interpolation functions are the same as used for incompressible
flows and defined in Chapter 10.
6.4.1 General descriptions
Using proper boundary conditions is an essential step towards obtaining

the solution of the fluid flow problem. The boundary conditions strongly
depend on the flow types.
In high-speed compressible flows, the velocity, pressure and

temperature are strongly coupled and must be treated like one variable as
far as the boundary conditions are concerned.
Generally, the unsteady viscous equations for compressible flows form

a hybrid parabolic-hyperbolic system. Problems to be solved based upon
such a system are boundary-value problems. Only a few special
mathematical results are known to ensure the existence and uniqueness of
the solution. This is contrary to the incompressible flow case for which the
mathematical theory is well established.
On the other hand, the Euler equations consist of a hyperbolic system,
for which the theory about boundary conditions has been established for
decades. The special boundary conditions developed here are mainly based
on the theory for inviscid flows. Practically speaking, a boundary condition
set that is suitable for inviscid flows may be a good first consideration for
viscid flows as well. We will use the following guidelines in the discussion
of our special boundary conditions. Let the inviscid flux (Eq.(6.5)) on the

boundary be equal to that in the right control volume (representing the

boundary volume)
1 1 1
Fn R = Fn L + Fn R P D
P U (6.6)
2 2
Defining
(p + cu ) (2c 2 )
1
2 (p cu ) (2c 2 )

= P 1U = p c 2
3
( v i )
4 1

5 ( v i 2 )
we have the characteristic forms of Eq.(6.6)
(di + di )i = 0, i = 1, 2,3, 4,5 (6.7)
where di are the eigenvalues defined in D, namely
d1 = u + c, d 2 = u c, di = u ( i = 3, 4,5 )
Eq.(6.7) states that the characteristic equation i must be zero if its

corresponding eigenvalue di is positive. The condition 4 = 5 = 0 is
trivial, which corresponds to the zero derivative of the tangential velocity
with respect to the normal direction. d1 , d 2 and d3 are the three wave
speeds along the characteristic lines. According to the signs of these wave
speeds, flows are classified into different regimes, namely supersonic at
inlet, subsonic at inlet, subsonic at outlet and supersonic at outlet, as
illustrated in the following figure. Boundary conditions strongly depend on
the status of flow regimes on the boundaries.

supersonic subsonic subsonic supersonic

at inlet at inlet at outlet at outlet
Du<-c -c<Du<0 0<Du<c Du>c
-c 0 c u
Figure 6.4 Definition of flow regimes in high-speed

compressible flows
The most frequently encountered boundary is probably the solid wall

boundary, which can be fixed, with prescribed motions or deformable
structures. Along a slip-wall, which is usually used in Euler equations, the
normal velocity is specified as the speed of the solid, while the tangential
velocities are left free to be determined by the solution. Along a no-slip
wall, all velocity components are prescribed to be equal to the velocity of
the solid. Thermal conditions on walls are usually either specified
temperature or adiabatic conditions. Since the relative velocity u is zero,
only the first wave speed is positive and, therefore, the additional condition
is provided by 1 = 0 .
Besides walls, other boundaries are exposed to the environment and
called open boundaries. A flow is called external if a part of the open
boundary is infinite. Otherwise it is called internal. Typical examples of
external flows are aerodynamic flows around aircrafts, space shuttles,
missiles, etc. Internal flows are widely visible in instruments that deal with
air.
In internal flows, an open boundary can be an inlet or an outlet if the
outward velocity is negative or positive, respectively. When a flow is
supersonic at an inlet, all wave speeds are negative, which corresponds to
incoming waves from outside of the computational domain. The boundary
values of the fluid variables are then completely determined by these
waves. Therefore three values must be specified on the boundary (see the
figure below).

t boundary
n
computational outside of
domain computational domain
Figure 6.5 Illustration of the supersonic condition at inlet
When a flow becomes subsonic at an inlet, two wave speeds are

negative while one ( d1 ) is positive. The negative wave speeds correspond
to two incoming waves from outside the computational domain while the
positive speed indicates that one wave leaves the domain. Therefore two
values must be specified on the boundary (see the figure below). The
additional condition is given by 1 = 0 .
t boundary
Du Du-c
n
domain computational
domain
Figure 6.6 Illustration of the subsonic condition at inlet

When a flow is subsonic at an outlet, only one wave speed is negative.

Therefore one value is expected on the boundary to represent the incoming
wave (see the figure below). This condition is sometimes called an outlet
control condition. Typically a pressure value is specified in most internal
flows. Corresponding to the two positive waves, 1 = 3 = 0 are used.
t boundary
n
domain
Figure 6.7 Illustration of the subsonic condition at outlet
No condition is required at the outlets where the flow becomes

supersonic since all waves leave the computational domain. In this case,
the boundary values are completely determined by the solution inside the
domain (see the figure below).

t boundary
n
domain
Figure 6.8 Illustration of the supersonic condition at outlet
In some situations where the required number of conditions cannot be

fulfilled, a zero first derivative condition can be used to replace it. This is
typically used along the outlet of subsonic external flows.
A typical set of fluid domain and associated boundary conditions for
external flows is shown in the following figure, where the whole boundary
S of the computational domain has been divided into the entrance Se
(where an external condition is applied), the boundary that bounds the solid
body S w (where a wall condition is applied) and the tail of the flow St
(where a uniform condition is applied). The Si here represents the interior
part of that part of the boundary, while the over bar indicates that it
contains the boundary of Si as well. Therefore, the common part of Si and
S j is the empty geometric set if i j , while Si and S j share their
common frontier where both conditions are applied. This boundary
partition can be expressed as
Se S w St = S = S
Si S j = i j; i, j = e, w, t

n
uniform flow
condition on St
v
n
external
condition
(v, , ) on Se wall condition on Sw
Figure 6.9 A typical boundary condition set for external high-

A similar boundary condition set for internal flows is shown in the

following figure, where the whole boundary S of the computational
domain has been divided into the inlet boundary Si (where an inlet
condition is applied), the wall boundary S w (where a wall condition is
applied) and the outlet boundary So (where an outlet condition is applied).
They also form the whole boundary and have no shared common part
except the locations where two boundaries meet each other.
inlet-subsonic, outlet-subsonic,
specified v and H on S i specified p on So
n
n
wall Sw
Figure 6.10 A typical fluid boundary condition set for internal

high-speed compressible flows

From these conditions, we observe that:
The type of the condition follows the flow regime on the boundary.
This observation is a fundamental guideline for specifying boundary

conditions in most high-speed compressible problems. Unlike the
conditions for incompressible flows, there is no default condition that is
implicitly applied to boundaries. Every boundary must be assigned a
condition.
Some variables on boundaries can directly be imposed if they do not or
only weakly depend on other solution variables. This type of condition is
referred to as a usual boundary condition. It is applied to the governing
equations in a simple manner. The conditions are either prescribed solution
variables (where the variables are forced to be the prescribed values and the
corresponding equations are removed) or forces, which are directly added
to the fluid equations as external boundary forces. However, caution is
necessary in using only the usual boundary conditions, since the set of
properly defined boundary conditions should describe somewhere a
complete flow condition rather than only a single variable. Therefore
special conditions are strongly recommended in high-speed compressible
flows.
Special boundary conditions may be more complicated or strongly
depend on solution variables and additional "boundary" equations must be
solved. They are discretized and assembled to the global system of
equations. Special boundary conditions play important roles in high-speed
compressible flows. In fact, the usual boundary conditions are rarely used.
Unless noted otherwise, special boundary conditions can only be
applied to geometries of one dimension lower than the computational
domain. More specifically, in 2D models, the geometries are lines or
nodes that can be connected to lines and in 3D models, the geometries are
surfaces or nodes that can be connected to surfaces. These geometries
may be located on the boundary or inside the domain, depending on the
type of boundary conditions.
Most boundary conditions impose only constraints on fluid variables
(velocity, pressure, temperature, etc.). Some other conditions impose
additional constraints on the boundary positions or boundary

displacements. These conditions are called kinematic conditions or moving

boundary conditions. When a moving boundary condition is applied, the
boundary nodal displacements are computed or specified. The interior
nodal positions are adjusted in an arbitrary way to ensure the validity and
quality of the mesh. In this case, an arbitrary Lagrangian-Eulerian (ALE)
coordinate system is used in the fluid model.
6.4.2 Usual boundary conditions for fluid
6.4.2.1 Prescribed specific discharge
In this condition, a time-dependent specific discharge is directly

prescribed
vi = vi ( t )
and applied to boundaries. The xi -momentum equations at the boundary

nodes are then replaced by this condition.
Be cautious when using this condition since there is no temperature
condition applied yet. This boundary condition alone does not give a well-
posed problem.
6.4.2.2 Zero specific discharge
Physically, this condition is equivalent to a prescribed specific

discharge. However, this condition removes the degree of freedom of the
specific discharge component.
Be cautious when using this condition since there is no temperature
condition applied yet. This boundary condition alone does not give a well-
posed problem.
6.4.2.3 Prescribed density
In this condition, a time-dependent density is directly prescribed
= (t )

and applied to the boundaries. The continuity equations at boundary nodes

are then replaced by this condition.
Be cautious when using this condition since there are no velocity and
temperature conditions applied yet. This condition alone does not give a
well-posed problem.
6.4.2.4 Concentrated force load
In this condition, a prescribed time-dependent force F ( t ) is directly

applied to the boundary. The force is added to the momentum equations of
the boundary nodes. Note that the force can be directly applied to nodes.
This force is also called an external nodal force.
This condition has no effect on the nodes where the specific discharge
condition is prescribed. Recall that in this case the momentum equation has
been replaced by the velocity condition.
Be cautious when using this condition. It functions like an additional
force load to the momentum equation. It does not provide any information
on the boundary variables. Therefore, to reach a well-posed boundary
condition set, this condition should not be included in that set. In other
words, another properly defined condition set must be applied to the same
boundary.
6.4.2.5 Distributed normal-traction loads
This condition can only be applied to boundary lines and surfaces,

respectively, of two-dimensional and three-dimensional computational
domains. In this condition, a time-dependent normal stress (called normal-
traction) nn ( t ) = n i i n is prescribed. The applied nodal force is then
computed by
F ( t ) = hv nn ( t ) dS
where hv is the virtual quantity of velocity on the boundary.

Note that the normal stress consists of the pressure and the normal shear
stress. Along open boundaries, the normal shear stress is usually negligible
compared with the pressure. Therefore, a normal-traction is usually applied
to open boundaries where the pressure is known. In particular, when

nn ( t ) = 0 , the application of the normal-traction load is trivial because it

can be reached by simply not specifying the condition.
This condition has no effect on the nodes where a normal specific
discharge condition is prescribed. Recall that in this case, the normal
momentum equation has been replaced by the normal specific discharge
condition.
Similar to a concentrated force condition, caution is necessary when
using this condition. It functions like an additional force load to the normal
momentum equation. It does not provide any information on the boundary
variables. Therefore, to reach a well-posed boundary condition set, this
condition should not be included in that set. In other words, another
properly defined condition set must be applied to the same boundary.
6.4.2.6 Field centrifugal loads
As described in Section (2.3), when the whole field is rotating, the flow
equations can be formulated in a rotational reference coordinate system.
The problem is then equivalent to a problem defined in a fixed coordinate
system subjected to additional centrifugal forces. These forces can be
written as
F ( t ) = hv fc dV
where hv is the virtual quantity of velocity and fc is defined in Chapter 2

and rewritten here, for the purpose of indicating the prescribed parameters
in the force
fc = x0 (t ) (t ) ( x x0 (t )) 2(t ) ( v)
(t ) (t ) ( x x0 (t ))
It must be kept in mind that the velocity solved in this coordinate system is
the relative velocity with respect to the rotational frame (which is denoted
as v r in Section 2.3).
Strictly speaking, this condition is not a "boundary" condition since the
force is applied to the whole domain.

6.4.2.7 Prescribed total energy
In this condition, a time-dependent total energy can be directly

prescribed and applied to boundaries:
E = E (t )
The energy equations at the boundary nodes are then replaced by this
condition.
Be cautious when using this condition since there are no density and
velocity conditions applied yet. If only this condition is applied to a
boundary, the problem may not be well-posed.
6.4.2.8 Concentrated heat flow load
In this condition, a prescribed time-dependent heat flow load Q ( t ) is

directly applied to the boundaries. The load is added to the energy
equations of the boundary nodes. Note that the heat flow loads can directly
be applied to nodes.
This condition has no effect if applied to the boundary where a total
energy is prescribed. Recall that in this case, the energy equation has been
replaced by the prescribed total energy condition.
Be cautious when using this condition. It functions like an additional
heat flow load to the energy equation. It does not provide any information
on the boundary variables. Therefore, to reach a well-posed boundary
condition set, this condition should not be included in that set. In other
words, another properly defined condition set must be applied to the same
boundary.
6.4.2.9 Distributed heat flux load
The condition can only be applied to boundary lines and surfaces, of

respectively. Distributed heat flux load is equivalent to a concentrated
force load obtained by specifying the normal component of the external
heat flux qn ( t ) imposed onto the fluid domain. The applied heat flow is
then computed by

Q ( t ) = h qn ( t ) dS
where h is the virtual quantity of temperature on the boundary. In

particular, when qn ( t ) = 0 , the application of the heat flux is trivial because
it can be reached by simply not specifying the condition.
Be cautious when using this condition. It functions only like an
additional heat flux load to the energy equation. It does not provide any
information on the boundary variables. Therefore, to reach a well-posed
boundary condition set, this condition should not be included in that set. In
other words, another properly defined condition set must be applied to the
same boundary.
6.4.3 Special boundary conditions
A special boundary condition is a combination of a few boundary

values specified together. Every value can either represent a Dirichlet
condition (prescribed variable value), or a Neumann condition giving the
zero value of first derivative. Let us denote any variable by
( = u , p, , H , e,... ). We use the following notation to represent the
possible conditions
= A(t ) + B (t )i
where i is the value of the same field solution variable adjacent to the
boundary.
When the variable is prescribed, choose B (t ) =0 and specify A(t ) to be
the prescribed value of the variable.
On the other hand, when a zero Neumann condition is applied, choose
A(t ) =0 and B (t ) =1. This means that a first order approximation to the
Neumann condition will be used.
This notation is used for most of the variables that are required in the
special boundary conditions for high-speed compressible flows.

6.4.3.1 Fixed wall
In the interface of a fluid and a fixed solid, no-slip or slip conditions are
usually applied. On fixed wall conditions, the boundary is fixed. In other
words, the boundary displacement is zero. Fixed wall conditions can only
be applied to boundary lines and surfaces of two-dimensional and three-
dimensional computational domains, respectively.
No-slip condition
When a no-slip condition on fixed walls is applied, the fluid velocity

vector on that wall is prescribed to be zero
v=0
For the temperature, two types of conditions are usually considered:

either the wall temperature is prescribed ( = (t ) ) or an adiabatic wall
condition is applied ( ni = 0 ). These two conditions can be specified by
choosing ( A, B ) = ( (t ), 0) and ( A, B ) = (0,1) respectively.
v=0;
_
=(t) or n =0
boundary is fixed
Figure 6.11 No-slip condition on fixed walls for high-speed

compressible flows
This condition is usually applied to wall boundaries in viscous flows.

Slip condition
In this condition, the normal component of the velocity vector is

prescribed to be zero
v in = 0

variables through the governing equations.
The temperature condition can be either a prescribed temperature or a
specified zero heat flux, in the same manner as in the no-slip condition.
v n=0;
_
=(t) or n =0
boundary is fixed
Figure 6.12 Slip condition on fixed walls for high-speed

compressible flows
This condition is usually applied to wall boundaries when the Euler

equations are solved (i.e., when the viscous effects are negligible).
6.4.3.2 Angular velocity
The condition is used to prescribe all velocity components in a special

manner: by means of a time-dependent angular velocity ( t ) (see figure
below). This condition can be applied to fixed or moving boundaries,
where the coordinates of the boundary nodes vary with time. The fluid
velocity is then computed as

v(t ) = ( t ) ( x(t ) x0 )
where x0 is the center of rotation and x(t ) denotes the boundary

coordinates.
The temperature condition can be prescribed or the zero heat flux is
specified, in the same manner as in the no-slip wall condition.
_ _
x moving or fixed
boundary nodes
_
x
_
(t)
Figure 6.13 Angular velocity condition for high-speed

compressible flows
6.4.3.3 Moving wall
This condition broadens applications of the fixed wall condition to

moving wall conditions. In this condition, the boundary displacement d(t )
is prescribed while the fluid velocity is computed based on the
displacement. This implicitly applies the Lagrange formulation to fluid
boundary nodes. A moving wall condition is therefore a moving boundary
condition. This condition can only be applied to boundary lines and
surfaces of two-dimensional and three-dimensional computational domains,
respectively.

No-slip condition on moving walls
The no-slip condition on a wall implies a consistency between the fluid

velocity and the solid velocity
v d(t ) = 0

_
_ t)
v=d(
=(t) or n =0
_
is prescribed: d(t)
Figure 6.14 No-slip condition on moving walls for high-speed

compressible flows
This condition is usually applied to moving wall boundaries in viscous

flows. The dominant motion of the moving wall must be in the normal
direction of the boundary. If the tangential movement is large, the moving
wall conditions with specified tangential or rotational velocity should be
more adequate.
Slip condition on moving walls
In slip wall conditions, only the normal component of the velocity is

enforced to be consistent with the solid counterpart

(v d(t ))in = 0
variables from the governing equations.
This condition is usually applied to moving wall boundaries where the
viscous effects are negligible.
_
v n=d(
_ t) n
=(t) or n =0
_
is prescribed: d(t)
Figure 6.15 Slip condition on moving walls for high-speed

compressible flows
In no-slip and slip moving wall conditions, the Lagrangian coordinate

system has been implicitly applied to both the normal and tangential
directions of moving boundaries. These conditions can present difficulties
with moving meshes if the tangential displacement of the moving wall is
large. The use of a Lagrangian coordinate system in the tangential
direction results in largely skewed meshes. On the other hand, the use of
the Lagrangian formulation to that direction is not absolutely necessary.
Most tangential movements do not change the shape of computational fluid
domains. Therefore, an ALE coordinate system can be used. In particular,
the displacement in the tangential direction can be completely avoided if
the Eulerian formulation is applied to that direction.

In this ALE formulation, both the tangential velocity and the

displacement vector are prescribed. They may or may not be consistent in
the tangential direction. When they are consistent, the Lagrangian
formulation is implicitly used. When the tangential displacement is zero,
the Eulerian formulation is then specified.
The new formulation is not only advantageous for some cases to reach
less distorted meshes, but is also necessary for other cases where otherwise
a valid mesh cannot be maintained. Two such formulations are introduced
here.
No-slip condition of moving walls (type=tangential)
In this condition, the displacement d(t ) is prescribed. This

displacement causes the boundary to move. The normal fluid velocity is
then computed based on the normal displacement. The tangential velocity
is prescribed separately (see the next figure). This condition is
v = v (t ) + d(t )i(I )
where v (t ) is the prescribed tangential velocity and is the tangential

direction computed by means of a specified direction a and the normal
direction on the boundary n
= a n
Notice here that the tangential direction is not directly required, since it is,
in general, not a constant vector.

_ _
_
_ =(t) or n =0
_
is prescribed: d(t)
Figure 6.16 Moving wall condition (type=tangential) for high-


No-slip condition of moving walls (type=rotational)
When the dominant boundary motion is determined by a rotation (see

the next figure), we can apply an ALE formulation to that direction while
we use the Lagrangian coordinate system for the other directions. In this
case, the displacement vector and the angular velocity is prescribed. The
velocity vector is then computed as
v(t ) = (t ) (x(t ) x0 ) + d(t )i(I )
where x0 is the center coordinate of the rotation, is the angular velocity

of the solid wall and the tangential direction is computed as
(t ) (x(t ) x0 )
=
(t ) (x(t ) x0 )

_ _ _
v=(t) (x(t)x0)+d(t) (I)
_
=(t) or n =0
x
_
is prescribed: d(t)
_
x0
_
(t)
Figure 6.17 Moving wall condition (type=rotational) for high-


6.4.3.4 External flow
Three boundary values and the direction of the specified velocity are
required in this condition. These values can be ( (t ), v (t ), M(t ) ) or
( (t ), (t ), v (t ) ), where v is the magnitude of the velocity in a specified
flow direction m .
In addition, a velocity condition imposed on the component in the
direction k can also be specified, where k is tangential to m . Let
vk vik and the coordinate in the normal direction n be n , this
condition can then be either vk = 0 (represented as s =0) or vk n = 0
(represented as s =1).
Based on these conditions, we can obtain a complete description of the
conservative variables.

specified _
flow direction m
Figure 6.18 External flow condition
In the case that ( (t ), v (t ), M(t ) ) is specified, we have

U= (v m + svi k )

1 (v 2 + sv 2 ) + v 2 M 2
2 i
In the case that ( (t ), (t ), v (t ) ) is specified, we have

U= (v m + svi k )

1 (v 2 + sv 2 ) + C
2 i v

respectively.

6.4.3.5 Supersonic at inlet
Three boundary values are required in this condition. These values can
be ( (t ), u (t ), M(t ) ), ( p (t ), (t ), u (t ) ) or ( p (t ), u (t ), M(t ) ), where u is the
magnitude of the velocity in the normal direction of the boundary n .
tangential direction can also be specified. Let v v i and the
coordinate in the direction n be n , this condition can be either v = 0
(represented as s =0) or v n = 0 (represented as s =1).
Based on these conditions, we can obtain a complete description of the
conservative variables.
Figure 6.19 Sketch for inlet and outlet flow conditions
In the case that ( (t ), u (t ), M(t ) ) is specified, we have

U= (un + svi )

1 (u 2 + sv 2 ) + u 2 M 2
2 i
In the case that ( p (t ), (t ), u (t ) ) is specified, we have

p (C p Cv ) ( )

U = (un + svi )

1 (u 2 + sv 2 ) + C
2 i v
In the case that ( p (t ), u (t ), M(t ) ) is specified, we have
pM 2 u 2 ( )

U = (un + svi )

1 (u 2 + sv 2 ) + p
2 i

respectively.
6.4.3.6 Subsonic at inlet
Two boundary values are required in this condition. They control the
flow condition in the normal direction to the boundary. Thus the velocity
specified in this condition is the magnitude of the velocity in the normal
direction n . Since the inlet flow condition is assumed to be subsonic, only
one wave speed ( d1 ) is positive. Therefore 1 = 0 is assumed.
tangential direction can also be specified. Let v v i and the
coordinate in the direction n be n , this condition can be either v = 0
(represented as s =0) or v n = 0 (represented as s =1).
For the purpose of convenience in the input, a few combinations of
these values can be used. They are
( p, H ), ( p, ), (, u ), (, u ), (, u ), ( H , u )

respectively.

6.4.3.7 Subsonic at outlet
One boundary value is required in this condition. It controls the flow

condition in the normal direction of the boundary. Thus the velocity
specified in this condition is the magnitude of the velocity in the normal
direction n . Since the outlet flow condition is assumed to be subsonic, two
wave speeds ( d1 and d3 ) are positive and therefore 1 = 3 = 0 are
assumed.
tangential direction can be specified. Let v v i and the coordinate
in the direction n be n , this condition can be either v = 0 (represented as
s =0) or v n = 0 (represented as s =1).
Most frequently a pressure condition is specified, although other
conditions are also applicable. The available boundary values that can be
specified are p, , , and e .

respectively.
6.4.3.8 Supersonic at outlet
When an outflow becomes supersonic, all wave speeds are positive, so

all i (i=1,2,3,4,5) are zero. This indicates that no boundary values are
required under this flow condition. The solutions of i = 0 are simply
equivalent to U = 0 . This is the same as specifying zero first derivatives
for all solution variables.
Although this type of condition is suitable for supersonic flows at
outlet, it is also frequently used as a uniform flow condition where all
variables are extrapolated.
respectively.

6.4.3.9 Symmetry boundary conditions
This condition is usually applied to the boundary where the flows on

both sides are symmetric (another side is not modeled). The conditions are
f
u = 0, =0 (f = p, and v un )
n
No parameters are required in this condition.

respectively.
6.4.3.10 Fluid-structure interface
The velocity and pressure conditions on a fluid-structure interface are

the same as for incompressible flows (see Section 3.4 for details). The
temperature condition can be either prescribed temperature or the zero heat
flux is specified in the same manner as in the no-slip wall condition.
respectively.
6.4.3.11 Gap
The velocity and pressure conditions are the same as in an

incompressible flow case (see Section 3.4 for details). If the status of the
gap is closed, an adiabatic wall is further assumed.
respectively.
The application of this condition is the same as in an incompressible

flow case (see Section 3.4) for details. The only difference is that
conservative variables are used here. That is, in the case I var = 1 , is
defined as ( u , v, w, , E ) .

Although the conservative variables are used in the solutions, the

primitive variables are still used in specifying initial conditions for the
convenience of input.
Recalling that pressure, temperature and density must satisfy the state
equation represented by the ideal gas law, zero pressure or zero temperature
must not be specified. Since the default initial conditions to all variables
are zero, actual initial conditions are always required in both transient and
steady-state analyses. Good initial conditions, as will be seen in Chapter
13, may accelerate the convergence during equilibrium iterations. In
certain cases, good initial conditions may become the key factor in
obtaining converged solutions.
solutions mapped from similar solutions.
6.6 Material models
6.6.1 Constant material model
This is the simplest yet most frequently used material model. In this
model, all fluid properties are assumed to be constant. They are
= fluid viscosity
2* = ratio of the second viscosity to the viscosity
k = thermal conductivity

6.6.2 Power-law model
This model contains the modifications of the viscosity and the heat
conductivity defined in Sutherlands formulae
m m
+ S + Sk
k
= 0 , k = k0 k

+ S k + Sk
The required fluid properties are
0 = fluid viscosity at temperature

= reference temperature used for calculating
S = constant temperature used for calculating
m = exponent used for calculating
k0 = thermal conductivity at temperature k
k = reference temperature used for calculating k
Sk = constant temperature used for calculating k
mk = exponent used for calculating k
2* = ratio of the second viscosity to the viscosity
B
q = rate of heat generated per unit volume
6.6.3 Temperature-dependent model
This model contains the same parameters as those in the power-law

model. However, the following parameters in this model depend on
temperature
0 , C p , Cv , k0 , q B = f ( )

6.6.4 Pressure-dependent model

pressure
0 , C p , Cv , k0 , q B = f ( p)
6.6.5 Pressure-temperature-dependent model

pressure and temperature
0 , C p , Cv , k0 , q B = f ( p, )
The input of these data assumes the following format:
f ( p1 ,11 ) f ( p1 ,12 ) ... f ( p1 ,1n1 )

f ( p , ) f ( p2 , 22 ) ... f ( p2 , 2 n2 )
2 21

... ... ... ...
f ( pm , m1 ) f ( pm , m 2 ) ... f ( pm , mnk )
where m is the number of pressure points and ni (i = 1, 2,..., m) is the

number of temperature points at the pressure point pi .
In this model, the user can specify the material data through a user-
supplied subroutine, which is designed specifically for high-speed
compressible flows. The subroutine will be invoked whenever material
data are required in the computations. Current pressure and temperature
will be passed in as arguments, while the program returns the following
fluid properties

f f
, k , q B , f , ,
p
( f = Cv & C p )
The following example demonstrates the use of a user-supplied material

data set for Re = 1000, = 1.4, Pr = 0.71 and the definitions
1 1 C p
= , Cv = , C p = 1 + Cv , k = , qB = 0
Re 1 Pr
SUBROUTINE CFVUSR(P,T,CMAT)
C======================================================================
C
C USER-SUPPLIED MATERIAL DATA FOR COMPRESSIBLE FLUID FLOWS
C
C INPUT : P,T
C OUTPUT: CMAT
C
C ------------------------------------------------------------------
C WHERE
C P = PRESSURE AT THE LOCATION
C T = TEMPERATURE AT THE LOCATION
C CMAT(1) = SPECIFIC HEAT AT CONSTANT PRESSURE (CP )
C CMAT(2) = DERIVATIVE OF CMAT(1) WITH RESPECT TO P (CPP)
C CMAT(3) = DERIVATIVE OF CMAT(1) WITH RESPECT TO T (CPT)
C CMAT(4) = SPECIFIC HEAT AT CONSTANT VOLUME (CV )
C CMAT(5) = DERIVATIVE OF CMAT(4) WITH RESPECT TO P (CVP)
C CMAT(6) = DERIVATIVE OF CMAT(4) WITH RESPECT TO T (CVT)
C CMAT(7) = VISCOSITY
C CMAT(8) = COEFFICIENT OF THERMAL CONDUCTIVITY
C CMAT(9) = RATE OF HEAT GENERATED PER UNIT VOLUME
C
C BE NOTED THAT THE STATE EQUATIONS ARE DEFINED AS
C
C P = (CP-CV)*D*T
C T = E/CV
C
C WHERE D = DENSITY
C E = INTERNAL ENERGY
C
C======================================================================
DIMENSION CMAT(*)
C----------------------------------------------------------------------
P = P
T = T
C----------------------------------------------------------------------
C
C FOR A PERFECT GAS, USING T_SCALE = (U_SCALE)**2/(CP-CV), THE
C MATERIAL DATA ARE REPRESENTED BY REYNOLDS NUMBER "RE", PRANDTL
C NUMBER "PR" AND THE RATIO OF SPECIFIC HEATS "GAMMA".
C
C XMU = 1/RE

C XKCON = XMU*CP/PR
C CV = 1/(GAMMA-1)
C CP = 1 + CV
C
C----------------------------------------------------------------------
RE = 1000.D0
PR = 0.71D0
GAMMA = 1.4D0
C
XMU = 1.D0/RE
CV = 1.D0/(GAMMA-1.D0)
CP = 1.D0 + CV
CMAT(1) = CP
CMAT(2) = 0.D0
CMAT(3) = 0.D0
CMAT(4) = CV
CMAT(5) = 0.D0
CMAT(6) = 0.D0
CMAT(7) = XMU
CMAT(8) = XMU*CP/PR
CMAT(9) = 0.D0
C----------------------------------------------------------------------
C FOR A REAL GAS, SUPPLY THE FUNCTIONS HERE AND COMMENT OUT THE
C UPPER SECTION
C----------------------------------------------------------------------
9999 CONTINUE
RETURN
END

Chapter 7: Formulations of turbulence in high-speed compressible flows
Chapter 7 Formulation of turbulence in high-speed

compressible flows
In this chapter, we introduce our two-equation turbulence model for

high-speed compressible flows.
The model introduced here can be coupled with ADINA solid models in
fluid-structure interaction analyses. They can also be associated with mass
transfer analyses.
are available for this turbulence model.
Table 7-1 Functionalities in the K - turbulence model for

high-speed compressible flows
Coupled models ADINA solid model
Note: Heat transfer heat transfer
must be coupled. mass transfer
3-node triangle
Elements
4-node tetrahedron
Material model turbulence K -
Initial conditions zero (default conditions)
specified. restart run

prescribed specific discharges

zero specific discharges
prescribed density v
field centrifugal load
prescribed total energy
Boundary conditions concentrated heat flow load E
Note: The column
furthest to the right fixed wall v, , K ,
represents the solution angular velocity v,
variables or equations
that are immediately moving wall v, , K , , d
affected by the external flow condition
specified condition. supersonic at inlet
The solution variable subsonic at inlet
d indicates a moving
, v, E
supersonic at outlet
boundary condition. subsonic at outlet
symmetric
fluid-structure interface v, , K , , d
gap
all
user-supplied
prescribed turbulence K - K,
zero flux of turbulence variables
sparse solver
equations
iterative methods (RPBCG, RPGMRES, and multi-grid)
Other capabilities skew system
pressure datum
The governing equations for the fluid flow are those of high-speed
compressible flows. However, the viscosity and heat conductivity are
modified to include the turbulence effect and the K - model is used to
model the required variables. These equations are listed in Table 2-1. Note
that the fluid solution variables are modified to include the effect of
compressibility. The solution variables are ( , v, ,, ) . In the case

of moving mesh problems, the ALE formulation described in Section 2.14

is used.
The time integration is performed as for the fluid flow equations. The
fluid equations and their discretization are described in Section 6.2.
The finite element method is also used to discretize the governing
equations of the turbulence solution variables. The finite element equations
are obtained by establishing a weak form of the turbulence equations using
the Galerkin procedure. The equations of K and are weighted with the
virtual quantities of K and , respectively. The equations are integrated
over the computational domain V . The divergence theorem is used to
lower the order of the derivatives of turbulence fluxes.
The variational forms of the K - equations are
(h G f + Q f ih f ) dV = 0
f
V
where f represents K and , h K and h are virtual quantities of K and

respectively, and
K
G K = y a + viK S K
t

G = y a + vi S
t
Q K = y aq K
Q K = y aq K

7.3 Elements
The elements that can be used for high-speed compressible turbulence

variables are the same as those for laminar model of high-speed
compressible flows.
The boundary conditions for fluid variables are usually the same as for
laminar models, although the treatment of the wall condition or its analogue
is modified as explained later in this section. The conditions introduced
here are only for the turbulence variables K and .
A typical turbulence condition may be prescribed as a turbulence
variable at the inlet, zero first derivative of the turbulence variables at the
outlet and a solid wall condition. In general, all parts of the boundary have
to be applied with one and only one turbulence boundary condition. There
is at least one wall condition or a prescribed turbulence variable condition
in steady-state analyses. The available conditions are introduced here.

In this condition, time-dependent turbulence variables are directly
prescribed
K = K ( t ) , = (t)
and applied to boundaries. The K and equations at the boundary nodes

are then replaced by these equations.
indirectly via

2
K= 3
2 (i v )
= K 2 (0.3L )
3
where, v , L and i are velocity scale, length scale and turbulence intensity
respectively.
7.4.2 Zero flux of turbulence variables
The natural boundary condition of any equation requires a specified

normal flux of that variable. The turbulence flux condition is currently
unavailable in ADINA-F (practically such conditions are not available).
Therefore, a zero flux is implicitly assumed along the boundary where no
turbulence condition is explicitly specified.
7.4.3 Solid walls
This condition is probably the most important one in modeling

turbulence. At a solid wall the no-slip condition applies so that both the
mean and fluctuating velocities are zero. In the near wall region where
0< y + <100, the wall function is applied. In this region, the Reynolds
stresses v ' v ' are nearly constant, so is the kinetic energy K. The
convection and diffusion in the K-equation are negligible so that local
equilibrium prevails. When buoyancy effects are absent, this implies that
all source terms in the K-equation are zero. We further recall the definition
of the shear stress and obtain
t (v ') 2 = , t v ' = v*2
The solutions of the above equations determine the variables K and in

the near-wall region
v*2
K= , = v*2 v ' (7.1)
c

The wall conditions are determined at a distance, yw , near the wall.

Although the location is not the same as the location of the physical wall,
we use the nodal position to store and then plot the solutions that are
actually located at yw . Applying the wall function to Eq.(7.1), we have a
complete set of conditions for the fluid flow with turbulence
v*2 v*4
v w w w = v*W ( yw+ ) , K w = , w = W '( yw+ )
c 0
where is the normalized tangential direction of the wall. The near-wall

location yw is evaluated as
yw+ = min{d w+ , 12 y + }
where d w+ is an empirical constant (initially set as 70) and y + is the

normalized near-wall element size.
Although similar research has been conducted for temperature
conditions (relating the heat flux at the wall to the difference between the
wall temperature and the temperature just outside the viscous sublayer), no
convincing results have been reported. We therefore have not included any
modification of the temperature value in wall conditions.
The near wall treatment of the fluid and the turbulence variables, as
discussed above, is automatically enforced whenever a wall condition or its
analogue is applied. These conditions are
No-slip fixed and moving wall conditions

Moving wall condition (type=tangential)
Moving wall condition (type=rotational)
No-slip fluid-structure interface
Fluid-structure interface (type=tangential)
Fluid-structure interface (type=rotational)

These conditions are defined in the same user-supplied subroutines and

are applied in the same manner as for incompressible fluid flows and
temperature conditions (see Section 3.4). The only difference is that
conservative variables are used here. That is, in the case I var = 1 , is
defined as ( u , v, w, , E) .
Recalling that pressure, temperature and density must satisfy the state
equation that is represented by the ideal gas law, zero pressure or zero
temperature must not be specified. Since the default initial condition to all
variables is zero, initial conditions are always required in both transient and
steady-state analyses. A good initial condition, as will be seen in Chapter
13, may accelerate the convergence during equilibrium iterations. In
certain cases, the initial condition may become a key factor in reaching a
converged solution.
7.6 Material models

material model: K - turbulence
All the material models that are available for high-speed compressible
flows are also applicable to the turbulence model described here. In these
material models, the parameters corresponding to fluid flows are the same
as described in Section 6.6. Additional empirically determined model
constants are required in the turbulence model. They are
c , c1 , c2 , c3 , K , , , d w+ and 0 . Their default values are

c = 0.09, c1 = 1.44, c2 = 1.92,

c3 = 0.8, K = 1, = 1.3,
= 0.9, d w+ = 70, 0 = 0.4

Chapter 8: Formulation of mass transfer in fluid flows
Chapter 8 Formulation of mass transfer in fluid flows
In this chapter, we introduce our formulation of mass transfer in fluid

flows.
The model introduced here can be coupled with all fluid models in
ADINA-F, including the models of incompressible, slightly compressible,
low and high speed compressible flows and flows in porous media. If these
fluid models are coupled with ADINA solid models in fluid-structure
interaction analyses, the mass transfer solution can also directly be used.
Solving a mass transfer problem involves the solution of the governing
equations associated with the input of physical fluid and species material
data, well-posed boundary conditions and initial conditions. The
formulation of the fluid models has been introduced in the previous
chapters. In this chapter, only the mass transfer equations are discussed.
The following table presents a quick overview of the capabilities
available for mass transfer analyses.
Table 8-1 Functionalities for mass transfer analyses
ADINA solid model
Coupled models incompressible, slightly compressible and compressible flows (including
laminar and turbulence models)
Note: Heat transfer
porous media flows
must be coupled in
compressible flow solid element groups
models. heat transfer
Electro-static and steady current conduction analyses
Galerkin Elements
constant
velocity-dependent
Material models
pressure-temperature-dependent
user-supplied

prescribed mass-ratio
Boundary conditions distributed mass flux
mass convection
user-supplied
specified
Initial conditions
restart run
sparse solver
equations
Other capabilities
others depending on the coupled fluid models.
The governing equations for mass-ratios and the bulk density are listed
in Table 2-1. In the current approach, we assume that the material data are
the same for all species. The solution variables are i ( i = 1, 2,..., n ) . Note
that the fluid model must be included since the mass transfer equations are
always coupled with the fluid flow equations. In the case of moving mesh
problems, the ALE formulation described in Section 2.14 is used.
Mass transfer may be active in the whole computational domain, or may
be inactive in certain regionsfor example, in solid element groups. The
mass-ratio can be deactivated in element groups. Once it is inactive, a zero
mass flux condition is assumed along the interface between the active-mass
and inactive mass regions.
The time integration schemes are those used for the fluid flow
equations.
The finite element method is used to discretize the governing equations.
The finite element equations are obtained by establishing a weak form of
the equations using the Galerkin procedure. The governing equations are

weighted with the virtual quantity of mass-ratios and integrated over the
computational domain V . The divergence theorem is used to lower the
order of the derivatives of the mass flux.
Corresponding to the governing equation, the variational form of the
mass transfer equation is
(h Gi + qi ihi ) dV = hi qi i dS
i
V S
where hi is the virtual quantity of i , and
i
Gi = + i( vi ) mi
t
qi = Dii
The upwinding techniques are used while constructing the finite

element equations.
In FCBI 2D 4-node and 3D 8-node elements, a finite volume method is
used. Using the related interpolation functions described in Chapter 10, the
discretized equations are
i
mi dV + ( v w ) i Dii i dS = 0
V t S
In the following cases, the mass solution variables may be coupled with
fluid variables.
any i 0, 1 i n
any i 0, 1 i n
any fluid material depends on mass-ratios
However, the choice of coupling between the fluid variables and the mass-
ratios is determined by the user. In certain applications, a decoupled
solution may be required.
The mass transfer equations are solved separately from the fluid
variables. However, in the case that they are coupled with fluid, the
iterative or partitioned procedure is applied. In the Newton-Rapson

8.2: Numerical method
method, the sparse solver is always used to solve the mass transfer
equations.
8.3 Elements
The elements that can be used for mass transfer are the same as those
for incompressible flows.
The boundary conditions for the fluid variables are the same as for fluid
models. The conditions introduced here are only for mass-ratios.
When n species are modeled, n mass transfer equations are solved.
For each species, a complete set of boundary conditions must be applied.
A typical set of boundary conditions for one species may include
prescribed mass-ratios at the inlet, and zero mass flux on the walls and on
the outlet. Also, a mass convection condition may be applied to some open
boundaries. A general guide for forming a well-posed species equation is
to apply one and only one boundary condition at each part of the boundary,
and at least one prescribed mass-ratio or mass convection condition is
applied if a steady-state analysis is performed. The available conditions are
introduced here.
8.4.1 Prescribed mass-ratio
In this condition, a time-dependent mass-ratio is directly prescribed
i = i ( t )
and applied to boundaries. The mass transfer equations corresponding to

the species i at the boundary nodes are then replaced by this equation.
This condition is usually applied at inlet boundaries.

8.4.2 Distributed mass flux
In this condition, a mass flux qi n ( t ) is imposed onto the fluid domain.

The applied mass flow is then computed by
Qi ( t ) = hi qi n ( t ) dS
where hi is the virtual quantity of mass-ratio on the boundary. The

computed mass flow Qi (t ) is directly applied to the right-hand side of the
mass transfer equation of the boundary nodes. In particular, when
qi n ( t ) = 0 , the application of the mass flux is trivial because it is equivalent
to no mass flux loads.
This condition is usually applied to boundaries where the mass flux is
known. It has no effect if applied to a boundary where the mass-ratio has
been prescribed.
8.4.3 Mass Convection
In this condition, the mass flux from an external convection condition is

specified as
(
qi n = hi i e i )
where hi is a mass convection coefficient and i e is an environmental
mass-ratio, both could being constant, time-dependent or mass-ratio
dependent. The applied mass flow is then computed by
Qi ( t ) = hi qi n dS
where hi is the virtual quantity of mass-ratio on the boundary. The mass

flow Qi (t ) is discretized and assembled accordingly into the right-hand
side and the stiffness matrix of the mass transfer equation.

This condition is usually applied to the boundary that is exposed to an

open environment. It has no effect if applied to a boundary where a mass-
ratio condition is prescribed.
This condition is described in Section 3.4.
In transient analyses, besides initial conditions for all fluid flow

variables, all mass-ratios must be specified. The default initial conditions
are zero. Although mass-ratio initial conditions are not required in steady-
state analyses, they are used as starting values in the equilibrium iterations.
Good initial conditions, as will be seen in Chapter 13, may accelerate the
convergence during equilibrium iterations. In certain cases, the initial
conditions may become the key factor in reaching a converged solution.

8.6 Material models
Fluid flow material conditions are defined when the fluid equations are
considered. In this section, only the material conditions for the species are
described. Mass material data are assumed to be the same for each of the
species.
8.6.1 Constant material data model
This model contains the following parameters:
D0 = mass diffusion coefficient

i = mass volume expansion coefficient for i-th species
i 0 = mass-ratio reference for i-th species
The mass diffusion coefficient tensor is computed by
D = D0 I
where I is the diagonal matrix.
8.6.2 Velocity-dependent model
In this model, the diffusion coefficient tensor is defined as
D = D0 I + Dnnn + D (I nn)
where n is the flow direction, D0 is a diffusion coefficient, Dn is the

diffusion coefficient in the flow direction and D is the diffusion
coefficient in the transverse direction of the flow.
The following parameters are required:
D0 = a homogeneous mass diffusion coefficient

Dn = mass diffusion coefficient in flow direction

D = mass diffusion coefficient in transverse direction
8.6.3 Pressure-temperature-dependent model
In this model, the mass diffusion coefficient matrix is defined as
np n
p
D = D0 I
p0 0
It is important to use proper units such that the pressure and temperature
solutions can never be zero.
The following parameters are required
D0 = mass diffusion coefficient at ( p0 , 0 )

p0 = reference pressure
0 = reference temperature
np = exponent of the pressure ratio
n = exponent of the temperature ratio
In this model, the user can specify the material data through a
user-supplied subroutine. The subroutine is invoked whenever material
data is required in the computations. Any necessary data is passed in as
arguments. The arguments are the coordinates of the current spatial point,
time, the element group label, the element label, current solution variables
and their derivatives at the point, etc.

In return, the following parameters need to be computed or specified in

the subroutine
D = mass diffusion coefficient matrix

= bulk density
S0 , S f = coefficients of the mass creation rate
M = mass buoyancy force ratio
where M is defined as
n
M = i (i i 0 )
i =1
and S0 and S f are used to compute the mass creation rate according to
mi = S0 S f i
The reason that the program requires the coefficients S0 and S f rather than
the mass creation rate mi itself is to assemble the mass-ratio in an implicit
way, therefore maintaining the stability of the computational procedure.
The following sample example demonstrates an application of a user-

supplied material set for mass transfer equations. In this problem, two
species are assumed.
For species-1:
= 1, S0 = S f = 0, M = 0
10 0 0
D = 0 0.2 0

0 0 0.2
and for species-2
= 1, S0 = S f = 0, M = 0

0.2 0 0
D = 0 1.2 0

0 0 0.2
In the commented lines, a velocity-dependent material is used.

Furthermore, the buoyancy term is defined. These conditions correspond to
the following data set
= 118.6, S0 = S f = 0, M = 0.11
D = 0.01I + 0.01nn + 0.003(I nn)
SUBROUTINE ADFURM(MODE,TIME,IELG,IELM,IMAS,NNOD,NM,H,DH,VESL,
. DEN,DIF,S0,SF,BM)
C=======================================================================
C
C USER SUPPLIED MATERIAL DATA AND SOURCE TERMS FOR MASS TRANSFER
C
C-----------------------------------------------------------------------
C
C INPUT:
C
C MODE = REFERRING PHASE (SEE EXPLANATIONS BELOW)
C TIME = CURRENT TIME
C IELG = ELEMENT GROUP NUMBER
C IELM = ELEMENT NUMBER
C IMAS = INDEX OF THE CURRENT SPECIES
C NNOD = NUMBER OF LOCAL NODES FOR NNOD-ELEMENT
C H(*) = INTERPOLATION FUNCTION
C DH(3,*) = DERIVATIVES OF H WITH RESPECT TO (X,Y,Z)
C VESL( 1, I) = X-VELOCITY AT LOCAL NODE I
C VESL( 2, I) = Y-VELOCITY AT LOCAL NODE I
C VESL( 3, I) = Z-VELOCITY AT LOCAL NODE I
C VESL( 4, I) = PRESSURE AT LOCAL NODE I
C VESL( 5, I) = TEMPERATURE AT LOCAL NODE I
C VESL( 6, I) = X-DISPLACEMENT AT LOCAL NODE I
C VESL( 7, I) = Y-DISPLACEMENT AT LOCAL NODE I
C VESL( 8, I) = Z-DISPLACEMENT AT LOCAL NODE I
C VESL( 9, I) = TURBULENCE_K AT LOCAL NODE I
C VESL(10, I) = TURBULENCE_E AT LOCAL NODE I
C VESL(11, I) = X-COORDINATE AT LOCAL NODE I
C VESL(12, I) = Y-COORDINATE AT LOCAL NODE I
C VESL(13, I) = Z-COORDINATE AT LOCAL NODE I
C VESL(14+K,I) = MASS RATIO OF SPECIES-K (K=0,1,2,...,NMASS)
C NM = LENGTH OF THE FIRST ENTRY OF VESL
C
C-----------------------------------------------------------------------
C
C OUTPUT MATERIAL DATA FOR MODE=1:
C
C DEN = BULK DENSITY OF MIXTURE
C (FOR INCOMP.FLUIDS, DEFAULT=SPECIFIED)
C (FOR COMP.FLUIDS, DEFAULT=COMPUTED USING STATE EQ.)

C DIF(3,3) = DIFFUSIVITY TENSOR (MUST BE SPECIFIED HERE)

C
C-----------------------------------------------------------------------
C
C OUTPUT MASS CREATION RATE FOR MODE=2:
C
C DEFINITION:
C
C MASS CREATION RATE = S0 - SF * (MASS RATIO OF SPECIES-IMAS)
C
C S0 = THE TERM THAT WILL BE PUT ON THE RIGHT-HAND SIDE
C SF = COEF. THAT WILL BE PUT INTO THE STIFF MATRIX
C FOR NUMERICAL STABILITY, ALWAYS MANAGE SUCH THAT SF>=0
C
C-----------------------------------------------------------------------
C
C OUTPUT BOUYANCY FACTOR BM IN MOMENTUM EQUATION FOR MODE=3
C
C DEFINITION:
C
C BM = SUM{ BETMI*(SI-SI_REF) }
C
C SI = MASS RATIO OF SPECIES-I
C SI_REF = REFERENCE MASS RATIO OF SPECIES-I
C BETMI = MASS VOLUME EXPANSION COEFFICIENT OF THE SPECIES-I
C
C======================================================================
DIMENSION H(NNOD),DH(3,NNOD),VESL(NM,NNOD)
DIMENSION DIF(3,3)
C----------------------------------------------------------------------
BM = 0.D0
DH(1,1) = DH(1,1)
H(1) = H(1)
IELG = IELG
IELM = IELM
TIME = TIME
VESL(1,1) = VESL(1,1)
C----------------------------------------------------------------------
IF(MODE.EQ.1) THEN
C +-------------------------------------------------+
C | |
C | SPECIFY DEN AND DIF HERE FOR SPECIES-IMAS |
C | |
C +-------------------------------------------------+
C
C EXAMPLE FOR CONSTANT MATERIAL
C
IF(IMAS.EQ.1) THEN
DEN = 1.D0
DIF(1,1) = 10.D0
DIF(2,2) = 0.2D0
DIF(3,3) = 0.2D0
DIF(1,2) = 0.0D0
DIF(1,3) = 0.0D0
DIF(2,1) = 0.0D0
DIF(2,3) = 0.0D0
DIF(3,1) = 0.0D0
DIF(3,2) = 0.0D0
ELSE IF(IMAS.EQ.2) THEN
DEN = 1.D0
DIF(1,1) = 0.2D0
DIF(2,2) = 1.2D0

DIF(3,3) = 0.2D0
DIF(1,2) = 0.0D0
DIF(1,3) = 0.0D0
DIF(2,1) = 0.0D0
DIF(2,3) = 0.0D0
DIF(3,1) = 0.0D0
DIF(3,2) = 0.0D0
ENDIF
C
C EXAMPLE FOR V-DEPENDENT MATERIAL
C
c IU = 1
c IV = 2
c IW = 3
c EPS = 1.D-20
C
c DEN = 118.6D0
c DIF0 = 0.01D0
c DIFVN = 0.01D0
c DIFVT = 0.003D0
C
c U = 0.D0
c V = 0.D0
c W = 0.D0
c DO 2 I = 1, NNOD
c U = U + VESL(IU,I)*H(I)
c V = V + VESL(IV,I)*H(I)
c W = W + VESL(IW,I)*H(I)
c2 CONTINUE
c F0FT = DIF0 + DIFVT
c FNFT = DIFVN - DIFVT
c U2 = U*U + V*V + W*W
c U2 = MAX(U2,EPS)
c UI2 = FNFT/U2
c UU = U*UI2
c VV = V*UI2
c WW = W*UI2
c DIF(1,1) = U*UU + F0FT
c DIF(2,2) = V*VV + F0FT
c DIF(3,3) = W*WW + F0FT
c DIF(1,2) = U*VV
c DIF(1,3) = U*WW
c DIF(2,3) = V*WW
c DIF(3,1) = DIF(1,3)
c DIF(3,2) = DIF(2,3)
c DIF(2,1) = DIF(1,2)
GO TO 9999
ENDIF
C
IF(MODE.EQ.2) THEN
C +-------------------------------------------------+
C | |
C | SPECIFY THE SOURCE IN TERMS OF S0 AND SF |
C | |
C +-------------------------------------------------+
S0 = 0.D0
SF = 0.D0
GO TO 9999
ENDIF
C
c IF(MODE.EQ.3) THEN
C +-------------------------------------------------+
C | |

C | SPECIFY BM = SUM{ BETM*(SI-SIREF) } HERE |

C | |
C +-------------------------------------------------+
c NMASS = 2
c BM = 0.D0
c DO 31 I = 1, NMASS
c IM = 14 + I
c SI = 0.D0
c SIREF = 0.D0
c BETMI = 0.1D0
c IF(I.EQ.2) BETMI = 0.D0
c DO 32 J = 1, NNOD
c SI = SI + H(J)*VESL(IM,J)
c32 CONTINUE
c BM = BM + BETMI*(SI-SIREF)
c31 CONTINUE
c GO TO 9999
c ENDIF
C-----------------------------------------------------------------------
9999 CONTINUE
RETURN
END


Chapter 9: Formulation of fluid-structure interactions
Chapter 9 Formulation of fluid-structure interactions
9.1 Introduction
In this chapter, we introduce the formulation used for fluid-structure

interactions (FSI) in the ADINA system. The fluid flow models used here
can be incompressible flows, slightly compressible flows, low and high
speed compressible flows and flows through porous media. The solid
models used here can be all those available in ADINA. Heat and mass
transfers can also be specified in fluid-structure interaction analyses.
In fluid-structure interaction analyses, fluid forces are applied onto the

solid and the solid deformation changes the fluid domain. For most
interaction problems, the computational domain is divided into the fluid
domain and solid domain, where a fluid model and a solid model are
defined respectively, through their material data, boundary conditions, etc.
The interaction occurs along the interface of the two domains. Having the
two models coupled, we can perform simulations and predictions of many
physical phenomena.
The ADINA system has a structural analysis capability as well as a
fluid analysis capability. The availability of both capabilities within the
same code provides the base for developing sophisticated fluid-structure
interaction tools.
A typical fluid-structure interaction problem is illustrated in the next
figure. The fluid model is defined in the fluid domain with wall boundary
conditions, prescribed velocity at the inlet, zero distributed normal-traction
at the outlet and, most importantly, the fluid-structure interface condition.
The solid model is defined in the structural domain, where its bottom is
fixed and its top is the fluid-structure interface. The two models must have
also been discretized using the elements that are available in ADINA Solids
& Structures and ADINA-F, respectively. The typical task of an analysis
of a fluid-structure model is to obtain the fluid and structure response
through the coupled solution. The structural model is based on a
Lagrangian coordinate system and the displacements are the primary

9.1: Introduction
unknowns. A pure fluid model is always analyzed using a Eulerian

coordinate system. However, for fluid-structure interaction problems, the
fluid model must be based on an arbitrary-Lagrangian-Eulerian coordinate
system since the fluid-structure interface is deformable. Therefore, the
solution variables of the fluid flow include the usual fluid variables
(pressure, velocity, etc.) as well as displacements.
FSI interface
wall FSI interface
v
fixed
solid model
fluid model
FSI interfaces
on fluid and solid
wall boundaries
v
fixed
fluid-structure interaction model
Figure 9.1 Illustration of fluid-structure interactions
9.2 Theory
ref. Zhang, H., and Bathe, K.J., Direct and Iterative Computing
of Fluid Flows fully Coupled with Structures,
Computational Fluid and Solid Mechanics, K.J. Bathe,
editor, Elsevier Science, 2001.

9.2.1 Kinematic and dynamic conditions
The fundamental conditions applied to the fluid-structure interfaces are

the kinematic condition (or displacement compatibility)
d f = ds
and the dynamic condition (or traction equilibrium)
ni f = ni s
where d f and d s are, respectively, the fluid and solid displacements and
f and s are, respectively, the fluid and solid stresses. The underlining
denotes that the values are defined on the fluid-structure interfaces only.
The fluid velocity condition is resulted from the kinematic condition
v = ds
if a no-slip condition is applied, or
n i v = n id s
if a slip condition is applied.

The fluid and solid models are coupled as follows.
The fluid nodal positions on the fluid-structure interfaces are
determined by the kinematic conditions. The displacements of the other
fluid nodes are determined automatically by the program to preserve the
initial mesh quality. The governing equations of fluid flow in their ALE
formulations are then solved. In steady-state analyses, the mesh velocities
are always set to zero even the fluid nodal displacements are updated.
Accordingly, the fluid velocities on the fluid-structure interfaces are zero.
According to the dynamic conditions, on the other hand, the fluid
traction is integrated into fluid force along fluid-structure interfaces and
exerted onto the structure node
F(t ) = hd f i dS (9.1)

9.2: Theory
where hd is the virtual quantity of the solid displacement.
9.2.2 Separate meshes of fluid and solid models
Completely different elements and meshes can be used in fluid and

solid models. The elements are solely limited by the availability of solid
and fluid models in the ADINA system. The nodal point positions of the
two models are therefore generally not the same on the fluid-structure
interface as illustrated in Fig. 9.1 and in the following figure.
5
1 1
6
2
2
7
3
3
4
Figure 9.2 Coupling of fluid and solid nodes
The fluid nodal displacements are interpolated using the solid nodal
displacements. For example, the displacement at fluid node 2 is
interpolated using the displacements at solid nodes 1 and 2. Given the
boundary nodal displacements, the other fluid nodal displacements are
calculated in a special procedure to preserve the initial mesh quality. The
procedure is generally applicable to all types of moving meshes and thus is
described in detail in Section 12.10.
Similarly, the fluid traction at a solid node is interpolated using the
stress of the fluid boundary element where the solid node is located. In the
example shown here, the fluid stress at solid node 2 is interpolated using
the fluid stresses at fluid nodes 2 and 3, while the stresses at solid nodes 1

and 3 equal the fluid stresses at fluid nodes 1 and 4 respectively.

According to Eq.(9.1), the fluid force at solid node 2 includes stress at solid
nodes 1, 2 and 3. Recall that the fluid stress is the sum of the pressure and
the shear stress. The fluid solution variables and the solid solution
variables at the numbered nodes are therefore eventually all coupled.
Since we allow separate meshes in the fluid and solid domains, it is
likely that the two meshes on the interfaces are not compatible. The
discretized representations of the same geometry in the two models may
therefore not be the same. However, the distance between the two
discretized boundaries must be within a small distance. We define a
relative distance from fluid FSI nodes to the solid FSI interface as
df
rf = max
Ds
where d f is the distance from a fluid node to the structural discretized

boundary and Ds is the length of the solid boundary element (see the figure
below). This relative distance is always checked. Whenever rf 1 , the
program sees this as an error and stops. Whenever 0.001 rf < 1 , a
warning message is printed and the program continues.
boundary
FSI interface geometry
in fluid model
df
Ds
FSI interface
in solid model
Figure 9.3 Measure of the distance between fluid and solid

FSI boundaries

9.2: Theory
If there are gaps between fluid FSI nodes and the solid FSI interface,
these fluid nodes are automatically moved onto the solid interface and the
other fluid nodes are moved accordingly in the same procedure as described
in Section 12.10.
Similarly a relative distance from the solid FSI nodes to the fluid FSI
interface is defined and checked
d
rs = max s
D f
where d s is the distance from a solid node to the fluid boundary and D f is
the size of the fluid boundary element. According to the value of rs , the
program prints a warning/error message and continues/stops as for fluid
nodes.
When the distance between the boundaries is too large, either the
geometries used in the fluid and solid models are apart from each other, or
the boundary elements have misrepresented the geometry in one or both
models.
To avoid possible discrepancies between the two interfaces, the
following tips could be useful.
Unnecessary details of the geometry must be cleaned out prior to

generating meshes.
The mesh should be fine enough to represent the boundary geometry,
particularly along locations of sharp curvatures. If fewer elements are
intentionally used along the FSI boundary where sharp curvatures exist,
or (because of the limitation of the computer capacity) the boundary
elements are unable to reasonably represent the geometry in any one of
the models, it is best to simplify the geometry.
Along boundaries of sharp curvatures, use compatible meshes. Note
that even when the elements have the same size for the fluid and solid
models, the boundary nodes of both models may not be at the same
locations. Incompatible meshes can generally be applied to smooth
boundaries.

9.2.3 Consistent time integration for fluid and solid models
The time integrations for both fluid and solid equations must be
consistent. Although different coordinate systems are used in the fluid and
solid models, the two systems are the same on fluid-structure interfaces
where the Lagrangian coordinate system is used. We therefore first focus
on the time integration on the interface and then apply the results to the
whole computational domain. Since the displacement, velocity and
acceleration on the interface are the same for both fluid and solid models,
we will not distinguish between them on the interfaces.
Let the fluid equations and the solid equations be represented by
G f f , f = 0 and G s d, d, d = 0 respectively, where the fluid variables
are represented by f and the solid displacements are represented by d .
We also consider that the fluid flow equations are balanced at time t + t
and the solid equations are satisfied at time t + t . Consider the Euler
method used for the fluid velocity and acceleration
t +t
t +t d td
v = t +t
v + t v (1 )
t
t +t
t +t v tv
a = t +t
a + t a (1 )
t
From these equations, the velocity and the acceleration at time t + t can
be obtained in terms of the unknown displacement
1 t +t 1
t +t
v=
t
( d t d ) t v 1 t +t d a + t

1 1 1
t +t
a = 2 2 ( t +t d t d ) t v 2 t a 1 t +t d b + t
t t
We then apply these equations to the coupled system. The final time
integration scheme can therefore be summarized as
t +t
G f G f t +t f , ( t +t f t f ) t = 0
t +t
G s G s t +t d, t +t d a + t , t +t d b + t = 0

9.2: Theory
It can be shown that the necessary stability condition of the consistent

time integration is 1 2 . The condition is obtained based on the
assumptions of a linear system and equal size elements. The scheme of
= 1 2 also presents a scheme of second order accuracy in time.
However, since the coupled system is nonlinear and the elements are
practically non-uniform, the scheme of = 1 2 could be unstable or
oscillating. For most applications, = 1 is recommended.
The Composite time integration method of second order accuracy (see

section 3.2.1 for details) is also available for FSI.
9.2.4 Finite element equations of the coupled system
Let the solution vector of the coupled system be X = ( X f , X s ) , where

X f and Xs are the fluid and solid solution vectors defined at the fluid and
solid nodes respectively. Thus, d s = d s ( X s ) and f = f ( X f ) . The finite
element equations of the coupled fluid-structure system can be expressed as
F f X f , d s ( X s )
F [ X] =0 (9.2)
Fs X s , f ( X f )

in which, F f and Fs are finite element equations corresponding to

equations G f and G s respectively. Note that the decoupled fluid and
solid equations can be represented by F f X f , 0 = 0 and Fs [ X s , 0] = 0
respectively.
Two-way fluid-structure coupling
For many coupled problems, the fluid traction affects the structural
deformations and the solid displacement affects the flow pattern. This fact
is the reason for performing fluid-structure interaction analyses. We call
this type of analysis two-way coupling.

The coupled fluid-structure equation is a nonlinear system regardless of

the solid model used (linear or nonlinear), since the fluid equations are
always nonlinear. An iteration procedure must therefore be used to obtain
the solution at a specific time. In other words, we have iterative solutions
X1 , X 2 ,... in fluid-structure interaction problems. We use criteria that are
based on either stress or displacement, or both of these, to check for
convergence of the iterations. The stress criterion is defined as
kf kf 1
r
{
max kf , 0 }
and the displacement criterion is defined as
d sk d sk 1
rd d
{
max d sk , 0 }
where and d are tolerances for stress and displacement convergence,
respectively, and 0 is a pre-determined constant ( 108 ) for the purpose
overriding of stresses or displacements in case they become too small to
measure convergence. The default option is that both criteria need to be
satisfied.
One-way fluid-structure coupling
In certain cases, the deformation of the solid is so small that its

influence on the fluid flow is negligible. Then only the fluid stress needs to
be applied onto the structure and no iteration between the fluid and solid
models is needed. We call this type of interaction one-way coupling.
To reach solution effectiveness, we can either directly or iteratively

solve the coupled system. In one-way coupling, we can also solve the
system together in one run or separately in a number of runs. All these
options are discussed in the next sections.
In all cases, the fluid and solid solutions are saved in different porthole
files, namely <fluid>.por and <solid>.por.

9.3: Iterative computing of two-way coupling
9.3 Iterative computing of two-way coupling
This computing method is also called the partitioned method. In this

solution, the fluid and solid solution variables are fully coupled. The fluid
equations and the solid equations are solved individually in succession,
always using the latest information provided from another part of the
coupled system. This iteration is continued until convergence in the
solution of the coupled equations is reached.
The computational procedure can be summarized as follows. To obtain
the solution at time t + t , we iterate between the fluid model and the solid
model. We start with the initial solution guess d s1 = d 0s = t d s and
0f = t f . For iterations k = 1, 2,... , the following equilibrium iteration is
t +t
performed to obtain the solution X.
1) Solve the fluid solution vector X kf from the fluid equation
F f X kf , d d sk 1 + (1 d )d sk 2 = 0
This solution is obtained in fluid flow analysis, using the prescribed

solid displacement. Note that the solid displacement has been relaxed
by using a displacement relaxation factor d (0 < d 1) , which may
be useful for many difficult problems since the fluid and solid models
are not solved in the same matrix system. The use of this factor helps
to reach convergence in the iterations.
2) If only the stress criterion needs to be satisfied, the stress
residual is computed and checked against the tolerance. If
the criterion is satisfied, steps 3 to 5 are skipped.
3) Solve the solid solution vector X ks from the structural
equation
Fs X ks , kf + (1 ) kf 1 = 0

Note that the fluid stresses have also been relaxed using a stress
relaxation factor (0 < 1) .
4) The fluid nodal displacements are computed with the
prescribed boundary conditions d kf = d d sk + (1 d )d sk 1 .
5) If only the displacement criterion needs to be satisfied, the
displacement residual is computed and checked against the
tolerance. If both the stress and the displacement criteria
need to be passed, both convergence conditions are checked.
If the iteration has not converged yet, the program goes back
to step 1 and continues for the next iteration unless a
maximum number of FSI iterations have been reached (in
this case, the program stops and prints out the divergence
information).
6) Print and save the fluid and solid solutions if required.
In this solution method, the time steps and solution times are controlled
in the fluid model. However, all time functions defined in the solid model
must cover the time range of the computation. The parameters that control
the convergence of the coupled system are also determined in the fluid
model. These parameters are the stress and displacement tolerances and
relaxation factors, convergence criteria (see above), etc.
The convergences reached in fluid and solid equations are controlled in
the fluid and solid models respectively. The controls of saving and printing
solutions are also determined in the individual models.
The iterative method requires less memory than the direct method in
two-way couplings. It is suitable for small to large problems that have or
have not contact conditions. It is also good for steady-state analyses.
9.4 Direct computing of two-way coupling
This computing method is also called the simultaneous solution

method. In this direct solution method, as in the above iterative solution,
the fluid and solid solution variables are also fully coupled. The fluid
equations and the solid equations are combined and treated in one system.
Therefore, they are linearized in a matrix system, as for a fluid model or a
solid model alone. This matrix system can be written as

9.4: Direct computing of two-way coupling
A ff A fs X kf B f
=

A sf A ss X ks B s
and
X k +1 = X k + X k
For example, when the Newton-Raphson method is used, we have
B f = F fk F f X kf , d d sk + (1 d )d sk 1
B s = Fsk Fs X ks , kf + (1 ) kf 1
Fik
A ij = ( i, j = f , s)
X j
Note that the fluid stress and solid displacement have also been relaxed, as
in the iterative coupling method, using a stress relaxation factor and a
displacement relaxation factor d respectively. These factors have been
practically proven useful for many difficult problems even in the direct
computing method.
The computational procedure can be summarized as follows. We start
with the initial solution guess X 0 = t X for the solution at time t + t . For
iterations k = 1,2,, the following equilibrium iteration procedure is
performed to obtain the solutions t +t X .
(1) Assemble the fluid and solid equations as is normally done in a single
fluid and a single solid model, followed by assembling the coupling
matrices A fs and A sf .
(2) Solve the linearized equation of the coupled system and update the
solution (including solving the fluid displacements). Compute and
check the stress and/or displacement residuals against the specified
tolerances. If the solution has not converged yet, the program goes
back to step 1 to continue for the next iteration unless a maximum
number of FSI iterations have been reached (in this case, the program
stops and prints out the divergence information).
(3) Print and save the fluid and solid solutions if required.

In the direct computing method, as in the iterative solution method, the

time steps are controlled by the fluid flow model. Hence the same control
parameters specified in the solid solver are ignored in the solution
procedure. However, all time functions defined in the solid model must
cover the time range of the computation. The parameters that control the
convergence of the coupled system are also processed in the fluid model.
These parameters are the stress and displacement tolerances, relaxation
factors, convergence criteria, etc.
The controls of saving and printing solutions are determined in the
individual solid and fluid models.
The direct method in general is faster than the iterative method in two-
way couplings. It is suitable for small to medium problems that have not
contact conditions. It is also good for transient analyses.
For obvious reasons, the direct FSI method is not applicable to the
segregated method, where even fluid variables are not directly coupled.
9.5 Direct computing of one-way coupling
In this case, the fluid stress is applied onto the structure while the
structure has no influence on the flow field. Choose one as the maximum
number of FSI iterations permitted in the iterative fluid-structure coupling
for this type of analyses. The program will then perform the following
steps in each time step:
1) Solve the fluid solution vector X f from the fluid equation

F f X f , 0 = 0 as for a fluid flow model alone with zero
solid displacements.
2) Solve the solid solution vector X s from the structural
equation Fs X s , f = 0 as for a solid model alone with the
prescribed fluid stress.
3) Print and save the fluid and solid solutions if required.

9.5: Indirect computing of one-way coupling
In the direct computing of one-way coupling problems, the time step

control is specified in the fluid model. The controls of saving and printing
solutions are determined in the individual solid and fluid models.
The direct method in one-way couplings is suitable for small to medium
problems that have negligibly small displacements to fluid fields.
9.6 Indirect computing of one-way coupling
When the fluid model is large in one-way FSI couplings, it can be

effective to first compute the fluid flows and save the stresses in a file at
selected time steps. The solid models can then be analyzed, with the fluid
stress input from the file, as many times as required. It is important to
mention here that the fluid and solid models are still prepared separately, so
that the meshes of the two models may not be compatible on the interface.
In this indirect computing method, the program performs the same
functions as in the direct computing of the one-way coupling.
In the indirect computing of one-way coupling problems, all the control
parameters are specified in the individual fluid and solid solvers. There are
therefore possible discrepancies between the fluid and solid solution times.
At certain times when the fluid stresses are not available, a linear
interpolation is performed to provide the fluid stress for the solid model
(see the figure below). If the solution time is out of range of the times for
which the fluid stresses are saved, a linear extrapolation is applied. In order
to have more accurate solutions, the fluid stresses should be saved more
frequently to cover the time steps that may be used for the solid model.
fluid stress saved fluid stresses
fluid stresses required

in solid model
Figure 9.4 Interpolation of fluid stresses in time in indirect

computing of one-way fluid-structure coupling
The indirect method in one-way couplings is suitable for small to large

problems that have negligibly small displacements to fluid fields.
9.7 Elements on fluid and solid interfaces
Any fluid element that can be used in a fluid model can also be used in
fluid-structure interaction models. These elements are either two-
dimensional elements (including axisymmetric elements) or three-
dimensional elements. The fluid-structure interfaces are always assigned to
boundaries of elements. In two dimensions and three dimensions, the
interfaces are then formed by line elements and surface elements,
respectively, which are generated automatically by the program.
Any solid element, on the other hand, that can be used in a separate
solid model can also be used in fluid-structure interaction models.
However, some of the elements cannot be directly connected with fluid
elements, which means that no fluid-structure interface can be assigned to
those elements. For example, the spring element cannot be directly
connected with fluid elements.
As a special case of directly connectable solid elements, the beam-type
elements can interact through both their sides with 2D-fluid boundary
elements and the shell-type elements can interact through both their
surfaces with 3D-fluid boundary elements. We call them F-S-F
connections. In this case, the two fluid boundaries are geometrically
identical but connected with the two sides/surfaces of the solid elements
respectively. On the contrary, when solid boundary elements or one
side/surface of the solid elements are connected with fluid boundary
elements, we call them F-S connections.
The possible connections of fluid and solid elements are listed in the
following table, in which the abbreviation n.c. stands for not
connectable.

9.7: Elements on fluid and solid interfaces
Table 9-1 Elements in connections of fluid and solid models in

fluid-structure interaction problems
2D fluid 3D fluid
solid elements fluid elements
elements elements

connection type
truss/cable elements
beam elements F-S
n.c.
iso-beam elements F-S-F
pipe elements
2D-solid elements
F-S
plate elements F-S
F-S-F
shell elements
3D-solid elements n.c. F-S
2D-displacement-based fluid elements
F-S n.c.
2D-potential-based fluid elements
3D-displacement-based fluid elements
n.c. F-S
3D-potential-based fluid elements
spring/damper/mass elements n.c.
In the case of F-S-F connections, although the two fluid boundaries are
geometrically identical, they are separated by the solid elements and
therefore generally nonidentical to the physical quantities, such as the
stresses and the pressures. The fluid elements must be carefully modeled in
this case and are discussed hereafter.
O-shape 2D F-S-F connection
The following figure shows a typical example of an O-shape two-

dimensional fluid element layout, where a beam is surrounded by fluid
elements and may be supported by springs that are invisible to fluids. Two
fluid boundaries and the solid beam locate at the same place. Since the
fluid variables on both sides of the beam are different, the fluid boundary
nodes must be different except, however, the two corner nodes that must be
coincident since the fluid variables are the same there.

fluid elements
separated
nodes
connected
coner nodes
beam
spring
Figure 9.5 Typical example of O-shape 2D F-S-F connections

in fluid-structure interaction analyses.
The O-shape two-dimensional fluid elements can be generated in the

AUI by explicitly forcing the corner nodes to be coincident while the
middle nodes are not checked in the coincident procedure. However, this
may be a tedious input if the geometry is complicated (this is particularly
true in three-dimensions). Two simple solutions are presented in the
following figure.
different
subdivisions
on both sides
c < n1
Figure 9.6 Possible methods for generating O-shape 2D fluid

elements in F-S-F connections

In the first solution, different element subdivisions are used on the two
boundaries. The key here is to make sure that there are no middle nodes at
the same locations.
The second solution is to move the two faces slightly apart, while
allowing the end-points to still share the same locations. The distance
between the two faces must be small enough (for example, less than ten
percent of the boundary element length), but must be larger than the
tolerance c that is used in the coincidence checking when the elements are
generated.
Two fluid-structure interfaces must be defined in the solid model and in
the fluid model. Each of them corresponds to one side of the beam. In this
example, the fluid-structure interface 1 in the fluid model is connected
logically with the fluid-structure interface 1 that is defined in the solid
model. They form a pair to model the left side of the beam. Similarly, the
pair of the interfaces 2 models the right side of the beam.
In general, the connectable boundaries should be specified in the same
fluid-structure interface, since the connected fluid corner nodes may be
over specified by different moving conditions. However, the corner nodes
in the example are uniquely defined in the two FSI conditions.
Truss/cable elements, iso-beam elements and pipe elements are similar
to beam elements in defining fluid-structure interaction problems.
FSI interface 2
FSI interface 1
FSI interface 2
beam FSI interface 1
solid model fluid model
Figure 9.7 Application of fluid-structure interface conditions to

beam elements

C-shape 2D F-S-F connection
The following figure shows a typical example of C-shape two-

dimensional fluid elements, where a beam is surrounded by fluid and has its
one end-point fixed. Two fluid boundaries share the same geometric
locations with the solid beam. Except at the fixed end-point, the fluid
pressures on both sides of the beam are different. The fluid boundary
nodes, therefore, must be separated on the two faces except that the end-
node points must be kept coincident.
separated
nodes fluid elements
beam
connected
coner node
Figure 9.8 Typical example of C-shape 2D F-S-F connections

The C-shape two-dimensional fluid elements can be generated in the

AUI by explicitly forcing the connected corner node to be coincident while
the other nodes are not checked in the coincident procedure. However, this
input may be tedious if the geometry is complicated. Two simple solutions
are presented in the following figure.
c < n1
c < n1
different
subdivisions
on both sides
Figure 9.9 Possible methods for generating C-shape 2D fluid

elements in F-S-F connections

The first solution is to move the end-point a small distance apart and to
use different subdivisions on the two boundaries. The key here is to make
sure there are no nodes at the same locations.
The second solution is to move the two faces slightly apart, while
allowing only the end-point (the connected one) to remain at the same
location. The distance between the two faces must be small enough (for
example, less than ten percent of the boundary element length), but must be
larger than the tolerance c that is used in the coincidence checking.
As in the case of the O-shape interface, two interfaces must be defined
in the solid model and in the fluid model, each pair of the fluid-solid
interfaces corresponding to one side of the beam.
Truss/cable elements, iso-beam elements and pipe elements are similar
to beam elements in defining fluid-structure interaction problems.
O-shape 3D F-S-F connection
In three dimensions, O-shape fluid elements enclose a plate or a shell.

The model needs to be constructed as for the O-shape 2D fluid-structure
interface. The solid shell is treated as two faces that are connected with
two fluid faces. The elements generated on the two fluid faces must be
separated inside the face, while connected on the boundary of the faces. A
fluid-structure interface must be assigned to one fluid face and connected to
its solid counterpart to form a pair of fluid-structure interfaces. A separate
pair is similarly assigned for another face.
top face
bottom face
connected fluid nodes
separated nodes on both faces
Figure 9.10 Typical example of O-shape 3D F-S-F connections


The fluid elements can be generated by enforcing coincidence checking

on the boundary of the two fluid faces, while not checking inside the faces.
However, other methods that are analogous to those in the two-dimensional
case can also be used. For example, by slightly changing the shape of one
face, the two faces would be moved apart by a small distance while keeping
their boundaries together. The distance between the two faces must be
small enough (for example, less than ten percent of the boundary element
length), but must be larger than the tolerance used in coincidence checking.
C-shape 3D F-S-F connection
Unlike the O-shape three-dimensional fluid elements, the C-shape fluid

elements leave an open part where the shell could be fixed or connected to
another part of the model. Along the open part, the fluid nodes must also
be separated. The solid shell is treated as two faces that are connected with
two fluid faces. The elements generated on the two fluid faces must be
separated inside the face and on the open face boundary, while connected
on the other boundaries of the two faces. A fluid-structure interface must
be assigned to one fluid face and connected to its solid counterpart to form
a pair of fluid-structure interfaces. A separate pair is similarly assigned for
another face.
top face
bottom face
connected fluid nodes

separated nodes on both faces
Figure 9.11 Typical example of C-shape 3D F-S-F connections


The fluid elements can be generated by enforcing coincidence checking

on the connected boundaries of the two fluid faces, while not checking the
rest of the faces. However, other methods that are analogous to those in the
two-dimensional case can also be used. For example, by slightly changing
the shape of one face, the two faces and the open boundary part would be
moved apart by a small distance while keeping the connected boundaries
together. The distance between the two faces must be small enough (for
example, less than ten percent of the boundary element length), but must be
larger than the tolerance used in coincidence checking.
9.8 Material models
Any fluid materials that can be used in separate fluid models can also
be used for fluid-structure interaction models.
Any solid materials that can be used in separate solid models can also
be used for fluid-structure interaction models.

9.9 Procedure of model preparation and testing
A fluid-structure interaction model is a combined model that includes a

fluid model and a solid model. They are connected through pairs of fluid-
structure interface conditions. It is understood that an FSI problem cannot
be solved if the individual fluid or solid model cannot be analyzed.
Therefore, an essential step towards a successful FSI computation is to have
the two individual models well-tested prior to coupling them together. A
good reason to perform the testing in separate models is that each problem
must be well-defined and no input errors should be present.
The solid model can be tested separately by applying a pressure
condition on the FSI boundary to simulate the fluid stress. FSI conditions
must be taken out in this test (by doing so, the solid model is separated
from the fluid model). The magnitude of the pressure should be in the
range of the possible fluid stresses that would be obtained in the FSI model.
The element, mesh, material data, etc., have to be tested well in this model
(refer to the ADINA Theory and Modeling Guide for the model preparation
and testing).
To test the fluid model, apply moving wall conditions on the FSI
boundary to simulate the displacements of the structure. Similarly, FSI
conditions must be taken out in this test. The range of boundary movement
should be as expected by the solid displacements. In addition to the
guidelines in Section 13.1 on model preparation and testing, the moving
mesh control needs special attention, particularly when large displacements
are encountered. Refer to the guidelines on moving mesh control in
Section 13.2 as well.
Once the fluid and solid models have been solved separately, the two
models can be coupled by having small input changes. Remember that the
control parameters of the coupled system are all specified in the fluid
model. The solid model should have minimum changes. In the solid
model, only the FSI condition is used to replace the pressure condition. In
the fluid model, replace the moving wall condition to a corresponding FSI
condition. In this stage, we focus on the interaction part, rather than on
the conditions related to the individual models. We have to, therefore,
avoid making additional changes in the coupled model, except those that
are related to coupling the system. In case the coupled system cannot be
solved, we know that only the coupling causes the difficulties.

9.10: Other types of fluid-structure interactions
Some guidelines are listed below.
In transient analyses, make sure a good initial condition is used. This

condition is more crucial in FSI computations than in individual model
solutions. Since the solid variables are displacements and the fluid
variables are velocities, etc., the initial conditions may not be
compatible. For example, recalling that the velocity is the time
derivative of the displacement, a steady-state solution of fluid velocity
does not correspond to a constant displacement. When convergence
difficulties arise in the first time step, it is possible that the initial
conditions are not well set.
If a pre-stretched structure is required, perform an artificial first time
step, in which the structure is stretched while the fluid conditions are
not yet applied (for example, zero inlet velocity, zero normal traction,
etc.) as controlled in the associated time functions. In this step, the
fluid mesh must be moved accordingly. The actual transient solution
starts then at the second time step.
Compute a steady-state solution as the initial condition if it is
physically correct. This can be achieved using a large time step size in
the first time step. Another clearer procedure is to perform two runs.
The first run is to obtain a steady-state solution (may be in a few time
steps). The second run is the restart execution of the first run using
transient analysis conditions.
It can be effective too to use a similar solution, as an initial condition,
from a mapping file.
The time step size must be properly determined as well. In steady-state
analyses, the time step size corresponds to the increments defined in
loads and boundary conditions. Therefore the time step size associated
with the time functions are important. In transient analyses, the time
step sizes must be small enough to capture the resolution of the solution
and large enough to allow the simulation to be completed in a
reasonable computer time. The general principles and suggestions are
described in Section 13.1. A special indicator of proper time step sizes
for FSI problems can be obtained from a frequency analysis of the
individual solid model. Since the fluid equations are always nonlinear,
no frequency analyses can be performed on FSI coupled models.
There are stress and displacement convergence criteria for FSI
iterations. The default option is to have both criteria satisfied. When

the displacements are very small or very large, the displacement

criterion should not be used. Similarly, when the stress is very small,
the stress criterion should not be used. In general, since the fluid forces
are not too small in FSI analyses, the stress criterion is preferred. Note
that, in iterative computing of two-way couplings, the stress and
displacement tolerances should be larger in general than the tolerances
used for equilibrium iterations used in the individual fluid and solid
models.
In two-way couplings, if the convergence is not achieved in the FSI
iterations, you can adjust the relaxation factors for the stresses and
displacements. The stress relaxation factor should be considered first
since it is usually the main source of convergence difficulty. The
factors should always be in the range from 0 to 1. In general, the closer
to 1 the factors are, the less the number of FSI iterations should
perform (if the convergence is achieved). On the other hand, the closer
to 0 the factors are, the greater the possibility of reaching convergence.
Large solution times are caused by factors being too small. The
optimal values should be obtained from numerical experiments,
preferably in a coarse mesh model.
In 2D planar FSI using beams, we recommend using the plane strain
iso-beam with unit thickness. This is because the fluid flow domain
corresponds to a unit thickness slice in the units selected. If the plane
stress or Hermitian beams are used, the stiffness of the iso-beam must
correspond to a unit thickness (i.e., the unit thickness implicitly
assumed in the 2D fluid flow).
9.10 Other types of fluid-structure interactions
In this section, we introduce more features on fluid and structural

interactions. Beside the interactions on fluid-solid interface, fluid and
structure can also be coupled in porous media or through thermal materials.
Consider a general fluid-structure-interaction model illustrated in Fig.
9.12. The whole computational domain is divided into three regions:
the pure fluid region f , a porous medium region p , and a pure solid
region s . Typical structural solution variables are displacements while
typical fluid solution variables are pressure, velocity and temperature. In

f , all fluid solution variables are considered. The displacement of the

boundary where it abuts the solid is also a variable that is formulated in the
solid model. The displacement inside the domain is arbitrary and is
formulated in the ALE coordinate used in the fluid governing equations. In
p , all fluid solution variables and the displacement are present, thus the
fluid and solid models sharing this domain. Finally in region s , the
displacement is present and the temperature variable is also included if the
thermal effect on the structure is significant. The interactions thus happen
in the regions where the two models are met. More precisely, the coupled
( (
regions are the interface S = f s p ) ) where traction and
displacement conditions must be satisfied; the region p where stress
condition must be satisfied; and finally the region s where thermal-
mechanical condition must be satisfied. The three types of interactions will
be called respectively, (conventional) Fluid-Structural Interaction (FSI),
Porous Fluid-Structural Interaction (PFSI) and Thermal Fluid-Structural
Interaction (TFSI). The coupled domain is then FSI = S p s .
pure fluid
domain
pure solid
domain
porous medium
domain
fluid-stucture
interface S
Figure 9.12 Illustration of overall fluid-structure interactions.

The coupled region is FSI = S p s
Currently, only the iterative method is applicable to PFSI and TFSI.

For the computational procedure, refer to Section 9.3.

9.10.1 Fluid-solid interaction in porous media
Unlike on the interface, the fluid and structural variables are coupled in
the whole porous medium. The microscope fluid stress must be added to
structural model as additional internal stress f ( pI ) . The total Cauchy
stress in solid model becomes
total
s = s pI
This equation is incorporated with the finite element procedure in solid

model, of course, in various elements and various structural materials.
On the other hand, the displacement compatibility between fluid mesh
and solid mesh is not required inside the medium (although it is must be
satisfied on the interface described in the previous section). However, as an
effective and convenient method, we still use
d f = ds
in the whole porous medium. This condition also indicates that the moving
mesh velocity in ALE formulation is identical with the solid velocity
w = ds
We note that the fluid mesh and solid mesh can be entirely different.
Interpolation is applied to the entire porous domain to obtain, respectively,
the fluid pressure on solid mesh and fluid nodal displacements.
The porosity that indicates the fluid volume percentage in the total
volume is no longer a constant in coupled porous media. Assuming that the
volume change of the solid part is much smaller than that in fluid part (or
the so called incompressible solid skeleton), the current porosity can be
obtained
J0
=1
J
(10 )

where J is the geometric element Jacobian and the superscript 0 indicates

the quantities at initial reference configuration. Accordingly, the current
fluid permeability is
0
=
0
9.10.2 Coupled thermal fluid-structural interaction
Thermal-mechanical interaction occurs in the solid region if structural

materials are temperature-dependent (taking rubber as an example). It can
also occur in porous media providing that the solid skeleton is temperature-
dependent as well. The thermal condition imposed on structure model is
realized through temperature-dependent materials
ms = ms ( f )
where, ms represents any temperature-dependent material properties in
structure. The temperature is a solution variable of the energy equation and
can be modeled either in the fluid or solid models.
Similar to porous media, the fluid nodal displacement is arbitrary, as
long as the boundary displacement compatibility is satisfied. For
convenience, we still use d f = d s to update fluid mesh and, accordingly,
fluid velocity in solid region is equal to the solid velocity, v f = d s .
There are two methods to perform TFSI computation:
(1) The standard two-way TFSI coupling is naturally included in two-

way FSI computation through standard setting. Temperature
variable is included in both fluid and solid models and coupled by
forcing heat-flux balance on FSI interface. This method accounts for
heat generation both in the structure and the fluid. That is, heat
generation due to plastic deformation or viscous effects in the
structure, and heat generation due to frictional contact are now
accounted for in the two-way thermal FSI analysis.

(2) The conjugate heat transfer is different from the standard TFSI by
applying temperature variable only in fluid model. The fluid model
must include solid element groups that are coincident with the
structural domain. Temperature is defined in both fluid and solid
element groups. In the case that the mesh in solid model is different
from the fluid solid element-group mesh, an interpolation method is
automatically applied to mapping variables that required. In order to
compute the thermal effect accurately when shell (in 3D problems)
or beam (in 2D problems) is used in TFSI analyses, the special
boundary condition called shell-thermal should be applied to the
fluid-shell interface. With this condition, ADINA-F can accurately
predict the temperature and its gradient on shells and beams and
then on nodal deformations.
9.10.3 Model one-way thermally coupled problem using mapping
In one-way thermally coupled models, it is assumed that structural

deformation has negligible effect on the fluid solution. To model those
problems, fluid solutions can be solved first without solving solid variables.
The solutions, including temperature are saved in fluid restart file. The
temperature solution is then mapped onto the solid mesh in the pre-
processing stage while the solid model is created. In the solid model,
temperature becomes a given condition.
Note that if the simulation times used in the fluid and solid models are
different, a linear interpolated temperature between two time steps is used.
For more accurate solutions, if temperature is required at multiple times in
the solid model, the fluid solutions at those times should be properly saved
into restart files.


Chapter 10: Elements
Chapter 10 Elements
10.1 Boundary line elements
The line elements are not directly available to users. They are only
used as boundary elements of two dimensional field elements. The
boundary elements are automatically generated by ADINA-F. Many
boundary conditions are formulated using boundary elements. Based on
the elements used, two types of line elements will be encountered: a 2-node
linear element and a 3-node parabolic element (see the figure below).
1 2 3 2
1
boundary element of boundary element of
2d 3-node element 6/9-node elements
Figure 10.1 Line elements
In line elements, any variable f is interpolated as
2
f = hi0 f i for 2 - node linear element
i =1
3
f = H i0 f i for 3 - node parabolic element
i =1
where
h10 = 1 r , h20 = r
(10.1)
H10 = h10 (h10 h20 ), H 20 = h20 (h20 h10 ), H 30 = 4h10 h20

10.1: Boundary line elements
Numerical integrations are unnecessary since all integrations can be

computed accurately in analytical forms. For example,
x 2 x1
( h ) = i
0
2 ( 1,1)
x 2 x1
x 2 x1
( H ) = i
0
2 (h
0
2 3h10 ,3h20 h10 , 4h10 4h20 )
x 2 x1
n m 1 n m n !m !
h10 h20 = h10 h20 dr =
0 (n + m + 1)!
and in particular,
1 1
(h ) = ( ), (h h ) =

i
0 1,1
i
0 0
j
3 6
2 2 1
sym. 3
2 1 1
15 30 15

( H ) = 16 , 16 , 23 ,
i
0
(H H ) =
i
0 0
j

2 1
15 15

sym. 8
15
10.2 2D triangular element (3-node)
The 3-node triangular element can be used for two-dimensional planar

and axisymmetric flows. The following figure shows the 3-node triangular
element. Note that all variables are defined at the corner nodes. The center
node is an auxiliary node where only the velocity (called bubble velocity)
is defined. The center velocity is locally condensed in the assemblage so
the degrees of freedom do not appear in the final equations.

3
F2 F1
s 0
1 2
r F3
Figure 10.2 2D/axisymmetric 3-node element
The triangular element satisfies the inf-sup condition of stability and

optimality. It is a linear element with first order accuracy in spatial
interpolation. The elements can be generated using a free-form mesh
generator for complicated geometries. The element can be used for high
and low Reynolds/Peclet number flows.
In axisymmetric flows, the model corresponds to one radian of the
physical domain (see Fig.2.2). Hence, when concentrated loads are
applied, they must also refer to one radian.
The interpolation functions for 2D 3-node triangular element are
defined as
h1 = 1 r s, h2 = r , h3 = s, h0 = h1 h2 h3 (10.2)
The solution variables are interpolated as
3 hi fi if f = v
i =0
f = 3
i =1 hi fi else

hi = J 1 Si i = 1, 2,3
1 r 2 n !m !k !
h1n h2m h3k = 2 h1n h2m h3k dsdr =
0 0 (n + m + k + 2)!

10.2: Triangular element (3-node)
and in particular,
1 1 1 1

6 12 12 180
1 1 1
( h , h , h , h ) = ( 13 , 13 , 13 , 601 ) , ( h h ) =
6 12 180
1 2 3 0 i j
1 1
6 180

sym. 1
2520
here, J is the Jacobian (twice the triangle area), Si is the outward facial
vector of the side opposite the local node i .
10.3 2D quadrilateral element (9-node)
The 9-node quadrilateral element can be used for two- dimensional

planar and axisymmetric flows. The following figure shows the 9-node
quadrilateral element. The pressure is defined at the four corner nodes and
all other variables are defined at all 9 nodes.
4 F3
3
7
F4 9 6 F2
8
s
5
1 r 2
F1
The quadrilateral element satisfies the inf-sup condition of stability and

optimality. The element is of second order accurate in the spatial

interpolations. The element is usually used for relatively low

Reynolds/Peclet number flows.
The interpolation functions are H i (i = 1,...,9) if all 9 nodes are
included and hi (i = 1,..., 4) if only 4 corner nodes are included. They are
defined as
( H1 , H 2 , H 5 ) = ( H10 ( r ) , H 20 ( r ) , H 30 ( r ) ) H10 ( s )
( H 4 , H 3 , H 7 ) = ( H10 ( r ) , H 20 ( r ) , H 30 ( r ) ) H 20 ( s )
( H 8 , H 6 , H 9 ) = ( H10 ( r ) , H 20 ( r ) , H 30 ( r ) ) H 30 ( s )
( h1 , h2 ) = ( h10 ( r ) , h20 ( r ) ) h10 ( s )
( h4 , h3 ) = ( h10 ( r ) , h20 ( r ) ) h20 ( s )
and the H 0 's and h0 's are given in Eq.(10.1). The solution variables are
interpolated as
4 hi f i if f = p
i =1
f = 9
i =1 H i fi else
No analytical integrations are available for 9-node elements. Numerical

Gauss integrations are used.
10.4 2D triangular element (6-node)
The 6-node triangular element can be used for two- dimensional planar
and axisymmetric flows. The following figure shows the 6-node triangular
element. This element is generated as a collapsed 9-node element. The
pressure is then only defined at the three corner nodes and all other
variables are defined at all 6 nodes.

10.4: Triangular element (6-node)
3
F1
F2
6 5
s
1 4 2
r
F3
The 6-node triangular element satisfies the inf-sup condition of stability

and optimality. It is second order accurate in the spatial interpolation. The
element is usually used for relatively low Reynolds/Peclet number flows.
The interpolation functions are H i (i = 1,..., 6) if all 6 nodes are
included
hi ( 2hi 1) i = 1, 2,3
Hi =
4h0 hki i = 4,5,6, ki = mod ( i + 1,3) + 1
where, the hs are given in Eq.(10.2). If only 3 corner nodes are included,
the linear interpolation functions defined in Eq.(10.2) are used. The
solution variables are interpolated as
3 hi f i if f = p
i =1
f = 6
i =1 H i fi else
No analytical integrations are available for 6-node elements. Numerical

Gauss integrations are used.

10.5 3D tetrahedral element (4-node)
The 4-node tetrahedral element can be used for three-dimensional

flows. The following figure shows the 4-node tetrahedral element. All
variables are defined at the corner nodes. The center node is an auxiliary
node where only the velocity (called bubble velocity) is defined. The
center velocity is locally condensed out prior to the assemblage so the
degrees of freedom do not appear in the final equations.
4
F2
F3
0 3
t
s
1 F1
r
F4 2
Figure 10.5 3D 4-node element
The tetrahedral element satisfies the inf-sup condition of stability and

optimality. It is a linear element with first order accuracy in the spatial
interpolations. The elements can be generated using a free-form mesh
generator for complicated geometries. This element can be used for
relatively high and low Reynolds/Peclet number flows.
The interpolation functions for 4-node tetrahedral element are
h1 = 1 r s t , h2 = r , h3 = s, h4 = t ,
h0 = h1 h2 h3 h4
The solution variables are interpolated as

10.5: 3D tetrahedral element (4-node)
4 hi f i if f = v
i =0
f = 4
i =1 hi f i else

hi = J 1 S i i = 1, 2,3, 4
1 1-r 1-r -s
h1n h2m h3k h4l = 6 h1n h2m h3k h4l dtdsdr
0 0 0
6 n !m !k !l !
=
(n + m + k + l + 3)!
and in particular,
1 1 1 1 1
10 20 20 20 3360
1 1 1 1
10 20 20 3360
( h , h , h , h , h ) = 14 , 14 , 14 , 14 , 8401 , (
1 2 3 4 5

hi h j = ) 1
10
1
20
1
3360
1 1
sym. 10 3360
1

415800
here, J is the Jacobian (which is six times the tetrahedral volume), Si is the
outward surface vector of the surface opposite the local node i.
10.6 3D brick element (27-node)
The 27-node brick element can be used for three-dimensional flows.

The following figure shows the 27-node brick element. The pressure is
defined at the eight corner nodes and all other variables are defined at all 27
nodes.

5
F2 7
14
6
13 27 15
8
F6 5 16
23 19
18 24
22 21
20 F4
17 2 25 10 3
t 9
s 26 11
1 r 4
12
F3 F1
Figure 10.6 27-node 3D element
The brick element satisfies the inf-sup condition of stability and

optimality. It has second order accuracy spatially while the variables are
interpolated. This element can be used for relatively low Reynolds/Peclet
number flows.
The interpolation functions are H i (i = 1,..., 27) if all 27 nodes are
included and hi (i = 1,...,8) if only 8 corner nodes are included. They are
defined as
( H1 , H 4 , H12 ) = H 0 (r ) H10 ( s) H10 (t )

( H 5 , H 8 , H16 ) = H 0 (r ) H10 ( s) H 20 (t )
( H17 , H 20 , H 25 ) = H 0 (r ) H10 ( s) H 30 (t )
( H 2 , H 3 , H10 ) = H 0 (r ) H 20 (s) H10 (t )
( H 6 , H 7 , H14 ) = H 0 (r ) H 20 ( s) H 20 (t )
( H18 , H19 , H 23 ) = H 0 (r ) H 20 (s) H 30 (t )
( H 9 , H11 , H 26 ) = H 0 (r ) H 30 ( s) H10 (t )
( H13 , H15 , H 27 ) = H 0 (r ) H 30 ( s) H 20 (t )
( H 22 , H 24 , H 21 ) = H 0 (r ) H 30 ( s) H 30 (t )

10.6: 3D brick element (27-node)
( h1 , h4 ) = h0 (r )h10 ( s)h10 (t )
( h5 , h8 ) = h0 (r )h10 ( s)h20 (t )
( h2 , h3 ) = h0 (r )h20 ( s)h10 (t )
( h6 , h7 ) = h0 (r )h20 ( s)h20 (t )
where
H 0 (r ) = ( H10 ( r ), H 20 ( r ), H 30 (r ))
h 0 (r ) = (h10 (r ), h20 (r ))
and the H 0 's and h0 's are given in Eq.(10.1). The solution variables are
interpolated as
8 hi f i if f = p
i =1
f = 27
i =1 H i fi else
No analytical integrations are available for 27-node elements.

Numerical Gauss integrations are used.
10.7 2D FCBI elements (4- and 3-node)
2D FCBI elements include 4-node and 3-node elements. They can be

used for two-dimensional planar and axisymmetric flows. The following
figures show the definitions of FCBI 2D elements.

3 (0,1) 4 (0,1) 3 (1,1)

3
2 4
2
s 1 s
3 1
1 2 1 2
(0,0) r (0,1) (0,0) r (1,0)
Figure 10.7 FCBI 2D elements
All solution variables are defined at corner nodes. Since step functions are
used for weighting functions, FCBI elements are eventually equivalent to
their counterpart: finite volume methods. Each element is thus equally
divided into sub-control-volumes and integrations on both faces and
surfaces are performed within elements. A flow-condition-based-
interpolation function for velocity on a face, say from point ( r , s ) = ( 12 , 0 )
to point ( r , s ) = ( 12 , 12 ) in a 4-node element, is defined as
h1v h4v
v v
= h ( x1 ) h ( x 2 ) h ( s ) h ( s )
h2 h3
with
1 y
k
eq r 1 vix k
k
x = , qk = , h( y) =
eq 1
k
y
where x1 = x 2 x1 , x 2 = x3 x 4 and v is the average velocity on the

face. With these functions, upwinding-effect is automatically captured in a
natural way.
On the other hand, pressure, temperature and coordinates are
interpolated using linear or bi-linear functions. In 4-node element, these
functions are the same as defined in Section 10.3. In 3-node element, they
are the same as defined in Section 10.2. In order to satisfy the Inf-sup

10.7: 2D FCBI elements (4- and 3-node)
condition, the mass flux along face is computed from a locally condensed
momentum equation.
10.8 3D FCBI elements (8-, 4-, 6- and 5-node)
These elements can be used for three- dimensional flows. The

following figures show the definitions of FCBI 3D elements.
6
(0,1,1)
4 (0,0,1) 5 (0,0,1)
5
4
5 (1,0,1)
3
4 3
(0,1,0)
t 1 3 3
(0,1,0) (1,1,0)
s
t
1 s 1
(0,0,0) 4 2
r
1 5
(0,0,0) r
2
2 (1,0,0) 7
6 (0,1,1) (1,1,1)
2 (1,0,0)
2 4 (0,0,1)
5 8 (1,0,1)
(0,0,1)
3
5
3 6 2 (0,1,0) 3
4 (1,1,0)
t 2
s
1
t 3
1 r 4 s (0,1,0)
(0,0,0) (1,0,0)
1 4
(0,0,0) r 1
2 (1,0,0)
Figure 10.8 FCBI 3D elements
All variables are defined at corner nodes. The interpolations functions for
velocity are similar to those used for FCBI 2D elements. The interpolation
functions for pressure, temperature and coordinates are linear or bi-linear.
In 8-node and 4-node elements, they are the same as defined in Section

10.6 and Section 10.5 respectively. In 5- and 6-node elements, they are
defined respectively as
(1 r )(1 s )

(1 t ) r (1 s )
( hi ) = rs

(1 r ) s

t
and
h1 h4 1 r s
h h5 = r [1 t t ]
2
h3 h6 s

10.9: FCBI-C elements
10.9 FCBI-C elements
These elements consist of 2D triangle and quadrilateral, and 3D

tetrahedron, pyramid, prism and brick. All degrees of freedoms are defined
at the center of elements. Solution variables are assumed to be piecewise
constant in elements during computation, while the final solution is
interpolated at corner nodes for post-processing purpose. The following
figures show the definitions of all FCBI-C elements.
Figure 10.9 FCBI-C elements

FCBI-C elements are used for incompressible, slightly compressible

and low-speed compressible flows. Only segregated method is used for
FCBI-C elements.
Since the degrees of freedom are defined at the center of elements, it is
natural that the following options are not applicable to FCBI-C element:
constraint condition, skew system and specular radiation.

10.9: FCBI-C elements

Chapter 11: Solutions of nonlinear equations
Chapter 11 Solutions of nonlinear equations
Most finite element or finite volume equations for fluid problems are
nonlinear. The solution of a nonlinear equation must be obtained iteratively
by solving some linearized algebraic equations. Iteration at this level is
called outer iteration. In ADINA-F, we provide two methods in outer
iteration: the Newton-Raphson and Segregated methods. The segregated
method is only used for FCBI-C elements, while the Newton-Raphson
method is used for other elements. For linearized equations, we provide
direct sparse solver and some iterative solvers. The iteration in iterative
solvers is called inner iteration. These nonlinear equation methods and
linear solvers are described in this chapter.
11.1 Newton-Raphson method
We use the Newton-Raphson method in ADINA-F for all types of

elements except the FCBI-C elements. This method is usually used for the
system of all solution variables, although it may be modified using a better
solution strategy for certain problems. The user needs take no action in
respect of the modification.
A typical procedure is summarized here. Consider a nonlinear equation

f ( x ) = 0 , where x is the solution vector that includes all active solution
variables x = ( xv ) = (u , v, w, p, ,...) . We take the solution at time t as the
initial guess of the solution x 0 at t + t . The following equilibrium
iteration procedure is performed for obtaining the solution:

11.1: Newton-Raphson method
solve for x k : (f k 1 x) x k = f ( x k 1 )

x k
= x k 1
+ x k

for k = 1, 2,...

{
r = max x k max { x k ,108 }
v
v v }

if ( r ) stop

where is the tolerance (its default value is 103 ).

Depending on the type of problem, the overall outer iteration may be
altered, say, combined with a few Newton-Raphson iterations for the
solution system. One example is the iterative coupling of fluid-structure
interation, where the fluid and solid variables are solved separately, treating
the others are fixed during the iteration.
The residuals obtained after iteration 1 are also checked. If they are
smaller than a steady-state tolerance s , the program stops since the steady-
state solution has been obtained. Otherwise, the program continues for the
solution at the next time step.
11.2 Segregated method
The segregated method is an iterative algorithm for solving the

nonlinear fluid system. It is used here only for FCBI-C elements. It is
natural that the segregated method cannot be used for direct FSI coupling.
This method also cannot be used for specular radiation problems.
With the exception of continuity, all linear equations are derived
directly by integrating each particular variable regarding all other variables
as fixed. For continuity equation, the discretized momentum equations are
used to derive a pressure-correction dependency on corrections to velocity.
The linear solver is then applied to each variable in a certain order in the
outer iteration. In each time step, the segregated method can be typically
described in the following diagram.

for outer -iteration (k = 1, K max )

for (k fluid = 1, K fluid )

for ( k = 1, K )

vp vp

solve for u , v , w , p one by one
k k k k

for (kvp = 1, K vp ) if ( converged ) break

solve for k

if ( converged ) break

solve for K k
and then for k
k

for ( k = 1, K )

K K


solve for other variables one by one

k

solve for d solid

solve for d k

for (kd = 1, K d ) fluid



Each solve for step involves the solution of linearized equations. This
solution can be obtained by use of either the direct sparse solver or the
iterative AMG1, AMG2 or RPBCG solvers.
Since the solution variables are not coupled directly in one matrix
system, it is essential that the solution be converged, or else a non-physical
solution may be obtained. If an iterative solver is used for the linear
equations, it is also important to properly control the convergence in the
inner iterations. In the next sub-sections, these issues will be discussed.

11.2: Segregated method
11.2.1 Basic parameters
The coupling of velocity and pressure is one of the important steps

towards a converged solution. Coupling schemes we used are either
SIMPLE (default) or SIMPLEC. In SIMPLE method, usually two under-
relaxation factors, v for velocity and p for pressure, must be applied to
ensure the overall convergence. In SIMPLEC method, the velocity
relaxation factor is usually unnecessary and can be set to one.
The FCBI-C element is naturally first order in space since all

variables are defined at the element centers. However, it can become
second order in space. Scheme order provides the choice of first (default)
or second order accuracy in integration of the flux on element face. The
second order scheme is normally less stable and requires more iterations,
but the solution is more accurate than the solution of the first-order with
fixed number of elements. When element sizes tend to zero, the difference
of the two solutions diminishes too.
The nodal variables are interpolated whenever they are needed in

computation. The first approach is the neighboring cell-size-weighted
average. The first order nodal variable can be improved to become second
order (which is our default option). However, in problems with high
difficulty and/or very poor mesh quality, where the second order method
hardly converges, first order method may be used.
The default scheme for pressure interpolation at the faces uses a natural
control volume approach. This procedure works well for most smooth-
varying pressure problems. For problems with large body forces or second-
order time integration scheme, the linear scheme should be used. In this
scheme, pressure is computed using a linear interpolation of neighboring
element values.
The Pressure-Implicit with Splitting of Operators (PISO) pressure-

velocity coupling scheme is based on a higher degree of approximate in
outer iterations. In general, with the choice of PISO, the solution
convergence can be achieved with less iteration, but slightly more
expensive in each iteration. PISO is highly recommended for turbulence
models.

By default, the solver for the displacement of the moving mesh is the
same as that for fluid variables. That is, the unknown displacement is
defined at element center and the matrix of linear equations is non-
symmetrical. The nodal displacement is then obtained using interpolation
of element-center displacements. This is true even when the fluid solver is
sparse solver. Until the convergence in the outer iteration, the displacement
solution is usually not converged yet (with respect to current conditions) in
each sub-iteration. If the intermediate mesh causes difficulty in overall
convergence, one can specify more sub-iterations for displacement ( K d in
the diagram). The advantage of this selection is that the program uses no
additional storage in solving for the displacement.
However, if the memory is enough, the sparse solver can be directly

applied to solving the nodal displacement. The mesh solver allows the user
to select this option. Although extra storage is used, the nodal
displacement is directly solved from a symmetrical matrix. In addition, if
the original mesh is used as the background mesh in solving for the
displacement, the matrix will be factorized only once, thus the solver uses
minimum CPU time. For most moving-mesh problems, it is recommended
to use the sparse solver as the mesh solver and the original mesh as the
background mesh. It is more efficient, more accurate and more stable.
For some problems, it is more efficient to divide the computational

procedure into a few stages. For example, in the first stage, one computes
the fluid variables and, in the second stage, only computes the temperature
(if the fluid is independent of temperature). The user can achieve this by
defining active-variables in an execution. By default, active variables are
all the solution variables required in the analysis. An inactive variable is
involved in all computations, with the exception that it will not be solved
and updated.
In certain transient problems, good initial conditions are difficult to

specify. In this case, user may choose a restart run, using the initial
condition that is obtained in another steady-state analysis. In the segregated
method, we allow the two executions be performed in one analysis. The
first step steady-state solution option forces the program to obtain a steady-
state solution in the first time step and transient solutions in the remaining
time steps. To turn on this option, the user needs to input the maximum

number of iterations specifically for the first time step. Of course, this
analysis can also be performed using restart run.
Limit allows the user to specify certain extreme values. These values
are minimum values of turbulence variables and maximum factor for
viscosity during iteration, maximum CPU time allowed in the execution,
etc.
11.2.2 Outer iteration control
11.2.2.1 Stopping criteria
The convergence criteria used in outer iteration may differ

depending on the simulation being carried out. The user may want to
completely or temporarily terminate an execution of the program for
different reasons: solution is converged; CPU time limit is reached;
being able to check the quality of an intermittent solution and adjust
relaxation factors; etc. The program can be stopped in a number of
ways.
Tolerances on residuals are used to identify the convergence of

the solution. In segregated method, we always use a modified
L1 -normal for an array x

x e N1e i =1 xi
Ne
for element quantities

x =

x N1f i =1 xi
Nf
for element -face quantities

f
where N e and N f are number of elements and number of element-faces

respectively.
Let a discretized conservation equation for a general variable v at an

element be written as
e = a0v0 + nb anb vnb b0 = 0

A relative equation residual (ER) is computed as
re = e emod
where for the continuity equation
{
emod = max emin , min
e
e f , em }
and for other equations
{
emod = max emin , min
e
{
e f , min max
e
e f , em , ed } }
e e
In which, the parameters emin , min , max are pre-specified, em is the
maximum residual after I 0 initial iterations (by default it is 5 and can be
modified), and

e f = max e (source-term) dV , f ( flux -vector )i dS

e f

ed = nb
a0v0
A relative variable residual (VR) is computed as
rv = v vmod
{
vmod = max vmin , min
v
v , min { max
v
v , vm }}
v v
Similarly, vmin , min , max are pre-specified constants, v is the absolute
variable residual, and vm is the maximum variable residual after the
initial iterations ( I 0 ) .

Let f = v, e represent the variable and equation. Obviously, the

pre-defined constants are problem dependent. Being different from the
dimensionless constants (min
f f
, max ) , f min also has a specific unit.
Assuming that all solution variable quantities are of order one, we initially
set, for all variables and equations
( f min , minf , max

f
) = (105 ,1,10)
Note that, if variable quantities are far different from order one, these
constants must be changed. For example, if p ( 12 v 2 ) 1 as usually
happen in low-speed compressible flows, the momentum equation residuals
could become too small because its equation scale (including the magnitude
of pressure) is too large. This will cause the program to skip from solving
for the velocity and then converged to a wrong solution. This symptom can
be observed if velocity residual is always zero. In order to solve the
problem, one can simply reduce (emin , min
e e
, max ) for momentum equations
such that a non-zero velocity residual is observed.
A single variable/equation fi is said to have converged if
(r fi < f ) & (k K min )
where f and K min are, respectively, the tolerance and minimum number
of iterations requested. The convergence of a variable or equation is
defined as convergence of some variables or equations of the users
preference. Variable type and equation type on tolerance are parameters
used to identify the possible combinations of variables/equations. VR/ER
type can be ALL, NO, or any one the solution variable. For example, if
ALL is the option for VR, the variable is said converged if all solution
variables are converged. Similarly, if NO is the option for ER, all solution
residuals will be ignored in convergence checking. Finally, the overall
convergence is defined as either VR or ER is converged. The default set is
ALL for VR, pressure for ER (mass conservation) and the tolerances are
(v , e ) = (103 ,104 ) . Obviously, the ER type may differ depending on

the simulation being carried out. For instance, it may be set to either NO or
temperature (energy conservation) in a pure conduction problem. Notice
that, as convention, velocity, pressure and energy equations are also called
force (or momentum), mass and energy conservations respectively.
Maximum number of iterations is used to stop the execution after that

number of iterations for the current time step. The number of iterations
required may be quite different, depending on the accuracy desired in the
solution and the complexity of the problem, from a couple of iterations for
simple and small transient problems, to even several thousands for some
large steady-state problems. When program stops due to the specified
maximum number of iterations, one must check if the solution has truly
converged. If it is obviously diverging, the model must be examined,
possibly modified, and executed again. If it is converging, the problem can
be continued in a restart-run. Differing from a normal restart-run, one
should specify the initial time the same as the last time at which the
solution did not converge. If you intend to do so, simply ignore the
warning message regarding the unmatched initial time. In the model of
restart runs, the relaxation factors can be slightly adjusted too, according to
their history, to accelerate the convergence.
Minimum number of iterations is used to force the program to iterate at

least to that number. If it is specified the same as the maximum, the
program will stop after that number of iterations regardless of convergence
or divergence.
CPU limit provides another option to stop the program in a certain time
period. Again, the solution must be checked and, if necessary, executed in
a restart run as described before.
11.2.2.2 Relaxation factors
Since the fluid variables are not fully coupled into a system (as in the
Newton-Raphson method) on each iteration, under-relaxation factors are
usually required. The factors may differ depending on the solution variable
being solved, the difficulty of the problem and the mesh quality of the
Considering a typical linearized equation

a0x0 + nb anbxnb = e0
there are two types of relaxation factors that are used in the program. The
variable relaxation factor (VRF) is applied to diagonal of the matrix
a0 a0 v . The equation relaxation factor (ERF) is applied to the right-
hand side of the equation (residual) e0 e e0 .
VRF is used to overcome the numerical difficulties raised from non-
linearity and to improve the conditioning of the linear equation. It is also
served equivalently as a form of pseudo-local-time step (see false time step
option), i.e., a small VRF corresponds to a small time step and large VRF
corresponds to a large time step. Hence a small VRF will not only lead to a
faster solution of the linear equation in iterative solver, but also help on
overall convergence. It is also understood that too small a VRF (like a
small time step) will lead to slower global convergence.
ERF is used to overcome the overall convergence difficulty raised from
a large source term in a linear equation. This source may be physically real
(e.g., in turbulence equations) or a numerical error (a bad solution
approximation at the start of each iteration). A small ERF adds the source
incrementally in the outer iteration and, therefore, improves the global
convergence.
Frequently, ERF is used for pressure while VRF is used for other
variables. We treat these factors as primary while others as secondary.
Notice that the primary pressure-relaxation factor is applied to the equation
to enforce the mass conservation from intermittent solutions.
Transient analyses differ from steady-state analyses by the presence of

the dynamic term that may stabilize solution equations (except the
continuity equation in incompressible flows). For small time step length
(used in short-wave problems, for instance), VRF is usually unnecessary,
meaning that it can be set to one. However, if the time step is not small
(used in long-wave problems, for instance), the solution could diverge. In
this case, the primary factors for transient analyses must be adjusted
accordingly, as those for steady-state analyses. Since we allow mixed
transient/steady analyses in a single model, both relaxation factors for
steady-state and transient analyses are required.

In summarizing, we have introduced the primary factors for steady-state

analyses (PRI-SS), the primary factors for transient analyses (PRI-TR) and
secondary factors (SEC). Their default values are listed in the table below.
Figure 11-1 Default set of outer relaxation factors in segregated

method
Variable PRI-SS PRI-TR SEC

Velocity 0.75 0.75 1
pressure 0.3 0.3 1
displacement 1 1 1
turbulence-K 0.97 0.97 1
turbulence- 0.97 0.97 1
other variables 0.99 0.99 1
11.2.2.3 Convergence information
At the end of each outer iteration, information is printed into the *.out
file. A typical output in outer iteration is in the format
ITER, ER-P, max(VRs), ER-V, VR-V, VR-P, (VAR, CELL, MAX-VR, CELL, MAX-ER)
where cell is the global element number (counted from the first element
of the element group 1 to the last element of the last element group) and
ITER = outer iteration number

ER-P/V/etc. = relative equation residuals for pressure, velocity, etc.
VR-P/V/etc. = relative variable residuals for pressure, velocity, etc.
Max(VRs) = maximum value of all relative variable residuals
VAR = variable name for which the maximum equation residual occurs
MAX-VR, CELL = maximum absolute variable residual for VAR
MAX-ER, CELL = maximum absolute equation residual for VAR
The user can choose the variable for which the maximum residuals are
printed. The default is the variable that has the maximum absolute residual.

11.2.3 Inner iteration control
A direct or an iterative solver can be used for solving the linearized

equations. Inner iteration refers to only iterative solvers. The available
iterative solvers associated with the segregated method are AMG1, AMG2
and RPBCG.
One simple strategy would be forcing a full convergence in iterative

solvers for every solution variable. However, since the goal of iteration is
to reach overall convergence of the nonlinear equation, an exact solution is
not required because this is just one step in the nonlinear outer iteration.
Managing the inner iteration to a minimum level with yet a reasonably
good intermittent solution may save much solution time.
11.2.3.1 Inner convergence criteria
A proper measurement of the quality of a solution leaving the iterative

solver is the first step to efficient inner iteration. We consider a solution
good enough if the residual in the inner iteration is smaller than a
percentage of the critical residual in outer iteration convergence. The
percentages used here are called variable and equation reduction numbers
( v and e ) respectively. Beside the reduction numbers, maximum
number of iterations I max and minimum number of iterations I min can also
be specified.
Consider an algebraic equation Ax = b and recall that the unknown
vector x and its right-hand side b are the variable and equation residuals
respectively in outer iteration. Let i be the inner iteration number, we
define three types of residuals in inner iteration
variable residual (VR ) : rvi = x i x i1 vmod

equation residual ( ER ) : rei = Ax i b emod
reducted residual ( DR ) : rdi = Ax i b b

where vmod and emod are the variable and equation scales, respectively,
determined in outer iteration. We define any one of the followings as the
convergence criterion in inner iteration:
(1) re0 ee
( 2) (rei 0.001ee ) or (rei ee ) & (i I min )
(3) (rdi 0.001e ) or (rdi e ) & (i I min )
( 4) (rvi 0.001vv ) or (rvi vv ) & (i I min )
where v and e are the tolerances defined in outer iteration. The
parameters that control the solution process are different for each
variable/equation. The default set is
Figure 11-2 Default set of inner relaxation factors in segregated

method
Variable e v I min I max

pressure 0.01 0.01 2 300
velocity 0.01 0.01 2 50
others variables 0.1 0.1 2 50
We note that, if the inner convergence criterion (1) is satisfied, no linear

equation solvers will be invoked (even a direct solver is selected), leaving
the solution unchanged at the end of that outer iteration. It is important to
enforce a tighter convergence of the continuity equation than other
equations. A false inner convergence of the continuity equation will
usually lead to a diverged solution in the outer iteration.
11.2.3.2 Inner convergence information
Whenever an iterative solver is invoked, inner iteration information is

printed into *.log file in the format shown below, at the end of every
selected number of outer iteration interval for print (it is 1 by default),

V: ITE, I, ED, ER, VR, EI, E0, EM, VI, VM, V0 =
where
V = abbreviation of variable name

ITE = outer iteration number
I = inner iterations performed
i i i
ED, ER, ER = rd , re , rv
EI, E0, EM = Ax i b , b , emod

VI, V0, VM = x i xi1 , x i , vmod
At the end of model execution, the total number of inner iterations

performed is printed.
11.2.4 Comments and tips on convergence
The segregated method requires far less memory than the Newton-
Raphson method. The memory usage is basically linear in relationship
with the number of elements if an iterative solver is used. The convergence
could be very fast too (say that CPU time is linearly related to the number
of elements), if the outer and inner solution processes are properly
controlled. These advantages make this method very useful in solving very
large problems. However, a drawback of this method is the relatively high
difficulty level of having the solution fully converged. Remember that not-
fully converged solutions may be physically unacceptable. For this reason,
we have given users more controls of this method than the Newton-
Raphson method.
Model preparation and testing is definitely the first important step

towards a good solution. Experience in model analyses has shown that the
most likely cause of a divergence process or of convergence to an
obviously incorrect solution is that the model has not been set up as
intended. If the model is believed to have a physically stable and unique
solution, the error is likely in the input. For the purpose of testing the
model, while keeping the time consuming part in the model (such as the
geometry, element groups, etc.), one can always use the smallest and
simplest model possible. For example, by reducing the number of
elements, using larger viscosity and a smaller buoyant force, etc. Using

this model, the problem can be easily and quickly identified. If the input is
correct, try to identify the cause of convergence difficulty.
Reducing variable relaxation factors can be an efficient way to

overcome the convergence difficulty if it is not a complex problem.
Usually the primary pressure relaxation factor need not be adjusted for
majority of problems. Note too that small relaxation factors may lead to
false convergence.
Unless there is a special purpose, avoid using variable under-relaxation

factor for pressure, since the intermittent solution may violate the mass
conservationan important quantity used in other convection-diffusion
equations. Notice that the variable relaxation factor for pressure is the
secondary factor.
To obtain a steady-state solution of very large size, try to break it into a

few consecutive runs. In each run, adjust control parameters if necessary
based on the printed convergence history.
If the cause of divergence is due to sudden large loads (normal traction,

prescribed velocity, etc.), add the loads incrementally in a few time steps,
say in 3 time steps each of them composing 1, 11 and 89 percents of loads
respectively. Usually the most difficult times are the transitions of flow
patterns, for example, in the first time step, when flow from stationary
changes to a complex flow pattern.
If a steady-state analysis is neither converging nor diverging and the

situation persists even the relaxation factors are reduced, it is possible that a
steady-state solution does not exist. More precisely, it may not exist for the
input conditions. A typical example is the laminar flow of high
Reynolds number in a backward-facing step, in which the circulation
reaches the outlet where a constant pressure is specified. The pressure
condition that is specified improperly inside the circulation may cause the
numerical oscillation. In this case, one has to use a turbulence model or a
smaller Reynolds number in the laminar model if physically acceptable.
Another efficient solution is to use time-marching technique. That is, one
can perform a real transient analysis with many time steps (may be in a few
runs as well), until the solution is visually unchanged. Since the final

solution is of interest, a loose tolerance can be used in outer iteration, using

equal minimum and maximum numbers of iterations or larger tolerance.
In transient analyses, an immediate divergence usually indicates a large

time step. Use CFL~1 as a criterion to check if the time step is small. For
small time step as in many short-wave problems, the default set of primary
relaxation factors for transient analyses is adequate. A not-fully converged
solution may also be acceptable depending the accuracy requirement. On
the other hand, if a large time step is used, as in many long-wave problems,
the set of relaxation factors must be adjusted to the set for steady-state
analyses. The convergence must be enforced as well in order to obtain
accurate transient solutions. One may also use the composite scheme of 2nd
order in time integration to improve the solution quality.
In a turbulence model, experience has revealed that local velocity-

pressure iteration ( K vp > 1) may improve the overall convergence. We
note that K vp = 2 corresponds to the Pressure-Implicit with Splitting
Operators (PISO) method if both predictor and corrector are implicitly
treated. Furthermore, local turbulence-variable iteration ( K K > 1) may
also improve the overall convergence.
In a fluid-structure-interaction model, a few iterations of fluid variables

is necessary ( K fluid = 2 ~ 10) . For very difficult problems, a large K fluid
may be specified. In this case, the iteration in fluid model is controlled by
the specified tolerance.
In buoyant-driven flows, if difficulty in convergence occurs, try to use

secondary under-relaxation factor for velocity. This is equivalent to
adding buoyant force gradually in iterations. One can also introduce a local
velocity-pressure-temperature ( K vp > 1) iteration, say 2 to 3 times, to
tighten the coupled relationship between the velocity field and the
temperature. In principle, if the order of the right-hand side of an equation
is much higher than the order of the magnitude of the matrix, the
corresponding secondary factor may be reduced.

A proper unit system can be efficient in minimizing the round-off errors

and reducing the chance of instability caused by round-off errors. One can
either input data in a proper unit system, or choose proper nondimensional
scales, such that all solution variables are of order one. If the linear solver
is always skipped from solving a variable (exact zero residual is printed in
the outer iteration), the residual of that equation is too small. This may be a
sign that an improper system of units is used. Of course, one can also
adjust the constants defined in the scale for that equation.
A mesh is said to be orthogonal if the line that connects the centers of

the two neighboring elements is perpendicular to the face they share. If a
2D 4-node element or 3D 8-node is applied to a simple-geometry problem,
the mesh is usually orthogonal or nearly orthogonal. For an orthogonal
mesh, large variable relaxation factors may be used (close to 0.99~1 if
possible without causing divergence) and, accordingly, fewer iterations
may be performed. On the contrary, for a free-form or distorted mesh,
smaller variable relaxation factors are usually necessary, and more
iterations will be performed.
For an equation with a dominant diffusive term, the convergence could

be very slow, unless a larger variable relaxation factor is used (set the factor
close to 0.99~1 if possible without causing divergence).
For pure conduction problems (without the presence of a fluid field),

the mass-conservation must not be checked, or, have the energy equation
residual option checked.
For moving mesh problems, try to use the original mesh as the
background mesh if possible (that means the choice of the original mesh
will not cause difficulty in overall convergence). If there is enough
memory, always use the sparse solver as the mesh-solver. The use of the
two in the same model can achieve the best in speed, accuracy and
convergence regarding moving mesh.
The solver AMG1 should be always the first choice of iterative solvers.
Breakdown in iterative solvers usually indicates a singular system, meaning
some errors exist in the model that should be always checked first. AMG2
can be used, in very rare situations, if AMG1 has failed in inner-iterations.
It uses slightly more memory.

A well-converged inner iteration is important in reliably obtaining

solutions of difficult problems. This is particularly true for the pressure
equation. Carefully check the information printed out in the *.log file. To
enforce a full convergence, reduce the reduction numbers and increase the
maximum number of iterations. For difficult problems, a much larger
number of inner iterations for pressure may be required (say 400~800).
Experience has showed that not-fully-converged pressure solutions in inner
iteration usually result in divergence in outer iteration, while not-fully-
converged velocity fields result in neither divergence nor convergence in
outer iteration.
Inner iteration tends to be most difficult at initial iterations (meaning

more iterations are performed) and becomes easier when solution is closer
to convergence. Too-difficult inner convergence indicates a poor
conditioning of the matrix, suggesting the application of the loads over a
few time steps or reducing the variable/equation relaxation factors. On the
contrary, a too-easy inner convergence indicates the tolerance is too loose.
A tighter reduction number or a larger minimum number of iteration may
be necessary.
11.2.5 Limitations
The segregated method is only used for FCBI-C elements. It cannot be

used for direct FSI coupling, specular radiation, constraint condition, skew
system and ATS option.

11.3 Solvers of algebraic equations
Solving a system of linear equations Ax = b is essential to the

computations in ADINA-F. Most of the storage assigned to the program is
occupied by the matrix system and most of the CPU times used by the
program is devoted to the solution of the algebraic equations. Various
solvers are available for solving the linear equations that are resulted from
the Newton-Raphson or Segregated methods. They are briefly introduced
here.
11.3.1 Gauss elimination method
This method is probably the most famous direct solution technique. In

ADINA-F, the active column solution is implemented (COLSOL). This
method should not be used, in general, because the sparse solver described
in the next section is much more effective. It is introduced here briefly to
be able to better describe other solvers later.
In this method, the matrix structure is initially optimized to have short
active columns (a low skyline) according to the finite element connectivity,
the constraint relations and to have sufficient fill-ins to not encounter zero
pivots. The area within the skyline stores all matrix elements before and
after the equation solution.
First, the matrix is decomposed as in the Gauss elimination (without
pivoting),
A = LDU
The diagonal elements of the matrices L and U are all one and the other
nonzero elements are located in the lower and upper positions of column i ,
respectively. The matrix D is diagonal. These structures allow us to store
all the matrices in the original matrix A.
Once the decomposition is complete, the solution of the equation can be
obtained by:
X = A1b = U 1 D 1 L1b

11.3: Solvers of algebraic equations
Note that the matrix has never been inverted.

Calculating the complete LDU decomposition without taking advantage
of possible no-fill-ins (see Section 11.2) is in general a very expensive
procedure according to current standards of matrix computations.
The matrix requires storage of N d mb words, where mb is the averaged
bandwidth.
The algorithm requires arithmetical operations of the order of N d mb2 .
Furthermore, the direct decomposition of the matrix may have a large
round-off error even if the matrix is well-conditioned. For example, for the
simple matrix
1
1

the process fails to give the solution when = 0 if no pivoting is used

(notice that the conditioning of the sample matrix is good). It is
understandable then that the LDU decomposition of this sample matrix will
have a large round-off error if is small.
The methods of column and/or row interchanges (pivoting) are very
effective in overcoming these problems but add much computational effort.
For this reason, in ADINA-F, the variables are ordered to not encounter a
singular system. The solution then does not allow for row/column
interchanges and hence, assumes that these are not needed.
When there is not enough memory to hold the matrix in core, this solver
automatically goes out-of-core. Then the computational speed is slowed
down by a factor of order nb2 , where nb is the number of blocks of the
matrix.
11.3.2 Sparse solver
The sparse solver is a direct solver based on the Gauss elimination

method. However, it differs from the traditional direct solvers in
preserving the sparsity of the matrix, thus reducing dramatically the storage
and computer time required by traditional direct solvers.
The assembled matrix is initially stored in compact form, according to
the finite element connectivity and the constraint relations. It contains only
logically nonzero elements in the assemblage level.

The sparse solver then reorders the finite element equations by keeping
the sparsity of the original matrix at its best. The sparse solver also
performs a symbolic factorization to locate those elements that will become
nonzero during the factorization. These elements are called fill-ins. The
number of fill-ins determines the additional storage that is required in the
solver. The number of fill-ins together with the structure of the reordered
matrix determines the effectiveness of the computations performed in
factorizing the matrix.
Here is a simple example that explains how the sparse solver works.
Consider a heat transfer problem with 5 degrees of freedom defined at the
corresponding nodes (see the following figures).
2 3 3 4
1 5
4 5 1 2
(a) (b)
Figure 11-3 Ordering in sparse solver
Fig.11.1(a) shows a possible numbering of equations. The corresponding

symbolic sparse matrix structure is

+ +

A= + +

+ +

+ +
where * indicates the original nonzero elements and + indicates the

additional fill-ins. The matrix in this example eventually becomes full.
This ordering is therefore not effective.
If the order of the equations as shown in Fig.11.1(b) is used, we have
the following symbolic sparse matrix structure


A=

where the empty elements are neither stored nor computed. There are no
fill-ins in this example; therefore, we have achieved the best order. This is
exactly what the sparse solver aims to achieve in reordering: to minimize
the number of fill-ins as shown in this example.
The minimum degree algorithm and the multi-graphics processes are
the most popular schemes of symbolic factorizations used in sparse solvers.
We use a hybrid of these two methods.
In addition to preserving the sparsity of the matrix system, we also
apply the super equation concept in regrouping equations into groups,
such that equations within each group have the same symbolic structure. A
group is further cut into panels that are formed by blocks, whose size is
determined based on the cache size of the machines used. The Gauss
elimination procedure is then performed in blocks instead of in columns.
This amalgamation reduces the data traffic and cache miss and therefore
reduces overall computation times.
The sparse solver is effective, although iterative solvers may be faster in
certain applications. It is usually about 10-100 times faster than traditional
direct solvers and requires storage of about 2-10 times than the iterative
solvers.
The sparse solver can also go out-of-core in case the machine memory
is not enough to hold the ordered matrix in the core. In this case, the
memory must be explicitly assigned to the sparse solver. The required hard
disk space will be automatically used whenever additional storage is
needed. Several important observations regarding the sparse solver are
listed here:
If no memory is assigned to the sparse solver, it will try to allocate the

memory that is required for an in-core solution. In case insufficient
memory is available, the sparse solver prints information and ADINA-
F terminates the execution. There is a possibility that the operating
system goes to swap on disk when the required allocated memory is not
sufficient. In this case, the sparse solver could be extremely slow (it is
out-of-core, but very inefficiently managed by the operating

system!). In iterative computing of the two-way fluid-structure

interaction problems, the sparse solver frees the memory every time
after the solid and fluid models are solved. There is a possibility that
insufficient in-core memory is encountered during FSI iterations if the
machine is shared by more than one job.
If a specific memory is assigned to the sparse solver, it assumes an in-
core procedure if the memory is sufficient. Otherwise, it will work on
the hard disk, enforcing an out-of-core solution.
The speed of the out-of-core sparse solver does not depend much on the
memory assigned to it and is about two times slower than its in-core
counterpart. Note that the original assembled matrix must always be
in-core, even when the out-of-core sparse solver is used.
Using these facts, we may efficiently use the sparse solver as suggested
below.
Only when it is absolutely certain that there is sufficient memory on the

machine for an in-core solution, it is not necessary to assign memory to
the sparse solver.
If it is doubtful whether the solution can be performed in-core, assign a
reasonable amount of memory to the sparse solver.
If the out-of-core solution becomes the only choice, assign just enough
memory to ADINA-F and assign the rest of available memory to the
sparse solver (see Chapter 13 for more details).
In iterative computing of two-way fluid-structure coupling problems,
always assign to the sparse solver the memory that is continuously
available on that machine.
If the operating system automatically goes to swapping in case of
memory shortage, always assign memory to the sparse solver.
11.3.3 Iterative methods
There are five iterative solvers available in ADINA-F:
(1) Right preconditioned biconjugate gradient method (RPBCG)

(2) Right preconditioned generalized minimal residual method
(RPGMRES)
(3) Algebraic multi-grid method (AMG) for system matrix

(4) Algebraic multi-grid method-1 (AMG1) for segregated method

(5) Algebraic multi-grid method-2 (AMG2) for segregated method
For illustration purpose, we will only present details of RPBCG in this

section.
When a preconditioner is required, we apply a right preconditioning to
the nonsymmetric linear system of equations. Instead of solving the
original equations Ax = b , we solve the equivalent system
Ax = b
where A = AB 1 and x = Bx .
A typical preconditioner is obtained using the incomplete LDU
decomposition of the matrix A. In this method, B is an approximation to
the complete factorization of the matrix A obtained in the Gauss elimination
method described in Section 11.1. The incomplete factorization is
performed as follows. Consider only those locations where A has nonzero
elements; at these locations, perform operations corresponding to the usual
decomposition procedure using elements only from those locations where A
has nonzero elements. The sparsity is then completely preserved in the
resulting incomplete factors.
In this structure, the required storage is dramatically reduced. The
matrix B only requires the same storage as the matrix A. With an additional
locator array, the total storage requirement is about 5 m nd Nd words in 32-
bit machines, where the factor nd varies from 1 to 7 corresponding to the
number of solution variables and m varies from 8 for 2D 3-node triangular
elements to 3125 for 3D 27-node brick elements, and Nd is the number of
equations.
In the RPBCG method, we apply the biconjugate gradient method to
solve the equation. The algorithm can be summarized as follows.

Algorithm RPBCG
x = 0, r = r = p = p = b, = (r , r )
for k = 1, 2,...
z = B1 p, z = Az , = ( z , p )
x = x + z
r = r z
if (converged ) stop
z = AT p, z = BT z
r = r z
= ( r , r ), = , =
p= r +p
p = r +p
The convergence criterion is
(1) r < 1 and r b < 2 or

(2) ( r < 3 )
in which 1 , 2 and 3 are predetermined tolerances. If the convergence

criteria are not satisfied after K iterations, ADINA-F prints a warning
message and terminates the current iterations in the solver. However, the
program continues.
The convergence control in the algebraic multi-grid method is very

much like in the RPBCG and RPGMRES methods. However, it requires
slightly more storage in manipulating the restriction and prolongation
operators in the coarsening and refining procedures respectively. The
default values of the control parameters for iterative solvers are
K = 3000, 1 = 106 , 2 = 104 , 3 = 108

Although the convergence of the iterative methods has been proven in

exact arithmetic for positive definite symmetric matrices, convergence is
not guaranteed when nonsymmetric indefinite matrices (such as in fluid
flow solutions) are considered. Nevertheless, numerical experiments show
that iterative methods can converge fast if the conditioning of the matrix is
fairly good and the number of the equations is not too large (typically in
less than 1000 iterations if the number of equations is less than 200,000).
For large equation systems, it is generally believed that the incomplete
factorized preconditioner is farther from the original matrix and thus more
iteration steps are required. Furthermore, the iterative methods require
much less memory than direct solution methods.
The ADINA AMG solver is designed for large systems (more than
200,000 equations) and can be efficiently applied to most fluid flows.
Currently, AMG solver cannot be used in direct computing of fluid-
structure interactions, and to problems with sliding-mesh, gap and slip-wall
boundary conditions. Its efficiency may be somewhat diminished in
turbulence models or if more than one coordinate system is used.
ADINA AMG1 and AMG2 are specially designed in association with

the segregated method. Their convergence control is introduced in the
description of inner iteration of the segregated method.

Chapter 12: Other capabilities
Chapter 12 Other capabilities
12.1 Automatic time-stepping CFL option
It is well known that explicit methods are unstable if the time step
length does not satisfy the CFL (Courant-Friedrichs-Lewy) condition
x
t min tc
w + 2 x
where w is given by u + c in high-speed compressible flows or the

magnitude of the velocity v in other flows.
Implicit methods have no such restriction. However, the CFL condition
is only a necessary condition obtained from a linear stability analysis.
Fluid problems are always nonlinear and the nonlinear terms are dominant
in many practical problems. Furthermore, experience has shown that a
reasonable time step length may not only accelerate the global convergence
in many cases, but also be crucial for stability.
We have developed the automatic time-stepping (CFL) option. In this
option, a user specified CFL number, CFL , plays a role of relaxation, rather
than a corresponding real time. The dynamic terms in the governing
equations are discretized
t +t k t +t
f f k t f k t +t
f k t+t f k1
= +
t t
where is a fictitious time step length and the right superscript

represents the iteration number in equilibrium iterations. The first term on
the right hand side is the Euler -method term that has been introduced in
the formulations of fluid models. The second term is introduced as a
fictitious time derivative that approaches zero when the iteration converges.
It is also clear that the fictitious time step increases the diagonal dominance
of the global matrix and thus improves its conditioning.

12.1: Automatic time-stepping CFL option
Focusing on the magnitude of the diagonal term (coefficient of the

current variable t +t f k ), we choose such that
1 1 1
+
t CFLtc
By default, when this option is not applied, the CFL number is infinity.
For most of the practical problems, CFL 102 to 107 can be used. For
explicit time integration, CFL = 0.8 to 0.9 can be a good choice. In
transient analyses, a proper use of the CFL number may improve the
equilibrium iteration convergence at each time step. In steady-state
analyses (where t is infinity), the convergence here means the final
converged solution and the equilibrium iterations act as time steps (known
as the time-marching method for steady-state solutions). In general, the
larger the CFL number, the less stable the scheme or the more difficult the
convergence, while the smaller the CFL number, the more stable the
scheme and the less difficulty in convergence. However, the drawback of
the smaller CFL number is a slower convergence. Considerably more
iterations are required if the CFL number is too small. The magnitude 1
should be regarded as the lower limit of CFL for implicit methods. An
optimized CFL number can be obtained by numerical experiments and
experiences. Usually one CFL number can be used for a group of models
that have similar geometry, materials and boundary conditions.
Using this option and the control of loading increments in time steps
can overcome many stability or convergence problems.
12.2 Automatic time-stepping ATS option
The automatic-time-stepping (ATS) option is used to obtain a

converged solution when equilibrium iterations fail because too large a time
step (giving too large load increments) is used.
In this option, the iterations in the implicit time integration are
performed as described in related chapters. In case convergence is not
reached, the program automatically subdivides the total load step increment
into two equal sub-time-steps and tries to compute solutions at these time

steps. This procedure continues if the iterations still do not converge until a
proper time step is found or the maximum subdivisions allowed have been
reached.
In the first case, the program continues the next sub-time-steps until the
user specified time step is reached. It may increase or decrease the sub-
time-step again in order to achieve the best convergence rate.
The user controls the maximum number of subdivisions. When too
many subdivisions are made, the program stops, indicating divergence of
the computations at the current time step.
The converged solutions are saved only at the user specified time steps,
regardless of the subdivisions that have been made.
The ATS option may help solution convergence in several ways as
listed here.
In transient analyses, the smaller time steps improve the matrix

conditioning.
When boundary conditions are time-dependent, smaller increments of
time steps are usually equivalent to smaller increments of loads
provided that the loads are monotonically increasing.
However, one must be cautious when using the ATS option. Since the
time step length is successively cut into two equal sizes, the process can be
very expensive if the original user-specified time step is far from the final
one used. For example, 10 subdivisions will result in a final step of 0.001
from an original user-specified step length 1. In each subdivision, the
maximum number of iterations will be performed to indicate an
unsuccessful time step size so that the next subdivision is necessary.
12.3 Restart analysis
During the execution of a problem, the step-by-step solutions are

automatically saved to a restart file every N time steps, where N is
chosen by the user. Output to the restart file always overwrites the current
contents of the file, i.e. the restart file only contains the solutions of the last
saved time step. Therefore, if it is expected to use a restart file more than
once, make sure to have a backup of that file since it will be overwritten in
a restart run.

12.6: Constraint conditions
Now let us call the previous problem A. Using the solution of the
problem A as an initial condition, another solution can be computed (call
the problem B), starting at the last time the solution A is saved. The
problem B is called a restart run. In subsequent restart runs, the geometry
and most element data must be the same as those in the first run. However,
the following changes are allowed:
Time step control data, including number of time steps, time step
lengths and time functions.
Global control flags, including analysis type (steady-state analysis and
transient analysis), CFL option, ATS option and their associated
parameters.
The parameters associated with the boundary conditions. For example,
the values of prescribed conditions, distributed loads, parameters in
special conditions, etc.
The parameters associated with the material models. For example, the
viscosity, the heat conductivity, etc.
Restart analyses can be used for moving mesh problems, including

particularly fluid-structure interactions.
12.4 Mapping solution
Upon the request of the user, a mapping file can be created that contains
the problem solutions in a neutral format. The solution includes the
independent nodal solutions and element connectivity. The solutions are
saved at each solution time whenever the porthole file is saved. The
mapping file can be in a binary or an ASCII format.
The mapping file can be used to transfer problem solutions from one program
data set to another. For example, the ADINA-F generated mapping file can be
directly loaded through the AUI and the temperature solutions in the file are then
transferred to nodes of an ADINA solid model for the purpose of a thermal stress
analysis. Of course, the overall geometric domains in ADINA-F solid element
groups and the ADINA solid model must be coincident.
The mapping files can also be used to transfer the solutions between
two problems of the same program. For example, the solution of a problem
in ADINA-F can be transferred (interpolated) to another problem in

ADINA-F to be used as initial conditions. Unlike restart analyses, the two

problems can be completely different except the coincident domains.
12.5 Skew system
Skew systems are right-handed orthogonal systems that have arbitrary

orientations with respect to the global Cartesian coordinate system. They
are applicable in ADINA-F for ease of modeling.
The a-b-c axes of a skew system are defined using the input of direction
cosines, Euler angles or axes.
Skew systems can be assigned to any nodal point or through the
associated geometry in AUI. Once a node is associated with a skew
system, the velocities and loads are referred to that skew system.
In many situations, conditions in skew systems can be reached by other
options. For example, on a slip wall where only the normal velocity is
zero, one could apply a skew system to its normal direction and then fix the
velocity in that direction. When the wall geometry is highly irregular, the
application of skew systems becomes not only tedious, but also almost
impossible. In this case, simply applying the special boundary condition
slip-wall is a much better choice.
In another example shown in the following figure, the velocity shall be
prescribed in the radial direction. One way, of course, is to apply skew
systems along the inlet line and then apply the velocity in the radial
direction. However, a simpler method is to use a proper spatial function.
R v
s
2
1
y
Figure 12.1 Prescribe a radial velocity

12.6: Constraint conditions
For example, we can prescribe the inlet velocity as
v = v f (t ) g 2 ( s )
w = v f (t ) g3 ( s )
where f (t ) is the associated time function and the gi ( s ) are spatial

functions which are, in this case,
s
g 2 ( s ) = R cos + 1
R
s
g3 ( s ) = R sin + 1
R
12.6 Constraint conditions
In some analyses, it is necessary to impose constraints between

solution variables. For example, a periodic boundary condition on
solution variables defined on a part of the boundary is to be identical to
the boundary condition defined on another part of the boundary.
In ADINA-F, the following conditions can be specified
Nk
fk = k j f j
j =1
where f k is a nodal solution variable such as velocity, pressure,

temperature, etc., k j are multiplying factors and N k is the number of
independent variables used. The solution variables on the right-hand side
are called master degrees of freedom or master solution variables, while
the one on the left-hand side is called the slave degree of freedom or slave
solution variable of these masters. The constraints can be explicitly
specified for any of the nodal variables.
Fluid flow problems are usually highly nonlinear. This is particularly
true in compressible flows. Special caution must be made when using

constraint equations. Constraining a single solution variable (say only

pressure) is generally not a good choice. When using constraint
equations, it is necessary to focus on the condition of entire geometric
entities rather than considering only individual solution variables. For
example, when considering a period boundary condition for surface 1 and
surface 2, all solution variables of surface 2 should be constrained to the
corresponding solution variables defined on surface 1.
12.7 Conjugate heat transfer problems
ADINA-F can be used to analyze problems of heat transfer in a domain

that includes both fluids and solids, the so-called conjugate heat transfer
problem. The heat transfer in solids occurs only by conduction. The heat
transfer in fluids consists of conduction, convection and specular radiation,
if specified.
The solid part can be defined through the definition of solid element
groups (a special type of fluid element group). Defining a solid element
group is the same as defining a fluid element group except that the solid
condition is explicitly specified. The velocity degrees of freedom on
nodes are then automatically deleted, including the interfaces with fluid
element groups. The pressure degrees of freedom are also deleted except
for the nodes shared with fluid element groups (see the illustration in the
following figure).

12.8: Element birth-death option
computational
domain for v, p,, etc.

group group
v=0
computational
domain for
Figure 12.2 Sketch of conjugate heat transfer
The computational domain for the continuity equation and the

momentum equations consists of all fluid element groups. Since the
velocity degrees of freedom have been deleted on the interfaces between
fluid and solid groups, a no-slip wall condition is implicitly applied. On the
other hand, since the computational domain for the energy equation
consists of the fluid and solid groups, the fluid-solid interface is not a
boundary and therefore, no condition is required. The temperature solution
is continuous across the interface.
12.8 Element birth-death option
The element birth and death option is available in ADINA-F. This

option enables the modeling of changes in the computational domain. For
example, an increased domain can be modeled by the element birth option,
a reduced domain can be modeled by the element death option and an
increased domain followed by a reduced domain can be modeled by the
element birth-then-death option.
The boundary conditions must be applied to the domain that covers all
possible live elements.

If the element birth option is used, the element is added to the total
system of finite elements at the time of birth and remains active for all
times thereafter. The boundary conditions that are linked to the element
will also be alive.
If the element death option is used, the element is taken out of the total
system of finite elements at the time of death and remains inactive for all
times thereafter. The boundary conditions that are linked to the element
will also be inactive.
If the element birth-then-death option is used, the element is added to
the total system of finite elements at the time of birth and remains active
until the time of death. The element is then taken out of the total system of
finite elements and remains inactive for all times thereafter.
12.9 Pressure datum
A pressure datum can be specified at a geometric point or node. Once

this feature is used, ADINA-F shifts all of the pressure solutions by a
pressure increment p (positive of negative)
p = p + p
where, the increment pressure can be either one of
(1) p = pd (t ) pi
(2) p = pd (t )
in which, i is the nodal label where the pressure datum pd (t ) is specified.

In the first case, the resultant pressure p is equal to the specified pressure
datum at the specific location. In the second case, it always adds the
specified pressure datum.
This procedure of shifting the pressure is only performed in the output
phase of the program. In particular, all stresses and forces that are printed,
saved and exported are calculated using the shifted pressure (including the
stress applied to structures in FSI problems). After output, the pressure
solution is shifted back to the original solutions.

12.9: Pressure datum
The pressure datum feature is useful in fluid-structure interaction

analysis as shown in Fig.12.3. In the physical model (a), the normal-
traction at the outlet could be a very large number. Since the pressure
solution in incompressible flows can be shifted by a constant, such a big
number can be avoided in the computation by specifying a zero normal-
traction as shown in (b). However, the total stress acting on the structure
must include the pressure that is omitted in the fluid model solution. This
is accomplished by simply adding the omitted pressure back into the fluid
stress calculated as shown in (c). Note that the pressure datum can be
negative too. In this case, it is subtracted from the computed pressure
solution that is applied to the structures.
structure
v fluid nn = p
(a)
structure
v fluid model: stress = nn = 0

(b)
specified pressure datum p
solid model is subject to fluid stress pI

(c)
Figure 12.3 Application of pressure-datum feature to fluid-

structure interaction problems.

12.10 ALE formulation and leader-follower option
The moving mesh conditions are encountered if any boundary condition

is a moving boundary condition. These conditions are given by moving
walls, free surfaces, fluid-fluid interfaces and, of course, most importantly
the fluid-structure interfaces. The discussion in this section covers all types
of moving mesh problems.
Physically, moving boundary conditions prescribe the boundary nodal
positions. For example, the displacements of nodes on FSI boundaries are
determined by the structural displacements; the displacements of nodes on
moving walls are determined by the time functions; the displacement of
nodes on free surfaces is determined by the free surface equations; and so
on. However, the boundary nodes can be moved along the geometric
boundary that changes position. The interior nodal movements are quite
arbitrary but need to be controlled to keep a good mesh quality. Of course,
there is no unique solution to establish a good mesh.
12.10.1 ALE formulation
In order to automatically determine the displacements of those nodes

that can arbitrarily be moved, we currently solve the Laplace equation
2 d = 0 (12.1)
where d is the increment of the displacement. The latest displacement is

then updated by adding the incremental solutions. A finite element method
is applied to solve equation (12.1) based on either the latest nodal positions
or the initial nodal positions.
However, Laplacian solutions cannot guarantee a valid mesh even in
some simple domains. The following figure shows such an example.

12.10: ALE formulation and leader-follower option
unacceptable
element
solid
fluid
Figure 12.4 A typical case of reaching an unacceptable

element.
When the fluid domain becomes more complicated, the situation can
become worse. From a mathematical point of view, the closer the domain
is to a convex domain, the greater the chance that the updated mesh is good.
Hence, it is a good strategy to divide the computational domain into a few
convex-like sub-domains and then apply the Laplace operator to these sub-
domains. A revised element would then be perfectly fine, as shown in the
following figure.
sub-domains
solid
fluid
Figure 12.5 Valid mesh with a proper subdivision of the

domain
Our general approach to reach a better solution of arbitrarily moving

meshes is based on the following steps:

1) Determine the displacements of the boundary nodes from the

physical conditions. These nodes are on fluid-structure
interfaces, free surfaces, phase-change interfaces and fluid-
fluid interfaces. They may also include some nodes
controlled by users (see the Leader-follower option
described in the next section).
2) Solve the Laplace equations on all lines that have been used
to generate fluid elements. The displacements obtained in
step 1 are used as prescribed conditions in this step. If the
displacements of any end nodal points of the lines have not
been prescribed from the physical moving conditions, they
are set to zero. These end point conditions provide Dirichlet
boundary conditions for the equations on lines. Note that
these lines may include those that are actually inside the
3) Solve the Laplace equations on all surfaces that have been
used to generate fluid elements. The displacements obtained
in step 2 are used as Dirichlet boundary conditions for these
equations. Since lines enclose each surface, the surface
equations are well defined.
4) Finally, if the problem is a three-dimensional problem, solve
the Laplace equations on all volumes, with the Dirichlet
boundary conditions obtained in step 3.
The program automatically performs these steps. However, the lines,

surfaces and volumes are implicitly determined during the procedure of
constructing elements. Therefore, the final mesh quality actually depends
on how the computational domain is subdivided while the elements are
generated.
12.10.2 Leader-follower option
In the ALE formulation, large displacements can be encountered. The

leader-follower option is designed to help in controlling the arbitrarily
moving mesh.
A leader and a follower form a pair. A leader is a point that is located
on a physically moving boundary (fluid-structure interface, moving wall,
free surface, etc.). A follower is a point whose movement is determined by

12.10: ALE formulation and leader-follower option
its leader using the method selected by the user. The motion of the moving
boundary determines the displacement of the leader. On the other hand,
any point that is not physically moving can be moved in an arbitrary
manner. Defining a leader-following pair, you can control a point by
forcing it to follow the leader. This procedure is equivalent to having some
nodal displacements prescribed, of course, in the special manner that is
described here.
There are three types of leader-follower options:
(1) Parallel option when a follower is not on the boundary of the

computational domain, its motion is completely determined by its
leader
d f = d l
where the superscripts indicate the leader and follower nodes and
is a factor for the user (the default is one). However, boundary
followers must always stay on the boundary while they follow their
leaders. Their motion is obtained from
d f = dl i(I nn)
where n is the locally normal direction of the boundary.
leader
follower rco
rpr rcl
Figure 12.6 Definition of the leader-follower options

(2) Closest option the boundary follower is located on the wall or

boundary geometry at the locally closest location to the leader
anywhere within the domain. In this option, the follower firstly
moves as in the parallel option and, then, adjusts locally until it
reaches the minimal distance.
(3) Cone option (for FCBI-C elements only) the follower is located
on the wall or boundary geometry at the locally closest location to
the leader within a conical search domain. The leader associated
with its normal to the boundary defines the apex of the conical
domain and the cone angle is a user-specified variable.
In the illustrative figure, rpr , rcl and rco represent follower positions with
the options parallel, closest and cone respectively.
The leader-follower option is a very useful option in controlling the

mesh quality in moving boundary problems. Some general guidelines for
this option are described as follows:
Leader and follower points are assigned to geometric points. The

program will then transfer the condition to their closest nodes.
The leader points must be on physically moving boundaries.
As a general rule, the follower points must not be on physically moving
boundaries. However, as an exception, they can be assigned to fluid-
structure interfaces or moving walls. In this case, care must be
exercised regarding the consistency of the physical and arbitrary
motions.
A follower cannot be assigned to be a leader.
Closest and Cone options are proven useful associated with gap
conditions. Caution: Avoid using these options if there are multiple
closest points. In this case, the program will pick up one randomly as
the followers location point. This procedure may result in oscillation
during iteration or even a diverged solution.
Recall that nodal displacements of lines are determined by the

displacements of its end-points in the automatic procedure; points of leader-
followers are sufficient to control line movements and, accordingly, to
control surface and volume movements.

12.11: Electro-static and steady current conduction analyses
12.11 Electro-static and steady current conduction analyses
This capability allows the user to solve problems where volumetric

heating of materials occur due to the existence of an electric current and
problems where a electrostatic force is important. For incompressible,
slightly compressible and low-speed compressible flows, we add the Joule
heat generation term to the energy equation,

Cv + ... = ... + 1 J iJ
t
and add the body force source term to the momentum equations,
v
+ ... = ... + i(DE) 12 (DiE)
t
where J = J + J 0 is the total current density, J = E is the unknown

current density(or induced eddy current density), D = E is the electric
flux density, E = is the electric field intensity, J 0 is applied source
current density, is the electrical potential, is the electrical
conductivity and is the electric permittivity.
There are two different models available to obtain the electrical

potential, namely electro-static analysis and steady current conduction
analysis. The governing equations for the two models are
i( ) = e
i( ) = iJ 0
where e is electric charge density.

The electric permittivity and the electrical conductivity can be

constant, time-dependent or temperature-dependent.
12.11.1 Discretization and method of solution
The variational form of Laplaces equation is, for example,
( ih ) dV = h J idS

V
The strategy of solution is as follows: Laplaces equation is solved for

the electric potential, , at each time step. The Joule heat generation term,
2
, is then calculated and added to the right-hand side of the energy
equation.
12.11.2 Boundary conditions
12.11.2.1 Prescribed electric potential
In this condition, a time-dependent electric potential can be directly

prescribed at the boundaries,
= (t )
12.11.2.2 Distributed current density and electric flux density loads
When steady current conduction model is adopted, the current density

condition can be active. With this condition, user specifies the normal
component of the unknown current density, J n (t ) = ni J .
When an electrostatic model is adopted, the electric flux density

condition can be active. With this condition, the user specifies the normal
component of the electric flux density, J n (t ) = ni D .

12.11: Electro-static and steady current conduction analyses
12.11.2.3 electric charge density and source current density
The field electric charge density or source current density can be

applied through user-supplied subroutine ESUSRS. The model type
(electrostatic or steady current), problem heading, element group ID,
element ID, current time, spatial coordinate are passed into the subroutine.
User writes its own program to calculate the source term. When the model
is electrostatic analysis, the electric charge density (scalar) is expected and,
when the model is steady current analysis, the source current density
(vector) is expected.

12.12 Phase change between liquid and vapor
ADINA-F can be used to analyze problems with phase change between

liquid and vapor. There are two different options available for phase
change. For the first option (general phase change model), the phase
change interfaces are determined by the temperature distribution in the
fluid. For the second option (cavitation model), the interfaces are
determined by the pressure distribution in the fluid. This feature is
available for incompressible flows, slightly compressible flows and low-
speed compressible flows. Different flow types can be chosen for different
phases. All elements that are currently used in ADINA-F can be used for
this feature.
Once the phase-change option is used, the density is always treated as a

variable even in incompressible flow models. The surface tension effect at
the phase-change interface is not available in current models.
We use the so-called vapor fraction function f that is defined as the

rate of the vapor volume and total volume. Any mixed property of a
material parameter m will be calculated as
m = ml + f ( m ml )
where the subscripts l and v represent, respectively, the liquid phase and
vapor phase.
12.12.1 General phase change model
In this model, the energy equation is modified to include the effect of

latent heat (L):
fL
Cv + = vi( fL)
t t

12.15: Steered adaptive meshing
First lets consider the phase change between liquid and vapor. Let lv
and vl be evaporation temperature and condensation temperature
respectively ( lv < vl ) and assume that the latent heat is absorbed when
liquid changes to vapor. The vapor fraction f thus equals zero under
evaporation temperature and equals one above condensation temperature.
A linear function is used to approximate the function between the two
temperature values. In order to obtain continuous derivatives, we also
smooth the linear function near the evaporation temperature and
condensation temperature within a region of length ( vl lv ) .
To be more specific, we first define a normalized function as
0 x0
x2 d 0< x

f ( x, ) = ( 2 x ) d < x 1
2
( d = 2 (1 ) )

1 ( x 1) d 1 < x 1
2
1 x >1

Currently, we choose = 0.1 . In the general phase-change model, we use

( )
f = f , , where = ( lv ) ( vl lv ) .
For phase-change between solid and liquid, the energy equation above
is still valid, but the above evaporation temperature lv and the
condensation temperature vl are replaced with the melting temperature
sl and the solidification temperature ls , respectively. To mimic solid
regions, the velocities and/or turbulent variables should be suppressed in
solid regions. This suppression is achieved by additional source terms
added into the momentum equations and the turbulence equations.
For melting and solidification applications, the additional term

A ( v v p ) is added onto the right hand side of the momentum equation,

and A is added onto the right hand side of the turbulence equations. In
these terms, v p is the solid region pulling velocity, and represents
k , , . The coefficient A is defined as
(1 f )
2
A = Cmushy
(f 3
+ mushy )
where f is the liquid fraction, Cmushy and mushy are user-defined

parameters. Note: Cmushy 0 and mushy 0 . If Cmushy or mushy is less
than zero, the computations may be unstable. Generally the value of
Cmushy is about 104 107 times larger than any other terms in the
momentum and turbulence equations to suppress velocities and turbulence
variables in solid regions. mushy is a small constant to avoid numerical
instability as f is very small. Larger Cmushy and/or smaller mushy is,
harder the convergence is.
12.12.2 Cavitation model
In liquid flows containing micro-size gas bubbles, cavitation generally

occurs when the pressure at some locations is below the vapor pressure of
the liquid. Cavitation appears in a wide variety of engineering applications,
such as nozzle, pumps, hydrofoils, propellers, and so on. In almost all
engineering applications, cavitation is not a desirable phenomenon. It can
cause structure damage, generate undesirable vibration and noise, and
lower system performance efficiency.
The default cavitation model in ADINA-CFD is a simplified phase

change model. The effect of latent heat is omitted. The phase change is
assumed primarily caused by the pressure change. Let plv and pvl be the
evaporation pressure and the condensation pressure respectively
( plv > pvl ), the vapor fraction function is evaluated by f = f ( p, ) ,
where p = ( plv p ) ( plv pvl ) .

A more sophisticated cavitation model associated with the VOF method

is also available (currently only for FCBI-C element). This model cannot
be used for compressible flows and free surface flows. Non-condensable
gas is not considered in this model. Also note that the problem must be in
transient.
To setup this model, one species of VOF must be activated. In the VOF
method, the vapor volume fraction f v and liquid volume fraction fl are
defined as
Vv
fv = , fl = 1 f v
Vv + Vl
where Vv and Vl are vapor volume and liquid volume respectively.
The mixture is a hypothetical fluid with variable density:
m = f v v + f l l
The mixture satisfies the continuity equation:
m
+ i( m v ) = 0
t
The transport of equation of the liquid volume fraction is
fl 1
+ i( fl v ) = m
t l
The mass transfer between the vapor and liquid is modeled through the
mass creation rate m .
In practical applications, the cavitation number is one of the most

important dimensionless numbers used to describe cavitating flow
structures. Its definition is

p pv
cav =
2 l v
1 2
where p and v are the far-field pressure and velocity respectively, and
pv is the vapor pressure of the liquid.
ADINA-CFD provides two models to simulate m : Kunz model and ZGB

model.
12.12.2.1 Kunz model
Kunz et al. modeled the vapor-liquid mass transfer rate as below:
Cdest v fl max ( pv p, 0 ) C prod v f l 2 f v

m=
2 l v t
2
1
t
where t ( = L v ) and L are the characteristic system scales of time and

length respectively, and the Cdest and C prod are two user-defined
parameters.
12.12.2.2 Zwart, Gerber and Belamri (ZGB) model
ZGB model is given as below:
3 f nuc fl 2 pv p
Cdest v if p pv
R 3 l
m=
C 3 f v 2 p pv if p > pv
prod v R 3
l
where f nuc is the nuclear site volume fraction, and R is the bubble radius.
The suggested values for Vnuc and R are 0.0005 and 10-6, respectively.

The coefficients Cdest and C prod are flow-condition dependent parameters.

The suggested values of Cdest and C prod are 50 and 0.01 respectively.
They are suitable for a variety of flows.
12.12.2.3 Turbulence viscosity modification
For cavitating flows, especially for the transitional flows from the sheet
cavitation to the cloud cavitation, the compressibility of the mixture near
the cavitation interfaces may need to be taken account in the turbulence
model. For such cases, the mixed density viscosity m is further modified
as
n
m
m = v v ( v l )
v l
where the exponent constant n should be n 1. For larger cavitation

number applications, n can be set as 1. Its value increases as the cavitation
number decreases and Reynolds number increases.
12.12.3 Phase-change model setup
To use this feature, you first declare that the element group is one with
phase change. You then define and apply two sets of material data and
additional control data. The flow types of the two phases can be different
inside one element group. The phase-change temperatures (or the phase-
change pressures), the latent heat and the type of phase (liquid or vapor) are
defined in the control data.
We note that when the difference between the two phase-change

temperatures (or the phase-change pressures) is very small, the convergence
may be difficult. It is therefore important to specify proper control
parameters.

12.13 VOF method
The VOF method is a surface-capturing method for solving flows

possessing multiple immiscible fluids. In the VOF method, the fluids share
momentum equations, and each of the fluids is convected through the
domain by solving scalar transport equations. The surface tension and wall
adhesion are simulated with a continuum surface force (CSF) model.
There are maximum 2 additional species that can be computed using the
VOF method. This method is available for all types of elements in
incompressible and slightly compressible flows. Currently it cannot be
coupled with the phase-change feature.
12.13.1 Transport equations
The scalar transport equation used can be written as
i
+ vii = 0 i = 1,.., n
t
where v is the velocity vector, i is mass ratio of the i-th species, and n is
total number of additional species (beside the primary fluid).
A finite volume method is used in discretizing the equation. A donor-
acceptor scheme is used to evaluate the value of the mass ratio at the face
of the control volume. In terms of accuracy, time step length must be
sufficiently small (usually Courant number vt x 0.1 ).
12.13.2 Material
Any material parameter m in the mixed fluid domain is evaluated as an

average among all the species,
n
m = i mi
i =0

n
We note that the mass-ratio for the primary fluid is 0 = 1 .
i =1 i

12.13.3 Surface tension
The surface tension and the wall adhesion are simulated with a
continuum surface force (CSF) model that models the surface force as a
body force. The body force per unit volume is then evaluated using
n 1 n
fs = (
i = 0 j = i +1
ij i j j + j ii )
where ij is surface tension coefficient between species i and j , and

i = i(i i ) .
12.13.4 VOF-wall angle condition
User can specify the angle conditions ij ( i < j ) at the boundary, where
ij denotes the angle between the boundary outward normal and the normal
of the interface (pointing from species j to the species i ), as defined in the
following figure.
Species j
interface between
Species i species i and species j
ij Boundary
nij
nwall
Figure 12.7 Definition of VOF-wall angle

12.13.5 Liquid-solid phase change for primary species
Phase change between liquid and solid can be considered for the
primary species. Let s and l be the solidus temperature and liquidus
temperature of the primary species ( s < l ) and assume that the latent heat
is absorbed when solid changes to liquid. The volume fraction of liquid
phase inside total volume of primary species f 0 thus equals zero under
solidus temperature and equals one above liquidus temperature. A linear
function is used to approximate the function between the two temperature
values.
The latent heat effect is modeled as a source term adding to the right-
hand side of the energy equation.
( f L)
+ v ( f 0 L)
0
Cv + ... = ... 0 f 0
t t

where 0 , f 0 and L are the density, volume fraction and latent heat of
primary species. Simultaneously, the additional drag force related terms
are added to the momentum equations and turbulence kinetic energy
equation respectively
v
+ ... = ... dCmush ( v v s )
t
K
+ ... = ... dCmush K
t
where,
f 02
d=
( )
3
1 f0 +

Cmush is a mushy zone constant (of unit density/time), v s is a prescribed

solid velocity, is a small constant introduced to avoid division by zero,
f 0 = (1 f 0 ) f 0 is the volume fraction of solid phase inside total volume
(including all species). Notice that when the solid phase of primary species
occupies the total volume, i.e. f 0 = 1 , the right-hand side source terms of
the momentum and turbulence kinetic energy equation dominate the
equations. In other words, the equations change to
v v s = 0, K = 0
12.13.6 Loading of the species
The mass-ratios i , ( i = 1,.., n ) can be prescribed at any lines, surfaces,

volumes and nodes.
12.14 Rigid motion of element group
In certain applications, it is useful to specify mesh movement. This

feature is particularly useful when associated with sliding-mesh boundary
conditions. There are two types of rigid motions, translation and rotation,
are available now. Under these conditions, all nodal positions in the
specified element groups are determined.
The translation is defined as
x ( t ) = x0 + x ( t )
where x 0 is initial nodal position and x ( t ) is the specified time-

dependent displacement.

In the condition of rigid rotation, the center of rotation x0 , rotation axis

e3 ( t ) and angular velocity ( t ) are specified. The nodal position is then
determined by
x ( t ) = x0 + J T R ( ) J ( x 0 x 0 )
where, J is the Jacobian that transforms global Cartesian system into a

( )
local system e1 ( t ) , e 2 ( t ) , e3 ( t ) , and R ( ) is the rotation matrix on
( )
t
plane e1 ( t ) , e 2 ( t ) . The rotated angle is computed from = ( t ) dt .
0
12.15 Steered adaptive meshing
12.15.1 Introduction
The steered adaptive meshing (SAM) capability is applicable for CFD and
FSI analyses. The capability is particularly useful for steady-state analysis.
In a SAM procedure, the analysis proceeds in the usual time stepping

solution and invokes the adaptive meshing step in the following cases.
Iteration convergence is not reached in a time step. This is frequently

due to a poor mesh, which is therefore repaired.
Mesh crashes during the solution iteration (typically in FSI analysis).
This adaptation is also called mesh repair.
Once a solution over the complete time span has been reached,
additional adaptive steps are prescribed to improve the solution
accuracy. This is referred to as mesh adjustment.
12.15.1.1 Adaptive mesh repair
Suppose that a CFD or FSI analysis is required for a number of time steps
from time t = t0 to time t = tend , where t0 < tend , and suppose that the usual

step-by-step procedure is performed, but the program does not converge at

a time step t = tc . Then with SAM and the prescribed remeshing criterion, a
new mesh is automatically generated from a mesh at a previous time step tp
(t0 < tp < tc). Thereafter the solution is automatically continued with the new
mesh. If necessary, this procedure of mesh repair is repeated at further time
steps until the computation finishes at the expected time t = tend .
12.15.1.2 Adaptive mesh adjustment
Suppose that a CFD or FSI analysis has been performed and the
computation has been completed for the total number of time steps.
However, the analyst judges that the result needs to be improved with a
better mesh quality. SAM can be set up for such mesh adaptation to achieve
solution accuracy. Let
NSTP = total number of time steps in original model

NADP = maximum number of adaptive steps (steps with
remeshing) that can be performed
(NSOL-1) = number of additional solution steps after the solution with
NSTP steps has been performed. NSOL is typically 1
(default), so that no further solution steps are performed
after the NSTP number of steps. But NSOL can be greater
than 1, if the analyst requires higher accuracy in the
solution. Then, in each such step a new mesh will be
spanned to reach that accuracy. Note that if during the
NSTP solutions, there have been M adaptive steps, then
(NADP-M) must not be less than (NSOL-1).
Here we give an example.

Load
3 mesh repairs performed to obtain solution at time 10
solution obtained at time step NSTP
mesh adjustments performed to obtain more accuracy
NSTP = 10
NADP = 10
NSOL = 4
Time
0 10 13
In this example, the number of initial total time steps NSTP=10. At steps 3,
6, and 8, remeshing has been performed to reach a solution at time 10;
hence M=3.
Here, NADP-M=7 and hence 3 = (NSOL-1) additional solution steps

corresponding to times 11, 12, 13 are performed, each with a remeshing in
an attempt to obtain the requested accuracy.
Note that the following steps are required in the ADINA User Interface
(AUI) for additional adaptive steps to achieve better accuracy.
In MeshingSteered Adaptive MeshControl, user must set

"Adaptive Timestep Sequence" to "Appends Original Timestep"
User must also define a time step with name ADAPTIVE with a
single step of size 1.0
The user can refer to Primer Problem 48 for more details on the use of
additional adaptive steps for better accuracy in the solution.

12.15.2 General procedure
A CFD/FSI computation with SAM may have one or more mesh repair or
mesh adaptation steps, all of which we refer to here as restarts. Each restart
is referred to as a new model because it has a new mesh, and also its
material properties, loads, initial and boundary conditions could be
changed at restart times (if the manual mode is selected). The model at
initial time t = t0 is referred to as the first model, and the model at the first
restart time as the second one, and so on.
The General procedure of both Manual mode and Automatic mode includes
two steps: Model setup and Running the CFD/FSI model with SAM.
12.15.2.1 Model Setup
Structure Model Setup for FSI problems
As mentioned before, SAM does not affect the structural models in

FSI problems. Therefore, there are no special restrictions on structural
models regarding geometry, element type, etc. for FSI problems with
SAM.
Fluid Model Setup for CFD/FSI problems
(1) Geometry and Mesh Preparations in Fluid Models
For CFD models and FSI fluid models, geometries have to be

defined as bodies via ADINA-M in AUI, or imported from a Nastran
file. The element type has to be 3-node triangular elements for 2D
problems or 4-node tetrahedral elements for 3D problems because
SAM only generates free-formed meshes with these element types.
These basic requirements of SAM are listed in the following table for
clarity:

Table 12. 1 Geometry generation tools and mesh element type

requirements for SAM
Geometry generation tools ADINA-M or Nastran

3-node triangular elements (2D)
Mesh element type
4-node tetrahedral elements (3D)
Note: internal and external boundaries in the first model will not
be changed when new meshes are generated via SAM. That is, even
though the nodal points on geometry boundaries can be re-
distributed, the boundaries cannot intercept with each other. Any
internal and/or external boundary interception will make the SAM
fail.
(2) Other Settings in Fluid Models
Other fluid model settings of CFD/FSI with SAM, such as initial

conditions, boundary conditions, materials and solution process
parameters, are the exactly same as those of general CFD problems.
(3) SAM Setup
In AUI, SAM can be accessed in

Meshing Steered Adaptive Meshing . There are three dialog
boxes for SAM: control, criterion and meshing. In SAM control
dialog box, Manual mode or Automatic mode can be selected; restart
time, at which SAM will generate a new mesh, can be specified;
CFD restart file (*.res) and adaptive command file (*.adp) can be
chosen; interpolation schemes of boundary geometry segments and
mapping schemes of boundary conditions and initial conditions can
be specified.
In SAM criterion dialog box, six criteria are available: Element

Quality, Element Size, Variable Gradient, Sphere Region, Brick
Region and Direct List. Each criterion can be used individually, or
they can be used in a combined way. See detail description for these
criteria in the sections of criteria for selection of preferable element
size and combination of criteria.

If Manual Mode is used, users will manually give all SAM

control information and criteria. Users may also change initial
conditions, boundary conditions, and solution process parameters for
restart models. In other words, users will manually generate the
second model, the third model and so on by themselves.
If Automatic Mode is used, users will prepare the first model,

specify criteria for re-meshing, and use AUI to generate an adaptive
file (*.adp). The adaptive file includes control information used to
create a new fluid model with a new mesh. For FSI problems, users
will generate two adaptive files via AUI: one is for the fluid model
and the other for the structural model.
More details of Manual Mode and Automatic Mode are given in

the next section.
12.15.2.2. Running CFD/FSI model with SAM
Manual Mode
Step 1. For CFD problems, in the first fluid model, Manual mode is
selected. Run the first CFD model as usual. Since the SAM
is turned on in the first model, ADINA will save the adaptive
file automatically. For FSI problems, besides in fluid model,
SAM is turned on and Manual Mode is also selected in the
first structural model. Run the first FSI model as usual.
AUI will generate two adaptive files; one is for the fluid
model and the other for the structural model. The structural
model solutions will be saved at the same times at which the
fluid model solutions are saved;
Step 2. Prepare restart model with adaptive mesh. At first, in SAM

control dialog boxes, restart time, fluid model restart file,
fluid model adaptive file can be specified, and then criteria
for re-meshing in SAM criterion dialog box. At the end, a
new mesh is generated via SAM meshing dialog box;
Step 3. Specify time step for the restart run;

Step 4. Save restart data file (*.dat);
Step 5. For FSI problems, use AUI to open the previous structural
model at restart time. In SAM dialog box, select Manual
Mode, specify restart time and specify restart file of the fluid
model. For CFD problems, this step is skipped;
Step 6. Generate new structural model data file. For CFD problems,
this step is skipped;
Step 7. Run CFD or FSI. For FSI problems, AUI will ask users to
copy the structure restart file at restart time.
Steps (2-7) can be repeated until final solution is obtained.
Automatic Mode
Step 1. Generate solid model (if it is an FSI problem) as usual in

AUI-ADINA-STRUCTURE;
Step 2. Still in AUI-ADINA-STRUCTURE, open SAM dialog box,

select SAM Automatic mode; give restart time (if the latest
mesh is used, give -1); give corresponding CFD restart file if
it is not a new run, otherwise leave it as blank;
Step 3. Save solid model;
For CFD problems, step 1-3 can be skipped.
Step 4. In AUI, select ADINA-CFD (with ADINA-Structure if it is a

FSI problem);
Step 5. Generate fluid model as usual (see other CFD cases in

primer). Notice that geometries can only be generated by
using ADINA-M, and the mesh must be free mesh with
triangles (2D) or tetrahedral elements (3D);

Step 6. Still in AUI-ADINA-CFD, open SAM control dialog box

( meshing SAM control ), select SAM Automatic
model; give restart time (if the latest mesh is used, specify
1 ); define SAM criteria in criterion dialog box;
Step 7. In Control time step , define adaptive time step;
Step 8. Save fluid model;
Step 9. Run automatic adaptive meshing for CFD (with/without

FSI): Run ai8.8 adp.
12.15.3 Criteria for selection of preferable element sizes
The novelty in our adaptation scheme is how we determine preferable

element sizes, hep , using various criteria, denoted as C ( Fe ) , where it may
depend on a flow-solution-variable Fe . Note that the criterion can be
calculated at a time tc (called criterion time) that could be earlier than or
the same as the restart time ts . The criteria can be applied to the whole
fluid model or only to certain element groups.
The simplest criterion is to directly specify element size (a length scale) in

a defined region (e.g., a sphere or a brick)
C ( Fe ) : hep = hep e
Another type of criteria is based on the mesh deformations r ( Fe ) : relative

element size variation or relative element quality variation
C ( Fe ) : hep = hep r ( Fe ) [ rmin , rmax ]
where rmin and rmax are, respectively, the acceptable minimum and
maximum variations.

A more sophisticated type of criteria is based on the well-known general

fact that, whatever field is to be predicted accurately, the element size times
the gradient of that field should be about constant over the fluid region
C ( Fe ) : hep Fe = c
where c is a constant, and Fe here represents pressure gradient, vorticity,

etc. The constant is determined by
c = r 1
Ne h Fe
e e
where r is a ratio factor, he is the local element size at criterion time, and
N e is the total number of elements in that mesh.
In practice, elements of too large or too small sizes must be prevented. The
calculated preferable element sizes are then modified by
c
hep =

c c
max min Fe , ,
hmin hmax

where hmin and hmax are, respectively, the minimum and maximum
element sizes allowed in the mesh.
The choice of Fe depends on the type of problem. For example, the

pressure-gradient criterion C (pe ) may be used for having a sufficient
number of elements in areas where large pressure gradients are present.
The vorticity criterion C ( v ) may be used to improve the solution
accuracy in boundary layers. In fact, the gradient of any other anticipated
solution variable could be used.
Based on a criterion, the program produces a to-be-repaired-element set

E of elements e associated with preferable element sizes hep

C ( Fe ) E {e | hep }
12.15.4 Combination of criteria
Usually, one criterion is not sufficient to produce effective meshes.

Therefore we allow certain operations on criteria to result in a combined set
of elements to be repaired. The procedure is performed as follows
E =
for i = 1, 2,... {E =: operation of E and Ei }
These operations are summarized in the following table
Operation Definition Note

Append E ( Ei \ E ) Append Ei to E
Replace Ei ( E \ Ei ) Append E to Ei
Subtract E \ Ei Subtract Ei from E
Smaller ( E \ Ei ) Join E and Ei hep = : min {hep , hepi }
Larger ( Ei \ E ) and update hep hep = : max {hep , hepi }
Average ( E Ei ) if e E Ei hep = : 12 (hep + hepi )
12.15.5 Smoothing technique
The preferable element sizes may be smoothed a few times using the
iteration procedure
h k +1
=
hh k
nb e ep
; k = 1, 2,...
h
ep
nb e

where as indicated by nb the sum is carried out over the neighboring

elements.
12.15.6 Control on the maximum number of elements
The above element criteria basically re-scale element sizes to obtain a mesh
in which the selected gradients times the element sizes are about constant
and the geometry and boundary conditions are properly represented. In
practice, the model size must be limited, that is, there must be an upper
limit on the maximum number of elements allowed. When the total
number of preferable elements N ep is larger than the maximum number of
elements allowed N max , we can increase the geometric size of all elements
by a factor , that is, we change hep to hep , with
N 1

ep
= max

,1

N max

where is the space dimension and
Ve
N ep = e
Vep
with Ve the current element volume and Vep the volume of the elements
computed using the preferable element size.
Of course, depending on the meshing algorithm used, the final number of

elements obtained after this adaptation could still be slightly larger or
smaller than N max . The purpose of this scheme is to apply a limit on the
size of the model.
12.15.7 Comments and limitations
In principle, the adaptive-mesh option can be used for solving transient

problems. In practice, however, the error accumulation due to the changes

in meshes over many time steps could be too large. The difficulty is that
any error introduced at a particular time may not be iterated away at a
later time. Therefore, in transient analyses, the solution should be carefully
checked once the mesh is repaired. Some times, use of the second-order
mapping scheme may also help on the accuracy. Note that the second-
order mapping scheme does not affect the solution accuracy in steady-state
analysis.
For steady-state solutions with moving meshes, the final solution may not
be able to reached in one or two steps. The key is, starting with a coarse
mesh, to refine the mesh gradually in steps. Similarly, the mesh should be
repaired earlier before it becomes too distorted. Consider FSI as an
example. A major difficulty arises when the required fluid element sizes
are very small, as in boundary layers, and the structural deformations are
rather large in an incremental step, measured on these fluid element sizes.
In such cases, the fluid mesh can become very distorted or, more severely,
element overlapped before converged solutions are obtained. A good
strategy is, to start the solution with a rather coarse mesh and allow the
large structural deformations to take place. Then, thereafter, the fineness of
the fluid mesh is increased. Since the structure will in this later phase only
adjust its deformations by relatively small changes in geometry, required
small fluid sizes, specifically in boundary layers, can now be
accommodated in the mesh adaptation to obtain the required solution
accuracy.
In the FSI solutions, we also use the arbitrary-Lagrangian-Eulerian

formulation for the fluid as usual, to allow reasonable mesh deformations in
each solution step.
If geometries of fluid models are imported into AUI from a Nastran file,
there is one restriction for the Nastran file. As known, Nastran files only
contain mesh information, but no geometry information. Therefore, for
such a model, SAM will do re-meshing based on the information of the
element groups. This requires every boundary should be grouped as a
single shell element group. For example, a model imported into AUI from
a Nastran file has 4 external boundaries and 2 internal boundaries. Users
should group each boundary as a different shell element group. Therefore
totally 6 shell element groups should be defined in the Nastran file.

12.15.8 Adaptive Meshing Modeling Tips
What follows is a brief discussion about meshing as it relates to mesh

adaptation. Focus is given to the 3-d case where interfaces are body faces,
but the same principles apply to the 2-d case where interfaces are face
edges.
Properly dealing with nodal coincidence is of prime importance when

modeling bodies with interfaces. If the flow is supposed to pass through,
nodes should not be duplicated across the interface. However, if there is no
intended flow, nodes should be duplicated across the interface. This is
routinely handled by the nodal coincidence parameter in the body meshing
commands.
The best way to handle situations where you have some interfaces (body
faces) with flow passing through and others with no flow is to use the
SELECTED nodal coincidence option. When meshing, this option
checks if a new node already exists among the nodes on the (previously
meshed) entities (typically, body faces) defined in the provided
DOMAIN. So, if you want flow passing only through some of the
interfaces between two bodies, you should add those interfaces to the
domain (when meshing).
Another important consideration is to make sure that nodal coincidence

checking is not order-dependent, in other words, the order in which the
bodies are meshed should be irrelevant. This means in particular that the
COINCIDE=NO directive should not be used. In the case mentioned in
the above paragraph, all the meshing commands involving such interfaces
(some with flow and some without) should make use of the same type of
nodal coincidence checking so that the meshing is order independent.
Thin solid structures should be avoided when modeling fluid-structure

interactions, whenever possible. It is always preferable to reduce the
dimension of the structural part. A thin structure should be modeled with
shell elements in three dimensions and with beam elements in two
dimensions. This is to avoid self-intersections of the fluid boundaries when
the mesh moves around the thin structures.


Chapter 13: Other topics
Chapter 13 Other topics
13.1 Model preparation and testing
Before any model is established and analysis performed, the following

issues should first be resolved:
What is the suitable fluid model (incompressible, slightly compressible,

low speed compressible, high speed compressible or porous medium)?
Should a steady-state analysis or a transient analysis be performed?
What time/load step size should be used?
Which part of the geometric domain should be modeled?
What boundary conditions are most adequate?
What element and mesh should be used?
Which material model describes the fluid behavior most adequately?
What is the procedure to use to reach a successful solution of a
complicated problem?
To answer the above questions requires an understanding of the

physical problem and familiarity with ADINA-F, including the available
capabilities, assumptions and limitations.
13.1.1 Choosing the fluid model
Most engineering applications of fluid flows are incompressible flows.

If a domain with open boundaries or an enclosure with rigid boundaries is
considered, the incompressible flow model is generally applicable.
Examples of fluids are oil, water, air (when the Mach number is less than
about 0.3), etc.

13.1:Model preparation and testing
rigid or deformable rigid
oil, water,
air (M<0.3), etc. oil, water,
open open air (M<0.3), etc.
Figure 13.1 Typical problems that incompressible flow model

can be applied
If the boundary of a domain is fully closed, the flow is called confined

or flows in an enclosure. If the boundary of confined flow is deformable,
as in a fluid-structure interaction problem or in problems where boundary
displacements are prescribed, the fluid cannot be treated as fully
incompressible.
Oil and water generally show negligible dependency of density on
temperature, so it is adequate to model their flows using the slightly
compressible flow option. When the boundary moves very slowly, the air
can be treated as a slightly compressible flow as well.
deformable
oil, water,
air (M<<1), etc.
Figure 13.2 Typical problems that slightly compressible flow

model can be applied
In many applications, the compressibility of air must be taken into

account. The problem must distinguish between low-speed and high-speed
fluid flows purely because of numerical limitations. Typical examples of

low-speed compressible flows are those pertaining to enclosures whose

boundaries are deformable.
fixed/flexable/moving
boundaries
air (M<0.3)
open or closed
open or closed
Figure 13.3 Typical problems that low-speed compressible

flow model can be applied
Whenever the Mach number is of order 1, the flow must be treated as a

high-speed compressible flow. Most flows of this type are open to the
environment.
closed
air (M ~ 1)
open open
Figure 13.4 Typical problems that high-speed compressible

flow model can be applied
The porous medium flow model has wide applications. In addition to

flow through porous solids such as soils, it can also be used to model many
other problems such as those involving air filters with small or large holes
(such as cigarettes). It is worth noting that the model can be incorporated

with incompressible, slightly compressible and low-speed compressible

flow models through element group options.
closed
mixture of
fluid and solid
open or closed
open or closed
Figure 13.5 Typical problems that porous media flow model

can be applied
When many small holes in a model need to be modeled simultaneously,

a porous medium model is the best choice. As shown in the following
figure, we can use a porous medium element group to simulate the pipes
instead of modeling the small pipes directly (which may not give good
results depending on the capacity of the computer used).
air flow fluid element

group air flow
environmental
pressure porous medium normal-traction
element group
Figure 13.6 Example of using porous media flow model

13.1.2 Choosing the analysis type
If a steady state physically exists, either a steady-state analysis or a

transient analysis can be performed to reach the steady-state solution. To
model such a problem, all materials, boundary conditions, etc., must be
time-independent at the last time step at which the steady solution is
expected. Recall that the definition of a steady-state analysis in ADINA-F
is to drop the dynamic terms t in the governing equations. We have the
following ways to reach a steady-state solution.
Perform one time step using a steady-state analysis and all materials
and boundary conditions are time-independent. This method can be
applied to some simple flow problems. The computational effort can
be demonstrated in the following chart:
(1b) for equilibrium iteration 1, 2, ...

solve the linearized steady - state equation
Use a proper CFL number associated with the method (1). Note that
the CFL option is, in terms of numerical stability, equivalent to the time
marching method in a transient analysis that reaches a steady-state
solution. The advantage of using the CFL option is to let the program
decide the proper auxiliary time step and the converged solution is
the steady-state solution. This method is recommended in almost all
fluid flow problems. In particular, method (1) can be obtained if the
CFL number is very large. The computational effort can be
demonstrated in the following chart:

solve the linearized unsteady - like equation
The advantage of iteration (2b) is that the matrix has a better

conditioning. With a properly chosen CFL number, the approach can
give a faster convergence as well. However, the CFL number cannot be
too small. The value 1 should be regarded as the lower limit for implicit
methods.

Perform a few time steps using the steady-state analysis option and
time-dependent materials or boundary conditions. The time step here
does not correspond to the real time; hence they are rather called load
steps. The solutions obtained at each time step are in steady state
corresponding to the conditions at that time. The solutions obtained
from the previous time step serve as initial conditions for the next time
step. This method can be applied to some flow problems with flow
transition stages. Typical examples are natural convection, flow in
curved channels, or any flows where fluid circulation zones may occur.
The solutions obtained at each time step are sometimes required to
represent solution sequences under different load conditions
(represented by specific Reynolds numbers, Rayleigh numbers, etc.).
The computational effort can be demonstrated in the following chart:
(3a) for time step 1, 2, ..., N t

solve the linearized steady - state equation

Usually, iteration (3b) has a better initial condition than (1b). Although
more time steps are performed, the overall CPU time may be less.
Above all, (3b) may converge in problems for which (1b) diverges.
Use a proper CFL number associated with the method (3). The
computational effort can be demonstrated in the following chart:
(4a ) for time step 1, 2, ..., N t

solve the linearized unsteady - like equation

Iteration (4b) has a better matrix conditioning than (3b) if a proper CFL
number is used. However, as explained for (2b), when the CFL number
is too small, convergence may become slow. Nevertheless, this is the
most frequently used method for steady-state analyses.
Use the time marching method, in which a transient analysis is

performed, to reach the steady-state solution. The steady-state solution
may be obtained during the time step marching or, in addition, a restart

run must be performed. Since only the final steady solution is of

interest, the equilibrium iterations can be omitted. The computational
effort can be demonstrated in the following chart:
(5a) for time step 1, 2, ..., N t

solve the linearized unsteady equation
The equilibrium iteration can be omitted by choosing either a large

iteration convergence tolerance or 1 as the maximum iteration number.
A steady-state tolerance can be specified as well to ensure that a
steady-state solution is reached. However, in the event that the steady-
state solution has not been reached, the final solution is always saved
and stored in the porthole file and in the restart file, so that the solution
can be visualized (for example, to check if indeed the steady-state
solution has been obtained) and, if necessary, a restart-run can be
performed.
When problems become unsteady in nature, transient analyses must be

performed. The initial conditions in this case must be actual physical
conditions. Physically meaningful time steps must also be chosen (see
Section 13.1.3). When time steps are large, convergence in equilibrium
iterations must be enforced for an accurate transient solution. In this case,
the computational effort can be demonstrated in the following chart:
(6a ) for time step 1, 2, ..., N t


solve the linearized unsteady equation
Remember that the truncation-error in time is always the first order (for
> 1 2 ). Hence, the accuracy in time frame is proportional to the time
step size. It is better to always iterate for equilibrium in implicit transient
analyses unless the CFL number is small (for example, in the order of 1). If
the time step size is very small, the user may dispense with the equilibrium
iterations but then the results must be carefully checked. The procedure
can be achieved by choosing either a large equilibrium iteration tolerance
or 1 as the maximum iteration number. In this case, the computational
effort is the same as given in the method (5).

13.1.3 Choosing the time/load step size
In steady-state analyses, the time step size has no physical meaning.

The time functions determine the conditions at each time steps.
However, in transient analyses, time steps are physical parameters.
Therefore, the time step sizes must be reasonable to visualize the solution
as well as to be able to accomplish the solution. Implicit methods have no
limitation of time step sizes in order to be numerically stable but
convergence in the equilibrium iterations must be reached. The choice of
the time step can solely be based on the physical conditions. The time steps
should be determined so that the solutions can represent the physical
phenomena of interest.
Many transient problems are fundamentally periodic in nature. This
property can be predicted based on the following: experience with the
problem, experimental data that is available, periodic boundary conditions
that are applied, a previously computed solution to a similar problem,
analytical results from a simplified problem, etc. A reasonable choice of
the smallest time step size for such problems is about 1 / 20th to 1/100th of
the smallest period of interest, depending on the accuracy that is sought and
the CPU time available.
solution behavior
in nature
20-100 time steps
Figure 13.7 Time step sizes in periodic solutions

Some transient problems are fundamentally determined by the waves

traveling in the domain of interest. For slightly compressible flows, the
wave speed can be predicted by w = and for compressible flows, the
wave speed can be predicted by the sound speed w = c = p . A
certain number of time steps is required in order to capture the wave
motion. Typically, 10 time steps are the minimum needed to allow the
wave to travel a distance L that characterizes the computational domain.
Therefore, a time step size of L (10 w) should be regarded as the largest
that is allowed. On the other hand, too many time steps are unnecessary as
well. A CFL number 1 should be regarded as the minimum time step size
that can be used. In other words, the time step should not be smaller than
x w .
L=10wDt x
Figure 13.8 Time step sizes in wave propagation problems
There are many problems that are fundamentally diffusive in nature.

Small Reynolds numbers and small Peclet numbers characterize these
problems. The nature of these problems is their similarity described by the
dimensionless variable x 2 Kt , where K is the diffusion coefficient
that represents in the momentum equations, k Cv in the heat
transfer equation, etc. In a one dimensional diffusion problem, for
example, the solution can be regarded as steady when > max 3 . We can
perform N t time steps of equal size (i.e., = max N t ) to cover the
solution resolution. The maximum time is therefore evaluated as
tmax = L2 2 K 2 and, accordingly, the time steps can be determined as
tn = tmax ( N t n + 1) 2 . We can see that the time step sizes ( tn = tn +1 tn )
vary quadratically from the smallest one, where the solution changes the
most to the largest one, where the solution is considered in steady state.

solution
2
ti /2K 2 (same )
1 2 3 x
Figure 13.9 Time steps in diffusion problems
When problems become more complicated in nature, neither of the

above factors is dominant and a natural choice is the smallest time step of
the possibilities if the whole history of the solution is required.
It is worthy to mention here again, whenever the time step is small (say
the CFL number is about order 1), the equilibrium iteration may be omitted.
Particularly when a large number of time steps are performed, this
approximation can save tremendous CPU time (typically 50% to 90%).
13.1.4 Choosing the computational domain
Obviously, we cannot include any details or sizes in a model. The

model must only contain the relevant data and phenomena of physics that
are suitable for the capacity of the computer. Some guidelines are listed
below.
The domain must be cut somewhere from the rest of the world.
Reasonable boundary conditions should be available to represent the
flow conditions on the boundary. For example, when a circulation zone
is expected behind a building, a long region should be used there if a
zero normal-traction is applied. If the circulation zone cannot be
included completely, frequently a transient analysis needs to be
performed.
When objects are small enough to be negligible compared to the major
parts of the model, they should be omitted or modeled in different
ways.

air flow
circulation is expected
behind the building
small objects are negligible
computational
domain
air flow
long enough to
include the
circulation zone
Figure 13.10 Choosing the computational domain
Avoid simulation of one-dimensional flows in two- or three-

dimensional models. Some one-dimensional parts can be cut from the
computational domain or modeled with a porous medium element
group. For example, in the following figure, the pressure drop in the
one-dimensional part can be evaluated by an analytical solution or by
the solution of modeling a portion of the one-dimensional part with a
given fluid velocity. The pressure based on the one-dimensional
analysis is then applied to the outlet of the separated three-dimensional
model. Sometimes the one-dimensional part can be modeled using a
short porous medium element group as well.

incoming
flow
environmental
pressure original model
incoming
flow
with or without
porous section
computational model
proper normal-traction
Figure 13.11 Avoid lower dimensional flows in the model
A similar situation happens when a porous medium is too thin

compared with the model dimension. One can actually model it using a
thicker porous medium with a modified permeability. This procedure
is equivalent to using different units in the porous media domain.

E
air flow air flow
air filter with air filter can be modeled

permeability using permeability E/
Figure 13.12 Using equivalent physical models

Special attention should be given to modeling external flows. The

physical boundary of such a problem is infinitely far away from the
region where the action takes place. The finite element model cannot
have an infinitely large domain; hence a truncated domain must be
used. However, the truncated (computational) domain should be large
enough to represent the infinity and, accordingly, represent the fluid
flow in the region of interest accurately.
13.1.5 Choosing proper boundary conditions
The use of appropriate boundary conditions is probably the most

important task in computational fluid dynamics. In addition to the general
considerations given for each flow model as described in the corresponding
chapters, here are some other guidelines on using ADINA-F.
When a pressure condition is to be specified at an open boundary, apply

a distributed normal-traction, rather than a prescribed pressure. A
normal traction is equivalent to external force acting on the boundary,
so the velocity and pressure are solved by the governing equations.
The pressure obtained from the solution will be slightly different from
the value of the normal-traction that is applied, which is reasonable
along open boundaries. On the other hand, if a prescribed pressure is
enforced, the discretized continuity equations on that boundary are
removed from the governing system. Then nonphysical phenomena
could be generated near the boundary.
In an incompressible flow solution, one open boundary can be left free
(no traction/pressure is specified), which corresponds to a zero normal-
traction. However, if multiple open boundaries are subjected to
different pressure conditions, the proper normal tractions must be
applied to each of these open boundaries (see the figure below).

p3 nn=p p
3 1
v v
physical incompressible
model p2 flow model
nn=p p
p1 nn=0 2 1
Figure 13.13 Multiple normal-traction conditions in

incompressible flow models
In compressible flows (in both low- and high-speed compressible flow

models), no zero normal-traction should be applied. The absolute
pressure value must be specified. Note that the same conditions can
also be applied to incompressible flows for the case considered in Fig.
13.13. In high-speed compressible flows, the pressure conditions are
usually defined and no additional normal-tractions should be applied.
nn=p
3
compressible and
incompressible nn=p
flow models nn=p 2
1
Figure 13.14 Multiple normal-traction conditions in

compressible and incompressible flow models

In confined incompressible flows, such as those in natural convection

problems, a fixed pressure must be applied to each of the singly
connected domains, see the following figure.
fluid element fluid element

groups groups
fixed
pressures
solid element
groups walls
Figure 13.15 Fixed pressure condition must be applied to

confined incompressible flows
Use special boundary conditions as much as possible in high-speed

compressible flow analyses. Remember that the prescribed conditions
are always applied to the conservative variables, rather than to the
primitive variables. For example, in internal flows, subsonic-inlet,
subsonic-outlet and supersonic-outlet are frequently used. In external
flows, external conditions are frequently used. If a prescribed velocity
in a certain direction, say a , is absolutely necessary, a few choices are
available. They are listed here.
1) If the normal direction of the boundary n corresponds to a ,

apply the subsonic-inlet condition with the slipc parameter
zero. The thermal condition can also be specified as either a
prescribed temperature or a zero heat flux.
2) Apply a wall condition of type tangential with the
tangential direction specified as a .
3) Assuming a zero first derivative of density, an external
condition of type ( , , v) , with specified velocity, zero heat
flux or prescribed temperature, and zero derivative of

density, can be applied (using slipc=0 to enforce the flow

direction).
4) Assuming a zero first derivative of pressure, a supersonic
condition of type ( p, , v) , with specified velocity, zero heat
flux or prescribed temperature, and zero derivative of
pressure, can also be applied (using slipc=0 as well).
5) When the velocity is the result of a boundary condition from
a rotating disk, for example, a wall condition of type
rotational must be applied.
Do not apply a moving boundary condition if the domain deformation

is not of interest. Instead, a proper boundary condition applied on a
fixed boundary may be more adequate. For example, a circular laminar
jet produces a free surface that represents the front of the jet flow. If a
free surface is specified (that is, a Lagrangian formulation is applied to
the free surface while an ALE formulation is applied to the whole
domain), the solution is difficult to reach. In fact, the steady-state
solution cannot be obtained in this model. Instead, a reasonable
boundary condition set applied to an axisymmetric plane can be used to
well approximate the problem, see the following figure.
z
nn =0
v=0
slip-wall w=0
v w y
axisymmetric
Figure 13.16 Solution of a circular laminar jet

13.1.6 Choosing the proper element
The 2D 3-node triangular and the 3D 4-node tetrahedral elements are

suitable for high Reynolds number and high Peclet number flows, although
they can also be used for low Reynolds number and low Peclet number
flows as well. For high-speed compressible flows, only these two elements
are available. The 2D 9-node quadrilateral and 3D 27-node brick elements
are very good elements for highly viscous flows, characterized by low
Reynolds number and low Peclet number flows. The 6-node triangular
element is usually used in conjunction with the 9-node quadrilateral
element as its triangular counterpart. Only the 9-node quadrilateral element
and the 6-node triangular element are compatible and therefore can be used
together.
On the other hand, the FCBI elements can be successfully applied to all
incompressible, slightly compressible and low-speed compressible flows.
These elements are very accurate provided the element quality (aligned to
the flow direction, smoothly varying in size, etc.) is reasonably good.
To obtain an accurate solution, the mesh size should be sufficiently
fine, especially in regions where there are high solution variable gradients.
Such regions usually occur near solid boundaries, circulation zones and
shocks in high-speed compressible flows. However, the number of
elements is also limited by the capacity of the computer in terms of storage
and speed. For a fixed number of elements that allows the analysis to be
completed with reasonable resources, in general, the elements should be
distributed such that x f const. , where x is the element size and
f is a target solution variable. This variable could be the pressure in high-
speed compressible flows where shocks are fundamentally important
phenomenon to be modeled; it also could be the velocity for flows in
boundary layers; etc. Larger element sizes should be used in regions where
a smooth varying solution is expected. On the other hand, wherever a
sharply changing solution is expected, the element sizes should be fine
enough to capture the solution gradients. Therefore, varying mesh sizes are
generally used in the discretization of the computational domain. However,
the transition of the element sizes from fine to coarse should be as smooth
as possible. For an optimal mesh, the gradient of the element sizes should
be approximately proportional to the magnitude of the second derivatives of
the solution variable.

13.1.7 Choosing the material model
The selected material model should pertain to the fluid that is modeled.
The material constitutive relations may be constant, time-dependent,
temperature-dependent, etc. A proper consistent unit system must be used
to avoid round-off errors. Here the feature of automatic
nondimensionalization in the fluid model of ADINA-F can be very useful.
In case gravity effects must be included, consider the option to remove
hydrostatic pressure.
13.1.8 Choosing the solver
There are basically two types of solvers that can be selected: direct
solver (sparse or SKYLINE) and iterative solvers (RPBCG, RPGMRES or
Multi-grid).
The Gauss elimination method (SKYLINE) should not be used for any
problem except for research studies.
If the matrix conditioning is always good (in the solution time range),
the iterative solvers are the best choices. Flows of low
Reynolds/Peclet/Rayleigh numbers, FSI problems with small
displacements, etc. could be such problems. Even for some sensitive
problems the nondimensionalization can be used to improve the matrix
conditioning (see Section 13.3). Note that improper units can cause failure
in convergence using the iterative solvers even for physically stable
problems. The ratios of the factorized diagonal elements are also a good
indicator of a good/bad matrix conditioning.
Usually a few hundred iterations performed with an iterative solver is
reasonable if the number of equations is less than 100,000. When the
number of equations becomes larger (say around 500,000 equations),
thousands of iterations may be required.
In all other cases, use the sparse solver. The sparse solver is a direct
solver and hence will always give the solution of the algebraic equations.
Of course, these equations may be ill conditioned and the sparse solver will
give messages to that effect, and indeed indicate physical instabilities in the
problem. When the conditioning of the matrix is poor, the solution from
direct solvers may also be misrepresented because of the round-off errors.
Use of proper units will always help in this situation.

A few numerical experiments can be performed as well to determine

which solver is best for your problem.
13.1.9 Model testing
Special attention must be given to ensuring that a model with

complexity has been generated correctly and with the appropriate physical
parameters. There are many possibilities in which a model could have been
constructed incorrectly. In order to obtain an accurate solution, testing is
essential. It usually turns out that a good approach in that regard saves time
as well. Here are steps that help progress toward a final model.
(1) Analyze the model with your experience, available experimental data
and known similar solutions, etc. Understand the nature of the problem
to be modeled. Is it a two-dimensional, axisymmetric, or three-
dimensional flow? Is it an incompressible, slightly compressible, low-
speed or high-speed compressible flow? Or is it a porous media flow?
What is the best way to model the problem in terms of accuracy,
numerical stability, computer storage and CPU available? Which
factors could be fundamentally difficult in the model? Can any part of
the model be simplified?
(2) If possible, test a simplified model first. For example, test a two-
dimensional model before creating the three-dimensional model. In
this model, follow the steps described below as well.
(3) Test the model using a coarse mesh and material data and loads that
permit a quick solution (e.g., large viscosities, small velocities, small
normal-tractions, etc.). The purpose of this test is to logically check the
input data. In this model, the types of materials and the types of
boundary conditions should be the same as those that will be used
eventually. Since the model is very easy to understand and to solve in
terms of physics as well as in terms of numerical computations, the
solutions will be easily obtained. In case of errors in the input, they can
be quickly identified and corrected as well. The solution is not
expected to represent the real solution, but must be physically
reasonable.
If any difficulty occurs, make sure the problem set is indeed easily
solved. You may even exaggerate the case by imposing a condition
that will produce a solution extremely close to the initial condition.
You may also iterate only once (by specifying a large convergence

tolerance) to visualize the immediate behavior of the solution. With

the help of the information printed in the *.out and *.log files and of the
visualization of the results, try to find out where the difficulty lies.
If an incorrect immediate solution is found near a boundary, a

boundary condition could have been applied incorrectly near that
location.
Velocity vectors that show fluid flowing into an interior location
usually indicates a false connection of the elements in that region as
generated in the incorrect use of the AUI. This error can also be
found if the mass is obviously not conserved.
Fluid flowing into the opposite direction than is physically expected
could possibly indicate that a wrong direction in the boundary
condition has been used. Those conditions could be a normal-
traction (remember that the normal-traction depends on the
orientation of the boundary geometry!), a negative prescribed
velocity, etc. In addition, a required wall boundary condition may
not be specified too.
Check if there is any error or warning message. If there is more
than one message, check the first one first. An extremely large ratio
of factorized diagonals (say 1020 ) in the first equilibrium iteration in
the first time step is an indicator that the model is ill posed, or
mathematically unstable. Perhaps the to be fixed pressure is not
specified in confined flows. Zero viscosity or heat conductivity
might have been used in the material data. In compressible flow
models, a zero pressure or temperature condition is probably used,
either explicitly specified or assigned as the default in the program.
If a zero unfactorized diagonal corresponds to a velocity or
temperature degree of freedom, the material data could be incorrect.
For example, a zero viscosity or zero heat conductivity might have
been specified.
1) Change some key parameters one by one to see if the solution

responds reasonably. The solution behavior can also be revealed
from this testing.
2) Further test the model using the real material data and loads. If
difficulties occur in convergence, find out which parameter is
most sensitive. Usually this can be a large prescribed velocity, a
small viscosity, a large heat expansion coefficient, a large heat

flux load or prescribed temperature and so on. The difficulty is

usually related to a high Reynolds number, high Peclet number,
high Rayleigh number, etc. The key is not to change the model
that has been tested in step 3 except for the values corresponding
to the types of materials and boundary conditions. This excludes
the possibility of creating other errors. Therefore any difficulty
that occurs in this stage must not result from the input errors that
could have happened and corrected in step 3. Use any of the
methods, or a combination of them, introduced in Section 13.3 to
achieve a converged solution. Since the mesh is coarse, the
solution can also be obtained quickly. The model here should be
tested thoroughly, exposing difficulties and overcoming them
accordingly. At the completion of this stage, the tested model
should be numerically stable and physically reasonable. We
emphasize the numerical stability here, because the model in the
final run (next step) could be slightly more difficult. In order to
reach a stable model in the final stage, some slightly more
difficult conditions can be introduced at the current stage. It is
worth spending more time in this stage, tuning the model to
make sure it is in good condition. In this test, knowledge in
choosing time step sizes or increments of loads, reasonable CFL
numbers, properly nondimensional parameters, and so on is
gained.
3) Only refine the mesh in the last stage of analysis and expect a
solution with high resolution. In general, a finer mesh will make
the model numerically more stable and therefore easier to
converge. However, in certain cases, the larger number of
equations may cause large round-off errors as well. If the
problem is sensitive, difficulties may occur again. If this is the
case, adjust the control parameters and tested again in step 5.
When iterative solvers are used, the maximum number of
iterations in these solvers may need to be increased if the number
of equations is very large (typically larger than 250,000). The
finer mesh solution may also resolve more complicated physical
phenomena, such as re-circulations, shocks etc., that could not be
simulated using coarse mesh in step 5. In this case, smaller
increments of loads, smaller time steps, a smaller CFL number
may need to be used.

13.3: Strategies toward obtaining converged solutions
13.2 Moving mesh control
In models that have moving meshes, nodal displacements are either

prescribed by means of physically moving boundary conditions or
arbitrarily moved in preserving the best mesh quality. The automatic
procedure for moving the mesh implemented in ADINA-F assigns
displacements to the nodes that can be arbitrarily moved. However, in
certain cases, the automatic procedure may not be able to control the mesh
quality well. An example is shown in Fig.13.17, where a ball is moving. A
typical element shown has been distorted due to the ball movement even
though the computational domain is simple. The reason is that there are no
moving conditions on the outer boundary of the domain, i.e., point 2 is
fixed and the line 1-2 becomes largely skewed.
1 1
Figure 13.17 Distorted element in a moving boundary problem
If we force point 2 to follow point 1, then line 1-2 will not be skewed
and the element will keep its quality. The revised situation is shown in
Fig.13.18. In this case, point 1 is the leader and point 2 is the follower.

1 1
leader
follower
2 2
Figure 13.18 Use of leader-follower option
It is clear that the specific geometry plays a key role in the ALE
formulation. In the AUI, the geometry is entered using points, lines,
surfaces and volumes; the mesh is generated based on the geometry.
It is also clear that there is no guarantee that the mesh will not overlap
during arbitrary movements of the boundaries. However, as demonstrated
earlier, if the domain is simple and adequate, the mesh will not overlap. A
good geometric domain is akin to a convex set. Recall that a convex set is
a domain in which a line joining any two interior points stays in the
domain. The simplest good convex domains are a brick and a
tetrahedron in three dimensions and a quadrilateral and a triangle in
two dimensions. For more complex domains, a good mesh for the ALE
formulation can be achieved only if the domain is properly divided into
simple and adequate sub-domains.
Note that a mesh overlap is likely to occur in those areas in which the
deformations correspond to compression of the mesh. Some other
helpful tips are:
Try to create surfaces/volumes that are as close as possible to convex

sub-domains.
If a certain region cannot be converted into adequate sub-domains, you
may define it as moving rigid body (that is, all the displacements of the
fluid nodes in this region are the same). You can achieve this by using
the leader-follower command to constrain all of the corner points of the
region.

For those regions within the domain that have regular shapes, try to
create brick-like sub-domains. For the remaining region, keep them
rigid. For example, if the fluid mesh near a deformable structure is
likely to overlap, you may set the surrounding surfaces/volumes to
move along with the structure, to minimize the deformations of the
fluid mesh near the structure.
Consider all possible mesh movements that can occur during the
solution process while you design the sub-domain topology. In
particular, consider critical mesh conditions in which you must control
the mesh quality so that the solution is represented accurately under
these conditions.
In certain cases, in order to prevent overlap, one can use the contact
surfaces (defined in the solid model). The contact surfaces may be
embedded there solely for the purpose of preventing the fluid mesh
from overlapping and will not affect the solution obtained. In the
example illustrated below, we have added two contact surfaces that
limit the region in which the deformable structure can move. The
lower contact surface has a short distance from the real wall that is
modeled in fluid model. This distance is small enough such that the
position of the structure is always above the wall and large enough such
that the fluid meshes will always be valid. If the contact surfaces are
placed in proper locations, they will not affect the final solution. In
transient analyses, the contact may physically happen. In this case, the
upper contact surface may be a real structure. The lower contact
surface would represent the physical wall.
rigid contact surfaces

(invisible in fluid domain)
deformable structure
fluid domain
Figure 13.19 Using contact surfaces in FSI analyses

Note that the above discussion may not need to be referred to if the
problem results in small displacements, say less than one element size. The
automatic procedure of moving meshes using the Laplacian procedure is
usually successful in the solution of such problems.
13.3 Strategies toward obtaining converged solutions
In order to obtain converged solutions, we must know what factors

could make the solution divergent.
Recall that our task is to obtain the solution at time t + t of the
nonlinear finite element equations represented by F(X) = 0 . We take the
Newton-Raphson method (see Section 11.1) as an example here.
The convergence of the Newton-Raphson iteration strongly depends on
the initial guess X 0 that is equal to the solution calculated in the last
solution step t X or the given initial conditions. A necessary condition for
the iteration to converge can be stated simply as follows:
Theorem: If the matrix A is not singular, there exists a set

{ }
D = X; X t +t X r , r > 0 and for any X 0 D , the Newton-Raphson
scheme converges.
This theorem indicates that, once the initial guess is close enough to the
solution, the Newton-Raphson iteration will converge. On the contrary,
therefore, if the final solution is very different from the initial condition, the
iteration procedure may diverge.
Certain solvers are used to solve the linearized equation. The sparse
solver is a direct solver that is based on the Gauss elimination method. The
iterative solvers are also linked to the Gauss elimination method through
their preconditioned matrices generated in the incomplete Cholesky
factorizations. Without mentioning other round-off errors that may occur
in the solution, the Gauss elimination method introduces round-off errors
that can be significant when the problem is ill-conditioned, expressed by
the condition number A A 1 , where we have not used the row/column
interchanging technique (for the purpose of efficiency in computing), the
round-off errors may be out of control and lead to a wrong solution or

destroy the convergence of the whole iteration. More specifically, what we

have solved is actually the solution of the perturbed equation
k = ( A + a) 1 (B + b)
X (13.1)
where a is a matrix containing the round-off errors during the LDU

decomposition and b is a vector containing the round-off errors while the
right hand side is computed. The solution error can be solved from
Eq.(13.1)
k X k = ( A + a)1 (b aX k )
X
This equation clearly indicates that the solution errors are proportional to
the round-off errors of the matrix as well as of the right-hand side.
The vector b is proportional to the right-hand side
bi 21m Bi
where m is the precision of the machine (e.g., m = 32 for 32 bit

machines). If our formulation results in a small right-hand side in
magnitude, the error of b is negligible.
Let the computed L, D and U be represented by L , D and U
respectively, we have the following evaluation of errors from the matrix
( )
aij N d 21m 3 Aij + 5 Lik D kk U kj + O ( (21m ) 2 )
where N d is the number of equations. There are three factors that can
contribute errors. Two of them are N d and m. When the number of
equations is not so large or the machine precision is very high (e.g., when
64 bit machines are used), the errors can be reduced. However, the two
factors are usually unchangeable. The third factor is L D U . If the
ik kk kj
matrix is diagonal dominant as is usually encountered in structural

analyses, it is the same order of Aij . In fluid flow problems, this factor
could become very large if small pivots are used. Small pivots could be
encountered if the matrix structure or the so-called symbolic ordering is not

suitable for the problem or the matrix is truly ill conditioned. ADINA-F
has been optimized to generate the matrix structure such that no artificial ill
conditionings may occur. Therefore the small pivots usually come from the
ill conditioned matrix.
The conditioning of the matrix is unknown unless we invert the matrix
and evaluate the norms of the two matrices. But if we had inversed the
matrix, we would not need to know the condition. ADINA-F prints out
some useful information during the LDU decomposition (in SKYLINE and
sparse solvers) or during the incomplete Cholesky factorization (in iterative
solvers), which can be used to predict the conditioning of the matrix. They
are
Amin , I min = minimum unfactorized diagonal and the

corresponding equation number defined as
Amin = AI min I min = min{ Aii } .
Amax , I max = maximum unfactorized diagonal and the
Amax = AI max I max = max{ Aii } .
Dmin , I min = minimum factorized diagonal and the
Dmin = DI min I min = min{ Dii } .
Dmax , I max = maximum factorized diagonal and the
Dmax = DI max I max = max{ Dii } .
The conditioning of the matrix can then be predicted by
Dmax
A A 1
Dmin
Hence, it must be the objective to have good initial conditions for the
solution at every time step, to improve the matrix conditioning and to
minimize the magnitude of the right-hand side. Some efficient methods for
these purposes are described in the next sections.

13.3.1 Incremental solution procedure
When the initial condition is far from the final solution, the problem can
be solved in a few time steps, such that the solutions at any two consecutive
time steps are close enough to satisfy the convergence condition in the
Newton-Raphson iteration method.
Consider the example of a forced convection problem as shown in
Fig.13.20, where the bottom wall is heated. The steady state solution is
required.
When the open boundaries are closed, it is the well-known Rayleigh-
Benard convection problem that is very unstable and has multiple solutions
when the Rayleigh number is larger than a critical value. The velocity that
is prescribed at the inlet forces the fluid to flow through the cavity and the
heat is therefore transported out of the domain primarily by convection.
The forced fluid boundary condition is thus a stabilizing factor in this
model compared with the Rayleigh-Benard convection problem. However,
the high Reynolds number presents a possible difficulty as well when only
the fluid flow is solved.
v n=1
=0
=0
v=0
4
Re=10
8
Ra=10
=1
v=0 nn =0
qn =0
Figure 13.20 Forced convection in a cavity heated at bottom
We may perform the computation in two stages: (1) compute the fluid
flow without the influence of heat transfer; (2) compute the heat transfer

with the fluid flow. In each stage, we may achieve the convergence in a
few time steps. The fluid flow is computed first because it is a stabilizing
factor in the whole model. At the same time, we take the hydrostatic
pressure out of the model (see chapter 2 for details regarding this issue).
In stage 1, we execute two time steps. In the first time step, we specify
conditions such that the Reynolds number is 103 , which is an easy problem
to solve, so the zero initial condition is fine in the Newton-Raphson
iteration. The solution obtained in step 1 is used as the initial condition for
time step 2, where we prescribe the full velocity load at the inlet and the
Reynolds number becomes the required one, Re = 104 . During this stage,
the bottom temperature is kept as zero, so the Rayleigh number is always
zero (see the figure below).
v n=0.1 v n=1
=0 =0
=0 =0
v=0 v=0
=0 =0
v=0 nn =0 v=0 nn =0
qn =0 qn =0
3 4
Re=10, Ra=0 Re=10, Ra=0
time step 1 time step 2
Figure 13.21 Computation in stage 1 for the forced convection

problem
In stage 2, we keep the fluid boundary conditions unchanged, while

slowly heating the bottom wall. This may be accomplished in a few time
steps as well, say two to six. Usually the greatest difficulty in converging
in the iteration occurs when the solution changes dramatically. It is
effective to impose, in the first step of this stage, a condition of a small
Rayleigh number, say Ra = 104 , with the initial conditions obtained at the
end of the stage 1. Then the heating can be increased rapidly. Of course,
the solution obtained at every time step is automatically used as the initial
condition for the next time step.

v n=1 v n=1
=0 =0
=0 =0
v=0 v=0
-4 ...
=10 =1
v=0 nn =0 v=0 nn =0
qn =0 qn =0
4 4 4 8
Re=10, Ra=10 Re=10, Ra=10
time step 3 time step n
Figure 13.22 Computation in stage 2 for the forced convection

problem
To achieve the described computational procedure, we may use time-

dependent materials, perform restart runs, etc. Here we use time-dependent
boundary conditions. First, we define the inlet velocity and the bottom
temperature:
v = v f1 (t ) ( v = 1)
(13.2)
= f 2 (t ) ( = 1)
where the time functions are defined as shown in the figure below.
t t
1 1
1 2 1 2
f1(t) f2(t)
Figure 13.23 Time functions used for the forced convection

problem

Secondly, we choose the time steps to obtain the solution (also shown here
are the corresponding inlet velocity and the bottom temperature computed
according to Eq.(13.2)):
time at the
time time step bottom
end of the inlet velocity
step length temperature
time step
1 0.1 t1 =0.1 0.1 0
2 0.9 t2 =1 1 0
3 0.0001 t3 = 1.0001 1 0.0001

n 2- tn1 tn = 2 1 1
Note that a steady-state analysis is performed; so the solution obtained

at the end of a time step is the steady-state solution under the various
conditions at the times considered. Remember that time has no physical
meaning in the model; it has merely been used as a parameter to control the
boundary conditions.
13.3.2 Improving initial conditions using restart runs
The initial condition in a solution provides the initial guess for the first
time step solution. In certain cases, this condition is critical for
convergence of the iterations. An example of transient fluid-structure
interaction analysis is shown in Fig.13.24, where a water pipe is subjected
to small excitation force acting at the bottom. In this example, we have
assumed that the gravity force is far larger than the excitation force. If we
were to start from the zero velocity condition and a straight pipe to
calculate the dynamic response, the solution would diverge (because of the
poor initial guess of the final solution). Let d g and d F be the deformation
scales of the pipe due to gravity and the excitation force, respectively, we
expect that d F dg .

vin g nn =0
FSI interface
F(t)
Figure 13.24 Water pipe subjected to a small periodic

excitation force
An effective solution will be obtained in two stages. In the first stage, a

steady-state analysis of the fluid and structure is performed solely under the
gravitational force. In the second stage, a transient analysis of the response
due to the excitation force is performed using the solution obtained in stage
1 as an initial condition. This procedure follows closely the actual physics,
and is also important for the numerical computations. If the gravity and the
excitation force are added simultaneously to the model, the excitation force
will be negligible and the pipe will hardly reach the deformation scale d g
because of the small time steps needed in the transient analysis.
We can achieve this result by either of the two methods:
(1) Use a very large time step in the first time step, during which only the
gravity force is added. The steady-state solution corresponding to only
the gravitational force will be obtained at the end of this time step.
Then, reasonable time step lengths are used for the real transient
analysis in which the excitation force becomes active. These steps are
performed to simulate the dynamic response due to the excitation force.
This method is simple and can be performed in one execution. The
drawback of this method is that the first time step length is arbitrary.
The transient analysis starting from an arbitrary large time step looks
unnatural as well.
(2) Perform two runs. The first run performs the steady-state analysis with
only the gravity force present. The second run performs the transient
analysis, based on the initial condition obtained in the first run.

13.3.3 Improving the initial conditions using the mapping file
A solution can always be used as the initial condition for another model
as long as the computational domain of the present model lies inside the
domain of the earlier model analyzed and a mapping file was saved for the
first model. If the two solutions are similar, it serves as a good initial
condition.
An advantage of this procedure is that the elements and other conditions
of the two runs are independent as well.
13.3.4 Improving the matrix conditioning using the CFL option
The conditioning of the matrix A in the Newton-Raphson iteration also

plays an important role. The matrix could be ill-conditioned due to
physical phenomena. This usually happens in high Reynolds number
flows. The conditioning can also be bad if the element sizes vary
dramatically in the computational domain. When an improper system of
units is used, the conditioning of the matrix could be poor as well (see next
section in this case). A simple solution is to add a relaxation factor in the
Newton-Raphson method
X = ( A + 1I )1 B
where is the relaxation factor. The factor changes the values of the
diagonal elements, thus improving the conditioning of the matrix. When
is small, it not only improves the matrix conditioning, but also reduces the
requirement to have good initial conditions. However, a small also
slows down the convergence rate in the Newton-Raphson method. There is
no universal factor that can be used for all cases. Obviously, the best factor
depends on the flow pattern, element sizes, etc. Furthermore, this
relaxation factor is not dimensionless. Therefore, used in a continuity
equation may not be good for the momentum equations, etc.
The automatic time-stepping CFL option in ADINA-F provides an
effective tool for improving the matrix conditioning. We have
1
X = ( A + CFL D)1 B

where D is a matrix close to a diagonal matrix. It is formulated during the

time discretization, on element sizes, velocities, viscosity, etc. More
importantly, the CFL number CFL is dimensionless and its magnitude can
be reasonably predicted from the stability analysis of explicit time
integrations. Its lower limit is 1 (the critical value for explicit methods) and
its upper limit is infinity when the matrix conditioning is very good. A
range of 102 to 107 could be an initially good choice for many cases.
13.3.5 Improving the matrix conditioning using proper units
An improper system of units can also result in a bad matrix

conditioning. For example, if we have a matrix system
103 103 v[m/s] ...

AX 3 =
10 0 p[Pa] ...
that is based on the velocity and pressure units as shown in the equation.
The conditioning of the matrix is
A
A 1 = 103 (103 + 103 )

If we only change meters to millimeters, we obtain a different matrix

system that reads
1 1 v[mm/s] ...
AX 2
=
1 0 p[kg/mm-s ] ...
and the conditioning of the matrix becomes
A A 1 =2

This procedure can be looked at from another point of view. If we use a

length scale L = 103 and velocity scale V = 103 in the procedure to
automatically nondimensionalize the variables, as available in ADINA-F,
we will achieve the same conditioning. Therefore, the task to choose a

proper unit can be accomplished by choosing proper scales for

nondimensionalization.
In order to obtain a general guideline for choosing the scales for
nondimensionalization, let us consider a model case for incompressible
flows
a c v ...
AX =
b 0 p ...
where we have
x 2 vx
a x d 2 , = 1 + +
t
b x d 1 , c x d 1
where d is 2 or 3 indicating a two dimensional or three dimensional

problem. The LDU decomposition is
1 0 a 0 1 ac
A = LDU = b
a 1 0 bca 0 1

Let us apply the scales D* , L* and V * to the model. We determine these

scales from the following conditions
amin amax
bc = 1, =1
(bc a ) min (bc a) max
The result will then be used for general incompressible flows, where the
element sizes, velocity magnitudes, etc. vary.
The final scales that can be used are

D* = , L* = x, V * =
x

where
x = xmin xmax , = min max

2 2
xmin xmax
min = 1 + , max = 1 + + Re
t t
with Re representing the maximum cell Reynolds number.

On the other hand, when temperature is included in the computations,
obvious choices of the related scales are
C * = Cv , T * = ( ) max
The temperature datum is a very important factor in minimizing the

magnitude of the right-hand side when there are radiation or specular
radiation boundary conditions in the model. Recall that the heat flux of
these boundary conditions is proportional to
qn (e4 4 ) (e ) (13.3)
If the environmental temperature is high, as is usually the case in radiation

heat transfer and the calculated temperature is low, as is usually the case in
the first time step, Eq.(13.3) leaves a term of extremely large magnitude on
the right-hand side of the equations. This difficulty can be avoided by
choosing the temperature datum
r* = e
In this case, the radiation heat flux will be
qn
This term will be absorbed in the matrix, leaving zero on the right-hand
side.
In compressible flows, ADINA-F forces the temperature scale and
datum as

2
V*
T = * , r* = 0
*
to keep consistency of the state equation, regardless of the input.

In the direct computing of FSI two-way coupling problems, the matrix
conditioning is worse than in individual fluid and solid models, not only
because more equations are solved, but also because of the large difference
in magnitude of the matrix elements. A typical diagonal term in the matrix
of the solid region is related to Youngs modulus that could be as high as
1010 in the unit systems used. A typical diagonal term in the matrix of the
fluid region is related to the viscosity that could be as low as 106 in the
same unit system. Therefore, a proper choice of nondimensionalization is
even more important in this case.
13.4 Consistent units, SI units and USCS units
ADINA-F solves the fluid equations based on Newtons second law

expressed in the form
F = Ma
where F , M and a are force, mass and acceleration, respectively. In the

case of compressible flows, the state equations in the form of the ideal gas
law are used. These two equations accept any consistent units. However,
there is sometimes confusion with some popular USCS (U.S. Customary
System) units, like pound-force ( lb f ) and pound-mass ( lb m ). We want to
discuss briefly on the use of consistent units.
In order to keep the units consistent for all data input for a model, one
must determine the primary dimensions. Any variable and parameter that is
not in the primary unit must be derived from these units. We show two
consistent units here as examples.
The first example starts with the primary units defined in the
(automatic) nondimensionalization procedure in fluid model: length ( L ),
velocity (V ), density ( D ) and temperature ( T ). Other variables and
parameters are then derived from the primary units. The following table

13.4: Consistent units, SI units and USCS units
gives the expressions of generally consistent units and typical examples in

SI and USCS units. In SI units, we use ( L,V , D, T ) = (m, m s,kg m3 , K)
and in USCS units, we use ( L,V , D, T ) = (in,in s,lb m in 3 , o R ) .
Table 13-1 Consistent units derived from the primary units:

length, velocity, density and temperature
Variables Units SI units USCS units

x L m in
v V m/s in/s
D kg/m3 lb m in 3
T K o
R
C p , Cv V2 T m 2 K-s 2 in 2 o R-s 2
t L /V s s
p DV 2 kg m-s 2 ( Pa) lb m in-s 2
DVL kg m-s lb m in-s
3 3
k DLV T kg-m K-s lb m -in o R-s3
F DV 2 L2 kg-m s 2 ( N) lb m -in s 2
e V2 m2 s2 in 2 s 2
The second example starts with the primary units length ( L ), time ( t ),
force ( F ) and temperature ( T ). Other variables and parameters are then
derived from the primary units. The following table gives the expressions
of generally consistent units and typical examples in SI and USCS units. In
SI units, we use ( L, t , F , T ) = (cm,s,dyne, K) and in USCS units, we use
( L, t , F , T ) = (ft ,s,lb f , R) .

Table 13-2 Consistent units derived from the primary units:

length, time, force and temperature
Variables Units SI units USCS units

x L cm ft
t t s s
F F dyne lb f
T K o
R
Ft 2 L4 g/cm 3
lb f -s 2 ft 4 ( slug ft 3 )
v Lt cm s ft s
C p , Cv L2 (t 2T ) cm 2 K-s 2 ft 2 o R-s 2
p F L2 dyne cm 2 lb f ft 2 ( psf)
Ft L2 g cm-s lb f -s ft 2 ( slug ft-s)
o
lb f R-s
k F (Tt ) g-cm K-s3
( slug-ft o R-s3 )
e L2 t 2 cm 2 s 2 ft 2 s 2
The key in keeping units consistent is to always work with the primary
units rather than conventional units. For example, in the second example
above, the density is in the unit lb f -s 2 ft 4 , not in the unit lb m ft 3 .
Another common approach is to work using nondimensionalized units.
That is, all data input has been made dimensionless. This method is
popular with scientific researchers. This approach then uses the Reynolds
number, Prandtl number, etc. in the definition of the problem to be solved.
13.5 Use of memory and disk
13.5.1 Memory usage
The memory used in ADINA-F falls into two categories: the basic
memory used to store the data input and generated data (including the

13.5: Use of memory and disk
matrix system) and additional memory that is required by the specific

solver selected. The memory that ADINA-F outputs is in words. One
word equals 4 bytes and 8 bytes, respectively, on 32 and 64 bit machines.
In the following descriptions, a 32-bit machine is assumed.
(2) Basic memory
This part includes the memory for the solution variables, element group
data, boundary conditions, etc. The major portion of the memory is of
course used to store the matrix system. To ease the explanation of the
memory usage here, we assume that these data are in-core as well.
Although the element group data and the matrix may be out-of-core, the in-
core solution, if it can be performed, is generally more effective.
The basic memory used can be divided into two portions
m = m1 + m2
where m1 and m2 are the memory for preliminary storage and for storing
and manipulating the fluid matrix. The amount needed can be predicted by
m1 = n N n + e N e words
m2 = 5 n N n = 5 m nd N d
2
m d words
where N n is the number of nodes, N e is the number of elements,

N d ( nd N n ) is the number of equations (or called the number of degrees
of freedom), nd is the number of solution variables, and the 's are
factors. n and e vary in the ranges
n 25-60, e 20-25
depending on the problem type. The factors that may increase the use of
memory are, for example, three-dimensional flows, including temperature
degrees of freedom, moving mesh conditions, turbulence models, etc. The
factor nd varies in the range
nd 1-7

corresponding to the number of variables (u , v, w, p, , K , ) included in

the model. It is 1 if temperature is the only variable and 7 if three-
dimensional turbulent flows including heat transfer are computed. Except
the solver SKYLINE, all solvers store the matrix in a compact form.
Assuming that either the sparse solver or iterative solver is used, the factor
m depends then only on the type and the connectivity of the elements
8 in triangular 3-node element

25 in quadrilateral 9-node element

m (13.4)
30 in tetrahedral 4-node element
125 in brick 27-node element
It is clear that, for most realistic problems, m2

m1 . Therefore, it is
reasonable to keep all data corresponding to the memory m1 in-core.
The direct solver SKYLINE and the iterative solvers RPBCG and
RPGMRES automatically use the hard disk (out-of-core) once the memory
m2 is not sufficient. However, the speed of solution can then be much
reduced. If the sparse solver is selected, the initially assembled matrix must
be in core.
(2) Memory used in solvers
This part includes the memory used in the procedure to solve the
equations. All iterative solvers do not require additional memory.
However, if the sparse solver is used, additional memory is required. In the
sparse solver, the original matrix is expanded to a new matrix whose size
depends on the fill-ins to the original matrix. This part of the memory is
highly dynamic and considered separately
M M m2
where the factor M would be typically 2 to10. This part of the memory
can be allocated out-of-core (using disk space) if the memory is not
sufficient and the out-of-core sparse solver is selected. The advantage of
the out-of-core sparse solver is that it does not depend much on the memory

assigned to it. Usually the out-of-core sparse solution takes twice the
amount of time as the in-core.
Regarding the memory usage in ADINA-F, we recommend the
following strategies to manage the memory efficiently:
If the computer has sufficient memory to hold both m and M in-core,

you need only to specify the memory m . The sparse solver, in case it
is used, will dynamically allocate the actual memory as required during
the procedure to solve the equations. If there is not enough memory
available, it will print out a message that indicates lack of memory and
the program stops.
If the computer has only sufficient memory for m , we have to use the
out-of-core sparse solver. In this case, the memory M must be
explicitly specified. The sparse solver will then manage to work in-
core if the memory M is sufficient and go out-of-core otherwise. Of
course, M must be less than the difference between the available
memory and m .
If the available memory is not large enough to hold the required m , the
sparse solver cannot be used. The out-of-core iterative solvers must be
used. Only the RPBCG and RPGMRES out-of-core solvers are
available. No special input is required to have these solvers work out-
of-core. They automatically use the disk space whenever the memory
m is not large enough. However, the speed of solution highly depends
on the blocks of the sub-matrices that are generated. The less memory
available, the larger the solution time.
The above prediction is only for the most frequently encountered cases.
Certain cases need further comments.
When mass transfer and moving meshes are included, the memory for
the solver is divided into three parts
M = M1 + M 2 + M 3
where M 1 , M 2 and M 3 are the memory assigned to the fluid

equations, mass transfer equations and moving mesh equations,
respectively. They are determined according to the ratios
N d : N n : N n 2 . Then, if the out-of-core sparse solver is used for the

fluid, the solvers for the mass ratios and moving meshes may or may
not be in-core.
In iterative computing of two-way fluid-structure coupled problems and
in direct computing of the one-way fluid-structure coupled problems,
the same memory M is assigned to the solvers that are used for both
the fluid and the solid models. In other words, the total memory used is
m + 2M .
In direct computing of two-way fluid-structure coupled problems, the
number of nodes, number of elements, and number of equations, etc.
are all referring to the coupling system.
When specular radiation boundary conditions are applied, the view
factor matrix and radiosity matrix require storage as well. Usually, this
part of storage is much larger than that for the fluid flow equations and
can be evaluated as
m3 = 2 N ns2 + 2 N es2 words
where N ns and N es are the number of specular nodes and elements,

respectively. This storage can be assigned to out-of-core memory. The
view factor matrix is computed first and stored on disk. It will be read
in while the radiosity matrix is assembled. Let nbc and nbi be the
number of sub-matrix blocks while the view factor matrix is computed
and then read-in, respectively, nbi is generally larger than nbc . Let nbr
be the number of sub-matrix blocks of the radiosity matrix; the storage
required in specular radiosity computations becomes
2 N ns2 2 N es2
m3 = +
nbi nbr
The basic memory is thus modified to
m = m1 + max {m2 , m3 }
The computational cost in specular radiosity calculations is directly

related to
CPU O(nbr2 , nbc2 )

When the assigned storage m is not sufficient to hold the view factor
matrix and/or radiosity matrix in-core, ADINA-F automatically works
with disk storage. The numbers, nbc , nbi and nbr are printed to inform
users of the possible computational effort. The allowed maximum
number of blocks is 1000. It would be computationally very inefficient
to have more blocks.
13.5.2 Disk usage
Some disk space is required dynamically during the execution of

ADINA-F. Part of that is occupied by the solution results while another
part is temporarily used during the program execution. The major disk
storage files are listed in the following table. The disk space requirement is
approximately given and may be more or less depending on the type of
problem solved. For example, if the fluid solver runs in-core, the matrix is
not saved on disk; if the specular radiation boundary conditions is not used,
no view factor and radiosity matrix disk storage is required; if the mapping
file is not selected, it will not be printed; etc.
The notation used in this table is
Nn = number of nodes
Ne = number of elements
N ts = number of time steps that solutions are saved
Nd = number of equations
m = the factor as defined in Eq.(13.4)
N ns = maximum number of specular nodes in one group
N es = maximum number of specular elements in one group
ne = number of nodes per element
nd = number of variables (1 to 7 for u , v, w, p, , K , )

Table 13-3 Major files used in ADINA CFD & FSI
File size
File name Explanations Data stored
in words
fort.80 nodal solutions 20 N n N ts
porthole file
(=*.por) element solutions 44 N e N ts
fort.74 nodal solutions 30 N n
restart file
(=*.res) element solutions 12 N e
fort.75
temperature file temperature solutions 2 N n N ts
(=*.tem)
nodal coordinates 6Nn
fort.77 element
mapping file ne N e
(=*.map) connectivities
nodal solutions 2 N d N ts
element group data 24 N e
fort.95 temporary file fluid matrix 2 m N d
specular matrix 2 N ns2
fort.78 temporary file specular view factors 2 N es2
Note that, in FSI computations, additional disk storage is required in solid

models. Please refer to the ADINA Solids & Structures Theory and
Modeling Guide for more information.
13.6 Brief summary of recommendations for solutions
The following is a brief summary of some recommendations regarding the

use of the CFD capabilities when modeling a physical problem. These
recommendations based on the flow to be solved are not exhaustive but
merely given to help a user in starting to employ the ADINA CFD solution
capabilities.

While we refer to the Re flow conditions, the same comments are

applicable to the Pe number flow conditions. Also, while we seem to imply
steady-state analyses, the same comments are applicable to transient
analyses.
For low Re incompressible flows
The 9-node 2D and 27-node 3D elements can be used, if the elements

support the boundary conditions, loading applied, and other features you
want to use. The direct sparse solver is used as the default option for
solution of the algebraic equations, hence the memory requirement and
solution time may be large if the model is large. In particular for 3D
problems, the bandwidth may be large. We hardly use the iterative solvers.
For FSI problems, the FCBI and FCBI-C elements are better used.
Use the FCBI elements, preferably the 4-node element in 2D and 8-node
element in 3D, instead of the triangular and tetrahedral elements (which
however are largely used in free-form meshing and must be used in
adaptive meshing). With these elements also by default the sparse solver is
used, but a smaller bandwidth is encountered. The memory and solution
time requirements may also be large when large models are solved, but not
as large as when using the 9-node and 27-node elements referred to above.
We hardly use the iterative solvers. These elements are recommended for
use in FSI problems.
For very large fluid models, the FCBI-C elements are an attractive option.
With these elements relatively fine meshes may need to be used. As for the
FCBI elements, here too, the 4-node element in 2D and 8-node element in
3D are preferred. Using the FCBI-C elements, the sparse solver can be
used, but the default is the algebraic iterative solver, type 1. Using this
iterative solver, memory requirements are much lower for large models
than using the sparse solver. But for the iteration, it is important to employ
tight enough convergence tolerances. In practice, you may solve the
algebraic equations using the default tolerances for the iterative solver,
study the solution, and if questions arise as to whether a converged solution
has been obtained, thereafter continue the iteration to reach a tighter
convergence tolerance.

For high Re incompressible flows
Using the 9-node 2D and 27-node 3D elements very fine meshes may be
required to have stability, that is, to have element Re numbers smaller than
about 6. The elements are formulated using the standard Galerkin method
(using the velocity-pressure formulation) without upwinding [1]. Hence
these elements are not very suitable for high Re flows.
The FCBI and FCBI-C elements should be used with the comments given
above.
Slightly compressible flows
For slightly compressible flows, the same comments as given above are
applicable.
High-speed and low-speed compressible flows
For high-speed and low-speed compressible flows, only the triangular 2D

and tetrahedral 3D elements are available. The algebraic equations are
solved by default using the sparse solver, but for large models the multi-
grid solver can be used.
Ref.
[1] KJ Bathe, Finite Element Procedures, Prentice-Hall, 1996.


Appendix-1: List of figures
Appendix-1 List of figures
Figure 2.1 Rotating coordinate frame........................................................................31

Figure 2.2 Coordinate system for axisymmetric flows..............................................34
Figure 2.3 Illustration of porous media models.........................................................42
Figure 2.4 A typical example of using ALE coordinates ..........................................69
Figure 2.5 Definition of the ALE coordinate system ................................................70
Figure 3.1 A typical boundary condition set for incompressible, slightly
compressible and low-speed compressible flows ..............................................89
Figure 3.2 A typical boundary condition set for temperature....................................90
Figure 3.3 Prescribed rotational velocity...................................................................93
Figure 3.4 No-slip condition on fixed walls for incompressible, slightly compressible
and low-speed compressible flows ....................................................................97
Figure 3.5 Slip condition on fixed walls for incompressible, slightly compressible
and low-speed compressible flows ....................................................................98
Figure 3.6 Angular velocity boundary condition for incompressible, slightly
compressible and low-speed compressible flows ..............................................99
Figure 3.7 No-slip condition on moving walls for incompressible, slightly
compressible and low-speed compressible flows ............................................ 100
Figure 3.8 Slip condition on moving walls for incompressible, slightly compressible
and low-speed compressible flows .................................................................. 101
Figure 3.9 Moving wall condition (type=tangential) for incompressible, slightly
Figure 3.10 Moving wall condition (type=rotational) for incompressible, slightly
Figure 3.11 Sketch of a free surface condition ........................................................105
Figure 3.12 A typical phase-change boundary condition ........................................ 108
Figure 3.13 A typical gap boundary condition ........................................................ 109
Figure 3.14 Diagram of gap open-close conditions................................................. 110
Figure 3.15 2D Rotor-Stator Interaction..................................................................111
Figure 3.16 Two Passing Trains/Cars..................................................................... 112
Figure 3.17 Sliding mesh interface coupling...........................................................113
Figure 3.18 2-D geometry for sliding mesh interface.............................................. 113
Figure 3.19 Physical conditions at edge of sliding mesh interface.......................... 114
Figure 3.20 A weak rotor-stator interaction ............................................................117
Figure 3.21 Two rotors in a chamber ...................................................................... 118

Figure 3.22 Model setup of a multiple-reference frame .......................................... 119

Figure 3.23 A typical convection boundary condition ............................................123
Figure 3.24 Schematic representation of radiation energies corresponding to an
incident ray ...................................................................................................... 125
Figure 3.25 A typical usage of specular groups ...................................................... 127
Figure 3.26 A piecewise linear function.................................................................. 144
Figure 4.1 A typical boundary condition set for porous media flows ..................... 162
Figure 5.1 A typical boundary condition set for turbulence variables.....................170
Figure 6.1 Definition of a two-dimensional control volume ................................... 190
Figure 6.2 Sub-control volumes and their interfaces within one triangular element
......................................................................................................................... 191
Figure 6.3 Sub-control volumes and their interfaces within one tetrahedral element
.........................................................................................................................192
Figure 6.4 Definition of flow regimes in high-speed compressible flows...............195
Figure 6.5 Illustration of the supersonic condition at inlet......................................196
Figure 6.6 Illustration of the subsonic condition at inlet .........................................196
Figure 6.7 Illustration of the subsonic condition at outlet .......................................197
Figure 6.8 Illustration of the supersonic condition at outlet....................................198
Figure 6.9 A typical boundary condition set for external high-speed compressible
flows ................................................................................................................199
Figure 6.10 A typical fluid boundary condition set for internal high-speed
compressible flows ..........................................................................................199
Figure 6.11 No-slip condition on fixed walls for high-speed compressible flows ..206
Figure 6.12 Slip condition on fixed walls for high-speed compressible flows........207
Figure 6.13 Angular velocity condition for high-speed compressible flows...........208
Figure 6.14 No-slip condition on moving walls for high-speed compressible flows
.........................................................................................................................209
Figure 6.15 Slip condition on moving walls for high-speed compressible flows....210
Figure 6.16 Moving wall condition (type=tangential) for high-speed compressible
flows ................................................................................................................212
Figure 6.17 Moving wall condition (type=rotational) for high-speed compressible
flows ................................................................................................................213
Figure 6.18 External flow condition........................................................................214
Figure 6.19 Sketch for inlet and outlet flow conditions ..........................................215
Figure 9.1 Illustration of fluid-structure interactions ..............................................247
Figure 9.2 Coupling of fluid and solid nodes ..........................................................249
Figure 9.3 Measure of the distance between fluid and solid FSI boundaries ..........250
Figure 9.4 Interpolation of fluid stresses in time in indirect computing of one-way
fluid-structure coupling....................................................................................259

Figure 9.5 Typical example of O-shape 2D F-S-F connections in fluid-structure

interaction analyses..........................................................................................262
Figure 9.6 Possible methods for generating O-shape 2D fluid elements in F-S-F
connections ......................................................................................................262
Figure 9.7 Application of fluid-structure interface conditions to beam elements ...263
Figure 9.8 Typical example of C-shape 2D F-S-F connections in fluid-structure
interaction analyses..........................................................................................264
Figure 9.9 Possible methods for generating C-shape 2D fluid elements in F-S-F
connections ...................................................................................................... 264
Figure 9.10 Typical example of O-shape 3D F-S-F connections in fluid-structure
interaction analyses.......................................................................................... 265
Figure 9.11 Typical example of C-shape 3D F-S-F connections in fluid-structure
interaction analyses.......................................................................................... 266
Figure 9.12 Illustration of overall fluid-structure interactions. The coupled region is
FSI = S p s ...................................................................................... 271
Figure 10.1 Line elements ....................................................................................... 276
Figure 10.2 2D/axisymmetric 3-node element ........................................................ 278
Figure 10.5 3D 4-node element ............................................................................... 282
Figure 10.6 27-node 3D element ............................................................................. 284
Figure 10.7 FCBI 2D elements................................................................................ 286
Figure 10.8 FCBI 3D elements................................................................................287
Figure 10.9 FCBI-C elements..................................................................................289
Figure 11-1 Default set of outer relaxation factors in segregated method...............302
Figure 11-2 Default set of inner relaxation factors in segregated method...............304
Figure 11-3 Ordering in sparse solver .....................................................................312
Figure 12.1 Prescribe a radial velocity ....................................................................322
Figure 12.2 Sketch of conjugate heat transfer .........................................................325
Figure 12.3 Application of pressure-datum feature to fluid-structure interaction
problems. .........................................................................................................327
Figure 12.4 A typical case of reaching an unacceptable element............................329
Figure 12.5 Valid mesh with a proper subdivision of the domain...........................329
Figure 12.6 Definition of the leader-follower options.............................................331
Figure 12.7 Definition of VOF-wall angle ..............................................................343
Figure 13.1 Typical problems that incompressible flow model can be applied ......361
Figure 13.2 Typical problems that slightly compressible flow model can be applied
.........................................................................................................................361

Figure 13.3 Typical problems that low-speed compressible flow model can be
applied..............................................................................................................362
Figure 13.4 Typical problems that high-speed compressible flow model can be
applied..............................................................................................................362
Figure 13.5 Typical problems that porous media flow model can be applied.........363
Figure 13.6 Example of using porous media flow model........................................ 363
Figure 13.7 Time step sizes in periodic solutions ................................................... 367
Figure 13.8 Time step sizes in wave propagation problems.................................... 368
Figure 13.9 Time steps in diffusion problems ......................................................... 369
Figure 13.10 Choosing the computational domain.................................................. 370
Figure 13.11 Avoid lower dimensional flows in the model .................................... 371
Figure 13.12 Using equivalent physical models...................................................... 371
Figure 13.13 Multiple normal-traction conditions in incompressible flow models 373
Figure 13.14 Multiple normal-traction conditions in compressible and
incompressible flow models ............................................................................373
Figure 13.15 Fixed pressure condition must be applied to confined incompressible
flows ................................................................................................................ 374
Figure 13.16 Solution of a circular laminar jet........................................................ 375
Figure 13.17 Distorted element in a moving boundary problem............................. 381
Figure 13.18 Use of leader-follower option ............................................................382
Figure 13.19 Using contact surfaces in FSI analyses .............................................. 383
Figure 13.20 Forced convection in a cavity heated at bottom................................. 387
Figure 13.21 Computation in stage 1 for the forced convection problem ...............388
Figure 13.22 Computation in stage 2 for the forced convection problem ...............389
Figure 13.23 Time functions used for the forced convection problem....................389
Figure 13.24 Water pipe subjected to a small periodic excitation force..................391


Appendix-2: List of tables
Appendix-2 List of tables
Table 1-1 Overview of capabilities of ADINA CFD & FSI......................................25

Table 3-1 Functionalities of incompressible, slightly compressible and low-speed
compressible flow models .................................................................................78
Table 3-2 Transformations of temperature units to absolute temperature units ...... 124
Table 4-1 Functionalities in the porous media flow model .....................................158
Table 5-1 Functionalities in K - and K - turbulence models for incompressible,
slightly compressible and low-speed compressible flows ...............................165
Table 6-1 Functionalities in the high-speed compressible flow model ...................180
Table 7-1 Functionalities in the K - turbulence model for high-speed compressible
flows ................................................................................................................224
Table 8-1 Functionalities for mass transfer analyses............................................... 232
Table 9-1 Elements in connections of fluid and solid models in fluid-structure
interaction problems ........................................................................................ 261
Table 13-1 Consistent units derived from the primary units: length, velocity, density
and temperature................................................................................................ 397
Table 13-2 Consistent units derived from the primary units: length, time, force and
temperature ...................................................................................................... 398
Table 13-3 Major files used in ADINA CFD & FSI ............................................... 404

Appendix-2: List of tables

Index
Appendix-3 List of error/warning messages
*** CODE ADF1005:

Too many error/warning messages for code number ADFxxxx have been
encountered.
This message indicates that the message of code number ADFxxxx

has been encountered more than 10 times, where xxxx is a 4-digit
number. Any error/warning message that has been printed 10 times will
no longer be printed. Note that in case the program terminates, the
cause could be related to ADFxxxx not printed. The message ADFxxxx
here represents any message in ADINA-F.
*** CODE ADF1006:

Too many error/warning messages have been encountered.
This message indicates that an error or warning message has been

encountered more than 10 times. Any error/warning message that has
been printed 10 times will no longer be printed. Note that in case the
program terminates, the cause could be related to any error or warning
messages that have been printed more than 10 times.
*** ERROR *** CODE ADF2002:

Model may be unstable! Ratio 9.19E+19 of largest to smallest diagonal
elements in factorized coefficient matrix is greater than 1.10E+19.
When this ratio is too large, machine round-off errors may become severe.
Possible causes are:
* Solution procedure has been divergent;

* Ratio of largest to smallest elements is too large;
* No fixed pressure condition in confined flows;
* Improper units are used in this model. If this is the reason, you
may either change the input to another proper unit system or choose
proper scales for the automatic nondimensionalization.

Appendix-3: List of error/warning messages
This message indicates a large ratio of the factorized diagonal in the

matrix LDU decomposition or incomplete decomposition. This
message is essentially the same as the message of ADF2006, except that
the corresponding solution variables here are unable to be identified.

Error encountered while reading data block (code 1111):
"CURRENT_DATA_ID".
Last input data block is:
"LAST_DATA_ID".
Data could be wrong some where between the two data blocks.
This could mean an internal error or a wrong data file was used.
This message indicates an error encountered in the *.dat file. This

file consists of a few data blocks. Each block has a brief identification
description. CURRENT_DATA_ID is the current block description
while LAST_DATA_ID is the last block description. The error could
be in the current block or the last block. The following data blocks are
sequentially stored in the data file:
MASTER DATA
TIME STEP BLOCK DATA
SAVE AND PRINT BLOCK DATA
TIME FUNTION DATA
ONE-D FUNCTION DATA
BULK DENSITY DEFINITION DATA
SKEW SYSTEM DATA
NODAL DATA
CONSTRAINT DATA
MATERIAL DATA OF ELEMENT GROUP 1
MASS MATERIAL DATA OF ELEMENT GROUP 1
ELEMENT GROUP DATA OF ELEMENT GROUP 1
GEOMETRY AND BCD HEAD DATA
GEOMETRY DATA
BOUNDARY CONDITION HEAD DATA
BOUNDARY CONDITION DATA
HEAD OF BOUNDARY NODE LIST DATA
BOUNDARY NODE LIST DATA
2D NORMAL-TRACTION LOAD DATA
2D HEAT FLUX LOAD DATA
3D NORMAL-TRACTION LOAD DATA
3D HEAT FLUX LOAD DATA
PRESCRIBED LOAD DATA
CONCENTRATE LOAD DATA

Index
2D MASS FLUX LOAD DATA

3D MASS FLUX LOAD DATA
PRESCRIBED MASS LOAD DATA
ROTATIONAL LOAD DATA
MASS INITIAL CONDITION DATA
HEAD OF FLUID INITIAL CONDITION DATA
FLUID INITIAL CONDITION DATA
When this message is encountered, first make sure a correct data file is used.
Refer to the suggestions for message ADF2111.
In case of an error that you may have made during a manual modification of
the data file, the last and the current data block information indicates where the
error could be.
*** WARNING *** CODE ADF2004:
Diagonal element 9.99E-21 of factorized effective stiffness matrix is

smaller than 1.99E-20. A constant pivot 1.00E+20 has been assigned
to equation 430000, which corresponds to PRESSURE at
node (label) 99999.
* Incorrect boundary conditions have been applied;

* Physical model is unstable;
* Solution procedure is divergent.
This message indicates a zero pivot encountered during the LDU

decomposition or incomplete decomposition. The corresponding
solution variable is also identified.
If this message appears in the first iteration in the first time step, a
direct possibility is that the model is ill posed.
If the corresponding solution variable is pressure, make sure that
the pressure degree of freedom is active.

the required fixed pressure condition (e.g., in confined flows) is
applied.
the viscosity is correctly specified.
no velocities are wrongly fixed (e.g. inside the computational
domain).
the transient analysis or a proper CFL number has been used in
Euler high-speed flows.
a proper unit system is used.

If the corresponding solution variable is velocity, make sure that
at least one wall or its analogue is applied.

the viscosity is correctly specified.
a proper unit system is used.
If the corresponding solution variable is temperature, make sure that
at least one prescribed temperature condition or its analogue

(convection, radiation, etc.) is applied if a steady-state analysis is
performed. If the only analogue of prescribed condition is a
convection boundary condition, the convection coefficient h (t )
must be nonzero. The associated time function must be checked
as well. Similarly, if the only analogue of prescribed condition
is a radiation boundary condition, the Stefan-Boltzmann constant
, the shape factor f and the emissivity e must be nonzero.
the heat conductivity is correctly specified.
If the corresponding solution variable is a turbulent variable, make

sure that
at least one prescribed condition has been applied to that

variable. A wall condition is considered as a prescribed
condition as well.
the laminar viscosity is correctly specified.
If the corresponding solution variable is related to moving mesh, the

equation number is the same as the nodal number. Make sure that a
valid mesh has been used, particularly near the node where the
singularity appears.
If this message appears after a few iterations, the iteration may be

divergent. It could be an indication that the model is improperly
defined, load increments are too large, the physical phenomenon is
unstable, etc. Refer to the topic Model preparation and testing for
helpful tips.

Index
*** ERROR/WARNING *** CODE ADF2005:

Iteration limit is reached without convergence.
Equilibrium is not established from time: 1.23450E-21
to time: 2.23449E-21
Here are some suggestions for improving convergence:
* Resolve all error/warning messages.

* Use proper units to avoid machine round-off errors.
* Apply a proper force or displacement relaxation factor in
fluid-structure interaction analyses.
* Use a proper CFL number in automatic time stepping method.
* In transient analyses, use smaller time steps.
* In steady-state analyses, incrementally apply loads (prescribed
velocities, normal-tractions, etc.) step-by-step, or/and use
time-dependent materials, starting with easier material values and
eventually approaching to the real values.
* In transient analyses of fluid-structure interactions, a good initial
condition of fluid variable is crucial. One can achieve that by
using a very large fictitious first time step, which provides a good
initial condition for the really transient analyses starting at
second time step. One can also achieve that by using a restart run,
in which the (steady-state) result of first run provides initial
conditions for the (transient) analysis in the second run.
This message indicates unsuccessful equilibrium iterations in the

current time step. If the automatic time step (ATS option) is selected,
the program will continue using a smaller time step.
Many possibilities could cause the divergence. Besides the
suggestions shown in this message, the topic Model preparation and
testing contains more detail and helpful tips. If the iteration residuals
show a converging tendency, you may increase the number of iterations
that is permitted to reach a converged solution.

Model may be unstable! Ratio 1.2339E-21 of largest to smallest diagonal
elements in factorized coefficient matrix is greater than 1.E+11. The
largest diagonal element -2.2339E-21 corresponds to
Y-VELOCITY at node (label) 999999 and the smallest
diagonal element 3.23449E-21 corresponds to PRESSURE
at node (label) 88888888.
When this ratio is too large, machine round-off errors may become so
severe that stable models become unstable.

* Solution procedure has been divergent;

* Ratio of largest to smallest elements is too large;
* No fixed pressure condition in enclosures;
* Improper units are used in this model. If this is the reason, you
may either change the input data in another unit system or use the
automatically nondimensional procedure by choosing proper scales.
This message indicates a large ratio of the factorized diagonal in the

matrix LDU decomposition or incomplete decomposition. The
corresponding solution variables, nodal numbers and equation number
are also identified. A large ratio usually indicates a poor conditioning of
the matrix. If the minimum factorized diagonal is zero or extremely
small (say smaller than 1020 ), this message is essentially the same as
the message of ADF2004. In this case, check the explanations for that
message.
Certain factors can contribute to the ill conditioning:
the problem is ill posed.

an improper unit system is used.
element sizes vary too much.
highly nonlinear problems (non-Newtonian fluid flows, high-
Reynolds/Peclet/Rayleigh number flows, etc.).
Refer to the topic Model preparation and testing for helpful tips to
exclude the possibility of ill posed problem. Usually, the matrix
conditioning can be improved by choosing a proper unit system. It can
also be done in the ADINA-F automatically nondimensional procedure
by providing proper scales and reference datums. However, since the
problem is nonlinear and the element sizes are likely non-uniform, this
improvement has its limit. In these cases, the warning message should
be ignored.

Turbulence model is incompatible with the material model.
Turbulence K-Epsilon material model can only be used for Turbulence
K-Epsilon fluid model.

Index
The message indicates that a turbulence K - material is used in an

element group while the fluid model is not turbulence K - . Refer to
the related chapters to see available materials.

Turbulence model is incompatible with the material model.
Turbulence K-Omega material model can only be used for Turbulence
K-Omega fluid model.
The message indicates that a turbulence K - material model is used

in an element group while the fluid model is not turbulence K - . Refer
to the related chapters to see the available materials.

Incorrect ADINA-F data file. This could occur if:
* The file is not an ADINA-F data file;

* The version of the data file does not match the program version;
* There is no permission to open the data file.
The message indicates that data file for the fluid model is not
recognizable.
When this message is encountered, first make sure a correct data file
is used. Make sure the version of AUI used to generate the data file is
consistent with the version of ADINA-F. It is best to use the latest
version. If the *.dat file was generated using an earlier version of AUI,
regenerate it using the consistent version.
If the data file was generated using another machine, make sure the
data is portable. In general, use a formatted (ASCII) data file if the AUI
and ADINA-F are installed in different machines, although most Unix
machines can use binary data files interchangeably.
If you have modified the *.dat file, make sure that the format in the
modification is correct. It is not recommended (although not restricted)
to edit some parameters in the data file.
Check also if the file has been protected so that it is not readable.

Incorrect ADINA data file. This could occur if:

* The file is not an ADINA data file;

* The version of the data file does not match the program version;
* There is no permission to open the data file.
This message is essentially the same as message of ADF2010,

except that the data file corresponds to the solid model. Only in
formulation of fluid-structure interactions, a solid data file is required.
In addition to the explanations made for the message ADF2010, the
consistency between the versions of ADINA and ADINA-F must also
be checked.

File name LONG_FILE_NAME is too long.
This message indicates that a file name is too long (more than 128
characters). Avoid using long names for your files.

Coarsening in multi-grid ended at level 11 since all inter-block links
are too weak.
This message indicates that a poor condition has been encountered

in multi-grid solver. Change the solver to the sparse solver or other
iterative solvers.

Multi-grid levels have been reduced to 11 due to memory limit.
This message indicates that multi-grid solver could not be working

properly because of the memory limitation. Change the solver to the
sparse solver or other iterative solvers.

Failed in multi-grid solver: status = "status_WHEN_ERROR_OCCURS".
This message indicates that multi-grid solver is unable to converge.

Change the solver to the sparse solver or other iterative solvers.

Index

No enough memory to continue. Currently available status is:
"status_WHEN_ERROR_OCCURS".
This message indicates that the memory assigned to ADINA-F is not

sufficient for the multi-grid solver. Usually other solvers may not work
as well because the memory limitation. More memory is required.

Incorrect data passed out from the user-supplied boundary condition
routine. B(I,VARIABLE_NAME) must be zero since there is no degree of
freedom for that variable.
This message indicates that an error has been found in the output of
the user-supplied boundary condition subroutine. Refer to the
description on how to use user-supplied boundary condition.
A user-supplied boundary condition is defined as a source term that
is added to the right hand side of the equation. This source is expressed
as A + Bii , where i are all possible solution variables. A nonzero Bi
must corresponds to an active variable i .

Incorrect data passed out from the user-supplied boundary condition
routine. B(I,VARIABLE_NAME) = 1.230E+19 is too large.
This message indicates that a possible error has been found in the
output of the user-supplied boundary condition subroutine. Refer to the
as A + Bii , where i are all possible solution variables. A very
large Bi usually indicates a non-initialized number.

Improper data passed out from the user-supplied boundary condition

routine. B(22) = 1.230E+19 may cause instability. Negative value

should be assigned.
This message indicates that a possible error has been found in the
output of the user-supplied boundary condition subroutine. Refer to the
as A + Bii , where i are all possible solution variables. A positive Bi
could cause numerical instability. Rewrite the source term such that the
coefficient is non-positive. For example, A = A + (1 + ) Bii and
Bi = Bi , where is a non-negative constant.

Unable to apply temperature-dependent-power-law material to the model
since there is no temperature degree of freedom.
This message indicates that the temperature degree of freedom is

inactive. If you intend to model the temperature, turn on the
temperature degree of freedom together with all properly defined
boundary conditions as well. If the temperature is not of interest, use
another material model. Power-law material model could be the one
that is closest to your original choice.

Parameter A 1.19E-21 in the temperature-dependent power-law material
model is invalid.
This message indicates that the value of parameter A is invalid in a

temperature-dependent power-law material model. It is usually out of
the range [1020 ,1020 ] . Check the definition of the material model for
more details.

Parameter N 1.19E-21 in the temperature-dependent power-law material
model is invalid.

Index
This message indicates that the value of parameter n is invalid in a

more details.

Parameter C 1.19E-21 in the temperature-dependent power-law material
model is invalid.
This message indicates that the value of parameter C is invalid in a

more details.

Unable to apply the second-order material to the model since there is no
temperature degree of freedom.
This message indicates that the temperature degree of freedom is

inactive. If you intend to model the temperature, turn on the
temperature degree of freedom together with all properly defined
boundary conditions. If the temperature is not of interest, use another
material model. The power-law material model could be the one that is
closest to your original choice.

Parameter A 1.19E-21 in the second-order material model is invalid.

second-order material model. It is usually out of the range [1020 ,1020 ] .
Check the definition of the material model for more details.

Parameter C 1.19E-21 in the second-order material model is invalid.
This message indicates that the value of parameter C is invalid in a

second-order material model. It is usually out of the range

[1020 ,1020 ] . Check the definition of the material model for more
details.

Parameter E 1.19E-21 in the second-order material model is invalid.
This message indicates that the value of parameter E is invalid in a

details.

Parameter F 1.19E-21 in the second-order material model is invalid.
This message indicates that the value of parameter F is invalid in a

details.

Parameter G 1.19E-21 in the second-order material model is invalid.
This message indicates that the value of parameter G is invalid in a

details.

Parameter H 1.19E-21 in the second-order material model is invalid.
This message indicates that the value of parameter H is invalid in a

details.

Index

Unable to apply the turbulence material to the model since there is no
turbulence degree of freedom.
This message indicates that the fluid model selected is not a two-
equation turbulence model ( K - or K - ) while a turbulence material
is used. Refer to the related chapters to see the available materials.

Phase change requires temperature being presented. There is no
temperature degree of freedom in this model.
This message indicates that the temperature degree of freedom is not

active in this model. Whenever phase change is required or phase
change boundary condition is applied, the temperature degree of
freedom must be present.

Parameter BetaT 1.19E-21 in the K-Omega material model is invalid.
This message indicates that the value of parameter is invalid in a

turbulent K - material model. It is usually out of the range
details.

Points are out of order in a curve. Program is unable to identify the type
of the curve. Currently available status is:
This message indicates that a material curve is invalid because some

points that define the curve are out of order. Tabulated material curves
must be input in an increasing order with no repeated points. The status
could be either temperature-dependent material curves or time-
dependent material curves.


Viscosity (XMU) 1.2345599E+20 is invalid.
This message indicates an invalid value of the fluid viscosity . It

is usually out of the range [1020 ,1020 ] . Check the definition of the
material model for more details.

Density (RHO) 1.2345599E+20 is invalid.
This message indicates an invalid value of the fluid or solid density

. It is usually out of the range [1020 ,1020 ] . Check the definition of
the material model for more details.

Specific heat at constant pressure (Cp) 1.2345599E+20 is invalid.
This message indicates an invalid value of the specific heat at

constant pressure C p . It is usually out of the range [1020 ,1020 ] . Check
the definition of the material model for more details.

Specific heat at constant volume (Cv) 1.2345599E+20 is invalid.
This message indicates an invalid value of the specific heat at

constant volume Cv . It is usually out of the range [1020 ,1020 ] . Check

Specific heat in solid element group (Cps) 1.2345599E+20 is invalid.
This message indicates an invalid value of the solid specific heat at

constant pressure C p s that is required in a porous medium material. It is

Index
usually out of the range [1020 ,1020 ] . Check the definition of the

Density in solid element group (ROHs) 1.2345599E+20 is invalid.
This message indicates an invalid value of the solid density s that

is required in a porous medium material. It is usually out of the range
details.

Heat conductivity (XKCON) 1.2345599E+20 is invalid.
This message indicates an invalid value of the heat conductivity k .

It is usually out of the range [1020 ,1020 ] . Check the definition of the

Porosity (PHI) 1.2345599E+20 is invalid.
This message indicates an invalid value of the porosity that is

required in a porous medium material. It is usually out of the range
[0,1] . Check the definition of the material model for more details.

Permeability in x-direction (PERMx) 1.2345599E+20 is invalid.
This message indicates an invalid value of the x -permeability 1

that is required in a porous medium material. It is usually out of the
range [1020 ,1020 ] . Check the definition of the material model for more
details.

Permeability in y-direction (PERMy) 1.2345599E+20 is invalid.
This message indicates an invalid value of the y -permeability 2

details.

Permeability in z-direction (PERMz) 1.2345599E+20 is invalid.
This message indicates an invalid value of the z -permeability 3

details.

Heat conductivity in solid element group (XKCONs) 1.2300E+20 is invalid.
This message indicates an invalid value of the solid heat

conductivity k s that is required in a porous medium material. It is

Parameter (A) 1.2345599E+20 in the power-law material model is invalid.

power-law material model. It is usually out of the range [1020 ,1020 ] .

Parameter (N) 1.2345599E+20 in the power-law material model is invalid.

Index
power-law material model. It is usually out of the range [1020 ,1020 ] .


The variable ID or/and equation ID have been replaced in the
user-supplied-boundary-condition subroutine. An error may exist
in the subroutine. The two IDs are stored in L(2) and L(3) respectively.
Current values of them are 11 and 22.
This message indicates that an error has been found in the output of
the user-supplied boundary condition subroutine. The two identification
numbers, I v and I c stored at L(2) and L(3) respectively, are read
only parameters in the user routine. However, their addresses have
been found used after the function call. Check the program to make
sure the passed-out parameters are saved at the correct locations in the
passed-out arrays.

Currently required element viscosity is not available. A wrong element
may have been used. Only triangular/tetrahedron elements can be used
if adaptive mesh option is applied.
This message indicates that possibly an unavailable element is used

while the mesh repair option is selected. Currently, only triangular 3-
node and tetrahedral 4-node elements are available if the mesh repair
option is selected.

Parameter (A) 1.2345599E+20 in the CARREAU material model is invalid.

Carreau material model. It is usually out of the range [1020 ,1020 ] .

Parameter (N) 1.2345599E+20 in the CARREAU material model is invalid.


Carreau material model. It is usually out of the range [1020 ,1020 ] .

X-gravitation (GRAVx) 1.2345599E+20 is invalid.
This message indicates an invalid value of the x -gravity g1 . It is


Y-gravitation (GRAVy) 1.2345599E+20 is invalid.
This message indicates an invalid value of the y -gravity g 2 . It is


Z-gravitation (GRAVz) 1.2345599E+20 is invalid.
This message indicates an invalid value of the z -gravity g3 . It is


Surface tension (SIGMA) 1.2345599E+20 is invalid.
This message indicates an invalid value of the surface tension . It

is usually out of the range [1020 ,1020 ] . Check the definition of the

Index
Reference temperature (Tc) 1.2345599E+20 is invalid.
This message indicates an invalid value of the reference temperature

0 in material data. It is usually out of the range [1020 ,1020 ] . Check

Rate of heat generated per unit volume (Qb) 1.2345599E+20 is invalid.
This message indicates an invalid value of the heat generation rate

q in material data. It is usually out of the range [1020 ,1020 ] . Check
B

Coefficient of volume expansion (BETA) 1.2345599E+20 is invalid.
This message indicates an invalid value of the coefficient of volume

expansion in material data. It is usually out of the range
details.

Parameter (C1) 1.2345599E+20 in the turbulence model is invalid.
This message indicates an invalid value of the model constant c1 in

turbulent material data. It is usually out of the range [1020 ,1020 ] .






Parameter (CMU) 1.2345599E+20 in the turbulence model is invalid.
This message indicates an invalid value of the model constant c in


Parameter (SIGMAK) 1.2345599E+20 in the turbulence model is invalid.
This message indicates an invalid value of the model constant K in


Parameter (SIGMAE) 1.2345599E+20 in the turbulence model is invalid.
This message indicates an invalid value of the model constant in


Parameter (SIGMAT) 1.2345599E+20 in the turbulence model is invalid.

Index
This message indicates an invalid value of the model constant in


Von Kaman constant (VON) 1.2345599E+20 in the turbulence model is invalid.
This message indicates an invalid value of the Von Karman constant

0 in turbulent material data. It is usually out of the range [1020 ,1020 ] .

Parameter (DW) 1.2345599E+20 in the turbulence model is invalid.
This message indicates an invalid value of the model constant d w+ in


Bulk modulus of elasticity (KAPA) 1.23400E+20 is invalid.
This message indicates an invalid value of the bulk modulus used

in a slightly compressible flow model. It is usually out of the range
details.

Ratio of second viscosity (XMU2) 1.23400E+20 is invalid.
This message indicates a possibly invalid value of the rate of the

second viscosity in turbulent material data. It is usually out of the
range [10,1] . Check the definition of the material model for more
details.


Exponent for viscosity (SEXMU) 1.23400E+20 is invalid.
This message indicates a possibly invalid value of the exponent m

in the Sutherlands formula used for high-speed compressible flows. It
is usually out of the range [10,10] . Check the definition of the

Parameter (S1XMU) -1.23400E+20 is invalid.
This message indicates a possibly invalid value of the constant

temperature S in Sutherlands formula used for high-speed
compressible flows. It is usually out of the range [1020 , ) . Check

Exponent for heat conductivity (SEXKC) 1.23400E+20 is invalid.
This message indicates a possibly invalid value of the exponent mk

in Sutherlands formula used for high-speed compressible flows. It is
usually out of the range [10,10] . Check the definition of the material
model for more details.

Parameter (S1XKC) -1.23400E+20 is invalid.
This message indicates a possibly invalid value of the constant

temperature S k in Sutherlands formula used for high-speed
compressible flows. It is usually out of the range [1020 , ) . Check

Index
Length scale 1.23400E+20 in nondimensionalization is invalid.
This message indicates an invalid value of the length scale L* in

automatically nondimensional procedure. It is usually out of the range
[1020 ,10+20 ] . Refer to the descriptions about nondimensional analyses.

Velocity scale 1.23400E+20 in nondimensionalization is invalid.
This message indicates an invalid value of the velocity scale V * in

an automatically nondimensional procedure. It is usually out of the
range [1020 ,10+20 ] . Refer to the descriptions about nondimensional
analyses.

Scale 1.2300E+20 for specific heats in nondimensionalization is invalid.
This message indicates an invalid value of the specific heat scale C *

in an automatically nondimensional procedure. It is usually out of the
analyses.

Density scale 1.23400E+20 in nondimensionalization is invalid.
This message indicates an invalid value of the density scale D* in

an automatically nondimensional procedure. It is usually out of the
analyses.

Temperature scale 1.23400E+20 in nondimensionalization is invalid.
This message indicates an invalid value of the temperature scale T *



analyses.

Mass-ratio scale 1.23400E+20 in nondimensionalization is invalid.
This message indicates an invalid value of the mass-ratio scale *

analyses.

X-coordinate datum 1.23400E+20 in nondimensionalization is invalid.
This message indicates an invalid value of the reference

x-coordinate xr in an automatically nondimensional procedure. It is
usually out of the range [1012 ,1012 ] . Refer to the descriptions about
nondimensional analyses.

Y-coordinate datum 1.23400E+20 in nondimensionalization is invalid.

y -coordinate yr in an automatically nondimensional procedure. It is

Z-coordinate datum 1.23400E+20 in nondimensionalization is invalid.

z -coordinate zr in an automatically nondimensional procedure. It is

Index

Temperature datum 1.23400E+20 in nondimensionalization is invalid.
This message indicates an invalid value of the reference temperature

in an automatically nondimensional procedure. It is usually out of

r
the range [1012 ,1012 ] . Refer to the descriptions about nondimensional

analyses.

Control flag (IFGNDM) -1 in nondimensionalization is invalid.
This message indicates an invalid control flag in an automatically

nondimensional procedure. A wrong data file may cause this error.
Whenever a possibly wrong data file is encountered, make sure the
following requirements are satisfied.
A correct data file is used.
The versions of AUI and ADINA-F are consistent. It is
suggested that always use the latest consistent versions.
If the data file was generated on a different machine where
ADINA-F is installed, the data file must be portable. In general,
use a formatted (ASCII) data file. Usually, binary data files are
compatible between Unix machines. They are not compatible
with Windows PCs and supercomputers.
If you have modified the *.dat file, make sure that the format is
correct. It is not recommended (although not restricted) to edit
parameters in the data file.

Analysis flag (IFGNDI) -1 in nondimensionalization is invalid.



Output flag (IFGNDO) -1 in nondimensionalization is invalid.


Starting time (TSTART) 1.23400E+20 is invalid.
This message indicates an invalid value of the restart time. It is

usually out of the range [1020 , ) . Check the latest time that the
solution is converged in the previous run. The restart time specified in
the restart run must be the same as that saved in the *.res file. Refer to
the descriptions about restart analyses.

Number of nodes (NNODE) -1 is invalid.
This message indicates an invalid number that represents the number

of nodes. It is usually a non-positive number. A wrong data file may
cause this error. Refer to the suggestions for message ADF2111.

Execution flag (MODEX) 1234 is invalid.
This message indicates an invalid control flag. A wrong data file

may cause this error. Refer to the suggestions for message ADF2111.

Number of element groups (NELMG) -1 is invalid.

of element groups. A wrong data file may cause this error. Refer to the
suggestions for message ADF2111.

Index

Number of time step blocks (NDTBK) -1 is invalid.

of time step blocks. A wrong data file may cause this error. Refer to
the suggestions for message ADF2111.

Number of skew systems (NSKEW) -1 is invalid.

of skew systems defined in this model. A wrong data file may cause
this error. Refer to the suggestions for message ADF2111.

Rotational flag (IROTA) 1234 is invalid.
This message indicates an invalid flag for centrifugal load data. A

wrong data file may cause this error. Refer to the suggestions for
message ADF2111.

Flag (ICOMP) 1234 for fluid flow type is invalid.
This message indicates an invalid flag that represents the flow model
selected. A wrong data file may cause this error. Refer to the

Turbulence model type (IFGTUR) 1234 is invalid.
This message indicates an invalid flag that represents the selected

turbulence model type. A wrong data file may cause this error. Refer to


Flag (IHYDRO) 1234 for including hydrostatic pressure is invalid.
This message indicates an invalid flag that represents the treatment

of the hydrostatic pressure. A wrong data file may cause this error.

Flag (METHOD) 1234 for equilibrium method is invalid.
This message indicates an invalid flag that represents the selected

equilibrium iteration method. A wrong data file may cause this error.

Maximum number of equilibrium iterations (ITEQLM) -1 is invalid.
This message indicates an invalid number that represents the

specified maximum number of iterations that are permitted in
equilibrium iterations. A wrong data file may cause this error. Refer to

Number of functions (NFN) -1 is invalid.

of one-dimensional functions defined in this model. A wrong data file

Maximum number of function points (NFP) -1 is invalid.
This message indicates an invalid number that represents the

maximum number of points defined in one-dimensional functions in this
model. A wrong data file may cause this error. Refer to the suggestions
for message ADF2111.

Index

Pressure points are too close in material data. Distance = 2.19E-21
Check the input related to material data.
This message indicates that a pressure-dependent material curve is

invalid because some pressure points that define the curve are out of
order. Tabulated material curves must be input in an increasing order
and no repeated points.

Temperature points are too close in material data. Distance = 2.19E-21
Check the input related to material data.
This message indicates that a temperature material curve is invalid

because some temperature points that define the curve are out of order.
Tabulated material curves must be input in an increasing order and no
repeated points.

78 errors have been found in NPAR parameters in element group data.
This message indicates that a few errors have been identified in the
element group control data. Check the earlier printed error/warning
messages. In case a wrong data file is used, also refer to the suggestions

Element birth time is larger than element death time:
element (label) = 1;
element group number = 2;
birth time = 3.0009E-02;
death time = 2.0020E-02.
This message reports an error found in the element group data. An

element birth time has been found larger than the element death time. It
is required that birth time is always smaller than the death time. It is
also possible that a wrong data file is used. In this case, also refer to the


Element number -1 in element group 2 is invalid.
Data file could be wrong.
This message reports an error found in the element group data. A

non-positive element label has been encountered. A wrong data file

Nodal label 1 in element connectivity data is invalid:
element group number = 2.
This message reports an error found in the element group data. A

node that forms an element cannot be recognized. A wrong data file

Inconsistency in element group control data (NPAR(6)) and element
connectivity data. In element 890, element group 213,
the node 11 is defined in a skew system, which was not shown
in control data.
This message reports inconsistent information shown in the element

group data and in global control data. The information is related to the
skew systems defined in this model. A wrong data file may cause this
error. Refer to the suggestions for message ADF2111.

Material data set number -1 is invalid in element 890,
element group 112.
This message reports an error found in the element group data. An

element group can have only one fluid material data set. However, the

Index
number shown here is inconsistent with the assumption. A wrong data

file may cause this error. Refer to the suggestions for message
ADF2111.

Model may be unstable because temperature or pressure is too small
(1.19E-21). Local temperature or pressure has been adjusted by applying
correction factor 1.230E-01 to node 890.
This message reports a possible error in the model. Whenever a

compressible flow model (low-speed or high-speed) is used, the
pressure and temperature must not be zero anywhere in the model.
Refer to the formulations for compressible flow models for more details
on the assumptions and limitations.
Recall that the state equation used in compressible flows is
p = (C p Cv ) . The pressure must be absolute pressure and the
temperature must be absolute temperature. The state equation must be
satisfied through all the input data, including materials, loads, boundary
conditions, initial conditions, etc.
If this message is encountered at the first iteration in the first time
step, usually some wrong data has been input. You may perform a test
by imposing conditions such that the solution is expected to be very
close to the initial conditions. In the same test model, specify a large
value for equilibrium iteration tolerance so the solution will be saved
after the first iteration. With the help of the printed information in the
*.out file and the help of visualization of the saved solution, find the
locations where the solutions are unreasonable.
If the pressure and the temperature are not too small, an improper
unit system is also possible. In this case, either use another proper
system of units or use the automatically nondimensional procedure by
specifying proper scales. Refer to the related topics for more details.

Incorrect initial conditions have been found at node 890. This may
occur if pressure and/or temperature are too small or Mach number is
too large.
Check the input related to initial conditions and material data.
Currently available information is:

pressure = 1.23450E-21;
x-velocity = 2.23449E-21;
y-velocity = 3.23449E-21;
z-velocity = 4.23449E-21;
temperature = 5.23449E-21;
Cp = 6.23449E-21;
Cv = 7.23449E-21.
This message reports a possible error in the initial condition.

Possibly a zero pressure or a zero temperature has been specified in
compressible flow models. Refer to the suggestions for message
ADF2224.

Incorrect constraint equation data: Constraint equation 890 is at
position 11112.
This message reports an error found in the constraint equation data.

A wrong data file may cause this error. Refer to the suggestions for
message ADF2111.

Incorrect constraint equation data: too many master nodes 2222 in
constraint equation 890. The largest possible number is 2333.

message ADF2111.

Inconsistency between constraint equation data and nodal data: the slave
node 11112 defined in the constraint equation 890 is unable to be
identified as a slave node based on nodal information.


Index
message ADF2111.

Inconsistency between constraint equation data and nodal data: a master
node defined in constraint equation 890 has no degree of freedom.

message ADF2111.

A node defined in constraint equation 890 is out of label range.

message ADF2111.

A variable identification direction in constraint equation 890 is
out of label range.

message ADF2111.
Variable identification directions in master nodes and slave nodes are not
identical in constraint equation 890.

message ADF2111.

In high speed compressible flows, all degrees of freedom of a slave node

must be constrained to the same degrees of freedom of its master node.
Error has been found in the following data set:
constraint equation = 11;
slave nodal label = 22;
variable identification direction = 33.
Check the input related to constraint equations.

In high-speed compressible flows, because of the highly nonlinear
property of the solution variables, ADINA-F does not accept that a
solution variable is constrained to a different solution variable. It is also
improper if only a part of solution variable is constrained. A wrong data
file may cause this error as well. In this case, refer to the suggestions

Incorrect combination of variables "S_V" in boundary condition 119.
This message reports an error found in a special boundary condition

in the high-speed compressible flow model. The combination of the
variables is not available. A wrong data file may cause this error. Refer
to the suggestions for message ADF2111.

Unreasonable solutions obtained at boundary node 1118 (index) in
boundary condition 19 where the combination of variables is "S_V".
Values of boundary variables may be modified in case it is necessary.
The values of inside primitive variables are:
p = 1.1234559E+11
u = 1.1234559E+11
v = 1.1234559E+11
w = 1.1234559E+11
t = 1.1234559E+11
The computed values of outside primitive variables are:
p = 1.1234559E+11
u = 1.1234559E+11
v = 1.1234559E+11

Index
w = 1.1234559E+11
t = 1.1234559E+11
Current values of Cp and Cv are:
Cp = 3.1277770E+11
Cv = 3.2277771E+11
Current boundary variables are:
3.32769E+11, 3.42769E+11, 3.52769E+11.
The above values may be in dimensionless form if automatic

nondimensionalization is used.
This message reports an unreasonable solution obtained in a special

boundary condition in the high-speed compressible flow model.
When this message is encountered at the first iteration in the first
time step, make sure the following conditions are satisfied:
The solution variables specified in boundary conditions satisfy

the state equation. In particular, the resultant Mach number is
correct.
The initial conditions are correctly specified. In particular, no

zero pressure or temperature are specified.
The material data are correctly input, particularly the C p and

Cv .
When this message is encountered in later iterations or in later time

steps, the solution procedure is probably divergent. Check the latest
converged solution to find any possible errors. If no solutions are saved
yet, use a large equilibrium tolerance to save the solution after the first
iteration. Refer to the suggestions described under the topic model
preparation and testing for more details.

Pivot has been assigned to a boundary equation, where the boundary
variable combination is "S_V" in boundary condition (index) 123456.

This message reports an ill conditioned matrix in a special boundary

condition in the high-speed compressible flow model. The possible
causes of this error are likely the same as those described in message
ADF2249.

Unidentified flow regime 1234 in boundary condition 222222.

in the high-speed compressible flow model. A wrong data file may

Unidentified variable ID 111 in boundary condition 1119.


Incorrect combination of variables in boundary condition 11. Their
identification flags are 22, 33 and 44.


Failed in solving boundary equations in boundary condition 11.


Index
This message reports a singularity found in a special boundary

ADF2249.

The boundary equation is singular in boundary condition 11.


ADF2249.

Failed in solving the fluid state equation.

D = 1.120E+11;
DI = 1.219E+11;
P = 1.319E+11;
T = 1.419E+11;
CP = 1.520E+11;
CPP = 2.119E+11;
CPT = 2.220E+11;
CV = 2.319E+11;
CVP = 2.420E+11;
CVT = 2.519E+11.
The above values may be in dimensionless form if automatic

nondimensionalization is used.


ADF2249.

Distributed loads (normal-traction, heat-flux, etc.) cannot be applied to
internal faces. Check the load applied to the face that is formed by the
Nodes: (1,8,3,78).
This message reports a load that has been applied to an internal face.
Distributed loads, such as normal-tractions, heat-fluxes, etc., should
never be applied to internal faces. This could be a input error in the
model.

The turbulence K-Omega model is currently not available for high speed
compressible flows. Turbulence K-Epsilon is available.
This message indicates that the turbulent K - is currently not

available in the high-speed compressible flow model. Only the
turbulent K - can be used for the high-speed compressible flows. You
may change the turbulence model type and rerun the model.

Force is not available at this stage.
This could be an internal error or the current working environment has been
accidentally broken.
This message indicates that the program is unable to compute the

fluid force or stress because of an unknown reason. One possible reason
is that some temporary files have been removed (not by the program).
Another possibility is that a wrong data file is used. In this case, refer to

The matrix size is too large. Current 32-bit program versions can only

Index
handle problems with less than 2147483647 matrix elements.
This message indicates that the size of the matrix surpasses the limit
+31
2 on a 32-bit machine. If the solver selected is the Gauss elimination
method (SKYLINE), choose the sparse or iterative solvers instead. The
iterative solver RPBCG uses the least memory.

Number of points 1119 in boundary geometry 56 is invalid.
This message reports an error found in the boundary geometry data.

message ADF2111.

Number of cells 1119 in boundary geometry 56 is invalid.

message ADF2111.

Time value in temperature file is out of order.
The file may not be generated by ADINA program or a wrong temperature file
was used.
This message reports an error found in the temperature file. Do you

really want to apply the temperature saved in the temperature file?
Probably not but have accidentally selected that choice. However, if
you want, the temperature file must be generated by ADINA program.
The temperature solutions are saved on nodes in every time step. The
time range saved must cover the time range in the current model. The
nodes saved must also be the same as those in the current model.
A wrong data file may cause this error as well. In this case, refer to


Fluid node (label) 99999999 is not attached to solid fluid-structure
interface 22. The relative distance to solid boundary element
33 is 9.99E-22.
The fluid nodal coordinates are (1.00E+01, 2.00E+01, 3.00E+01).
The points and their coordinates of the solid boundary element are:
point 444444 = ( 1.400E+01, 2.400E+01, 3.400E+01)
point 555555 = ( 1.500E+01, 2.500E+01, 3.500E+01)
point 666666 = ( 1.600E+01, 2.600E+01, 3.600E+01)
* Fluid-structure interfaces in fluid and solid models are not

coincident;
* Solid boundary elements are too coarse compared with fluid boundary
elements on the same fluid-structure interface.
Recall that fluid-structure interfaces are specified in both fluid and

solid models. They must be coincident with each other. However, it is
normal to have a small distance between fluid nodes and solid
fluid-structure interfaces if meshes of interfaces in both models are
not matched.
The relative distance here is defined as the ratio of the distance between
the fluid node and the closest solid boundary element to the length of the
solid boundary element.
This message reports a distance, in three dimensions, that exists

between two fluid-structure interfaces that are defined in the fluid model
and the solid model respectively. Recall that the two interfaces
specified in the fluid and the solid model form a pair to define the
interaction; they represent the same geometric surface.
Since the two models are allowed to be generated separately, the
meshes on the interfaces are likely to be unmatched. Reasonably small
gaps are therefore expected. However, when a relative distance is too
large, the program treat it as an error. This situation may occur if two
discrepant geometries are used in fluid and solid models. It may also
occur if the sizes of the two meshes are too deviated, particularly along
boundary locations of sharp curvatures. Excessively fine details on the
boundary geometry may also cause this problem if the fluid and solid
nodes are unmatched. Refer to the related chapters for more details.

Index
Fluid node (label) 99999999 is not attached to solid fluid-structure

interface 22. The relative distance to solid boundary element
33 is 9.99E-22.
The fluid nodal coordinates are (1.00E+01, 2.00E+01, 3.00E+01)
The points and their coordinates of the solid boundary element are:
point 444444 = ( 1.400E+01, 2.400E+01, 3.400E+01)
point 555555 = ( 1.500E+01, 2.500E+01, 3.500E+01)
* Fluid-structure interfaces in fluid and solid models are not

coincident;
* Solid boundary elements are too coarse compared with fluid boundary
elements on the same fluid-structure interface.
Recall that fluid-structure interfaces are specified in both fluid and

solid models. They must be coincident with each other. However, it is
normal to have a small distance between fluid nodes and solid
fluid-structure interfaces if meshes of interfaces in both models are
not matched.
The relative distance here is defined as the ratio of the distance between
the fluid node and the closest solid boundary element to the length of the
solid boundary element.
This message reports a distance, in two dimensions, that exists

between two fluid-structure interfaces that are defined in the fluid model
and the solid model respectively. Recall that the two interfaces
specified in the fluid and the solid model form a pair to define the
interaction; they represent the same geometric surface.
This problem is the same as described for message ADF2701.

Fluid node (label) 999999 is not attached to boundary geometry 22.
The relative distance to boundary cell 33 is 0.10E-20.
The points and their coordinates of the boundary cell are:
point 444444 = ( 1.400E+01, 2.400E+01, 3.400E+01)
point 555555 = ( 1.500E+01, 2.500E+01, 3.500E+01)
point 666666 = ( 1.600E+01, 2.600E+01, 3.600E+01)
It is normal to have a small distance between fluid nodes and boundary

geometry. The relative distance here is defined as the ratio of the
distance between the fluid node and the closest boundary cell to the
length of the cell.

Note: The boundary geometry is internally generated by the program.
This message reports an error found in the three-dimensional

boundary geometry data. A wrong data file may cause this error. Refer

Fluid node (label) 999999 is not attached to boundary geometry 22.
The relative distance to boundary cell 33 is 9.99E-22.
The points and their coordinates of the boundary cell are:
point 444444 = ( 1.400E+01, 2.400E+01, 3.400E+01)
point 555555 = ( 1.500E+01, 2.500E+01, 3.500E+01)
It is normal to have a small distance between fluid nodes and boundary

geometry. The relative distance here is defined as the ratio of the
distance between the fluid node and the closest boundary cell to the
length of the cell.
Note: The boundary geometry is internally generated by the program.
This message reports an error found in the two-dimensional

boundary geometry data. A wrong data file may cause this error. Refer

Dimension 111 in boundary geometry 56 is invalid.

message ADF2111.

Identification type 111 in boundary geometry 56 is invalid.

message ADF2111.

Index

Time function number 111 in boundary geometry 56 is invalid.
time functions are applied.

message ADF2111.

Birth time -1.0E-12 or/and death time -2.1E-12 in boundary geometry
56 are invalid.

message ADF2111.

Neighbor cells of point 13456 are not connected properly in boundary
geometry 156. 22 neighbor cells are found, but only 11
of them are connected.
This could be an internal error or a wrong data file was used.

message ADF2111.

Point number 55 in boundary geometry 22 is out of range
[1,11]. Current point index is 44 in cell 33.

message ADF2111.

Incorrect data was found in boundary condition 21: invalid combination

of the following parameters:
identification type = 21;
moving type = 31;
geometry type = 41;
boundary geometry (label) = 51.
This message reports an error found in boundary condition data. A

message ADF2111.

Cell 22 in boundary geometry 11 has zero length/area.

message ADF2111.

Fluid-structure interface (label) 22 in boundary condition 11 is
invalid. This could be an internal error or wrong data file was used.
This message reports an error found in a boundary condition data. A

message ADF2111.

Boundary geometry (label) 22 in boundary condition 11 is invalid.

message ADF2111.

Boundary condition (label) 11 has been applied to an improper location

Index
The left side of the location is fluid element group

The right side of the location is fluid element group
A boundary condition can only be applied to certain locations where they
are physically meaningful.
This message reports an error found in a boundary condition data.

This condition has been applied to a location that has no physical
meaning. For example, a wall condition cannot be applied to lines
inside a 2D fluid domain.

No conditions for moving mesh are found in boundary condition 5.

message ADF2111.

Parameter SLIPC (=55) in boundary condition 11 is invalid.
Check the input related to this condition.
This message reports an error found in boundary condition data.

The slipc parameter indicates the type of velocity condition in a defined
direction. You may check the input of that boundary condition. A
wrong data file may also cause this error. In this case, refer to the

BCD_TYPE (slave) and the attached solid boundaries (master) are
parallel. Current available information is
slave condition label = 1111;
slave nodal label = 2222;
slave normal = 1.000E+03, 2.000E+03, 3.000E+03
master condition label = 3333;
master normal = 4.000E+03, 5.000E+03, 6.000E+03
This message reports an error found while treating a boundary

condition data. The boundary condition could a free-surface condition,
a fluid-fluid condition or a phase-change condition. When a free

surface, for example, touches a wall, the wall functions as a frontier.

However, program finds out the normal direction of the free surface is
coincident with the normal direction of the wall and, therefore, unable to
continue the computation.
n n
fluid wall

Moving type 22 in boundary condition 11 is invalid.

message ADF2111.

Type 22 in boundary condition 11 is invalid.

message ADF2111.

The parameter 22 that indicates number of real numbers in boundary
condition 11 is invalid.

Index

message ADF2111.

Number 22 for the integers in boundary condition 11 is invalid.

message ADF2111.

Number of boundary nodes 22 in boundary condition 11 is invalid.

message ADF2111.

Boundary node (label) 99999 is unable to be connected to other boundary
nodes in boundary condition 11.
* While a boundary condition is applied to nodes (e.g., defined in

PATRAN, I-DEAS, etc.) incorrect or incomplete boundary nodes may
have been picked.
* An internal line/surface is applied with a condition that should only
applied to boundary.
* A wrong data file was used.

Boundary nodes in the same boundary condition must be able to be
connected to each other and form boundary elements that represent
discretized boundary lines (in two dimensions) or surfaces (in three
dimensions).
If the model is generated using other programs, the conditions are

usually applied to nodes rather than on geometry lines and surfaces.

When there is data missed while they are imported to ADINA system
(usually through the TRANSOR program), the boundary nodes may not
be connectable. Sometimes, these programs allow conditions only
applied to nodes, you may pick incomplete nodes along a boundary.
This may also cause this problem.
The boundary condition of the reported type here can only be
applied to boundaries. Therefore, the boundary elements are facing
fluid elements on only one side of them. If a boundary condition is
applied to internal lines or surfaces, where both sides of them facing
fluid elements, the boundary elements are unable to be identified and
generated. In this case, the boundary condition is applied to a wrong
location.

Variable identification number 22 is out of range [1,3] in
user-supplied boundary condition 11.
This message reports an error found in a user-supplied boundary

condition data. A wrong data file may cause this error. Refer to the

Follower boundary node 33 is defined at a wrong location. It should only
appear in moving boundary condition where the geometry is line.
Current boundary geometry type is 22 in boundary condition 11.

message ADF2111.

Geometry label 33 in boundary condition 11 is invalid. The
condition type is 22. In conditions of this type, no geometries are

Index
expected. This could be an internal error or a wrong data file was used.
This message reports a potential error encountered in boundary

condition data. A wrong data file may cause this error. Refer to the

Geometry type 33 in boundary condition 11 is invalid.
Current boundary condition type is 22. Geometry type should be 2
or 3 indicating a line or a surface respectively.

message ADF2111.

Follower nodes in boundary condition 11 are impossible to follow
leader nodes that are defined in boundary condition 22.

message ADF2111.

Inconsistent boundary dimensions 22 and 33 are generated. Current
boundary condition is 11.
This message reports an error found in boundary condition data. In

general, a boundary condition (particularly a special boundary
condition) should be applied to lines (in two-dimensional models) or
surfaces (in three-dimensional models). It is not allowed that a
boundary condition is applied to lines and surfaces in the same model.
A wrong data file may also cause this error. In this case, refer to the

Boundary node (label) 999999 defined in boundary condition 1111

is out of possible range.

message ADF2111.

Follower node (label) 99999999 in boundary condition 11 is invalid.

message ADF2111.

Leader node (label) 99999999 in boundary condition 11 is invalid.

message ADF2111.

Leader nodal index 22 in boundary condition 11 is invalid.

message ADF2111.

Open value 1.1200E+00 is smaller than close value 2.2200E+00 in gap
boundary condition 11. This may result in unreasonable oscillations in
open/close status in this gap.
As a general guide, always let open value be slightly bigger than close

Index
value.
This message reports a potential error found in a gap boundary

condition data. The open value and close value in a gap condition
controls the open/close status of the gap. If the open value is smaller
than the close value, the gap may alternatively open and close in each
time steps. Usually, this is unwanted physical situation. As a general
rule, the open value must be slightly larger than the close value. Refer
to the description of the gap condition.

Close value 1.1230E-13 is too small in gap boundary condition 22.
It has been changed to 2.2221E+00.
This message reports a potential error found in a gap boundary

condition data. When the close value is smaller than 1012 , this warning
message is printed and the close value is modified to 1012 . When this
value is too small, the meshes near the gap may be overlapped as the
gap approaches to the closed position.

Type 22 in gap boundary condition 11 is invalid.

message ADF2111.

Initial open/close status 22 in gap boundary condition 11 is
invalid.
This message reports an error found in a gap boundary condition

data. A wrong data file may cause this error. Refer to the suggestions

Inconsistent dimensions of boundary elements in boundary condition 11.

Each two-dimensional boundary element is expected to be connected by two
nodes.
Current boundary element is connected by three boundary nodes (their
indices are 22, 33 and 44), corresponding to nodes (labels)
55, 66 and 77.

The cause of this error is essentially the same as described for message
ADF2762.

Inconsistent dimensions of boundary elements in boundary condition 11.
Each three-dimensional boundary element is expected to be connected by
three nodes.
Current boundary element is connected by two boundary nodes (their indices
are 22 and 33), corresponding to nodes (labels) 55
and 66.

ADF2762.

Inconsistent boundary conditions have been found. Any two nodes on the
same side of an element (node=9 or 27) must be assigned same boundary
conditions. Currently available information is:
boundary condition = 11;
element number = 33;
first node on the side = 44;
second node on the side = 55.
If boundary conditions are applied to geometries, this error should never

occur unless a wrong data file was used. However, if they are applied to
nodes (e.g., defined in PATRAN, I-DEAS, etc.), incorrect or incomplete
boundary nodes may have been picked.

ADF2762.

Index

Memory is too little for the computation to continue. The block length
1119 is smaller than the number of equations.
This message indicates that memory assigned to ADINA-F is far

less than required. Refer to the explanations under the topic Use of
memory and disk for more details.

Memory is too little for the computation to continue. Too many blocks
(123456) are generated.
This message indicates that memory assigned to ADINA-F is far


Boundary side is connected to two elements in boundary condition 11.
Nodes that form the side are 22 and 33.
This side will not be treated as boundary element.
This message reports a possible error in the boundary condition data.

If the elements are generated using a free mesh generator, it is possible
there are some elements in which each corner nodes have been assigned
the same boundary condition. If the elements are generated using a rule
based mesh generator, this situation rarely occurs although still possible.
However, certain incorrect applications of boundary condition are
also possible. In the example shown in the next figure (a), if the same
wall condition is applied to both top and bottom lines, the program will
be confused. To solve the problem, you may either apply two wall
conditions of different labels to the top and to the bottom respectively as
shown in (a) or use more than one layer elements in the fluid domain as
shown in (b).

wall 1 wall 1
(a) wall 2 (b) wall 1
If the model is generated using other programs, the conditions are

usually applied to nodes rather than on geometry lines and surfaces.
When some internal nodes are picked while a boundary condition is
applied, this error may also occur.

Boundary face is connected to two elements in boundary condition 11.
Nodes that form the side are 22, 33 and 44.
This face will not be treated as boundary element.

This is the three-dimensional case of the message ADF2804.

Boundary side is connected to two elements in boundary condition 11.
The center node on the side is 22.
This side will not be treated as boundary element.

The cause of it is possibly the same as described for message ADF2804.
However, this message indicates that element is either quadrilateral 9-
node or brick 27-node. In these elements, this message usually points to
an error in the element. It is possible that a boundary condition that can
only be applied to boundaries has been applied to an internal line or
surface.

No enough memory to continue.
This message indicates that the memory assigned to ADINA-F is far

Index

No enough memory to continue. Currently available information about the
required array is:
type (LW )= 11;
dimension (LD )= 22;
device unit (IUN )= 33;
short of words (IAM-IA2-1) = 44;
identification string = ID_STRING.

enough. Refer to the explanations under the topic Use of memory and
disk for more details.

No pressure degree of freedom in the model.
Check the flow assumptions or the definitions in "MASTER" command.
This message indicates that there is no pressure degree of freedom in

this model. Whenever fluid velocity is considered, the pressure must be
included in the model.

There are no active degrees of freedom.
This message indicates that there are no solution variables included

in this model. It is possible that you may have created or modified your
model in commands. In MASTER command, you probably have turned
off all degrees of freedom.


Variable identification flag 11 (IDOF) is invalid.
It is also possible that a wrong data file has been used.
This message reports an error that is related to the overall solution

degrees of freedom. A wrong data file may cause this error. Refer to

Flag for temperature file is invalid.
Possibly a wrong data file was used.
This message reports an error that is related to the usage of the

temperature file. A wrong data file may cause this error. Refer to the

Pressure equation 111111 at node 222222 is unable to be reasonably
connected to its neighbor velocity degrees of freedom. System matrix
could be singular or the boundary condition near the node could be
incorrect.
This message reports a potential error in this model. The solution

variable pressure must always be coupled with the solution variable
velocities. However, program has identified a pressure that has no
connections with velocities. Possibly the velocities around that pressure
node have been wrongly fixed. This error can only occur when
quadrilateral 9-node and brick 27-node elements are used. A typical
example is shown in the figure below.

Index
v=0 at the 9 nodes

p has no
connection
with v
wall

"Dead" element 111111 encountered in element group 22.
Note: An element is called "dead" if the same boundary condition has been
applied to all nodes in this element. In order to avoid "dead" elements,
split the boundary condition into two boundary conditions of the same type
and apply them to different sides of elements.
This message reports a possible error found in boundary condition

data. The cause of this error is essentially the same as described for
message ADF2804.

A boundary has been applied twice-different compressible flow conditions.
The side is facing the local nodal index 33 of element 11 in
element group 22. The two boundary condition labels are 44 and
55. Check if the two conditions have been applied correctly.
As a general guide in modeling of high speed compressible flows, every

boundary must have one and only one boundary condition for fluid and
energy equations.
This message reports a possible error found in boundary condition

data in the high-speed compressible flow model. Some special
boundary conditions are particularly designed for high-speed
compressible flows. As a general rule, each boundary can be applied
only one such boundary condition. If two conditions are applied to the

same location, a thorough understanding of these boundary conditions is

required.

No pressure or temperature degree of freedom in the model.
For compressible flows, both pressure and temperature degrees of freedom
must be active.
This message reports an error in compressible flows. Recall that in

the state equation, pressure, temperature and density are coupled.
Therefore, pressure, temperature and velocity must be all active in
compressible flow models.

Error encountered in I/O operation.
device unit = 11;

I/O flag = 22;
length of data = 33;
I/O status = 44.
* File is too big;

* Disk space is not big enough for the current problem;
* Some working files are accidentally deleted.
This message reports an unidentified error while accessing the hard

disk. Make sure that the possible causes in this message are fixed.

Thermal resistance boundary condition cannot be applied to high-speed
compressible flows.
The thermal resistance boundary condition is currently not available

for high-speed compressible flows.

Index

Line search is not available.
This message reports an error that is related to an unavailable option.

message ADF2111.

Incorrect flag 11 for direct coupling of fluid-structure interactions.
This message reports an error that is related to an unavailable option.

message ADF2111.

Pressure datum has been ignored since there is no pressure degree of
freedom in the model. The results may be incorrect.
This message reports a potential error in the model. If you are intent
selecting the option of pressure datum, you must include the pressure as
a variable. Or you may remove the selection of the option.

No pressure degree of freedom at node 11 where pressure datum is
applied. Pressure datum has been ignored. The results may be incorrect.
This message reports an error in the pressure datum data. The

pressure datum must be applied to a point or node that has pressure
degree of freedom.

Only the sparse solver can be used for direct coupling of fluid-structure

interactions.
This message reports an unavailable option has been selected. Only

the sparse solver can be used in direct computing of two-way fluid-
structure coupling problems.

One-way fluid-structure interactions cannot be performed using direct
coupling.
This message reports an unavailable option has been selected.

Recall that the fluid solution variables are not affected by the solid
displacements in one-way fluid-structure coupling problems. There is
no need to use the direct computing method. Refer to the related
chapters for more details.

Two- and three-dimensional elements cannot be used together in this model.
This message reports an error in the selected elements. Mixed

elements are generally unacceptable in the same model. Separate the 2-
dimensional and 3-dimensional models.

Adaptive mesh option cannot be applied to models that include solid
element groups (e.g., conjugate heat transfer problems).

Currently the mesh repair or adaptive mesh cannot be applied to
conjugate heat transfer problems.

Memory is not enough to keep element group data in-core. Use of adaptive
mesh option requires in-core element group data.

Index
Whenever the mesh repair or mesh adaptive option is used, we

assume that the memory of the computer is enough and available during
the whole computation process. Since the procedure of mesh repair is
dynamic, the required memory is unable to be predicted in initiation
stage of computation.

Only the sparse solver can be used if adaptive mesh option is applied.
This message reports an unavailable option has been selected. Only

the sparse solver can be used if mesh repair or adaptive mesh option is
selected.

Only triangular/tetrahedron elements can be used if adaptive mesh option
is applied.
This message reports an unavailable option has been selected. The

mesh repair or adaptive mesh option can only be applied to triangular 3-
node or tetrahedral 4-node elements.

Adaptive mesh requires in core solution. Memory is not enough to continue.

enough. Once the mesh repair or adaptive mesh option is selected, the
solution procedure is in core and, therefore, more memory may be
required. Refer to the descriptions under the topic Use of memory and
disk.

Fluid-fluid or phase change boundary conditions are not applicable if
adaptive mesh option is applied.

mesh repair or adaptive mesh option is currently not allowed in the
problems that have fluid-fluid or/and phase-change boundary

conditions.

Unsuccessful in mesh repair. Current function is "ADP_FUNCTION_NAME".
This message reports an unsuccessful attempt to repair an invalid

mesh. When the solution is divergent, the mesh could be too far from
being able to be repaired. Try to use small time steps, small load
increments, etc. to have the solution changed smoothly.

Unsuccessful in fluid model for one-way coupling (through files) of
fluid-structure interaction.
This message indicates that the solution procedure in ADINA-F is

unsuccessful in a separate run of one-way fluid-structure coupling
problems. Test a direct run of the same problem. If it is successful, you
probably had used a wrong syntax in the separate run. If it is not
successful, retest your model. Refer to the suggestions under the topic
Model preparation and testing.

Phase-change boundary conditions cannot be applied to compressible flows.
This message reports an unavailable option has been selected. In

high-speed compressible flows, the phase-change boundary condition
option is not available. Refer to the related chapters to see the available

Different element types cannot be used in a problem if it contains
moving mesh boundary conditions.
This message reports an error in the selected elements. Mixed

elements are generally unacceptable in the same model, particularly if

Index
there is a moving boundary condition applied. You must also not apply
special boundary conditions to lines and surfaces in the same model.
Resolve these issues before rerunning this problem.

Multi-grid solver can only be applied to problems with triangular or
tetrahedron elements.

multi-grid solver can only be applied to problems with triangular 3-node
or tetrahedral 4-node elements.

Multi-grid solver cannot be applied to this problem since there is no
pressure degree of freedom.
This message indicates that there is no pressure degree of freedom in

this model. Whenever fluid velocity is considered, the pressure must be
included in the model. In pure heat transfer problems where only
temperature is included, use other solvers.

Explicit method cannot be applied to problems with phase-change boundary
conditions.
This message indicates a limitation of the explicit methods. These

methods are generally not as stable as their implicit counterparts. You
cannot use explicit methods in problems that include moving boundary

Convection boundary conditions cannot be applied to compressible flows.

high-speed compressible flows, the convection boundary condition

option is not available. Refer to the related chapters to see the available

Radiation boundary conditions cannot be applied to compressible flows.

high-speed compressible flows, the radiation boundary condition option
is not available. Refer to the available boundary conditions in chapter 7.

Unsuccessful in solid model for fluid-structure interaction. Last status
in solid model is "CALLING_STAGE".
This message indicates that the solution procedure in the ADINA

solid model is unsuccessful in fluid-structure coupling problems. You
can find out more information from the <solid>.log file. It is important
to refer to the suggestions under the topics Model preparation and
testing and Formulation of fluid-structure interactions.
If this message is printed at the first iteration in the first time step, a

Unsuccessful in fluid model for fluid-structure interaction. Last status
in fluid model is "CALLING_STAGE".
This message indicates that the solution procedure in the ADINA-F

fluid model is unsuccessful in fluid-structure coupling problems. You
can find out more information from the <fluid>.log file. It is important
to refer to the suggestions under the topics Model preparation and

Index

Unsuccessful during iterations for fluid-structure interaction. Last
status is "CALLING_STAGE".
This message indicates that the FSI iteration is divergent in an

iterative computing of two-way fluid-structure coupling problems.
Refer to the suggestions under the topics Model preparation and

Unsuccessful while firstly initiate fluid model.
This message indicates that the initiation in the ADINA-F fluid

model is unsuccessful. You can find out more information from the
<fluid>.log file. It is likely that a wrong data file causes this error. In
this case, refer to the suggestions for message ADF2111. It is also
important to refer to the suggestions under the topics Model
preparation and testing and Formulation of fluid-structure
interactions.

Unsuccessful while firstly initiate solid model.
This message indicates that the initiation in the ADINA solid model
is unsuccessful. You can find out more information from the
<solid>.log file. It is likely that a wrong data file causes this error. In
interactions.

Unsuccessful while secondly initiate fluid model.

This message indicates that the initiation in the ADINA-F fluid

model is unsuccessful. You can find out more information from the
<fluid>.log file. It is likely that a wrong data file causes this error. In
interactions.

Unsuccessful while secondly initiate solid model.
This message indicates that the initiation in the ADINA solid model
is unsuccessful. You can find out more information from the
<solid>.log file. It is likely that a wrong data file causes this error. In
interactions.

Current adaptive mesh cannot be applied to problems with more than one
type of materials. Material type in element group 1 is 11 and material
type in element group 22 is 33.

Currently the mesh repair is carried out through the whole
computational domain. The element groups with different materials
cannot therefore be preserved.

Current adaptive mesh cannot be used together with the element birth/death
option. Element birth/death was found in element group 11.

Currently the mesh repair is carried out for every element. The element
birth/death information cannot therefore be preserved.

Index

Too many times of mesh repairs have been performed in current time step.
This message usually indicates an unsuccessful attempting on mesh

repair. Too many times of repair in a time step cannot lead to a
converged solution. The solution at this time step may be crucial,
sensitive, going to diverge, etc. Try to use small time steps, small load
increments around the current time. It is also important to refer to the
suggestions under the topics Model preparation and testing and
Formulation of fluid-structure interactions.

Unsuccessful while fluid walls are moved.

fluid model is unsuccessful in solving moving meshes. You can find
out more information from the <fluid>.log file.
If this occurs before the FSI iterations, it is likely that the mesh is
wrong. It could be that the big gaps existing between the interfaces of
fluid and solid are too large. A wrong data file may also cause this
error. In this case, refer to the suggestions for message ADF2111.
If this occurs during the FSI iterations, the structure displacements
may be incorrect or too distorted. It is usually a sign that the solution
procedure is divergent. It is important to refer to the suggestions under
the topics Model preparation and testing and Formulation of fluid-
structure interactions.

Unsuccessful while assemble fluid equations.

fluid model is unsuccessful in assembling the fluid equations. You can
find out more information from the <fluid>.log file.
Usually a wrong data file may cause this error. Refer to the
suggestions for message ADF2111. It is important to refer to the


Unsuccessful while assemble solid equations.

solid model is unsuccessful in assembling the fluid equations. You can
find out more information from the <solid>.log file.
Usually a wrong data file may cause this error. Refer to the
suggestions for message ADF2111. It is important to refer to the

Unsuccessful while fluid nodes on fluid-structure interfaces are moved.

fluid model is unsuccessful in solving moving meshes. You can find
out more information from the <fluid>.log file.
If this occurs before the FSI iterations, it is likely that the mesh is
wrong. It could be that the gaps existing between the interfaces of fluid
and solid are too large. A wrong data file may also cause this error. In
this case, refer to the suggestions for message ADF2111.
If this occurs during the FSI iterations, the structure displacements
may be incorrect or too distorted. It is usually a sign that the solution
procedure is divergent. It is important to refer to the suggestions under

Unsuccessful while reset data in the solid model.

solid model is unsuccessful. You can find out more information from
the <solid>.log file. An accidentally changing working environment
can cause this problem. When there is no space left on hard disk, this
error may also occur.

Index
Unsuccessful while reset data in the fluid model.

fluid model is unsuccessful. You can find out more information from
the <fluid>.log file. An accidentally changing working environment can
cause this problem. When there is no space left on hard disk, this error
may also occur.

Unsuccessful while solid solutions are being updated.
Refer to the explanations for message ADF3061.

Unsuccessful while fluid solutions are being updated.

Unsuccessful while output solutions in the fluid model.

Unsuccessful while output solutions in the solid model.

Unsuccessful in the solid model during FSI iteration.

solid model is unsuccessful during FSI iterations. You can find out
more information from the <solid>.log file. Usually this indicates that
solution is divergent. It is important to refer to the suggestions under



Unsuccessful in the fluid model during FSI iteration.

fluid model is unsuccessful during FSI iterations. You can find out
more information from the <fluid>.log file. Usually this indicates that
solution is divergent. It is important to refer to the suggestions under

Unsuccessful during FSI iteration.
This message indicates that the solution procedure in FSI iteration is

unsuccessful. You can find out more information from the <fluid>.log
and <solid>.log files. Check the stress and displacement residual
history. If they indicate a converging behavior, you may increase the
maximum number of FSI iterations and allow it to converge. If they
indicate a diverging behavior, considering the choices below:
Always check the solutions obtained so far to see if they are

physically reasonable. Try to exclude possibilities of errors in
input data.
Reduce the stress relaxation factor if it is not too small.
Reduce the displacement relaxation factor if it is not too small.
It is generally suggested that tolerances in fluid and solid models
are smaller than the tolerance for FSI convergence tolerance. If
you have not done it, make some adjustments.
Perform direct computing rather than the iterative computing.
It is also important to refer to the suggestions under the topics

Model preparation and testing and Formulation of fluid-structure
interactions.

Index

Element group type 111 (NPAR(1)) is invalid.
This message reports an error in the element group data. A wrong

data file may cause this error. Refer to the suggestions for message
ADF2111.

Number of elements 11 (NPAR(2)) is invalid.

Flag 1 for the solid element group (NPAR(3)) is invalid.

Flag 1 for the element birth/death option (NPAR(4)) is invalid.

Element subtype 1 (NPAR(5)) is invalid.

Flag 11 (NPAR(6)) for skewed nodes is invalid.


Element type 1 (= NPAR(7) = number of nodes per element) is invalid.

Element type 1 (= NPAR(7) = number of nodes per element) is invalid.

Element type 1 (= NPAR(7) = number of nodes per element) is not
available in two dimensions.

Element type 1 (= NPAR(7) = number of nodes per element) is not
available in three dimensions.

Too many points are used in the pressure/temperature dependent
material data.
number of pressure points (NCP ) = 1;
number of temperature points (NCT ) = 2;
maximum allowable (NCP*NCT) = 3.

Index

Material type 1 (NPAR(15)) is invalid.

Number 11 (NPAR(16)) of pressure points in the pressure dependent
material data is invalid.

Number 11 (NPAR(18)) of temperature points in the temperature
dependent material data is invalid.

Upwinding type 1 (NPAR(19)) is invalid.

Bubble flag 1 (NPAR(20)) is invalid.

Flag 1 for clapped 9-node element (NPAR(8)) is invalid.


Level 1 of numerical integration (NPAR(10)) is invalid.

Level 1 of numerical integration in t-direction (NPAR(11)) is invalid

Material set number 1 (NPAR(16)) is invalid.

Flag 1 for energy dissipation (NPAR(20)) is invalid.

Too many specular nodes exist in this model.
Current 32-bit program versions can only handle problems with number
of specular nodes less than 46341.
This message indicates that the size of the radiosity matrix surpasses
the limit 2+31 on a 32-bit machine. Remember that the matrix is full and
therefore its size is 8nn2 bites, where nn is the number of specular nodes.
Too many specular nodes not only require large memory but also very
cost in computation. Refer to the descriptions under the topic Memory
usage.

Index

Too many specular elements exist in this model.
Current 32-bit program versions can only handle problems with number
of specular elements less than 46341.
This message indicates that the size of the view factor matrix
surpasses the limit 2+31 on a 32-bit machine. Remember that the matrix
is full and therefore its size is 8ne2 bites, where ne is the number of
specular elements. Too many specular elements not only require large
memory but also very cost in computation. Refer to the descriptions
under the topic Memory usage.

PARAMETER_NAME 123456 is invalid.
This message reports an invalid integer during run time. If this

occurs in the first iteration in the first time, a wrong data file may cause
this error. In this case, refer to the suggestions for message ADF2111.
If this occurs later in the computation, the solution may be divergent. It
is also important to refer to the suggestions under the topics Model
interactions.

PARAMETER_NAME 1.2345599E+18 is invalid.
This message reports an invalid real number during run time. If this
occurs in the first iteration in the first time, a wrong data file may cause
this error. In this case, refer to the suggestions for message ADF2111.
If this occurs later in the computation, the solution may be divergent. It
is also important to refer to the suggestions under the topics Model
interactions.

Current iterative solver fails during iteration. Temporary variable DENA

or DENB 1.23450E-21 is too small. Try to use other solvers.
This message indicates that the iterative solver fails in the current
computation.
If this occurs in the first iteration in the first time, a wrong data file
may cause this error. In this case, refer to the suggestions for message
ADF2111. It may also be caused by a poor matrix condition or poor
initial condition. The solution may be divergent as well. For all these
reasons, refer to the suggestions under the topics Model preparation
and testing and Strategies toward obtaining convergent solutions.
You may use the sparse solver instead.

The following element is overlapped:
local nodal indices and their coordinates =
1 .10000000E+01 .20000000E+01 .30000000E+01
2 .40000000E+01 .50000000E+01 .60000000E+01
3 .70000000E+01 .80000000E+01 .90000000E+01
This message reports an invalid or a too distorted element

encountered.
If this occurs at the first iteration in the first time, make sure the
following conditions are met.
A correct data file has been used. Refer to the suggestions for
message ADF2111.
Check the reported element and its neighbor element while the
element is generated. If it is overlapped, correct it.
If the reported element is near an FSI interface, consider the
possibility that the distance between the interfaces of fluid and
solid models is too large. Refer to the suggestions under the
topics Model preparation and testing and Formulation of
fluid-structure interaction.
If this occurs later in the computation, consider the following

suggestions:
Use proper leader-follower relations.

Index
Subdivide the domain into more regular domains that are close to
being convex while the mesh is generated. Refer to the
suggestions under the topic Control of moving mesh in ALE
formulation.
Use a good initial condition in transient analyses. Refer to the
suggestions under the topic Strategies toward obtaining
convergent solutions.

The following boundary element is overlapped:
boundary element (label) = 11;
boundary condition (label) = 22;
local nodal indices and their coordinates =
1 .10000000E+01 .20000000E+01 .30000000E+01
2 .40000000E+01 .50000000E+01 .60000000E+01
3 .70000000E+01 .80000000E+01 .90000000E+01
This message is essentially the same as message ADF5010.

However, the invalid element encountered here is a boundary element.
This may narrow the possible causes.

Storage is too small to save the coefficient matrix for semi-explicit
method. Computation continues using low-speed module.
This message indicates the memory assigned to ADINA-F is not

enough. The low-speed module of the semi-explicit method could be
very slow. It is recommend using implicit methods. For example, if the
sparse solver or iterative solvers are selected, the implicit time
integrations are automatically enforced.

Crashed mesh has been repaired on using the factor 1.23450E-21.
This message indicates the moving mesh in ALE formulation is not

fully satisfied. Increments of displacement have been reduced. It is
usually a sign that the solution is divergent. Refer to the suggestions
under the topics Model preparation and testing and Strategies toward

obtaining convergent solutions.

Unsuccessful in solving wall function equation.
Possibly the computation is divergent.
This message indicates the wall functions used in two-equation

turbulence models are unable to be calculated. It is usually a sign that
the solution is divergent. Refer to the suggestions under the topics
Model preparation and testing and Strategies toward obtaining
convergent solutions.

Unsuccessful in bubble condensation.
Possibly the computation is divergent.
This message indicates the bubble velocity degree of freedom is

unable to be condensed. If this occurs at the first iteration in the first
time step, usually a zero viscosity has been defined in a fluid element
group. It also could be caused by an element that is too distorted or a
divergent solution. Refer to the suggestions under the topics Model
preparation and testing and Strategies toward obtaining convergent
solutions.

Improper boundary values for pressure in boundary condition 11 are
specified. The A and B values are 1.23450E-21 and 2.23449E-21.
This message reports possibly invalid parameters specified in a

boundary condition for high-speed compressible flows. Refer to the
related chapters to see the available boundary conditions and their
limitations.

Improper values for temperature in boundary condition 11 are specified
The A and B values are 1.23450E-21 and 2.23449E-21.

Index

Improper values for density in boundary condition 11 are specified.

Improper values for internal energy in boundary condition 11 are

Improper values for enthalpy in boundary condition 11 are specified.

Improper values for normal velocity in boundary condition 11 are

Improper values for flow rate in boundary condition 11 are specified.

Improper values for Mach number in boundary condition 11 are specified


Fluid flows inward at an outlet boundary. Local normal velocity and sound
speed are 9.99999E-22 and 1.99999E-21 respectively. Model may be
unstable or improper boundary conditions and/or initial conditions are
applied.
This message reports that a flow condition becomes contrary to what

the boundary condition is designed for. If this occurs at the first
iteration in the first time step, probably wrong parameters have been
defined in that boundary condition. If you are intent using this type of
boundary condition as reported, ignore this warning. Refer to the
related chapters for more details as well.

Fluid flows outward at an inlet boundary. Local normal velocity and sound
speed are 9.99999E-22 and 1.99999E-21 respectively. Model may be
unstable or improper boundary conditions and/or initial conditions are
applied.


Fluid flow becomes supersonic at an outlet boundary. Local normal velocity
and sound speed are 9.99999E-22 and 1.99999E-21 respectively. Model
may be unstable, or improper boundary conditions and/or initial conditions
are applied.


Index

Fluid flow becomes supersonic at an inlet boundary. Local normal velocity
and sound speed are 9.99999E-22 and 1.99999E-21 respectively. Model
may be unstable, or improper boundary conditions and/or initial conditions
are applied.


End-of-file encountered in reading the data file.
Possibly a wrong file was used.
This message reports that the data file does not contain enough
information that is required. A wrong data file may cause this error.

No authority to use the program on this computer.
Please contact ADINA R&D, Inc.

Telephone (617)-926-5199
Telefax (617)-926-0238
Email support@adina.com
This message indicates that the license is invalid or expired.

Too few records in the data file.
Possibly a wrong file was used.
This message reports that the data file does not contain enough
required information. A wrong data file may cause this error. Refer to


High-speed storage is too small to fit one column matrix data in core.
This message reports that the memory assigned to ADINA-F is far

less than required. Refer to the descriptions under the topic Memory
usage for more details.

No storage available to continue. Increase storage (MTOT).

Too many blocks are generated. Maximum possible number of blocks is 11
This message reports that the memory assigned to ADINA-F is far

less than it required. If the Gauss elimination method (SKYLINE) has
been used, change it to the sparse solver or iterative solvers. Also refer
to the descriptions under the topic Memory usage for more details.

Iteration in mass transfer is divergent. Current iteration step is 11
and the residual is 1.23450E-21.
This message indicates that the iteration in mass transfer equations is

unsuccessful. The mass transfer equations are linear unless some
nonlinear source terms have been applied through user-supplied material
data. Check the residual history. If the tendency is convergent, increase
the maximum number of iterations to allow it to converge. If the
tendency is divergent, the source terms must be added in smaller
increments. It is also important to refer to the suggestions under the

Index
topics Model preparation and testing and Formulation of fluid-


Time step length 1.23450E-21 in time step block 111 is invalid.
This message indicates that a time step length is too small or

negative. Check the input in the time step control.

No fluid-structure-interaction boundary conditions are found.
Possible errors exist in the model.
This message indicates inconsistent information. In Formulation of

fluid-structure interactions, some fluid-structure interfaces must be
defined in both fluid and solid models. Current fluid data file does not
contain such information.

The sparse solver is not available on this computer.
This message indicates that the hardware sparse solver is not

available from the computer vendor. You can change it to the ADINA
sparse solver.

Error encountered while reading data from device 1 at local index
222.
This message indicates that the program is unable to read from a

device because of an unknown reason. One possible reason is that some

temporary files have been removed (not by the program). Another

possibility is that a wrong data file is used. In this case, refer to the

Error encountered in record "IDENTIFICATION_STRING" in restart file.
Possibly a wrong restart file was used.
This message indicates that the restart file is invalid. Possible

reasons are
A wrong restart file is used.

The file has been replaced in last execution of the problem.
Remember that the restart file is replaced once the execution
starts. It must always keep a backup of that file if multiple
attempts may be made.
The current version of ADINA-F is inconsistent with the one that
created the restart file.
Some limitations are not satisfied. Refer to the related chapters
for more details.

Norm of VARIABLE_NAME solution vector is smaller than 2.30E-14.
This message indicates that the reported solution is too small. An

inadequate modeling may cause this situation. If you are intent doing
so, ignore this message. It is also important to refer the suggestions
under the topic Model preparation and testing and other related
chapters.

Relative residual 2.30E+12 of VARIABLE_NAME is too large.
This message indicates that the relative residual of the reported

solution is too large. It could be that the solution is too small or the
solution is divergent. An inadequate modeling may cause this situation.
If you are intent doing so, ignore this message. It is also important to

Index
refer the suggestions under the topic Model preparation and testing
and other related chapters.

Absolute residual 2.30E+12 of VARIABLE_NAME is too large.
This message indicates that the absolute residual of the reported

solution is too large. The solution procedure is probably divergent. An
inadequate modeling may cause this situation too. It is also important to
refer the suggestions under the topic Model preparation and testing
and other related chapters.

No convergence is reached after the maximum number of iterations in the
iterative solver. Maximum number of iterations = 11, present time
step = 22 and present equilibrium iteration = 33.
This message indicates that the iterative solver is unsuccessful while

the current matrix system is solved. If the rest of the iterations are
successful, this message can be ignored although it is recommended to
find out the reasons. The default of the maximum number of iterations
(=3000) in iterative solvers may not be enough in the following
situations:
The condition of the matrix is poor (In this case, refer to the
suggestions under the topic Strategies toward obtaining
convergent solutions).
Too many equations are in the current system, usually more than
half million. In this case you may increase the maximum
number of iterations in iterative solvers.
The sparse solver is an alternative if desired.

No convergence is reached after the maximum number of iterations in the
iterative solver while solving pressure equations in semi-explicit method.
Present iterations = 11 and residual = 1.23450E-21.

This message indicates that the semi-explicit method is unsuccessful

for this model. Use other implicit methods, such as the sparse solver or
iterative solvers.

Incremental factor becomes too small.
present iteration = 11;
present residual = 2.00E-22;
present factor (AL) = 1.00E-22.
Refer to the message ADF6041.

Incremental factor becomes too large.
present iteration = 11;
present relative residual = 2.00E-22;
present absolute residual = 1.00E-22.
Refer to the message ADF6041.

Last record in the boundary condition data block is incorrect.
Data file is wrong.

message ADF2111.

Inconsistency between the old and new values in "ID_STRING":
the old value = 1;
the new value = 2.
This message reports inconsistent information found from different

sources, typically in restart runs.
An incorrect restart file could have been used. In this case, refer to
the explanations for message ADF6023.

Index

Inconsistency between the old and new values in "ID_STRING":
the old value = 2.0000000474844E-33;
the new value = 1.0000000237422E-33.
This message reports inconsistent information found from different sources,

typically in restart runs.
An incorrect restart file could have been used. In this case, refer to
the explanations for message ADF6023.

Data is different in restart file and data file.
The inconsistency occurs in = "IDENTIFICATION_STRING";
the value in restart file = 11;
current value = 22.
* The input for the restart run does not match the previous run;
* Incorrect restart file has been used.

Data is different in restart file and data file.
The inconsistency occurs in = "IDENTIFICATION_STRING";
the value in restart file = 2.0000000474844E-33;
current value = 1.0000000237422E-33.
* The input for the restart run does not match the previous run;
* Incorrect restart file has been used.

Unable to open the file of unit 11. Currently available information is

ICREAT, IFORM, IACCES, LENGTH = 22, 33, 44, 55.
This message indicates that the reported file is unable to be opened.

Try first to find out the file definitions corresponding to the unit number
from the table under the topic Memory usage. The file that is required
may not be in the current working directory. If the file is a temporary
file or unidentified from that table, probably the disk is full or you dont
have the authority to write into the current directory.

Inconsistency between old and new records.
index of the record = 11;
the value in restart file = 22;
current value = 33.
Possibly a wrong restart file was used.

Gaps were found in equation numbers. Total number of equations is 11,
but the maximum equation number is 22.

message ADF2111.

Inconsistent records were found regarding number of constraint equations.
number of constraint equations available = 1;
number of constraint equations in master data = 2.

message ADF2111.

Index

Inconsistent records were found regarding the total number of equations.
number of fixed degrees of freedom = 1;
number of constraint equations = 2;
number of active degrees of freedom = 3;
sum of them must be equal to = 4.

message ADF2111.

Fluid model could be wrong! The factorized diagonal 1.0000E-24 of
equation 11 is too small.
A constant pivot has been assigned to this equation.
This message is essentially the same as ADF2004.

Fluid model could be unstable! The factorized diagonal 1.0000E-24 of
equation 11 is too small.
This message is essentially the same as ADF2004.

Mass-ratio model could be wrong! The factorized diagonal 1.0000E-24 at
node (index) 11 is too small.
A constant pivot has been assigned to the equation at this node.

decomposition or incomplete decomposition in mass transfer equations.
high possibility is that the model is ill posed. The possible reasons are
as follows.
No prescribed mass-ratio conditions or mass convection

conditions are applied if a steady-state analysis is performed. If

mass convection condition is applied, the mass transfer
coefficient is wrongly specified as zero. The associated time
function must be checked as well.
The mass transfer diffusion coefficient is wrongly specified as
zero.
An improper unit system is used.
If this message appears after a few iterations, possibly the iteration is

divergent in the fluid model. It could be an indication that the model is
improperly defined, load increments are too large, the physical
phenomenon is unstable, etc. Refer to the topic Model preparation and
testing for helpful tips.

Mass-ratio model could be unstable! The factorized diagonal 1.0000E-24
at node (index) 11 is too small.

Current mesh could be invalid! The factorized diagonal 1.0000E-24 at

decomposition or incomplete decomposition in moving meshes.
high possibility is that mesh is invalid. In this case, check the elements
around the reported node.
If this message appears after a few iterations, possibly the iteration is
divergent in the fluid model. It could be an indication that the model is
improperly defined, load increments are too large, the physical
phenomenon is unstable, etc. Refer to the topic Model preparation and
testing for helpful tips. Refer to the related chapters on mesh control as
well.

Index
Current mesh could be invalid! The factorized diagonal 1.0000E-24 at

node (index) 11 is negative.

Node (label) 22 is out of range [1,11].
This message reports an invalid nodal label encountered. A wrong

ADF2111.

Skew system 111 is out of range [33,44]. This skew
system has been applied to node 22.
This message reports an invalid skew system label encountered. A

message ADF2111.

Node (label) 111 is out of range [22,33].
See message ADF6070.

Nodal index 9 is out of range [1,2].

Node (label) 11 cannot be identified since it is not defined in
nodal data. Data file could be wrong.


Axis has zero length in "status_WHEN_ERROR_OCCURS".
This message reports an invalid axis defined in a boundary

condition. Check the reported boundary conditions and correct errors
that occur. A wrong data file may cause also this error. In this case,
refer to the suggestions for message ADF2111.

Axis 2 has zero length in skew system 11.
This message reports an invalid skew system encountered. A wrong

ADF2111.

Skew system 11 is invalid: A-axis and B-axis are not orthogonal.
dot of the two axes = 3.30E-21
This message reports an invalid skew system encountered. A wrong

ADF2111.

Program is unable to find out the function (label) 11 defined in
"FUNCTION_NAME".
This message reports a time function or a one-dimensional function

cannot be found while it is referred. A wrong data file may cause this

Function (label) 11 in "FUNCTION_NAME" is invalid.


Index

Function (label) 11 in "FUNCTION_NAME" is invalid.
Operators related to this function have been ignored.


Points are out of order in time function 11.
This message indicates that a time function curve is invalid because

some points that define the curve are out of order. Tabulated time
function curves must be input in an increasing order and no repeated
points. A wrong data file may cause also this error. In this case, refer to

Too many points exist in time functions.
number of points in the current time function = 1;
maximum number of points defined in master data = 2.
This message reports inconsistent information existing in time

functions. A wrong data file may cause this error. Refer to the

time function points do not cover computational time range.
computational time range = [1.00E-22,2.00E+00];
minimum and maximum time function points = [2.99E-22,4.00E-03].
This message indicates that the range covered by the specified time
functions is smaller than the range the computation performed. Modify
these time functions or modify the time step control data.


Variable identification number 11 in ONE-D-FUNCTION 22 is invalid
Possible values are between 33 and 44.
This message reports an invalid parameter in the reported one-

dimensional function. A wrong data file may cause this error. Refer to

Argument identification number 11 in ONE-D-FUNCTION 22 is invalid
Possible values are between 0 and 2.
This message reports an invalid parameter in the reported one-

dimensional function. A wrong data file may cause this error. Refer to

An input flag indicates that the flow is dependent of mass-ratios.
However, all mass expansion coefficients in the buoyant force term are
zero. The input flag has been ignored.
This message reports inconsistent control in the input. Mass

transfers may not affect the fluid solutions unless one of the following
conditions is met:
At least one mass expansion coefficient is nonzero.

Fluid materials are mass-ratio dependent.
When mass transfers do not affect the fluid solutions, they do not need
to be coupled with fluid solutions.

An input flag indicates that the flow is independent of mass-ratios.
However, non zero mass expansion coefficients in the buoyant force term
are found. The input flag has been changed to indicate the mass-ratio
dependency.

Index
This message reports inconsistent control in the input. Mass

transfers will affect the fluid solutions if one of the following conditions
is met:
At least one mass expansion coefficient is nonzero.

Fluid materials are mass-ratio dependent.
In this case, the coupling of the fluid and mass-ratios must be

performed.

Incorrect data was found in the initial mass-ratios. Data file could be
wrong. Currently available information is:
N1, N2, NN = 11, 22, 33.
This message reports an invalid record in the mass initial condition

data. A wrong data file may cause this error. Refer to the suggestions

An error message passed out from SLVMSG. The current status is:
This message reports an error was encountered in the sparse solver.

message ADF2111.

Next time step length 1.00E-22 is too small.
This message reports a very small time step length has been
encountered during an automatic time step cut. This step will be
ignored.

Explicit methods may be unstable if the time steps are larger than the
critical time step length. It is suggested to use automatic time stepping
with the CFL number smaller than one.
This message reports a potential unstable factor in explicit methods.

When time steps are specified, they must be small enough to satisfy the
CFL condition at all solution times. If the step sizes are too small, you
will perform unnecessary computations. Therefore, it is recommended
to use the automatic time step (CFL) option for all explicit methods.
The selected CFL number must be smaller than 1, preferably around 0.8.

Explicit methods cannot be used for moving-mesh problems.
(example problems are fluid-structure interaction, moving wall, etc.).

methods are generally not as stable as their implicit counterparts. You
cannot use explicit methods in problems that include moving
boundaries.

Explicit methods may be unstable if the CFL number is larger than one.
Current CFL number is 1.00E-22. Check your input related to automatic
time stepping.

Maximum number of CFL related time step cuts has been reached without
arriving at the required time.
This message indicates that the required solution at the specified

time has not been obtained. This is normal if you use time integration
methods with the automatic time step (CFL) option. The solution
obtained in the last time step has been saved. In case a continuous
transient simulation is required, you can perform a restart run.

Index
Maximum number of ATS related time step cuts has been reached without
arriving at the required time.
This message indicates that the ATS method is unsuccessful. We

recommend the suggestions described under the topics Model
preparation and testing and Strategies toward obtaining convergent
solutions. Theoretically speaking, you can increase the maximum
number of cuts that is allowed in the ATS method. You should know
other choices as well.

Solid and fluid models are not connected properly. This could occur if:
* Solid file is not specified;
* Solid file is incorrect;
* No fluid-structure interface specified in solid model (where it
is called FSBOUNDARY in command);
* This is not a fluid-structure-interaction problem, but a
fluid-structure interface is defined in fluid model.
Recall that fluid-structure interaction needs both fluid and solid models
connected internally through fluid-structure interfaces. The interfaces
are defined as FSBOUNDARY and FLUID-STUCTURE BOUNDARY-CONDITION
in solid
and fluid models respectively.
This message indicates that the fluid model and solid model cannot
be connected through their interfaces. Refer to the related chapters for
more details on fluid-structure interactions.

Convergence is obtained in one fluid-structure iteration. This may be
caused by an inadequate convergence criterion. It could also occur if the
model is not created properly.
This message indicates that the converged solution is obtained after

one FSI iteration. If this is what you are intent modeling, ignore this
message.
However, inadequate modeling frequently causes this situation. For
example, the fluid stress is too small to affect the structure. If this is
physically true, no coupling model is necessary. On the other hand, if
the driven factor comes from the solid model, the displacement is too

small to affect fluid solutions. In this case, a one-way FSI coupling is

more suitable.

Automatic-time-stepping method ATS is currently not available for
one-way coupling (through files) of fluid-structure iteration.
This message reports a limitation on ATS method. When one-way

fluid-structure coupling analyses are performed separately, the solutions
information is exchanged through the file <solid>.fsi. The solutions
therefore must be computed as specified. ATS cut information cannot
be enforced in the solid model in later runs.

Explicit methods cannot be used together with ATS option.
This message indicates a limitation of the explicit methods. ATS is

very unpractical if it is used in explicit methods and therefore are not
allowed.

Explicit methods cannot be used for mass transfer.

methods are generally not as stable as their implicit counterparts. It is
not allowed to use explicit methods in problems that include mass
transfers. You can perform the analysis using implicit methods. The
sparse solver and iterative solvers will automatically enforce the implicit
time integrations.

Explicit methods cannot be used for steady-state analyses.

methods are generally not as stable as their implicit counterparts.
Always perform transient analyses in explicit time integrations.

Index

Explicit methods cannot be used for turbulent flows.

methods are generally not as stable as their implicit counterparts. It is
not allowed to use explicit methods for modeling turbulent flows.

Joule-Heat model could be wrong! The factorized diagonal 1.0000E-24 at

Joule-Heat model could be unstable! The factorized diagonal 1.0000E-12 at
These two messages report possible modeling errors in Joule-heat

input. For examples, the electrical conductivity is zero or no electric
potential is prescribed in the model.
*** ERROR *** CODE ADF8001- ADF8999:

Please contact ADINA R&D, Inc.
Telephone (617)-926-5199
Telefax (617)-926-0238
Email support@adina.com
These messages report the run time status when an error is

encountered. Program is unable to identify the exact cause of the error
when it occurs.
Make sure that the ADINA system has been installed properly, the
working environment has not been accidentally changed (file removed,
directory removed, etc.), disk has enough space for the problem, correct
data file has been used, correct restart file has been used in case of
restart runs, both fluid and solid data files are properly generated and
specified in FSI problems, etc.
A wrong data file may cause this type of error/warning message too.
In this case, refer to the suggestions for message ADF2111.
Contact ADINA R&D if necessary.

Index
Index
body force, 27, 39, 44
2
boundary condition, 32, 72, 85, 372
2D axisymmetric flow. See axisymmetric angular velocity. See angular velocity
flow condition
2D planar flow. See planar flow concentrated load. See concentrated
convection. See convection condition
3 distributed load. See distributed
3D fluid flow, 27, 29, 30, 36, 41, 45, 47, external. See external condition
48, 50 fluid-fluid. See fluid-fluid interface
free surface. See free surface
A FSI. See fluid-structure interface
absolute pressure, 40 gap. See gap condition
absolute temperature, 123, 126 high-speed compressible, 193218
absorption, 124 incompressible, 88136
acoustic flow, 46 low-speed compressible, 88136
active mass-ratio, 233 mass transfer, 23537
adaptive mesh, 346 phase-change. See phase-change
ADINA composite scheme, 81 boundary
air bag, 45 porous medium, 161
ALE formulation, 49, 69, 91, 328, 381 prescribed variable. See prescribed
differential form, 70 radiation. See radiation condition
integral form, 71 shell-thermal, 136
analysis type, 364. See transient and slightly compressible, 88136
steady-state specular radiation. See specular
angular velocity, 30, 93, 95, 98, 99, 103, radiation
207, 212 subsonic. See subsonic at inlet/outlet
angular velocity condition, 93, 98, 103, supersonic. See supersonic at
207, 212 inlet/outlet
ASME steam table, 149 symmetric. See symmetric condition
ATS option, 319 thermal resistance, 128
maximum subdivisions, 320 turbulence, 169, 227
Aungier Redlich Kwong model, 48 uniform flow. See uniform flow
Aungier Redlich Kwong model, 156 condition
averaged pressure, 110 user-supplied, 128, 218, 237. See user-
averaged velocity, 43, 52 supplied boundary condition
axisymmetric flow, 34, 36 wall. See fixed wall and moving wall
zero variable. See zero
B
Benard convection, 387

Index
boundary element, 85, 86, 94, 122, 126, constraint equation. See constraint
202, 205, 236, 249, 251, 261, 264, 265, condition
266, 267, 276 contact condition, 383
Boundary friction condition, 115 contact discontinuity, 47
boundary layer continuity equation, 27, 29, 30, 33, 34, 36,
thickness, 74, 75 41, 44, 45, 47, 48, 50, 82, 202
Boussinesq approximation, 39, 44 control volume, 85, 182, 186, 190, 192
Boussinesq number, 74 convection, 75, 122
Brinkman number, 76 condition, 235, 236
bulk density, 62, 63, 65, 66, 67, 68, 240 forced, 75, 387
bulk modulus of elasticity, 36, 45, 138, natural, 73, 74, 374
139, 140, 163, 175 convective velocity, 71
bulk velocity, 62 convergence criteria in inner iteration, 303
buoyant force, 40, 68, 74, 75 convergence criteria in outer iteration, 297
Coriolis velocity, 31
C
coupled fluid-solid system. See fluid-
capabilities of ADINA-F, 24 structure interaction>coupled system
Capillary number, 76 cylindrical coordinate, 34
Cartesian coordinate, 27, 30
D
cell Peclet number, 84
centrifugal force, 33 Darcys law, 43, 44
CFL condition, 318 Darcy-Forchheimer equation, 43
CFL number. See CFL option decouple of fluid from temperature, 39
CFL option, 318, 364, 392 deformable structure, 69
compressibility, 36, 37, 45, 47, 54 deformation rate, 29, 35, 51, 139, 140,
compressible flow, 47, 73, 232, 246, 318, 141, 143
361 degree of freedom. See number of
high-speed. See high-speed equations
compressible flow density, 27, 36, 37, 39, 47, 55, 138, 140,
low-speed. See low-speed compressible 143, 175
flow fluid, 44, 163
computational domain, 69, 82, 88, 161, solid, 44, 163
167, 169, 198, 226, 234, 246, 252, 325, Detached Eddy Simulation (DES) model,
329, 330, 369, 392 59
concentrated force, 86, 94, 202 diffusion
concentrated heat flow, 86, 121, 204 fluid, 76
conduction, 75, 76 mass, 76
confined flow, 37, 45 thermal, 74, 75
conjugate heat transfer, 274, 324 viscosity, 74
conservative form, 27, 36 diffusivity, 124
conservative variable, 28, 182, 186, 188, displacement compatibility, 248
191, 192, 218, 219, 230, 374 displacement criterion, 254
constraint condition, 323 dissipation, 40

Index
distributed current density load, 334 Euler equations, 49

distributed heat flux, 121, 204 explicit time integration, 188
distributed mass flux, 233, 236 explosion, 45
distributed normal-traction, 86, 94, 202 exponent of the pressure ratio, 239
divergence theorem, 82 exponent of the temperature ratio, 239
dynamic term, 80 external condition, 213
E F
Eckert number, 40, 75 FCBI elements, 82, 88
eigenvalue, 49 FCBI-C elements, 88
eigenvector, 49 Ficks law, 63, 64
Electro-static and steady current fictitious time, 318
conduction analyses, 333 field centrifugal force, 86, 203
element, 276, 376 field centrifugal load, 95
3D brick element (27-node), 283 field element. See element
FCBI brick element (8-node), 287 Field friction, 151
FCBI prism element (6-node), 287 file, 403
FCBI pyramid element (5-node), 287 finite volume method, 160, 168, 234
FCBI quadrilateral element (4-node), fixed wall, 96, 206
285 no-slip, 96, 206
FCBI tetrahedral element (4-node), 287 slip, 97, 207
FCBI triangular element (3-node), 285 fixity condition. See zero variable
FCBI-C element, 289 fluid equation, 253
fluid-structure interface, 260 fluid model, 104, 249, 252, 253, 360
high-speed compressible, 190 fluid potential, 46
line, 276 fluid-fluid interface, 107
quadrilateral element (9-node), 279 fluid-structure interaction, 35, 41, 77, 80,
tetrahedral element (4-node), 282 246
triangular element (3-node), 277 coupled system, 253
triangular element (6-node), 280 direct computing, 256
emissivity, 123 in porous media, 270, 272
emittance, 124 iterative computing, 255
energy equation, 27, 29, 30, 33, 34, 36, one-way coupling, 253, 258, 259
41, 44, 45, 47, 48, 50 thermal effect, 270, 273
enthalpy, 48 time integration, 252
environment mass-ratio, 236 two-way coupling, 253
environment temperature, 122, 123, 126 fluid-structure interface, 104, 218, 261
equilibrium iteration, 257, 292, 366, 369, body connection, 261
379 2D C-shape, 264
FSI, 255, 256, 257 2D O-shape, 261
maximum number, 256, 257 3D C-shape, 266
Euler -method , 81 3D O-shape, 265
Euler backward method, 81 boundary connection, 261

Index
corner node, 261, 264, 265, 266 axisymmetric, 183

displacement interpolation, 249 planar, 183
no-slip, 104, 218 three dimenisons, 183
relative distance, 250 governing equation, 2776
rotational, 104, 218 compact form, 48
separated boundary, 261, 264, 265, 266 conservative form, 27
slip, 104, 218 nonconservative form, 29
stress interpolation, 249 Grashof number, 75
tangential, 104, 218 gravitational acceleration, 28, 138, 140,
unmatched meshes, 249 163, 175
flux-splitting method, 185 gravitational force, 28, 39, 40
follower. See leader-follower option
H
Fourier's law, 28
free surface, 69, 105 heat conduct coefficient, 28, 51, 138, 139,
free surface curvature, 106 140, 141, 175, 219, 220, 221
functionality fluid, 44, 163
high-speed compressible flow, 180 laminar, 51
incompressible flow, 78 solid, 44, 163
low-speed compressible flow, 78 turbulent, 51
mass transfer, 232 turbulent, 51, 52, 53, 54
porous media flow, 158 heat conductivity, 49
slightly compressible flow, 78 heat convection coefficient, 122
turbulence in high-speed compressible heat flux, 27, 82, 87
flow, 224 heat flux condition. See distributed heat
turbulence in incompressible flow, 165 flux
turbulence in low-speed compressible heat transfer, 80
flow, 165 heat transfer equation, 83
turbulence in slightly compressible discrete, 84
flow, 165 exact solution, 83
numerical solution, 84
G
oscillate, 84
Galerkin elements, 87 high-speed compressible flow, 47, 48,
Galerkin method, 82, 85 180, 360
gap hydrostatic pressure, 40, 41
boundary condition, 108, 218
I
close, 110
discontinous solution, 108 ideal gas. See perfect gas
open, 110 Ideal gas models, 155
size, 110 inactive variable, 296
status, 110 incompressible flow, 37, 45, 47, 77, 80,
Gas models, 154 246, 360
Gauss elimination method, 310, 377 incremental procedure, 387
geometric conservation, 183 inertia force, 74

Index
initial condition, 81, 137, 163, 175, 189, mass diffusion coefficient, 67, 238, 239,
219, 230, 237, 269, 322, 365, 366, 384, 240
386, 387, 390, 392, 445 mass expansion coefficient, 238, 239, 240
inner iteration, 292, 303 mass expansion coefficients, 68
internal energy, 47, 75 mass flow rate, 63
inviscid, 49 mass flux condition. See distributed mass
inviscid fluid flow, 48 flux
isentropic condition, 37 mass partial density, 62
iterative solver AMG1, 314 mass ratio, 63
iterative solver AMG2, 314 mass transfer, 62, 77, 232
iterative solvers, 314 mass transfer equation, 63, 64, 65
mass-ratio, 63, 65, 67
K
master degree, 323
kinetic energy, 75 material, 377
kinetic pressure, 40 high-speed compressible, 21923
kinetic stress, 41 incompressible, 13851
Kunz model, 340 low-speed compressible, 13851
mass transfer, 23744
L porous medium, 163
large-eddy-simulation model, 51 slightly compressible, 13851
latent heat, 108 material curves, 152
leader-follower closest option, 332 material model
leader-follower cone option, 332 K - high-Reynolds turbulence, 177
leader-follower option, 328, 330 K - low-Reynolds turbulence, 177
leader-follower parallel option, 331
Lewis number, 76 K - turbulence, 175, 230
liquid-vapor phase change, 336 Carreau, 140
load step size. See time step size constant, 138, 219, 238
low-speed compressible flow, 47, 77, 80, DES model, 178
360 large-eddy-simulation, 141
non-Newtonian, 13941
M porous medium, 163
Mach number, 37, 40, 47, 50, 74, 75 power-law, 139, 220
mapping file, 269, 392, 403 pressure-temperature-dependent, 239
format, 321 RNG K - turbulence, 177
mapping solution. See mapping file second order, 141
mass buoyant force ratio, 240 shear stress transport model, 178
mass concentration, 62 Southerlands formular, 220
mass conservation, 62 Spalart-Allmaras, 178
mass convection, 236 temperature-dependent, 39, 140
mass creation rate, 62, 63, 67, 240 temperature-dependent power-law, 140
mass density, 62, 65 time-dependent, 139
mass density flux, 63 two-layer zonal turbulence model, 179

Index
user-supplied, 143, 221, 239 noise, 45

velocity-dependent, 238 nonconservative form, 36, 45
mathematical formulation, 38 nondimension, 72
matrix condition, 318, 364, 365, 377, 386, coordinate reference, 72
392, 393 density scale, 72
memory, 398 length scale, 72
basic, 398 mass-ratio scale, 72
disk, 403 reference, 72
mass transfer, 401 scale, 72
moving mesh, 401 specific heat scale, 72
sparse solver, 400 temperature reference, 72
specular, 401 temperature scale, 72
mesh solver, 295 velocity scale, 72
mixing length, 143 normal-traction condition. See distributed
molar concentration, 64 normal-traction
molar density, 64, 65 number of equations, 403
molar density flux, 64 number of specifies, 62
molar-ratio, 64, 65 numerical instability, 83
molecular weight, 64 numerical methods
momentum equation, 27, 29, 30, 33, 34, high-speed compressible flow, 182
36, 41, 43, 45, 47, 48, 50 incompressible flow, 80
moving boundary, 69 low-speed compressible flow, 80
moving coordinate system, 95 mass transfer, 233
moving mesh, 328, 330, 381 porous medium, 160
moving velocity, 70 slightly compressible flow, 80
moving wall, 69, 99, 208 turbulence, 167, 226
displacement, 99, 100, 102, 103, 209, numerical oscillation, 111
211, 212
O
no-slip, 99, 209
rotational, 103, 212 one-way coupled thermal-mechanical
slip, 87, 100, 209 interaction, 274
tangential, 102, 211 outer iteration, 292, 293, 297
multi-grid solver. See solver>multi-grid
P
N
Peclet number, 75, 83
natural boundary condition, 82 Peng Robison model, 48
Navier-Stokes equation, 27, 29, 33, 41, Peng Robison model, 157
50, 70, 72 perfect gas, 47, 75, 137, 175, 219, 230,
Navier-Stokes equations, 71 396
neutral species, 62 periodic boundary condition, 119
Newton-Raphson method, 129, 160, 257, permeability, 41, 43, 163, 273
292, 384, 387, 392 PFSI, 271, 272
nodal interpolation, 295 phase change, 336

Index
phase-change boundary, 107 rarefaction, 47

physical stress, 41 ratio of specific heats, 37
physical traction. See mathematical ray tracing, 87
formulation Rayleigh number, 74
PISO, 295 real gas models, 48
planar flow, 33 Real gas models, 154
Poisson equation, 44 Redlich Kwong models, 155
pore pressure. See pressure reference datum of coordinate. See
porosity, 43, 163, 272 nondimension>coordinate reference
porous fluid-structure-interaction, 272 reference mass-ratio, 68, 238, 239
porous media, 272 reference pressure, 239
porous media flow, 41, 158, 246, 360 reference temperature, 39, 138, 140, 163,
porthole file, 403 175, 220, 239
Prandtl number, 40, 51 Reichardt law, 55
prescribed density, 201 relative velocity, 32, 71
prescribed electric potential, 334 Relaxation factors in outer iteration, 300
prescribed mass-ratio, 235 restart analysis, 320, 390
prescribed pressure, 92 Reynolds number, 40, 50, 76
prescribed rotational velocity, 93 rigid motion of element groups, 345
prescribed specific discharge, 201 RNG-based subgrid-scale model, 142
prescribed tangential velocity, 102, 211 rotating frame, 30, 33
prescribed temperature, 120 rotation center, 93, 95, 99, 103, 208, 212
prescribed total energy, 204 rotation velocity:, 32
prescribed turbulence K - , 170, 171, 227 rotational velocity condition. See
prescribed variable, 86 prescribed rotational velocity
prescribed velocity, 91 RPBCG iterative solver. See
pressure, 28, 36, 40, 44, 45, 47 solver>RPBCG
pressure datum, 326 RPGMRES iterative solver. See
pressure interpolation, 295 solver>RPGMRES
pressure-dependent condition, 130 S
pressure-dependent material, 221
pressure-temperature-dependent material, SAM, 346
221 saturated porous media, 41
primary unit, 396 Schmidt number, 76
primitive variable, 28, 80, 160, 182, 191, second viscosity, 28, 220
193, 219 ratio, 219
P-T-dependent material, 150 second-order scheme in Segregated
method, 295
R Segregated method, 293
radiation condition, 87, 123 shape factor, 123, 126
radiosity, 87 shear rate. See deformation rate
radiosity equation, 87, 126 shear stress, 48
random motions of fluid, 50, 51 shear stress transport model, 55

Index
shear viscous layer, 75 in-core, 311, 313, 400

shock absorber, 45 multi-grid, 314, 377
shock wave, 47 out-of-core, 311, 313, 400
SIMPLE, 295 RPBCG, 314, 377
SIMPLEC, 295 RPGMRES, 314, 377
singular, 50 SKYLINE, 377
equation, 50 sparse, 311, 377
matrix, 384 sonic surface, 47
skew system, 322 sound speed, 37, 74, 368
slave degree, 323 Spalart-Allmaras model, 57
Sliding-mesh interface, 111 sparse solver. See solver>sparse
slightly compressible flow, 37, 45, 77, 80, special boundary conditions, 47, 86, 96,
246, 360 122, 205
Smagorinsky-Lilly subgrid-scale model, species coupled with fluid, 234
142 specific heat, 138, 139, 140, 175, 219,
Soave Redlich Kwong model, 156 220, 221
Soave Redlich Kwong model, 48 at constant pressure, 47
solid density, 108 at constant volume, 30, 47
solid elements, 261 fluid, 44, 163
2D- potential-based fluid, 261 solid, 44, 163
2D-displacement-based fluid, 261 specific internal energy, 28
2D-solid, 261 specific kinetic energy, 28
3D- potential-based fluid, 261 specific rate of heat generation, 27, 138,
3D-displacement-based fluid, 261 139, 140, 163, 175, 219, 220, 221
3D-solid, 261 specific total energy, 27, 33
beam, 261, 263, 264 specific total energy in rotating frame, 33
iso-beam, 261, 265 specular group, 126
pipe, 265 specular radiation, 87, 124
pipe, 261 specular reflection, 124
plate, 261 stability, 41, 81, 83, 130, 188, 189, 240,
shell, 261 253, 318, 319, 367, 387
truss/cable, 261, 265 stagnation point, 50
solid equation, 252, 253 standard Redlich Kwong model, 48
solid model, 104, 249, 252, 253 standard Redlich Kwong model, 156
solid velocity, 55, 71 standard TFSI, 273
solid wall, 172, 228 state equation, 28
solution increment. See solution vector compressible flow, 47
solution vector, 128, 160, 253, 292 incompressible flow, 37
fluid, 253, 255, 258 slightly compressible flow, 45
solid, 253, 255, 258 steady-state, 50, 68, 80
solver, 377 steady-state analysis, 364
COLSOL, 310 steered adaptive mesh, 346
explicit. See explit time integration Stefan-Boltzmann constant, 123, 125

Index
stiffness matrix, 310, 318 displacement, 254, 256, 270

FSI, 256 equilibrium, 366, 379
stopping criteria in Segregated method, iterative solver, 316
297 steady-state, 293, 366
storage. See memory stress, 254, 256, 257, 270
stress, 27, 33, 35, 38, 54, 82 traction equilibrium, 248
stress criterion, 254 transient analysis, 80, 364
structural model. See solid model transmisivity, 124
subsonic at inlet, 216 trapezoidal rule, 81
subsonic at outlet, 217 turbulence, 51
successive substitution method, 160 dissipation rate, 51, 55
supersonic at inlet, 215 empirical law, 55
supersonic at outlet, 217 fluctuating velocity, 52
surface tension, 76, 105, 106, 138, 139, kinetic energy, 51
140, 163, 175 universal law, 55
symmetric condition, 218 viscous sub-layer, 55
system unit, 72, 393, 396 wall function, 55
dimensionless. See nondimension turbulence model, 50, 51, 165, 224
T
K - high-Reynolds number model,
53
tangential velocity, 55 K - low-Reynolds number model, 54
temperature, 28, 47 K - model, 51, 52, 54, 55
temperature boundary condition, 120 K - RNG model, 52
temperature file, 403
large-eddy-simulation model, 51
temperature variation, 39
turbulence model constants, 176, 177, 230
temperature-dependent boundary
two-layer zonal turbulence model, 59
condition, 122, 123, 125
temperature-dependent material, 39, 68, U
88, 140, 161, 220
unfactorized diagonal, 386
temporary file, 403
uniform flow condition, 98
TFSI, 271, 273
upwinding, 84
thermal expansion coefficient, 39, 138,
finite element approach, 85
139, 140, 163, 175
finite volume approach, 85
thermal-mechanical interaction, 273
multidimensions, 85
time, 27, 30, 69, 80
USCS unit, 396
time integration, 80
User-modified-materials, 151
time step size, 367
user-supplied boundary condition. See
time-dependent boundary condition, 91,
boundary condition>user-supplied
92, 96, 122, 123, 125, 170, 171, 201,
user-supplied material. See material
227, 235, 236, 320
model>user-supplied
time-dependent material, 80, 96
usual boundary condition, 86, 91, 120,
tolerance
201
coincident checking, 265, 266, 267

Index
V VOF method, 342

angle condition, 343
vapor fraction, 336
equations, 342
vapor-liquid phase change, 336
loading, 345
variation of internal energy, 30
materials, 342
variational form, 82
surface tension, 343
velocity, 27
Von Karman constant, 55
velocity of moving frame, 31
velocity of species, 62 W
velocity strain tensor, 28
wave speed, 36, 37, 50
velocity-pressure coupling, 295
weak form of equation, 82
vent boundary condition, 115
view factor, 87 Z
view factor matrix, 126
virtual pressure, 82 zero flux of turbulence, 172, 228
virtual temperature, 82 zero pressure, 93
zero specific discharge, 201
virtual velocity, 82
viscosity, 28, 38, 49, 51, 75, 138, 139, zero temperature, 121
140, 141, 143, 163, 175, 219, 220, 221 zero variable, 86
zero velocity, 92
laminar, 51, 55
turbulent, 51, 52, 53, 54 ZGB model, 340
viscous force, 74, 75, 76

TMG-F 92

Uploaded by

Copyright:

Available Formats

TMG-F 92

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

TMG-F 92

Uploaded by

Copyright:

Available Formats

UTOMATIC

Theory and Modeling Guide

ADINA 9.2 July 2016

ADINA R & D, Inc.

ADINA R&D, Inc.

tel. (617) 926-5199

The information contained in this document is subject to change without notice.

ADINA is a registered trademark of K.J. Bathe /ADINA R & D, Inc.

ADINA R & D, Inc. 1987-2016

Chapter 1 Introduction .........................................................................................23

Chapter 2 Various forms of fluid equations ........................................................27

Chapter 3 Formulation of incompressible, slightly compressible and low-speed

ADINA R & D, Inc. 3

3.4.1 General descriptions................................................................................88

Chapter 4 Formulation of fluid flows through porous media..........................158

Chapter 5 Formulation of turbulence in incompressible, slightly compressible

4 ADINA CFD & FSI Theory and Modeling Guide

5.3 Elements ........................................................................................................169

Chapter 6 Formulation of high-speed compressible flows ...............................180

ADINA R & D, Inc. 5

6.6.1 Constant material model .......................................................................219

Chapter 7 Formulation of turbulence in high-speed compressible flows .......224

Chapter 8 Formulation of mass transfer in fluid flows ....................................232

Chapter 9 Formulation of fluid-structure interactions ....................................246

6 ADINA CFD & FSI Theory and Modeling Guide

9.2 Theory ...........................................................................................................247

Chapter 10 Elements .............................................................................................276

Chapter 11 Solutions of nonlinear equations ......................................................292

ADINA R & D, Inc. 7

11.3 Solvers of algebraic equations.....................................................................310

Chapter 12 Other capabilities...............................................................................318

8 ADINA CFD & FSI Theory and Modeling Guide

12.15.3 Criteria for selection of preferable element sizes..............................353

Chapter 13 Other topics ........................................................................................360

Appendix-1 List of figures .....................................................................................408

ADINA R & D, Inc. 9

Appendix-2 List of tables .......................................................................................413

Appendix-3 List of error/warning messages ........................................................415

10 ADINA CFD & FSI Theory and Modeling Guide

Typical units used here are:

Scales and reference datums used here in nondimensional quantities are:

ADINA R & D, Inc. 11

12 ADINA CFD & FSI Theory and Modeling Guide

Er used as specific total m 2 s 2 Er V

ADINA R & D, Inc. 13

f gB gravitational force per m3

k heat conductivity [ W m-K ] k ( DV LC )

Le D C k , Lewis number [.] Le

14 ADINA CFD & FSI Theory and Modeling Guide

number of blocks while

ADINA R & D, Inc. 15

first local coordinate in

16 ADINA CFD & FSI Theory and Modeling Guide

ADINA R & D, Inc. 17

18 ADINA CFD & FSI Theory and Modeling Guide

ADINA R & D, Inc. 19

CFL CFL number [.] CFL

20 ADINA CFD & FSI Theory and Modeling Guide

specific dissipation rate