Partial Differential Equation Toolbox

Users Guide
R2013b
How to Contact MathWorks
www.mathworks.com Web
comp.soft-sys.matlab Newsgroup
www.mathworks.com/contact_TS.html Technical Support
suggest@mathworks.com Product enhancement suggestions
bugs@mathworks.com Bug reports
doc@mathworks.com Documentation error reports
service@mathworks.com Order status, license renewals, passcodes
info@mathworks.com Sales, pricing, and general information
508-647-7000 (Phone)
508-647-7001 (Fax)
The MathWorks, Inc.
3 Apple Hill Drive
Natick, MA 01760-2098
For contact information about worldwide offices, see the MathWorks Web site.
Partial Differential Equation Toolbox Users Guide
COPYRIGHT 19952013 by The MathWorks, Inc.
The software described in this document is furnished under a license agreement. The software may be used
or copied only under the terms of the license agreement. No part of this manual may be photocopied or
reproduced in any form without prior written consent from The MathWorks, Inc.
FEDERAL ACQUISITION: This provision applies to all acquisitions of the Program and Documentation
by, for, or through the federal government of the United States. By accepting delivery of the Program
or Documentation, the government hereby agrees that this software or documentation qualifies as
commercial computer software or commercial computer software documentation as such terms are used
or defined in FAR 12.212, DFARS Part 227.72, and DFARS 252.227-7014. Accordingly, the terms and
conditions of this Agreement and only those rights specified in this Agreement, shall pertain to and govern
the use, modification, reproduction, release, performance, display, and disclosure of the Program and
Documentation by the federal government (or other entity acquiring for or through the federal government)
and shall supersede any conflicting contractual terms or conditions. If this License fails to meet the
governments needs or is inconsistent in any respect with federal procurement law, the government agrees
to return the Program and Documentation, unused, to The MathWorks, Inc.
Trademarks
MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See
www.mathworks.com/trademarks for a list of additional trademarks. Other product or brand
names may be trademarks or registered trademarks of their respective holders.
Patents
MathWorks products are protected by one or more U.S. patents. Please see
www.mathworks.com/patents for more information.
Revision History
August 1995 First printing New for Version 1.0
February 1996 Second printing Revised for Version 1.0.1
July 2002 Online only Revised for Version 1.0.4 (Release 13)
September 2002 Third printing Minor Revision for Version 1.0.4
June 2004 Online only Revised for Version 1.0.5 (Release 14)
October 2004 Online only Revised for Version 1.0.6 (Release 14SP1)
March 2005 Online only Revised for Version 1.0.6 (Release 14SP2)
August 2005 Fourth printing Minor Revision for Version 1.0.6
September 2005 Online only Revised for Version 1.0.7 (Release 14SP3)
March 2006 Online only Revised for Version 1.0.8 (Release 2006a)
March 2007 Online only Revised for Version 1.0.10 (Release 2007a)
September 2007 Online only Revised for Version 1.0.11 (Release 2007b)
March 2008 Online only Revised for Version 1.0.12 (Release 2008a)
October 2008 Online only Revised for Version 1.0.13 (Release 2008b)
March 2009 Online only Revised for Version 1.0.14 (Release 2009a)
September 2009 Online only Revised for Version 1.0.15 (Release 2009b)
March 2010 Online only Revised for Version 1.0.16 (Release 2010a)
September 2010 Online only Revised for Version 1.0.17 (Release 2010b)
April 2011 Online only Revised for Version 1.0.18 (Release 2011a)
September 2011 Online only Revised for Version 1.0.19 (Release 2011b)
March 2012 Online only Revised for Version 1.0.20 (Release 2012a)
September 2012 Online only Revised for Version 1.1 (Release 2012b)
March 2013 Online only Revised for Version 1.2 (Release 2013a)
September 2013 Online only Revised for Version 1.3 (Release 2013b)
Contents
Getting Started
1
Partial Differential Equation Toolbox Product
Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2
Key Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2
Prerequisite Knowledge for Using This Toolbox . . . . . . 1-3
Types of PDE Problems You Can Solve . . . . . . . . . . . . . . 1-4
Common Toolbox Applications . . . . . . . . . . . . . . . . . . . . . . 1-7
Typical Steps to Solve PDEs . . . . . . . . . . . . . . . . . . . . . . . . 1-9
Visualize and Animate Solutions . . . . . . . . . . . . . . . . . . . . 1-11
Poissons Equation with Complex 2-D Geometry . . . . . . 1-12
PDE App Shortcuts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-28
Solving 3-D Problems Using 2-D Models . . . . . . . . . . . . . . 1-31
Finite Element Method (FEM) Basics . . . . . . . . . . . . . . . . 1-32
Setting Up Your PDE
2
Open the PDE App . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-3
v
Specify Geometry Using a CSG Model . . . . . . . . . . . . . . . 2-5
Select Graphical Objects Representing Your
Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-7
Rounded Corners Using CSG Modeling . . . . . . . . . . . . . . 2-8
Enter Parameter Values as MATLAB Expressions . . . . 2-12
Systems of PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-13
Scalar PDE Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-14
Scalar PDE Coefficients in String Form . . . . . . . . . . . . . 2-16
Coefficients for Scalar PDEs in PDE App . . . . . . . . . . . . 2-19
Scalar PDE Coefficients in Function Form . . . . . . . . . . . 2-23
Coefficients as the Result of a Program . . . . . . . . . . . . . . . . 2-23
Calculate Coefficients in Function Form . . . . . . . . . . . . . . . 2-24
Scalar PDE Functional Form and Calling Syntax . . . . . 2-26
Enter Coefficients in the PDE App . . . . . . . . . . . . . . . . . . 2-32
Coefficients for Systems of PDEs . . . . . . . . . . . . . . . . . . . . 2-42
2-D Systems in the PDE App . . . . . . . . . . . . . . . . . . . . . . . . 2-44
f for Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-48
c for Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-50
c as Tensor, Matrix, and Vector . . . . . . . . . . . . . . . . . . . . . . 2-50
Scalar c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-52
Two-Element Column Vector c . . . . . . . . . . . . . . . . . . . . . . . 2-52
Three-Element Column Vector c . . . . . . . . . . . . . . . . . . . . . 2-53
Four-Element Column Vector c . . . . . . . . . . . . . . . . . . . . . . 2-53
vi Contents
N-Element Column Vector c . . . . . . . . . . . . . . . . . . . . . . . . . 2-54
2N-Element Column Vector c . . . . . . . . . . . . . . . . . . . . . . . . 2-55
3N-Element Column Vector c . . . . . . . . . . . . . . . . . . . . . . . . 2-56
4N-Element Column Vector c . . . . . . . . . . . . . . . . . . . . . . . . 2-57
2N(2N+1)/2-Element Column Vector c . . . . . . . . . . . . . . . . 2-57
4N
2
-Element Column Vector c . . . . . . . . . . . . . . . . . . . . . . . 2-58
a or d for Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-59
Coefficients a or d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-59
Scalar a or d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-60
N-Element Column Vector a or d . . . . . . . . . . . . . . . . . . . . . 2-60
N(N+1)/2-Element Column Vector a or d . . . . . . . . . . . . . . . 2-60
N
2
-Element Column Vector a or d . . . . . . . . . . . . . . . . . . . . 2-61
Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-62
Types of Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . 2-65
No Boundary Conditions Between Subdomains . . . . . . 2-66
Identify Boundary Labels . . . . . . . . . . . . . . . . . . . . . . . . . . 2-69
Boundary Conditions Overview . . . . . . . . . . . . . . . . . . . . . 2-71
Boundary Conditions for Scalar PDE . . . . . . . . . . . . . . . . 2-72
Boundary Conditions for PDE Systems . . . . . . . . . . . . . . 2-77
Tooltip Displays for Mesh and Plots . . . . . . . . . . . . . . . . . 2-84
Mesh Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-85
Adaptive Mesh Refinement . . . . . . . . . . . . . . . . . . . . . . . . . 2-86
Improving Solution Accuracy Using Mesh Refinement . . . 2-86
Error Estimate for the FEM Solution . . . . . . . . . . . . . . . . . 2-87
Mesh Refinement Functions . . . . . . . . . . . . . . . . . . . . . . . . . 2-88
Mesh Refinement Termination Criteria . . . . . . . . . . . . . . . 2-88
vii
Solving PDEs
3
Set Up and Solve Your PDE Problem . . . . . . . . . . . . . . . . 3-2
Structural Mechanics Plane Stress . . . . . . . . . . . . . . . . 3-6
Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-9
Using the PDE App . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-9
Structural Mechanics Plane Strain . . . . . . . . . . . . . . . 3-13
Clamped, Square Isotropic Plate With a Uniform
Pressure Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-14
Deflection of a Piezoelectric Actuator . . . . . . . . . . . . . . . 3-19
Electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-34
Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-35
Using the PDE App . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-35
Magnetostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-37
Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-39
Using the PDE App . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-40
AC Power Electromagnetics . . . . . . . . . . . . . . . . . . . . . . . . 3-44
Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-46
Using the PDE App . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-47
Conductive Media DC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-50
Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-50
Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-57
Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-58
Using the PDE App . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-58
Nonlinear Heat Transfer In a Thin Plate . . . . . . . . . . . . . 3-60
Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-71
viii Contents
Elliptic PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-72
Solve Poissons Equation on a Unit Disk . . . . . . . . . . . . . . . 3-72
Scattering Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-76
Minimal Surface Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-81
Domain Decomposition Problem . . . . . . . . . . . . . . . . . . . . . 3-83
Parabolic PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-88
Heat Equation for Metal Block with Cavity . . . . . . . . . . . . 3-88
Heat Distribution in a Radioactive Rod . . . . . . . . . . . . . . . . 3-92
Hyperbolic PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-95
Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-95
Eigenvalue Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-100
Eigenvalues and Eigenfunctions for the L-Shaped
Membrane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-100
L-Shaped Membrane with a Rounded Corner . . . . . . . . . . . 3-104
Eigenvalues and Eigenmodes of a Square . . . . . . . . . . . . . . 3-106
Vibration Of a Circular Membrane Using The MATLAB
eigs Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-109
Solve PDEs Programmatically . . . . . . . . . . . . . . . . . . . . . . 3-113
When You Need Programmatic Solutions . . . . . . . . . . . . . . 3-113
Data Structures in Partial Differential Equation
Toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-113
Tips for Solving PDEs Programmatically . . . . . . . . . . . . . . 3-117
Solve Poissons Equation on a Grid . . . . . . . . . . . . . . . . . . 3-119
PDE App
4
PDE App Menus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-2
File Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-4
New . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-4
ix
Open . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
Save As . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
Export Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-5
Print . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-6
Edit Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-7
Paste . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-8
Options Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-9
Grid Spacing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-10
Axes Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-11
Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-11
Draw Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-13
Rotate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-14
Boundary Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-15
Specify Boundary Conditions in the PDE App . . . . . . . . . . 4-16
PDE Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-19
PDE Specification in the PDE App . . . . . . . . . . . . . . . . . . . 4-20
Mesh Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-21
Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-23
Solve Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-25
Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-26
Plot Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-31
Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-31
Window Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-38
Help Menu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-39
x Contents
Finite Element Method
5
Elliptic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-2
Systems of PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-10
Parabolic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-13
Reducing Parabolic Equations to Elliptic Equations . . . . . 5-13
Solve a Parabolic Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 5-15
Hyperbolic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-18
Eigenvalue Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-19
Nonlinear Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-24
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-30
Functions Alphabetical List
6
Index
xi
xii Contents
1
Getting Started
Partial Differential Equation Toolbox Product Description on page 1-2
Prerequisite Knowledge for Using This Toolbox on page 1-3
Types of PDE Problems You Can Solve on page 1-4
Common Toolbox Applications on page 1-7
Typical Steps to Solve PDEs on page 1-9
Visualize and Animate Solutions on page 1-11
Poissons Equation with Complex 2-D Geometry on page 1-12
PDE App Shortcuts on page 1-28
Solving 3-D Problems Using 2-D Models on page 1-31
Finite Element Method (FEM) Basics on page 1-32
1 Getting Started
Partial Differential Equation Toolbox Product Description
Solve partial differential equations using finite element methods
The Partial Differential Equation Toolbox product contains tools for the
study and solution of partial differential equations (PDEs) in two space
dimensions (2-D) and time. A PDE app and functions let you preprocess,
solve, and postprocess generic 2-D PDEs for a broad range of engineering and
science applications.
Key Features
PDE app for pre- and post-processing 2-D PDEs
Automatic and adaptive meshing
Geometry creation using constructive solid geometry (CSG) paradigm
Boundary condition specification: Dirichlet, generalized Neumann, and
mixed
Flexible coefficient and PDE problem specification using MATLAB

syntax
Fully automated mesh generation and refinement
Nonlinear and adaptive solvers that handle systems with multiple
dependent variables
Simultaneous visualization of multiple solution properties, FEM-mesh
overlays, and animation
1-2
Prerequisite Knowledge for Using This Toolbox
Prerequisite Knowledge for Using This Toolbox
Partial Differential Equation Toolbox software is designed for both beginners
and advanced users.
The minimal requirement is that you can formulate a PDE problem on paper
(draw the domain, write the boundary conditions, and the PDE). At the
MATLAB command line, type
pdetool
This invokes the PDE app, which is a self-contained graphical environment
for PDE solving. For common applications you can use the specific physical
terms rather than abstract coefficients.
Using the PDE app requires no knowledge of the mathematics behind the
PDE, the numerical schemes, or MATLAB. Poissons Equation with Complex
2-D Geometry on page 1-12 guides you through an example step by step.
Advanced applications are also possible by downloading the domain geometry,
boundary conditions, and mesh description to the MATLAB workspace. You
can use functions to, for example, generate meshes, discretize your problem,
interpolate, and plot data on unstructured grids.
1-3
1 Getting Started
Types of PDE Problems You Can Solve
This toolbox applies to the following PDE type:
V V
( )
+ = c u au f ,
expressed in , which we shall refer to as the elliptic equation, regardless of
whether its coefficients and boundary conditions make the PDE problem
elliptic in the mathematical sense. Analogously, we shall use the terms
parabolic equation and hyperbolic equation for equations with spatial
operators like the previous one, and first and second order time derivatives,
respectively. is a bounded domain in the plane. c, a, f, and the unknown u
are scalar, complex valued functions defined on . c can be a 2-by-2 matrix
function on . The software can also handle the parabolic PDE
d c au f
u
t
u
c
c
V V
( )
+ = ,
the hyperbolic PDE
d
u
t
c u au f
c
c
V V
( )
+ =
2
2
,
and the eigenvalue problem
V V
( )
+ = c u au du ,
where d is a complex valued function on , and is an unknown eigenvalue.
For the parabolic and hyperbolic PDE the coefficients c, a, f, and d can
depend on time, on the solution u, and on its gradient u. A nonlinear solver
(pdenonlin) is available for the nonlinear elliptic PDE
V V ( ) + = c u u a u u f u ( ) ( ) ( ),
where c, a, and f are functions of the unknown solution u and on its gradient
u. The parabolic and hyperbolic equation solvers also solve nonlinear and
time-dependent problems.
1-4
Types of PDE Problems You Can Solve
Note Before solving a nonlinear elliptic PDE, from the Solve menu in the
PDE app, select Parameters. Then, select the Use nonlinear solver check
box and click OK.
All solvers can handle the system case of N coupled equations. You can solve
N = 1 or 2 equations using the PDE app, and any number of equations using
command-line functions. For example, N = 2 elliptic equations:


V V ( ) V V ( ) + + =
V V ( ) V V (


c u c u a u a u f
c u c u
11 1 12 2 11 1 12 2 1
21 1 22 2
)) + + = a u a u f
21 1 22 2 2
.
For a description of N > 1 PDE systems and their coefficients, see Coefficients
for Systems of PDEs on page 2-42.
For the elliptic problem, an adaptive mesh refinement algorithm is
implemented. It can also be used in conjunction with the nonlinear solver. In
addition, a fast solver for Poissons equation on a rectangular grid is available.
The following boundary conditions are defined for scalar u:
Dirichlet: hu = r on the boundary .
Generalized Neumann:

n c u qu g V
( )
+ = on .

n is the outward unit normal. g, q, h, and r are complex-valued functions

defined on . (The eigenvalue problem is a homogeneous problem, i.e., g= 0,
r = 0.) In the nonlinear case, the coefficients g, q, h, and r can depend on u,
and for the hyperbolic and parabolic PDE, the coefficients can depend on time.
For the two-dimensional system case, Dirichlet boundary condition is
h u h u r
h u h u r
11 1 12 2 1
21 1 22 2 2
+ =
+ = ,
the generalized Neumann boundary condition is
1-5
1 Getting Started


n c u n c u q u q u g
n c u n c u


11 1 12 2 11 1 12 2 1
21 1 22
V ( ) V ( ) + + =
V ( ) + V
+
22 21 1 22 2 2
( ) + + = q u q u g .
and the mixed boundary condition is
h u h u r
h n c u n c u q u q u g
11 1 12 2 1
11 11 1 12 2 11 1 12 2 1
+ =
+ V ( ) V ( ) + + = +

nn c u n c u q u q u g h ,
21 1 22 2 21 1 22 2 2 12
V ( ) + V ( ) + + = +

where is computed such that the Dirichlet boundary condition is satisfied.

Dirichlet boundary conditions are also called essential boundary conditions,
and Neumann boundary conditions are also called natural boundary
conditions.
For advanced, nonstandard applications you can transfer the description of
domains, boundary conditions etc. to your MATLAB workspace. From there
you use Partial Differential Equation Toolbox functions for managing data
on unstructured meshes. You have full access to the mesh generators, FEM
discretizations of the PDE and boundary conditions, interpolation functions,
etc. You can design your own solvers or use FEM to solve subproblems of more
complex algorithms. See also Solve PDEs Programmatically on page 3-113.
1-6
Common Toolbox Applications
Common Toolbox Applications
Elliptic and parabolic equations are used for modeling:
Steady and unsteady heat transfer in solids
Flows in porous media and diffusion problems
Electrostatics of dielectric and conductive media
Potential flow
Steady state of wave equations
Hyperbolic equation is used for:
Transient and harmonic wave propagation in acoustics and
electromagnetics
Transverse motions of membranes
Eigenvalue problems are used for:
Determining natural vibration states in membranes and structural
mechanics problems
In addition to solving generic scalar PDEs and generic systems of PDEs
with vector valued u, Partial Differential Equation Toolbox provides tools
for solving PDEs that occur in these common applications in engineering
and science:
Structural Mechanics Plane Stress on page 3-6
Structural Mechanics Plane Strain on page 3-13
Electrostatics on page 3-34
Magnetostatics on page 3-37
AC Power Electromagnetics on page 3-44
Conductive Media DC on page 3-50
Heat Transfer on page 3-57
Diffusion on page 3-71
1-7
1 Getting Started
The PDE app lets you specify PDE coefficients and boundary conditions in
terms of physical entities. For example, you can specify Youngs modulus
in structural mechanics problems.
The application mode can be selected directly from the pop-up menu in the
upper right part of the PDE app or by selecting an application from the
Application submenu in the Options menu. Changing the application
resets all PDE coefficients and boundary conditions to the default values for
that specific application mode.
When using an application mode, the generic PDE coefficients are replaced
by application-specific parameters such as Youngs modulus for problems in
structural mechanics. The application-specific parameters are entered by
selecting Parameters from the PDE menu or by clicking the PDE button.
You can also access the PDE parameters by double-clicking a subdomain, if
you are in the PDE mode. That way it is possible to define PDE parameters
for problems with regions of different material properties. The Boundary
condition dialog box is also altered so that the Description column reflects the
physical meaning of the different boundary condition coefficients. Finally,
the Plot Selection dialog box allows you to visualize the relevant physical
variables for the selected application.
Note In the User entry options in the Plot Selection dialog box, the solution
and its derivatives are always referred to as u, ux, and uy (v, vx, and vy for
the system cases) even if the application mode is nongeneric and the solution
of the application-specific PDE normally is named, e.g., V or T.
The PDE app lets you solve problems with vector valued u of dimension two.
However, you can use functions to solve problems for any dimension of u.
1-8
Typical Steps to Solve PDEs
Typical Steps to Solve PDEs
Partial Differential Equation Toolbox provides the PDE app that you can
use to:
1 Define the 2-D geometry.
You create , the geometry, using the constructive solid geometry (CSG)
model paradigm. A set of solid objects (rectangle, circle, ellipse, and
polygon) is provided. You can combine these objects using set formulas.
2 Define the boundary conditions.
You can have different types of boundary conditions on different boundary
segments. See Types of Boundary Conditions on page 2-65.
3 Define the PDE coefficients. See Scalar PDE Coefficients on page 2-14
and Coefficients for Systems of PDEs on page 2-42.
You interactively specify the type of PDE and the coefficients c, a, f, and d.
You can specify the coefficients for each subdomain independently. This
may ease the specification of, e.g., various material properties in a PDE
model.
4 Create the triangular mesh.
Generate the mesh to a fineness that adequately resolves the important
features in the geometry, but is coarse enough to run in a reasonable
amount of time and memory.
5 Solve the PDE.
You can invoke and control the nonlinear and adaptive solvers for elliptic
problems. For parabolic and hyperbolic problems, you can specify the
initial values, and the times for which the output should be generated.
For the eigenvalue solver, you can specify the interval in which to search
for eigenvalues.
6 Plot the solution and other physical properties calculated from the solution
(post processing).
1-9
1 Getting Started
After solving a problem, you can return to the mesh mode to further refine
your mesh and then solve again. You can also employ the adaptive mesh
refiner and solver, adaptmesh. This option tries to find a mesh that fits the
solution.
1-10
Visualize and Animate Solutions
Visualize and Animate Solutions
From the PDE app you can use plot mode, where you have a wide range of
visualization possibilities. You can visualize both inside the PDE app and in
separate figures. You can plot three different solution properties at the same
time, using color, height, and vector field plots.
Surface, mesh, contour, and arrow (quiver) plots are available. For surface
plots, you can choose between interpolated and flat rendering schemes. The
mesh may be hidden or exposed in all plot types.
For parabolic and hyperbolic equations, you can even produce an animated
movie of the solutions time dependence. All visualization functions are also
accessible from the command line.
1-11
1 Getting Started
Poissons Equation with Complex 2-D Geometry
This example shows how to solve the Poissons equation, u = f using the
PDE app. This problem requires configuring a 2-D geometry with Dirichlet
and Neumann boundary conditions.
To start the PDE app, type the command pdetool at the MATLAB prompt.
The PDE app looks similar to the following figure, with exception of the grid.
Turn on the grid by selecting Grid from the Options menu. Also, enable
the snap-to-grid feature by selecting Snap from the Options menu. The
snap-to-grid feature simplifies aligning the solid objects.
1-12
Poissons Equation with Complex 2-D Geometry
1-13
1 Getting Started
The first step is to draw the geometry on which you want to solve the PDE.
The PDE app provides four basic types of solid objects: polygons, rectangles,
circles, and ellipses. The objects are used to create a Constructive Solid
Geometry model (CSG model). Each solid object is assigned a unique label,
and by the use of set algebra, the resulting geometry can be made up of a
combination of unions, intersections, and set differences. By default, the
resulting CSG model is the union of all solid objects.
To select a solid object, either click the button with an icon depicting the
solid object that you want to use, or select the object by using the Draw
pull-down menu. In this case, rectangle/square objects are selected. To draw
a rectangle or a square starting at a corner, click the rectangle button without
a + sign in the middle. The button with the + sign is used when you want
to draw starting at the center. Then, put the cursor at the desired corner,
and click-and-drag using the left mouse button to create a rectangle with the
desired side lengths. (Use the right mouse button to create a square.) Click
and drag from (1,.2) to (1,.2). Notice how the snap-to-grid feature forces
the rectangle to line up with the grid. When you release the mouse, the CSG
model is updated and redrawn. At this stage, all you have is a rectangle. It is
assigned the label R1. If you want to move or resize the rectangle, you can
easily do so. Click-and-drag an object to move it, and double-click an object to
open a dialog box, where you can enter exact location coordinates. From the
dialog box, you can also alter the label. If you are not satisfied and want to
restart, you can delete the rectangle by clicking the Delete key or by selecting
Clear from the Edit menu.
Next, draw a circle by clicking the button with the ellipse icon with the + sign,
and then click-and-drag in a similar way, starting near the point (.5,0) with
radius .4, using the right mouse button, starting at the circle center.
1-14
Poissons Equation with Complex 2-D Geometry
1-15
1 Getting Started
The resulting CSG model is the union of the rectangle R1 and the circle C1,
described by set algebra as R1+C1. The area where the two objects overlap
is clearly visible as it is drawn using a darker shade of gray. The object that
you just drewthe circlehas a black border, indicating that it is selected.
A selected object can be moved, resized, copied, and deleted. You can select
more than one object by Shift+clicking the objects that you want to select.
Also, a Select All option is available from the Edit menu.
Finally, add two more objects, a rectangle R2 from (.5,.6) to (1,1), and a
circle C2 centered at (.5,.2) with radius .2. The desired CSG model is formed
by subtracting the circle C2 from the union of the other three objects. You
do this by editing the set formula that by default is the union of all objects:
C1+R1+R2+C2. You can type any other valid set formula into Set formula
edit field. Click in the edit field and use the keyboard to change the set
formula to
(R1+C1+R2)-C2
1-16
Poissons Equation with Complex 2-D Geometry
1-17
1 Getting Started
If you want, you can save this CSG model as a file. Use the Save As option
from the File menu, and enter a filename of your choice. It is good practice
to continue to save your model at regular intervals using Save. All the
additional steps in the process of modeling and solving your PDE are then
saved to the same file. This concludes the drawing part.
You can now define the boundary conditions for the outer boundaries. Enter
the boundary mode by clicking the icon or by selecting Boundary Mode
from the Boundary menu. You can now remove subdomain borders and
define the boundary conditions.
The gray edge segments are subdomain borders induced by the intersections
of the original solid objects. Borders that do not represent borders between,
e.g., areas with differing material properties, can be removed. From the
Boundary menu, select the Remove All Subdomain Borders option. All
borders are then removed from the decomposed geometry.
The boundaries are indicated by colored lines with arrows. The color reflects
the type of boundary condition, and the arrow points toward the end of the
boundary segment. The direction information is provided for the case when
the boundary condition is parameterized along the boundary. The boundary
condition can also be a function of x and y, or simply a constant. By default,
the boundary condition is of Dirichlet type: u = 0 on the boundary.
Dirichlet boundary conditions are indicated by red color. The boundary
conditions can also be of a generalized Neumann (blue) or mixed (green) type.
For scalar u, however, all boundary conditions are either of Dirichlet or the
generalized Neumann type. You select the boundary conditions that you want
to change by clicking to select one boundary segment, by Shift+clicking to
select multiple segments, or by using the Edit menu option Select All to
select all boundary segments. The selected boundary segments are indicated
by black color.
For this problem, change the boundary condition for all the circle arcs. Select
them by using the mouse and Shift+click those boundary segments.
1-18
Poissons Equation with Complex 2-D Geometry
1-19
1 Getting Started
Double-clicking anywhere on the selected boundary segments opens the
Boundary Condition dialog box. Here, you select the type of boundary
condition, and enter the boundary condition as a MATLAB expression.
Change the boundary condition along the selected boundaries to a Neumann
condition, u/n = 5. This means that the solution has a slope of 5 in the
normal direction for these boundary segments.
In the Boundary Condition dialog box, select the Neumann condition type,
and enter -5 in the edit box for the boundary condition parameter g. To define
a pure Neumann condition, leave the q parameter at its default value, 0.
When you click the OK button, notice how the selected boundary segments
change to blue to indicate Neumann boundary condition.
Next, specify the PDE itself through a dialog box that is accessed by clicking
the button with the PDE icon or by selecting PDE Specification from the
PDE menu. In PDE mode, you can also access the PDE Specification dialog
box by double-clicking a subdomain. That way, different subdomains can
have different PDE coefficient values. This problem, however, consists of
only one subdomain.
1-20
Poissons Equation with Complex 2-D Geometry
In the dialog box, you can select the type of PDE (elliptic, parabolic, hyperbolic,
or eigenmodes) and define the applicable coefficients depending on the PDE
type. This problem consists of an elliptic PDE defined by the equation
V V
( )
+ = c u au f ,
with c = 1.0, a = 0.0, and f = 10.0.
Finally, create the triangular mesh that Partial Differential Equation Toolbox
software uses in the Finite Element Method (FEM) to solve the PDE. The
triangular mesh is created and displayed when clicking the button with the
icon or by selecting the Mesh menu option Initialize Mesh. If you want
a more accurate solution, the mesh can be successively refined by clicking the
button with the four triangle icon (the Refine button) or by selecting the
Refine Mesh option from the Mesh menu.
Using the Jiggle Mesh option, the mesh can be jiggled to improve the
triangle quality. Parameters for controlling the jiggling of the mesh, the
refinement method, and other mesh generation parameters can be found in
a dialog box that is opened by selecting Parameters from the Mesh menu.
1-21
1 Getting Started
You can undo any change to the mesh by selecting the Mesh menu option
Undo Mesh Change.
Initialize the mesh, then refine it once and finally jiggle it once.
1-22
Poissons Equation with Complex 2-D Geometry
1-23
1 Getting Started
We are now ready to solve the problem. Click the = button or select Solve
PDE from the Solve menu to solve the PDE. The solution is then plotted. By
default, the plot uses interpolated coloring and a linear color map. A color bar
is also provided to map the different shades to the numerical values of the
solution. If you want, the solution can be exported as a vector to the MATLAB
main workspace.
1-24
Poissons Equation with Complex 2-D Geometry
1-25
1 Getting Started
There are many more plot modes available to help you visualize the solution.
Click the button with the 3-D solution icon or select Parameters from the
Plot menu to access the dialog box for selection of the different plot options.
Several plot styles are available, and the solution can be plotted in the PDE
app or in a separate figure as a 3-D plot.
Now, select a plot where the color and the height both represent u. Choose
interpolated shading and use the continuous (interpolated) height option. The
default colormap is the cool colormap; a pop-up menu lets you select from
a number of different colormaps. Finally, click the Plot button to plot the
solution; click the Close button to save the plot setup as the current default.
The solution is plotted as a 3-D plot in a separate figure window.
1-26
Poissons Equation with Complex 2-D Geometry
The following solution plot is the result. You can use the mouse to rotate
the plot in 3-D. By clicking-and-dragging the axes, the angle from which the
solution is viewed can be changed.
1-27
1 Getting Started
PDE App Shortcuts
PDE app toolbar provide quick access to key operations that are also available
in the menus.
The toolbar consists of three different parts: the five leftmost buttons for draw
mode functions, the next six buttons for different boundary, mesh, solution,
and plot functions, and the rightmost button for activating the zoom feature.
Five buttons on the left let you draw the geometry. Double-click a button
makes it stick, and you can then continue to draw solid objects of the
selected type until you single-click the button to release it.
In draw mode, you can create the 2-D geometry using the constructive solid
geometry (CSG) model paradigm. A set of solid objects (rectangle, circle,
ellipse, and polygon) is provided. These objects can be combined using set
formulas in a flexible way.
Draw a rectangle/square starting at a corner.
Using the left mouse button, click-and-drag to create a
rectangle. Using the right mouse button (or Ctrl+click),
click-and-drag to create a square.
Draw a rectangle/square starting at the center.
Using the left mouse button, click-and-drag to create a
rectangle. Using the right mouse button (or Ctrl+click),
click-and-drag to create a square.
Draw an ellipse/circle starting at the perimeter.
Using the left mouse button, click-and-drag to create an
ellipse. Using the right mouse button (or Ctrl+click),
click-and-drag to create a circle.
1-28
PDE App Shortcuts
Draw an ellipse/circle starting at the center.
Using the left mouse button, click-and-drag to create an
ellipse. Using the right mouse button (or Ctrl+click),
click-and-drag to create a circle.
Draw a polygon. Click-and-drag to create polygon edges. You
can close the polygon by pressing the right mouse button.
Clicking at the starting vertex also closes the polygon.
The remaining buttons represent, from left to right:
Enters the boundary mode.
In boundary mode, you can specify the boundary conditions.
You can have different types of boundary conditions on
different boundaries. In this mode, the original shapes of the
solid objects constitute borders between subdomains of the
model. Such borders can be eliminated in this mode.
Opens the PDE Specification dialog box.
In PDE mode, you can interactively specify the type of PDE
problem, and the PDE coefficients. You can specify the
coefficients for each subdomain independently. This makes
it easy to specify, e.g., various material properties in a PDE
model.
Initializes the triangular mesh
In mesh mode, you can control the automated mesh generation
and plot the mesh.
Refines the triangular mesh.
Solves the PDE.
In solve mode, you can invoke and control the nonlinear
and adaptive solver for elliptic problems. For parabolic and
hyperbolic PDE problems, you can specify the initial values,
and the times for which the output should be generated. For
the eigenvalue solver, you can specify the interval in which to
search for eigenvalues.
1-29
1 Getting Started
3-D solution opens the Plot Selection dialog box.
In plot mode, there is a wide range of visualization
possibilities. You can visualize both in the PDE app and in
a separate figure window. You can visualize three different
solution properties at the same time, using color, height,
and vector field plots. There are surface, mesh, contour, and
arrow (quiver) plots available. For parabolic and hyperbolic
equations, you can animate the solution as it changes with
time.
Toggles zoom.
1-30
Solving 3-D Problems Using 2-D Models
Solving 3-D Problems Using 2-D Models
Partial Differential Equation Toolbox software solves problems in two space
dimensions and time, whereas reality has three space dimensions. The
reduction to 2-D is possible when variations in the third space dimension
(taken to be z) can be accounted for in the 2-D equation. In some cases, like
the plane stress analysis, the material parameters must be modified in the
process of dimensionality reduction.
When the problem is such that variation with z is negligible, all z-derivatives
drop out and the 2-D equation has exactly the same units and coefficients
as in 3-D.
Slab geometries are treated by integration through the thickness. The result
is a 2-D equation for the z-averaged solution with the thickness, say D(x,y),
multiplied onto all the PDE coefficients, c, a, d, and f, etc. For instance, if
you want to compute the stresses in a sheet welded together from plates of
different thickness, multiply Youngs modulus E, volume forces, and specified
surface tractions by D(x,y), Similar definitions of the equation coefficients are
called for in other slab geometry examples and application modes.
1-31
1 Getting Started
Finite Element Method (FEM) Basics
The core Partial Differential Equation Toolbox algorithm is a PDE solver
that uses the Finite Element Method (FEM) for problems defined on bounded
domains in the plane.
The solutions of simple PDEs on complicated geometries can rarely be
expressed in terms of elementary functions. You are confronted with two
problems: First you need to describe a complicated geometry and generate a
mesh on it. Then you need to discretize your PDE on the mesh and build an
equation for the discrete approximation of the solution. The PDE app provides
you with easy-to-use graphical tools to describe complicated domains and
generate triangular meshes. It also discretizes PDEs, finds discrete solutions
and plots results. You can access the mesh structures and the discretization
functions directly from the command line (or from a file) and incorporate
them into specialized applications.
Here is an overview of the Finite Element Method (FEM). The purpose of
this presentation is to get you acquainted with the elementary FEM notions.
Here you find the precise equations that are solved and the nature of the
discrete solution. Different extensions of the basic equation implemented in
Partial Differential Equation Toolbox software are presented. A more detailed
description can be found in Elliptic Equations on page 5-2, with variants for
specific types in Systems of PDEs on page 5-10, Parabolic Equations on
page 5-13, Hyperbolic Equations on page 5-18, Eigenvalue Equations on
page 5-19, and Nonlinear Equations on page 5-24.
You start by approximating the computational domain with a union of
simple geometric objects, in this case triangles. The triangles form a mesh and
each vertex is called a node. You are in the situation of an architect designing
a dome. He has to strike a balance between the ideal rounded forms of the
original sketch and the limitations of his simple building-blocks, triangles or
quadrilaterals. If the result does not look close enough to a perfect dome, the
architect can always improve his work using smaller blocks.
Next you say that your solution should be simple on each triangle.
Polynomials are a good choice: they are easy to evaluate and have good
approximation properties on small domains. You can ask that the solutions in
neighboring triangles connect to each other continuously across the edges.
You can still decide how complicated the polynomials can be. Just like an
1-32
Finite Element Method (FEM) Basics
architect, you want them as simple as possible. Constants are the simplest
choice but you cannot match values on neighboring triangles. Linear functions
come next. This is like using flat tiles to build a waterproof dome, which
is perfectly possible.
A Triangular Mesh (left) and a Continuous Piecewise Linear Function on That Mesh
Now you use the basic elliptic equation (expressed in )
V V
( )
+ = c u au f ,
If u
h
is the piecewise linear approximation to u, it is not clear what the second
derivative term means. Inside each triangle, u
h
is a constant (because u
h
is
flat) and thus the second-order term vanishes. At the edges of the triangles,
cu
h
is in general discontinuous and a further derivative makes no sense.
What you are looking for is the best approximation of u in the class of
continuous piecewise polynomials. Therefore you test the equation for u
h
against all possible functions v of that class. Testing means formally to
multiply the residual against any function and then integrate, i.e., determine
u
h
such that
1-33
1 Getting Started

( )
= V V ( ) +
}
c u au f vdx
h h
O
0
for all possible v. The functions v are usually called test functions.
Partial integration (Greens formula) yields that u
h
should satisfy
c u v au v c u dx n vds fvdx v
h h h
V ( ) V + V ( ) ( )
=
} } }
c
,
O O O

where is the boundary of and

n is the outward pointing normal on
. The integrals of this formulation are well-defined even if u
h
and v are
piecewise linear functions.
Boundary conditions are included in the following way. If u
h
is known at some
boundary points (Dirichlet boundary conditions), we restrict the test functions
to v = 0 at those points, and require u
h
to attain the desired value at that
point. At all the other points we ask for Neumann boundary conditions, i.e.,
c u n qu g
h h
( ) + =

. The FEM formulation reads: Find u

h
such that
c u v au v dx qu vds gvds f vdx v
h h h
( ) +
( )
+ = +


1 1
,
where
1
is the part of the boundary with Neumann conditions. The test
functions v must be zero on
1
.
Any continuous piecewise linear u
h
is represented as a combination
u x U x
h i i
i
N
( ) ( ), =
=

|
1
where
i
are some special piecewise linear basis functions and U
i
are scalar
coefficients. Choose
i
like a tent, such that it has the height 1 at the node
i and the height 0 at all other nodes. For any fixed v, the FEM formulation
yields an algebraic equation in the unknowns U
i
. You want to determine N
unknowns, so you need N different instances of v. What better candidates
than v =
i
, i = 1, 2, ... , N? You find a linear system KU = F where the matrix
1-34
Finite Element Method (FEM) Basics
K and the right side F contain integrals in terms of the test functions
i
,
j
,
and the coefficients defining the problem: c, a, f, q, and g. The solution vector
U contains the expansion coefficients of u
h
, which are also the values of u
h
at
each node x
i
, since u
h
(x
i
) = U
i
.
If the exact solution u is smooth, then FEM computes u
h
with an error of the
same size as that of the linear interpolation. It is possible to estimate the
error on each triangle using only u
h
and the PDE coefficients (but not the
exact solution u, which in general is unknown).
There are Partial Differential Equation Toolbox functions that assemble K
and F. This is done automatically in the PDE app, but you also have direct
access to the FEM matrices from the command-line function assempde.
To summarize, the FEM approach is to approximate the PDE solution u by
a piecewise linear function u
h
. The function u
h
is expanded in a basis of
test-functions
i
, and the residual is tested against all the basis functions.
This procedure yields a linear system KU = F. The components of U are the
values of u
h
at the nodes. For x inside a triangle, u
h
(x) is found by linear
interpolation from the nodal values.
FEM techniques are also used to solve more general problems. The following
are some generalizations that you can access both through the PDE app and
with command-line functions.
Time-dependent problems are easy to implement in the FEM context. The
solution u(x,t) of the equation
d c au f
u
t
u
c
c
V V
( )
+ = ,
can be approximated by
u x t U t x
h i i
i
N
( , ) ( ) ( ). =
=

|
1
This yields a system of ordinary differential equations (ODE)
1-35
1 Getting Started
M
dU
dt
KU F + = ,
which you integrate using ODE solvers. Two time derivatives yield a
second order ODE
M
d U
dt
KU F
2
2
+ = ,
etc. The toolbox supports problems with one or two time derivatives (the
functions parabolic and hyperbolic).
Eigenvalue problems: Solve
V V
( )
+ = c u au du ,
for the unknowns u and ( is a complex number). Using the FEM
discretization, you solve the algebraic eigenvalue problem KU =
h
MU to
find u
h
and
h
as approximations to u and . A robust eigenvalue solver
is implemented in pdeeig.
If the coefficients c, a, f, q, or g are functions of u or u, the PDE is called
nonlinear and FEM yields a nonlinear system K(U)U = F(U). You can use
iterative methods for solving the nonlinear system. For elliptic equations,
the toolbox provides a nonlinear solver called pdenonlin using a damped
Gauss-Newton method. The parabolic and hyperbolic functions call
the nonlinear solver automatically.
Small triangles are needed only in those parts of the computational domain
where the error is large. In many cases the errors are large in a small
region and making all triangles small is a waste of computational effort.
Making small triangles only where needed is called adapting the mesh
refinement to the solution. An iterative adaptive strategy is the following:
For a given mesh, form and solve the linear system KU = F. Then estimate
the error and refine the triangles in which the error is large. The iteration
is controlled by adaptmesh and the error is estimated by pdejmps.
Although the basic equation is scalar, systems of equations are also handled
by the toolbox. The interactive environment accepts u as a scalar or 2-vector
function. In command-line mode, systems of arbitrary size are accepted.
1-36
Finite Element Method (FEM) Basics
If c > 0 and a 0, under rather general assumptions on the domain and
the boundary conditions, the solution u exists and is unique. The FEM linear
system has a unique solution which converges to u as the triangles become
smaller. The matrix K and the right side F make sense even when u does
not exist or is not unique. It is advisable that you devise checks to problems
with questionable solutions.
References
[1] Cook, Robert D., David S. Malkus, and Michael E. Plesha, Concepts and
Applications of Finite Element Analysis, 3rd edition, John Wiley & Sons,
New York, 1989.
1-37
1 Getting Started
1-38
2
Setting Up Your PDE
Open the PDE App on page 2-3
Specify Geometry Using a CSG Model on page 2-5
Select Graphical Objects Representing Your Geometry on page 2-7
Rounded Corners Using CSG Modeling on page 2-8
Enter Parameter Values as MATLAB Expressions on page 2-12
Systems of PDEs on page 2-13
Scalar PDE Coefficients on page 2-14
Scalar PDE Coefficients in String Form on page 2-16
Coefficients for Scalar PDEs in PDE App on page 2-19
Scalar PDE Coefficients in Function Form on page 2-23
Scalar PDE Functional Form and Calling Syntax on page 2-26
Enter Coefficients in the PDE App on page 2-32
Coefficients for Systems of PDEs on page 2-42
2-D Systems in the PDE App on page 2-44
f for Systems on page 2-48
c for Systems on page 2-50
a or d for Systems on page 2-59
Initial Conditions on page 2-62
Types of Boundary Conditions on page 2-65
No Boundary Conditions Between Subdomains on page 2-66
Identify Boundary Labels on page 2-69
2 Setting Up Your PDE
Boundary Conditions Overview on page 2-71
Boundary Conditions for Scalar PDE on page 2-72
Boundary Conditions for PDE Systems on page 2-77
Tooltip Displays for Mesh and Plots on page 2-84
Mesh Data on page 2-85
Adaptive Mesh Refinement on page 2-86
2-2
Open the PDE App
Open the PDE App
Partial Differential Equation Toolbox software includes the PDE app, which
covers all aspects of the PDE solution process. You start it by typing
pdetool
at the MATLAB command line. It may take a while the first time you launch
the PDE app during a MATLAB session. The following figure shows the PDE
app as it looks when you start it.
At the top, the PDE app has a pull-down menu bar that you use to control the
modeling. Below the menu bar, a toolbar with icon buttons provide quick and
easy access to some of the most important functions.
To the right of the toolbar is a pop-up menu that indicates the current
application mode. You can also use it to change the application mode. The
upper right part of the PDE app also provides the x- and y-coordinates of the
2-3
2 Setting Up Your PDE
current cursor position. This position is updated when you move the cursor
inside the main axes area in the middle of the PDE app.
The edit box for the set formula contains the active set formula.
In the main axes you draw the 2-D geometry, display the mesh, and plot
the solution.
At the bottom of the PDE app, an information line provides information about
the current activity. It can also display help information about the toolbar
buttons.
2-4
Specify Geometry Using a CSG Model
Specify Geometry Using a CSG Model
You can specify complex geometries by overlapping solid objects. This
approach to representing geometries is called Constructive Solid Geometry
(CSG).
Use these four solid objects to specify a geometry for your problem:
Circle Represents the set of points inside and on a circle.
Polygon Represents the set of points inside and on a polygon given by a
set of line segments.
Rectangle Represents the set of points inside and on a rectangle.
Ellipse Represents the set of points inside and on an ellipse. The ellipse
can be rotated.
When you draw a solid object in the PDE app, each solid object is automatically
given a unique name. Default names are C1, C2, C3, etc., for circles; P1, P2,
P3, etc. for polygons; R1, R2, R3, etc., for rectangles; E1, E2, E3, etc., for
ellipses. Squares, although a special case of rectangles, are named SQ1, SQ2,
SQ3, etc. The name is displayed on the solid object itself. You can use any
unique name, as long as it contains no blanks. In draw mode, you can alter
the names and the geometries of the objects by double-clicking them, which
opens a dialog box. The following figure shows an object dialog box for a circle.
You can use the name of the object to refer to the corresponding set of points
in a set formula. The operators +, *, and are used to form the set of points
in the plane over which the differential equation is solved. The operators
+, the set union operator, and *, the set intersection operator, have the
2-5
2 Setting Up Your PDE
same precedence. The operator , the set difference operator, has higher
precedence. The precedence can be controlled by using parentheses. The
resulting geometrical model, , is the set of points for which the set formula
evaluates to true. By default, it is the union of all solid objects. We often refer
to the area as the decomposed geometry.
2-6
Select Graphical Objects Representing Your Geometry
Select Graphical Objects Representing Your Geometry
Throughout the PDE app, similar principles apply for selecting objects such
as solid objects, subdomains, and boundaries.
To select a single object, click it using the left mouse button.
To select several objects and to deselect objects, Shift+click (or click using
the middle mouse button) on the desired objects.
Clicking in the intersection of several objects selects all the intersecting
objects.
To open an associated dialog box, double-click an object. If the object is not
selected, it is selected before opening the dialog box.
In draw mode and PDE mode, clicking outside of objects deselects all
objects.
To select all objects, use the Select All option from the Edit menu.
When defining boundary conditions and the PDE via the menu items from
the Boundary and PDE menus, and no boundaries or subdomains are
selected, the entered values applies to all boundaries and subdomains by
default.
2-7
2 Setting Up Your PDE
Rounded Corners Using CSG Modeling
This example shows how to represent a geometry that includes rounded
corners (fillets) using Constructive Solid Geometry (CSG) modeling. You learn
how to draw several overlapping solid objects, and specify how these objects
should combine to produce the desired geometry.
Start the PDE app using pdetool and turn on the grid and the snap-to-grid
feature using the Options menu. Also, change the grid spacing to
-1.5:0.1:1.5 for the x-axis and -1:0.1:1 for the y-axis.
Select Rectangle/square from the Draw menu or click the button with the
rectangle icon. Then draw a rectangle with a width of 2 and a height of 1
using the mouse, starting at (1,0.5). To get the round corners, add circles,
one in each corner. The circles should have a radius of 0.2 and centers at
a distance that is 0.2 units from the left/right and lower/upper rectangle
boundaries ((0.8,0.3), (0.8,0.3), (0.8,0.3), and (0.8,0.3)). To draw several
circles, double-click the button for drawing ellipses/circles (centered). Then
draw the circles using the right mouse button or Ctrl+click starting at the
circle centers. Finally, at each of the rectangle corners, draw four small
squares with a side of 0.2.
The following figure shows the complete drawing.
2-8
Rounded Corners Using CSG Modeling
Now you have to edit the set formula. To get the rounded corners, subtract
the small squares from the rectangle and then add the circles. As a set
formula, this is expressed as
R1-(SQ1+SQ2+SQ3+SQ4)+C1+C2+C3+C4
2-9
2 Setting Up Your PDE
Enter the set formula into the edit box at the top of the PDE app. Then enter
the Boundary mode by clicking the button or by selecting the Boundary
Mode option from the Boundary menu. The CSG model is now decomposed
using the set formula, and you get a rectangle with rounded corners, as shown
in the following figure.
2-10
Rounded Corners Using CSG Modeling
Because of the intersection of the solid objects used in the initial CSG model,
a number of subdomain borders remain. They are drawn using gray lines. If
this is a model of, e.g., a homogeneous plate, you can remove them. Select the
Remove All Subdomain Borders option from the Boundary menu. The
subdomain borders are removed and the model of the plate is now complete.
2-11
2 Setting Up Your PDE
Enter Parameter Values as MATLAB Expressions
When entering parameter values, e.g., as a function of x and y, the entered
string must be a MATLAB expression to be evaluated for x and y defined on
the current mesh, i.e., x and y are MATLAB row vectors. For example, the
function 4xy should be entered as 4*x.*y and not as 4*x*y, which normally
is not a valid MATLAB expression.
2-12
Systems of PDEs
Systems of PDEs
As described in Types of PDE Problems You Can Solve on page 1-4, Partial
Differential Equation Toolbox can solve systems of PDEs. This means you can
have N coupled PDEs, with coupled boundary conditions. The solvers such
as assempde and hyperbolic can solve systems of PDEs with any number
N of components.
Scalar PDEs are those with N = 1, meaning just one PDE. Systems of PDEs
generally means N > 1. The documentation sometimes refers to systems as
multidimensional PDEs or as PDEs with vector solution u.
In all cases, PDE systems have a single 2-D geometry and mesh. It is only N,
the number of equations, that can vary.
2-13
2 Setting Up Your PDE
Scalar PDE Coefficients
A scalar PDE is one of the following:
Elliptic
V V
( )
+ = c u au f ,
Parabolic
d c au f
u
t
u
c
c
V V
( )
+ = ,
Hyperbolic
d
u
t
c u au f
c
c
V V
( )
+ =
2
2
,
Eigenvalue
V V
( )
+ = c u au du ,
In all cases, the coefficients d, c, a, and f can be functions of position (x and
y) and the subdomain index. For all cases except eigenvalue, the coefficients
can also depend on the solution u and its gradient. And for parabolic and
hyperbolic equations, the coefficients can also depend on time.
The question is how to represent the coefficients for the toolbox.
There are three ways of representing each coefficient. You can use different
ways for different coefficients.
Numeric If a coefficient is numeric, give the value.
String formula See Scalar PDE Coefficients in String Form on page
2-16.
MATLAB function See Scalar PDE Coefficients in Function Form on
page 2-23.
2-14
Scalar PDE Coefficients
For an example incorporating each way to represent coefficients, see Scalar
PDE Functional Form and Calling Syntax on page 2-26.
Note If any coefficient depends on time or on the solution u or its gradient,
then that coefficient should be NaN when either time or the solution u is NaN.
This is the way that solvers check to see if the equation depends on time or
on the solution.
2-15
2 Setting Up Your PDE
Scalar PDE Coefficients in String Form
Write a text expression using these conventions:
'x' x-coordinate
'y' y-coordinate
'u' Solution of equation
'ux' Derivative of u in the x-direction
'uy' Derivative of u in the y-direction
't' Time (parabolic and hyperbolic equations)
'sd' Subdomain number
For example, you could use this string to represent a coefficient:
'(x+y)./(x.^2 + y.^2 + 1) + 3 + sin(t)./(1+u.^4)'
Note Use .*, ./, and .^ for multiplication, division, and exponentiation
operations. The text expressions operate on row vectors, so the operations
must make sense for row vectors. The row vectors are the values at the
triangle centroids in the mesh.
You can write MATLAB functions for coefficients as well as plain text
expressions. For example, suppose your coefficient f is given by the file
fcoeff.m:
function f = fcoeff(x,y,t,sd)
f = (x.*y)./(1+x.^2+y.^2); % f on subdomain 1
f = f + log(1+t); % include time
r = (sd == 2); % subdomain 2
f(r) = cos(x+y); % f on subdomain 2
Represent this function in the parabolic solver, for example:
u1 = parabolic(u0,tlist,b,p,e,t,c,a,'fcoeff(x,y,t,sd)',d)
2-16
Scalar PDE Coefficients in String Form
Caution In function form, t represents triangles, and time represents time.
In string form, t represents time, and triangles do not enter into the form.
There is a simple way to write a text expression for multiple subdomains
without using 'sd' or a function. Separate the formulas for the different
subdomains with the '!' character. Generally use the same number of
expressions as subdomains. However, if an expression does not depend on the
subdomain number, you can give just one expression.
For example, an expression for an input (a, c, f, or d) with three subdomains:
'(1+x.^2)./(1+x.^2+y.^2)!(1+(x+y).^2)./(1+x.^2+y.^2)!(1+x.^2)./(1+x.^2+y.
The coefficient c is a 2-by-2 matrix. You can give c in any of the following
forms:
Scalar or single string The software interprets c as a diagonal matrix:
c
c
0
0
|
\

|
.
|
Two-element column vector or two-row text array The software
interprets c as a diagonal matrix:
c
c
( )
( )
1 0
0 2
|
\

|
.
|
Three-element column vector or three-row text array The software
interprets c as a symmetric matrix:
c c
c c
( ) ( )
( ) ( )
1 2
2 3
|
\

|
.
|
Four-element column vector or four-row text array The software
interprets c as a full matrix:
2-17
2 Setting Up Your PDE
c c
c c
( ) ( )
( ) ( )
1 3
2 4
|
\

|
.
|
For example, c as a symmetric matrix with cos(xy) on the off-diagonal terms:
c = char('x.^2+y.^2',...
'cos(x.*y)',...
'u./(1+x.^2+y.^2)')
To include subdomains separated by '!', include the '!' in each row. For
example,
c = char('1+x.^2+y.^2!x.^2+y.^2',...
'cos(x.*y)!sin(x.*y)',...
'u./(1+x.^2+y.^2)!u.*(x.^2+y.^2)')
Caution Do not include spaces in your coefficient strings in the PDE app.
The string parser can misinterpret a space as a vector separator, as when a
MATLAB vector uses a space to separate elements of a vector.
For elliptic problems, when you include 'u', 'ux', or 'uy', you must use
the pdenonlin solver instead of assempde. In the PDE app, select Solve >
Parameters > Use nonlinear solver.
Related
Examples
Scalar PDE Functional Form and Calling Syntax on page 2-26
Enter Coefficients in the PDE App on page 2-32
Scalar PDE Coefficients in Function Form on page 2-23
Concepts Scalar PDE Coefficients on page 2-14
2-18
Coefficients for Scalar PDEs in PDE App
Coefficients for Scalar PDEs in PDE App
To enter coefficients for your PDE, select PDE > PDE Specification.
Enter text expressions using these conventions:
x x-coordinate
y y-coordinate
u Solution of equation
ux Derivative of u in the x-direction
uy Derivative of u in the y-direction
t Time (parabolic and hyperbolic equations)
sd Subdomain number
For example, you could use this expression to represent a coefficient:
(x+y)./(x.^2+y.^2+1)+3+sin(t)./(1+u.^4)
2-19
2 Setting Up Your PDE
For elliptic problems, when you include u, ux, or uy, you must use the
nonlinear solver. Select Solve > Parameters > Use nonlinear solver.
Note
Do not use quotes or unnecessary spaces in your entries. The string parser
can misinterpret a space as a vector separator, as when a MATLAB vector
uses a space to separate elements of a vector.
Use .*, ./, and .^ for multiplication, division, and exponentiation
operations. The text expressions operate on row vectors, so the operations
must make sense for row vectors. The row vectors are the values at the
triangle centroids in the mesh.
You can write MATLAB functions for coefficients as well as plain text
expressions. For example, suppose your coefficient f is given by the file
fcoeff.m.
function f = fcoeff(x,y,t,sd)
f = (x.*y)./(1+x.^2+y.^2); % f on subdomain 1
f = f + log(1+t); % include time
r = (sd == 2); % subdomain 2
f(r) = cos(x+y); % f on subdomain 2
Use fcoeff(x,y,t,sd) as the f coefficient in the parabolic solver.
2-20
Coefficients for Scalar PDEs in PDE App
Alternatively, you can represent a coefficient in function form rather than in
string form. See Scalar PDE Coefficients in Function Form on page 2-23.
The coefficient c is a 2-by-2 matrix. You can give 1-, 2-, 3-, or 4-element
matrix expressions. Separate the expressions for elements by spaces. These
expressions mean:
1-element expression:
c
c
0
0
|
\

|
.
|
2-element expression:
c
c
( )
( )
1 0
0 2
|
\

|
.
|
3-element expression:
c c
c c
( ) ( )
( ) ( )
1 2
2 3
|
\

|
.
|
4-element expression:
c c
c c
( ) ( )
( ) ( )
1 3
2 4
|
\

|
.
|
For example, c is a symmetric matrix with constant diagonal entries and
cos(xy) as the off-diagonal terms:
1.1 cos(x.*y) 5.5
This corresponds to coefficients for the parabolic equation
2-21
2 Setting Up Your PDE
c
c
V
|
\

|
.
|
V
|
\

|
.
|
=
u
t
xy
xy
u
. cos( )
cos( ) .
.
1 1
5 5
10
Related
Examples
Enter Coefficients in the PDE App on page 2-32
Concepts Scalar PDE Coefficients on page 2-14
2-22
Scalar PDE Coefficients in Function Form
Scalar PDE Coefficients in Function Form
In this section...
Coefficients as the Result of a Program on page 2-23
Calculate Coefficients in Function Form on page 2-24
Coefficients as the Result of a Program
Usually. the simplest way to give coefficients as the result of a program is
to use a string expression as described in Scalar PDE Coefficients in String
Form on page 2-16. For the most detailed control over coefficients, though,
you can write a function form of coefficients.
A coefficient in function form has the syntax
coeff = coeffunction(p,t,u,time)
coeff represents any coefficient: c, a, f, or d.
Your program evaluates the return coeff as a row vector of the function
values at the centroids of the triangles t. For help calculating these values,
see Calculate Coefficients in Function Form on page 2-24.
p and t are the node points and triangles of the mesh. For a description of
these data structures, see Mesh Data on page 2-85. In brief, each column
of p contains the x- and y-values of a point, and each column of t contains
the indices of three points in p and the subdomain label of that triangle.
u is a row vector containing the solution at the points p. u is [] if the
coefficients do not depend on the solution or its derivatives.
time is the time of the solution, a scalar. time is [] if the coefficients do
not depend on time.
Caution In function form, t represents triangles, and time represents time.
In string form, t represents time, and triangles do not enter into the form.
2-23
2 Setting Up Your PDE
Pass the coefficient function to the solver as a string 'coeffunction' or
as a function handle @coeffunction. In the PDE app, pass the coefficient
as a string coeffunction without quotes, because the PDE app interprets
all entries as strings.
If your coefficients depend on u or time, then when u or time are NaN, ensure
that the corresponding coeff consist of a vector of NaN of the correct size.
This signals to solvers, such as parabolic, to use a time-dependent or
solution-dependent algorithm.
For elliptic problems, if any coefficient depends on u or its gradient, you must
use the pdenonlin solver instead of assempde. In the PDE app, select Solve >
Parameters > Use nonlinear solver.
Calculate Coefficients in Function Form
X- and Y-Values
The x- and y-values of the centroid of a triangle t are the mean values of the
entries of the points p in t. To get row vectors xpts and ypts containing
the mean values:
% Triangle point indices
it1=t(1,:);
it2=t(2,:);
it3=t(3,:);
% Find centroids of triangles
xpts=(p(1,it1)+p(1,it2)+p(1,it3))/3;
ypts=(p(2,it1)+p(2,it2)+p(2,it3))/3;
Interpolated u
The pdeintrp function linearly interpolates the values of u at the centroids of
t, based on the values at the points p.
uintrp = pdeintrp(p,t,u); % Interpolated values at centroids
The output uintrp is a row vector with the same number of columns as t. Use
uintrp as the solution value in your coefficient calculations.
2-24
Scalar PDE Coefficients in Function Form
Gradient or Derivatives of u
The pdegrad function approximates the gradient of u.
[ux,uy] = pdegrad(p,t,u); % Approximate derivatives
The outputs ux and uy are row vectors with the same number of columns as t.
Subdomains
If your coefficients depend on the subdomain label, check the subdomain
number for each triangle. Subdomains are the last (fourth) row of the triangle
matrix. So the row vector of subdomain numbers is:
subd = t(4,:);
You can see the subdomain labels by using the pdegplot function with the
subdomainLabels name-value pair set to 'on':
pdegplot(g,'subdomainLabels','on')
Related
Examples
Scalar PDE Functional Form and Calling Syntax on page 2-26
Enter Coefficients in the PDE App on page 2-32
Scalar PDE Coefficients in String Form on page 2-16
Deflection of a Piezoelectric Actuator on page 3-19
Concepts Scalar PDE Coefficients on page 2-14
2-25
2 Setting Up Your PDE
Scalar PDE Functional Form and Calling Syntax
This example shows how to write PDE coefficients in string form and in
functional form.
The geometry is a rectangle with a circular hole.
Code for generating the figure
% Rectangle is code 3, 4 sides,
% followed by x-coordinates and then y-coordinates
R1 = [3,4,-1,1,1,-1,-.4,-.4,.4,.4]';
2-26
Scalar PDE Functional Form and Calling Syntax
% Circle is code 1, center (.5,0), radius .2
C1 = [1,.5,0,.2]';
% Pad C1 with zeros to enable concatenation with R1
C1 = [C1;zeros(length(R1)-length(C1),1)];
geom = [R1,C1];
% Names for the two geometric objects
ns = (char('R1','C1'))';
% Set formula
sf = 'R1-C1';
% Create geometry
gd = decsg(geom,sf,ns);
% View geometry
pdegplot(gd,'edgeLabels','on')
xlim([-1.1 1.1])
axis equal
The PDE is parabolic,
d c au f
u
t
u
c
c
V V
( )
+ = ,
with the following coefficients:
d = 5
a = 0
f is a linear ramp up to 10, holds at 10, then ramps back down to 0:
f
t t
t
t t
=
s s
s s
s s

10
10 0 0 1
1 0 1 0 9
10 10 0 9 1
*
.
. .
.
c = 1 +.x
2
+ y
2
Write a function for the f coefficient.
2-27
2 Setting Up Your PDE
function f = framp(t)
if t <= 0.1
f = 10*t;
elseif t <= 0.9
f = 1;
else
f = 10-10*t;
end
f = 10*f;
The boundary conditions are the same as in Boundary Conditions for Scalar
PDE on page 2-72.
Boundary conditions
Suppose the boundary conditions on the outer boundary (segments 1 through
4) are Dirichlet, with the value u(x,y) = t(x y), where t is time. Suppose
the circular boundary (segments 5 through 8) has a generalized Neumann
condition, with q = 1 and g = x
2
+ y
2
.
function [qmatrix,gmatrix,hmatrix,rmatrix] = pdebound(p,e,u,time)
ne = size(e,2); % number of edges
qmatrix = zeros(1,ne);
gmatrix = qmatrix;
hmatrix = zeros(1,2*ne);
rmatrix = hmatrix;
for k = 1:ne
x1 = p(1,e(1,k)); % x at first point in segment
x2 = p(1,e(2,k)); % x at second point in segment
xm = (x1 + x2)/2; % x at segment midpoint
y1 = p(2,e(1,k)); % y at first point in segment
y2 = p(2,e(2,k)); % y at second point in segment
ym = (y1 + y2)/2; % y at segment midpoint
switch e(5,k)
case {1,2,3,4} % rectangle boundaries
hmatrix(k) = 1;
hmatrix(k+ne) = 1;
2-28
Scalar PDE Functional Form and Calling Syntax
if ~isempty(time)
rmatrix(k) = time*(x1 - y1);
rmatrix(k+ne) = time*(x2 - y2);
end
otherwise % same as case {5,6,7,8}, circle boundaries
qmatrix(k) = 1;
gmatrix(k) = xm^2 + ym^2;
end
end
The initial condition is u(x,y) = 0 at t = 0.
After running the code for creating the geometry, create the mesh, refine
it twice, and jiggle it once.
[p,e,t] = initmesh(gd);
[p,e,t] = refinemesh(gd,p,e,t);
[p,e,t] = refinemesh(gd,p,e,t);
p = jigglemesh(p,e,t);
Set the time steps for the parabolic solver to 50 steps from time 0 to time 1.
tlist = linspace(0,1,50);
Solve the parabolic PDE.
b = @pdebound;
d = 5;
a = 0;
f = 'framp(t)';
c = '1+x.^2+y.^2';
u = parabolic(0,tlist,b,p,e,t,c,a,f,d);
View an animation of the solution.
for tt = 1:size(u,2) % number of steps
pdeplot(p,e,t,'xydata',u(:,tt),'zdata',u(:,tt),'colormap','jet')
axis([-1 1 -1/2 1/2 -1.5 1.5 -1.5 1.5]) % use fixed axis
title(['Step ' num2str(tt)])
view(-45,22)
drawnow
pause(.1)
2-29
2 Setting Up Your PDE
end
Equivalently, you can write a function for the coefficient f in the syntax
described in Scalar PDE Coefficients in Function Form on page 2-23.
function f = framp2(p,t,u,time)
if time <= 0.1
f = 10*time;
elseif time <= 0.9
f = 1;
else
f = 10-10*time;
2-30
Scalar PDE Functional Form and Calling Syntax
end
f = 10*f;
Call this function by setting
f = @framp2;
u = parabolic(0,tlist,b,p,e,t,c,a,f,d);
You can also write a function for the coefficient c, though it is more
complicated than the string formulation.
function c = cfunc(p,t,u,time)
% Triangle point indices
it1=t(1,:);
it2=t(2,:);
it3=t(3,:);
% Find centroids of triangles
xpts=(p(1,it1)+p(1,it2)+p(1,it3))/3;
ypts=(p(2,it1)+p(2,it2)+p(2,it3))/3;
c = 1 + xpts.^2 + ypts.^2;
Call this function by setting
c = @cfunc;
u = parabolic(0,tlist,b,p,e,t,c,a,f,d);
Related
Examples
Enter Coefficients in the PDE App on page 2-32
Scalar PDE Coefficients in String Form on page 2-16
Scalar PDE Coefficients in Function Form on page 2-23
Nonlinear Heat Transfer In a Thin Plate on page 3-60
Deflection of a Piezoelectric Actuator on page 3-19
Concepts Scalar PDE Coefficients on page 2-14
2-31
2 Setting Up Your PDE
Enter Coefficients in the PDE App
This example shows how to enter coefficients in the PDE app.
Caution: Do not include spaces in your coefficient strings in the PDE app.
The string parser can misinterpret a space as a vector separator, as when a
MATLAB vector uses a space to separate elements of a vector.
The PDE is parabolic,
d c au f
u
t
u
c
c
V V
( )
+ = ,
with the following coefficients:
d = 5
a = 0
f is a linear ramp up to 10, holds at 10, then ramps back down to 0:
f
t t
t
t t
=
s s
s s
s s

10
10 0 0 1
1 0 1 0 9
10 10 0 9 1
*
.
. .
.
c = 1 +.x
2
+ y
2
These coefficients are the same as in Scalar PDE Functional Form and
Calling Syntax on page 2-26.
Write the following file framp.m and save it on your MATLAB path.
function f = framp(t)
if t <= 0.1
f = 10*t;
elseif t <= 0.9
f = 1;
else
f = 10-10*t;
2-32
Enter Coefficients in the PDE App
end
f = 10*f;
Open the PDE app, either by typing pdetool at the command line, or selecting
PDE from the Apps menu.
Select PDE > PDE Specification.
Select Parabolic equation. Fill in the coefficients as pictured:
c = 1+x.^2+y.^2
a = 0
f = framp(t)
d = 5
The PDE app interprets all inputs as strings. Therefore, do not include quotes
for the c or f coefficients.
Select Options > Grid and Options > Snap.
2-33
2 Setting Up Your PDE
Select Draw > Draw Mode, then draw a rectangle centered at (0,0) extending
to 1 in the x-direction and 0.4 in the y-direction.
Draw a circle centered at (0.5,0) with radius 0.2
Change the set formula to R1-C1.
2-34
Enter Coefficients in the PDE App
2-35
2 Setting Up Your PDE
Select Boundary > Boundary Mode
Click a segment of the outer rectangle, then Shift-click the other three
segments so that all four segments of the rectangle are selected.
Double-click one of the selected segments.
Fill in the resulting dialog box as pictured, with Dirichlet boundary conditions
h = 1 and r = t*(x-y). Click OK.
Select the four segments of the inner circle using Shift-click, and double-click
one of the segments.
Select Neumann boundary conditions, and set g = x.^2+y.^2 and q = 1.
Click OK.
2-36
Enter Coefficients in the PDE App
Click to initialize the mesh.
Click to refine the mesh. Click again to get an even finer mesh.
Select Mesh > Jiggle Mesh to improve the quality of the mesh.
Set the time interval and initial condition by selecting Solve > Parameters
and setting Time = linspace(0,1,50) and u(t0) = 0. Click OK.
2-37
2 Setting Up Your PDE
Solve and plot the equation by clicking the button.
2-38
Enter Coefficients in the PDE App
2-39
2 Setting Up Your PDE
Match the following figure using Plot > Parameters.
Click the Plot button.
2-40
Enter Coefficients in the PDE App
Related
Examples
Scalar PDE Functional Form and Calling Syntax on page 2-26
Concepts Coefficients for Scalar PDEs in PDE App on page 2-19
Scalar PDE Coefficients on page 2-14
2-41
2 Setting Up Your PDE
Coefficients for Systems of PDEs
As describe in Systems of PDEs on page 2-13, toolbox functions can address
the case of systems of N PDEs. How do you represent the coefficients of your
PDE in the correct form? In general, an elliptic system is
V V
( )
+ = c u au f,
The notation V V ( ) c u means the N-by-1 matrix with (i,1)-component
c
c
c
c
+
c
c
c
c
+
c
c
c
c
+
c
c
c
c x
c
x x
c
y y
c
x y
c
i j i j i j i j , , , , , , , , , , , , 1 1 1 2 2 1 2 2
yy
u
j
N
j
|
\

|
.
|
=

1
Other problems with N > 1 are the parabolic system
d c au f
u
u
c
c
V V
( )
+ =
t
,
the hyperbolic system
d
u
c u au f
c
c
V V
( )
+ =
2
2
t
,
and the eigenvalue system
V V
( )
+ = c u au du .
To solve a PDE using this toolbox, you convert your problem into one of the
forms the toolbox accepts. Then express your problem coefficients in a form
the toolbox accepts.
The question is how to express each coefficient: d, c, a, and f. For answers,
see f for Systems on page 2-48, c for Systems on page 2-50, and a or d
for Systems on page 2-59.
2-42
Coefficients for Systems of PDEs
Note If any coefficient depends on time or on the solution u or its gradient,
then all coefficients should be NaN when either time or the solution u is NaN.
This is the way that solvers check to see if the equation depends on time or
on the solution.
2-43
2 Setting Up Your PDE
2-D Systems in the PDE App
You can enter coefficients for a system with N = 2 in the PDE app. To do so,
open the PDE app and select Generic System.
Then select PDE > PDE Specification.
2-44
2-D Systems in the PDE App
Enter string expressions for coefficients using the form in Coefficients for
Scalar PDEs in PDE App on page 2-19.
Do not use quotes or unnecessary spaces in your entries.
For higher-dimensional systems, do not use the PDE app. Represent your
problem coefficients at the command line.
You can enter scalars into the c matrix, corresponding to these equations:


V V ( ) V V ( ) + + =
V V ( ) V V (


c u c u a u a u f
c u c u
11 1 12 2 11 1 12 2 1
21 1 22 2
)) + + = a u a u f
21 1 22 2 2
.
2-45
2 Setting Up Your PDE
If you need matrix versions of any of the cij coefficients, enter expressions
separated by spaces. You can give 1-, 2-, 3-, or 4-element matrix expressions.
These mean:
1-element expression:
c
c
0
0
|
\

|
.
|
2-element expression:
c
c
( )
( )
1 0
0 2
|
\

|
.
|
3-element expression:
c c
c c
( ) ( )
( ) ( )
1 2
2 3
|
\

|
.
|
4-element expression:
c c
c c
( ) ( )
( ) ( )
1 3
2 4
|
\

|
.
|
For details, see c for Systems on page 2-50.
For example, these expressions show one of each type (1-, 2-, 3-, and 4-element
expressions)
2-46
2-D Systems in the PDE App
These expressions correspond to the equations
V
+
+
|
\

|
.
|
V
|
\

|
.
|
V
|
\

|
.
|
V
|
\

cos( )
cos( )

4 0
0 4
1 0
0 1
1 2
xy
xy
u u
||
.
|
=
V
|
\

|
.
|
V
|
\

|
.
|
V

|
\

|
.
|
V
|

1
1 2
2 3
7 6
5
1 2

. .
. .

.
. exp( )
u
x y
u
\\

|
.
|
= 2.
Related
Examples
Coefficients for Scalar PDEs in PDE App on page 2-19
f for Systems on page 2-48
c for Systems on page 2-50
a or d for Systems on page 2-59
Concepts Coefficients for Systems of PDEs on page 2-42
2-47
2 Setting Up Your PDE
f for Systems
This section describes how to write the coefficient f in the equation
V V
( )
+ = c u au f,
or in similar equations. The number of rows in f indicates N, the number of
equations. Give f as any of the following:
A scalar or single string expression. Solvers expand the single input to a
vector of N elements.
A column vector with N components. For example, if N = 3, f could be:
f = [3;4;10];
A character array with N rows. The rows of the character array are
MATLAB expressions as described in Scalar PDE Coefficients in String
Form on page 2-16. Pad the rows with spaces so each row has the same
number of characters (char does this automatically). For example, if N = 3,
f could be:
f = char('sin(x) + cos(y)','cosh(x.*y)','x.*y./(1+x.^2+y.^2)')
f =
sin(x) + cos(y)
cosh(x.*y)
x.*y./(1+x.^2+y.^2)
A function of the form as described in Scalar PDE Coefficients in Function
Form on page 2-23. The function should return a matrix of size N-by-Nt,
where Nt is the number of triangles in the mesh. The function should
evaluate f at the triangle centroids, as in Scalar PDE Coefficients in
Function Form on page 2-23. Give solvers the function name as a string
'filename', or as a function handle @filename, where filename.m is a file
on your MATLAB path. For details on writing the function, see Calculate
Coefficients in Function Form on page 2-24.
For example, if N = 3, f could be:
2-48
f for Systems
function f = fcoeffunction(p,t,u,time)
N = 3; % Number of equations
% Triangle point indices
it1=t(1,:);
it2=t(2,:);
it3=t(3,:);
% Find centroids of triangles
xpts=(p(1,it1)+p(1,it2)+p(1,it3))/3;
ypts=(p(2,it1)+p(2,it2)+p(2,it3))/3;
[ux,uy] = pdegrad(p,t,u); % Approximate derivatives
uintrp = pdeintrp(p,t,u); % Interpolated values at centroids
nt = size(t,2); % Number of columns
f = zeros(N,nt); % Allocate f
% Now the particular functional form of f
f(1,:) = xpts - ypts + uintrp(1,:);
f(2,:) = 1 + tanh(ux(1,:)) + tanh(uy(3,:));
f(3,:) = (5+uintrp(3,:)).*sqrt(xpts.^2+ypts.^2);
Because this function depends on the solution u, if the equation is elliptic,
use the pdenonlin solver. The initial value can be all 0s in the case of
Dirichlet boundary conditions:
np = size(p,2); % number of points
u0 = zeros(N*np,1); % initial guess
Related
Examples
a or d for Systems on page 2-59
c for Systems on page 2-50
Deflection of a Piezoelectric Actuator on page 3-19
2-49
2 Setting Up Your PDE
c for Systems
In this section...
c as Tensor, Matrix, and Vector on page 2-50
Scalar c on page 2-52
Two-Element Column Vector c on page 2-52
Three-Element Column Vector c on page 2-53
Four-Element Column Vector c on page 2-53
N-Element Column Vector c on page 2-54
2N-Element Column Vector c on page 2-55
3N-Element Column Vector c on page 2-56
4N-Element Column Vector c on page 2-57
2N(2N+1)/2-Element Column Vector c on page 2-57
4N
2
-Element Column Vector c on page 2-58
c as Tensor, Matrix, and Vector
This section describes how to write the coefficient c in the equation
V V
( )
+ = c u au f,
or in similar equations. The coefficient c is an N-by-N-by-2-by-2 tensor with
components c(i,j,k,l).
The notation V V ( ) c u means the N-by-1 matrix with (i,1)-component.
c
c
c
c
+
c
c
c
c
+
c
c
c
c
+
c
c
c
c x
c
x x
c
y y
c
x y
c
i j i j i j i j , , , , , , , , , , , , 1 1 1 2 2 1 2 2
yy
u
j
N
j
|
\

|
.
|
=

1
There are many ways to represent the coefficient c. All representations begin
with a flattening of the N-by-N-by-2-by-2 tensor to a 2N-by-2N matrix,
where the matrix is logically an N-by-N matrix of 2-by-2 blocks.
2-50
c for Systems
c c c c c N c N ( , , , ) ( , , , ) ( , , , ) ( , , , ) ( , , , ) ( , , 1 1 1 1 1 1 1 2 1 2 1 1 1 2 1 2 1 1 1 1 1 ,, )
( , , , ) ( , , , ) ( , , , ) ( , , , ) ( , , , ) (
2
1 1 2 1 1 1 2 2 1 2 2 1 1 2 2 2 1 2 1 1 c c c c c N c ,, , , )
( , , , ) ( , , , ) ( , , , ) ( , , , ) ( , , ,
N
c c c c c N
2 2
2 1 1 1 2 1 1 2 2 2 1 1 2 2 1 2 2 1 1 )) ( , , , )
( , , , ) ( , , , ) ( , , , ) ( , , , ) ( ,
c N
c c c c c N
2 1 2
2 1 2 1 2 1 2 2 2 2 2 1 2 2 2 2 2 ,, , ) ( , , , )
( , , , ) ( , , , ) ( , , , ) ( ,
2 1 2 2 2
1 1 1 1 1 2 2 1 1 2
c N
c N c N c N c N

,, , ) ( , , , ) ( , , , )
( , , , ) ( , , , ) ( , , , )
1 2 1 1 1 2
1 2 1 1 2 2 2 2 1
c N N c N N
c N c N c N c(( , , , ) ( , , , ) ( , , , ) N c N N c N N 2 2 2 2 1 2 2
|
\

|
.
|
|
|
|
|
|
|
|
|
|
|
||
The matrix further gets flattened to a vector, where the N-by-N matrix of
2-by-2 blocks is first transformed to a vector of 2-by-2 blocks, and then the
2-by-2 blocks are turned into vectors in the usual column-wise way.
The coefficient vector c relates to the tensor c as follows:
c c c N c N c N N c N N
c c c
( ) ( ) ( ) ( ) ( ( ) ) ( ( ) )
( ) ( )
1 3 2 1 2 3 2 2 1 1 2 2 1 3
2 4
+ + + +
(( ) ( ) ( ( ) ) ( ( ) )
( ) ( ) ( ) (
2 2 2 4 2 2 1 2 2 2 1 4
5 7 2 5
N c N c N N c N N
c c c N c
+ + + +
+

22 7 2 2 1 5 2 2 1 7
6 8 2 6 2 8
N c N N c N N
c c c N c N c
+ + +
+ +
) ( ( ) ) ( ( ) )
( ) ( ) ( ) ( ) (

22 2 1 6 2 2 1 8
2 3 2 1 4 3 4
N N c N N
c N c N c N c N
( ) ) ( ( ) )
( ) ( ) ( ) (
+ +


11 4 3 4 1
2 2 2 4 2 4 4 2 4
2 2
2 2
) ( ) ( )
( ) ( ) ( ) ( ) ( ) (

c N c N
c N c N c N c N c N c N

))
|
\

|
.
|
|
|
|
|
|
|
|
|
|
|
|
|
Coefficient c(i,j,k,l) is in row (4N(j1)+4i+2l+k6) of the vector c.
Express c as numbers, text expressions, or functions, as in f for Systems
on page 2-48.
Often, your tensor c has structure, such as symmetric or block diagonal. In
many cases, you can represent c using a smaller vector than one with 4N
2
components.
2-51
2 Setting Up Your PDE
The number of rows in the matrix can differ from 4N
2
, as described in the
next few sections.
In function form, the number of columns is Nt, which is the number of
triangles in the mesh. The function should evaluate c at the triangle centroids,
as in Scalar PDE Coefficients in Function Form on page 2-23. Give solvers
the function name as a string 'filename', or as a function handle @filename,
where filename.m is a file on your MATLAB path. For details on writing the
function, see Calculate Coefficients in Function Form on page 2-24.
Scalar c
The software interprets a scalar c as a diagonal matrix, with c(i,i,1,1) and
c(i,i,2,2) equal to the scalar, and all other entries 0.
c
c
c
c
c
c
0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
0 0 0 0 0
0 0 0 0 0

|
\

|
.
|
|
|
|
|
|
|
|
|
|
|
|
Two-Element Column Vector c
The software interprets a two-element column vector c as a diagonal matrix,
with c(i,i,1,1) and c(i,i,2,2) as the two entries, and all other entries 0.
2-52
c for Systems
c
c
c
c
c
( )
( )
( )
( )
( )
1 0 0 0 0 0
0 2 0 0 0 0
0 0 1 0 0 0
0 0 0 2 0 0
0 0 0 0 1 0
0

00 0 0 0 2 c( )
|
\

|
.
|
|
|
|
|
|
|
|
|
|
|
|
Three-Element Column Vector c
The software interprets a three-element column vector c as a symmetric
block diagonal matrix, with c(i,i,1,1) = c(1), c(i,i,2,2) = c(3), and
c(i,i,1,2) = c(i,i,2,1) = c(2).
c c
c c
c c
c c
( ) ( )
( ) ( )
( ) ( )
( ) ( )
1 2 0 0 0 0
2 3 0 0 0 0
0 0 1 2 0 0
0 0 2 3 0 0

0 0 0 0 1 2
0 0 0 0 2 3
c c
c c
( ) ( )
( ) ( )
|
\

|
.
|
|
|
|
|
|
|
|
|
|
|
|
Four-Element Column Vector c
The software interprets a four-element column vector c as a block diagonal
matrix.
2-53
2 Setting Up Your PDE
c c
c c
c c
c c
( ) ( )
( ) ( )
( ) ( )
( ) ( )
1 3 0 0 0 0
2 4 0 0 0 0
0 0 1 3 0 0
0 0 2 4 0 0

0 0 0 0 1 3
0 0 0 0 2 4
c c
c c
( ) ( )
( ) ( )
|
\

|
.
|
|
|
|
|
|
|
|
|
|
|
|
N-Element Column Vector c
The software interprets an N-element column vector c as a diagonal matrix.
c
c
c
c
c N
( )
( )
( )
( )
( )
1 0 0 0 0 0
0 1 0 0 0 0
0 0 2 0 0 0
0 0 0 2 0 0
0 0 0 0 0
0

00 0 0 0 c N ( )
|
\

|
.
|
|
|
|
|
|
|
|
|
|
|
|
2-54
c for Systems
Caution If N = 2, 3, or 4, the 2-, 3-, or 4-element column vector form takes
precedence over the N-element form. So, for example, if N = 3, and you have a
c matrix of the form
c
c
c
c
c
c
1 0 0 0 0 0
0 1 0 0 0 0
0 0 2 0 0 0
0 0 0 2 0 0
0 0 0 0 3 0
0 0 0 0 0 3
|
\

|
.
|
|
|
|
||
|
|
|
,
you cannot use the N-element form of c, and instead would have to use
the 2N-element form. If you give c as the vector [c1;c2;c3], the software
interprets it as a 3-element form, namely
c c
c c
c c
c c
c c
c c
1 2 0 0 0 0
2 3 0 0 0 0
0 0 1 2 0 0
0 0 2 3 0 0
0 0 0 0 1 2
0 0 0 0 2 3
|
\

||
.
|
|
|
|
|
|
|
|
.
Instead, use the 2N-element form [c1;c1;c2;c2;c3;c3].
2N-Element Column Vector c
The software interprets a 2N-element column vector c as a diagonal matrix.
2-55
2 Setting Up Your PDE
c
c
c
c
c N
( )
( )
( )
( )
(
1 0 0 0 0 0
0 2 0 0 0 0
0 0 3 0 0 0
0 0 0 4 0 0
0 0 0 0 2 1

))
( )
0
0 0 0 0 0 2 c N
|
\

|
.
|
|
|
|
|
|
|
|
|
|
|
|
Caution If N = 2, the 4-element form takes precedence over the 2N-element
form. For example, if your c matrix is
c
c
c
c
1 0 0 0
0 2 0 0
0 0 3 0
0 0 0 4
|
\

|
.
|
|
|
|
,
you cannot give c as [c1;c2;c3;c4], because the software interprets this
vector as the 4-element form
c c
c c
c c
c c
1 3 0 0
2 4 0 0
0 0 1 3
0 0 2 4
|
\

|
.
|
|
|
|
.
Instead, use the 3N-element form [c1;0;c2;c3;0;c4] or the 4N-element
form [c1;0;0;c2;c3;0;0;c4].
3N-Element Column Vector c
The software interprets a 3N-element column vector c as a symmetric block
diagonal matrix.
2-56
c for Systems
c c
c c
c c
c c
( ) ( )
( ) ( )
( ) ( )
( ) ( )
1 2 0 0 0 0
2 3 0 0 0 0
0 0 4 5 0 0
0 0 5 6 0 0

0 0 0 0 3 2 3 1
0 0 0 0 3 1 3
c N c N
c N c N
( ) ( )
( ) ( )

|
\

|
.
|
|
|
|
||
|
|
|
|
|
|
|
Coefficient c(i,j,k,l) is in row (3i+k+l4) of the vector c.
4N-Element Column Vector c
The software interprets a 4N-element column vector c as a block diagonal
matrix.
c c
c c
c c
c c
( ) ( )
( ) ( )
( ) ( )
( ) ( )
1 3 0 0 0 0
2 4 0 0 0 0
0 0 5 7 0 0
0 0 6 8 0 0

0 0 0 0 4 3 4 1
0 0 0 0 4 2 4
c N c N
c N c N
( ) ( )
( ) ( )

|
\

|
.
|
|
|
|
||
|
|
|
|
|
|
|
Coefficient c(i,j,k,l) is in row (4i+2l+k6) of the vector c.
2N(2N+1)/2-Element Column Vector c
The software interprets a 2N(2N+1)/2-element column vector c as a symmetric
matrix. In the following diagram, means the entry is symmetric.
2-57
2 Setting Up Your PDE
c c c c c N N c N N
c c
( ) ( ) ( ) ( ) (( )( ) ) (( )( ) )
( ) ( )
1 2 4 6 1 2 1 1 1 2 1 3
3 5

+ +
cc c N N c N N
c c c N
( ) (( )( ) ) (( )( ) )
( ) ( ) (( )(
7 1 2 1 2 1 2 1 4
8 9 1 2

+ +
NN c N N
c c N N c N
+ +
+
1 5 1 2 1 7
10 1 2 1 6 1 2
) ) (( )( ) )
( ) (( )( ) ) (( )( NN
c N N c N N
c N N
+
+ +
+
1 8
2 1 2 2 1 1
2 1
) )
( ( ) ) ( ( ) )
( (


)))
|
\

|
.
|
|
|
|
|
|
|
|
|
|
|
|
Coefficient c(i,j,k,l), for i < j, is in row (2j
2
3j+4i+2l+k5) of the vector c. For
i = j, coefficient c(i,j,k,l) is in row (2i
2
+i+l+k4) of the vector c.
4N
2
-Element Column Vector c
The software interprets a 4N
2
-element column vector c as a matrix.
c c c N c N c N N c N N
c c c
( ) ( ) ( ) ( ) ( ( ) ) ( ( ) )
( ) ( )
1 3 2 1 2 3 2 2 1 1 2 2 1 3
2 4
+ + + +
(( ) ( ) ( ( ) ) ( ( ) )
( ) ( ) ( ) (
2 2 2 4 2 2 1 2 2 2 1 4
5 7 2 5
N c N c N N c N N
c c c N c
+ + + +
+

22 7 2 2 1 5 2 2 1 7
6 8 2 6 2 8
N c N N c N N
c c c N c N c
+ + +
+ +
) ( ( ) ) ( ( ) )
( ) ( ) ( ) ( ) (

22 2 1 6 2 2 1 8
2 3 2 1 4 3 4
N N c N N
c N c N c N c N
( ) ) ( ( ) )
( ) ( ) ( ) (
+ +


11 4 3 4 1
2 2 2 4 2 4 4 2 4
2 2
2 2
) ( ) ( )
( ) ( ) ( ) ( ) ( ) (

c N c N
c N c N c N c N c N c N

))
|
\

|
.
|
|
|
|
|
|
|
|
|
|
|
|
|
Coefficient c(i,j,k,l) is in row (4N(j1)+4i+2l+k6) of the vector c.
Related
Examples
f for Systems on page 2-48
a or d for Systems on page 2-59
Deflection of a Piezoelectric Actuator on page 3-19
2-58
a or d for Systems
a or d for Systems
In this section...
Coefficients a or d on page 2-59
Scalar a or d on page 2-60
N-Element Column Vector a or d on page 2-60
N(N+1)/2-Element Column Vector a or d on page 2-60
N
2
-Element Column Vector a or d on page 2-61
Coefficients a or d
This section describes how to write the coefficients a or d in the equation
d c au f
u
u
c
c
V V
( )
+ =
t
,
or in similar equations. a and d are N-by-N matrices, where N is the number
of equations.
Express the coefficients as numbers, text expressions, or functions, as in f
for Systems on page 2-48.
The number of rows in the matrix is either 1, N, N(N+1)/2, or N
2
, as described
in the next few sections. If you choose to express the coefficients in functional
form, the number of columns is Nt, which is the number of triangles in the
mesh. The function should evaluate a or d at the triangle centroids, as in
Scalar PDE Coefficients in Function Form on page 2-23. Give solvers the
function name as a string 'filename', or as a function handle @filename,
where filename.m is a file on your MATLAB path. For details on how to write
the function, see Calculate Coefficients in Function Form on page 2-24.
Often, a or d have structure, either as symmetric or diagonal. In these cases,
you can represent a or d using fewer than N
2
rows.
2-59
2 Setting Up Your PDE
Scalar a or d
The software interprets a scalar a or d as a diagonal matrix.
a
a
a
0 0
0 0
0 0

|
\

|
.
|
|
|
|
N-Element Column Vector a or d
The software interprets an N-element column vector a or d as a diagonal
matrix.
d
d
d N
( )
( )
( )
1 0 0
0 2 0
0 0

|
\

|
.
|
|
|
|
For example, if N = 3, a or d could be
a = char('sin(x) + cos(y)','cosh(x.*y)','x.*y./(1+x.^2+y.^2)') % or d
a =
sin(x) + cos(y)
cosh(x.*y)
x.*y./(1+x.^2+y.^2)
N(N+1)/2-Element Column Vector a or d
The software interprets an N(N+1)/2-element column vector a or d as a
symmetric matrix. In the following diagram, means the entry is symmetric.
2-60
a or d for Systems
a a a a N N
a a a N N
a a N
( ) ( ) ( ) ( ( ) / )
( ) ( ) ( ( ) / )
( ) ( (
1 2 4 1 2
3 5 1 2 1
6

+
NN
a N N
+
+
|
\

|
.
|
|
|
|
|
|
1 2 2
1 2
) / )
( ( ) / )


Coefficient a(i,j) is in row (j(j1)/2+i) of the vector a.
N
2
-Element Column Vector a or d
The software interprets an N
2
-element column vector a or d as a matrix.
d d N d N N
d d N d N N
d N d N d N
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) (
1 1 1
2 2 2
2
2
2
2
+ +
+ +

))
|
\

|
.
|
|
|
|
|
Coefficient a(i,j) is in row (N(j1)+i) of the vector a.
Related
Examples
f for Systems on page 2-48
c for Systems on page 2-50
Deflection of a Piezoelectric Actuator on page 3-19
2-61
2 Setting Up Your PDE
Initial Conditions
Initial conditions has two meanings:
For the parabolic and hyperbolic solvers, the initial condition u0 is the
solution u at the initial time. You must specify the initial condition for
these solvers. Pass the initial condition in the first argument or arguments.
u = parabolic(u0,...
or
u = hyperbolic(u0,ut0,...
For the hyperbolic solver, you must also specify ut0, which is the value
of the derivative of u with respect to time at the initial time. ut0 has the
same form as u0.
For nonlinear elliptic problems, the initial condition u0 is a guess or
approximation of the solution u at the initial iteration of the pdenonlin
nonlinear solver. You pass u0 in the 'U0' name-value pair.
u = pdenonlin(b,p,e,t,c,a,f,'U0',u0)
If you do not specify initial conditions, pdenonlin uses the zero function for
the initial iteration.
Pass u0 as a column vector of values at the points p in the usual p, t, e mesh.
See Mesh Data on page 2-85. You can also pass a scalar, which means the
initial condition is a constant value.
Tip For reliability, the initial conditions and boundary conditions should
be consistent.
The size of the column vector u0 depends on the number of equations, N, and
on the number of points in the mesh, N
p
.
For scalar u, specify a column vector of length N
p
. The value of element k
corresponds to the point p(k).
2-62
Initial Conditions
For a system of N equations, specify a column vector of N*N
p
elements.
The first N
p
elements contain the values of component 1, where the value
of element k corresponds to point p(k). The next N
p
points contain the
values of component 2, etc. It can be convenient to first represent the initial
conditions u0 as an N
p
-by-N matrix, where the first column contains entries for
component 1, the second column contains entries for component 2, etc. The
final representation of the initial conditions is u0(:).
For example, suppose you have a function myfun(x,y) that calculates the
value of the initial condition u0(x,y) as a row vector of length N. Suppose
that p is the usual mesh point data (see Mesh Data on page 2-85). Compute
the initial conditions for all mesh points p.
% Assume N and p exist; N = 1 for a scalar problem
np = size(p,2); % Number of mesh points
u0 = zeros(np,N); % Allocate initial matrix
for k = 1:np
x = p(1,k);
y = p(2,k);
u0(k,:) = myfun(x,y); % Fill in row k
end
u0 = u0(:); % Convert to column form
Specify u0 as the initial condition.
For the parabolic and hyperbolic solvers with scalar problems, you can
also specify text expressions for the initial conditions. The initial conditions
are functions of x and y alone.
For example, if you have an initial condition
u x y
xy x
x y
( , )
cos( )
, =
+ + 1
2 2
then you can use this expression for the initial condition.
'x.*y.*cos(x)./(1+x.^2+y.^2)'
See Also hyperbolic | parabolic | pdenonlin
2-63
2 Setting Up Your PDE
Concepts Mesh Data on page 2-85
2-64
Types of Boundary Conditions
Types of Boundary Conditions
The general mixed-boundary conditions for PDE systems (see Systems of
PDEs on page 2-13) are
hu r
n c qu g h u
=

( )
+ = + ' V .
The notation n c u
( )
V means the N-by-1 matrix with (i,1)-component
cos( ) cos( ) sin( ) sin(
, , , , , , , , ,
o o o c
x
c
y
c
x
i j i j i j 1 1 1 2 2 1
c
c
+
c
c
+
c
c
+ oo)
, , , ,
c
y
u
i j
j
N
j 2 2
1
c
c
|
\

|
.
|
=

where the outward normal vector of the boundary n = ( ) cos( ),sin( ) o o . For
each edge segment, there are M Dirichlet conditions and the h-matrix is
M-by-N, M 0. The generalized Neumann condition contains a source ' h
where the solver computes Lagrange multipliers such that the Dirichlet
conditions are satisfied.
Boundary Conditions Overview on page 2-71 describes a recommended
approach for writing boundary conditions.
The problem of how to specify boundary conditions is somewhat simpler in
the case of scalar solutions. Find the directions for your specific problem in
Boundary Conditions for Scalar PDE on page 2-72 or Boundary Conditions
for PDE Systems on page 2-77.
2-65
2 Setting Up Your PDE
No Boundary Conditions Between Subdomains
There are two types of boundaries:
Boundaries between the interior of the region and the exterior of the region
Boundaries between subdomainsthese are boundaries in the interior
of the region
Boundary conditions, either Dirichlet or generalized Neumann, apply only to
boundaries between the interior and exterior of the region. This is because
the toolbox formulation uses the weak form of PDEs; see Finite Element
Method (FEM) Basics on page 1-32. In the weak formulation you do not
specify boundary conditions between subdomains, even if coefficients are
discontinuous between subdomains. So both ways of specifying boundary
conditions, boundary matrix and boundary file, do not support defining
boundary conditions on subdomain boundaries.
For example, look at a rectangular region with a circular subdomain. The red
numbers are the subdomain labels, the black numbers are the edge segment
labels.
2-66
No Boundary Conditions Between Subdomains
Code for generating the figure
% Rectangle is code 3, 4 sides,
% followed by x-coordinates and then y-coordinates
R1 = [3,4,-1,1,1,-1,-.4,-.4,.4,.4]';
% Circle is code 1, center (.5,0), radius .2
C1 = [1,.5,0,.2]';
% Pad C1 with zeros to enable concatenation with R1
C1 = [C1;zeros(length(R1)-length(C1),1)];
geom = [R1,C1];
% Names for the two geometric objects
ns = (char('R1','C1'))';
2-67
2 Setting Up Your PDE
% Set formula
sf = 'R1+C1';
% Create geometry
gd = decsg(geom,sf,ns);
% View geometry
pdegplot(gd,'edgeLabels','on','subdomainLabels','on')
xlim([-1.1 1.1])
axis equal
You need not give boundary conditions on segments 5, 6, 7, and 8, because
these are subdomain boundaries, not exterior boundaries.
However, if the circle is a hole, meaning it is not part of the region, then you
do give boundary conditions on segments 5, 6, 7, and 8. For an example, see
Scalar PDE Functional Form and Calling Syntax on page 2-26.
2-68
Identify Boundary Labels
Identify Boundary Labels
You can see the edge labels by using the pdegplot function with the
edgeLabels name-value pair set to 'on':
pdegplot(g,'edgeLabels','on')
For example, look at the edge labels for a simple annulus geometry:
e1 = [4;0;0;1;.5;0]; % Outside ellipse
e2 = [4;0;0;.5;.25;0]; % Inside ellipse
ee = [e1 e2]; % Both ellipses
lbls = char('outside','inside'); % Ellipse labels
lbls = lbls'; % Change to columns
sf = 'outside-inside'; % Set formula
dl = decsg(ee,sf,lbls); % Geometry now done
pdegplot(dl,'edgeLabels','on')
2-69
2 Setting Up Your PDE
2-70
Boundary Conditions Overview
Boundary Conditions Overview
There are two ways to specify boundary conditions:
Boundary matrix
Boundary file
Except for the simplest cases, the easiest way to specify boundary conditions
at the command line is with a boundary file (see pdebound). Write a function
file, say pdebound.m, with the following syntax:
[qmatrix,gmatrix,hmatrix,rmatrix] = pdebound(p,e,u,time)
Your function returns matrices qmatrix, gmatrix, hmatrix, and rmatrix,
based on these inputs:
p Points in the mesh (Mesh Data on page 2-85)
e Finite element edges in the mesh, a subset of all the edges (Mesh
Data on page 2-85)
u Solution of the PDE
time Time, for parabolic or hyperbolic PDE only
If your boundary conditions do not depend on u or time, those inputs are
[]. If your boundary conditions do depend on u or time, then when u or
time are NaN, ensure that the outputs such as qmatrix consist of matrices of
NaN of the correct size. This signals to solvers, such as parabolic, to use a
time-dependent or solution-dependent algorithm.
Before specifying boundary conditions, you need to know the boundary labels.
See Identify Boundary Labels on page 2-69.
Alternatively, to generate a boundary matrix, use the PDE app to draw your
geometry and specify your boundary conditions, and then export the boundary
conditions as a boundary matrix.
2-71
2 Setting Up Your PDE
Boundary Conditions for Scalar PDE
For a scalar PDE, some boundary segments can have Dirichlet conditions, and
some boundary segments can have generalized Neumann conditions.
Dirichlet boundary conditions are
hu = r,
where h and r can be functions of x, y, the solution u, the edge segment index,
and, for parabolic and hyperbolic equations, time.
Generalized Neumann boundary conditions are

n c u qu g V
( )
+ = on .

n is the outward unit normal. g and q are functions defined on , and can
be functions of x, y, the solution u, the edge segment index, and, for parabolic
and hyperbolic equations, time.
The PDE solver, such as assempde or adaptmesh, passes a matrix p of points
and e of edges. e has seven rows and ne columns, where you do not necessarily
know in advance the size ne.
p is a 2-by-N
p
matrix, where p(1,k) is the x-coordinate of point k, and
p(2,k) is the y-coordinate of point k.
e is a 7-by-ne matrix, where
- e(1,k) is the index of the first point of edge k.
- e(2,k) is the index of the second point of edge k.
- e(5,k) is the label of the geometry edge of edge k (see Identify
Boundary Labels on page 2-69).
e contains an entry for every finite element edge that lies on an exterior
boundary.
Use the following template for your boundary file.
function [qmatrix,gmatrix,hmatrix,rmatrix] = pdebound(p,e,u,time)
ne = size(e,2); % number of edges
2-72
Boundary Conditions for Scalar PDE
qmatrix = zeros(1,ne);
gmatrix = qmatrix;
hmatrix = zeros(1,2*ne);
rmatrix = hmatrix;
for k = 1:ne
x1 = p(1,e(1,k)); % x at first point in segment
x2 = p(1,e(2,k)); % x at second point in segment
xm = (x1 + x2)/2; % x at segment midpoint
y1 = p(2,e(1,k)); % y at first point in segment
y2 = p(2,e(2,k)); % y at second point in segment
ym = (y1 + y2)/2; % y at segment midpoint
switch e(5,k)
case {some_edge_labels}
% Fill in hmatrix,rmatrix or qmatrix,gmatrix
case {another_list_of_edge_labels}
% Fill in hmatrix,rmatrix or qmatrix,gmatrix
otherwise
% Fill in hmatrix,rmatrix or qmatrix,gmatrix
end
end
For each column k in e, entry k of rmatrix is the value of rmatrix at the first
point in the edge, and entry ne + k is the value at the second point in the edge.
For example, if r = x
2
+ y
4
, then write these lines:
rmatrix(k) = x1^2 + y1^4;
rmatrix(k+ne) = x2^2 + y2^4;
The syntax for hmatrix is identical: entry k of hmatrix is the value of r at
the first point in the edge, and entry k + ne is the value at the second point
in the edge.
For each column k in e, entry k of qmatrix is the value of qmatrix at the
midpoint in the edge. For example, if q = x
2
+ y
4
, then write these lines:
qmatrix(k) = xm^2 + ym^4;
The syntax for gmatrix is identical: entry k of gmatrix is the value of gmatrix
at the midpoint in the edge.
2-73
2 Setting Up Your PDE
If the coefficients depend on the solution u, use the element u(e(1,k)) as the
solution value at the first point of edge k, and u(e(2,k)) as the solution
value at the second point of edge k.
For example, consider the following geometry, a rectangle with a circular hole.
Code for generating the figure
% Rectangle is code 3, 4 sides,
% followed by x-coordinates and then y-coordinates
R1 = [3,4,-1,1,1,-1,-.4,-.4,.4,.4]';
% Circle is code 1, center (.5,0), radius .2
C1 = [1,.5,0,.2]';
2-74
Boundary Conditions for Scalar PDE
% Pad C1 with zeros to enable concatenation with R1
C1 = [C1;zeros(length(R1)-length(C1),1)];
geom = [R1,C1];
% Names for the two geometric objects
ns = (char('R1','C1'))';
% Set formula
sf = 'R1-C1';
% Create geometry
gd = decsg(geom,sf,ns);
% View geometry
pdegplot(gd,'edgeLabels','on')
xlim([-1.1 1.1])
axis equal
Suppose the boundary conditions on the outer boundary (segments 1 through
4) are Dirichlet, with the value u(x,y) = t(x y), where t is time. Suppose
the circular boundary (segments 5 through 8) has a generalized Neumann
condition, with q = 1 and g = x
2
+ y
2
.
Write the following boundary file to represent the boundary conditions:
function [qmatrix,gmatrix,hmatrix,rmatrix] = pdebound(p,e,u,time)
ne = size(e,2); % number of edges
qmatrix = zeros(1,ne);
gmatrix = qmatrix;
hmatrix = zeros(1,2*ne);
rmatrix = hmatrix;
for k = 1:ne
x1 = p(1,e(1,k)); % x at first point in segment
x2 = p(1,e(2,k)); % x at second point in segment
xm = (x1 + x2)/2; % x at segment midpoint
y1 = p(2,e(1,k)); % y at first point in segment
y2 = p(2,e(2,k)); % y at second point in segment
ym = (y1 + y2)/2; % y at segment midpoint
2-75
2 Setting Up Your PDE
switch e(5,k)
case {1,2,3,4} % rectangle boundaries
hmatrix(k) = 1;
hmatrix(k+ne) = 1;
rmatrix(k) = time*(x1 - y1);
rmatrix(k+ne) = time*(x2 - y2);
otherwise % same as case {5,6,7,8}, circle boundaries
qmatrix(k) = 1;
gmatrix(k) = xm^2 + ym^2;
end
end
Related
Examples
Scalar PDE Functional Form and Calling Syntax on page 2-26
Deflection of a Piezoelectric Actuator on page 3-19
2-76
Boundary Conditions for PDE Systems
Boundary Conditions for PDE Systems
The general mixed-boundary conditions for PDE systems (see Systems of
PDEs on page 2-13) are
hu r
n c qu g h u
=

( )
+ = + ' V .
The notation n c u
( )
V means the N-by-1 matrix with (i,1)-component
cos( ) cos( ) sin( ) sin(
, , , , , , , , ,
o o o c
x
c
y
c
x
i j i j i j 1 1 1 2 2 1
c
c
+
c
c
+
c
c
+ oo)
, , , ,
c
y
u
i j
j
N
j 2 2
1
c
c
|
\

|
.
|
=

where the outward normal vector of the boundary n = ( ) cos( ),sin( ) o o . For
each edge segment there are M Dirichlet conditions and the h-matrix is
M-by-N, M 0. The generalized Neumann condition contains a source ' h
where the solver computes Lagrange multipliers such that the Dirichlet
conditions are satisfied.
A PDE solver, such as assempde or adaptmesh, passes a matrix p of points and
e of edges. e has seven rows and ne columns, where you do not necessarily
know in advance the size ne.
p is a 2-by-N
p
matrix, where p(1,k) is the x-coordinate of point k, and
p(2,k) is the y-coordinate of point k.
e is a 7-by-ne matrix, where
- e(1,k) is the index of the first point of edge k.
- e(2,k) is the index of the second point of edge k.
- e(5,k) is the label of the geometry edge of edge k (see Identify
Boundary Labels on page 2-69).
e contains an entry for every finite element edge that lies on an exterior
boundary.
2-77
2 Setting Up Your PDE
Let N be the dimension of the system of PDEs; see Systems of PDEs on page
2-13. Use the following template for your boundary file.
function [qmatrix,gmatrix,hmatrix,rmatrix] = pdebound(p,e,u,time)
N = 3; % Set N = the number of equations
ne = size(e,2); % number of edges
qmatrix = zeros(N^2,ne);
gmatrix = zeros(N,ne);
hmatrix = zeros(N^2,2*ne);
rmatrix = zeros(N,2*ne);
for k = 1:ne
x1 = p(1,e(1,k)); % x at first point in segment
x2 = p(1,e(2,k)); % x at second point in segment
xm = (x1 + x2)/2; % x at segment midpoint
y1 = p(2,e(1,k)); % y at first point in segment
y2 = p(2,e(2,k)); % y at second point in segment
ym = (y1 + y2)/2; % y at segment midpoint
switch e(5,k)
case {some_edge_labels}
% Fill in hmatrix,rmatrix or qmatrix,gmatrix
case {another_list_of_edge_labels}
% Fill in hmatrix,rmatrix or qmatrix,gmatrix
otherwise
% Fill in hmatrix,rmatrix or qmatrix,gmatrix
end
end
For the boundary file, you represent the matrix h for each edge segment as a
vector, taking the matrix column-wise, as hmatrix(:). Column k of hmatrix
corresponds to the matrix at the first edge point e(1,k), and column k + ne
corresponds to the matrix at the second edge point e(2,k).
Similarly, you represent each vector r for an edge as a column in the matrix
rmatrix. Column k corresponds to the vector at the first edge point e(1,k),
and column k + ne corresponds to the vector at the second edge point e(2,k).
2-78
Boundary Conditions for PDE Systems
Represent the entries for the matrix q for each edge segment as a vector,
qmatrix(:), similar to the matrix hmatrix(:). Similarly, represent g for
each edge segment is a column vector in the matrix gmatrix. Unlike h and
r, which have two columns for each segment, q and g have just one column
for each segment, which is the value of the function at the midpoint of the
edge segment.
For example, consider the following geometry, a rectangle with a circular hole.
Code for generating the figure
% Rectangle is code 3, 4 sides,
% followed by x-coordinates and then y-coordinates
2-79
2 Setting Up Your PDE
R1 = [3,4,-1,1,1,-1,-.4,-.4,.4,.4]';
% Circle is code 1, center (.5,0), radius .2
C1 = [1,.5,0,.2]';
% Pad C1 with zeros to enable concatenation with R1
C1 = [C1;zeros(length(R1)-length(C1),1)];
geom = [R1,C1];
% Names for the two geometric objects
ns = (char('R1','C1'))';
% Set formula
sf = 'R1-C1';
% Create geometry
gd = decsg(geom,sf,ns);
% View geometry
pdegplot(gd,'edgeLabels','on')
xlim([-1.1 1.1])
axis equal
Suppose N = 3. Suppose the boundary conditions are mixed. There is M =
1 Dirichlet condition:
The first component of u = 0 on the rectangular segments (numbers 14).
So h(1,1) = 1 and r(1) = 0 for those segments.
The second components of u = 0 on the circular segments (numbers 58).
So h(2,2) = 1 and r(2) = 0 for those segments.
On the rectangular segments (numbers 14),
q =
|
\

|
.
|
|
|
0 1 1
0 0 0
1 1 0
and
2-80
Boundary Conditions for PDE Systems
g =
+
+
|
\

|
.
|
|
|
|
1
0
1
2
2
x
y
On the circular segments (numbers 58),
q =
+ +
+ +
|
\

|
.
|
|
|
|
0 1 2
0 0 0
1 1 0
2 2
4 4
x y
x y
and
g =
+
|
\

|
.
|
|
|
cos( )
tanh( )
t x
x y
0
Write the following boundary file to represent the boundary conditions:
function [qmatrix,gmatrix,hmatrix,rmatrix] = pdebound(p,e,u,time)
N = 3;
ne = size(e,2); % number of edges
qmatrix = zeros(N^2,ne);
gmatrix = zeros(N,ne);
hmatrix = zeros(N^2,2*ne);
rmatrix = zeros(N,2*ne);
for k = 1:ne
x1 = p(1,e(1,k)); % x at first point in segment
x2 = p(1,e(2,k)); % x at second point in segment
xm = (x1 + x2)/2; % x at segment midpoint
y1 = p(2,e(1,k)); % y at first point in segment
y2 = p(2,e(2,k)); % y at second point in segment
ym = (y1 + y2)/2; % y at segment midpoint
switch e(5,k)
case {1,2,3,4}
hk = zeros(N);
2-81
2 Setting Up Your PDE
hk(1,1) = 1;
hk = hk(:);
hmatrix(:,k) = hk;
hmatrix(:,k+ne) = hk;
rk = zeros(N,1); % Not strictly necessary
rmatrix(:,k) = rk; % These are already 0
rmatrix(:,k+ne) = rk;
qk = zeros(N);
qk(1,2) = 1;
qk(1,3) = 1;
qk(3,1) = 1;
qk(3,2) = 1;
qk = qk(:);
qmatrix(:,k) = qk;
gk = zeros(N,1);
gk(1) = 1+xm^2;
gk(3) = 1+ym^2;
gmatrix(:,k) = gk;
case {5,6,7,8}
hk = zeros(N);
hk(2,2) = 1;
hk = hk(:);
hmatrix(:,k) = hk;
hmatrix(:,k+ne) = hk;
rk = zeros(N,1); % Not strictly necessary
rmatrix(:,k) = rk; % These are already 0
rmatrix(:,k+ne) = rk;
qk = zeros(N);
qk(1,2) = 1+xm^2;
qk(1,3) = 2+ym^2;
qk(3,1) = 1+xm^4;
qk(3,2) = 1+ym^4;
qk = qk(:);
qmatrix(:,k) = qk;
2-82
Boundary Conditions for PDE Systems
gk = zeros(N,1);
gk(1) = cos(pi*xm);
gk(3) = tanh(xm*ym);
gmatrix(:,k) = gk;
end
end
Related
Examples
Scalar PDE Functional Form and Calling Syntax on page 2-26
Deflection of a Piezoelectric Actuator on page 3-19
2-83
2 Setting Up Your PDE
Tooltip Displays for Mesh and Plots
In mesh mode, you can use the mouse to display the node number and the
triangle number at the position where you click. Press the left mouse button
to display the node number on the information line. Use the left mouse button
and the Shift key to display the triangle number on the information line.
In plot mode, you can use the mouse to display the numerical value of the
plotted property at the position where you click. Press the left mouse button
to display the triangle number and the value of the plotted property on the
information line.
The information remains on the information line until you release the mouse
button.
2-84
Mesh Data
Mesh Data
A mesh consists of three data structures:
p (points) is a 2-by-Np matrix of points, where Np is the number of points in
the mesh. Each column p(:,k) consists of the x-coordinate of point k in
p(1,k), and the y-coordinate of point k in p(2,k).
e (edges) is a 7-by-Ne matrix of edges, where Ne is the number of edges in
the mesh. An edge is a pair of points in p containing a boundary between
subdomains, or containing an outer boundary, as follows:
- e(1,k) is the index of the first point in edge k.
- e(2,k) is the index of the second point in edge k.
- e(3,k) is the parameter value at the first point of edge k. The parameter
value is related to arc length along the edge.
- e(4,k) is the parameter value at the second point of edge k.
- e(5,k) is the segment number of the geometry containing the edge.
The segment number is inherited from the edge segments in the base
geometry. You can see geometry segment numbers using the command
pdegplot(geom,'edgeLabels','on').
- e(6,k) is the subdomain number on the left of the edge (subdomain 0 is
the exterior of the geometry), where direction along the edge is given by
increasing parameter value.
- e(7,k) is the subdomain number on the right of the edge.
t (triangles) is a 4-by-Nt matrix of triangles, where Nt is the number of
triangles in the mesh. t(1,k), t(2,k), and t(3,k) contain indices to the
three points in p that form triangle k. The points are in counterclockwise
order. t(4,k) contains the subdomain number of the triangle.
Generate an initial mesh using the initmesh function. Refine the mesh using
the refinemesh function. Improve mesh quality without introducing new
points using the jigglemesh function.
Related
Examples
Scalar PDE Functional Form and Calling Syntax on page 2-26
Solve Poissons Equation on a Unit Disk on page 3-72
2-85
2 Setting Up Your PDE
Adaptive Mesh Refinement
In this section...
Improving Solution Accuracy Using Mesh Refinement on page 2-86
Error Estimate for the FEM Solution on page 2-87
Mesh Refinement Functions on page 2-88
Mesh Refinement Termination Criteria on page 2-88
Improving Solution Accuracy Using Mesh Refinement
Partial Differential Equation Toolbox software has a function for global,
uniform mesh refinement. It divides each triangle into four similar triangles
by creating new corners at the midsides, adjusting for curved boundaries. You
can assess the accuracy of the numerical solution by comparing results from a
sequence of successively refined meshes. If the solution is smooth enough,
more accurate results may be obtained by extrapolation.
The solutions of equations often have geometric features like localized strong
gradients. An example of engineering importance in elasticity is the stress
concentration occurring at reentrant corners such as the MATLAB L-shaped
membrane. Then it is more economical to refine the mesh selectively, i.e.,
only where it is needed. When the selection is based on estimates of errors
in the computed solutions, a posteriori estimates, we speak of adaptive mesh
refinement. See adaptmesh for an example of the computational savings
where global refinement needs more than 6000 elements to compete with an
adaptively refined mesh of 500 elements.
The adaptive refinement generates a sequence of solutions on successively
finer meshes, at each stage selecting and refining those elements that are
judged to contribute most to the error. The process is terminated when the
maximum number of elements is exceeded or when each triangle contributes
less than a preset tolerance. You need to provide an initial mesh, and choose
selection and termination criteria parameters. The initial mesh can be
produced by the initmesh function. The three components of the algorithm
are the error indicator function, which computes an estimate of the element
error contribution, the mesh refiner, which selects and subdivides elements,
and the termination criteria.
2-86
Adaptive Mesh Refinement
Error Estimate for the FEM Solution
The adaptation is a feedback process. As such, it is easily applied to a
larger range of problems than those for which its design was tailored. You
want estimates, selection criteria, etc., to be optimal in the sense of giving
the most accurate solution at fixed cost or lowest computational effort for
a given accuracy. Such results have been proved only for model problems,
but generally, the equidistribution heuristic has been found near optimal.
Element sizes should be chosen such that each element contributes the same
to the error. The theory of adaptive schemes makes use of a priori bounds for
solutions in terms of the source function f. For nonelliptic problems such a
bound may not exist, while the refinement scheme is still well defined and
has been found to work well.
The error indicator function used in the software is an elementwise estimate of
the contribution, based on the work of C. Johnson et al. [5], [6]. For Poissons
equation u = f on , the following error estimate for the FEM-solution
u
h
holds in the L
2
-norm :
V s + ( ) ( ), u u hf D u
h h h
o |
where h = h(x) is the local mesh size, and
D v h
v
n
h
E
( ) .
/
=

(
|
\

|
.
|
|
c
c
e
t
t
t
2
2
1 2
1
The braced quantity is the jump in normal derivative of v across edge ,
h

is the length of edge , and the sum runs over E

i
, the set of all interior
edges of the triangulation. The coefficients and are independent of the
triangulation. This bound is turned into an elementwise error indicator
function E(K) for element K by summing the contributions from its edges. The
final form for the equation
(cu) + au = f
becomes
2-87
2 Setting Up Your PDE
E K h f au h c u
K
h
K
( )
= ( ) + ( )
|
\

|
.
|
V
ec

o |
t t
t
1
2
2
2
1 2
n ,
/
where n
t
is the unit normal of edge and the braced term is the jump in flux
across the element edge. The L
2
norm is computed over the element K. This
error indicator is computed by the pdejmps function.
Mesh Refinement Functions
Partial Differential Equation Toolbox software is geared to elliptic problems.
For reasons of accuracy and ill-conditioning, they require the elements
not to deviate too much from being equilateral. Thus, even at essentially
one-dimensional solution features, such as boundary layers, the refinement
technique must guarantee reasonably shaped triangles.
When an element is refined, new nodes appear on its midsides, and if the
neighbor triangle is not refined in a similar way, it is said to have hanging
nodes. The final triangulation must have no hanging nodes, and they are
removed by splitting neighbor triangles. To avoid further deterioration of
triangle quality in successive generations, the longest edge bisection scheme
Rosenberg-Stenger [8] is used, in which the longest side of a triangle is always
split, whenever any of the sides have hanging nodes. This guarantees that no
angle is ever smaller than half the smallest angle of the original triangulation.
Two selection criteria can be used. One, pdeadworst, refines all elements
with value of the error indicator larger than half the worst of any element.
The other, pdeadgsc, refines all elements with an indicator value exceeding a
user-defined dimensionless tolerance. The comparison with the tolerance is
properly scaled with respect to domain and solution size, etc.
Mesh Refinement Termination Criteria
For smooth solutions, error equidistribution can be achieved by the
pdeadgsc selection if the maximum number of elements is large enough.
The pdeadworst adaptation only terminates when the maximum number of
elements has been exceeded. This mode is natural when the solution exhibits
singularities. The error indicator of the elements next to the singularity may
2-88
Adaptive Mesh Refinement
never vanish, regardless of element size, and equidistribution is too much
to hope for.
2-89
2 Setting Up Your PDE
2-90
3
Solving PDEs
Set Up and Solve Your PDE Problem on page 3-2
Structural Mechanics Plane Stress on page 3-6
Structural Mechanics Plane Strain on page 3-13
Clamped, Square Isotropic Plate With a Uniform Pressure Load on page
3-14
Deflection of a Piezoelectric Actuator on page 3-19
Electrostatics on page 3-34
Magnetostatics on page 3-37
AC Power Electromagnetics on page 3-44
Conductive Media DC on page 3-50
Heat Transfer on page 3-57
Nonlinear Heat Transfer In a Thin Plate on page 3-60
Diffusion on page 3-71
Elliptic PDEs on page 3-72
Parabolic PDEs on page 3-88
Hyperbolic PDEs on page 3-95
Eigenvalue Problems on page 3-100
Vibration Of a Circular Membrane Using The MATLAB eigs Function on
page 3-109
Solve PDEs Programmatically on page 3-113
Solve Poissons Equation on a Grid on page 3-119
3 Solving PDEs
Set Up and Solve Your PDE Problem
The layout of the PDE app represents the sequence of steps you perform to
solve a PDE. Specifically, the order of the PDE app menu and toolbar items
represent these actions you perform:
Note Platform-dependent keyboard accelerators are available for the most
common PDE app activities. Learning to use the accelerator keys may
improve the efficiency of your PDE app sessions.
1 Start the PDE app using pdetool.
At this point, the PDE app is in draw mode, where you can use the four
basic solid objects to draw your Constructive Solid Geometry (CSG) model.
You can also edit the set formula. The solid objects are selected using the
five leftmost buttons (or from the Draw menu).
To the right of the draw mode buttons you find buttons through which you
can access all the functions that you need to define and solve the PDE
problem: define boundary conditions, design the triangular mesh, solve the
PDE, and plot the solution.
2 Use the PDE app as a drawing tool to make a drawing of the 2-D geometry
on which you want to solve your PDE. Make use of the four basic solid
objects and the grid and the snap-to-grid feature. The PDE app starts in
the draw mode, and you can select the type of object that you want to use
by clicking the corresponding button or by using the Draw menu. Combine
the solid objects and the set algebra to build the desired CSG model.
3 Save the geometry to a model file. If you want to continue working using
the same geometry at your next Partial Differential Equation Toolbox
session, simply type the name of the model file at the MATLAB prompt.
The PDE app then starts with the model files solid geometry loaded. If you
save the PDE problem at a later stage of the solution process, the model
file also contains commands to recreate the boundary conditions, the PDE
coefficients, and the mesh.
3-2
Set Up and Solve Your PDE Problem
4 Move to the next step in the PDE solving process by clicking the button.
The outer boundaries of the decomposed geometry are displayed with the
default boundary condition indicated. If the outer boundaries do not match
the geometry of your problem, reenter the draw mode. You can then correct
your CSG model by adding, removing or altering any of the solid objects, or
change the set formula used to evaluate the CSG model.
Note The set formula can only be edited while you are in the draw mode.
If the drawing process resulted in any unwanted subdomain borders,
remove them by using the Remove Subdomain Border or Remove All
Subdomain Borders option from the Boundary menu.
You can now define your problems boundary conditions by selecting
the boundary to change and open a dialog box by double-clicking the
boundary or by using the Specify Boundary Conditions option from
the Boundary menu.
5 Initialize the triangular mesh. Click the button or use the corresponding
Mesh menu option Initialize Mesh. Normally, the mesh algorithms
default parameters generate a good mesh. If necessary, they can be
accessed using the Parameters menu item.
6 If you need a finer mesh, the mesh can be refined by clicking the Refine
button. Clicking the button several times causes a successive refinement
of the mesh. The cost of a very fine mesh is a significant increase in the
number of points where the PDE is solved and, consequently, a significant
increase in the time required to compute the solution. Do not refine unless
it is required to achieve the desired accuracy. For each refinement, the
number of triangles increases by a factor of four. A better way to increase
the accuracy of the solution to elliptic PDE problems is to use the adaptive
solver, which refines the mesh in the areas where the estimated error of
the solution is largest. See the adaptmesh reference page for an example of
how the adaptive solver can solve a Laplace equation with an accuracy that
requires more than 10 times as many triangles when regular refinement
is used.
3-3
3 Solving PDEs
7 Specify the PDE from the PDE Specification dialog box. You can access
that dialog box using the PDE button or the PDE Specification menu
item from the PDE menu.
Note This step can be performed at any time prior to solving the PDE
since it is independent of the CSG model and the boundaries. If the PDE
coefficients are material dependent, they are entered in the PDE mode by
double-clicking the different subdomains.
8 Solve the PDE by clicking the = button or by selecting Solve PDE from
the Solve menu. If you do not want an automatic plot of the solution, or if
you want to change the way the solution is presented, you can do that from
the Plot Selection dialog box prior to solving the PDE. You open the Plot
Selection dialog box by clicking the button with the 3-D solution plot icon or
by selecting the Parameters menu item from the Plot menu.
9 Now, from here you can choose one of several alternatives:
Export the solution and/or the mesh to the MATLAB main workspace
for further analysis.
Visualize other properties of the solution.
Change the PDE and recompute the solution.
Change the mesh and recompute the solution. If you select Initialize
Mesh, the mesh is initialized; if you select Refine Mesh, the current
mesh is refined. From the Mesh menu, you can also jiggle the mesh
and undo previous mesh changes.
Change the boundary conditions. To return to the mode where you can
select boundaries, use the button or the Boundary Mode option
from the Boundary menu.
Change the CSG model. You can reenter the draw mode by selecting
Draw Mode from the Draw menu or by clicking one of the Draw Mode
icons to add another solid object. Back in the draw mode, you are able to
add, change, or delete solid objects and also to alter the set formula.
3-4
Set Up and Solve Your PDE Problem
In addition to the recommended path of actions, there are a number of
shortcuts, which allow you to skip over one or more steps. In general, the
PDE app adds the necessary steps automatically.
If you have not yet defined a CSG model, and leave the draw mode with an
empty model, the PDE app creates an L-shaped geometry with the default
boundary condition and then proceeds to the action called for, performing
all the steps necessary.
If you are in draw mode and click the button to initialize the mesh, the
PDE app first decomposes the geometry using the current set formula and
assigns the default boundary condition to the outer boundaries. After that,
an initial mesh is created.
If you click the refine button to refine the mesh before the mesh has been
initialized, the PDE app first initializes the mesh (and decomposes the
geometry, if you were still in the draw mode).
If you click the = button to solve the PDE and you have not yet created a
mesh, the PDE app initializes a mesh before solving the PDE.
If you select a plot type and choose to plot the solution, the PDE app checks
to see if there is a solution to the current PDE available. If not, the PDE
app first solves the current PDE. The solution is then displayed using the
selected plot options.
If you have not defined your PDE, the PDE app solves the default PDE,
which is Poissons equation:
u = 10.
(This corresponds to the generic elliptic PDE with c = 1, a = 0, and f = 10.)
For the different application modes, different default PDE settings apply.
3-5
3 Solving PDEs
Structural Mechanics Plane Stress
In structural mechanics, the equations relating stress and strain arise from
the balance of forces in the material medium. Plane stress is a condition that
prevails in a flat plate in the x-y plane, loaded only in its own plane and
without z-direction restraint.
The stress-strain relation can then be written, assuming isotropic and
isothermal conditions
o
o
t
v
v
v
v
c
c

x
y
xy
x
y
xy
E
|
\

|
.
|
|
|
=

|
\

|
.
|
|
|
|
|
\

1
1 0
1 0
0 0
1
2
2

|
.
|
|
|
,
where
x
and
y
are the normal stresses in the x and y directions, and
xy
is the
shear stress. The material properties are expressed as a combination of E, the
elastic modulus or Youngs modulus, and , Poissons ratio.
The deformation of the material is described by the displacements in the x
and y directions, u and v, from which the strains are defined as
c
c

x
y
xy
u
x
v
y
u
y
v
x
=
=
= +
c
c
c
c
c
c
c
c
.
The balance of force equations are
=
=
c
c
c
c
c
c
c
c
o
t
t o
x
xy
xy y
x
y
x y
x y
K
K ,
3-6
Structural Mechanics Plane Stress
where K
x
and K
y
are volume forces (body forces).
Combining the preceding relations, we arrive at the displacement equations,
which can be written
V V
( )
= , c u k
where c is a rank four tensor (see c for Systems on page 2-50), which can be
written as four 2-by-2 matrices c
11
, c
12
, c
21
, and c
22
:
c
G
G
c
G
c
G
c
G
G
11
12
21
22
2 0
0
0
0
0
0
0
0 2
=
+ |
\

|
.
|
=
|
\

|
.
|
=
|
\

|
.
|
=
+
|
\

|
.
|
,
where G, the shear modulus, is defined by
G
E
=
+
( )
2 1 v
,
and in turn is defined by

v
v
=

=
|
\

|
.
|
|
2
1
G
K
K
x
y
.
k
are volume forces.
This is an elliptic PDE of system type (u is two-dimensional), but you need
only to set the application mode to Structural Mechanics, Plane Stress
3-7
3 Solving PDEs
and then enter the material-dependent parameters E and and the volume
forces k into the PDE Specification dialog box.
In this mode, you can also solve the eigenvalue problem, which is described by
V V =
=
|
\

|
.
|
( )
.
c d
d
u

u
0
0
, the density, can also be entered using the PDE Specification dialog box.
In the Plot Selection dialog box, the x- and y-displacements, u and v, and
the absolute value of the displacement vector (u, v) can be visualized using
color, contour lines, or z-height, and the displacement vector field (u, v) can
be plotted using arrows or a deformed mesh. In addition, for visualization
using color, contour lines, or height, you can choose from 15 scalar tensor
expressions:
u
u
x
x
=
c
c
u
u
y
y
=
c
c
v
v
x
x
=
c
c
v
v
y
y
=
c
c
exx, the x-direction strain (
x
)
eyy, the y-direction strain (
y
)
exy, the shear strain (
xy
)
sxx, the x-direction stress (
x
)
syy, the y-direction stress (
y
)
sxy, the shear stress (
xy
)
3-8
Structural Mechanics Plane Stress
e1, the first principal strain (
1
)
e2, the second principal strain (
2
)
s1, the first principal stress (
1
)
s2, the second principal stress (
2
)
von Mises, the von Mises effective stress
o o o o
1
2
2
2
1 2
+ .
For a more detailed discussion on the theory of stress-strain relations and
applications of FEM to problems in structural mechanics, see Cook, Robert D.,
David S. Malkus, and Michael E. Plesha, Concepts and Applications of Finite
Element Analysis, 3rd edition, John Wiley & Sons, New York, 1989.
Example
Consider a steel plate that is clamped along a right-angle inset at the
lower-left corner, and pulled along a rounded cut at the upper-right corner.
All other sides are free.
The steel plate has the following properties: Dimension: 1-by-1 meters;
thickness 1 mm; inset is 1/3-by-1/3 meters. The rounded cut runs from (2/3, 1)
to (1, 2/3). Youngs modulus: 196 10
3
(MN/m
2
), Poissons ratio: 0.31.
The curved boundary is subjected to an outward normal load of 500 N/m.
We need to specify a surface traction; we therefore divide by the thickness
1 mm, thus the surface tractions should be set to 0.5 MN/m
2
. We will use
the force unit MN in this example.
We want to compute a number of interesting quantities, such as the x- and
y-direction strains and stresses, the shear stress, and the von Mises effective
stress.
Using the PDE App
Using the PDE app, set the application mode to Structural Mechanics,
Plane Stress.
3-9
3 Solving PDEs
The CSG model can be made very quickly by drawing a polygon with corners
in x=[0 2/3 1 1 1/3 1/3 0] and y=[1 1 2/3 0 0 1/3 1/3] and a circle
with center in x=2/3, y=2/3 and radius 1/3.
The polygon is normally labeled P1 and the circle C1, and the CSG model
of the steel plate is simply P1+C1.
Next, select Boundary Mode to specify the boundary conditions. First,
remove all subdomain borders by selecting Remove All Subdomain
Borders from the Boundary menu. The two boundaries at the inset in the
lower left are clamped, i.e., Dirichlet conditions with zero displacements. The
rounded cut is subject to a Neumann condition with q=0 and g1=0.5*nx,
g2=0.5*ny. The remaining boundaries are free (no normal stress), that is, a
Neumann condition with q=0 and g=0.
The next step is to open the PDE Specification dialog box and enter the PDE
parameters.
The E and (nu) parameters are Youngs modulus and Poissons ratio,
respectively. There are no volume forces, so Kx and Ky are zero. (rho) is
not used in this mode. The material is homogeneous, so the same E and [[
apply to the whole 2-D domain.
Initialize the mesh by clicking the button. If you want, you can refine the
mesh by clicking the Refine button.
3-10
Structural Mechanics Plane Stress
The problem can now be solved by clicking the = button.
A number of different strain and stress properties can be visualized, such as
the displacements u and v, the x- and y-direction strains and stresses, the
shear stress, the von Mises effective stress, and the principal stresses and
strains. All these properties can be selected from pop-up menus in the Plot
Selection dialog box. A combination of scalar and vector properties can be
plotted simultaneously by selecting different properties to be represented by
color, height, vector field arrows, and displacements in a 3-D plot.
Select to plot the von Mises effective stress using color and the displacement
vector field (u,v) using a deformed mesh. Select the Color and Deformed
mesh plot types. To plot the von Mises effective stress, select von Mises from
the pop-up menu in the Color row.
In areas where the gradient of the solution (the stress) is large, you need to
refine the mesh to increase the accuracy of the solution. Select Parameters
from the Solve menu and select the Adaptive mode check box. You can use
the default options for adaptation, which are the Worst triangles triangle
selection method with the Worst triangle fraction set to 0.5. Now solve
the plane stress problem again. Select the Show Mesh option in the Plot
Selection dialog box to see how the mesh is refined in areas where the stress
is large.
3-11
3 Solving PDEs
Visualization of the von Mises Effective Stress and the Displacements Using
Deformed Mesh
3-12
Structural Mechanics Plane Strain
Structural Mechanics Plane Strain
A deformation state where there are no displacements in the z-direction,
and the displacements in the x- and y-directions are functions of x and y
but not z is called plane strain. You can solve plane strain problems with
Partial Differential Equation Toolbox software by setting the application
mode to Structural Mechanics, Plane Strain. The stress-strain relation
is only slightly different from the plane stress case, and the same set of
material parameters is used. The application interfaces are identical for the
two structural mechanics modes.
The places where the plane strain equations differ from the plane stress
equations are:
The parameter in the c tensor is defined as

v
v
=

2
1 2
G .
The von Mises effective stress is computed as
o o v v o o v v
1
2
2
2 2
1 2
2
1 2 2 1 + ( ) +
( )
+
( )
.
Plane strain problems are less common than plane stress problems. An
example is a slice of an underground tunnel that lies along the z-axis. It
deforms in essentially plane strain conditions.
3-13
3 Solving PDEs
Clamped, Square Isotropic Plate With a Uniform Pressure
Load
This example shows how to calculate the deflection of a structural plate acted
on by a pressure loading using the Partial Differential Equation Toolbox.
PDE and Boundary Conditions For A Thin Plate
The partial differential equation for a thin, isotropic plate with a pressure
loading is
where is the bending stiffness of the plate given by
and is the modulus of elasticity, is Poissons ratio, and is the plate
thickness. The transverse deflection of the plate is and is the pressure
load.
The boundary conditions for the clamped boundaries are and
where is the derivative of in a direction normal to the boundary.
The Partial Differential Equation Toolbox cannot directly solve the fourth
order plate equation shown above but this can be converted to the following
two second order partial differential equations.
where is a new dependent variable. However, it is not obvious how to specify
boundary conditions for this second order system. We cannot directly specify
boundary conditions for both and . Instead, we directly prescribe to be
zero and use the following technique to define in such a way to insure that
also equals zero on the boundary. Stiff "springs" that apply a transverse
shear force to the plate edge are distributed along the boundary. The shear
3-14
Clamped, Square Isotropic Plate With a Uniform Pressure Load
force along the boundary due to these springs can be written
where is the normal to the boundary and is the stiffness of the springs.
The value of must be large enough that is approximately zero at all points
on the boundary but not so large that numerical errors result because the
stiffness matrix is ill-conditioned. This expression is a generalized Neumann
boundary condition supported by Partial Differential Equation Toolbox
In the Partial Differential Equation Toolbox definition for an elliptic
system, the and dependent variables are u(1) and u(2). The two second
order partial differential equations can be rewritten as
which is the form supported by the toolbox. The input corresponding to this
formulation is shown in the sections below.
Problem Parameters
E = 1.0e6; % modulus of elasticity
nu = .3; % Poisson's ratio
thick = .1; % plate thickness
len = 10.0; % side length for the square plate
hmax = len/20; % mesh size parameter
D = E*thick^3/(12*(1 - nu^2));
pres = 2; % external pressure
Geometry and Mesh
For a single square, the geometry and mesh are easily defined as shown below.
gdmTrans = [3 4 0 len len 0 0 0 len len];
sf = 'S1';
nsmTrans = 'S1';
g = decsg(gdmTrans', sf, nsmTrans');
[p, e, t] = initmesh(g, 'Hmax', hmax);
Boundary Conditions
b = @boundaryFileClampedPlate;
3-15
3 Solving PDEs
type boundaryFileClampedPlate
function [ q, g, h, r ] = boundaryFileClampedPlate( p, e, u, time )
%BOUNDARYFILECLAMPEDPLATE Boundary conditions for heatTransferThinPlateExam
% [ q, g, h, r ] = BOUNDARYFILECLAMPEDPLATE( p, e, u, time ) returns the
% Neumann BC (q, g) and Dirichlet BC (h, r) matrices for the
% clampedSquarePlateExample example.
% p is the point matrix returned from INITMESH
% e is the edge matrix returned from INITMESH
% u is the solution vector (used only for nonlinear cases)
% time (used only for parabolic and hyperbolic cases)
%
% See also PDEBOUND, INITMESH
% Copyright 2012 The MathWorks, Inc.
N = 2;
ne = size(e,2);
% Apply a shear force along the boundary due to the transverse
% deflection of stiff, distributed springs
k = 1e7; % spring stiffness
q = [0 k 0 0]'*ones(1,ne);
g = zeros(N, ne);
h = zeros(N^2, 2*ne);
r = zeros(N, 2*ne);
end
Coefficient Definition
The documentation for assempde shows the required formats for the a and c
matrices in the section titled "PDE Coefficients for System Case". The most
convenient form for c in this example is from the table where is
the number of differential equations. In this example . The tensor, in
the form of an matrix of submatrices is shown below.
3-16
Clamped, Square Isotropic Plate With a Uniform Pressure Load
The six-row by one-column c matrix is defined below. The entries in the full
a matrix and the f vector follow directly from the definition of the
two-equation system shown above.
c = [1; 0; 1; D; 0; D];
a = [0; 0; 1; 0];
f = [0; pres];
Finite Element and Analytical Solutions
The solution is calculated using the assempde function and the transverse
deflection is plotted using the pdeplot function. For comparison, the
transverse deflection at the plate center is also calculated using an analytical
solution to this problem.
u = assempde(b,p,e,t,c,a,f);
numNodes = size(p,2);
pdeplot(p, e, t, 'xydata', u(1:numNodes), 'contour', 'on');
title 'Transverse Deflection'
numNodes = size(p,2);
fprintf('Transverse deflection at plate center(PDE Toolbox)=%12.4e\n', min(
% compute analytical solution
wMax = -.0138*pres*len^4/(E*thick^3);
fprintf('Transverse deflection at plate center(analytical)=%12.4e\n', wMax)
Transverse deflection at plate center(PDE Toolbox)= -2.7563e-01
Transverse deflection at plate center(analytical)= -2.7600e-01
3-17
3 Solving PDEs
3-18
Deflection of a Piezoelectric Actuator
Deflection of a Piezoelectric Actuator
This example shows how to solve a coupled elasticity-electrostatics problem
using the Partial Differential Equation Toolbox. Piezoelectric materials
deform when a voltage is applied. Conversely, a voltage is produced when a
piezoelectric material is deformed.
Analysis of a piezoelectric part requires the solution of a set of coupled partial
differential equations with deflections and electrical potential as dependent
variables. One of the main objectives of this example is to show how such a
system of coupled partial differential equations can be solved using PDE
Toolbox.
PDE For a Piezoelectric Solid
The elastic behavior of the solid is described by the equilibrium equations
where is the stress tensor and is the body force vector. The electrostatic
behavior of the solid is described by Gauss Law
where is the electric displacement and is the distributed, free charge.
These two PDE systems can be combined into the following single system
In 2D, has the components and and has the components
and .
The constitutive equations for the material define the stress tensor and
electric displacement vector in terms of the strain tensor and electric field.
For a 2D, orthotropic, piezoelectric material under plane stress conditions
these are commonly written as
3-19
3 Solving PDEs
where are the elastic coefficients, are the electrical permittivities, and
are the piezoelectric stress coefficients. The piezoelectric stress coefficients
are written to conform to conventional notation in piezoelectric materials
where the z-direction (3-direction) is aligned with the "poled" direction of
the material. For the 2D analysis, we want the poled direction to be aligned
with the y-axis.
Finally, the strain vector can be written in terms of the x-displacement, ,
and y-displacement, as
and the electric field written in terms of the electrical potential, , as
See reference 2, for example, for a more complete description of the
piezoelectric equations.
The strain-displacement equations and electric field equations above can be
substituted into the constitutive equations to yield a system of equations
for the stresses and electrical displacements in terms of displacement and
electrical potential derivatives. If the resulting equations are substituted into
the PDE system equations, we have a system of equations that involve the
divergence of the displacement and electrical potential derivatives. Arranging
these equations to match the form required by PDE Toolbox will be the topic
for the next section.
Converting the Equations To PDE Toolbox Form
3-20
Deflection of a Piezoelectric Actuator
The PDE Toolbox requires a system of elliptic equations to be expressed in
the form
or in tensor form
where summation is implied by repeated indices. For the 2D piezoelectric
system described above, the PDE Toolbox system vector is
This is an system. The gradient of is given by
The documentation for the function assempde shows that it is convenient to
view the tensor as an matrix of submatrices. The most
convenient form for the input argument for this symmetric, system
has 21 rows in and is described in detail in the PDE Toolbox documentation.
It is repeated here for convenience.
3-21
3 Solving PDEs
For the purposes of mapping terms from constitutive equations to the form
required by PDE Toolbox it is useful to write the tensor and solution
gradient in the following form
From this equation the traditional constitutive coefficients can be mapped to
the form required for the PDE Toolbox matrix. Note the minus sign in the
equations for electric field. This minus must be incorporated into the matrix
to match the PDE Toolbox convention. This is shown explicitly below.
Piezoelectric Bimorph Actuator Model
Now that we have defined the equations for a 2D piezoelectric material,
we are ready to apply these to a specific model. The model is a two-layer
3-22
Deflection of a Piezoelectric Actuator
cantilever beam that has been extensively studied (e.g. refs 1 and 2). It is
defined as a "bimorph" because although both layers are made of the same
Polyvinylidene Fluoride (PVDF) material, in the top layer the polarization
direction points down (minus y direction) and in the bottom layer, it points
up. A schematic of the cantilever beam is shown in the figure below.
This figure is not to scale; the actual thickness/length ratio is 100 so the beam
is very slender. When a voltage is applied between the lower and upper
surfaces of the beam, it deflects in the y-direction; one layer shortens and the
other layer lengthens. Devices of this type can be designed to provide the
required motion or force for different applications.
Geometry and Mesh
The simple two-layer geometry of the beam can be created by defining the
sum of two rectangles.
L = 100e-3; % beam length in meters
H = 1e-3; % overall height of the beam
H2 = H/2; % height of each layer in meters
3-23
3 Solving PDEs
% The two lines below contain the columns of the
% geometry description matrix (GDM) for the two rectangular layers.
% The GDM is the first input argument to decsg and describes the
% basic geometric entities in the model.
topLayer = [3 4 0 L L 0 0 0 H2 H2];
bottomLayer = [3 4 0 L L 0 -H2 -H2 0 0];
gdm = [topLayer; bottomLayer]';
g = decsg(gdm, 'R1+R2', ['R1'; 'R2']');
pdegplot(g); axis equal; title 'Two-layer Piezoelectric Cantilever Beam'
xlabel 'X-coordinate, meters'
ylabel 'Y-coordinate, meters'
% We need a relatively fine mesh with maximum element size roughly equal H/
% to accurately model the bending of the beam.
hmax = H/16;
[p, e, t] = initmesh(g, 'Hmax', hmax, 'MesherVersion', 'R2013a');
Warning: Approximately 51200 triangles will be generated.
3-24
Deflection of a Piezoelectric Actuator
Material Properties
The material in both layers of the beam is Polyvinylidene Fluoride (PVDF), a
thermoplastic polymer with piezoelectric behavior.
E = 2.0e9; % Elastic modulus, N/m^2
NU = 0.29; % Poisson's ratio
G = 0.775e9; % Shear modulus, N/m^2
d31 = 2.2e-11; % Piezoelectric strain coefficients, C/N
d33 = -3.0e-11;
% relative electrical permittivity of the material
3-25
3 Solving PDEs
relPermittivity = 12; % at constant stress
% electrical permittivity of vacuum
permittivityFreeSpace = 8.854187817620e-12; % F/m
C11 = E/(1-NU^2); C12 = NU*C11;
c2d = [C11 C12 0; C12 C11 0; 0 0 G];
pzeD = [0 d31; 0 d33; 0 0];
% The piezoelectric strain coefficients for PVDF are
% given above but the constitutive relations in the
% finite element formulation require the
% piezoelectric stress coefficients. These are calculated on the next
% line (for details see, for example, reference 2).
pzeE = c2d*pzeD;
D_const_stress = [relPermittivity 0; 0 relPermittivity]*permittivityFreeSpa
% Convert dielectric matrix from constant stress to constant strain
D_const_strain = D_const_stress - pzeD'*pzeE;
% As discussed above, it is convenient to view the 21 coefficients
% required by assempde as a 3 x 3 array of 2 x 2 submatrices.
% The cij matrices defined below are the 2 x 2 submatrices in the upper
% triangle of this array.
c11 = [c2d(1,1) c2d(1,3) c2d(3,3)];
c12 = [c2d(1,3) c2d(1,2); c2d(3,3) c2d(2,3)];
c22 = [c2d(3,3) c2d(2,3) c2d(2,2)];
c13 = [pzeE(1,1) pzeE(1,2); pzeE(3,1) pzeE(3,2)];
c23 = [pzeE(3,1) pzeE(3,2); pzeE(2,1) pzeE(2,2)];
c33 = [D_const_strain(1,1) D_const_strain(2,1) D_const_strain(2,2)];
Function To Return C Coefficients
The c-matrix for this N=3 system is symmetric. From the documentation for
assempde, we see that the most convenient form for defining the c-matrix has
21 rows defining the upper triangle of the matrix.
c = @(p, t, u, t0) calcCMatPiezoActuator(p, t, c11, c12, c22, c13, c23, c33
% The function shown below is called by the PDE Toolbox routines to
% return the required 21 entries in the c-matrix.
type calcCMatPiezoActuator
function c = calcCMatPiezoActuator( p, t, c11, c12, c22, c13, c23, c33 )
%CALCCMATPIEZOACTUATOR C-matrix for piezoelectric actuator example
3-26
Deflection of a Piezoelectric Actuator
% c = CALCCMATPIEZOACTUATOR( p, t, c11, c12, c22, c13, c23, c33 )
% returns the 'c' coefficient matrix for the piezoelectric actuator
% example given the point and element matrices along with the
% constitutive submatrices (cij) for the PVDF material.
% Copyright 2012 The MathWorks, Inc.
numElems = size(t,2);
c=zeros(21,numElems);
%
% Although the material in both layers is PVDF, in the top layer
% the polarization direction points down (minus y direction) and in the
% bottom layer, it points up. That is, the top layer has d-coefficients
% that are the negative of those in the bottom layer.
%
% The code below examines the y-location of the centroid of each
% triangular element and assigns the correct material properties to
% element depending on whether it is in the top or bottom layer.
%
ctop = [c11(:); c12(:); c22(:); -c13(:); -c23(:); -c33(:)];
cbot = [c11(:); c12(:); c22(:); c13(:); c23(:); -c33(:)];
% calculate y-coordinate of triangle centers
yCenter=(p(2,t(1,:)) + p(2,t(2,:)) + p(2,t(3,:)))/3;
for i=1:numElems
if(yCenter(i) < 0)
c(:,i) = cbot;
else
c(:,i) = ctop;
end
end
end
Boundary Condition Definition
The function below is referred to as a "boundary file" in the PDE Toolbox
documentation. It returns the appropriate q and g matrices defining the
Neumann boundary conditions and the h and r matrices defining Dirichlet
boundary conditions on each exterior, finite element edge.
3-27
3 Solving PDEs
For this example, the top geometry edge (edge 1) has the voltage prescribed as
100 volts. The bottom geometry edge (edge 2) has the voltage prescribed as 0
volts (i.e. grounded). The left geometry edge (edges 6 and 7) have the u and v
displacements equal zero (i.e. clamped). The stress and charge are zero on
the right geometry edge (i.e. q=0).
V = 100;
b = @(p, e, u, time) boundaryFilePiezoActuator(p, e, V);
type boundaryFilePiezoActuator
function [ q, g, h, r ] = boundaryFilePiezoActuator( p, e, V )
%BOUNDARYFILEPIEZOACTUATOR Boundary conditions for piezoelectric actuator e
% [ q, g, h, r ] = BOUNDARYFILEPIEZOACTUATOR( p, e, V ) returns the
% Neumann BC (q, g) and Dirichlet BC (h, r) matrices for the
% piezoelectric actuator example example.
% p is the point matrix returned from INITMESH
% e is the edge matrix returned from INITMESH
% V is the voltage applied to the top layer of the beam
% Copyright 2012 The MathWorks, Inc.
% $Revision: 1.1.6.3 $ $Date: 2012/10/24 04:26:47 $
N = 3;
ne = size(e,2);
q = zeros(N^2, ne);
g = zeros(N, ne);
h = zeros(N^2, 2*ne);
r = zeros(N, 2*ne);
voltage = V;
for i=1:ne
geomEdgeID = e(5,i);
if(geomEdgeID == 1)
% top geometry edge
% set the voltage to V at both vertices on all element edges
% on this geometry edge
h(9,i) = 1;
h(9,i+ne) = 1;
r(3,i) = voltage;
r(3,i+ne) = voltage;
3-28
Deflection of a Piezoelectric Actuator
elseif(geomEdgeID == 2)
% bottom geometry edge
% set the voltage to zero at both vertices on all element edges
% on this geometry edge
% (the entries in r have already been set to zero above)
h(9,i) = 1;
h(9,i+ne) = 1;
elseif(geomEdgeID == 6 || geomEdgeID == 7)
% left geometry edge
% set the u and v displacements to zero at both vertices on all
% element edges on this geometry edge
% (the entries in r have already been set to zero above)
h(1,i) = 1;
h(1,i+ne) = 1;
h(5,i) = 1;
h(5,i+ne) = 1;
end
end
end
Finite Element Solution
a = 0;
f = [0;0;0];
u = assempde(b, p, e, t, c, a, f);
%
% For display and plotting purposes, it is convenient to reshape the
% solution vector as three columns containing the x-displacement,
% y-displacement, and electrical potential, respectively.
%
n = size(p,2); % number of finite element nodes
uu = reshape(u, n, []);
feTipDeflection = uu(1,2);
fprintf('Finite element tip deflection is: %12.4e\n', feTipDeflection);
varsToPlot = char('X-Deflection, meters', 'Y-Deflection, meters', ...
'Electrical Potential, Volts');
for i=1:size(varsToPlot,1)
figure;
pdeplot(p, e, t, 'xydata', uu(:,i), 'contour', 'on');
3-29
3 Solving PDEs
title(varsToPlot(i,:));
% scale the axes to make it easier to view the contours
axis([0, L, -4*H2, 4*H2]);
xlabel 'X-Coordinate, meters'
ylabel 'Y-Coordinate, meters'
end
Finite element tip deflection is: -3.2772e-05
3-30
Deflection of a Piezoelectric Actuator
3-31
3 Solving PDEs
Analytical Solution
A simple, approximate, analytical solution was obtained for this problem in
reference 1.
tipDeflection = -3*d31*V*L^2/(8*H2^2);
fprintf('Analytical tip deflection is: %12.4e\n', tipDeflection);
Analytical tip deflection is: -3.3000e-05
Summary
3-32
Deflection of a Piezoelectric Actuator
The color contour plots of x-deflection and y-deflection show the standard
behavior of the classical cantilever beam solution. The linear distribution
of voltage through the thickness of the beam is as expected. There is
good agreement between the PDE Toolbox finite element solution and the
analytical solution from reference 1.
Although this example shows a very specific coupled elasticity-electrostatic
model, the general approach here can be used for many other systems of
coupled PDEs. The key to applying PDE Toolbox to these types of coupled
systems is the systematic, multi-step coefficient mapping procedure described
above.
References
1 Hwang, W. S.; Park, H. C; Finite Element Modeling of Piezoelectric Sensors
and Actuators. AIAA Journal, 31(5), pp 930-937, 1993.
2 Pieford, V; Finite Element Modeling of Piezoelectric Active Structures.
PhD Thesis, Universite Libre De Bruxelles, 2001.
3-33
3 Solving PDEs
Electrostatics
Applications involving electrostatics include high voltage apparatus,
electronic devices, and capacitors. The statics implies that the time rate of
change is slow, and that wavelengths are very large compared to the size of
the domain of interest. In electrostatics, the electrostatic scalar potential V is
related to the electric field E by E = V and, using one of Maxwells equations,
D = and the relationship D = E, we arrive at the Poisson equation
(V) = ,
where is the coefficient of dielectricity and is the space charge density.
Note should really be written as
0
, where
0
is the coefficient of
dielectricity or permittivity of vacuum (8.854 10
-12
farad/meter) and is
the relative coefficient of dielectricity that varies among different dielectrics
(1.00059 in air, 2.24 in transformer oil, etc.).
Using the Partial Differential Equation Toolbox electrostatics application
mode, you can solve electrostatic problems modeled by the preceding equation.
The PDE Specification dialog box contains entries for and .
The boundary conditions for electrostatic problems can be of Dirichlet or
Neumann type. For Dirichlet conditions, the electrostatic potential V is
specified on the boundary. For Neumann conditions, the surface charge
n (V) is specified on the boundary.
For visualization of the solution to an electrostatic problem, the plot selections
include the electrostatic potential V, the electric field E, and the electric
displacement field D.
For a more in-depth discussion of problems in electrostatics, see Popovic,
Branko D., Introductory Engineering Electromagnetics, Addison-Wesley,
Reading, MA, 1971.
3-34
Electrostatics
Example
Let us consider the problem of determining the electrostatic potential in an
air-filled quadratic frame, bounded by a square with side length of 0.2 in
the center and by outer limits with side length of 0.5. At the inner boundary,
the electrostatic potential is 1000V. At the outer boundary, the electrostatic
potential is 0V. There is no charge in the domain. This leads to the problem of
solving the Laplace equation
V = 0
with the Dirichlet boundary conditions V = 1000 at the inner boundary, and V
= 0 at the outer boundary.
Using the PDE App
After setting the application mode to Electrostatics, the 2-D area is most
easily drawn by first drawing a square with sides of length 0.2 (use the Snap
option and adjust the grid spacing if necessary). Then draw another square
with sides of length 0.5 using the same center position. The 2-D domain is
then simply SQ2-SQ1, if the first square is named SQ1 and the second square
is named SQ2. Enter the expression into the Set formula edit box, and
proceed to define the boundary conditions. Use Shift+click to select all the
inner boundaries. Then double-click an inner boundary and enter 1000 as the
Dirichlet boundary condition for the inner boundaries.
Next, open the PDE Specification dialog box, and enter 0 into the space charge
density (rho) edit field. The coefficient of dielectricity can be left at 1, since it
does not affect the result as long as it is constant.
Initialize the mesh, and click the = button to solve the equation. Using the
adaptive mode, you can improve the accuracy of the solution by refining
the mesh close to the reentrant corners where the gradients are steep. For
example, use the triangle selection method picking the worst triangles and set
the maximum number of triangles to 500. Add one uniform mesh refinement
by clicking the Refine button once. Finally turn adaptive mode off, and click
the = button once more.
To look at the equipotential lines, select a contour plot from the Plot
Selection dialog box. To display equipotential lines at every 100th volt, enter
0:100:1000 into the Contour plot levels edit box.
3-35
3 Solving PDEs
Equipotential Lines in Air-Filled Frame
3-36
Magnetostatics
Magnetostatics
Magnets, electric motors, and transformers are areas where problems
involving magnetostatics can be found. The statics implies that the time
rate of change is slow, so we start with Maxwells equations for steady cases,
= H J
= B 0
and the relationship
B H =
where B is the magnetic flux density, H is the magnetic field intensity, J is the
current density, and is the materials magnetic permeability.
Since = B 0 , there exists a magnetic vector potential A such that
B A =
and
V V
|
\

|
.
|
=
1

A J
The plane case assumes that the current flows are parallel to the z-axis, so
only the z component of A is present,
A J = = ( , , ), ( , , ) 0 0 0 0 A J
and the preceding equation can be simplified to the scalar elliptic PDE
V V
|
\

|
.
|
= ,
1

A J
where J = J(x,y).
3-37
3 Solving PDEs
For the 2-D case, we can compute the magnetic flux density B as
B =

A
y
A
x
, ,0
and the magnetic field H, in turn, is given by
H B =
1

The interface condition across subdomain borders between regions of different

material properties is that H x n be continuous. This implies the continuity of
1

c
c
A
n
and does not require special treatment since we are using the variational
formulation of the PDE problem.
In ferromagnetic materials, is usually dependent on the field strength |B|
= |A|, so the nonlinear solver is needed.
The Dirichlet boundary condition specifies the value of the magnetostatic
potential A on the boundary. The Neumann condition specifies the value
of the normal component of
n
|
\

|
.
|
V
1

A
on the boundary. This is equivalent to specifying the tangential value of the
magnetic field H on the boundary.
Visualization of the magnetostatic potential A, the magnetic field H, and the
magnetic flux density B is available. B and H can be plotted as vector fields.
3-38
Magnetostatics
For a more detailed discussion on Maxwells equations and magnetostatics,
see Popovic, Branko D., Introductory Engineering Electromagnetics,
Addison-Wesley, Reading, MA, 1971.
Example
As an example of a problem in magnetostatics, consider determining the
static magnetic field due to the stator windings in a two-pole electric motor.
The motor is considered to be long, and when end effects are neglected, a 2-D
computational model suffices.
The domain consists of three regions:
Two ferromagnetic pieces, the stator and the rotor
The air gap between the stator and the rotor
The armature coil carrying the DC current
The magnetic permeability is 1 in the air and in the coil. In the stator and
the rotor, is defined by

=
+ V
+
max
min
.
1
2
c A

max
= 5000,
min
= 200, and c = 0.05 are values that could represent
transformer steel.
The current density J is 0 everywhere except in the coil, where it is 1.
The geometry of the problem makes the magnetic vector potential A
symmetric with respect to y and antisymmetric with respect to x, so you can
limit the domain to x 0,y 0 with the Neumann boundary condition
n
|
\

|
.
|
= V
1
0

A
3-39
3 Solving PDEs
on the x-axis and the Dirichlet boundary condition A = 0 on the y-axis.
The field outside the motor is neglected leading to the Dirichlet boundary
condition A = 0 on the exterior boundary.
Using the PDE App
The geometry is complex, involving five circular arcs and two rectangles.
Using the PDE app, set the x-axis limits to [-1.5 1.5] and the y-axis limits to
[-1 1]. Set the application mode to Magnetostatics, and use a grid spacing
of 0.1. The model is a union of circles and rectangles; the reduction to the first
quadrant is achieved by intersection with a square. Using the snap-to-grid
feature, you can draw the geometry using the mouse, or you can draw it by
entering the following commands:
pdecirc(0,0,1,'C1')
pdecirc(0,0,0.8,'C2')
pdecirc(0,0,0.6,'C3')
pdecirc(0,0,0.5,'C4')
pdecirc(0,0,0.4,'C5')
pderect([-0.2 0.2 0.2 0.9],'R1')
pderect([-0.1 0.1 0.2 0.9],'R2')
pderect([0 1 0 1],'SQ1')
You should get a CSG model similar to the one in the following plot.
3-40
Magnetostatics
Enter the following set formula to reduce the model to the first quadrant:
(C1+C2+C3+C4+C5+R1+R2)*SQ1
In boundary mode you need to remove a number of subdomain borders. Using
Shift+click, select borders and remove them using the Remove Subdomain
Border option from the Boundary menu until the geometry consists of four
subdomains: the stator, the rotor, the coil, and the air gap. In the following
plot, the stator is subdomain 1, the rotor is subdomain 2, the coil is subdomain
3, and the air gap is subdomain 4. The numbering of your subdomains may
be different.
3-41
3 Solving PDEs
Before moving to the PDE mode, select the boundaries along the x-axis and
set the boundary condition to a Neumann condition with g = 0 and q = 0.
In the PDE mode, turn on the labels by selecting the Show Subdomain
Labels option from the PDE menu. Double-click each subdomain to define
the PDE parameters:
In the coil both and J are 1, so the default values do not need to be
changed.
In the stator and the rotor is nonlinear and defined by the preceding
equation. Enter as
5000./(1+0.05*(ux.^2+uy.^2))+200
ux.^2+uy.^2 is equal to |A |
2
. J is 0 (no current).
3-42
Magnetostatics
In the air gap is 1, and J is 0.
Initialize the mesh, and continue by opening the Solve Parameters dialog box
by selecting Parameters from the Solve menu. Since this is a nonlinear
problem, the nonlinear solver must be invoked by checking the Use
nonlinear solver. If you want, you can adjust the tolerance parameter. The
adaptive solver can be used together with the nonlinear solver. Solve the
PDE and plot the magnetic flux density B using arrows and the equipotential
lines of the magnetostatic potential A using a contour plot. The plot clearly
shows, as expected, that the magnetic flux is parallel to the equipotential
lines of the magnetostatic potential.
Equipotential Lines and Magnetic Flux in a Two-Pole Motor
3-43
3 Solving PDEs
AC Power Electromagnetics
AC power electromagnetics problems are found when studying motors,
transformers and conductors carrying alternating currents.
Let us start by considering a homogeneous dielectric, with coefficient of
dielectricity and magnetic permeability , with no charges at any point. The
fields must satisfy a special set of the general Maxwells equations:
V =
c
c
V =
c
c
+
E
H
H
E
J

c
t
t
.
For a more detailed discussion on Maxwells equations, see Popovic, Branko
D., Introductory Engineering Electromagnetics, Addison-Wesley, Reading,
MA, 1971.
In the absence of current, we can eliminate H from the first set and E from
the second set and see that both fields satisfy wave equations with wave
speed c :
A
A
E
H
H
E
=
=
c
c
c
c
c
c
2
2
2
2
0
0
t
t
.
We move on to studying a charge-free homogeneous dielectric, with coefficient
of dielectrics , magnetic permeability , and conductivity . The current
density then is
J E = o
and the waves are damped by the Ohmic resistance,
3-44
AC Power Electromagnetics
AE
E E
=
c
c
c
c
o c
t
t
2
2
0
and similarly for H.
The case of time harmonic fields is treated by using the complex form,
replacing E by
E
c
e
j t
The plane case of this Partial Differential Equation Toolbox mode has
E J
c c
j t
E Je = ( ) =
( )
0 0 0 0 , , , , ,
e
, and the magnetic field
H H H =
( )
=

V
x y c
j
E , , . 0
1
o
The scalar equation for E
c
becomes
V V
|
\

|
.
|
+ ( ) = .
1
0
2

eo e c E j E
c c
This is the equation used by Partial Differential Equation Toolbox software in
the AC power electromagnetics application mode. It is a complex Helmholtzs
equation, describing the propagation of plane electromagnetic waves in
imperfect dielectrics and good conductors ( ). A complex permittivity

c
can be defined as
c
= -j/. The conditions at material interfaces with
abrupt changes of and are the natural ones for the variational formulation
and need no special attention.
The PDE parameters that have to be entered into the PDE Specification
dialog box are the angular frequency , the magnetic permeability , the
conductivity , and the coefficient of dielectricity .
The boundary conditions associated with this mode are a Dirichlet boundary
condition, specifying the value of the electric field E
c
on the boundary, and a
3-45
3 Solving PDEs
Neumann condition, specifying the normal derivative of E
c
. This is equivalent
to specifying the tangential component of the magnetic field H:
H
j
E
t c
=
|
\

|
.
|
V
e
n .
1
Interesting properties that can be computed from the solutionthe electric
field Eare the current density J = E and the magnetic flux density
B E = V
j
e
.
The electric field E, the current density J, the magnetic field H and the
magnetic flux density B are available for plots. Additionally, the resistive
heating rate
Q E
c
=
2
/ o
is also available. The magnetic field and the magnetic flux density can be
plotted as vector fields using arrows.
Example
The example shows the skin effect when AC current is carried by a wire
with circular cross section. The conductivity of copper is 57 10
6
, and the
permeability is 1, i.e., = 410
7
. At the line frequency (50 Hz) the
2
-term
is negligible.
Due to the induction, the current density in the interior of the conductor is
smaller than at the outer surface where it is set to J
S
= 1, a Dirichlet condition
for the electric field, E
c
= 1/. For this case an analytical solution is available,
J J
J kr
J kR
S
=
( )
( )
0
0
,
where
3-46
AC Power Electromagnetics
k j = eo .
R is the radius of the wire, r is the distance from the center line, and J
0
(x) is
the first Bessel function of zeroth order.
Using the PDE App
Start the PDE app and set the application mode toAC Power
Electromagnetics. Draw a circle with radius 0.1 to represent a cross section
of the conductor, and proceed to the boundary mode to define the boundary
condition. Use the Select All option to select all boundaries and enter
1/57E6 into the r edit field in the Boundary Condition dialog box to define the
Dirichlet boundary condition (E = J/).
Open the PDE Specification dialog box and enter the PDE parameters. The
angular frequency = 2 50.
Initialize the mesh and solve the equation. Due to the skin effect, the current
density at the surface of the conductor is much higher than in the conductors
interior. This is clearly visualized by plotting the current density J as a 3-D
plot. To improve the accuracy of the solution close to the surface, you need
to refine the mesh. Open the Solve Parameters dialog box and select the
Adaptive mode check box. Also, set the maximum numbers of triangles to
Inf, the maximum numbers of refinements to 1, and use the triangle selection
method that picks the worst triangles. Recompute the solution several times.
Each time the adaptive solver refines the area with the largest errors. The
3-47
3 Solving PDEs
number of triangles is printed on the command line. The following mesh is
the result of successive adaptations and contains 1548 triangles.
The Adaptively Refined Mesh
The solution of the AC power electromagnetics equation is complex. The
plots show the real part of the solution (a warning message is issued), but
the solution vector, which can be exported to the main workspace, is the full
complex solution. Also, you can plot various properties of the complex solution
by using the user entry. imag(u) and abs(u) are two examples of valid user
entries.
3-48
AC Power Electromagnetics
The skin effect is an AC phenomenon. Decreasing the frequency of the
alternating current results in a decrease of the skin effect. Approaching DC
conditions, the current density is close to uniform (experiment using different
angular frequencies).
The Current Density in an AC Wire
3-49
3 Solving PDEs
Conductive Media DC
For electrolysis and computation of resistances of grounding plates, we have
a conductive medium with conductivity and a steady current. The current
density J is related to the electric field E through J = E. Combining the
continuity equation J = Q, where Q is a current source, with the definition
of the electric potential V yields the elliptic Poissons equation
(V) = Q.
The only two PDE parameters are the conductivity and the current source Q.
The Dirichlet boundary condition assigns values of the electric potential V
to the boundaries, usually metallic conductors. The Neumann boundary
condition requires the value of the normal component of the current density
(n (V)) to be known. It is also possible to specify a generalized Neumann
condition defined by n (V) + qV = g, where q can be interpreted as a film
conductance for thin plates.
The electric potential V, the electric field E, and the current density J are all
available for plotting. Interesting quantities to visualize are the current lines
(the vector field of J) and the equipotential lines of V. The equipotential lines
are orthogonal to the current lines when is isotropic.
Example
Two circular metallic conductors are placed on a plane, thin conductor like
a blotting paper wetted by brine. The equipotentials can be traced by a
voltmeter with a simple probe, and the current lines can be traced by strongly
colored ions, such as permanganate ions.
The physical model for this problem consists of the Laplace equation
(V) = 0
for the electric potential V and the boundary conditions:
V = 1 on the left circular conductor
V = 1 on the right circular conductor
3-50
Conductive Media DC
The natural Neumann boundary condition on the outer boundaries
c
c
=
V
n
0.
The conductivity = 1 (constant).
1 Open the PDE app by typing
pdetool
at the MATLAB command prompt.
2 Click Options > Application > Conductive Media DC.
3 Click Options > Grid Spacing..., deselect the Auto check boxes for
X-axis linear spacing and Y-axis linear spacing, and choose a spacing
of 0.3, as pictured. Ensure the Y-axis goes from 0.9 to 0.9. Click Apply,
and then Done.
3-51
3 Solving PDEs
4 Click Options > Snap
5 Click and draw the blotting paper as a rectangle with corners in
(-1.2,-0.6), (1.2,-0.6), (1.2,0.6), and (-1.2,0.6).
6 Click and add two circles with radius 0.3 that represent the circular
conductors. Place them symmetrically with centers in (-0.6,0) and (0.6,0).
7 To express the 2-D domain of the problem, enter
R1-(C1+C2)
for the Set formula parameter.
8 To decompose the geometry and enter the boundary mode, click .
9 Hold down Shift and click to select the outer boundaries. Double-click the
last boundary to open the Boundary Condition dialog box.
10 Select Neumann and click OK.
11 Hold down Shift and click to select the left circular conductor boundaries.
Double-click the last boundary to open the Boundary Condition dialog box.
12 Set the parameters as follows and then click OK:
Condition type = Dirichlet
3-52
Conductive Media DC
h = 1
r = 1
13 Hold down Shift and click to select the right circular conductor boundaries.
Double-click the last boundary to open the Boundary Condition dialog box.
14 Set the parameters as follows and then click OK:
Condition type = Dirichlet
h = 1
r = -1
3-53
3 Solving PDEs
15 Open the PDE Specification dialog box by clicking PDE > PDE
Specification.
16 Set the current source, q, parameter to 0.
17 Initialize the mesh by clicking Mesh > Initialize Mesh.
18 Refine the mesh by clicking Mesh > Refine Mesh twice.
19 Improve the triangle quality by clicking Mesh > Jiggle Mesh.
20 Solve the PDE by clicking .
The resulting potential is zero along the y-axis, which is a vertical line
of anti-symmetry for this problem.
3-54
Conductive Media DC
21 Visualize the current density
J
by clicking Plot > Parameters, selecting
Contour and Arrows check box, and clicking Plot.
The current flows, as expected, from the conductor with a positive potential
to the conductor with a negative potential.
3-55
3 Solving PDEs
The Current Density Between Two Metallic Conductors
3-56
Heat Transfer
Heat Transfer
The heat equation is a parabolic PDE:
C
T
t
k T Q h T T
c
c
V V
( )
= + ( )
ext
.
It describes the heat transfer process for plane and axisymmetric cases, and
uses the following parameters:
Density
Heat capacity C
Coefficient of heat conduction k
Heat source Q
Convective heat transfer coefficient h
External temperature T
ext
The term h(T
ext
T) is a model of transversal heat transfer from the
surroundings, and it may be useful for modeling heat transfer in thin cooling
plates etc.
For the steady state case, the elliptic version of the heat equation,
V V
( )
= + ( ) k T Q h T T
ext
is also available.
The boundary conditions can be of Dirichlet type, where the temperature on
the boundary is specified, or of Neumann type where the heat flux, n V ( ) k T ,
is specified. A generalized Neumann boundary condition can also be used.
The generalized Neumann boundary condition equation is n V + = ( ) k T qT g ,
where q is the heat transfer coefficient.
3-57
3 Solving PDEs
Visualization of the temperature, the temperature gradient, and the heat
flux kT is available. Plot options include isotherms and heat flux vector
field plots.
Example
In the following example, a heat transfer problem with differing material
parameters is solved.
The problems 2-D domain consists of a square with an embedded diamond (a
square with 45 degrees rotation). The square region consists of a material with
coefficient of heat conduction of 10 and a density of 2. The diamond-shaped
region contains a uniform heat source of 4, and it has a coefficient of heat
conduction of 2 and a density of 1. Both regions have a heat capacity of 0.1.
Using the PDE App
Start the PDE app and set the application mode to Heat Transfer. In draw
mode, set the x- and y-axis limits to [-0.5 3.5] and select the Axis Equal
option from the Options menu. The square region has corners in (0,0), (3,0),
(3,3), and (0,3). The diamond-shaped region has corners in (1.5,0.5), (2.5,1.5),
(1.5,2.5), and (0.5,1.5).
The temperature is kept at 0 on all the outer boundaries, so you do not have
to change the default boundary conditions. Move on to define the PDE
parameters (make sure to set the application mode to Heat Transfer in the
PDE mode by double-clicking each of the two regions and enter the PDE
parameters. You want to solve the parabolic heat equation, so make sure that
the Parabolic option is selected. In the square region, enter a density of 2,
a heat capacity of 0.1, and a coefficient of heat conduction of 10. There is no
heat source, so set it to 0. In the diamond-shaped region, enter a density of 1,
a heat capacity of 0.1, and a coefficient of heat conduction of 2. Enter 4 in the
edit field for the heat source. The transversal heat transfer term h(T
ext
T) is
not used, so set h, the convective heat transfer coefficient, to 0.
Since you are solving a dynamic PDE, you have to define an initial value, and
the times at which you want to solve the PDE. Open the Solve Parameters
dialog box by selecting Parameters from the Solve menu. The dynamics for
this problem is very fastthe temperature reaches steady state in about
0.1 time units. To capture the interesting part of the dynamics, enter
3-58
Heat Transfer
logspace(-2,-1,10) as the vector of times at which to solve the heat
equation. logspace(-2,-1,10) gives 10 logarithmically spaced numbers
between 0.01 and 0.1. Set the initial value of the temperature to 0. If the
boundary conditions and the initial value differ, the problem formulation
contains discontinuities.
Solve the PDE. By default, the temperature distribution at the last time is
plotted. The best way to visualize the dynamic behavior of the temperature
is to animate the solution. When animating, turn on the Height (3-D plot)
option to animate a 3-D plot. Also, select the Plot in x-y grid option. Using a
rectangular grid instead of a triangular grid speeds up the animation process
significantly.
Other interesting visualizations are made by plotting isothermal lines using a
contour plot, and by plotting the heat flux vector field using arrows.
Visualization of the Temperature and the Heat Flux
3-59
3 Solving PDEs
Nonlinear Heat Transfer In a Thin Plate
This example shows how to perform a heat transfer analysis of a thin plate
using the Partial Differential Equation Toolbox.
The plate is square and the temperature is fixed along the bottom edge. No
heat is transferred from the other three edges (i.e. they are insulated). Heat
is transferred from both the top and bottom faces of the plate by convection
and radiation. Because radiation is included, the problem is nonlinear. One
of the purposes of this example is to show how to handle nonlinearities in
PDE problems.
Both a steady state and a transient analysis are performed. In a steady state
analysis we are interested in the final temperature at different points in the
plate after it has reached an equilibrium state. In a transient analysis we are
interested in the temperature in the plate as a function of time. One question
that can be answered by this transient analysis is how long does it take for
the plate to reach an equilibrium temperature.
Heat Transfer Equations for the Plate
The plate has planar dimensions one meter by one meter and is 1 cm thick.
Because the plate is relatively thin compared with the planar dimensions,
the temperature can be assumed constant in the thickness direction; the
resulting problem is 2D.
Convection and radiation heat transfer are assumed to take place between
the two faces of the plate and a specified ambient temperature.
The amount of heat transferred from each plate face per unit area due to
convection is defined as
where is the ambient temperature, is the temperature at a particular x
and y location on the plate surface, and is a specified convection coefficient.
The amount of heat transferred from each plate face per unit area due to
radiation is defined as
3-60
Nonlinear Heat Transfer In a Thin Plate
where is the emissivity of the face and is the Stefan-Boltzmann constant.
Because the heat transferred due to radiation is proportional to the fourth
power of the surface temperature, the problem is nonlinear.
The PDE describing the temperature in this thin plate is
where is the material density, is the specific heat, is the plate thickness,
and the factors of two account for the heat transfer from both plate faces.
It is convenient to rewrite this equation in the form expected by PDE Toolbox
Problem Parameters
The plate is composed of copper which has the following properties
k = 400; % thermal conductivity of copper, W/(m-K)
rho = 8960; % density of copper, kg/m^3
specificHeat = 386; % specific heat of copper, J/(kg-K)
thick = .01; % plate thickness in meters
stefanBoltz = 5.670373e-8; % Stefan-Boltzmann constant, W/(m^2-K^4)
hCoeff = 1; % Convection coefficient, W/(m^2-K)
% The ambient temperature is assumed to be 300 degrees-Kelvin.
ta = 300;
emiss = .5; % emissivity of the plate surface
Definition of PDE Coefficients
The expressions for the coefficients required by PDE Toolbox can easily
be identified by comparing the equation above with the scalar parabolic
equation in the PDE Toolbox documentation.
c = thick*k;
3-61
3 Solving PDEs
% Because of the radiation boundary condition, the "a" coefficient
% is a function of the temperature, u. It is defined as a MATLAB
% expression so it can be evaluated for different values of u
% during the analysis.
a = sprintf('2*%g + 2*%g*%g*u.^3', hCoeff, emiss, stefanBoltz)
f = 2*hCoeff*ta + 2*emiss*stefanBoltz*ta^4;
d = thick*rho*specificHeat;
a =
2*1 + 2*0.5*5.67037e-08*u.^3
Geometry and Mesh
For a square, the geometry and mesh are easily defined as shown below.
width = 1; height = 1;
% define the square by giving the 4 x-locations followed by the 4
% y-locations of the corners.
gdmTrans = [3 4 0 width width 0 0 0 height height];
g = decsg(gdmTrans', 'S1', ('S1')');
% Create the triangular mesh on the square with approximately
% ten elements in each direction.
hmax = .1; % element size
[p, e, t] = initmesh(g, 'Hmax', hmax);
pdeplot(p,e,t);
title 'Plate With Triangular Element Mesh'
xlabel 'X-coordinate, meters'
ylabel 'Y-coordinate, meters'
3-62
Nonlinear Heat Transfer In a Thin Plate
Boundary Conditions
The bottom edge of the plate is set to 1000 degrees-Kelvin.
The boundary conditions are defined in the function below which is referred
to in the PDE Toolbox documentation as a boundary file. Three of the plate
edges are insulated. Because a Neumann boundary condition equal zero is
the default in the finite element formulation, the boundary conditions on
these edges do not need to be set explicitly. A Dirichlet condition is set on
all nodes on the bottom edge, edge 1,
3-63
3 Solving PDEs
b=@boundaryFileThinPlate;
type boundaryFileThinPlate
function [ q, g, h, r ] = boundaryFileThinPlate( p, e, u, time )
%BOUNDARYFILETHINPLATE Boundary conditions for heatTransferThinPlateExample
% [ q, g, h, r ] = BOUNDARYFILETHINPLATE( p, e, u, time ) returns the
% Neumann BC (q, g) and Dirichlet BC (h, r) matrices for the
% heatTransferThinPlateExample example.
% p is the point matrix returned from INITMESH
% e is the edge matrix returned from INITMESH
% u is the solution vector (used only for nonlinear cases)
% time (used only for parabolic and hyperbolic cases)
%
% See also PDEBOUND, INITMESH
% Copyright 2012 The MathWorks, Inc.
% $Revision: 1.1.6.2 $ $Date: 2012/08/21 01:12:25 $
N = 1;
ne = size(e,2);
q = zeros(N^2, ne);
g = zeros(N, ne);
h = zeros(N^2, 2*ne);
r = zeros(N, 2*ne);
for i=1:ne
ei = e(5,i);
if(ei == 1)
% Set the temperature at both vertices on the edge
h(1,i) = 1;
h(1,i+ne) = 1;
r(1,i) = 1000;
r(1,i+ne) = 1000;
end
end
end
3-64
Nonlinear Heat Transfer In a Thin Plate
Steady State Solution
Because the a and f coefficients are functions of temperature (due to the
radiation boundary conditions), the nonlinear solver pdenonlin must be used
to obtain the solution.
u = pdenonlin(b,p,e,t,c,a,f, 'jacobian', 'lumped');
figure;
pdeplot(p, e, t, 'xydata', u, 'contour', 'on', 'colormap', 'jet');
title 'Temperature In The Plate, Steady State Solution'
xlabel 'X-coordinate, meters'
ylabel 'Y-coordinate, meters'
plotAlongY(p, u, 0);
title 'Temperature As a Function of the Y-Coordinate'
xlabel 'X-coordinate, meters'
ylabel 'Temperature, degrees-Kelvin'
fprintf('Temperature at the top edge of the plate = %5.1f degrees-K\n', .
u(4));
Temperature at the top edge of the plate = 448.9 degrees-K
3-65
3 Solving PDEs
3-66
Nonlinear Heat Transfer In a Thin Plate
Transient Solution
endTime = 5000;
tlist = 0:50:endTime;
numNodes = size(p,2);
% Set the initial temperature of all nodes to ambient, 300 K
u0(1:numNodes) = 300;
% Find all nodes along the bottom edge and set their initial temperature
% to the value of the constant BC, 1000 K
nodesY0 = abs(p(2,:)) < 1.0e-5;
u0(nodesY0) = 1000;
3-67
3 Solving PDEs
rtol = 1.0e-3; atol = 1.0e-4;
% The transient solver parabolic automatically handles both linear
% and nonlinear problems, such as this one.
u = parabolic(u0, tlist, b,p,e,t,c,a,f,d,rtol,atol);
figure;
plot(tlist, u(3, :)); grid;
title 'Temperature Along the Top Edge of the Plate as a Function of Time'
xlabel 'Time, seconds'
ylabel 'Temperature, degrees-Kelvin'
%
figure;
pdeplot(p, e, t, 'xydata', u(:,end), 'contour', 'on', 'colormap', 'jet');
title(sprintf('Temperature In The Plate, Transient Solution( %d seconds)\
tlist(1,end)));
xlabel 'X-coordinate, meters'
ylabel 'Y-coordinate, meters'
%
fprintf('\nTemperature at the top edge of the plate(t=%5.1f secs) = %5.1f
tlist(1,end), u(4,end));
65 successful steps
0 failed attempts
95 function evaluations
1 partial derivatives
16 LU decompositions
94 solutions of linear systems
Temperature at the top edge of the plate(t=5000.0 secs) = 447.2 degrees-K
3-68
Nonlinear Heat Transfer In a Thin Plate
3-69
3 Solving PDEs
Summary
As can be seen, the plots of temperature in the plate from the steady state
and transient solution at the ending time are very close. That is, after around
5000 seconds, the transient solution has reached the steady state values.
The temperatures from the two solutions at the top edge of the plate agree
to within one percent.
3-70
Diffusion
Diffusion
Since heat transfer is a diffusion process, the generic diffusion equation has
the same structure as the heat equation:
c
c
V V
( )
=
c
t
c D Q ,
where c is the concentration, D is the diffusion coefficient and Q is a volume
source. If diffusion process is anisotropic, in which case D is a 2-by-2 matrix,
you must solve the diffusion equation using the generic system application
mode of the PDE app. For more information, see PDE Menu on page 4-19.
The boundary conditions can be of Dirichlet type, where the concentration on
the boundary is specified, or of Neumann type, where the flux, n V ( ) D c , is
specified. It is also possible to specify a generalized Neumann condition. It is
defined by n V + = ( ) D c qc g , where q is a transfer coefficient.
Visualization of the concentration, its gradient, and the flux is available from
the Plot Selection dialog box.
3-71
3 Solving PDEs
Elliptic PDEs
This topic describes the solution of some elliptic PDE problems. The last
problem, a minimal surface problem, is nonlinear and illustrates the use of
the nonlinear solver. The problems are solved using both the PDE app and
command-line functions. The topics include:
In this section...
Solve Poissons Equation on a Unit Disk on page 3-72
Scattering Problem on page 3-76
Minimal Surface Problem on page 3-81
Domain Decomposition Problem on page 3-83
Solve Poissons Equation on a Unit Disk
This example shows how to solve a simple elliptic PDE in the form of Poissons
equation on a unit disk.
The problem formulation is
U = 1 in , U = 0 on ,
where is the unit disk. In this case, the exact solution is
U x y
x y
, , ( ) =
1
4
2 2
so the error of the numeric solution can be evaluated for different meshes.
Using the PDE App
With the PDE app started, perform the following steps using the generic
scalar mode:
1 Using some of the Option menu features, add a grid and turn on the
snap-to-grid feature. Draw a circle by clicking the button with the ellipse
icon with the + sign, and then click-and-drag from the origin, using the
3-72
Elliptic PDEs
right mouse button, to a point at the circles perimeter. If the circle that you
create is not a perfect unit circle, double-click the circle. This opens a dialog
box where you can specify the exact center location and radius of the circle.
2 Enter the boundary mode by clicking the button with the icon. The
boundaries of the decomposed geometry are plotted, and the outer
boundaries are assigned a default boundary condition (Dirichlet boundary
condition, u = 0 on the boundary). In this case, this is what we want. If the
boundary condition is different, double-click the boundary to open a dialog
box through which you can enter and display the boundary condition.
3 To define the partial differential equation, click the PDE button. This
opens a dialog box, where you can define the PDE coefficients c, a, and f.
In this simple case, they are all constants: c = 1, f = 1, and a = 0.
4 Click the button or select Initialize Mesh from the Mesh menu.
This initializes and displays a triangular mesh.
5 Click the button or select Refine Mesh from the Mesh menu. This
causes a refinement of the initial mesh, and the new mesh is displayed.
6 To solve the system, just click the = button. The toolbox assembles the
PDE problem and solves the linear system. It also provides a plot of the
solution. Using the Plot Selection dialog box, you can select different types
of solution plots.
3-73
3 Solving PDEs
7 To compare the numerical solution to the exact solution, select the user
entry in the Property pop-up menu for Color in the Plot Selection dialog
box. Then input the MATLAB expression u-(1-x.^2-y.^2)/4 in the user
entry edit field. You obtain a plot of the absolute error in the solution.
You can also compare the numerical solution to the exact solution by entering
some simple command-line-oriented commands. It is easy to export the mesh
data and the solution to the MATLAB main workspace by using the Export
options from the Mesh and Solve menus. To refine the mesh and solve the
PDE successively, simply click the refine and = buttons until the desired
accuracy is achieved. (Another possibility is to use the adaptive solver.)
3-74
Elliptic PDEs
Using Command-Line Functions
First you must create a MATLAB function that parameterizes the 2-D
geometryin this case a unit circle.
The circleg.m file returns the coordinates of points on the unit circles
boundary. The file conforms to the file format described on the reference page
for pdegeom. You can display the file by typing type circleg.
Also, you need a function that describes the boundary condition. This is a
Dirichlet boundary condition where u = 0 on the boundary. The circleb1.m
file provides the boundary condition. The file conforms to the file format
described on the reference page for pdebound. You can display the file by
typing type circleb1.
Now you can start working from the MATLAB command line:
[p,e,t]=initmesh('circleg','Hmax',1);
error=[]; err=1;
while err > 0.001,
[p,e,t]=refinemesh('circleg',p,e,t);
u=assempde('circleb1',p,e,t,1,0,1);
exact=-(p(1,:).^2+p(2,:).^2-1)/4;
err=norm(u-exact',inf);
error=[error err];
end
pdemesh(p,e,t)
pdesurf(p,t,u)
pdesurf(p,t,u-exact')
The first MATLAB command creates the initial mesh using the
parameterizing function circleg.
Also, initialize a vector error for the maximum norm errors of the successive
solutions and set the initial error err to 1. The loop then runs until the error
of the solution is smaller than 10
-3
.
1 Refine the mesh. The current triangular mesh, defined by the geometry
circleg, the point matrix p, the edge matrix e, and the triangle matrix t,
is refined, and the mesh is returned using the same matrix variables.
3-75
3 Solving PDEs
2 Assemble and solve the linear system. The coefficients of the elliptic PDE
are constants (c = f = 1, a = 0) for this simple case. circleb1 contains a
description of the boundary conditions, and p, e, and t define the triangular
mesh.
3 Find the error of the numerical solution produced by Partial Differential
Equation Toolbox solver. The vector exact contains the exact solution at
the nodes, and what you actually find is the max-norm error of the solution
at the nodes.
4 Plot the mesh, the solution, and the error. The plot function pdesurf as
third argument can take any vector of values on the mesh given by p and t,
not just the solution. In this case you are also plotting the error function.
1
0.5
0
0.5
1
1
0.5
0
0.5
1
0
1
2
3
4
x 10
4
The Error Function
pdedemo1 performs all the previous steps.
Scattering Problem
This example shows how to solve a simple scattering problem, where you
compute the waves reflected from an object illuminated by incident waves.
For this problem, assume an infinite horizontal membrane subjected to small
vertical displacements U. The membrane is fixed at the object boundary.
3-76
Elliptic PDEs
r
V
We assume that the medium is homogeneous so that the wave speed is
constant, c.
Note Do not confuse this c with the parameter c appearing in Partial
Differential Equation Toolbox functions.
When the illumination is harmonic in time, we can compute the field by
solving a single steady problem. With
U(x,y,t) = u(x,y)e
it
,
the wave equation

=
2
2
2
0
U
t
c U
turns into

2
u c
2
u = 0
or the Helmholtzs equation
u k
2
u = 0,
3-77
3 Solving PDEs
where k, the wave number, is related to the angular frequency , the
frequency f, and the wavelength by
k
c
f
c
= = =
e t t

2 2
.
We have yet to specify the boundary conditions. Let the incident wave be a
plane wave traveling in the direction

a = (cos(a), sin(a)):
V x y t e v x y e
i ka x t i t
( , , ) ( , ) , = =
( )

e e
where
v x y e
ika x
( , ) . =


u is the sum of v and the reflected wave r,
u = v + r.
The boundary condition for the objects boundary is easy: u = 0, i.e.,
r = v(x,y)
For acoustic waves, where v is the pressure disturbance, the proper condition
would be

=
u
n
0.
The reflected wave r travels outward from the object. The condition at the
outer computational boundary should be chosen to allow waves to pass
without reflection. Such conditions are usually called nonreflecting, and we
use the classical Sommerfeld radiation condition. As

x approaches infinity, r
approximately satisfies the one-way wave equation
3-78
Elliptic PDEs
c
c
V = +
r
t
r c

, 0
which allows waves moving in the positive -direction only ( is the radial
distance from the object). With the time-harmonic solution, this turns into the
generalized Neumann boundary condition

. V = r ikr
For simplicity, let us make the outward normal of the computational domain
approximate the outward -direction.
Using the PDE App
You can now use the PDE app to solve this scattering problem. Using the
generic scalar mode, start by drawing the 2-D geometry of the problem. Let
the illuminated object be a square SQ1 with a side of 0.1 units and center
in [0.8 0.5] and rotated 45 degrees, and let the computational domain be
a circle C1 with a radius of 0.45 units and the same center location. The
Constructive Solid Geometry (CSG) model is then given by C1-SQ1.
For the outer boundary (the circle perimeter), the boundary condition is a
generalized Neumann condition with q = ik. The wave number k = 60, which
corresponds to a wavelength of about 0.1 units, so enter -60i as a constant q
and 0 as a constant g.
For the square objects boundary, you have a Dirichlet boundary condition:
r v x y e
ika x
= ( ) =

, .

In this problem, the incident wave is traveling in the x direction, so the
boundary condition is simply
r = e
ikx
.
Enter this boundary condition in the Boundary Condition dialog box as
a Dirichlet condition: h=1, r=-exp(-i*60*x). The real part of this is a
sinusoid.
3-79
3 Solving PDEs
For sufficient accuracy, about 10 finite elements per wavelength are needed.
The outer boundary should be located a few object diameters from the object
itself. An initial mesh generation and two successive mesh refinements give
approximately the desired resolution.
Although originally a wave equation, the transformation into a Helmholtzs
equation makes itin the Partial Differential Equation Toolbox context,
but not strictly mathematicallyan elliptic equation. The elliptic PDE
coefficients for this problem are c = 1, a = -k
2
= -3600, and f = 0. Open the
PDE Specification dialog box and enter these values.
The problem can now be solved, and the solution is complex. For a complex
solution, the real part is plotted and a warning message is issued.
The propagation of the reflected waves is computed as
Re(r(x,y)e
it
),
which is the reflex of
Re . e
i ka x t

( )
( )
e
To see the whole field, plot
Re ( , ) . r x y e e
ika x i t
+
( ) ( )


e
The reflected waves and the shadow behind the object are clearly visible
when you plot the reflected wave.
To make an animation of the reflected wave, the solution and the mesh data
must first be exported to the main workspace. Then make a script file or type
the following commands at the MATLAB prompt:
h=newplot; hf=get(h,'Parent'); set(hf,'Renderer','zbuffer')
axis tight, set(gca,'DataAspectRatio',[1 1 1]); axis off
M=moviein(10,hf);
maxu=max(abs(u));
colormap(cool)
3-80
Elliptic PDEs
for j=1:10,
ur=real(exp(-j*2*pi/10*sqrt(-1))*u));
pdeplot(p,e,t,'xydata',ur,'colorbar','off','mesh','off');
caxis([-maxu maxu]);
axis tight, set(gca,'DataAspectRatio',[1 1 1]); axis off
M(:,j)=getframe;
end
movie(hf,M,50);
pdedemo2 contains a full command-line implementation of the scattering
problem.
Minimal Surface Problem
This example shows how to solve a nonlinear problem for this equation:

=
1
1
0
2
u
u
where the coefficients c, a, and f do not depend only on x and y, but also on
the solution u.
The problem geometry is a unit disk, specified as = {(x, y) | x
2
+ y
2
1},
with u = x
2
on .
This nonlinear and cannot be solved with the regular elliptic solver. Instead,
the nonlinear solver pdenonlin is used.
3-81
3 Solving PDEs
This example show how to solve this minimal surface problem using both the
PDE app and command-line functions.
Using the PDE App
Make sure that the application mode in the PDE app is set to Generic
Scalar. The problem domain is simply a unit circle. Draw it and move to the
boundary mode to define the boundary conditions. Use Select All from the
Edit menu to select all boundaries. Then double-click a boundary to open the
Boundary Condition dialog box. The Dirichlet condition u = x
2
is entered by
typing x.^2 into the r edit box. Next, open the PDE Specification dialog box
to define the PDE. This is an elliptic equation with
c
u
a f =
+
= =
1
1
0 0
2
, , . and
The nonlinear c is entered into the c edit box as
1./sqrt(1+ux.^2+uy.^2)
Initialize a mesh and refine it once.
Before solving the PDE, select Parameters from the Solve menu and check
the Use nonlinear solver option. Also, set the tolerance parameter to 0.001.
Click the = button to solve the PDE. Use the Plot Selection dialog box to plot
the solution in 3-D (check u and continuous selections in the Height column)
to visualize the saddle shape of the solution.
Using Command-Line Functions
Working from the command line, the following sequence of commands solves
the minimal surface problem and plots the solution. The files circleg
and circleb2 contain the geometry specification and boundary condition
functions, respectively.
g='circleg';
b='circleb2';
c='1./sqrt(1+ux.^2+uy.^2)';
rtol=1e-3;
3-82
Elliptic PDEs
[p,e,t]=initmesh(g);
[p,e,t]=refinemesh(g,p,e,t);
u=pdenonlin(b,p,e,t,c,0,0,'Tol',rtol);
pdesurf(p,t,u)
You can run this example by typing pdedemo3.
1
0.5
0
0.5
1
1
0.5
0
0.5
1
0
0.2
0.4
0.6
0.8
1
Domain Decomposition Problem
This example shows how to perform one-level domain decomposition for
complicated geometries, where you can decompose this geometry into the
union of more subdomains of simpler structure. Such structures are often
introduced by the PDE app.
Assume now that is the disjoint union of some subdomains
1
,
2
, . . . ,
n. Then you could renumber the nodes of a mesh on such that the indices
of the nodes of each subdomain are grouped together, while all the indices of
nodes common to two or more subdomains come last. Since K has nonzero
3-83
3 Solving PDEs
entries only at the lines and columns that are indices of neighboring nodes,
the stiffness matrix is partitioned as follows:
K
K B
K B
K B
B B B C
T
T
n n
T
n
=
|
\

|
.
|
|
|
|
|
|
|
1 1
2 2
1 2
0 0
0 0
0 0

while the right side is

F
f
f
f
f
n
c
=
|
\

|
.
|
|
|
|
|
|
1
2

The Partial Differential Equation Toolbox function assempde can assemble

the matrices K
j
, B
j
, f
j
, and C separately. You have full control over the storage
and further processing of these matrices.
Furthermore, the structure of the linear system
Ku = F
is simplified by decomposing K into the partitioned matrix.
Now consider the geometry of the L-shaped membrane. You can plot the
geometry of the membrane by typing
pdegplot('lshapeg')
Notice the borders between the subdomains. There are three subdomains.
Thus the matrix formulas with n = 3 can be used. Now generate a mesh for
the geometry:
[p,e,t]=initmesh('lshapeg');
3-84
Elliptic PDEs
[p,e,t]=refinemesh('lshapeg',p,e,t);
[p,e,t]=refinemesh('lshapeg',p,e,t);
So for this case, with n = 3, you have
K B
K B
K B
B B B C
u
u
u
u
T
T
T
c
1 1
2 2
3 3
1 2 3
1
2
3
0 0
0 0
0 0
|
\

|
.
|
|
|
|
|
|
\

|
..
|
|
|
|
|
=
|
\

|
.
|
|
|
|
|
f
f
f
f
c
1
2
3
,
and the solution is given by block elimination:
( ) C B K B B K B B K B u f B K f B K f B
T T T
c c
=

1 1
1
1 2 2
1
2 3 3
1
3 1 1
1
1 2 2
1
2 33 3
1
3
1 1
1
1 1
K f
u K f B u
T
c

= ( )

In the following MATLAB solution, a more efficient algorithm using Cholesky

factorization is used:
time=[];
np=size(p,2);
% Find common points
c=pdesdp(p,e,t);
nc=length(c);
C=zeros(nc,nc);
FC=zeros(nc,1);
[i1,c1]=pdesdp(p,e,t,1);ic1=pdesubix(c,c1);
[K,F]=assempde('lshapeb',p,e,t,1,0,1,time,1);
K1=K(i1,i1);d=symamd(K1);i1=i1(d);
K1=chol(K1(d,d));B1=K(c1,i1);a1=B1/K1;
C(ic1,ic1)=C(ic1,ic1)+K(c1,c1)-a1*a1';
f1=F(i1);e1=K1'\f1;FC(ic1)=FC(ic1)+F(c1)-a1*e1;
[i2,c2]=pdesdp(p,e,t,2);ic2=pdesubix(c,c2);
[K,F]=assempde('lshapeb',p,e,t,1,0,1,time,2);
3-85
3 Solving PDEs
K2=K(i2,i2);d=symamd(K2);i2=i2(d);
K2=chol(K2(d,d));B2=K(c2,i2);a2=B2/K2;
C(ic2,ic2)=C(ic2,ic2)+K(c2,c2)-a2*a2';
f2=F(i2);e2=K2'\f2;FC(ic2)=FC(ic2)+F(c2)-a2*e2;
[i3,c3]=pdesdp(p,e,t,3);ic3=pdesubix(c,c3);
[K,F]=assempde('lshapeb',p,e,t,1,0,1,time,3);
K3=K(i3,i3);d=symamd(K3);i3=i3(d);
K3=chol(K3(d,d));B3=K(c3,i3);a3=B3/K3;
C(ic3,ic3)=C(ic3,ic3)+K(c3,c3)-a3*a3';
f3=F(i3);e3=K3'\f3;FC(ic3)=FC(ic3)+F(c3)-a3*e3;
% Solve
u=zeros(np,1);
u(c)=C\ FC;
u(i1)=K1\(e1-a1'*u(c1));
u(i2)=K2\(e2-a2'*u(c2));
u(i3)=K3\(e3-a3'*u(c3));
The problem can also be solved by typing
% Compare with solution not using subdomains
[K,F]=assempde('lshapeb',p,e,t,1,0,1);u1=K\F;
norm(u-u1,'inf')
pdesurf(p,t,u)
You can run this entire example by typing pdedemo4.
3-86
Elliptic PDEs
1
0.5
0
0.5
1
1
0.5
0
0.5
1
0
0.05
0.1
0.15
3-87
3 Solving PDEs
Parabolic PDEs
This section describes the solution of some parabolic PDE problems. The
problems are solved using both the PDE app and the command-line functions.
The topics include:
In this section...
Heat Equation for Metal Block with Cavity on page 3-88
Heat Distribution in a Radioactive Rod on page 3-92
Heat Equation for Metal Block with Cavity
This example shows how to solve a heat equation that describes the diffusion
of heat in a body. The heat equation has the form:
d
u
t
u
c
c
= A 0.
Consider a metal block containing rectangular crack or cavity. The left side of
the block is heated to 100 degrees centigrade. At the right side of the metal
block, heat is flowing from the block to the surrounding air at a constant rate.
All the other block boundaries are isolated. This leads to the following set of
boundary conditions (when proper scaling of t is chosen):
u = 100 on the left side (Dirichlet condition)
u/n = 10 on the right side (Neumann condition)
u/n = 0 on all other boundaries (Neumann condition)
Also, for the heat equation we need an initial value: the temperature in the
metal block at the starting time t
0
. In this case, the temperature of the block
is 0 degrees at the time we start applying heat.
Finally, to complete the problem formulation, we specify that the starting
time is 0 and that we want to study the heat distribution during the first
five seconds.
3-88
Parabolic PDEs
Using the PDE App
Once you have started the PDE app and selected the Generic Scalar mode,
drawing the CSG model can be done very quickly: Draw a rectangle (R1) with
the corners in x = [-0.5 0.5 0.5 -0.5] and y = [-0.8 -0.8 0.8 0.8].
Draw another rectangle (R2) to represent the rectangular cavity. Its corners
should have the coordinates x = [-0.05 0.05 0.05 -0.05] and y = [-0.4
-0.4 0.4 0.4]. To assist in drawing the narrow rectangle representing the
cavity, open the Grid Spacing dialog box from the Options and enter x-axis
extra ticks at -0.05 and 0.05. Then turn on the grid and the snap-to-grid
feature. A rectangular cavity with the correct dimensions is then easy to draw.
The CSG model of the metal block is now simply expressed as the set formula
R1-R2.
Leave the draw mode and enter the boundary mode by clicking the button,
and continue by selecting boundaries and specifying the boundary conditions.
Using the Select All option from the Edit menu and then defining the
Neumann condition

=
u
n
0
for all boundaries first is a good idea since that leaves only the leftmost and
rightmost boundaries to define individually.
3-89
3 Solving PDEs
The next step is to open the PDE Specification dialog box and enter the PDE
coefficients.
The generic parabolic PDE that Partial Differential Equation Toolbox
functions solve is
d c au f
u
t
u
c
c
V V
( )
+ = ,
with initial values u
0
= u(t
0
) and the times at which to compute a solution
specified in the array tlist.
For this case, you have d = 1, c = 1, a = 0, and f = 0.
Initialize the mesh by clicking the button. If you want, you can refine the
mesh by clicking the Refine button.
The initial values u0 = 0, and the list of times is entered as the MATLAB
array [0:0.5:5]. They are entered into the Solve Parameters dialog box,
which is accessed by selecting Parameters from the Solve menu.
The problem can now be solved. Pressing the = button solves the heat
equation at 11 different times from 0 to 5 seconds. By default, an interpolated
plot of the solution, i.e., the heat distribution, at the end of the time span
is displayed.
A more interesting way to visualize the dynamics of the heat distribution
process is to animate the solution. To start an animation, check the
Animation check box in the Plot selection dialog box. Also, select the
colormap hot. Click the Plot button to start a recording of the solution
plots in a separate figure window. The recorded animation is then played
five times.
The temperature in the block rises very quickly. To improve the animation
and focus on the first second, try to change the list of times to the MATLAB
expression logspace(-2,0.5,20).
Also, try to change the heat capacity coefficient d and the heat flow at the
rightmost boundary to see how they affect the heat distribution.
3-90
Parabolic PDEs
Using Command-Line Functions
First, you must create geometry and boundary condition files. The files used
here were created using the PDE app. The geometry of the metal block is
described in crackg.m, and the boundary conditions can be found in crackb.m.
To create an initial mesh, call initmesh:
[p,e,t]=initmesh('crackg');
The heat equation can now be solved using the Partial Differential Equation
Toolbox function parabolic. The generic parabolic PDE that parabolic
solves is
d c au f
u
t
u
c
c
V V
( )
+ = ,
with initial value u0 = u(t
0
) and the times at which to compute a solution
specified in the array tlist. For this case, you have d = 1, c = 1, a = 0, and
f = 0. The initial value u0 = 0, and the list of times, tlist, is set to the
MATLAB array 0:0.5:5.
To compute the solution, call parabolic:
3-91
3 Solving PDEs
u=parabolic(0,0:0.5:5,'crackb',p,e,t,1,0,0,1);
The solution u created this way a matrix with 11 columns, where each column
corresponds to the solution at the 11 points in time 0, 0.5, . . . , 4.5, 5.0.
Let us plot the solution at t = 5.0 seconds using interpolated shading and a
hidden mesh. Use the hot colormap:
pdeplot(p,e,t,'xydata',u(:,11),'mesh','off',...
'colormap','hot')
Heat Distribution in a Radioactive Rod
This example shows how to solve a 3-D parabolic PDE problem by reducing
the problem to 2-D using coordinate transformation. For a step-by-step
command-line solution, see Heat Distribution in Radioactive Rod.
Consider a cylindrical radioactive rod. At the left end, heat is continuously
added. The right end is kept at a constant temperature. At the outer
boundary, heat is exchanged with the surroundings by transfer. At the
same time, heat is uniformly produced in the whole rod due to radioactive
processes. Assume that the initial temperature is zero. This leads to the
following problem:
C k f
u
t
u
c
c
V V
( )
= ,
where is the density, C is the rods thermal capacity, k is the thermal
conductivity, and f is the radioactive heat source.
The density for this metal rod is 7800 kg/m
3
, the thermal capacity is 500
Ws/kgC, and the thermal conductivity is 40 W/mC. The heat source is
20000 W/m
3
. The temperature at the right end is 100 C. The surrounding
temperature at the outer boundary is 100 C, and the heat transfer coefficient
is 50 W/m
2
C. The heat flux at the left end is 5000 W/m
2
.
But this is a cylindrical problem, so you need to transform the equation, using
the cylindrical coordinates r, z, and . Due to symmetry, the solution is
independent of , so the transformed equation is
3-92
Parabolic PDEs
r C kr kr fr
u
t r
u
r z
u
z

c
c
c
c
c
c
c
c
c
c

|
\

|
.
|

|
\

|
.
|
= .
The boundary conditions are:

n k u V
( )
= 5000 at the left end of the rod (Neumann condition). Since the
generalized Neumann condition in Partial Differential Equation Toolbox
software is

n k u V
( )
+ qu = g, and c depends on r in this problem (c = kr),
this boundary condition is expressed as

n c u V
( )
= 5000r.
u = 100 at the right end of the rod (Dirichlet condition).

n k u V
( )
= 50(100-u) at the outer boundary (generalized Neumann
condition). In Partial Differential Equation Toolbox software, this must be
expressed as

n c u V
( )
+ 50r u = 50r 100.
The cylinder axis r = 0 is not a boundary in the original problem, but in
our 2-D treatment it has become one. We must give the artificial boundary
condition

n c u V
( )
here.
The initial value is u(t
0
) = 0.
Using the PDE App
Solve this problem using the PDE app. Model the rod as a rectangle with its
base along the x-axis, and let the x-axis be the z direction and the y-axis be the
r direction. A rectangle with corners in (-1.5,0), (1.5,0), (1.5,0.2), and (-1.5,0.2)
would then model a rod with length 3 and radius 0.2.
Enter the boundary conditions by double-clicking the boundaries to open the
Boundary Condition dialog box. For the left end, use Neumann conditions
with 0 for q and 5000*y for g. For the right end, use Dirichlet conditions
with 1 for h and 100 for r. For the outer boundary, use Neumann conditions
with 50*y for q and 50*y*100 for g. For the axis, use Neumann conditions
with 0 for q and g.
Enter the coefficients into the PDE Specification dialog box: c is 40*y, a is
zero, d is 7800*500*y, and f is 20000*y.
3-93
3 Solving PDEs
Animate the solution over a span of 20000 seconds (computing the solution
every 1000 seconds). We can see how heat flows in over the right and outer
boundaries as long as u < 100, and out when u > 100. You can also open
the PDE Specification dialog box, and change the PDE type to Elliptic.
This shows the solution when u does not depend on time, i.e., the steady
state solution. The profound effect of cooling on the outer boundary can be
demonstrated by setting the heat transfer coefficient to zero.
3-94
Hyperbolic PDEs
Hyperbolic PDEs
This section describes the solution of a hyperbolic PDE problem. The problem
is solved using the PDE app and command-line functions.
Wave Equation
As an example of a hyperbolic PDE, let us solve the wave equation
c
c
=
2
2
0
u
t
u A
for transverse vibrations of a membrane on a square with corners in (1,1),
(1,1), (1,1), and (1,1). The membrane is fixed (u = 0) at the left and right
sides, and is free (u/n = 0) at the upper and lower sides. Additionally, we
need initial values for u(t
0
) and u(t
0
)/t
The initial values need to match the boundary conditions for the solution to
be well-behaved. If we start at t = 0,
u x y x , , arctan cos 0
2
( ) =
|
\

|
.
|
|
\

|
.
|
t
and
c
c
=
( )
|
\

|
.
|
|
\

|
.
|
( )
=
u
t
x y
x y t
t
, ,
sin exp sin
0
3
2
t
t
are initial values that satisfy the boundary conditions. The reason for the
arctan and exponential functions is to introduce more modes into the solution.
Using the PDE App
Use the PDE app in the generic scalar mode. Draw the square using the
Rectangle/square option from the Draw menu or the button with the
rectangle icon. Proceed to define the boundary conditions by clicking the
button and then double-click the boundaries to define the boundary conditions.
3-95
3 Solving PDEs
Initialize the mesh by clicking the button or by selecting Initialize mesh
from the Mesh menu.
Also, define the hyperbolic PDE by opening the PDE Specification dialog box,
selecting the hyperbolic PDE, and entering the appropriate coefficient values.
The general hyperbolic PDE is described by
d
u
t
c u au f
c
c
V V
( )
+ =
2
2
,
so for the wave equation you get c = 1, a = 0, f = 0, and d = 1.
Before solving the PDE, select Parameters from the Solve menu to open the
Solve Parameters dialog box. As a list of times, enter linspace(0,5,31)
and as initial values for u:
atan(cos(pi/2*x))
and for u/t , enter
3*sin(pi*x).*exp(sin(pi/2*y))
Finally, click the = button to compute the solution. The best plot for viewing
the waves moving in the x and y directions is an animation of the whole
3-96
Hyperbolic PDEs
sequence of solutions. Animation is a very real time and memory consuming
feature, so you may have to cut down on the number of times at which to
compute a solution. A good suggestion is to check the Plot in x-y grid option.
Using an x-y grid can speed up the animation process significantly.
Using Command-Line Functions
From the command line, solve the equation with the preceding boundary
conditions and the initial values, starting at time 0 and then every 0.05
seconds for five seconds.
The geometry is described in the file squareg.m and the boundary conditions
in the file squareb3.m. The following sequence of commands then generates a
solution and animates it. First, create a mesh and define the initial values
and the times for which you want to solve the equation:
[p,e,t]=initmesh('squareg');
x=p(1,:)'; y=p(2,:)';
u0=atan(cos(pi/2*x));
ut0=3*sin(pi*x).*exp(sin(pi/2*y));
n=31;
tlist=linspace(0,5,n); % list of times
You are now ready to solve the wave equation. The general form for the
hyperbolic PDE in Partial Differential Equation Toolbox software is
d
u
t
c u au f
c
c
V V
( )
+ =
2
2
,
so here you have d = 1, c = 1, a = 0, and f = 0:
uu=hyperbolic(u0,ut0,tlist,'squareb3',p,e,t,1,0,0,1);
To visualize the solution, you can animate it. Interpolate to a rectangular
grid to speed up the plotting:
delta=-1:0.1:1;
3-97
3 Solving PDEs
[uxy,tn,a2,a3]=tri2grid(p,t,uu(:,1),delta,delta);
gp=[tn;a2;a3];
umax=max(max(uu));
umin=min(min(uu));
newplot
M=moviein(n);
for i=1:n,
pdeplot(p,e,t,'xydata',uu(:,i),'zdata',uu(:,i),...
'mesh','off','xygrid','on','gridparam',gp,...
'colorbar','off','zstyle','continuous');
axis([-1 1 -1 1 umin umax]); caxis([umin umax]);
M(:,i)=getframe;
end
movie(M,10);
You can find a complete example of this problem, including animation, in
pdedemo6. If you have lots of memory, you can try increasing n, the number of
frames in the movie.
3-98
Hyperbolic PDEs
Animation of the Solution to the Wave Equation
3-99
3 Solving PDEs
Eigenvalue Problems
This section describes the solution of some eigenvalue PDE problems. The
problems are solved using the PDE app and command-line functions. The
problems include:
In this section...
Eigenvalues and Eigenfunctions for the L-Shaped Membrane on page
3-100
L-Shaped Membrane with a Rounded Corner on page 3-104
Eigenvalues and Eigenmodes of a Square on page 3-106
Eigenvalues and Eigenfunctions for the L-Shaped
Membrane
The problem of finding the eigenvalues and the corresponding eigenfunctions
of an L-shaped membrane is of interest to all MATLAB users, since the plot of
the first eigenfunction is the MathWorks

logo. In fact, you can compare the

eigenvalues and eigenfunctions computed by Partial Differential Equation
Toolbox software to the ones produced by the MATLAB function membrane.
The problem is to compute all eigenmodes with eigenvalues < 100 for the
eigenmode PDE problem
u = u
on the geometry of the L-shaped membrane. u = 0 on the boundary (Dirichlet
condition).
Using the PDE App
With the PDE app active, check that the current mode is set to Generic
Scalar. Then draw the L-shape as a polygon with corners in (0,0), (1,0),
(1,1), (1,1), (1,1), and (0,1).
There is no need to define any boundary conditions for this problem since the
default conditionu = 0 on the boundaryis the correct one. Therefore, you
can continue to the next step: to initialize the mesh. Refine the initial mesh
3-100
Eigenvalue Problems
twice. Defining the eigenvalue PDE problem is also easy. Open the PDE
Specification dialog box and select Eigenmodes. The default values for the
PDE coefficients, c = 1, a = 0, d = 1, all match the problem description, so you
can exit the PDE Specification dialog box by clicking the OK button.
Open the Solve Parameters dialog box by selecting Parameters from the
Solve menu. The dialog box contains an edit box for entering the eigenvalue
search range. The default entry is [0 100], which is just what you want.
Finally, solve the L-shaped membrane problem by clicking the = button.
The solution displayed is the first eigenfunction. The value of the first
(smallest) eigenvalue is also displayed. You find the number of eigenvalues
on the information line at the bottom of the PDE app. You can open the Plot
Selection dialog box and choose which eigenfunction to plot by selecting from
a pop-up menu of the corresponding eigenvalues.
Using Command-Line Functions
The geometry of the L-shaped membrane is described in the file lshapeg.m
and the boundary conditions in the file lshapeb.m.
First, initialize the mesh and refine it twice using the command line functions
at the MATLAB prompt:
[p,e,t]=initmesh('lshapeg');
[p,e,t]=refinemesh('lshapeg',p,e,t);
[p,e,t]=refinemesh('lshapeg',p,e,t);
Recall the general eigenvalue PDE problem description:
V V
( )
+ = c u au du ,
This means that in this case you have c = 1, a = 0, and d = 1. The syntax of
pdeeig, the Partial Differential Equation Toolbox eigenvalue solver, is
[v,l]=pdeeig(b,p,e,t,c,a,d,r)
The input argument r is a two-element vector indicating the interval on the
real axis where pdeeig searches for eigenvalues. Here you are looking for
eigenvalues < 100, so the interval you use is [0 100].
3-101
3 Solving PDEs
Now you can call pdeeig and see how many eigenvalues you find:
[v,l]=pdeeig('lshapeb',p,e,t,1,0,1,[0 100]);
There are 19 eigenvalues smaller than 100. Plot the first eigenmode and
compare it to the MATLAB membrane function:
pdesurf(p,t,v(:,1))
figure
membrane(1,20,9,9)
membrane can produce the first 12 eigenfunctions for the L-shaped membrane.
Compare also the 12th eigenmodes:
figure
pdesurf(p,t,v(:,12))
figure
membrane(12,20,9,9)
Looking at the following eigenmodes, you can see how the number of
oscillations increases. The eigenfunctions are symmetric or antisymmetric
around the diagonal from (0,0) to (1,-1), which divides the L-shaped membrane
into two mirror images. In a practical computation, you could take advantage
of such symmetries in the PDE problem, and solve over a region half the size.
The eigenvalues of the full L-shaped membrane are the union of those of the
half with Dirichlet boundary condition along the diagonal (eigenvalues 2, 4, 7,
11, 13, 16, and 17) and those with Neumann boundary condition (eigenvalues
1, 3, 5, 6, 10, 12, 14, and 15).
The eigenvalues
8
and
9
make up a double eigenvalue for the PDE at around
49.64. Also, the eigenvalues
18
and
19
make up another double eigenvalue at
around 99.87. You may have gotten two different but close values. The default
triangulation made by initmesh is not symmetric around the diagonal, but
a symmetric grid gives a matrix with a true double eigenvalue. Each of the
eigenfunctions u
8
and u
9
consists of three copies of eigenfunctions over the
unit square, corresponding to its double second eigenvalue. You may not have
obtained the zero values along a diagonal of the squareany line through
the center of the square may have been computed. This shows a general
fact about multiple eigenvalues for symmetric matrices; namely that any
vector in the invariant subspace is equally valid as an eigenvector. The two
3-102
Eigenvalue Problems
eigenfunctions u
8
and u
9
are orthogonal to each other if the dividing lines
make right angles. Check your solutions for that.
Actually, the eigenvalues of the square can be computed exactly. They are
(m
2
+ n
2
)
2
e.g., the double eigenvalue
18
and
19
is 10
2
, which is pretty close to 100.
If you compute the FEM approximation with only one refinement, you
would only find 16 eigenvalues, and you obtain the wrong solution to the
original problem. You can of course check for this situation by computing the
eigenvalues over a slightly larger range than the original problem.
You get some information from the printout in the MATLAB command window
that is printed during the computation. For this problem, the algorithm
computed a new set of eigenvalue approximations and tested for convergence
every third step. In the output, you get the step number, the time in seconds
since the start of the eigenvalue computation, and the number of converged
eigenvalues with eigenvalues both inside and outside the interval counted.
Here is what MATLAB wrote:
Basis= 10, Time= 2.70, New conv eig= 0
Basis= 13, Time= 3.50, New conv eig= 0
Basis= 16, Time= 4.36, New conv eig= 0
Basis= 19, Time= 5.34, New conv eig= 1
Basis= 22, Time= 6.46, New conv eig= 2
Basis= 25, Time= 7.61, New conv eig= 3
Basis= 28, Time= 8.86, New conv eig= 3
Basis= 31, Time= 10.23, New conv eig= 5
Basis= 34, Time= 11.69, New conv eig= 5
Basis= 37, Time= 13.28, New conv eig= 7
Basis= 40, Time= 14.97, New conv eig= 8
Basis= 43, Time= 16.77, New conv eig= 9
Basis= 46, Time= 18.70, New conv eig= 11
Basis= 49, Time= 20.73, New conv eig= 11
Basis= 52, Time= 22.90, New conv eig= 13
Basis= 55, Time= 25.13, New conv eig= 14
Basis= 58, Time= 27.58, New conv eig= 14
3-103
3 Solving PDEs
Basis= 61, Time= 30.13, New conv eig= 15
Basis= 64, Time= 32.83, New conv eig= 16
Basis= 67, Time= 35.64, New conv eig= 18
Basis= 70, Time= 38.62, New conv eig= 22
End of sweep: Basis= 70, Time= 38.62, New conv eig= 22
Basis= 32, Time= 43.29, New conv eig= 0
Basis= 35, Time= 44.70, New conv eig= 0
Basis= 38, Time= 46.22, New conv eig= 0
Basis= 41, Time= 47.81, New conv eig= 0
Basis= 44, Time= 49.52, New conv eig= 0
Basis= 47, Time= 51.35, New conv eig= 0
Basis= 50, Time= 53.27, New conv eig= 0
Basis= 53, Time= 55.30, New conv eig= 0
End of sweep: Basis= 53, Time= 55.30, New conv eig= 0
You can see that two Arnoldi runs were made. In the first, 22 eigenvalues
converged after a basis of size 70 was computed; in the second, where
the vectors were orthogonalized against all the 22 converged vectors, the
smallest eigenvalue stabilized at a value outside of the interval [0, 100], so
the algorithm signaled convergence. Of the 22 converged eigenvalues, 19
were inside the search interval.
L-Shaped Membrane with a Rounded Corner
An extension of this problem is to compute the eigenvalues for an L-shaped
membrane where the inner corner at the knee is rounded. The roundness
is created by adding a circle so that the circles arc is a part of the L-shaped
membranes boundary. By varying the circles radius, the degree of roundness
can be controlled. The lshapec file is an extension of an ordinary model file
created using the PDE app. It contains the lines
pdepoly([-1, 1, 1, 0, 0, -1],...
[-1, -1, 1, 1, 0, 0],'P1');
pdecirc(-a,a,a,'C1');
pderect([-a 0 a 0],'SQ1');
The extra circle and rectangle that are added using pdecirc and pderect
to create the rounded corner are affected by the added input argument a
through a couple of extra lines of MATLAB code. This is possible since
3-104
Eigenvalue Problems
Partial Differential Equation Toolbox software is a part of the open MATLAB
environment.
With lshapec you can create L-shaped rounded geometries with different
degrees of roundness. If you use lshapec without an input argument, a
default radius of 0.5 is used. Otherwise, use lshapec(a), where a is the
radius of the circle.
Experimenting using different values for the radius a shows you that the
eigenvalues and the frequencies of the corresponding eigenmodes decrease
as the radius increases, and the shape of the L-shaped membrane becomes
more rounded. In the following figure, the first eigenmode of an L-shaped
membrane with a rounded corner is plotted.
First Eigenmode for an L-Shaped Membrane with a Rounded Corner
3-105
3 Solving PDEs
Eigenvalues and Eigenmodes of a Square
Let us study the eigenvalues and eigenmodes of a square with an interesting
set of boundary conditions. The square has corners in (-1,-1), (-1,1), (1,1), and
(1,-1). The boundary conditions are as follows:
On the left boundary, the Dirichlet condition u = 0.
On the upper and lower boundary, the Neumann condition
c
c
=
u
n
0.
On the right boundary, the generalized Neumann condition
c
c
=
u
n
u
3
4
0.
The eigenvalue PDE problem is
u = u .
We are interested in the eigenvalues smaller than 10 and the corresponding
eigenmodes, so the search range is [-Inf 10]. The sign in the generalized
Neumann condition is such that there are negative eigenvalues.
Using the PDE App
Using the PDE app in the generic scalar mode, draw the square using the
Rectangle/square option from the Draw menu or the button with the
rectangle icon. Then define the boundary conditions by clicking the button
and then double-click the boundaries to define the boundary conditions. On
the right side boundary, you have the generalized Neumann conditions, and
you enter them as constants: g = 0 and g = 3/4.
Initialize the mesh and refine it once by clicking the and refine buttons or
by selecting the corresponding options from the Mesh menu.
Also, define the eigenvalue PDE problem by opening the PDE Specification
dialog box and selecting the Eigenmodes option. The general eigenvalue
PDE is described by
3-106
Eigenvalue Problems
V V
( )
+ = c u au du ,
so for this problem you use the default values c = 1, a = 0, and d = 1. Also, in
the Solve Parameters dialog box, enter the eigenvalue range as the MATLAB
vector [-Inf 10].
Finally, click the = button to compute the solution. By default, the first
eigenfunction is plotted. You can plot the other eigenfunctions by selecting the
corresponding eigenvalue from a pop-up menu in the Plot Selection dialog box.
The pop-up menu contains all the eigenvalues found in the specified range.
You can also export the eigenfunctions and eigenvalues to the MATLAB main
workspace by using the Export Solution option from the Solve menu.
Using Command-Line Functions
The geometry description file and boundary condition file for this problem
are called squareg.m and squareb2.m, respectively. Use the following
sequence of commands to find the eigenvalues in the specified range and the
corresponding eigenfunctions:
[p,e,t]=initmesh('squareg');
[p,e,t]=refinemesh('squareg',p,e,t);
The eigenvalue PDE coefficients c, a, and d for this problem are c = 1, a = 0,
and d = 1. You can enter the eigenvalue range r as the MATLAB vector [-Inf
10]. pdeeig returns two output arguments, the eigenvalues as an array l and
a matrix v of corresponding eigenfunctions:
[v,l]=pdeeig('squareb2',p,e,t,1,0,1,[-Inf 10]);
To plot the fourth eigenfunction as a surface plot, type
pdesurf(p,t,v(:,4))
This problem is separable, i.e.,
u(x,y) = f(x)g(y).
The functions f and g are eigenfunctions in the x and y directions, respectively.
In the x direction, the first eigenmode is a slowly increasing exponential
function. The higher modes include sinusoids. In the y direction, the first
3-107
3 Solving PDEs
eigenmode is a straight line (constant), the second is half a cosine, the third is
a full cosine, the fourth is one and a half full cosines, etc. These eigenmodes in
the y direction are associated with the eigenvalues:
0
4
4
4
9
4
2 2 2
, , , ,...
t t t
There are five eigenvalues smaller than 10 for this problem, and the first one
is even negative (0.4145). It is possible to trace the preceding eigenvalues in
the eigenvalues of the solution. Looking at a plot of the first eigenmode, you
can see that it is made up of the first eigenmodes in the x and y directions.
The second eigenmode is made up of the first eigenmode in the x direction and
the second eigenmode in the y direction.
Look at the difference between the first and the second eigenvalue:
l(2)-l(1)
ans =
2.4740
pi*pi/4
ans =
2.4674
Likewise, the fifth eigenmode is made up of the first eigenmode in the x
direction and the third eigenmode in the y direction. As expected, l(5)-l(1)
is approximately equal to
2
. You can explore higher modes by increasing the
search range to include eigenvalues greater than 10.
3-108
Vibration Of a Circular Membrane Using The MATLAB eigs Function
Vibration Of a Circular Membrane Using The MATLAB
eigs Function
This example shows the calculation of the vibration modes of a circular
membrane. The calculation of vibration modes requires the solution of the
eigenvalue partial differential equation (PDE). In this example the solution of
the eigenvalue problem is performed using both the PDE Toolbox pdeeig
solver and the core MATLAB eigs eigensolver.
The main objective of this example is to show how eigs can be used with
PDE Toolbox. Generally, the eigenvalues calculated by pdeeig and eigs
are practically identical. However, sometimes, it is simply more convenient to
use eigs than pdeeig. One example of this is when it is desired to calculate a
specified number of eigenvalues in the vicinity of a user-specified target value.
pdeeig requires that a lower and upper bound surrounding this target value
be specified. eigs requires only that the target eigenvalue and the desired
number of eigenvalues be specified.
Geometry And Mesh
The geometry for a circle can easily be defined as shown below.
radius = 2;
g = decsg([1 0 0 radius]', 'C1', ('C1')');
[p,e,t] = initmesh(g, 'hmax', .2);
Define the PDE Coefficients and Boundary Conditions
c = 1e2;
a = 0;
f = 0;
d = 10;
% Define boundary conditions using a boundary file function.
b = @boundaryFileZeroDirichlet;
% This boundary file function sets the solution to zero at r=radius.
type boundaryFileZeroDirichlet.m
function [ q, g, h, r ] = boundaryFileZeroDirichlet( p, e, u, time )
%BOUNDARYFILEZERODIRICHLET Solution is zero on all edges
3-109
3 Solving PDEs
% Define a Dirichlet boundary condition making the solution equal zero
% on all boundary edges.
N = 1; % Only a single scalar PDE
ne = size(e,2); % number of boundary edges
q = zeros(N^2, ne); % Neumann coefficient q is zero on all edges
g = zeros(N, ne); % Neumann coefficient g is zero on all edges
h = ones(N^2, 2*ne); % Dirichlet h coefficient is one at both ends of all e
r = zeros(N,2*ne); % Dirichlet r coefficient is zero at both ends of all ed
end
Solve the eigenvalue problem using eigs
Use assempde and assema to calculate the global finite element mass and
stiffness matrices.
[K,~,B] = assempde(b,p,e,t,c,a,f);
[~,M] = assema(p,t,c,d,f);
M = B'*M*B; % apply the constraints to the mass matrix from |assema|
sigma = 1e2; numberEigenvalues = 5;
[eigenvectorsEigs,eigenvaluesEigs] = eigs(K,M,numberEigenvalues,sigma);
% eigs orders the eigenvalues (and their eigenvectors) from highest to
% lowest. Reorder these from lowest to highest to be consistent with |pdeei
eigenvaluesEigs = flipud(diag(eigenvaluesEigs));
% Reorder the eigenvectors. Also transform the eigenvectors with constraine
% equations removed to the full eigenvector including constrained equations
eigenvectorsEigs = B*fliplr(eigenvectorsEigs);
Solve the eigenvalue problem using pdeeig
Define the eigenvalue range for pdeeig from the output eigenvalues from eigs
so that it computes the same ones.
r = [eigenvaluesEigs(1)*.99 eigenvaluesEigs(end)*1.01];
[eigenvectorsPde,eigenvaluesPde] = pdeeig(b,p,e,t,c,a,d,r);
Basis= 10, Time= 0.23, New conv eig= 1
Basis= 19, Time= 0.25, New conv eig= 3
Basis= 28, Time= 0.25, New conv eig= 8
Basis= 37, Time= 0.33, New conv eig= 12
3-110
Vibration Of a Circular Membrane Using The MATLAB eigs Function
End of sweep: Basis= 37, Time= 0.33, New conv eig= 12
Basis= 22, Time= 0.58, New conv eig= 0
Basis= 31, Time= 0.78, New conv eig= 0
End of sweep: Basis= 31, Time= 0.78, New conv eig= 0
Compare the solutions computed by eigs and pdeeig
eigenValueDiff = eigenvaluesPde - eigenvaluesEigs;
fprintf('Maximum difference in eigenvalues from pdeeig and eigs: %e\n', ...
norm(eigenValueDiff,inf));
%
% As can be seen, both functions calculate the same eigenvalues. For any
% eigenvalue, the eigenvector can be multiplied by an arbitrary scalar.
% eigs and pdeeigs choose a different arbitrary scalar for normalizing
% their eigenvectors as shown in the figure below.
%
h = figure; pos = get(h,'position'); set(h,'position',[1 1 2 1].*pos);
subplot(1,2,1); axis equal;
pdeplot(p,e,t,'xydata', eigenvectorsEigs(:,end), 'contour', 'on');
title(sprintf('eigs eigenvector, eigenvalue: %12.4e', eigenvaluesEigs(end))
xlabel('x'); ylabel('y');
subplot(1,2,2); axis equal;
pdeplot(p,e,t,'xydata', eigenvectorsPde(:,end), 'contour', 'on');
title(sprintf('pdeeig eigenvector, eigenvalue: %12.4e', eigenvaluesPde(end)
xlabel('x'); ylabel('y');
Maximum difference in eigenvalues from pdeeig and eigs: 1.421085e-13
3-111
3 Solving PDEs
3-112
Solve PDEs Programmatically
Solve PDEs Programmatically
In this section...
When You Need Programmatic Solutions on page 3-113
Data Structures in Partial Differential Equation Toolbox on page 3-113
Tips for Solving PDEs Programmatically on page 3-117
When You Need Programmatic Solutions
Although the PDE app provides a convenient working environment, there
are situations where the flexibility of using the command-line functions is
needed. These include:
Geometrical shapes other than straight lines, circular arcs, and elliptical
arcs
Nonstandard boundary conditions
Complicated PDE or boundary condition coefficients
More than two dependent variables in the system case
Nonlocal solution constraints
Special solution data processing and presentation itemize
The PDE app can still be a valuable aid in some of the situations presented
previously, if part of the modeling is done using the PDE app and then made
available for command-line use through the extensive data export facilities
of the PDE app.
Data Structures in Partial Differential Equation
Toolbox
The process of defining your problem and solving it is reflected in the design
of the PDE app. A number of data structures define different aspects of the
problem, and the various processing stages produce new data structures out
of old ones. See the following figure.
3-113
3 Solving PDEs
The rectangles are functions, and ellipses are data represented by matrices or
files. Arrows indicate data necessary for the functions.
As there is a definite direction in this diagram, you can cut into it by
presenting the needed data sets, and then continue downward. In the
following sections, we give pointers to descriptions of the precise formats of
the various data structures and files.
3-114
Solve PDEs Programmatically
Geometry
Description
mat r i x
Decomposed
Geometry
mat r i x
Mesh
data
Geometry
M- f i l e
Coefficient
mat r i x
Coefficient
M- f i l e
initmesh
refinemesh
assempde
Boundary
M- f i l e
Boundary
Condition
mat r i x
Solution
data
decsg
pdeplot
3-115
3 Solving PDEs
Constructive Solid Geometry Model
A Constructive Solid Geometry (CSG) model is specified by a Geometry
Description matrix, a set formula, and a Name Space matrix. For a description
of these data structures, see the reference page for decsg. At this level,
the problem geometry is defined by overlapping solid objects. These can be
created by drawing the CSG model in the PDE app and then exporting the
data using the Export Geometry Description, Set Formula, Labels
option from the Draw menu.
Decomposed Geometry
A decomposed geometry is specified by either a Decomposed Geometry matrix,
or by a Geometry file. Here, the geometry is described as a set of disjoint
minimal regions bounded by boundary segments and border segments. A
Decomposed Geometry matrix can be created from a CSG model by using
the function decsg. It can also be exported from the PDE app by selecting
the Export Decomposed Geometry, Boundary Conds option from
the Boundary menu. A Geometry file equivalent to a given Decomposed
Geometry matrix can be created using the wgeom function. A decomposed
geometry can be visualized with the pdegplot function. For descriptions of
the data structures of the Decomposed Geometry matrix and Geometry file,
see the respective reference pages for decsg and pdegeom.
Boundary Conditions
These are specified by either a Boundary Condition matrix, or a Boundary
file. Boundary conditions are given as functions on boundary segments. A
Boundary Condition matrix can be exported from the PDE app by selecting
the Export Decomposed Geometry, Boundary Conds option from the
Boundary menu. For a description of the data structures of the Boundary
Condition matrix and Boundary file, see the respective reference pages for
assemb and pdebound.
Equation Coefficients
The PDE is specified by either a Coefficient matrix or a Coefficient file for each
of the PDE coefficients c, a, f, and d. The coefficients are functions on the
subdomains. Coefficients can be exported from the PDE app by selecting the
Export PDE Coefficient option from the PDE menu. For the details on the
equation coefficient data structures, see the reference page for assempde.
3-116
Solve PDEs Programmatically
Mesh
A triangular mesh is described by the mesh data which consists of a Point
matrix, an Edge matrix, and a Triangle matrix. In the mesh, minimal regions
are triangulated into subdomains, and border segments and boundary
segments are broken up into edges. Mesh data is created from a decomposed
geometry by the function initmesh and can be altered by the functions
refinemesh and jigglemesh. The Export Mesh option from the Mesh menu
provides another way of creating mesh data. The adaptmesh function creates
mesh data as part of the solution process. The mesh may be plotted with
the pdemesh function. For details on the mesh data representation, see the
reference page for initmesh.
Solution
The solution of a PDE problem is represented by the solution vector. A
solution gives the value at each mesh point of each dependent variable,
perhaps at several points in time, or connected with different eigenvalues.
Solution vectors are produced from the mesh, the boundary conditions, and
the equation coefficients by assempde, pdenonlin, adaptmesh, parabolic,
hyperbolic, and pdeeig. The Export Solution option from the Solve menu
exports solutions to the workspace. Since the meaning of a solution vector
is dependent on its corresponding mesh data, they are always used together
when a solution is presented. For details on solution vectors, see the reference
page for assempde.
Post Processing and Presentation
Given a solution/mesh pair, a variety of tools is provided for the visualization
and processing of the data. pdeintrp and pdeprtni can be used to interpolate
between functions defined at triangle nodes and functions defined at triangle
midpoints. tri2grid interpolates a functions from a triangular mesh to a
rectangular grid. pdegrad and pdecgrad compute gradients of the solution.
pdeplot has a large number of options for plotting the solution. pdecont and
pdesurf are convenient shorthands for pdeplot.
Tips for Solving PDEs Programmatically
Use the export facilities of the PDE app as much as you can. They provide
data structures with the correct syntax, and these are good starting points
that you can modify to suit your needs.
3-117
3 Solving PDEs
Working with the system matrices and vectors produced by assema and
assemb can sometimes be valuable. When solving the same equation for
different loads or boundary conditions, it pays to assemble the stiffness matrix
only once. Point loads on a particular node can be implemented by adding the
load to the corresponding row in the right side vector. A nonlocal constraint
can be incorporated into the H and R matrices.
An example of a handwritten Coefficient file is circlef.m, which produces a
point load. You can find the full example in pdedemo7 and on the assempde
reference page.
The routines for adaptive mesh generation and solution are powerful but can
lead to dense meshes and thus long computation times. Setting the Ngen
parameter to one limits you to a single refinement step. This step can then be
repeated to show the progress of the refinement. The Maxt parameter helps
you stop before the adaptive solver generates too many triangles. An example
of a handwritten triangle selection function is circlepick, used in pdedemo7.
Remember that you always need a decomposed geometry with adaptmesh.
Deformed meshes are easily plotted by adding offsets to the Point matrix p.
Assuming two variables stored in the solution vector u:
np=size(p,2);
pdemesh(p+scale*[u(1:np) u(np+1:np+np)]',e,t)
The time evolution of eigenmodes is obtained by, e.g.,
u1=u(:,mode)*cos(sqrt(l(mode))*tlist) % hyperbolic
for positive eigenvalues in hyperbolic problems, or
u1=u(:,mode)*exp(-l(mode)*tlist); % parabolic
in parabolic problems. This makes nice animations, perhaps together with
deformed mesh plots.
3-118
Solve Poissons Equation on a Grid
Solve Poissons Equation on a Grid
While the general strategy of Partial Differential Equation Toolbox software
is to use the MATLAB built-in solvers for sparse systems, there are situations
where faster solution algorithms are available. One such example is found
when solving Poissons equation
u = f in
with Dirichlet boundary conditions, where is a rectangle.
For the fast solution algorithms to work, the mesh on the rectangle must be a
regular mesh. In this context it means that the first side of the rectangle is
divided into N
1
segments of length h
1
, the second into N
2
segments of length
h
2
, and (N
1
+ 1) by (N
2
+ 1) points are introduced on the regular grid thus
defined. The triangles are all congruent with sides h
1
, h
2
and a right angle
in between.
The Dirichlet boundary conditions are eliminated in the usual way, and the
resulting problem for the interior nodes is Kv = F. If the interior nodes are
numbered from left to right, and then from bottom to top, the K matrix is
block tridiagonal. The N
2
1 diagonal blocks, here called T, are themselves
tridiagonal (N
1
1) by (N
1
1) matrices, with 2(h
1
/h
2
+ h
2
/h
1
) on the diagonal
and h
2
/h
1
on the subdiagonals. The subdiagonal blocks, here called I, are
h
1
/h
2
times the unit N
1
1 matrix.
The key to the solution of the problem Kv = F is that the problem Tw = f
is possible to solve using the discrete sine transform. Let S be the
(N
1
1) by (N
1
1) matrix with Sij= sin(ij/N
1
). Then S
1
TS = , where is
a diagonal matrix with diagonal entries 2(h
1
/h
2
+ h
2
/h
1
) 2h
2
/h
1
cos(i/N
1
).
w = S
1
S
1
f, but multiplying with S is nothing more than taking the
discrete sine transform, and multiplying with S
1
is the same as taking the
inverse discrete sine transform. The discrete sine transform can be efficiently
calculated using the fast Fourier transform on a sequence of length 2N
1
.
Solving Tw = f using the discrete sine transform would not be an advantage in
itself, since the system is tridiagonal and should be solved as such. However,
for the full system Ky = F, a transformation of the blocks in K turns it into
N
1
1 decoupled tridiagonal systems of size N
2
1. Thus, a solution algorithm
would look like
3-119
3 Solving PDEs
1 Divide F into N
2
1 blocks of length N
1
1, and perform an inverse discrete
sine transform on each block.
2 Reorder the elements and solve N
1
1 tridiagonal systems of size N
2
1,
with 2(h
1
/h
2
+ h
2
/h
1
) 2h
2
/h
1
cos(i/N
1
) on the diagonal, and h
1
/h
2
on the
subdiagonals.
3 Reverse the reordering, and perform N
2
1 discrete sine transforms on
the blocks of length N
1
1.
When using a fast solver such as this one, time and memory are also saved
since the matrix K in fact never has to be assembled. A drawback is that since
the mesh has to be regular, it is impossible to do adaptive mesh refinement.
The fast elliptic solver for Poissons equation is implemented in poisolv. The
discrete sine transform and the inverse discrete sine transform are computed
by dst and idst, respectively.
3-120
4
PDE App
You open the PDE app by entering pdetool at the command line. The main
components of the PDE app are the menus, the dialog boxes, and the toolbar.
PDE App Menus on page 4-2
File Menu on page 4-4
Edit Menu on page 4-7
Options Menu on page 4-9
Draw Menu on page 4-13
Boundary Menu on page 4-15
PDE Menu on page 4-19
Mesh Menu on page 4-21
Solve Menu on page 4-25
Plot Menu on page 4-31
Window Menu on page 4-38
Help Menu on page 4-39
4 PDE App
PDE App Menus
PDE app menus let you perform the following operations:
File menu. From the File menu you can Open and Save model files that
contain a command sequence that reproduces your modeling session. You
can also print the current graphics and exit the PDE app.
Edit menu. From the Edit menu you can cut, clear, copy, and paste the
solid objects. There is also a Select All option.
Options menu. The Options menu contains options such as toggling the
axis grid, a snap-to-grid feature, and zoom. You can also adjust the axis
limits and the grid spacing, select the application mode, and refresh the
PDE app.
Draw menu. From the Draw menu you can select the basic solid objects
such as circles and polygons. You can then draw objects of the selected
type using the mouse. From the Draw menu you can also rotate the solid
objects and export the geometry to the MATLAB main workspace.
Boundary menu. From the Boundary menu you access a dialog box
where you define the boundary conditions. Additionally, you can label
edges and subdomains, remove borders between subdomains, and export
the decomposed geometry and the boundary conditions to the workspace.
PDE menu. The PDE menu provides a dialog box for specifying the PDE,
and there are menu options for labeling subdomains and exporting PDE
coefficients to the workspace.
Mesh menu. From the Mesh menu you create and modify the triangular
mesh. You can initialize, refine, and jiggle the mesh, undo previous mesh
changes, label nodes and triangles, display the mesh quality, and export
the mesh to the workspace.
Solve menu. From the Solve menu you solve the PDE. You can also open
a dialog box where you can adjust the solve parameters, and you can export
the solution to the workspace.
Plot menu. From the Plot menu you can plot a solution property. A dialog
box lets you select which property to plot, which plot style to use and
several other plot parameters. If you have recorded a movie (animation) of
the solution, you can export it to the workspace.
4-2
PDE App Menus
Window menu. The Window menu lets you select any currently open
MATLAB figure window. The selected window is brought to the front.
Help menu. The Help menu provides a brief help window.
4-3
4 PDE App
File Menu
New on page
4-4
Create a new (empty) Constructive Solid Geometry
(CSG) model.
Open on page
4-5
Load a model file from disk.
Save Save the PDE app session to a model file.
Save As on page
4-5
Save the PDE app session to a new model file.
Export Image
on page 4-5
Save the current figure in one of a variety of image
formats.
Print on page
4-6
Print a hardcopy of a figure.
Exit Exit the PDE app.
New
New deletes the current CSG model and creates a new, empty model called
Untitled.
4-4
File Menu
Open
Open displays a dialog box with a list of existing files from which you can
select the file that you want to load. You can list the contents of a different
folder by changing the path in the Selection text box. You can use the scroll
bar to display more filenames. You can select a file by double-clicking the
filename or by clicking the filename and then clicking the Done button. When
you select a file, the CSG model that is stored in the model file is loaded into
the workspace and displayed. Also, the equation, the boundary conditions,
and information about the mesh and the solution are loaded if present, and
the modeling and solution process continues to the same status as when you
saved the file.
Save As
Save As displays a dialog box in which you can specify the name of the file in
which to save the CSG model and other information regarding the PDE app
session. You can also change the folder in which it is saved. If the filename is
given without a .m extension, .m is appended automatically.
The PDE app session is stored in a model file, which contains a sequence of
drawing commands and commands to recreate the modeling environment
(axes scaling, grid, etc.). If you have already defined boundary conditions,
PDE coefficients, created a triangular mesh, and solved the PDE, further
commands to recreate the modeling and solution of the PDE problem are also
included in the model file. The PDE app can be started from the command
line by entering the name of a model file. The model in the file is then directly
loaded into the PDE app.
Export Image
Save the current figure as a file in your choice of formats. Available formats
include:
Bitmap (.bmp)
EPS (.eps)
JPEG (.jpg)
Portable Document Format (.pdf)
Portable Network Graphics (.png)
4-5
4 PDE App
TIFF (.tif)
Print
Print displays a dialog box for printing a hardcopy of a figure. Only the main
part of the figure is printed, not the upper and lower menu and information
parts. In the dialog box, you can enter any device option that is available for
the MATLAB print command. The default device option is -dps (PostScript

for black and white printers). The paper orientation can be set to portrait,
landscape, or tall, and you can print to a printer or to file.
4-6
Edit Menu
Edit Menu
Undo Undo the last line when drawing a polygon.
Cut Move the selected solid objects to the Clipboard.
Copy Copy the selected objects to the Clipboard, leaving them
intact in their original location.
Paste on
page 4-8
Copy the contents of the Clipboard to the current CSG model.
Clear Delete the selected objects.
Select All Select all solid objects in the current CSG model. Also, select
all outer boundaries or select all subdomains.
4-7
4 PDE App
Paste
Paste displays a dialog box for pasting the contents of the Clipboard on to the
current CSG model. The Clipboard contents can be repeatedly pasted adding
a specified x- and y-axis displacement to the positions of the Clipboard objects.
Using the default valueszero displacement and one repetitionthe
Clipboard contents is inserted at its original position.
4-8
Options Menu
Options Menu
Grid Turn grid on/off.
Grid Spacing on page
4-10
Adjust the grid spacing.
Snap Turn the snap-to-grid feature on/off.
Axes Limits on page
4-11
Change the scaling of the drawing axes.
Axis Equal Turn the axis equal feature on/off.
Turn off Toolbar Help Turn off help texts for the toolbar buttons.
Zoom Turn zoom feature on/off.
Application Select application mode.
Refresh Redisplay all graphical objects in the PDE app.
4-9
4 PDE App
Grid Spacing
In the Grid Spacing dialog box, you can adjust the x-axis and y-axis grid
spacing. By default, the MATLAB automatic linear grid spacing is used. If
you turn off the Auto check box, the edit fields for linear spacing and extra
ticks are enabled. For example, the default linear spacing -1.5:0.5:1.5 can
be changed to -1.5:0.2:1.5. In addition, you can add extra ticks so that the
grid can be customized to aid in drawing the desired 2-D domain. Extra tick
entries can be separated using spaces, commas, semicolons, or brackets.
Examples:
pi
2/3, 0.78, 1.1
-0.123; pi/4
Clicking the Apply button applies the entered grid spacing; clicking the Done
button ends the Grid Spacing dialog.
4-10
Options Menu
Axes Limits
In the Axes Limits dialog box, the range of the x-axis and the y-axis can be
adjusted. The axis range should be entered as a 1-by-2 MATLAB vector such
as [-10 10]. If you select the Auto check box, automatic scaling of the axis
is used.
Clicking the Apply button applies the entered axis ranges; clicking the Close
button ends the Axes Limits dialog.
Application
4-11
4 PDE App
From the Application submenu, you can select from 10 available application
modes. The application modes can also be selected using the pop-up menu in
the upper right corner of the PDE app.
The available application modes are:
Generic Scalar (the default mode)
Generic System
Structural Mechanics Plane Stress
Structural Mechanics Plane Strain
Electrostatics
Magnetostatics
AC Power Electromagnetics
Conductive Media DC
Heat Transfer
Diffusion
4-12
Draw Menu
Draw Menu
Draw Mode Enter draw mode.
Rectangle/square Draw a rectangle/square starting at a
corner. Using the left mouse button,
click-and-drag to create a rectangle.
Using the right mouse button (or
Ctrl+click), click-and-drag to create a
square.
Rectangle/square (centered) Draw a rectangle/square starting at the
center. Using the left mouse button,
click-and-drag to create a rectangle.
Using the right mouse button (or
Ctrl+click), click-and-drag to create a
square.
Ellipse/circle Draw an ellipse/circle starting at the
perimeter. Using the left mouse button,
click-and-drag to create an ellipse. Using
the right mouse button (or Ctrl+click),
click-and-drag to create a circle.
4-13
4 PDE App
Ellipse/circle (centered) Draw an ellipse/circle starting at the
center. Using the left mouse button,
click-and-drag to create an ellipse. Using
the right mouse button (or Ctrl+click),
click-and-drag to create a circle.
Polygon Draw a polygon. You can close the
polygon by pressing the right mouse
button. Clicking at the starting vertex
also closes the polygon.
Rotate on page 4-14 Rotate selected objects.
Export Geometry
Description, Set Formula,
Labels
Export the Geometry Description matrix
gd, the set formula string sf, and the
Name Space matrix ns (labels) to the
main workspace.
Rotate
Rotate opens a dialog box where you can enter the angle of rotation in
degrees. The selected objects are then rotated by the number of degrees
that you specify. The rotation is done counter clockwise for positive rotation
angles. By default, the rotation center is the center-of-mass of the selected
objects. If the Use center-of-mass option is not selected, you can enter a
rotation center (xc,yc) as a 1-by-2 MATLAB vector such as [-0.4 0.3].
4-14
Boundary Menu
Boundary Menu
Boundary Mode Enter the boundary mode.
Specify Boundary Conditions Specify boundary conditions for the
selected boundaries. If no boundaries
are selected, the entered boundary
condition applies to all boundaries.
Show Edge Labels Toggle the labeling of the edges (outer
boundaries and subdomain borders)
on/off. The edges are labeled using the
column number in the Decomposed
Geometry matrix.
Show Subdomain Labels Toggle the labeling of the subdomains
on/off. The subdomains are labeled
using the subdomain numbering in the
Decomposed Geometry matrix.
Remove Subdomain Border Remove selected subdomain borders.
Remove All Subdomain
Borders
Remove all subdomain borders.
Export Decomposed
Geometry, Boundary Conds
Export the Decomposed Geometry
matrix g and the Boundary Condition
matrix b to the main workspace.
4-15
4 PDE App
Specify Boundary Conditions in the PDE App
Specify Boundary Conditions opens a dialog box where you can specify
the boundary condition for the selected boundary segments. There are three
different condition types:
Generalized Neumann conditions, where the boundary condition is
determined by the coefficients q and g according to the following equation:

n c qu g u . V
( )
+ =
In the system cases, q is a 2-by-2 matrix and g is a 2-by-1 vector.
Dirichlet conditions: u is specified on the boundary. The boundary
condition equation is hu = r, where h is a weight factor that can be applied
(normally 1).
In the system cases, h is a 2-by-2 matrix and r is a 2-by-1 vector.
Mixed boundary conditions (system cases only), which is a mix of Dirichlet
and Neumann conditions. q is a 2-by-2 matrix, g is a 2-by-1 vector, h is
a 1-by-2 vector, and r is a scalar.
The following figure shows the dialog box for the generic system PDE
(Options > Application > Generic System).
4-16
Boundary Menu
For boundary condition entries you can use the following variables in a valid
MATLAB expression:
The 2-D coordinates x and y.
A boundary segment parameter s, proportional to arc length. s is 0 at
the start of the boundary segment and increases to 1 along the boundary
segment in the direction indicated by the arrow.
The outward normal vector components nx and ny. If you need the
tangential vector, it can be expressed using nx and ny since t
x
= n
y
and
t
y
= n
x
.
The solution u.
The time t.
4-17
4 PDE App
Note If the boundary condition is a function of the solution u, you must use
the nonlinear solver. If the boundary condition is a function of the time t, you
must choose a parabolic or hyperbolic PDE.
Examples: (100-80*s).*nx, and cos(x.^2)
In the nongeneric application modes, the Description column contains
descriptions of the physical interpretation of the boundary condition
parameters.
4-18
PDE Menu
PDE Menu
PDE Mode Enter the partial differential equation
mode.
Show Subdomain Labels Toggle the labeling of the subdomains
on/off. The subdomains are labeled
using the subdomain numbering in the
decomposed geometry matrix.
PDE Specification in the
PDE App on page 4-20
Open the dialog box for specifying PDE
coefficients and types.
Export PDE Coefficients Export current PDE coefficients to the
main workspace. The resulting workspace
variables are strings.
4-19
4 PDE App
PDE Specification in the PDE App
PDE Specification opens a dialog box where you enter the type of partial
differential equation and the applicable parameters. The dimension of
the parameters depends on the dimension of the PDE. The following
description applies to scalar PDEs. If you select a nongeneric application
mode, application-specific PDEs and parameters replace the standard PDE
coefficients.
Each of the coefficients c, a, f, and d can be given as a valid MATLAB
expression for computing coefficient values at the triangle centers of mass.
These variables are available:
x and y The x- and y-coordinates
u The solution
sd The subdomain number
ux and uy The x and y derivatives of the solution
t The time
For details, see Coefficients for Scalar PDEs in PDE App on page 2-19 and
2-D Systems in the PDE App on page 2-44.
4-20
Mesh Menu
Mesh Menu
Mesh Mode Enter mesh mode.
Initialize Mesh Build and display an initial triangular mesh.
Refine Mesh Uniformly refine the current triangular mesh.
Jiggle Mesh Jiggle the mesh.
Undo Mesh Change Undo the last mesh change. All mesh
generations are saved, so repeated Undo
Mesh Change eventually brings you back to
the initial mesh.
Display Triangle Quality Display a plot of the triangular mesh where
the individual triangles are colored according
to their quality. The quality measure is a
number between 0 and 1, where triangles
with a quality measure greater than 0.6 are
acceptable. For details on the triangle quality
measure, see pdetriq.
Show Node Labels Toggle the mesh node labels on/off. The node
labels are the column numbers in the Point
matrix p.
4-21
4 PDE App
Show Triangle Labels Toggle the mesh triangle labels on/off. The
triangle labels are the column numbers in the
triangle matrix t.
Parameters on page 4-23 Open dialog box for modification of mesh
generation parameters.
Export Mesh Export Point matrix p, Edge matrix e, and
Triangle matrix t to the main workspace.
4-22
Mesh Menu
Parameters
Parameters opens a dialog box containing mesh generation parameters. The
parameters used by the mesh initialization algorithm initmesh are:
4-23
4 PDE App
Maximum edge size: Largest triangle edge length (approximately). This
parameter is optional and must be a real positive number.
Mesh growth rate: The rate at which the mesh size increases away from
small parts of the geometry. The value must be between 1 and 2. The
default value is 1.3, i.e., the mesh size increases by 30%.
Mesher version: Choose the geometry triangulation algorithm. R2013a is
faster, and can mesh more geometries. preR2013a gives the same mesh as
previous toolbox versions.
Jiggle mesh: Toggles automatic jiggling of the initial mesh on/off.
The parameters used by the mesh jiggling algorithm jigglemesh are:
Jiggle mode: Select a jiggle mode from a pop-up menu. Available modes
are on, optimize minimum, and optimize mean. on jiggles the mesh once.
Using the jiggle mode optimize minimum, the jiggling process is repeated
until the minimum triangle quality stops increasing or until the iteration
limit is reached. The same applies for the optimize mean option, but it
tries to increase the mean triangle quality.
Number of jiggle iterations: Iteration limit for the optimize minimum
and optimize mean modes. Default: 20.
Finally, for the mesh refinement algorithm refinemesh, the Refinement
method can be regular or longest. The default refinement method is
regular, which results in a uniform mesh. The refinement method longest
always refines the longest edge on each triangle.
4-24
Solve Menu
Solve Menu
Solve PDE Solve the partial differential equation for the
current CSG model and triangular mesh, and plot
the solution (the automatic solution plot can be
disabled).
Parameters on page
4-26
Open dialog box for entry of PDE solve parameters.
Export Solution Export the PDE solution vector u and, if applicable,
the computed eigenvalues l to the main workspace.
4-25
4 PDE App
Parameters
Elliptic Equations
Parameters opens a dialog box where you can enter the solve parameters.
The set of solve parameters differs depending on the type of PDE.
Elliptic PDEs. By default, no specific solve parameters are used, and
the elliptic PDEs are solved using the basic elliptic solver assempde.
4-26
Solve Menu
Optionally, the adaptive mesh generator and solver adaptmesh can be used.
For the adaptive mode, the following parameters are available:
- Adaptive mode. Toggle the adaptive mode on/off.
- Maximum number of triangles. The maximum number of new
triangles allowed (can be set to Inf). A default value is calculated based
on the current mesh.
- Maximum number of refinements. The maximum number of
successive refinements attempted.
- Triangle selection method. There are two triangle selection methods,
described below. You can also supply your own function.
Worst triangles. This method picks all triangles that are worse than
a fraction of the value of the worst triangle (default: 0.5). For more
details, see pdetriq.
Relative tolerance. This method picks triangles using a relative
tolerance criterion (default: 1E-3). For more details, see pdeadgsc.
User-defined function. Enter the name of a user-defined triangle
selection method. See pdedemo7 for an example of a user-defined
triangle selection method.
- Function parameter. The function parameter allows fine-tuning of the
triangle selection methods. For the worst triangle method (pdeadworst),
it is the fraction of the worst value that is used to determine which
triangles to refine. For the relative tolerance method, it is a tolerance
parameter that controls how well the solution fits the PDE.
- Refinement method. Can be regular or longest. See the Parameters
dialog box description in Mesh Menu on page 4-21.
If the problem is nonlinear, i.e., parameters in the PDE are directly
dependent on the solution u, a nonlinear solver must be used. The following
parameters are used:
- Use nonlinear solver. Toggle the nonlinear solver on/off.
- Nonlinear tolerance. Tolerance parameter for the nonlinear solver.
- Initial solution. An initial guess. Can be a constant or a function of
x and y given as a MATLAB expression that can be evaluated on the
nodes of the current mesh.
4-27
4 PDE App
Examples: 1, and exp(x.*y). Optional parameter, defaults to zero.
- Jacobian. Jacobian approximation method: fixed (the default), a fixed
point iteration, lumped, a lumped (diagonal) approximation, or full,
the full Jacobian.
- Norm. The type of norm used for computing the residual. Enter as
energy for an energy norm, or as a real scalar p to give the lp norm. The
default is Inf, the infinity (maximum) norm.
Note The adaptive mode and the nonlinear solver can be used together.
- Parabolic PDEs. The solve parameters for the parabolic PDEs are:
Time. A MATLAB vector of times at which a solution to the parabolic
PDE should be generated. The relevant time span is dependent on the
dynamics of the problem.
Examples: 0:10, and logspace(-2,0,20)
u(t0). The initial value u(t
0
) for the parabolic PDE problem The initial
value can be a constant or a column vector of values on the nodes
of the current mesh.
Relative tolerance. Relative tolerance parameter for the ODE
solver that is used for solving the time-dependent part of the parabolic
PDE problem.
Absolute tolerance. Absolute tolerance parameter for the ODE
solver that is used for solving the time-dependent part of the parabolic
PDE problem.
4-28
Solve Menu
Hyperbolic Equations
- Hyperbolic PDEs. The solve parameters for the hyperbolic PDEs are:
Time. A MATLAB vector of times at which a solution to the hyperbolic
PDE should be generated. The relevant time span is dependent on the
dynamics of the problem.
Examples: 0:10, and logspace(-2,0,20)
u(t0). The initial value u(t
0
) for the hyperbolic PDE problem. The
initial value can be a constant or a column vector of values on the
nodes of the current mesh.
u(t0). The initial value u(t
0
) for the hyperbolic PDE problem. You
can use the same formats as for u(t0).
4-29
4 PDE App
Relative tolerance. Relative tolerance parameter for the ODE solver
that is used for solving the time-dependent part of the hyperbolic
PDE problem.
Absolute tolerance. Absolute tolerance parameter for the ODE
solver that is used for solving the time-dependent part of the
hyperbolic PDE problem.
Eigenvalue Equations
- Eigenvalue problems. For the eigenvalue PDE, the only solve parameter
is the Eigenvalue search range, a two-element vector, defining an
interval on the real axis as a search range for the eigenvalues. The left
side can be -Inf.
Examples: [0 100], [-Inf 50]
4-30
Plot Menu
Plot Menu
Plot Solution Display a plot of the solution.
Parameters on
page 4-31
Open dialog box for plot selection.
Export Movie If a movie has been recorded, the movie matrix M is
exported to the main workspace.
Parameters
Plot Selection Dialog Box
4-31
4 PDE App
Parameters opens a dialog box containing options controlling the plotting
and visualization.
The upper part of the dialog box contains four columns:
Plot type (far left) contains a row of six different plot types, which can be
used for visualization:
- Color. Visualization of a scalar property using colored surface objects.
- Contour. Visualization of a scalar property using colored contour lines.
The contour lines can also enhance the color visualization when both
plot types (Color and Contour) are checked. The contour lines are
then drawn in black.
- Arrows. Visualization of a vector property using arrows.
- Deformed mesh. Visualization of a vector property by deforming the
mesh using the vector property. The deformation is automatically scaled
to 10% of the problem domain. This plot type is primarily intended for
visualizing x- and y-displacements (u and v) for problems in structural
mechanics. If no other plot type is selected, the deformed triangular
mesh is displayed.
- Height (3-D plot). Visualization of a scalar property using height
(z-axis) in a 3-D plot. 3-D plots are plotted in separate figure windows. If
the Color and Contour plot types are not used, the 3-D plot is simply
a mesh plot. You can visualize another scalar property simultaneously
using Color and/or Contour, which results in a 3-D surface or contour
plot.
- Animation. Animation of time-dependent solutions to parabolic and
hyperbolic problems. If you select this option, the solution is recorded
and then animated in a separate figure window using the MATLAB
movie function.
A color bar is added to the plots to map the colors in the plot to the magnitude
of the property that is represented using color or contour lines.
Property contains four pop-up menus containing lists of properties that
are available for plotting using the corresponding plot type. From the first
pop-up menu you control the property that is visualized using color and/or
contour lines. The second and third pop-up menus contain vector valued
4-32
Plot Menu
properties for visualization using arrows and deformed mesh, respectively.
From the fourth pop-up menu, finally, you control which scalar property to
visualize using z-height in a 3-D plot. The lists of properties are dependent
on the current application mode. For the generic scalar mode, you can
select the following scalar properties:
- u. The solution itself.
- abs(grad(u)). The absolute value of u, evaluated at the center of each
triangle.
- abs(c*grad(u)). The absolute value of c u, evaluated at the center
of each triangle.
- user entry. A MATLAB expression returning a vector of data defined on
the nodes or the triangles of the current triangular mesh. The solution
u, its derivatives ux and uy, the x and y components of c u, cux and
cuy, and x and y are all available in the local workspace. You enter the
expression into the edit box to the right of the Property pop-up menu
in the User entry column.
Examples: u.*u, x+y
The vector property pop-up menus contain the following properties in the
generic scalar case:
- -grad(u). The negative gradient of u, u.
- -c*grad(u). c times the negative gradient of u, c u.
- user entry. A MATLAB expression [px; py] returning a 2-by-ntri
matrix of data defined on the triangles of the current triangular mesh
(ntri is the number of triangles in the current mesh). The solution u, its
derivatives ux and uy, the x and y components of c u, cux and cuy,
and x and y are all available in the local workspace. Data defined on the
nodes is interpolated to triangle centers. You enter the expression into
the edit field to the right of the Property pop-up menu in the User
entry column.
Examples: [ux;uy], [x;y]
For the generic system case, the properties available for visualization using
color, contour lines, or z-height are u, v, abs(u,v), and a user entry. For
visualization using arrows or a deformed mesh, you can choose (u,v) or a
4-33
4 PDE App
user entry. For applications in structural mechanics, u and v are the x- and
y-displacements, respectively.
The variables available in the local workspace for a user entered expression
are the same for all scalar and system modes (the solution is always referred
to as u and, in the system case, v).
User entry contains four edit fields where you can enter your own
expression, if you select the user entry property from the corresponding
pop-up menu to the left of the edit fields. If the user entry property is not
selected, the corresponding edit field is disabled.
Plot style contains three pop-up menus from which you can control the
plot style for the color, arrow, and height plot types respectively. The
available plot styles for color surface plots are
- Interpolated shading. A surface plot using the selected colormap and
interpolated shading, i.e., each triangular area is colored using a linear,
interpolated shading (the default).
- Flat shading. A surface plot using the selected colormap and flat
shading, i.e., each triangular area is colored using a constant color.
You can use two different arrow plot styles:
- Proportional. The length of the arrow corresponds to the magnitude of
the property that you visualize (the default).
- Normalized. The lengths of all arrows are normalized, i.e., all arrows
have the same length. This is useful when you are interested in the
direction of the vector field. The direction is clearly visible even in areas
where the magnitude of the field is very small.
For height (3-D plots), the available plot styles are:
- Continuous. Produces a smooth continuous plot by interpolating data
from triangle midpoints to the mesh nodes (the default).
- Discontinuous. Produces a discontinuous plot where data and z-height
are constant on each triangle.
A total of three properties of the solutiontwo scalar properties and one
vector fieldcan be visualized simultaneously. If the Height (3-D plot)
option is turned off, the solution plot is a 2-D plot and is plotted in the main
4-34
Plot Menu
axes of the PDE app. If the Height (3-D plot) option is used, the solution
plot is a 3-D plot in a separate figure window. If possible, the 3-D plot uses
an existing figure window. If you would like to plot in a new figure window,
simply type figure at the MATLAB command line.
Additional Plot Control Options
In the middle of the dialog box are a number of additional plot control options:
Plot in x-y grid. If you select this option, the solution is converted from
the original triangular grid to a rectangular x-y grid. This is especially
useful for animations since it speeds up the process of recording the movie
frames significantly.
Show mesh. In the surface plots, the mesh is plotted using black color if
you select this option. By default, the mesh is hidden.
Contour plot levels. For contour plots, the number of level curves, e.g.,
15 or 20 can be entered. Alternatively, you can enter a MATLAB vector of
levels. The curves of the contour plot are then drawn at those levels. The
default is 20 contour level curves.
Examples: [0:100:1000], logspace(-1,1,30)
Colormap. Using the Colormap pop-up menu, you can select from a
number of different colormaps: cool, gray, bone, pink, copper, hot, jet,
hsv, and prism.
Plot solution automatically. This option is normally selected. If turned
off, there will not be a display of a plot of the solution immediately upon
solving the PDE. The new solution, however, can be plotted using this
dialog box.
For the parabolic and hyperbolic PDEs, the bottom right portion of the Plot
Selection dialog box contains the Time for plot parameter.
Time for plot. A pop-up menu allows you to select which of the solutions
to plot by selecting the corresponding time. By default, the last solution is
plotted.
4-35
4 PDE App
Also, the Animation plot type is enabled. In its property field you find an
Options button. If you press it, an additional dialog box appears. It contains
parameters that control the animation:
Animation rate (fps). For the animation, this parameter controls the
speed of the movie in frames per second (fps).
Number of repeats. The number of times the movie is played.
Replay movie. If you select this option, the current movie is replayed
without rerecording the movie frames. If there is no current movie, this
option is disabled.
4-36
Plot Menu
For eigenvalue problems, the bottom right part of the dialog box contains a
pop-up menu with all eigenvalues. The plotted solution is the eigenvector
associated with the selected eigenvalue. By default, the smallest eigenvalue is
selected.
You can rotate the 3-D plots by clicking the plot and, while keeping the mouse
button down, moving the mouse. For guidance, a surrounding box appears.
When you release the mouse, the plot is redrawn using the new viewpoint.
Initially, the solution is plotted using -37.5 degrees horizontal rotation and
30 degrees elevation.
If you click the Plot button, the solution is plotted immediately using the
current plot setup. If there is no current solution available, the PDE is first
solved. The new solution is then plotted. The dialog box remains on the screen.
If you click the Done button, the dialog box is closed. The current setup is
saved but no additional plotting takes place.
If you click the Cancel button, the dialog box is closed. The setup remains
unchanged since the last plot.
4-37
4 PDE App
Window Menu
From the Window menu, you can select all currently open MATLAB figure
windows. The selected window is brought to the front.
4-38
Help Menu
Help Menu
PDETool Help Open documentation to pdetool entry.
PDE Toolbox Help Open documentation to Partial Differential Equation
Toolbox.
Examples Examples using the software.
About the PDE
Toolbox
Display a window with some program information.
4-39
4 PDE App
4-40
5
Finite Element Method
Elliptic Equations on page 5-2
Systems of PDEs on page 5-10
Parabolic Equations on page 5-13
Hyperbolic Equations on page 5-18
Eigenvalue Equations on page 5-19
Nonlinear Equations on page 5-24
References on page 5-30
5 Finite Element Method
Elliptic Equations
The basic elliptic equation handled by the software is
V V
( )
+ = c u au f ,
in , where is a bounded domain in the plane. c, a, f, and the unknown
solution u are complex functions defined on . c can also be a 2-by-2 matrix
function on . The boundary conditions specify a combination of u and its
normal derivative on the boundary:
Dirichlet: hu = r on the boundary .
Generalized Neumann:

n (cu) + qu = g on .
Mixed: Only applicable to systems. A combination of Dirichlet and
generalized Neumann.

n is the outward unit normal. g, q, h, and r are functions defined on .

Our nomenclature deviates slightly from the tradition for potential theory,
where a Neumann condition usually refers to the case q = 0 and our Neumann
would be called a mixed condition. In some contexts, the generalized Neumann
boundary conditions is also referred to as the Robin boundary conditions. In
variational calculus, Dirichlet conditions are also called essential boundary
conditions and restrict the trial space. Neumann conditions are also called
natural conditions and arise as necessary conditions for a solution. The
variational form of the Partial Differential Equation Toolbox equation with
Neumann conditions is given below.
The approximate solution to the elliptic PDE is found in three steps:
1 Describe the geometry of the domain and the boundary conditions. This
can be done either interactively using the PDE app or through MATLAB
files (see pdegeom and pdebound).
2 Build a triangular mesh on the domain . The software has mesh
generating and mesh refining facilities. A mesh is described by three
matrices of fixed format that contain information about the mesh points,
the boundary segments, and the triangles.
5-2
Elliptic Equations
3 Discretize the PDE and the boundary conditions to obtain a linear system
Ku = F. The unknown vector u contains the values of the approximate
solution at the mesh points, the matrix K is assembled from the coefficients
c, a, h, and q and the right-hand side F contains, essentially, averages of
f around each mesh point and contributions from g. Once the matrices K
and F are assembled, you have the entire MATLAB environment at your
disposal to solve the linear system and further process the solution.
More elaborate applications make use of the Finite Element Method (FEM)
specific information returned by the different functions of the software.
Therefore we quickly summarize the theory and technique of FEM solvers to
enable advanced applications to make full use of the computed quantities.
FEM can be summarized in the following sentence: Project the weak form of
the differential equation onto a finite-dimensional function space. The rest of
this section deals with explaining the preceding statement.
We start with the weak form of the differential equation. Without restricting
the generality, we assume generalized Neumann conditions on the whole
boundary, since Dirichlet conditions can be approximated by generalized
Neumann conditions. In the simple case of a unit matrix h, setting g = qr and
then letting q yields the Dirichlet condition because division with a very
large q cancels the normal derivative terms. The actual implementation is
different, since the preceding procedure may create conditioning problems.
The mixed boundary condition of the system case requires a more complicated
treatment, described in Systems of PDEs on page 5-10.
Assume that u is a solution of the differential equation. Multiply the equation
with an arbitrary test function v and integrate on :
( ) = V V ( ) +
} }
. c u v auv dx fv dx
O O
Integrate by parts (i.e., use Greens formula) to obtain
c fv u v auv dx n c u v ds dx V V + ( ) V
( ) ( )
=
} } }
c
.
O O O

The boundary integral can be replaced by the boundary condition:

5-3
5 Finite Element Method
c fv u v auv dx qu g v ds dx V V + ( ) + ( )
( )
=
} } }
c
.
O O O
Replace the original problem with Find u such that
c u v auv fv dx qu g v ds v V V + ( ) + ( )
( )
=
} }
c
.
O O
0
This equation is called the variational, or weak, form of the differential
equation. Obviously, any solution of the differential equation is also a solution
of the variational problem. The reverse is true under some restrictions on
the domain and on the coefficient functions. The solution of the variational
problem is also called the weak solution of the differential equation.
The solution u and the test functions v belong to some function space V. The
next step is to choose an Np-dimensional subspace V V
N
p
c . Project the
weak form of the differential equation onto a finite-dimensional function space
simply means requesting u and v to lie in V
N
p
rather than V. The solution of
the finite dimensional problem turns out to be the element of V
N
p
that lies
closest to the weak solution when measured in the energy norm. Convergence
is guaranteed if the space V
N
p
tends to V as N
p
. Since the differential
operator is linear, we demand that the variational equation is satisfied for
N
p
test-functions
i
V
N
p
that form a basis, i.e.,
c u au f dx qu g ds i N
i i i i p
V V + ( ) + ( ) =
( )
=
} }
c
,..., . , | | | |
O O
0 1
Expand u in the same basis of V
N
p
elements
u x U x
j j
j
N
p
( ) ( ), =
=

|
1
and obtain the system of equations
5-4
Elliptic Equations
c U a dx q ds f dx g
j i j i j i i
j
N
j
p
V V +
( ) ( )
+
|
\

|
.
|
= +
} }

}
c
=
| | | | | | | |
O O O
1
ii p
ds i N
c
}
=
O
, , ... ,
.
1
Use the following notations:
K c dx
i j j i ,
=
( )
V V
}
| |
O
(stiffness matrix)
M a dx
i j j i ,
=
}
| |
O
(mass matrix)
Q q ds
i j j i ,
=
c
}
| |
O
F f dx
i i
=
}
|
O
G g ds
i i
=
c
}
|
O
and rewrite the system in the form
(K + M + Q)U = F + G.
K, M, and Q are N
p
-by-N
p
matrices, and F and G are N
p
-vectors. K, M, and F
are produced by assema, while Q, G are produced by assemb. When it is not
necessary to distinguish K, M, and Q or F and G, we collapse the notations to
KU = F, which form the output of assempde.
When the problem is self-adjoint and elliptic in the usual mathematical
sense, the matrix K + M + Q becomes symmetric and positive definite. Many
common problems have these characteristics, most notably those that can also
be formulated as minimization problems. For the case of a scalar equation, K,
M, and Q are obviously symmetric. If c(x) > 0, a(x) 0 and q(x) 0 with
q(x) > 0 on some part of , then, if U 0.
5-5
5 Finite Element Method
U K M Q U c u au qu dx ds U
T
+ + ( ) = +
( )
+ > =
} }
c
2
2 2
0 0
O O
, . if
U
T
(K + M + Q)U is the energy norm. There are many choices of the
test-function spaces. The software uses continuous functions that are linear
on each triangle of the mesh. Piecewise linearity guarantees that the integrals
defining the stiffness matrix K exist. Projection onto V
N
p
is nothing more
than linear interpolation, and the evaluation of the solution inside a triangle
is done just in terms of the nodal values. If the mesh is uniformly refined,
V
N
p
approximates the set of smooth functions on .
A suitable basis for V
N
p
is the set of tent or hat functions
i
. These are
linear on each triangle and take the value 0 at all nodes x
j
except for x
i
.
Requesting
i
(x
i
) = 1 yields the very pleasant property
u x U x U
i j j i
j
N
i
p
( ) = ( ) =
=

|
1
.
That is, by solving the FEM system we obtain the nodal values of the
approximate solution. The basis function
i
vanishes on all the triangles that
do not contain the node x
i
. The immediate consequence is that the integrals
appearing in K
i,j
, M
i,j
, Q
i,j
, F
i
and G
i
only need to be computed on the triangles
that contain the node x
i
. Secondly, it means that K
i,j
andM
i,j
are zero unless
x
i
and x
j
are vertices of the same triangle and thus K and M are very sparse
matrices. Their sparse structure depends on the ordering of the indices of
the mesh points.
The integrals in the FEM matrices are computed by adding the contributions
from each triangle to the corresponding entries (i.e., only if the corresponding
mesh point is a vertex of the triangle). This process is commonly called
assembling, hence the name of the function assempde.
The assembling routines scan the triangles of the mesh. For each triangle
they compute the so-called local matrices and add their components to the
correct positions in the sparse matrices or vectors. (The local 3-by-3 matrices
contain the integrals evaluated only on the current triangle. The coefficients
5-6
Elliptic Equations
are assumed constant on the triangle and they are evaluated only in the
triangle barycenter.) The integrals are computed using the midpoint rule.
This approximation is optimal since it has the same order of accuracy as
the piecewise linear interpolation.
Consider a triangle given by the nodes P
1
, P
2
, and P
3
as in the following figure.
P
P
P
1
2
3
P
c
P
b
x
y
1
1
The Local Triangle P1P2P3
Note The local 3-by-3 matrices contain the integrals evaluated only on the
current triangle. The coefficients are assumed constant on the triangle and
they are evaluated only in the triangle barycenter.
The simplest computations are for the local mass matrix m:
m a P x x a P
P P P
dx
i j c i j
P P P
c i j , ,
= ( )
( ) ( )
= ( )
( )
+
(
}
| | o
A
A
1 2 3
1 2 3
12
1
area
))
,
where P
c
is the center of mass of P
1
P
2
P
3
, i.e.,
P
P P P
c
=
+ +
1 2 3
3
.
5-7
5 Finite Element Method
The contribution to the right side F is just
f f P
P P P
i c
= ( )
( ) area A
1 2 3
3
.
For the local stiffness matrix we have to evaluate the gradients of the basis
functions that do not vanish on P
1
P
2
P
3
. Since the basis functions are linear on
the triangle P
1
P
2
P
3
, the gradients are constants. Denote the basis functions

1
,
2
, and
3
such that (P
i
) = 1. If P
2
P
3
= [x
1
,y
1
]
T
then we have that
V =
( )

(
|
1
1 2 3
1
1
1
2area AP P P
y
x
and after integration (taking c as a constant matrix on the triangle)
k
P P P
y x c P
y
x
i j j j c ,
, . =
( )

( )

(
1
4
1 2 3
1
1
area A
If two vertices of the triangle lie on the boundary , they contribute to the
line integrals associated to the boundary conditions. If the two boundary
points are P
1
and P
2
, then we have
Q q P
P P
i j
i j b i j , ,
, , , = ( )

+
( )
=
1 2
6
1 1 2 o
and
G g P
P P
i
i b
= ( )

=
1 2
2
1 2 , ,
where P
b
is the midpoint of P
1
P
2
.
For each triangle the vertices P
m
of the local triangle correspond to the indices
i
m
of the mesh points. The contributions of the individual triangle are added
to the matrices such that, e.g.,
5-8
Elliptic Equations
K t K k m n
i i i i m n
m n m n
, , ,
, , , , . + = 1 2 3
This is done by the function assempde. The gradients and the areas of the
triangles are computed by the function pdetrg.
The Dirichlet boundary conditions are treated in a slightly different manner.
They are eliminated from the linear system by a procedure that yields a
symmetric, reduced system. The function assempde can return matrices K,
F, B, and ud such that the solution is u = Bv + ud where Kv = F. u is an
N
p
-vector, and if the rank of the Dirichlet conditions is rD, then v has N
p
rD
components.
5-9
5 Finite Element Method
Systems of PDEs
Partial Differential Equation Toolbox software can also handle systems of N
partial differential equations over the domain . We have the elliptic system
V V
( )
+ = c u au f,
the parabolic system
d c au f
u
u
c
c
V V
( )
+ =
t
,
the hyperbolic system
d
u
c u au f
c
c
V V
( )
+ =
2
2
t
,
and the eigenvalue system
V V
( )
+ = c u au du ,
where c is an N-by-N-by-2-by-2 tensor. By the notation V V ( ) c u , we mean
the N-by-1 matrix with (i,1)-component.
c
c
c
c
+
c
c
c
c
+
c
c
c
c
+
c
c
c
c x
c
x x
c
y y
c
x y
c
i j i j i j i j , , , , , , , , , , , , 1 1 1 2 2 1 2 2
yy
u
j
N
j
|
\

|
.
|
=

1
The symbols a and d denote N-by-N matrices, and u denotes column vectors
of lengthN.
The elements c
ijkl
, a
ij
, d
ij
, and f
i
of c, a, d, and f are stored row-wise in
the MATLAB matrices c, a, d, and f. The case of identity, diagonal, and
symmetric matrices are handled as special cases. For the tensor c
ijkl
this
applies both to the indices i and j, and to the indices k and l.
5-10
Systems of PDEs
Partial Differential Equation Toolbox software does not check the ellipticity of
the problem, and it is quite possible to define a system that is not elliptic in
the mathematical sense. The preceding procedure that describes the scalar
case is applied to each component of the system, yielding a symmetric positive
definite system of equations whenever the differential system possesses
these characteristics.
The boundary conditions now in general are mixed, i.e., for each point on the
boundary a combination of Dirichlet and generalized Neumann conditions,
hu r
n c qu g h u
=

( )
+ = + ' V .
By the notation n c u
( )
V we mean the N-by-1 matrix with (i,1)-component
cos( ) cos( ) sin( ) sin(
, , , , , , , , ,
o o o c c c
x y x
i j i j i j 1 1 1 2 2 1
c
c
c
c
c
c
+ + + oo)
, , ,
c
y
i j
j
N
j 2 2
1
c
c
|
\

|
.
|
=

u
where the outward normal vector of the boundary is n = ( ) cos( ),sin( ) o o .
There are M Dirichlet conditions and the h-matrix is M-by-N, M 0. The
generalized Neumann condition contains a source ' h , where the Lagrange
multipliers are computed such that the Dirichlet conditions become
satisfied. In a structural mechanics problem, this term is exactly the reaction
force necessary to satisfy the kinematic constraints described by the Dirichlet
conditions.
The rest of this section details the treatment of the Dirichlet conditions and
may be skipped on a first reading.
Partial Differential Equation Toolbox software supports two implementations
of Dirichlet conditions. The simplest is the Stiff Spring model, so named for
its interpretation in solid mechanics. See Elliptic Equations on page 5-2 for
the scalar case, which is equivalent to a diagonal h-matrix. For the general
case, Dirichlet conditions
hu = r
5-11
5 Finite Element Method
are approximated by adding a term
L( ) ' ' h h h u r
to the equations KU = F, where L is a large number such as 10
4
times a
representative size of the elements of K.
When this number is increased, hu = r will be more accurately satisfied,
but the potential ill-conditioning of the modified equations will become more
serious.
The second method is also applicable to general mixed conditions with
nondiagonal h, and is free of the ill-conditioning, but is more involved
computationally. Assume that there are N
p
nodes in the triangulation. Then
the number of unknowns is N
p
N = N
u
. When Dirichlet boundary conditions
fix some of the unknowns, the linear system can be correspondingly reduced.
This is easily done by removing rows and columns when u values are given,
but here we must treat the case when some linear combinations of the
components of u are given, hu = r. These are collected into HU = R where H
is an M-by-N
u
matrix and R is an M-vector.
With the reaction force term the system becomes
KU +H = F
HU = R.
The constraints can be solved for M of the U-variables, the remaining called
V, an N
u
M vector. The null space of H is spanned by the columns of B,
and U = BV + u
d
makes U satisfy the Dirichlet conditions. A permutation
to block-diagonal form exploits the sparsity of H to speed up the following
computation to find B in a numerically stable way. can be eliminated by
premultiplying by B since, by the construction, HB = 0 or BH = 0. The
reduced system becomes
B KBV = B F BKu
d
which is symmetric and positive definite if K is.
5-12
Parabolic Equations
Parabolic Equations
In this section...
Reducing Parabolic Equations to Elliptic Equations on page 5-13
Solve a Parabolic Equation on page 5-15
Reducing Parabolic Equations to Elliptic Equations
The elliptic solver allows other types of equations to be more easily
implemented. In this section, we show how the parabolic equation can
be reduced to solving elliptic equations. This is done using the function
parabolic.
Consider the equation
d c au f
u
t
u
c
c
V V
( )
+ = in O,
with the initial condition
u(x,0) = u
0
(x) for x
and boundary conditions of the same kind as for the elliptic equation on .
The heat equation reads
C k h u u f
u
t
u
c
c
V V
( )
+ ( ) =

in the presence of distributed heat loss to the surroundings. is the density, C

is the thermal capacity, k is the thermal conductivity, h is the film coefficient,
u

is the ambient temperature, and f is the heat source.

For time-independent coefficients, the steady-state solution of the equation is
the solution to the standard elliptic equation
(cu) + au = f.
5-13
5 Finite Element Method
Assuming a triangular mesh on and t 0, expand the solution to the PDE
(as a function of x) in the Finite Element Method basis:
u x t U t x
i i
i
( , ) ( ) ( ). =

|
Plugging the expansion into the PDE, multiplying with a test function

j
, integrating over , and applying Greens formula and the boundary
conditions yield
d
dU t
dt
dx c a dx q ds
j i
i
i
j i j i j i
| | | | | | | |
( )
+ V V ( ) +
( )
+
|
\

|
.
}

} }
c O O O
||
= +

} }
c
U t
f dx g ds j
i
i
j j
( )
. | |
O O
In matrix notation, we have to solve the linear, large and sparse ODE system
M
dU
dt
KU F + = .
This method is traditionally called method of lines semidiscretization.
Solving the ODE with the initial value
U
i
(0) = u
0
(x
i
)
yields the solution to the PDE at each node x
i
and time t. Note that K and F
are the stiffness matrix and the right-hand side of the elliptic problem
(cu) + au = f in
with the original boundary conditions, while M is just the mass matrix of
the problem
(0u) + du = 0 in .
5-14
Parabolic Equations
When the Dirichlet conditions are time dependent, F contains contributions
from time derivatives of h and r. These derivatives are evaluated by finite
differences of the user-specified data.
The ODE system is ill conditioned. Explicit time integrators are forced by
stability requirements to very short time steps while implicit solvers can
be expensive since they solve an elliptic problem at every time step. The
numerical integration of the ODE system is performed by the MATLAB
ODE Suite functions, which are efficient for this class of problems. The time
step is controlled to satisfy a tolerance on the error, and factorizations of
coefficient matrices are performed only when necessary. When coefficients
are time dependent, the necessity of reevaluating and refactorizing the
matrices each time step may still make the solution time consuming, although
parabolic reevaluates only that which varies with time. In certain cases a
time-dependent Dirichlet matrix h(t) may cause the error control to fail, even
if the problem is mathematically sound and the solution u(t) is smooth. This
can happen because the ODE integrator looks only at the reduced solution v
with u = Bv + ud. As h changes, the pivoting scheme employed for numerical
stability may change the elimination order from one step to the next. This
means that B, v, and ud all change discontinuously, although u itself does not.
Solve a Parabolic Equation
This example shows how to solve a parabolic equation and to set an initial
condition as a variable.
1 At the MATLAB command prompt, type pdetool.
2 Draw a rectangle in the PDE app axes.
3 From the Draw menu, select Export Geometry Description, Set
Formula, Labels.
4 In the Export dialog box, enter gd sf ns. Click OK.
The exported variables are available in the MATLAB workspace.
5 From the Boundary menu, select Boundary Mode.
6 From the Boundary menu, select Specify Boundary Conditions.
5-15
5 Finite Element Method
7 Set the Neumann and Dirichlet boundary conditions. If these conditions
are not the same for all the stages, set the conditions accordingly.
8 From the Boundary menu, select Export Decomposed Geometry,
Boundary Conds.
9 In the Export dialog box, enter g b. Click OK.
10 From the PDE menu, select PDE Mode.
11 From the PDE menu, select PDE Specification.
12 Set the partial differential equation (PDE) coefficients, which are the same
for any value of time.
13 From the PDE menu, select Export PDE Coefficients.
14 In the Export dialog box, enter c a f d. Click OK.
15 From the Mesh menu, select Mesh Mode.
16 From the Mesh menu, select Parameters.
17 Verify the initial mesh, jiggle mesh, and refine mesh values. The mesh
is fixed for all stages.
18 From the Mesh menu, select Export Mesh.
19 In the Export dialog box, enter p e t. Click OK.
20 Save the workspace variables into a MAT-file by typing save data.mat
at the MATLAB command prompt.
21 Save the following code as a file:
clear all;
close all;
load data
%For the first stage you need to specify an
%initial condition, U0.
U0 = 0; %U0 expands to the correct size automatically.
5-16
Parabolic Equations
%Divide the time range into 4 stages.
time = {0:.01:1, 1:.05:3, 3:.1:5, 5:.5:20};
for i = 1:4
U1 = parabolic(U0,time{i},b,p,e,t,c,a,f,d);
for j = 1:size(U1,2)
H =pdeplot(p,e,t,'xydata',U1(:,j),'zdata',...
U1(:,j),'mesh','off');
set(gca,'ZLim',[-80 0]);
drawnow
end
%Reset the initial condition at all points.
U0 = U1(:,1);
end
This file uses the variables you defined in the MATLAB workspace to solve
a parabolic equation in stages. Within this file, you set the initial condition
as a variable.
5-17
5 Finite Element Method
Hyperbolic Equations
Using the same ideas as for the parabolic equation, hyperbolic implements
the numerical solution of
d
u
t
c u au f
c
c
V V
( )
+ =
2
2
,
for x in , with the initial conditions
u x u x
x v x
u
t
,
,
0
0
0
0
( ) =
( )
( ) =
( )
c
c
for all x in , and usual boundary conditions. In particular, solutions of the
equation u
tt
- cu = 0 are waves moving with speed c .
Using a given triangulation of , the method of lines yields the second order
ODE system
M
d U
dt
KU F
2
2
+ =
with the initial conditions
U u x
d
dt
U v x
i
i
i i
i i
0
0
0
0
( )
= ( )
( )
= ( )

after we eliminate the unknowns fixed by Dirichlet boundary conditions. As

before, the stiffness matrix K and the mass matrix M are assembled with the
aid of the function assempde from the problems
(cu) + au = f and (0u) + du = 0.
5-18
Eigenvalue Equations
Eigenvalue Equations
Partial Differential Equation Toolbox software handles the following basic
eigenvalue problem:
V V
( )
+ = c u au du ,
where is an unknown complex number. In solid mechanics, this is a problem
associated with wave phenomena describing, e.g., the natural modes of a
vibrating membrane. In quantum mechanics is the energy level of a bound
state in the potential well a(x).
The numerical solution is found by discretizing the equation and solving
the resulting algebraic eigenvalue problem. Let us first consider the
discretization. Expand u in the FEM basis, multiply with a basis element, and
integrate on the domain . This yields the generalized eigenvalue equation
KU = MU
where the mass matrix corresponds to the right side, i.e.,
M d x x x dx
i j j i ,
( ) ( ) ( ) =
}
| |
O
The matrices K and M are produced by calling assema for the equations
(cu) + au = 0 and (0u) + du = 0
In the most common case, when the function d(x) is positive, the mass matrix
M is positive definite symmetric. Likewise, when c(x) is positive and we have
Dirichlet boundary conditions, the stiffness matrix K is also positive definite.
The generalized eigenvalue problem, KU = MU, is now solved by the Arnoldi
algorithm applied to a shifted and inverted matrix with restarts until all
eigenvalues in the user-specified interval have been found.
Let us describe how this is done in more detail. You may want to look at
the example provided in the section Eigenvalue Problems on page 3-100,
where an actual run is reported.
5-19
5 Finite Element Method
First a shift is determined close to where we want to find the eigenvalues.
When both K and M are positive definite, it is natural to take = 0, and
get the smallest eigenvalues; in other cases take any point in the interval
[lb,ub] where eigenvalues are sought. Subtract M from the eigenvalue
equation and get (K - M)U = ( - )MU. Then multiply with the inverse of
this shifted matrix and get
1 1

= ( )

U K M MU.
This is a standard eigenvalue problem AU = U, with the matrix
A = (K M)
-1
M and eigenvalues
u

i
i
=

1
where i = 1, . . ., n. The largest eigenvalues
i
of the transformed matrix A
now correspond to the eigenvalues
i
= + 1/
i
of the original pencil (K,M)
closest to the shift .
The Arnoldi algorithm computes an orthonormal basis V where the shifted
and inverted operator A is represented by a Hessenberg matrix H,
AV
j
= V
j
H
j,j
+ E
j
.
(The subscripts mean that V
j
and E
j
have j columns and H
j,j
has j rows and
columns. When no subscripts are used we deal with vectors and matrices of
size n.)
Some of the eigenvalues of this Hessenberg matrix H
j,j
eventually give good
approximations to the eigenvalues of the original pencil (K,M) when the
basis grows in dimension j, and less and less of the eigenvector is hidden in
the residual matrix E
j
.
The basis V is built one column v
j
at a time. The first vector v
1
is chosen at
random, as n normally distributed random numbers. In step j, the first j
vectors are already computed and form the n j matrix V
j
. The next vector
v
j+1
is computed by first letting A operate on the newest vector v
j
, and then
making the result orthogonal to all the previous vectors.
5-20
Eigenvalue Equations
This is formulated as h v Av V h
j j j j j + +
=
1 1
, where the column vector h
j
consists of the Gram-Schmidt coefficients, and h
j+1,j
is the normalization factor
that gives v
j+1
unit length. Put the corresponding relations from previous
steps in front of this and get
AV V H v h e
j j j j j j j j
T
= +
+ + , , 1 1
where H
j,j
is a jj Hessenberg matrix with the vectors h
j
as columns. The
second term on the right-hand side has nonzeros only in the last column; the
earlier normalization factors show up in the subdiagonal of H
j,j
.
The eigensolution of the small Hessenberg matrix H gives approximations to
some of the eigenvalues and eigenvectors of the large matrix operator A
j,j
in
the following way. Compute eigenvalues
i
and eigenvectors s
i
of H
j,j
,
H s s i j
j j i i i ,
, ,..., . = = u 1
Then y
i
= V
j
s
i
is an approximate eigenvector of A, and its residual is
r Ay y AV s V s AV V H s v h s
i i i i j i j i i j j j j i j j j i j
= = = =
+ +
( )
, , , 1 1
This residual has to be small in norm for
i
to be a good eigenvalue
approximation. The norm of the residual is
r h s
i j j j i
=
+1, ,
,
the product of the last subdiagonal element of the Hessenberg matrix and the
last element of its eigenvector. It seldom happens that h
j+1,j
gets particularly
small, but after sufficiently many steps j there are always some eigenvectors
s
i
with small last elements. The long vector V
j+1
is of unit norm.
It is not necessary to actually compute the eigenvector approximation y
i
to get
the norm of the residual; we only need to examine the short vectors s
i
, and flag
those with tiny last components as converged. In a typical case n may be 2000,
while j seldom exceeds 50, so all computations that involve only matrices and
vectors of size j are much cheaper than those involving vectors of length n.
5-21
5 Finite Element Method
This eigenvalue computation and test for convergence is done every few
steps j, until all approximations to eigenvalues inside the interval [lb,ub]
are flagged as converged. When n is much larger than j, this is done very
often, for smaller n more seldom. When all eigenvalues inside the interval
have converged, or when j has reached a prescribed maximum, the converged
eigenvectors, or more appropriately Schur vectors, are computed and put
in the front of the basis V.
After this, the Arnoldi algorithm is restarted with a random vector, if all
approximations inside the interval are flagged as converged, or else with the
best unconverged approximate eigenvector y
i
. In each step j of this second
Arnoldi run, the vector is made orthogonal to all vectors in V including the
converged Schur vectors from the previous runs. This way, the algorithm
is applied to a projected matrix, and picks up a second copy of any double
eigenvalue there may be in the interval. If anything in the interval converges
during this second run, a third is attempted and so on, until no more
approximate eigenvalues
i
show up inside. Then the algorithm signals
convergence. If there are still unconverged approximate eigenvalues after a
prescribed maximum number of steps, the algorithm signals nonconvergence
and reports all solutions it has found.
This is a heuristic strategy that has worked well on both symmetric,
nonsymmetric, and even defective eigenvalue problems. There is a tiny
theoretical chance of missing an eigenvalue, if all the random starting vectors
happen to be orthogonal to its eigenvector. Normally, the algorithm restarts p
times, if the maximum multiplicity of an eigenvalue is p. At each restart a
new random starting direction is introduced.
The shifted and inverted matrix A = (K M)
1
M is needed only to operate
on a vector v
j
in the Arnoldi algorithm. This is done by computing an LU
factorization,
P(K M)Q = LU
using the sparse MATLAB command lu (P and Q are permutations that make
the triangular factors L and U sparse and the factorization numerically
stable). This factorization needs to be done only once, in the beginning, then x
= Av
j
is computed as,
x = QU
1
L
1
PMv
j
5-22
Eigenvalue Equations
with one sparse matrix vector multiplication, a permutation, sparse forward-
and back-substitutions, and a final renumbering.
5-23
5 Finite Element Method
Nonlinear Equations
The low-level Partial Differential Equation Toolbox functions are aimed at
solving linear equations. Since many interesting computational problems
are nonlinear, the software contains a nonlinear solver built on top of the
assempde function.
Note Before solving a nonlinear elliptic PDE, from the Solve menu in the
PDE app, select Parameters. Then, select the Use nonlinear solver check
box and click OK. At the command line, use pdenonlin instead of assempde.
The parabolic and hyperbolic functions automatically detect when they
need to use a nonlinear solver.
The basic idea is to use Gauss-Newton iterations to solve the nonlinear
equations. Say you are trying to solve the equation
r(u) = (c(u)u) + a(u)u - f(u) = 0.
In the FEM setting you solve the weak form of r(u) = 0. Set as usual
u x U
j j
( ) =

then, multiply the equation by an arbitrary test function
i
, integrate on the
domain , and use Greens formula and the boundary conditions to obtain
0 =
( )
= ( )
( )
+ ( )
|
\

V V
+
}

| | | | U c x U x x a x U x x dx
q x
j j j i
j
, ( ) ( ) , ( ) ( )
,
O
UU ds U
dx g x U ds
x x
f x U x x
j i
i i
j
( )
|
.
|
( ) ( )
c
c
}
} }
| |
| |
( ) ( )
, ( ) ( ) ,
O
O O
which has to hold for all indices i.
5-24
Nonlinear Equations
The residual vector (U) can be easily computed as
(U) = (K + M + Q)U (F + G)
where the matrices K, M, Q and the vectors F and G are produced by
assembling the problem
(c(U)u) + a(U)u = f(U).
Assume that you have a guess U
(n)
of the solution. If U
(n)
is close enough to
the exact solution, an improved approximation U
(n+1)
is obtained by solving
the linearized problem
c
( )
c

( )
=
( )
+

o
U
U
U U U
n
n n n
( )
( ) ( ) ( )
,
1
where is a positive number. (It is not necessary that (U) = 0 have a solution
even if (u) = 0 has.) In this case, the Gauss-Newton iteration tends to be the
minimizer of the residual, i.e., the solution of min
U
( ) U .
It is well known that for sufficiently small
U U
n n ( ) ( ) +
( )
<
( )
1
and
p
U
U
U
n
n
n
=
c
( )
c
|
\

|
.
|
( )

( )
( )
1
is called a descent direction for ( ) U , where is the L
2
-norm. The iteration
is
U
(n+1)
= U
(n)
+ p
n
,
where 1 is chosen as large as possible such that the step has a reasonable
descent.
5-25
5 Finite Element Method
The Gauss-Newton method is local, and convergence is assured only when U
(0)
is close enough to the solution. In general, the first guess may be outside the
region of convergence. To improve convergence from bad initial guesses, a
damping strategy is implemented for choosing , the Armijo-Goldstein line
search. It chooses the largest damping coefficient out of the sequence 1, 1/2,
1/4, . . . such that the following inequality holds:
o
o
U U p U
n n
n
n ( ) ( ) ( )
( )

( )
+
( )
>
2
which guarantees a reduction of the residual norm by at least 1 /2. Each
step of the line-search algorithm requires an evaluation of the residual (U
(n)
+ p
n
).
An important point of this strategy is that when U
(n)
approaches the solution,
then 1 and thus the convergence rate increases. If there is a solution to
(U) = 0, the scheme ultimately recovers the quadratic convergence rate of the
standard Newton iteration.
Closely related to the preceding problem is the choice of the initial guess
U
(0)
. By default, the solver sets U
(0)
and then assembles the FEM matrices K
and F and computes
U
(1)
= K
1
F
The damped Gauss-Newton iteration is then started with U
(1)
, which should
be a better guess than U
(0)
. If the boundary conditions do not depend on the
solution u, then U
(1)
satisfies them even if U
(0)
does not. Furthermore, if the
equation is linear, then U
(1)
is the exact FEM solution and the solver does not
enter the Gauss-Newton loop.
There are situations where U
(0)
= 0 makes no sense or convergence is
impossible.
In some situations you may already have a good approximation and the
nonlinear solver can be started with it, avoiding the slow convergence regime.
This idea is used in the adaptive mesh generator. It computes a solution

U on
a mesh, evaluates the error, and may refine certain triangles. The interpolant
of

U is a very good starting guess for the solution on the refined mesh.
5-26
Nonlinear Equations
In general the exact Jacobian
J
U
U
n
n
=
( )
c
c

( )
is not available. Approximation of J
n
by finite differences in the following way
is expensive but feasible. The ith column of J
n
can be approximated by
c|
c
U U
n
i
n ( ) ( )
+ ( )
( )
which implies the assembling of the FEM matrices for the triangles containing
grid point i. A very simple approximation to J
n
, which gives a fixed point
iteration, is also possible as follows. Essentially, for a given U
(n)
, compute
the FEM matrices K and F and set
U
(n+1)
= K
1
F .
This is equivalent to approximating the Jacobian with the stiffness matrix.
Indeed, since (U
(n)
) = KU
(n)
F, putting J
n
= K yields
U U J U U K KU F K F
n n
n
n n n ( ) ( ) ( ) ( ) ( )
.
+
=
( )
=
( )
=
1 1 1 1

In many cases the convergence rate is slow, but the cost of each iteration
is cheap.
The Partial Differential Equation Toolbox nonlinear solver also provides for
a compromise between the two extremes. To compute the derivative of the
mapping UKU, proceed as follows. The a term has been omitted for clarity,
but appears again in the final result.
5-27
5 Finite Element Method
c
c
V V

( )
V
( )
= +
( )
+
( )
|
\

KU
c U U
U
dx
c U
i
j l i l l j
l
j
lim
,
c c
c| | | co
|
0
1
O
ll i l
j i j l i l
l
dxU
c U dx
c
u
dxU
V
|
.
|
=
( )
V V +
c
c
V V
}
} }

|
| | | | |
O
O O
.
The first integral term is nothing more than K
i,j
.
The second term is lumped, i.e., replaced by a diagonal matrix that contains
the row sums. Since
j

j
= 1, the second term is approximated by
o | |
i j l i l
l
c
u
dxU
,
c
c
V V
}

O
which is the ith component of K
(c)
U, where K
(c)
is the stiffness matrix
associated with the coefficient c/u rather than c. The same reasoning can
be applied to the derivative of the mapping UMU. The derivative of the
mapping U F is exactly

c
c
}
f
u
dx
i j
| |
O
which is the mass matrix associated with the coefficient f/u. Thus the
Jacobian of the residual (U) is approximated by
J K M K M U
c a f c a
= + + +
( ) ( )
' ' ' ( ) ( ) ( ) ( )
diag
where the differentiation is with respect to u, K and M designate stiffness and
mass matrices, and their indices designate the coefficients with respect to
which they are assembled. At each Gauss-Newton iteration, the nonlinear
solver assembles the matrices corresponding to the equations
5-28
Nonlinear Equations
+ =
+ =
( ) ( )
( )
c u a f u
c u a u
'
' '
0
0
and then produces the approximate Jacobian. The differentiations of the
coefficients are done numerically.
In the general setting of elliptic systems, the boundary conditions are
appended to the stiffness matrix to form the full linear system:

KU
K H
H
U F
R
F =
'

(
=

(
=
0
,
where the coefficients of

K and

F may depend on the solution

U . The
lumped approach approximates the derivative mapping of the residual by
J H
H
'

(
0
The nonlinearities of the boundary conditions and the dependencies of the
coefficients on the derivatives of

U are not properly linearized by this
scheme. When such nonlinearities are strong, the scheme reduces to the
fix-point iteration and may converge slowly or not at all. When the boundary
conditions are linear, they do not affect the convergence properties of the
iteration schemes. In the Neumann case they are invisible (H is an empty
matrix) and in the Dirichlet case they merely state that the residual is zero on
the corresponding boundary points.
5-29
5 Finite Element Method
References
[1] Bank, Randolph E., PLTMG: A Software Package for Solving Elliptic
Partial Differential Equations, Users Guide 6.0, Society for Industrial and
Applied Mathematics, Philadelphia, PA, 1990.
[2] Dahlquist, Germund, and Bjrk, ke, Numerical Methods, 2nd edition,
1995, in print.
[3] Golub, Gene H., and Charles F. Van Loan, Matrix Computations, 2nd
edition, John Hopkins University Press, Baltimore, MD, 1989.
[4] George, P.L., Automatic Mesh Generation Application to Finite Element
Methods, Wiley, 1991.
[5] Johnson, C., Numerical Solution of Partial Differential Equations by the
Finite Element Method, Studentlitteratur, Lund, Sweden, 1987.
[6] Johnson, C., and Eriksson, K., Adaptive Finite Element Methods for
Parabolic Problems I: A Linear Model Problem, SIAM J. Numer. Anal, 28,
(1991), pp. 4377.
[7] Saad, Yousef, Variations on Arnoldis Method for Computing
Eigenelements of Large Unsymmetric Matrices, Linear Algebra and its
Applications, Vol 34, 1980, pp. 269295.
[8] Rosenberg, I.G., and F. Stenger, A lower bound on the angles of triangles
constructed by bisecting the longest side, Math. Comp. 29 (1975), pp 390395.
[9] Strang, Gilbert, Introduction to Applied Mathematics,
Wellesley-Cambridge Press, Cambridge, MA, 1986.
[10] Strang, Gilbert, and Fix, George, An Analysis of the Finite Element
Method, Prentice-Hall Englewoood Cliffs, N.J., USA, 1973.
5-30
6
Functions Alphabetical
List
adaptmesh
Purpose Adaptive mesh generation and PDE solution
Syntax [u,p,e,t]=adaptmesh(g,b,c,a,f)
[u,p,e,t]=adaptmesh(g,b,c,a,f,'PropertyName',PropertyValue,)
Description [u,p,e,t]=adaptmesh(g,b,c,a,f)
[u,p,e,t]=adaptmesh(g,b,c,a,f,'PropertyName',PropertyValue,)
performs adaptive mesh generation and PDE solution. Optional
arguments are given as property name/property value pairs.
The function produces a solution u to the elliptic scalar PDE problem
V V
( )
+ = c u au f ,
for (x,y) , or the elliptic system PDE problem
V V
( )
+ = c u au f,
with the problem geometry and boundary conditions given by g and b.
The mesh is described by the p, e, and t.
The solution u is represented as the solution vector u. For details on the
representation of the solution vector, see assempde.
The algorithm works by solving a sequence of PDE problems using
refined triangular meshes. The first triangular mesh generation is
obtained either as an optional argument to adaptmesh or by a call to
initmesh without options. The following generations of triangular
meshes are obtained by solving the PDE problem, computing an error
estimate, selecting a set of triangles based on the error estimate, and
then finally refining these triangles. The solution to the PDE problem
is then recomputed. The loop continues until no triangles are selected
by the triangle selection method, or until the maximum number
of triangles is attained, or until the maximum number of triangle
generations has been generated.
g describes the decomposed geometry of the PDE problem. g can be
a Decomposed Geometry matrix, the name of a Geometry file, or a
function handle to a Geometry file. The formats of the Decomposed
6-2
adaptmesh
Geometry matrix and Geometry file are described in the entries on
decsg and pdegeom, respectively.
b describes the boundary conditions of the PDE problem. b can be a
Boundary Condition matrix, the name of a Boundary file, or a function
handle to a Boundary file. The formats of the Boundary Condition
matrix and Boundary file are described in the entries on assemb and
pdebound, respectively.
The adapted triangular mesh of the PDE problem is given by the mesh
data p, e, and t. For details on the mesh data representation, see Mesh
Data on page 2-85.
The coefficients c, a, and f of the PDE problem can be given in a wide
variety of ways. In the context of adaptmesh the coefficients can depend
on u if the nonlinear solver is enabled using the property nonlin. The
coefficients cannot depend on t, the time. For a complete listing of all
options, see Scalar PDE Coefficients on page 2-14 and Coefficients for
Systems of PDEs on page 2-42.
The following table lists the property name/value value pairs, their
default values, and descriptions of the properties.
Property Value Default Description
Maxt positive integer inf Maximum number of new
triangles
Ngen positive integer 10 Maximum number of
triangle generations
Mesh p1, e1, t1 initmesh Initial mesh
Tripick MATLAB function pdeadworst Triangle selection method
Par numeric 0.5 Function parameter
Rmethod 'longest' |
'regular'
'longest' Triangle refinement method
Nonlin 'on' | 'off' 'off' Use nonlinear solver
Toln numeric 1e-4 Nonlinear tolerance
6-3
adaptmesh
Property Value Default Description
Init u0 0 Nonlinear initial value
Jac 'fixed | 'lumped'
| 'full'
'fixed' Nonlinear Jacobian
calculation
norm numeric | inf inf Nonlinear residual norm
MesherVersion 'R2013a' |
'preR2013a'
'preR2013a' Algorithm for generating
initial mesh
Par is passed to the Tripick function, which is described later.
Normally it is used as tolerance of how well the solution fits the
equation.
No more than Ngen successive refinements are attempted. Refinement
is also stopped when the number of triangles in the mesh exceeds Maxt.
p1, e1, and t1 are the input mesh data. This triangular mesh is used as
starting mesh for the adaptive algorithm. For details on the mesh data
representation, see initmesh. If no initial mesh is provided, the result
of a call to initmesh with no options is used as the initial mesh.
The triangle selection method, Tripick, is a user-definable triangle
selection method. Given the error estimate computed by the function
pdejmps, the triangle selection method selects the triangles to be
refined in the next triangle generation. The function is called using
the arguments p, t, cc, aa, ff, u, errf, and par. p and t represent
the current generation of triangles, cc, aa, and ff are the current
coefficients for the PDE problem, expanded to triangle midpoints, u is
the current solution, errf is the computed error estimate, and par, the
function parameter, given to adaptmesh as optional argument. The
matrices cc, aa, ff, and errf all have Nt columns, where Nt is the
current number of triangles. The number of rows in cc, aa, and ff are
exactly the same as the input arguments c, a, and f. errf has one
row for each equation in the system. There are two standard triangle
selection methodspdeadworst and pdeadgsc. pdeadworst selects
triangles where errf exceeds a fraction (default: 0.5) of the worst value,
and pdeadgsc selects triangles using a relative tolerance criterion.
6-4
adaptmesh
The refinement method is either longest or regular. For details on the
refinement method, see refinemesh.
The MesherVersion property chooses the algorithm for mesh
generation. The 'R2013a' algorithm runs faster, and can triangulate
more geometries than the 'preR2013a' algorithm. Both algorithms
use Delaunay triangulation.
The adaptive algorithm can also solve nonlinear PDE problems. For
nonlinear PDE problems, the Nonlin parameter must be set to on.
The nonlinear tolerance Toln, nonlinear initial value u0, nonlinear
Jacobian calculation Jac, and nonlinear residual norm Norm are passed
to the nonlinear solver pdenonlin. For details on the nonlinear solver,
see pdenonlin.
Examples Solve the Laplace equation over a circle sector, with Dirichlet boundary
conditions u = cos(2/3atan2(y,x)) along the arc, and u = 0 along the
straight lines, and compare to the exact solution. We refine the
triangles using the worst error criterion until we obtain a mesh with at
least 500 triangles:
[u,p,e,t]=adaptmesh('cirsg','cirsb',1,0,0,'maxt',500,...
'tripick','pdeadworst','ngen',inf);
x=p(1,:); y=p(2,:);
exact=((x.^2+y.^2).^(1/3).*cos(2/3*atan2(y,x)))';
max(abs(u-exact))
ans =
0.0028
size(t,2)
ans =
629
pdemesh(p,e,t)
6-5
adaptmesh
The maximum absolute error is 0.0028, with 629 triangles.
We test how many refinements we have to use with a uniform triangle
net:
[p,e,t]=initmesh('cirsg');
[p,e,t]=refinemesh('cirsg',p,e,t);
u=assempde('cirsb',p,e,t,1,0,0);
x=p(1,:); y=p(2,:);
exact=((x.^2+y.^2).^(1/3).*cos(2/3*atan2(y,x)))';
max(abs(u-exact))
ans =
0.0121
size(t,2)
6-6
adaptmesh
ans =
788
[p,e,t]=refinemesh('cirsg',p,e,t);
u=assempde('cirsb',p,e,t,1,0,0);
x=p(1,:); y=p(2,:);
exact=((x.^2+y.^2).^(1/3).*cos(2/3*atan2(y,x)))';
max(abs(u-exact))
ans =
0.0078
size(t,2)
ans =
3152
pdemesh(p,e,t)
6-7
adaptmesh
Uniform refinement with 3152 triangles produces an error of 0.0078.
This error is over three times as large as that produced by the adaptive
method (0.0028) with many fewer triangles (629). For a problem with
regular solution, we expect a O(h
2
) error, but this solution is singular
since u r
1 3 /
at the origin.
Diagnostics Upon termination, one of the following messages is displayed:
Adaption completed (This means that the Tripick function
returned zero triangles to refine.)
Maximum number of triangles obtained
Maximum number of refinement passes obtained
See Also assempde | initmesh | pdeadgsc | pdeadworst | pdejmps |
refinemesh
6-8
assema
Purpose Assemble area integral contributions
Syntax [K,M,F]=assema(p,t,c,a,f)
[K,M,F]=assema(p,t,c,a,f,u0)
[K,M,F]=assema(p,t,c,a,f,u0,time)
[K,M,F]=assema(p,t,c,a,f,u0,time,sdl)
[K,M,F]=assema(p,t,c,a,f,time)
[K,M,F]=assema(p,t,c,a,f,time,sdl)
Description [K,M,F]=assema(p,t,c,a,f) assembles the stiffness matrix K, the
mass matrix M, and the right-hand side vector F.
The input parameters p, t, c, a, f, u0, time, and sdl have the same
meaning as in assempde.
See Also assempde
How To Scalar PDE Coefficients on page 2-14
Coefficients for Systems of PDEs on page 2-42
6-9
assemb
Purpose Assemble boundary condition contributions
Syntax [Q,G,H,R]=assemb(b,p,e)
[Q,G,H,R]=assemb(b,p,e,u0)
[Q,G,H,R]=assemb(b,p,e,u0,time)
[Q,G,H,R]=assemb(b,p,e,u0,time,sdl)
[Q,G,H,R]=assemb(b,p,e,time)
[Q,G,H,R]=assemb(b,p,e,time,sdl)
Description [Q,G,H,R]=assemb(b,p,e) assembles the matrices Q and H, and the
vectors G and R. Q should be added to the system matrix and contains
contributions from mixed boundary conditions. G should be added to
the right side and contains contributions from generalized Neumann
and mixed boundary conditions. The equation H*u=R represents the
Dirichlet type boundary conditions.
The input parameters p, e, u0, time, and sdl have the same meaning
as in assempde.
b describes the boundary conditions of the PDE problem. b can be a
Boundary Condition matrix, the name of a Boundary file, or a function
handle to a Boundary file. The format of the Boundary Condition
matrix is described later.
Partial Differential Equation Toolbox software treats the following
boundary condition types:
On a generalized Neumann boundary segment, q and g are related to
the normal derivative value by:
n c
( )
+ = Vu qu g
On a Dirichlet boundary segment, hu = r.
The software can also handle systems of partial differential equations
over the domain . Let the number of variables in the system be N. The
general boundary condition is hu = r.
n c qu g h u .
( )
+ = + ' V
6-10
assemb
The notation n c u
( )
V indicates that the N by 1 matrix with
(i,1)-component
cos( ) cos( ) sin( ) sin(
, , , , , , , , ,
o o o c
x
c
y
c
x
i j i j i j 1 1 1 2 2 1

+ oo)
, , ,
c
y
u
i j
j
N
2 2
1

where is the angle of the normal vector of the boundary, pointing in

the direction out from , the domain.
The Boundary Condition matrix is created internally in the PDE app
(actually a function called by the PDE app) and then used from the
function assemb for assembling the contributions from the boundary to
the matrices Q, G, H, and R. The Boundary Condition matrix can also
be saved onto a file as a boundary file for later use with the wbound
function.
For each column in the Decomposed Geometry matrix there must be a
corresponding column in the Boundary Condition matrix. The format of
each column is according to the following list:
Row one contains the dimension N of the system.
Row two contains the number M of Dirichlet boundary conditions.
Row three to 3 + N
2
1 contain the lengths for the strings
representing q. The lengths are stored in column-wise order with
respect to q.
Row 3 + N
2
to 3 + N
2
+N 1 contain the lengths for the strings
representing g.
Row 3 + N
2
+ N to 3 + N
2
+ N + MN 1 contain the lengths for the
strings representing h. The lengths are stored in columnwise order
with respect to h.
Row 3 + N
2
+ N + MN to 3 + N
2
+ N + MN + M 1 contain the lengths
for the strings representing r.
The following rows contain text expressions representing the actual
boundary condition functions. The text strings have the lengths
6-11
assemb
according to above. The MATLAB text expressions are stored in
columnwise order with respect to matrices h and q. There are no
separation characters between the strings. You can insert MATLAB
expressions containing the following variables:
The 2-D coordinates x and y.
A boundary segment parameter s, proportional to arc length. s is 0
at the start of the boundary segment and increases to 1 along the
boundary segment in the direction indicated by the arrow.
The outward normal vector components nx and ny. If you need the
tangential vector, it can be expressed using nx and ny since t
x
= n
y
and t
y
= n
x
.
The solution u (only if the input argument u has been specified).
The time t (only if the input argument time has been specified).
It is not possible to explicitly refer to the time derivative of the solution
in the boundary conditions.
Examples Example 1
The following examples describe the format of the boundary condition
matrix for one column of the Decomposed Geometry matrix. For a
boundary in a scalar PDE (N = 1) with Neumann boundary condition
(M = 0)
n c x u V
( )
=
2
the boundary condition would be represented by the column vector
[1 0 1 5 '0' '-x.^2']'
No lengths are stored for h or r.
Also for a scalar PDE, the Dirichlet boundary condition
u = x
2
y
2
6-12
assemb
is stored in the column vector
[1 1 1 1 1 9 '0' '0' '1' 'x.^2-y.^2']'
For a system (N = 2) with mixed boundary conditions (M = 1):
h h r
q q
q q
g
g
11 12 1
11 12
21 22
1
2
( ) =

( )
+
|
\

|
.
|
=
|
\

|
.
|
+ V
u
n c u u s
the column appears similar to the following example:
2
1
lq11
lq21
lq12
lq22
lg1
lg2
lh11
lh12
lr1
q11 ...
q21 ...
q12 ...
q22 ...
g1 ...
g2 ...
h11 ...
h12 ...
r1 ...
Where lq11, lq21, . . . denote lengths of the MATLAB text expressions,
and q11, q21, . . . denote the actual expressions.
6-13
assemb
You can easily create your own examples by trying out the PDE
app. Enter boundary conditions by double-clicking on boundaries
in boundary mode, and then export the Boundary Condition matrix
to the MATLAB workspace by selecting the Export Decomposed
Geometry, Boundary Conds option from the Boundary menu.
Example 2
The following example shows you how to find the boundary condition
matrices for the Dirichlet boundary condition u x y =
2 2
on the
boundary of a circular disk.
1 Create the following function in your working folder:
function [x,y]=circ_geom(bs,s)
%CIRC_GEOM Creates a geometry file for a unit circle.
% Number of boundary segments
nbs=4;
if nargin==0 % Number of boundary segments
x=nbs;
elseif nargin==1 % Create 4 boundary segments
dl=[0 pi/2 pi 3*pi/2
pi/2 pi 3*pi/2 2*pi
1 1 1 1
0 0 0 0];
x=dl(:,bs);
else % Coordinates of edge segment points
z=exp(i*s);
x=real(z);
y=imag(z);
end
2 Create a second function in your working folder that finds the
boundary condition matrices, Q, G, H, and R:
6-14
assemb
function assemb_example
% Use ASSEMB to find the boundary condition matrices.
% Describe the geometry using four boundary segments
figure(1)
pdegplot('circ_geom')
axis equal
% Initialize the mesh
[p,e,t]=initmesh('circ_geom','Hmax',0.4);
figure(2)
% Plot the mesh
pdemesh(p,e,t)
axis equal
% Define the boundary condition vector, b,
% for the boundary condition u=x^2-y^2.
% For each boundary segment, the boundary
% condition vector is
b=[1 1 1 1 1 9 '0' '0' '1' 'x.^2-y.^2']';
% Create a boundary condition matrix that
% represents all of the boundary segments.
b = repmat(b,1,4);
% Use ASSEMB to find the boundary condition
% matrices. Since there are only Dirichlet
% boundary conditions, Q and G are empty.
[Q,G,H,R]=assemb(b,p,e)
3 Run the function assemb_example.m.
The function returns the four boundary condition matrices.
Q =
6-15
assemb
All zero sparse: 41-by-41
G =
All zero sparse: 41-by-1
H =
(1,1) 1
(2,2) 1
(3,3) 1
(4,4) 1
(5,5) 1
(6,6) 1
(7,7) 1
(8,8) 1
(9,9) 1
(10,10) 1
(11,11) 1
(12,12) 1
(13,13) 1
(14,14) 1
(15,15) 1
(16,16) 1
R =
(1,1) 1.0000
(2,1) -1.0000
(3,1) 1.0000
(4,1) -1.0000
(5,1) 0.0000
(6,1) -0.0000
(7,1) 0.0000
(8,1) -0.0000
(9,1) 0.7071
(10,1) -0.7071
6-16
assemb
(11,1) -0.7071
(12,1) 0.7071
(13,1) 0.7071
(14,1) -0.7071
(15,1) -0.7071
(16,1) 0.7071
Q and G are all zero sparse matrices because the problem has only
Dirichlet boundary conditions and neither generalized Neumann nor
mixed boundary conditions apply.
See Also assempde | pdebound
How To Boundary Conditions for Scalar PDE on page 2-72
Boundary Conditions for PDE Systems on page 2-77
6-17
assempde
Purpose Assemble stiffness matrix and right side of PDE problem
Syntax u=assempde(b,p,e,t,c,a,f)
u=assempde(b,p,e,t,c,a,f,u0)
u=assempde(b,p,e,t,c,a,f,u0,time)
u=assempde(b,p,e,t,c,a,f,time)
[K,F]=assempde(b,p,e,t,c,a,f)
[K,F]=assempde(b,p,e,t,c,a,f,u0)
[K,F]=assempde(b,p,e,t,c,a,f,u0,time)
[K,F]=assempde(b,p,e,t,c,a,f,u0,time,sdl)
[K,F]=assempde(b,p,e,t,c,a,f,time)
[K,F]=assempde(b,p,e,t,c,a,f,time,sdl)
[K,F,B,ud]=assempde(b,p,e,t,c,a,f)
[K,F,B,ud]=assempde(b,p,e,t,c,a,f,u0)
[K,F,B,ud]=assempde(b,p,e,t,c,a,f,u0,time)
[K,F,B,ud]=assempde(b,p,e,t,c,a,f,time)
[K,M,F,Q,G,H,R]=assempde(b,p,e,t,c,a,f)
[K,M,F,Q,G,H,R]=assempde(b,p,e,t,c,a,f,u0)
[K,M,F,Q,G,H,R]=assempde(b,p,e,t,c,a,f,u0,time)
[K,M,F,Q,G,H,R]=assempde(b,p,e,t,c,a,f,u0,time,sdl)
[K,M,F,Q,G,H,R]=assempde(b,p,e,t,c,a,f,time)
[K,M,F,Q,G,H,R]=assempde(b,p,e,t,c,a,f,time,sdl)
u=assempde(K,M,F,Q,G,H,R)
[K1,F1]=assempde(K,M,F,Q,G,H,R)
[K1,F1,B,ud]=assempde(K,M,F,Q,G,H,R)
Description assempde is the basic Partial Differential Equation Toolbox function. It
assembles a PDE problem by using the FEM formulation described in
Elliptic Equations on page 5-2. The command assempde assembles
the scalar PDE problem
V V
( )
+ = c u au f ,
for (x,y) , or the system PDE problem
V V
( )
+ = c u au f.
6-18
assempde
The command can optionally produce a solution to the PDE problem.
For the scalar case the solution vector u is represented as a column
vector of solution values at the corresponding node points from p. For
a system of dimension N with n
p
node points, the first n
p
values of u
describe the first component of u, the following n
p
values of u describe
the second component of u, and so on. Thus, the components of u are
placed in the vector u as N blocks of node point values.
u=assempde(b,p,e,t,c,a,f) assembles and solves the PDE problem
by eliminating the Dirichlet boundary conditions from the system of
linear equations.
[K,F]=assempde(b,p,e,t,c,a,f) assembles the PDE problem by
approximating the Dirichlet boundary condition with stiff springs (see
Systems of PDEs on page 5-10 for details). K and F are the stiffness
matrix and right-hand side, respectively. The solution to the FEM
formulation of the PDE problem is u=K\F.
[K,F,B,ud]=assempde(b,p,e,t,c,a,f) assembles the PDE problem
by eliminating the Dirichlet boundary conditions from the system of
linear equations. u1=K\F returns the solution on the non-Dirichlet
points. The solution to the full PDE problem can be obtained as the
MATLAB expression u=B*u1+ud.
[K,M,F,Q,G,H,R]=assempde(b,p,e,t,c,a,f) gives a split
representation of the PDE problem.
u=assempde(K,M,F,Q,G,H,R) collapses the split representation into
the single matrix/vector form, and then solves the PDE problem by
eliminating the Dirichlet boundary conditions from the system of linear
equations.
[K1,F1]=assempde(K,M,F,Q,G,H,R) collapses the split representation
into the single matrix/vector form, by fixing the Dirichlet boundary
condition with large spring constants.
[K1,F1,B,ud]=assempde(K,M,F,Q,G,H,R) collapses the split
representation into the single matrix/vector form by eliminating the
Dirichlet boundary conditions from the system of linear equations.
6-19
assempde
b describes the boundary conditions of the PDE problem. b can be a
Boundary Condition matrix, the name of a Boundary file, or a function
handle to a Boundary file. The formats of the Boundary Condition
matrix and Boundary file are described in the entries on assemb and
pdebound, respectively.
The geometry of the PDE problem is given by the mesh data p, e, and t.
For details on the mesh data representation, see initmesh.
The optional list of subdomain labels, sdl, restricts the assembly
process to the subdomains denoted by the labels in the list. The optional
input arguments u0 and time are used for the nonlinear solver and time
stepping algorithms, respectively. The tentative input solution vector
u0 has the same format as u.
Examples Example 1
Solve the equation u = 1 on the geometry defined by the L-shaped
membrane. Use Dirichlet boundary conditions u = 0 on . Finally
plot the solution.
[p,e,t]=initmesh('lshapeg','Hmax',0.2);
[p,e,t]=refinemesh('lshapeg',p,e,t);
u=assempde('lshapeb',p,e,t,1,0,1);
pdesurf(p,t,u)
Example 2
Consider Poissons equation on the unit circle with unit point source
at the origin. The exact solution
u r =
1
2t
log( )
is known for this problem. We define the function
f=circlef(p,t,u,time) for computing the right-hand side.
circlef returns zero for all triangles except for the one located at the
origin; for that triangle it returns 1/a, where a is the triangle area.
pdedemo7 executes an adaptive solution for this problem.
6-20
assempde
1
0.5
0
0.5
1
1
0.5
0
0.5
1
0
0.2
0.4
0.6
0.8
1
See Also assema | assemb | initmesh | pdebound | refinemesh
How To Scalar PDE Coefficients on page 2-14
Coefficients for Systems of PDEs on page 2-42
Boundary Conditions for Scalar PDE on page 2-72
Boundary Conditions for PDE Systems on page 2-77
6-21
csgchk
Purpose Check validity of Geometry Description matrix
Syntax gstat=csgchk(gd,xlim,ylim)
gstat=csgchk(gd)
Description gstat=csgchk(gd,xlim,ylim) checks if the solid objects in the
Geometry Description matrix gd are valid, given optional real numbers
xlim and ylim as current length of the x- and y-axis, and using a special
format for polygons. For a polygon, the last vertex coordinate can be
equal to the first one, to indicate a closed polygon. If xlim and ylim are
specified and if the first and the last vertices are not equal, the polygon
is considered as closed if these vertices are within a certain closing
distance. These optional input arguments are meant to be used only
when calling csgchk from the PDE app.
gstat=csgchk(gd) is identical to the preceding call, except for using
the same format of gd that is used by decsg. This call is recommended
when using csgchk as a command-line function.
gstat is a row vector of integers that indicates the validity status of the
corresponding solid objects, i.e., columns, in gd.
For a circle solid, gstat=0 indicates that the circle has a positive radius,
1 indicates a nonpositive radius, and 2 indicates that the circle is not
unique.
For a polygon, gstat=0 indicates that the polygon is closed and does
not intersect itself, i.e., it has a well-defined, unique interior region.
1 indicates an open and non-self-intersecting polygon, 2 indicates
a closed and self-intersecting polygon, and 3 indicates an open and
self-intersecting polygon.
For a rectangle solid, gstat is identical to that of a polygon. This is so
because a rectangle is considered as a polygon by csgchk.
For an ellipse solid, gstat=0 indicates that the ellipse has positive
semiaxes, 1 indicates that at least one of the semiaxes is nonpositive,
and 2 indicates that the ellipse is not unique.
6-22
csgchk
If gstat consists of zero entries only, then gd is valid and can be used as
input argument by decsg.
See Also decsg
6-23
csgdel
Purpose Delete borders between minimal regions
Syntax [dl1,bt1]=csgdel(dl,bt,bl)
[dl1,bt1]=csgdel(dl,bt)
Description [dl1,bt1]=csgdel(dl,bt,bl) deletes the border segments in the
list bl. If the consistency of the Decomposed Geometry matrix is not
preserved by deleting the elements in the list bl, additional border
segments are deleted. Boundary segments cannot be deleted.
For an explanation of the concepts or border segments, boundary
segments, and minimal regions, see decsg.
dl and dl1 are Decomposed Geometry matrices. For a description of the
Decomposed Geometry matrix, see decsg. The format of the Boolean
tables bt and bt1 is also described in the entry on decsg.
[dl1,bt1]=csgdel(dl,bt) deletes all border segments.
See Also csgchk | decsg
6-24
decsg
Purpose Decompose Constructive Solid Geometry into minimal regions
Syntax dl=decsg(gd,sf,ns)
dl=decsg(gd)
[dl,bt]=decsg(gd)
[dl,bt]=decsg(gd,sf,ns)
[dl,bt,dl1,bt1,msb]=decsg(gd)
[dl,bt,dl1,bt1,msb]=decsg(gd,sf,ns)
Description This function analyzes the Constructive Solid Geometry model (CSG
model) that you draw. It analyzes the CSG model, constructs a set of
disjoint minimal regions, bounded by boundary segments and border
segments, and optionally evaluates a set formula in terms of the objects
in the CSG model. We often refer to the set of minimal regions as the
decomposed geometry. The decomposed geometry makes it possible for
other Partial Differential Equation Toolbox functions to understand
the geometry you specify. For plotting purposes a second set of minimal
regions with a connected boundary is constructed.
The PDE app uses decsg for many purposes. Each time a new solid
object is drawn or changed, the PDE app calls decsg to be able to
draw the solid objects and minimal regions correctly. The Delaunay
triangulation algorithm, initmesh, also uses the output of decsg to
generate an initial mesh.
dl=decsg(gd,sf,ns) decomposes the CSG model gd into the
decomposed geometry dl. The CSG model is represented by the
Geometry Description matrix, and the decomposed geometry is
represented by the Decomposed Geometry matrix. decsg returns the
minimal regions that evaluate to true for the set formula sf. The Name
Space matrix ns is a text matrix that relates the columns in gd to
variable names in sf.
dl=decsg(gd) returns all minimal regions. (The same as letting sf
correspond to the union of all objects in gd.)
[dl,bt]=decsg(gd) and [dl,bt]=decsg(gd,sf,ns) additionally
return a Boolean table that relates the original solid objects to the
6-25
decsg
minimal regions. A column in bt corresponds to the column with the
same index in gd. A row in bt corresponds to a minimal region index.
[dl,bt,dl1,bt1,msb]=decsg(gd) and
[dl,bt,dl1,bt1,msb]=decsg(gd,sf,ns) return a second set of
minimal regions dl1 with a corresponding Boolean table bt1. This
second set of minimal regions all have a connected boundary. These
minimal regions can be plotted by using MATLAB patch objects. The
second set of minimal regions have borders that may not have been
induced by the original solid objects. This occurs when two or more
groups of solid objects have nonintersecting boundaries.
The calling sequences additionally return a sequence msb of drawing
commands for each second minimal region. The first row contains
the number of edge segment that bounds the minimal region. The
additional rows contain the sequence of edge segments from the
Decomposed Geometry matrix that constitutes the bound. If the
index edge segment label is greater than the total number of edge
segments, it indicates that the total number of edge segments should be
subtracted from the contents to get the edge segment label number and
the drawing direction is opposite to the one given by the Decomposed
Geometry matrix.
Geometry Description Matrix
The Geometry Description matrix gd describes the CSG model that you
draw using the PDE app. The current Geometry Description matrix can
be made available to the MATLAB workspace by selecting the Export
Geometry Description, Set Formula, Labels option from the Draw
menu in the PDE app.
Each column in the Geometry Description matrix corresponds to an
object in the CSG model. Four types of solid objects are supported. The
object type is specified in row 1:
For the circle solid, row one contains 1, and the second and third
row contain the center x- and y-coordinates, respectively. Row four
contains the radius of the circle.
6-26
decsg
For a polygon solid, row one contains 2, and the second row contains
the number, n, of line segments in the boundary of the polygon. The
following n rows contain the x-coordinates of the starting points of
the edges, and the following n rows contain the y-coordinates of the
starting points of the edges.
For a rectangle solid, row one contains 3. The format is otherwise
identical to the polygon format.
For an ellipse solid, row one contains 4, the second and third row
contains the center x- and y-coordinates, respectively. Rows four
and five contain the semiaxes of the ellipse. The rotational angle (in
radians) of the ellipse is stored in row six.
Set Formula
sf contains a set formula expressed with the set of variables listed
in ns. The operators +, *, and - correspond to the set operations
union, intersection, and set difference, respectively. The precedence
of the operators + and * is the same. - has higher precedence. The
precedence can be controlled with parentheses.
Name Space Matrix
The Name Space matrix ns relates the columns in gd to variable
names in sf. Each column in ns contains a sequence of characters,
padded with spaces. Each such character column assigns a name to the
corresponding geometric object in gd. This way we can refer to a specific
object in gd in the set formula sf.
Decomposed Geometry Matrix
The Decomposed Geometry matrix dl contains a representation of the
decomposed geometry in terms of disjointed minimal regions that
have been constructed by the decsg algorithm. Each edge segment
of the minimal regions corresponds to a column in dl. We refer to
edge segments between minimal regions as border segments and outer
boundaries as boundary segments. In each such column rows two and
three contain the starting and ending x-coordinate, and rows four and
five the corresponding y-coordinate. Rows six and seven contain left
and right minimal region labels with respect to the direction induced
6-27
decsg
by the start and end points (counter clockwise direction on circle and
ellipse segments). There are three types of possible edge segments in a
minimal region:
For circle edge segments row one is 1. Rows eight and nine contain
the coordinates of the center of the circle. Row 10 contains the radius.
For line edge segments row one is 2.
For ellipse edge segments row one is 4. Rows eight and nine contain
the coordinates of the center of the ellipse. Rows 10 and 11 contain
the semiaxes of the ellipse, respectively. The rotational angle of the
ellipse is stored in row 12.
Examples The following command sequence starts the PDE app and draws a unit
circle and a unit square.
pdecirc(0,0,1)
pderect([0 1 0 1])
Insert the set formula C1-SQ1. Export the Geometry Description matrix,
set formula, and Name Space matrix to the MATLAB workspace by
selecting the Export Geometry Description option from the Draw
menu. Then type
[dl,bt]=decsg(gd,sf,ns);
dl =
2.0000 2.0000 1.0000 1.0000 1.0000
0 0 -1.0000 0.0000 0.0000
1.0000 0 0.0000 1.0000 -1.0000
0 1.0000 -0.0000 -1.0000 1.0000
0 0 -1.0000 0 -0.0000
0 0 1.0000 1.0000 1.0000
1.0000 1.0000 0 0 0
0 0 0 0 0
0 0 0 0 0
0 0 1.0000 1.0000 1.0000
bt =
6-28
decsg
1 0
There is one minimal region, with five edge segments, three circle edge
segments, and two line edge segments.
Algorithms The algorithm consists of the following steps:
1 Determine the intersection points between the borders of the model
objects.
2 For each intersection point, sort the incoming edge segments on
angle and curvature.
3 Determine if the induced graph is connected. If not, add some
appropriate edges, and redo algorithm from step 1.
4 Cycle through edge segments of minimal regions.
5 For each original region, determine minimal regions inside it.
6 Organize output and remove the additional edges.
Note The input CSG model is not checked for correctness. It is assumed
that no circles or ellipses are identical or degenerated and that no
lines have zero length. Polygons must not be self-intersecting. Use the
function csgchk to check the CSG model.
Diagnostics NaN is returned if the set formula sf cannot be evaluated.
See Also csgchk | csgdel | pdebound | pdecirc | pdeellip | pdegeom |
pdepoly | pderect | pdetool | wbound | wgeom
6-29
dst
Purpose Discrete sine transform
Syntax y=dst(x)
y=dst(x,n)
x=idst(y)
x=idst(y,n)
Description The dst function implements the following equation:
y k x n
kn
N
k N
n
N
( ) ( ) sin , ,..., . =
+
|
\

|
.
|
=
=

t
1
1
1
y=dst(x) computes the discrete sine transform of the columns of x.
For best performance speed, the number of rows in x should be 2
m
1,
for some integer m.
y=dst(x,n) pads or truncates the vector x to length n before
transforming.
If x is a matrix, the dst operation is applied to each column.
The idst function implements the following equation:
y k
N
x n
kn
N
k N
n
N
( ) ( ) sin , ,..., . =
+ +
|
\

|
.
|
=
=

2
1 1
1
1
t
x=idst(y) calculates the inverse discrete sine transform of the columns
of y. For best performance speed, the number of rows in y should be
2
m
1, for some integer m.
x=idst(y,n) pads or truncates the vector y to length n before
transforming.
If y is a matrix, the idst operation is applied to each column.
For more information about this algorithm, see Solve Poissons
Equation on a Grid on page 3-119.
6-30
idst
See Also poiasma | poiindex | poisolv
6-31
hyperbolic
Purpose Solve hyperbolic PDE problem
Syntax u1=hyperbolic(u0,ut0,tlist,b,p,e,t,c,a,f,d)
u1=hyperbolic(u0,ut0,tlist,b,p,e,t,c,a,f,d,rtol)
u1=hyperbolic(u0,ut0,tlist,b,p,e,t,c,a,f,d,rtol,atol)
u1=hyperbolic(u0,ut0,tlist,K,F,B,ud,M)
u1=hyperbolic(u0,ut0,tlist,K,F,B,ud,M,rtol)
u1=hyperbolic(u0,ut0,tlist,K,F,B,ud,M,rtol,atol)
u1=hyperbolic( ___ ,'Stats','off')
Description u1=hyperbolic(u0,ut0,tlist,b,p,e,t,c,a,f,d) produces the
solution to the FEM formulation of the scalar PDE problem
d
u
t
c u au f
c
c
V V
( )
+ =
2
2
,
for (x,y) , or the system PDE problem
d
u
c u au f
c
c
V V
( )
+ =
2
2
t
,
on a mesh described by p, e, and t, with boundary conditions given by b,
and with initial value u0 and initial derivative ut0.
In the scalar case, each row in the solution matrix u1 is the solution at
the coordinates given by the corresponding column in p. Each column in
u1 is the solution at the time given by the corresponding item in tlist.
For a system of dimension N with n
p
node points, the first n
p
rows of u1
describe the first component of u, the following n
p
rows of u1 describe
the second component of u, and so on. Thus, the components of u are
placed in blocks u as N blocks of node point rows.
b describes the boundary conditions of the PDE problem. b can be a
Boundary Condition matrix, the name of a Boundary file, or a function
handle to a Boundary file. The boundary conditions can depend on t,
the time. The formats of the Boundary Condition matrix and Boundary
file are described in the entries on assemb and pdebound, respectively.
6-32
hyperbolic
The geometry of the PDE problem is given by the mesh data p, e, and t.
For details on the mesh data representation, see initmesh.
The coefficients c, a, d, and f of the PDE problem can be given in a
variety of ways. The coefficients can depend on t, the time. They can
also depend on u, the solution, and on the components of the gradient of
u, namely ux and uy. For a complete listing of all options, see Scalar
PDE Coefficients on page 2-14 and Coefficients for Systems of PDEs
on page 2-42.
atol and rtol are absolute and relative tolerances that are passed
to the ODE solver.
u1=hyperbolic(u0,ut0,tlist,K,F,B,ud,M) produces the solution
to the ODE problem
' + =
= +
B MB
d u
dt
K F
u Bu u
u
i
i
i d
2
2

with initial values for u being u0 and ut0.
Add the Stats name-value pair at the end of any syntax to control the
display of internal ODE solver statistics. Valid values for Stats are
'off' and 'on' (default).
Examples Solve the wave equation
c
c
=
2
2
u
t
u A
on a square geometry -1 x,y 1 (squareg), with Dirichlet boundary
conditions u = 0 for x = 1, and Neumann boundary conditions
c
c
=
u
n
0
for y = 1 (squareb3). Choose
6-33
hyperbolic
u(0) = atan(cos(x))
and
du
dt
x y ( ) sin( ) exp cos( ) . 0 3 = ( ) t t
Compute the solution at times 0, 1/6, 1/3, ... , 29/6, 5.
[p,e,t]=initmesh('squareg');
x=p(1,:)';
y=p(2,:)';
u0=atan(cos(pi/2*x));
ut0=3*sin(pi*x).*exp(cos(pi*y));
tlist=linspace(0,5,31);
uu=hyperbolic(u0,ut0,tlist,'squareb3',p,e,t,1,0,0,1);
The file pdedemo6 contains a complete example with animation.
Note In expressions for boundary conditions and PDE coefficients,
the symbol t is used to denote time. The variable t is often used to
store the triangle matrix of the mesh. You can use any variable to store
the triangle matrix, but in the Partial Differential Equation Toolbox
expressions, t always denotes time.
See Also assempde | parabolic
How To Scalar PDE Coefficients on page 2-14
Coefficients for Systems of PDEs on page 2-42
Boundary Conditions for Scalar PDE on page 2-72
Boundary Conditions for PDE Systems on page 2-77
6-34
initmesh
Purpose Create initial triangular mesh
Syntax [p,e,t]=initmesh(g)
[p,e,t]=initmesh(g,'PropertyName',PropertyValue,...)
Description [p,e,t]=initmesh(g) returns a triangular mesh using the geometry
specification function g. It uses a Delaunay triangulation algorithm.
The mesh size is determined from the shape of the geometry.
g describes the geometry of the PDE problem. g can be a Decomposed
Geometry matrix, the name of a Geometry file, or a function handle
to a Geometry file. The formats of the Decomposed Geometry matrix
and Geometry file are described in the entries on decsg and pdegeom,
respectively.
The outputs p, e, and t are the mesh data.
In the Point matrix p, the first and second rows contain x- and
y-coordinates of the points in the mesh.
In the Edge matrix e, the first and second rows contain indices of
the starting and ending point, the third and fourth rows contain the
starting and ending parameter values, the fifth row contains the edge
segment number, and the sixth and seventh row contain the left- and
right-hand side subdomain numbers.
In the Triangle matrix t, the first three rows contain indices to the
corner points, given in counter clockwise order, and the fourth row
contains the subdomain number.
The following property name/property value pairs are allowed.
Property Value Default Description
Hmax numeric estimate Maximum edge size
Hgrad numeric,
strictly
between 1
and 2
1.3 Mesh growth rate
6-35
initmesh
Property Value Default Description
Box 'on' |
'off'
'off' Preserve bounding box
Init 'on' |
'off'
'off' Edge triangulation
Jiggle 'off' |
'mean' |
'minimum'
| 'on'
'mean' Call jigglemesh after
creating the mesh, with
the Opt name-value
pair set to the stated
value. Exceptions:
'off' means do not call
jigglemesh, and 'on'
means call jigglemesh
with Opt = 'off'.
JiggleIter numeric 10 Maximum iterations
MesherVersion 'R2013a' |
'preR2013a'
'preR2013a' Algorithm for
generating initial mesh
The Hmax property controls the size of the triangles on the mesh.
initmesh creates a mesh where triangle edge lengths are approximately
Hmax or less.
The Hgrad property determines the mesh growth rate away from a
small part of the geometry. The default value is 1.3, i.e., a growth rate
of 30%. Hgrad cannot be equal to either of its bounds, 1 and 2.
Both the Box and Init property are related to the way the mesh
algorithm works. By turning on Box you can get a good idea of how
the mesh generation algorithm works within the bounding box. By
turning on Init you can see the initial triangulation of the boundaries.
By using the command sequence
[p,e,t]=initmesh(dl,'hmax',inf,'init','on');
[uxy,tn,a2,a3]=tri2grid(p,t,zeros(size(p,2)),x,y);
n=t(4,tn);
6-36
initmesh
you can determine the subdomain number n of the point xy. If the
point is outside the geometry, tn is NaN and the command n=t(4,tn)
results in a failure.
The Jiggle property is used to control whether jiggling of the mesh
should be attempted (see jigglemesh for details). Jiggling can be
done until the minimum or the mean of the quality of the triangles
decreases. JiggleIter can be used to set an upper limit on the number
of iterations.
The MesherVersion property chooses the algorithm for mesh
generation. The 'R2013a' algorithm runs faster, and can triangulate
more geometries than the 'preR2013a' algorithm. Both algorithms
use Delaunay triangulation.
Examples Make a simple triangular mesh of the L-shaped membrane in the PDE
app. Before you do anything in the PDE app, set the Maximum edge
size to inf in the Mesh Parameters dialog box. You open the dialog box
by selecting the Parameters option from the Mesh menu. Also select
the items Show Node Labels and Show Triangle Labels in the
Mesh menu. Then create the initial mesh by pressing the A button.
(This can also be done by selecting the Initialize Mesh option from
the Mesh menu.)
The following figure appears.
6-37
initmesh
1.5 1 0.5 0 0.5 1 1.5
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
1 2
3 4
5 6
1
2
3
4
The corresponding mesh data structures can be exported to the main
workspace by selecting the Export Mesh option from the Mesh menu.
p
p =
-1 1 1 0 0 -1
-1 -1 1 1 0 0
e
e =
1 2 3 4 5 6
2 3 4 5 6 1
0 0 0 0 0 0
1 1 1 1 1 1
1 2 3 4 5 6
1 1 1 1 1 1
0 0 0 0 0 0
6-38
initmesh
t
t =
1 2 3 1
2 3 4 5
5 5 5 6
1 1 1 1
References George, P. L., Automatic Mesh Generation Application to Finite
Element Methods, Wiley, 1991.
See Also decsg | jigglemesh | pdegeom | refinemesh
How To Mesh Data on page 2-85
6-39
jigglemesh
Purpose Jiggle internal points of triangular mesh
Syntax p1=jigglemesh(p,e,t)
p1=jigglemesh(p,e,t,'PropertyName',PropertyValue,...)
Description p1=jigglemesh(p,e,t) jiggles the triangular mesh by adjusting the
node point positions. The quality of the mesh normally increases.
The following property name/property value pairs are allowed.
Property Value Default Description
Opt 'off' |
'mean' |
'minimum'
'mean' Optimization
method, described
in the following
bullets
Iter numeric 1 or 20 (see the
following bullets)
Maximum
iterations
Each mesh point that is not located on an edge segment is moved
toward the center of mass of the polygon formed by the adjacent
triangles. This process is repeated according to the settings of the Opt
and Iter variables:
When Opt is set to 'off' this process is repeated Iter times (default:
1).
When Opt is set to 'mean' the process is repeated until the mean
triangle quality does not significantly increase, or until the bound
Iter is reached (default: 20).
When Opt is set to 'minimum' the process is repeated until the
minimum triangle quality does not significantly increase, or until the
bound Iter is reached (default: 20).
Examples Create a triangular mesh of the L-shaped membrane, first without
jiggling, and then jiggle the mesh.
[p,e,t]=initmesh('lshapeg','jiggle','off');
6-40
jigglemesh
q=pdetriq(p,t);
pdeplot(p,e,t,'xydata',q,'colorbar','on','xystyle','flat')
p1=jigglemesh(p,e,t,'opt','mean','iter',inf);
q=pdetriq(p1,t);
pdeplot(p1,e,t,'xydata',q,'colorbar','on','xystyle','flat')
See Also initmesh | pdetriq
How To Mesh Data on page 2-85
6-41
parabolic
Purpose Solve parabolic PDE problem
Syntax u1=parabolic(u0,tlist,b,p,e,t,c,a,f,d)
u1=parabolic(u0,tlist,b,p,e,t,c,a,f,d,rtol)
u1=parabolic(u0,tlist,b,p,e,t,c,a,f,d,rtol,atol)
u1=parabolic(u0,tlist,K,F,B,ud,M)
u1=parabolic(u0,tlist,K,F,B,ud,M,rtol)
u1=parabolic(u0,tlist,K,F,B,ud,M,rtol,atol)
u1=parabolic( ___ ,'Stats','off')
Description u1=parabolic(u0,tlist,g,b,p,e,t,c,a,f,d) produces the solution
to the FEM formulation of the scalar PDE problem
d c au f
u
t
u
c
c
V V
( )
+ = ,
for (x,y) , or the system PDE problem
d c au f
u
u
c
c
V V
( )
+ =
t
,
on a mesh described by p, e, and t, with boundary conditions given by b,
and with initial value u0.
For the scalar case, each row in the solution matrix u1 is the solution at
the coordinates given by the corresponding column in p. Each column in
u1 is the solution at the time given by the corresponding item in tlist.
For a system of dimension N with n
p
node points, the first n
p
rows of u1
describe the first component of u, the following n
p
rows of u1 describe
the second component of u, and so on. Thus, the components of u are
placed in the vector u as N blocks of node point rows.
b describes the boundary conditions of the PDE problem. b can be a
Boundary Condition matrix, the name of a Boundary file, or a function
handle to a Boundary file. The boundary conditions can depend on t,
the time. The formats of the Boundary Condition matrix and Boundary
file are described in the entries on assemb and pdebound, respectively.
6-42
parabolic
The geometry of the PDE problem is given by the mesh data p, e, and t.
For details on the mesh data representation, see initmesh.
The coefficients c, a, d, and f of the PDE problem can be given in a
variety of ways. The coefficients can depend on t, the time. They can
also depend on u, the solution, and on the components of the gradient of
u, namely ux and uy. For a complete listing of all options, see Scalar
PDE Coefficients on page 2-14 and Coefficients for Systems of PDEs
on page 2-42.
atol and rtol are absolute and relative tolerances that are passed
to the ODE solver.
u1=parabolic(u0,tlist,K,F,B,ud,M) produces the solution to the
ODE problem
' + =
= +
B MB
du
dt
K F
u Bu u
u
i
i
i d

with initial value for u being u0.

Add the Stats name-value pair at the end of any syntax to control the
display of internal ODE solver statistics. Valid values for Stats are
'off' and 'on' (default).
Examples Solve the heat equation
c
c
=
u
t
u A
on a square geometry 1 x,y 1 (squareg). Choose u(0) = 1 on the disk
x
2
+y
2
< 0.4
2
, and u(0) = 0 otherwise. Use Dirichlet boundary conditions
u = 0 (squareb1). Compute the solution at times linspace(0,0.1,20).
[p,e,t]=initmesh('squareg');
[p,e,t]=refinemesh('squareg',p,e,t);
u0=zeros(size(p,2),1);
ix=find(sqrt(p(1,:).^2+p(2,:).^2)<0.4);
6-43
parabolic
u0(ix)=ones(size(ix));
tlist=linspace(0,0.1,20);
u1=parabolic(u0,tlist,'squareb1',p,e,t,1,0,0,1);
Note In expressions for boundary conditions and PDE coefficients,
the symbol t is used to denote time. The variable t is often used to
store the triangle matrix of the mesh. You can use any variable to store
the triangle matrix, but in the Partial Differential Equation Toolbox
expressions, t always denotes time.
See Also assempde | hyperbolic
How To Scalar PDE Coefficients on page 2-14
Coefficients for Systems of PDEs on page 2-42
Boundary Conditions for Scalar PDE on page 2-72
Boundary Conditions for PDE Systems on page 2-77
6-44
pdeadgsc
Purpose Select triangles using relative tolerance criterion
Syntax bt=pdeadgsc(p,t,c,a,f,u,errf,tol)
Description bt=pdeadgsc(p,t,c,a,f,u,errf,tol) returns indices of triangles to
be refined in bt. Used from adaptmesh to select the triangles to be
further refined. The geometry of the PDE problem is given by the mesh
data p and t. For more details, see Mesh Data on page 2-85.
c,a, and f are PDE coefficients. For details, see Scalar PDE
Coefficients on page 2-14 and Coefficients for Systems of PDEs on
page 2-42.
u is the current solution, given as a column vector. For details, see
assempde.
errf is the error indicator, as calculated by pdejmps.
tol is a tolerance parameter.
Triangles are selected using the criterion errf>tol*scale, where
scale is calculated as follows:
Let cmax, amax, fmax, and umax be the maximum of c, a, f, and u,
respectively. Let l be the side of the smallest axis-aligned square that
contains the geometry.
Then scale=max(fmax*l^2,amax*umax*l^2,cmax*umax). The scaling
makes the tol parameter independent of the scaling of the equation
and the geometry.
See Also adaptmesh | pdejmps
6-45
pdeadworst
Purpose Select triangles relative to worst value
Syntax bt=pdeadworst(p,t,c,a,f,u,errf,wlevel)
Description bt=pdeadworst(p,t,c,a,f,u,errf,wlevel) returns indices of
triangles to be refined in bt. Used from adaptmesh to select the
triangles to be further refined.
The geometry of the PDE problem is given by the mesh data p and t.
For details, see Mesh Data on page 2-85.
c, a, and f are PDE coefficients. For details, see Scalar PDE
Coefficients on page 2-14.
u is the current solution, given as a column vector. For details, see
assempde.
errf is the error indicator, as calculated by pdejmps.
wlevel is the error level relative to the worst error. wlevel must be
between 0 and 1.
Triangles are selected using the criterion errf>wlevel*max(errf).
See Also adaptmesh | assempde | initmesh | pdejmps
6-46
pdearcl
Purpose Interpolation between parametric representation and arc length
Syntax pp=pdearcl(p,xy,s,s0,s1)
Description pp=pdearcl(p,xy,s,s0,s1) returns parameter values for a
parameterized curve corresponding to a given set of arc length values.
p is a monotone row vector of parameter values and xy is a matrix with
two rows giving the corresponding points on the curve.
The first point of the curve is given the arc length value s0 and the last
point the value s1.
On return, pp contains parameter values corresponding to the arc
length values specified in s.
The arc length values s, s0, and s1 can be an affine transformation of
the arc length.
Examples See the example cardg on the reference page for pdegeom.
See Also pdegeom
6-47
pdebound
Purpose Write custom function for defining boundary conditions
Syntax [qmatrix,gmatrix,hmatrix,rmatrix]=pdebound(p,e,u,time)
Description The Boundary file specifies the boundary conditions of a PDE problem.
The most general form of boundary conditions that we can handle is
hu r
n c qu g h u
=

( )
+ = + ' V .
By the notation n c u
( )
V we mean the N-by-1 matrix, where N is the
dimension of the system, with (i,1)-component
cos( ) cos( ) sin( ) sin(
, , , , , , , , ,
o o o c
x
c
y
c
x
i j i j i j 1 1 1 2 2 1
c
c
+
c
c
+
c
c
+ oo)
, , , ,
c
y
u
i j
j
N
j 2 2
1
c
c
|
\

|
.
|
=

where the outward normal vector of the boundary n = ( ) cos( ),sin( ) o o .

There are M Dirichlet conditions and the h-matrix is M-by-N, M 0.
The generalized Neumann condition contains a source ' h where the
Lagrange multipliers is computed such that the Dirichlet conditions
become satisfied.
The data that you specify is q, g, h, and r.
For M = 0 we say that we have a generalized Neumann boundary
condition, for M = N a Dirichlet boundary condition, and for 0 < M < N a
mixed boundary condition.
The Boundary file [qmatrix,gmatrix,hmatrix,rmatrix] =
pdebound(p,e,u,time) computes the values of q, g, h, and r, on the a
set of edges e.
The matrices p and e are mesh data. e needs only to be a subset of the
edges in the mesh. Details on the mesh data representation can be
found in the entry on initmesh.
6-48
pdebound
The input arguments u and time are used for the nonlinear solver and
time stepping algorithms, respectively. u and time are empty matrices
if the corresponding parameter is not passed to assemb. If time is NaN
and any of the function q, g, h, and r depends on time, pdebound must
return a matrix of correct size, containing NaNs in all positions, in the
corresponding output argument. It is not possible to explicitly refer to
the time derivative of the solution in the boundary conditions.
Similarly, if an output argument depends on u, then the output
argument must return a matrix of NaN of the correct size if the input
u is NaN. This requirement signals to the solver, such a parabolic or
hyperbolic, that the argument depends on time or the solution u.
The solution u is represented by the solution vector u. Details on the
representation can be found in the entry on assempde.
qmatrix and gmatrix must contain the value of q and g on the midpoint
of each boundary. Thus we have size(qmatrix)=[N^2 ne], where N
is the dimension of the system, and ne the number of edges in e, and
size(gmatrix)=[N ne]. For the Dirichlet case, the corresponding
values must be zeros.
hmatrix and rmatrix must contain the values of h and r at the first
point on each edge followed by the value at the second point on
each edge. Thus we have size(hmatrix)=[N^2 2*ne], where N is
the dimension of the system, and ne the number of edges in e, and
size(rmatrix)=[N 2*ne]. When M < N, h and r must be padded with
N M rows of zeros.
The elements of the matrices q and h are stored in column-wise ordering
in the MATLAB matrices qmatrix and hmatrix.
Examples For the boundary conditions
1 1 2
1 2
2 0
3
4
( ) =

( )
+
|
\

|
.
|
=
|
\

|
.
|
+ ' V
u
n c u h u
6-49
pdebound
the following values should be stored in qmatrix, gmatrix, hmatrix,
and rmatrix
1
qmatrix = [ ... 2 ... ]
2
0
gmatrix = [ ... 3 ... ]
4
1 1
hmatrix = [ ... 0 ... 0 ... ]
-1 -1
0 0
rmatrix = [ ... 2 ... 2 ... ]
0 0
See Also initmesh | pdeent | pdegeom | pdesdt
How To Boundary Conditions for Scalar PDE on page 2-72
Boundary Conditions for PDE Systems on page 2-77
6-50
pdecgrad
Purpose Flux of PDE solution
Syntax [cgxu,cgyu]=pdecgrad(p,t,c,u)
[cgxu,cgyu]=pdecgrad(p,t,c,u,time)
[cgxu,cgyu]=pdecgrad(p,t,c,u,time,sdl)
Description [cgxu,cgyu]=pdecgrad(p,t,c,u) returns the flux, c u , evaluated
at the center of each triangle.
Row i of cgxu contains
c
u
x
c
u
y
ij
j
j
N
ij
j
11
1
12
c
c
+
c
c
=

Row i of cgyu contains

c
u
x
c
u
y
ij
j
j
N
ij
j
21
1
22
c
c
+
c
c
=

There is one column for each triangle in t in both cgxu and cgyu.
The geometry of the PDE problem is given by the mesh data p and t.
Details on the mesh data representation can be found in the entry on
initmesh.
The coefficient c of the PDE problem can be given in a variety of
ways. A complete listing of all options can be found in the entry on
assempdeScalar PDE Coefficients on page 2-14 and c for Systems
on page 2-50.
The format for the solution vector u is described in assempde.
The scalar optional argument time is used for parabolic and hyperbolic
problems, if c depends on t, the time.
The optional argument sdl restricts the computation to the subdomains
in the list sdl.
6-51
pdecgrad
See Also assempde
6-52
pdecirc
Purpose Draw circle
Syntax pdecirc(xc,yc,radius)
pdecirc(xc,yc,radius,label)
Description pdecirc(xc,yc,radius) draws a circle with center in (xc,yc) and
radius radius. If the PDE app is not active, it is automatically started,
and the circle is drawn in an empty geometry model.
The optional argument label assigns a name to the circle (otherwise
a default name is chosen).
The state of the Geometry Description matrix inside the PDE app is
updated to include the circle. You can export the Geometry Description
matrix from the PDE app by using the Export Geometry Description
option from the Draw menu. For a details on the format of the
Geometry Description matrix, see decsg.
Examples The following command starts the PDE app and draws a unit circle.
pdecirc(0,0,1)
See Also pdeellip | pdepoly | pderect | pdetool
6-53
pdecont
Purpose Shorthand command for contour plot
Syntax pdecont(p,t,u)
pdecont(p,t,u,n)
pdecont(p,t,u,v)
h=pdecont(p,t,u)
h=pdecont(p,t,u,n)
h=pdecont(p,t,u,v)
Description pdecont(p,t,u) draws 10 level curves of the PDE node or triangle
data u. h = pdecont(p,t,u) additionally returns handles to the drawn
axes objects.
If u is a column vector, node data is assumed. If u is a row vector,
triangle data is assumed. Triangle data is converted to node data using
the function pdeprtni.
The geometry of the PDE problem is given by the mesh data p and t.
For details on the mesh data representation, see initmesh.
pdecont(p,t,u,n) plots using n levels.
pdecont(p,t,u,v) plots using the levels specified by v.
This command is just shorthand for the call
pdeplot(p,[],t,'xydata',u,'xystyle','off','contour',...
'on','levels',n,'colorbar','off');
If you want to have more control over your contour plot, use pdeplot
instead of pdecont.
Examples Plot the contours of the solution to the equation u = 1 over the
geometry defined by the L-shaped membrane. Use Dirichlet boundary
conditions u = 0 on .
[p,e,t]=initmesh('lshapeg');
[p,e,t]=refinemesh('lshapeg',p,e,t);
u=assempde('lshapeb',p,e,t,1,0,1);
pdecont(p,t,u)
6-54
pdecont
See Also pdemesh | pdeplot | pdesurf
6-55
pdeeig
Purpose Solve eigenvalue PDE problem
Syntax [v,l]=pdeeig(b,p,e,t,c,a,d,r)
[v,l]=pdeeig(K,B,M,r)
Description [v,l]=pdeeig(b,p,e,t,c,a,d,r) produces the solution to the FEM
formulation of the scalar PDE eigenvalue problem
(cu) + au = du on
or the system PDE eigenvalue problem
V V + = ( ) c a u u du on O
on a geometry described by p, e, and t, and with boundary conditions
given by b.
r is a two-element vector, indicating an interval on the real axis. (The
left-hand side can be -Inf.) The algorithm returns all eigenvalues in
this interval in l, up to a maximum of 99 eigenvalues.
v is an eigenvector matrix. For the scalar case each column in v is an
eigenvector of solution values at the corresponding node points from p.
For a system of dimension N with n
p
node points, the first n
p
rows of
v describe the first component of v, the following n
p
rows of v describe
the second component of v, and so on. Thus, the components of v are
placed in blocks v as N blocks of node point rows.
b describes the boundary conditions of the PDE problem. b can be a
Boundary Condition matrix, the name of a Boundary file, or a function
handle to a Boundary file. The formats of the Boundary Condition
matrix and Boundary file are described in the entries on assemb and
pdebound, respectively. The eigenvalue PDE problem is a homogeneous
problem, i.e., only boundary conditions where g = 0 and r = 0 can be
used. The nonhomogeneous part is removed automatically.
The geometry of the PDE problem is given by the mesh data p, e, and
t. For details on the mesh data representation, see Mesh Data on
page 2-85.
6-56
pdeeig
The coefficients c, a, d of the PDE problem can be given in a wide
variety of ways. In the context of pdeeig the coefficients cannot depend
on u nor t, the time. For a complete listing of all options, see Scalar
PDE Coefficients on page 2-14 and Coefficients for Systems of PDEs
on page 2-42.
[v,l]=pdeeig(K,B,M,r) produces the solution to the generalized
sparse matrix eigenvalue problem
K u
i
= BMBu
i
u = Bu
i
with Real() in the interval in r.
Examples Compute the eigenvalues less than 100 and corresponding eigenmodes
for
u = u,
on the geometry of the L-shaped membrane. Then display the first
and sixteenth eigenmodes.
[p,e,t]=initmesh('lshapeg');
[p,e,t]=refinemesh('lshapeg',p,e,t);
[p,e,t]=refinemesh('lshapeg',p,e,t);
[v,l]=pdeeig('lshapeb',p,e,t,1,0,1,[-Inf 100]);
l(1) % first eigenvalue
pdesurf(p,t,v(:,1)) % first eigenmode
figure
membrane(1,20,9,9) % the MATLAB function
figure
l(16) % sixteenth eigenvalue
pdesurf(p,t,v(:,16)) % sixteenth eigenmode
Algorithms pdeeig calls sptarn to calculate eigenvalues. For details of the
algorithm, see the sptarn reference pages.
6-57
pdeeig
Caution In the standard case c and d are positive in the entire region. All
eigenvalues are positive, and 0 is a good choice for a lower bound of the
interval. The cases where either c or d is zero are discussed next.
If d = 0 in a subregion, the mass matrix M becomes singular. This
does not cause any trouble, provided that c > 0 everywhere. The
pencil (K,M) has a set of infinite eigenvalues.
If c = 0 in a subregion, the stiffness matrix K becomes singular,
and the pencil (K,M) has many zero eigenvalues. With an interval
containing zero, pdeeig goes on for a very long time to find all the
zero eigenvalues. Choose a positive lower bound away from zero but
below the smallest nonzero eigenvalue.
If there is a region where both c = 0 and d = 0, we get a singular
pencil. The whole eigenvalue problem is undetermined, and any
value is equally plausible as an eigenvalue.
Some of the awkward cases are detected by pdeeig. If the shifted matrix
is singular, another shift is attempted. If the matrix with the new shift
is still singular a good guess is that the entire pencil (K,M) is singular.
If you try any problem not belonging to the standard case, you must use
your knowledge of the original physical problem to interpret the results
from the computation.
See Also sptarn
How To Scalar PDE Coefficients on page 2-14
Coefficients for Systems of PDEs on page 2-42
Boundary Conditions for Scalar PDE on page 2-72
Boundary Conditions for PDE Systems on page 2-77
6-58
pdeellip
Purpose Draw ellipse
Syntax pdeellip(xc,yc,a,b,phi)
pdeellip(xc,yc,a,b,phi,label)
Description pdeellip(xc,yc,a,b,phi) draws an ellipse with center in (xc,yc)
and semiaxes a and b. The rotation of the ellipse (in radians) is given by
phi. If the PDE app is not active, it is automatically started, and the
ellipse is drawn in an empty geometry model.
The optional argument label assigns a name to the ellipse (otherwise
a default name is chosen.)
The state of the Geometry Description matrix inside the PDE app is
updated to include the ellipse. You can export the Geometry Description
matrix from the PDE app by selecting the Export Geometry
Description option from the Draw menu. For a details on the format
of the Geometry Description matrix, see decsg.
Examples The following command starts the PDE app and draws an ellipse.
pdeellip(0,0,1,0.3,pi/4)
See Also pdecirc | pdepoly | pderect | pdetool
6-59
pdeent
Purpose Indices of triangles neighboring given set of triangles
Syntax ntl=pdeent(t,tl)
Description Given triangle data t and a list of triangle indices tl, ntl contains
indices of the triangles in tl and their immediate neighbors, i.e., those
whose intersection with tl is nonempty.
See Also refinemesh
6-60
pdegeom
Purpose Write custom function for defining geometry
Syntax ne=pdegeom
d=pdegeom(bs)
[x,y]=pdegeom(bs,s)
Description We represent 2-D regions by parameterized edge segments. Both the
regions and edge segments are assigned unique positive numbers
as labels. The edge segments cannot overlap. The full 2-D problem
description can contain several nonintersecting regions, and they can
have common border segments. The boundary of a region can consist of
several edge segments. All edge segment junctions must coincide with
edge segment endpoints. We sometimes refer to an edge segment as a
boundary segment or a border segment. A boundary segment is located
on the outer boundary of the union of the minimal regions, and a border
segment is located on the border between minimal regions.
There are two options for specifying the problem geometry:
Create a Decomposed Geometry matrix with the function decsg.
This is done automatically from the PDE app. Using the Decomposed
Geometry matrix restricts the edge segments to be straight lines,
circle, or ellipse segments. The Decomposed Geometry matrix can be
used instead of the Geometry file.
Create a Geometry file. By creating your own Geometry file, you can
create a geometry that follows any mathematical function exactly.
The following is an example of how to create a cardioid.
ne=pdegeom is the number of edge segments.
d=pdegeom(bs) is a matrix with one column for each edge segment
specified in bs.
Row 1 contains the start parameter value.
Row 2 contains the end parameter value.
Row 3 contains the label of the left-hand region (left with respect to
direction induced by start and end from row 1 and 2).
6-61
pdegeom
Row 4 contains the label of the right-hand region.
The complement of the union of all regions is assigned the region
number 0.
[x,y]=pdegeom(bs,s) produces coordinates of edge segment points.
bs specifies the edge segments and s the corresponding parameter
values. bs can be a scalar. The parameter s should be approximately
proportional to the curve length. All minimal regions should have at
least two, and preferably three, edge segments in their boundary.
Examples The function cardg defines the geometry of a cardioid
r = + 2 1 ( cos( )). u
function [x,y]=cardg(bs,s)
%CARDG Geometry File defining the geometry of a cardioid.
nbs=4;
if nargin==0
x=nbs;
return
end
dl=[ 0 pi/2 pi 3*pi/2
pi/2 pi 3*pi/2 2*pi
1 1 1 1
0 0 0 0];
if nargin==1
x=dl(:,bs);
return
end
x=zeros(size(s));
y=zeros(size(s));
[m,n]=size(bs);
if m==1 & n==1,
bs=bs*ones(size(s)); % expand bs
6-62
pdegeom
elseif m~=size(s,1) | n~=size(s,2),
error('bs must be scalar or of same size as s');
end
nth=400;
th=linspace(0,2*pi,nth);
r=2*(1+cos(th));
xt=r.*cos(th);
yt=r.*sin(th);
th=pdearcl(th,[xt;yt],s,0,2*pi);
r=2*(1+cos(th));
x(:)=r.*cos(th);
y(:)=r.*sin(th);
We use the function pdearcl to make the parameter s proportional to
arc length. You can test the function by typing
pdegplot('cardg'), axis equal
[p,e,t]=initmesh('cardg');
pdemesh(p,e,t), axis equal
Then solve the PDE problem u = 1 on the geometry defined by the
cardioid. Use Dirichlet boundary conditions u = 0 on . Finally plot
the solution.
u=assempde('cardb',p,e,t,1,0,1);
pdesurf(p,t,u);
Caution The parameter s should be approximately proportional to the curve
length. All minimal regions should have at least two, and preferably
three, edge segments in their boundary.
See Also initmesh | pdearcl | refinemesh
How To Boundary Conditions for Scalar PDE on page 2-72
Boundary Conditions for PDE Systems on page 2-77
6-63
pdegeom
Mesh Data on page 2-85
6-64
pdegplot
Purpose Plot PDE geometry
Syntax pdegplot(g)
h = pdeplot(g)
h = pdeplot(g,Name,Value)
Description pdegplot(g) plots the geometry of a PDE problem.
h = pdeplot(g) returns handles to the figure axes.
h = pdeplot(g,Name,Value) plots with additional options specified by
one or more Name,Value pair arguments.
Input
Arguments
g
Decomposed geometry matrix, as produced by decsg, or by selecting
Boundary > Export Decomposed Geometry, Boundary Conds
from the PDE app. g can also be the name of a geometry file or a
function handle to a geometry file (see wgeom for details).
Name-Value Pair Arguments
Specify optional comma-separated pairs of Name,Value arguments.
Name is the argument name and Value is the corresponding
value. Name must appear inside single quotes (' '). You can
specify several name and value pair arguments in any order as
Name1,Value1,...,NameN,ValueN.
edgeLabels
'on' shows the label for each boundary edge. These are the same as
the labels that the PDE app produces when you select Boundary >
Show Edge Labels.
Default: 'off'
subdomainLabels
6-65
pdegplot
'on' shows the label for each subdomain. These are the same as the
labels that the PDE app produces when you select PDE > Show
Subdomain Labels.
Default: 'off'
Output
Arguments
h
Vector of handles to the figure axes.
Examples Plot Geometry
Plot the geometry of a region defined by a few simple shapes.
g = [2 1 1 1 1 1 1 1 1 4 4;
-1 -0.55 -0.5 -0.45 -0.5 0.45 0.5 0.55 0.5 -1 0.169101978725763;
1 -0.5 -0.45 -0.5 -0.55 0.5 0.55 0.5 0.45 0.169101978725763 1;
0 -0.25 -0.3 -0.25 -0.2 -0.25 -0.3 -0.25 -0.2 0 -0.739198919740117;
0 -0.3 -0.25 -0.2 -0.25 -0.3 -0.25 -0.2 -0.25 -0.739198919740117 0;
0 0 0 0 0 0 0 0 0 1 1;
1 1 1 1 1 1 1 1 1 0 0;
0 -0.5 -0.5 -0.5 -0.5 0.5 0.5 0.5 0.5 0 0;
0 -0.25 -0.25 -0.25 -0.25 -0.25 -0.25 -0.25 -0.25 0 0;
0 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05 1 1;
0 0 0 0 0 0 0 0 0 0.75 0.75;
0 0 0 0 0 0 0 0 0 0 0];
pdegplot(g)
6-66
pdegplot
View the edge labels and the subdomain label.
pdegplot(g,'edgeLabels','on','subdomainLabels','on')
ylim([-.8,.1]) % to see the top edge clearly
6-67
pdegplot
Alternatives If you create the geometry in the PDE app, you can view the geometry
from Boundary Mode. You can see the edge labels by selecting
Boundary > Show Edge Labels, and you can see the subdomain
labels by selecting PDE > Show Subdomain Labels.
See Also decsg | pdetool | wgeom
Related
Examples
Scalar PDE Functional Form and Calling Syntax on page 2-26
6-68
pdegrad
Purpose Gradient of PDE solution
Syntax [ux,uy]=pdegrad(p,t,u)
[ux,uy]=pdegrad(p,t,u,sdl)
Description [ux,uy]=pdegrad(p,t,u) returns the gradient of u evaluated at the
center of each triangle.
Row i from 1 to N of ux contains
c
c
u
x
i
Row i from 1 to N of uy contains
c
c
u
y
i
There is one column for each triangle in t in both ux and uy.
The geometry of the PDE problem is given by the mesh data p and t.
For details on the mesh data representation, see initmesh.
For a description of the format for the solution vector u, see assempde.
The optional argument sdl restricts the computation to the subdomains
in the list sdl.
See Also assempde
How To Gradient or Derivatives of u on page 2-25
f for Systems on page 2-48
6-69
pdeintrp
Purpose Interpolate from node data to triangle midpoint data
Syntax ut=pdeintrp(p,t,un)
Description ut=pdeintrp(p,t,un) gives linearly interpolated values at triangle
midpoints from the values at node points.
The geometry of the PDE problem is given by the mesh data p and t.
For details on the mesh data representation, see initmesh.
Let N be the dimension of the PDE system, n
p
the number of node
points, and n
t
the number of triangles. The components of the node
data are stored in un either as N columns of length n
p
or as an ordinary
solution vector. The first n
p
values of un describe the first component,
the following n
p
values of un describe the second component, and so on.
The components of triangle data are stored in ut as N rows of length n
t
.
Caution pdeprtni and pdeintrp are not inverse functions. The interpolation
introduces some averaging.
See Also assempde | initmesh | pdeprtni
How To Interpolated u on page 2-24
f for Systems on page 2-48
6-70
pdejmps
Purpose Error estimates for adaptation
Syntax errf=pdejmps(p,t,c,a,f,u,alfa,beta,m)
Description errf=pdejmps(p,t,c,a,f,u,alfa,beta,m) calculates the error
indication function used for adaptation. The columns of errf correspond
to triangles, and the rows correspond to the different equations in the
PDE system.
p andt are mesh data. For details, see initmesh.
c, a, and f are PDE coefficients. See Scalar PDE Coefficients on page
2-14 and Coefficients for Systems of PDEs on page 2-42 for details. c,
a, and f must be expanded, so that columns correspond to triangles.
u is the solution vector. For details, see assempde.
The formula for computing the error indicator E(K) for each triangle K is
E K h f au h c u
m
K
m
h
K
( )
= ( ) +
|
\

|
.
|
V
ec

o |
t t
t
1
2
2 2
1 2
[ , ( )]
/
n
where n
t
is the unit normal of edge t and the braced term is the jump
in flux across the element edge, where and are weight indices and
m is an order parameter. The norm is an L
2
norm computed over the
element K. The error indicator is stored in errf as column vectors, one
for each triangle in t. More information can be found in the section
Adaptive Mesh Refinement on page 2-86.
See Also adaptmesh | pdeadgsc | pdeadworst
6-71
pdemdlcv
Purpose Convert Partial Differential Equation Toolbox 1.0 model files to 1.0.2
format
Syntax pdemdlcv(infile,outfile)
Description pdemdlcv(infile,outfile) converts the Partial Differential Equation
Toolbox 1.0 model file infile to a Partial Differential Equation Toolbox
1.0.2 compatible model file. The resulting file is saved as outfile. If
the .m extension is missing in outfile, it is added automatically.
Examples pdedmdlcv('model42.m','model5.m') converts the Partial Differential
Equation Toolbox 1.0 Model file model42.m and saves the converted
model in model5.m.
6-72
pdemesh
Purpose Plot PDE triangular mesh
Syntax pdemesh(p,e,t)
pdemesh(p,e,t,u)
h=pdemesh(p,e,t)
h=pdemesh(p,e,t,u)
Description pdemesh(p,e,t) plots the mesh specified by the mesh data p, e, and t.
h=pdemesh(p,e,t) additionally returns handles to the plotted axes
objects.
pdemesh(p,e,t,u) plots PDE node or triangle data u using a mesh
plot. If u is a column vector, node data is assumed. If u is a row vector,
triangle data is assumed. This command plots substantially faster than
the pdesurf command.
The geometry of the PDE problem is given by the mesh data p, e, and t.
For details on the mesh data representation, see initmesh.
This command is just shorthand for the calls
pdeplot(p,e,t)
pdeplot(p,e,t,'zdata',u)
If you want to have more control over your mesh plot, use pdeplot
instead of pdemesh.
Examples Plot the mesh for the geometry of the L-shaped membrane.
[p,e,t]=initmesh('lshapeg');
[p,e,t]=refinemesh('lshapeg',p,e,t);
pdemesh(p,e,t)
Now solve Poissons equation u = 1 over the geometry defined by the
L-shaped membrane. Use Dirichlet boundary conditions u = 0 on ,
and plot the result.
u=assempde('lshapeb',p,e,t,1,0,1);
6-73
pdemesh
pdemesh(p,e,t,u)
See Also pdecont | pdeplot | pdesurf
How To Mesh Data on page 2-85
Scalar PDE Functional Form and Calling Syntax on page 2-26
6-74
pdenonlin
Purpose Solve nonlinear PDE problem
Syntax [u,res]=pdenonlin(b,p,e,t,c,a,f)
[u,res]=pdenonlin(b,p,e,t,c,a,f,'PropertyName', 'PropertyValue',...)
Description [u,res]=pdenonlin(b,p,e,t,c,a,f) solves the nonlinear scalar PDE
problem
(cu) + au = f on
or the nonlinear system PDE problem
V V + = O ( ) c u u f a on
where the coefficients c, a, and f can depend on the solution u. The
algorithm solves the equation by using damped Newton iteration with
the Armijo-Goldstein line search strategy.
The solution u is represented as the solution vector u. For details on the
representation of the solution vector, see assempde. res is the norm of
the Newton step residuals.
The triangular mesh of the PDE problem is given by the mesh data p, e,
and t. For details on the mesh data representation, see initmesh.
b describes the boundary conditions of the PDE problem. b can be a
Boundary Condition matrix, the name of a Boundary file, or a function
handle to a Boundary file. The formats of the Boundary Condition
matrix and Boundary file are described in the entries on assemb and
pdebound, respectively. For the general call to pdebound, the boundary
conditions can also depend on u. A fixed-point iteration strategy is
employed to solve for the nonlinear boundary conditions.
The coefficients c, a, f of the PDE problem can be given in a wide
variety of ways. In the context of pdenonlin the coefficients can depend
on u. The coefficients cannot depend on t, the time. For a complete
listing of all format options, see Scalar PDE Coefficients on page 2-14
and Coefficients for Systems of PDEs on page 2-42.
6-75
pdenonlin
The solver can be fine-tuned by setting some of the options described
next.
Property Name Property Value Default Description
Jacobian fixed|lumped|full fixed Approximation of Jacobian
U0 string or numeric 0 Initial solution guess
Use the syntax of Initial
Conditions on page 2-62
Tol positive scalar 1e-4 Residual size at termination
MaxIter positive integer 25 Maximum Gauss-Newton
iterations
MinStep positive scalar 1/2^16 Minimum damping of search
direction
Report on|off off Print convergence
information
Norm string or numeric Inf Residual norm
There are three methods currently implemented to compute the
Jacobian:
Numerical evaluation of the full Jacobian based on the sparse version
of the function numjac
A lumped approximation described in Nonlinear Equations on
page 5-24 based on the numerical differentiation of the coefficients
A fixed-point iteration matrix where the Jacobian is approximated by
the stiffness matrix
Select the desired method by setting the Jacobian property to full,
lumped, or fixed, bearing in mind that the more precise methods are
computationally more expensive.
U0 is the starting guess that can be given as an expression, a generic
scalar, or a vector. By default it is set to 0, but this is useless in
problems such as (1/uu) = 0 with Dirichlet boundary conditions
6-76
pdenonlin
u = e
x+y
. Tol fixes the exit criterion from the Gauss-Newton iteration,
i.e., the iterations are terminated when the residual norm is less than
Tol. The norm in which the residual is computed is selected through
Norm. This can be any admissible MATLAB vector norm or energy for
the energy norm.
MaxIter and MinStep are safeguards against infinite Gauss-Newton
loops and they bound the number of iterations and the step size used
in each iteration. Setting Report to on forces printing of convergence
information.
Diagnostics If the Newton iteration does not converge, the error message Too many
iterations or Stepsize too small is displayed. If the initial guess
produces matrices containing NaN or Inf elements, the error message
Unsuitable initial guess U0 (default: U0=0) is printed.
See Also assempde | pdebound
How To Scalar PDE Coefficients on page 2-14
Coefficients for Systems of PDEs on page 2-42
Boundary Conditions for Scalar PDE on page 2-72
Boundary Conditions for PDE Systems on page 2-77
Initial Conditions on page 2-62
6-77
pdeplot
Purpose Plot solution
Syntax pdeplot(p,e,t)
pdeplot(p,e,t,Name,Value)
h = pdeplot( ___ )
Description pdeplot(p,e,t) plots the (p,e,t) mesh.
pdeplot(p,e,t,Name,Value) plots data on the (p,e,t) mesh using one
or more Name,Value pair arguments.
Give at least one of the flowdata (vector field plot), xydata (colored
surface plot), or zdata (3-D height plot) name-value pairs. Otherwise,
pdeplot plots the mesh with no data. You can combine any number of
plot types.
h = pdeplot( ___ ) returns handles to the axis objects using any of the
input arguments in the previous syntaxes.
Input
Arguments
p - Mesh points
matrix
2-by-Np matrix of points, where Np is the number of points in the mesh.
For a description of the (p,e,t) matrices, see Mesh Data on page 2-85.
Typically, you use the p, e, and t data exported from the PDE app, or
generated by initmesh or refinemesh.
Example: [p,e,t] = initmesh(gd)
Data Types
double
e - Mesh edges
matrix
6-78
pdeplot
7-by-Ne matrix of edges, where Ne is the number of edges in the mesh.
For a description of the (p,e,t) matrices, see Mesh Data on page 2-85.
Typically, you use the p, e, and t data exported from the PDE app, or
generated by initmesh or refinemesh.
Example: [p,e,t] = initmesh(gd)
Data Types
double
t - Mesh triangles
matrix
4-by-Nt matrix of triangles, where Nt is the number of triangles in the
mesh. For a description of the (p,e,t) matrices, see Mesh Data on
page 2-85.
Typically, you use the p, e, and t data exported from the PDE app, or
generated by initmesh or refinemesh.
Example: [p,e,t] = initmesh(gd)
Data Types
double
Name-Value Pair Arguments
Specify optional comma-separated pairs of Name,Value arguments.
Name is the argument name and Value is the corresponding
value. Name must appear inside single quotes (' '). You can
specify several name and value pair arguments in any order as
Name1,Value1,...,NameN,ValueN.
Tip Give at least one of the flowdata (vector field plot), xydata
(colored surface plot), or zdata (3-D height plot) name-value pairs.
Otherwise, pdeplot plots the mesh with no data.
6-79
pdeplot
Example: pdeplot(p,e,t,'xydata',u,'zdata',u) sets surface plot
coloring to the solution u, and sets the heights for a 3-D plot to the
solution u.
colorbar - Indicator to include color bar
'on' (default) | 'off'
Indicator to include color bar, specified as the comma-separated pair
consisting of 'colorbar' and a string. 'on' displays a bar giving the
numeric values of colors in the plot. For details, see colorbar. pdeplot
uses the colormap specified in the colormap name-value pair.
Example: 'colorbar','off'
Data Types
char
colormap - Colormap
'cool' (default) | colormap string or matrix
Colormap, specified as the comma-separated pair consisting of
'colormap' and a string representing a built-in colormap, or a colormap
matrix. For details, see colormap.
colormap relates to the xydata name-value pair.
Example: 'colormap','jet'
Data Types
double | char
contour - Indicator to plot level curves
'off' (default) | 'on'
Indicator to plot level curves, specified as the comma-separated pair
consisting of 'contour' and a string. 'on' plots level curves for the
xydata data. Specify the levels with the levels name-value pair.
Example: 'contour','on'
Data Types
char
6-80
pdeplot
flowdata - Data for quiver plot
matrix
Data for quiver plot, specified as the comma-separated pair consisting
of 'flowdata' and a matrix of vector data. flowdata can be M-by-2
or 2-by-M, where M is the number of mesh points p or the number of
triangles t. flowdata contains the x and y values of the field at the
mesh points or at the triangle centroids.
Typically, you set flowdata to the gradient of the solution. For example:
[ux,uy] = pdegrad(p,t,u); % Calculate gradient
ugrad = [ux;uy];
pdeplot(p,e,t,'flowdata',ugrad)
In a 3-D plot, the quiver plot appears in the z = 0 plane.
pdeplot plots the real part of complex data.
Example: 'flowdata',ugrad
Data Types
double
Complex Number Support: Yes
flowstyle - Indicator to show quiver plot
'arrow' (default) | 'off'
Indicator to show quiver plot, specified as the comma-separated pair
consisting of 'flowstyle' and a string. 'arrow'displays the quiver
plot specified by the flowdata name-value pair.
Example: 'flowstyle','off'
Data Types
char
gridparam - Customized grid for xygrid name-value pair
[tn;a2;a3] from an earlier call to tri2grid
6-81
pdeplot
Customized grid for the xygrid name-value pair, specified as the
comma-separated pair consisting of 'gridparam' and the matrix
[tn;a2;a3]. For example:
[~,tn,a2,a3] = tri2grid(p,t,u,x,y);
pdeplot(p,e,t,'xygrid','on','gridparam',[tn;a2;a3],'xydata',u)
For details on the grid data and its x and y arguments, see tri2grid.
Example: 'gridparam',[tn;a2;a3]
Data Types
double
levels - Levels for contour plot
10 (default) | Positive integer | Vector of level values
Levels for contour plot, specified as the comma-separated pair consisting
of 'levels' and a positive integer or a vector.
Positive integer Plot levels equally-spaced contours.
Vector Plot contours at the values in levels.
To obtain a contour plot, set the contour name-value pair to 'on'.
Example: 'levels',16
Data Types
double
mesh - Indicator to show mesh
'off' (default) | 'on'
Indicator to show mesh, specified as the comma-separated pair
consisting of 'mesh' and a string. 'on' shows the mesh in the plot.
Example: 'mesh','on'
Data Types
char
title - Title of plot
6-82
pdeplot
string
Title of plot, specified as the comma-separated pair consisting of
'title' and a string.
Example: 'title','Solution Plot'
Data Types
char
xydata - Colored surface plot data
vector
Colored surface plot data, specified as the comma-separated pair
consisting of 'xydata' and a vector. Give data for points in a vector of
length size(p,2), or data for triangles in a vector of length size(t,2).
Typically, you set xydata to u, the solution. pdeplot uses xydata for
coloring both 2-D and 3-D plots.
pdeplot uses the colormap specified in the colormap name-value pair,
using the style specified in the xystyle name-value pair.
When the contour name-value pair is 'on', pdeplot also plots level
curves of xydata.
pdeplot plots the real part of complex data.
To plot the kth component of a solution to a PDE system, extract the
relevant part of the solution. For example:
np = size(p,2); % number of node points
uk = reshape(u,np,[]); % each uk column has one component of u
pdeplot(p,e,t,'xydata',uk(:,k)) % data for column k
Example: 'xydata',u
Data Types
double
Complex Number Support: Yes
xygrid - Indicator to convert to x-y grid before plotting
6-83
pdeplot
'off' (default) | 'on'
Indicator to convert mesh data to x-y grid before plotting, specified as
the comma-separated pair consisting of 'xygrid' and a string.
Note This conversion can change the geometry, and can lessen the
quality of the plot.
By default, the grid has about sqrt(size(t,2)) elements in each
direction. Exercise more control over the x-y grid by generating it
with the tri2grid function, and passing it in with the gridparam
name-value pair.
Example: 'xygrid','on'
Data Types
char
xystyle - Coloring choice
'interp' (default) | 'off' | 'flat'
Coloring choice, specified as the comma-separated pair consisting of
'xystyle' and a string.
'off' No shading, shows the mesh only.
'flat' Each triangle in the mesh has a uniform color.
'interp' Plot coloring is smoothly interpolated.
The coloring choice relates to the xydata name-value pair.
Example: 'xystyle','flat'
Data Types
char
zdata - Data for 3-D plot heights
matrix
6-84
pdeplot
Data for 3-D plot heights, specified as the comma-separated pair
consisting of 'zdata' and a vector. Give data for points in a vector of
length size(p,2), or data for triangles in a vector of length size(t,2).
Typically, you set zdata to u, the solution. The xydata name-value pair
sets the coloring of the 3-D plot. The zstyle name-value pair specifies
whether the plot is continuous or discontinuous.
pdeplot plots the real part of complex data.
To plot the kth component of a solution to a PDE system, extract the
relevant part of the solution. For example:
np = size(p,2); % number of node points
uk = reshape(u,np,[]); % each uk column has one component of u
pdeplot(p,e,t,'xydata',uk(:,k),'zdata',uk(:,k)) % data for column k
Example: 'zdata',u
Data Types
double
Complex Number Support: Yes
zstyle - 3-D plot style
'continuous' (default) | 'off' | 'discontinuous'
3-D plot style, specified as the comma-separated pair consisting of
'zstyle' and a string.
'off' No 3-D plot.
'discontinuous' Each triangle in the mesh has a uniform height
in a 3-D plot.
'continuous' 3-D surface plot is continuous.
zstyle relates to the zdata name-value pair.
Example: 'zstyle','discontinuous'
Data Types
char
6-85
pdeplot
Output
Arguments
h - Handles to axis objects in the plot
vector of handles
Handles to axis objects in the plot, returned as a vector.
Definitions Quiver Plot
Plot of vector field
Plot of a vector field, also called a flow plot. Arrows show the direction
of the field, with the lengths of the arrows showing the relative sizes of
the field strength. For details on quiver plots, see quiver.
Examples Mesh Plot
Plot the (p,e,t) mesh.
Create the geometry and mesh.
[p,e,t] = initmesh('lshapeg');
Plot the mesh.
pdeplot(p,e,t)
6-86
pdeplot
2-D Solution Plot
Plot a PDE solution as a 2-D colored surface plot.
Create the geometry, mesh, boundary conditions, PDE coefficients, and
solution.
[p,e,t] = initmesh('lshapeg');
u = assempde('lshapeb',p,e,t,1,0,1);
Plot the solution.
6-87
pdeplot
pdeplot(p,e,t,'xydata',u)
3-D Solution Plot
Plot a PDE solution as a 3-D colored plot.
Create the geometry, mesh, boundary conditions, PDE coefficients, and
solution.
[p,e,t] = initmesh('lshapeg');
u = assempde('lshapeb',p,e,t,1,0,1);
6-88
pdeplot
Plot the solution.
pdeplot(p,e,t,'xydata',u,'zdata',u)
Solution Quiver Plot
Plot the gradient of a PDE solution as a quiver plot.
Create the geometry, mesh, boundary conditions, PDE coefficients, and
solution.
6-89
pdeplot
[p,e,t] = initmesh('lshapeg');
u = assempde('lshapeb',p,e,t,1,0,1);
Calculate the gradient of the solution. Put the gradient in a matrix for
inclusion in the quiver plot.
[ux,uy] = pdegrad(p,t,u); % Calculate gradient
ugrad = [ux;uy];
Plot the gradient as a quiver plot.
pdeplot(p,e,t,'flowdata',ugrad)
6-90
pdeplot
Composite Plot
Plot the solution of a PDE in 3-D with the 'jet' coloring and a mesh,
and include a quiver plot. Get handles to the axis objects.
Create the geometry, mesh, boundary conditions, PDE coefficients, and
solution.
[p,e,t] = initmesh('lshapeg');
u = assempde('lshapeb',p,e,t,1,0,1);
6-91
pdeplot
Calculate the gradient of the solution. Put the gradient in a matrix for
inclusion in the quiver plot.
[ux,uy] = pdegrad(p,t,u); % Calculate gradient
ugrad = [ux;uy];
Plot the solution in 3-D with the 'jet' coloring and a mesh, and include
the gradient as a quiver plot.
h = pdeplot(p,e,t,'xydata',u,'zdata',u,...
'colormap','jet','mesh','on','flowdata',ugrad)
h =
174.0082
175.0077
179.0077
6-92
pdeplot
Look underneath to see the quiver plot.
view(20,-20)
6-93
pdeplot
See Also initmesh | pdecont | pdemesh | pdesurf | refinemesh
Related
Examples
Deflection of a Piezoelectric Actuator on page 3-19
Scalar PDE Functional Form and Calling Syntax on page 2-26
Concepts Mesh Data on page 2-85
6-94
pdepoly
Purpose Draw polygon
Syntax pdepoly(x,y)
pdepoly(x,y,label)
Description pdepoly(x,y) draws a polygon with corner coordinates defined by x
and y. If the PDE app is not active, it is automatically started, and the
polygon is drawn in an empty geometry model.
The optional argument label assigns a name to the polygon (otherwise
a default name is chosen).
The state of the Geometry Description matrix inside the PDE app
is updated to include the polygon. You can export the Geometry
Description matrix from the PDE app by using the Export Geometry
Description option from the Draw menu. For a details on the format
of the Geometry Description matrix, see decsg.
Examples The command
pdepoly([-1 0 0 1 1 -1],[0 0 1 1 -1 -1]);
creates the L-shaped membrane geometry as one polygon.
See Also pdecirc | pderect | pdetool
6-95
pdeprtni
Purpose Interpolate from triangle midpoint data to node data
Syntax un=pdeprtni(p,t,ut)
Description un=pdeprtni(p,t,ut) gives linearly interpolated values at node points
from the values at triangle midpoints.
The geometry of the PDE problem is given by the mesh data p and t.
For details on the mesh data representation, see initmesh.
Let N be the dimension of the PDE system, n
p
the number of node
points, and n
t
the number of triangles. The components of triangle data
in ut are stored as N rows of length n
t
. The components of the node
data are stored in un as N columns of length n
p
.
Caution pdeprtni and pdeintrp are not inverse functions. The interpolation
introduces some averaging.
See Also assempde | initmesh | pdeintrp
6-96
pderect
Purpose Draw rectangle
Syntax pderect(xy)
pderect(xy,label)
Description pderect(xy) draws a rectangle with corner coordinates defined by
xy=[xmin xmax ymin ymax]. If the PDE app is not active, it is
automatically started, and the rectangle is drawn in an empty geometry
model.
The optional argument label assigns a name to the rectangle
(otherwise a default name is chosen).
The state of the Geometry Description matrix inside the PDE app
is updated to include the rectangle. You can export the Geometry
Description matrix from the PDE app by selecting the Export
Geometry Description option from the Draw menu. For details on
the format of the Geometry Description matrix, see decsg.
Examples The following command sequence starts the PDE app and draws the
L-shaped membrane as the union of three squares.
pderect([-1 0 -1 0])
pderect([0 1 -1 0])
pderect([0 1 0 1])
See Also pdecirc | pdeellip | pdepoly | pdetool
6-97
pdesdp
Purpose Indices of points/edges/triangles in set of subdomains
Syntax c=pdesdp(p,e,t)
[i,c]=pdesdp(p,e,t)
c=pdesdp(p,e,t,sdl)
[i,c]=pdesdp(p,e,t,sdl)
i=pdesdt(t)
i=pdesdt(t,sdl)
i=pdesde(e)
i=pdesde(e,sdl)
Description [i,c]=pdesdp(p,e,t,sdl) given mesh data p, e, and t and a list of
subdomain numbers sdl, the function returns all points belonging
to those subdomains. A point can belong to several subdomains,
and the points belonging to the domains in sdl are divided into two
disjoint sets. i contains indices of the points that wholly belong to the
subdomains listed in sdl, and c lists points that also belongs to the
other subdomains.
c=pdesdp(p,e,t,sdl) returns indices of points that belong to more
than one of the subdomains in sdl.
i=pdesdt(t,sdl) given triangle data t and a list of subdomain numbers
sdl, i contains indices of the triangles inside that set of subdomains.
i=pdesde(e,sdl) given edge data e, it extracts indices of outer
boundary edges of the set of subdomains.
If sdl is not given, a list of all subdomains is assumed.
6-98
pdesmech
Purpose Calculate structural mechanics tensor functions
Syntax ux=pdesmech(p,t,c,u,'PropertyName',PropertyValue,...)
Description ux=pdesmech(p,t,c,u,p1,v1,...) returns a tensor expression
evaluated at the center of each triangle. The tensor expressions are
stresses and strains for structural mechanics applications with plane
stress or plane strain conditions. pdesmech is intended to be used for
postprocessing of a solution computed using the structural mechanics
application modes of the PDE app, after exporting the solution, the
mesh, and the PDE coefficients to the MATLAB workspace. Poissons
ratio, nu, has to be supplied explicitly for calculations of shear stresses
and strains, and for the von Mises effective stress in plane strain mode.
Valid property name/property value pairs include the following.
Property Name Property Value/Default Description
tensor ux|uy|vx|vy|exx|eyy|exy|sxx|syy|sxy|e1|
e2|s1|s2|{von Mises}
Tensor expression
application {ps}|pn Plane stress|plane
strain
nu Scalar or string expression {0.3} Poissons ratio
The available tensor expressions are
ux, which is
c
c
u
x
uy, which is
c
c
u
y
vx, which is
c
c
v
x
vy, which is
c
c
v
y
6-99
pdesmech
exx, the x-direction strain (
x
)
eyy, the y-direction strain (
y
)
exy, the shear strain (
xy
)
sxx, the x-direction stress (
x
)
syy, the y-direction stress (
y
)
sxy, the shear stress (
xy
)
e1, the first principal strain (
1
)
e2, the second principal strain (
2
)
s1, the first principal stress (
1
)
s2, the second principal stress (
2
)
von Mises, the von Mises effective stress, for plane stress conditions
o o o o
1
2
2
2
1 2
+
or for plane strain conditions
( )( ) o o o o
1
2
2
2 2
1 2
2
1 2 2 1 + + +
( )
v v v v
where v is Poissons ratio nu.
Examples Assuming that a problem has been solved using the application
mode Structural Mechanics, Plane Stress, discussed in Structural
Mechanics Plane Stress on page 3-6, and that the solution u, the
mesh data p and t, and the PDE coefficient c all have been exported to
the MATLAB workspace, the x-direction strain is computed as
sx=pdesmech(p,t,c,u,'tensor','sxx');
To compute the von Mises effective stress for a plane strain problem
with Poissons ratio equal to 0.3, type
6-100
pdesmech
mises=pdesmech(p,t,c,u,'tensor','von Mises',...
'application','pn','nu',0.3);
6-101
pdesurf
Purpose Shorthand command for surface plot
Syntax pdesurf(p,t,u)
Description pdesurf(p,t,u) plots a 3-D surface of PDE node or triangle data. If
u is a column vector, node data is assumed, and continuous style and
interpolated shading are used. If u is a row vector, triangle data is
assumed, and discontinuous style and flat shading are used.
h=pdesurf(p,t,u) additionally returns handles to the drawn axes
objects.
For node data, this command is just shorthand for the call
pdeplot(p,[],t,'xydata',u,'xystyle','interp',...
'zdata',u,'zstyle','continuous',...
'colorbar','off');
and for triangle data it is
pdeplot(p,[],t,'xydata',u,'xystyle','flat',...
'zdata',u,'zstyle','discontinuous',...
'colorbar','off');
If you want to have more control over your surface plot, use pdeplot
instead of pdesurf.
Examples Surface plot of the solution to the equation u = 1 over the geometry
defined by the L-shaped membrane. Use Dirichlet boundary conditions
u = 0 on .
[p,e,t]=initmesh('lshapeg');
[p,e,t]=refinemesh('lshapeg',p,e,t);
u=assempde('lshapeb',p,e,t,1,0,1);
pdesurf(p,t,u)
See Also pdecont | pdemesh | pdeplot
6-102
pdetool
Purpose Open PDE app
Syntax pdetool
Description pdetool starts the PDE app. You should not call pdetool with
arguments.
The PDE app helps you to draw the 2-D domain and to define boundary
conditions for a PDE problem. It also makes it possible to specify the
partial differential equation, to create, inspect and refine the mesh, and
to compute and display the solution from the PDE app.
The PDE app contains several different modes:
In draw mode, you construct a Constructive Solid Geometry model (CSG
model) of the geometry. You can draw solid objects that can overlap.
There are four types of solid objects:
Circle object represents the set of points inside a circle.
Polygon object represents the set of points inside the polygon
given by a set of line segments.
Rectangle object represents the set of points inside the rectangle
given by a set of line segments.
Ellipse object represents the set of points inside an ellipse. The
ellipse can be rotated.
The solid objects can be moved and rotated. Operations apply to
groups of objects by doing multiple selects. (A Select all option is also
available.) You can cut and paste among the selected objects. The model
can be saved and restored. the PDE app can be started by just typing
the name of the model. (This starts the corresponding file that contains
the MATLAB commands necessary to create the model.)
The solid objects can be combined by typing a set formula. Each object
is automatically assigned a unique name, which is displayed in the PDE
app on the solid object itself. The names refer to the object in the set
formula. More specifically, in the set formula, the name refers to the set
of points inside the object. The resulting geometrical model is the set of
6-103
pdetool
points for which the set formula evaluates to true. (For a description
of the syntax of the set formula, see decsg.) By default, the resulting
geometrical model is the union of all objects.
A snap-to-grid function is available. This means that objects align to
the grid. The grid can be turned on and off, and the scaling and the
grid spacing can be changed.
In boundary mode, you can specify the boundary conditions. You can
have different types of boundary conditions on different boundaries. In
this mode, the original shapes of the solid building objects constitute
borders between subdomains of the model. Such borders can be
eliminated in this mode. The outer boundaries are color coded to indicate
the type of boundary conditions. A red outer boundary corresponds to
Dirichlet boundary conditions, blue to generalized Neumann boundary
conditions, and green to mixed boundary conditions. You can return to
the boundary condition display by clicking the button or by selecting
Boundary Mode from the Boundary menu.
In PDE mode, you can specify the type of PDE problem, and the
coefficients c, a, f and d. You can specify the coefficients for each
subdomain independently. This makes it easy to specify, e.g.,
various material properties in one PDE model. The PDE to be solved
can be specified by clicking the PDE button or by selecting PDE
Specification from the PDE menu. This brings up a dialog box.
In mesh mode, you can control the automated mesh generation and plot
the mesh. An initial mesh can be generated by clicking the button
or by selecting Initialize Mesh from the Mesh menu. Choose the
meshing algorithm using the Mesh > Parameters > Mesher version
menu. The 'R2013a' algorithm runs faster, and can triangulate more
geometries than the 'preR2013a' algorithm. The initial mesh can be
repeatedly refined by clicking the refine button or by selecting Refine
Mesh from the Mesh menu.
In solve mode, you can specify solve parameters and solve the PDE. For
parabolic and hyperbolic PDE problems, you can also specify the initial
conditions, and the times at which the output should be generated.
For eigenvalue problems, the search range can be specified. Also, the
6-104
pdetool
adaptive and nonlinear solvers for elliptic PDEs can be invoked. The
PDE problem is solved by clicking the = button or by selecting Solve
PDE from the Solve menu. By default, the solution is plotted in the
PDE app axes.
In plot mode, you can select a wide variety of visualization methods
such as surface, mesh, contour, and quiver (vector field) plots. For
surface plots, you can choose between interpolated and flat rendering
schemes. The mesh can be hidden in all plot types. For parabolic and
hyperbolic equations, you can animate the solution as it changes with
time. You can show the solution both in 2-D and 3-D. 2-D plots are
shown inside the PDE app. 3-D plots are plotted in separate figure
windows. Different types of plots can be selected by clicking the button
with the solution plot icon or by selecting Parameters from the Plot
menu. This opens a dialog box.
Boundary Condition Dialog Box
In this dialog box, the boundary condition for the selected boundaries is
entered. The following boundary conditions can be handled:
Dirichlet: hu = r on the boundary.
Generalized Neumann:

n c u qu g V
( )
+ = on the boundary.
Mixed: a combination of Dirichlet and generalized Neumann
condition.

n is the outward unit length normal.

The boundary conditions can be entered in a variety of ways. (See
assemb and Boundary Menu on page 4-15.)
PDE Specification Dialog Box
In this dialog box, the type of PDE and the PDE coefficients are entered.
The following types of PDEs can be handled:
Elliptic PDE: (cu) + au = f
Parabolic PDE: d c au f
u
t
u
c
c
V V
( )
+ =
6-105
pdetool
Hyperbolic PDE: d
u
t
c u au f
c
c
V V + =
2
2
( )
Eigenvalue PDE: (cu) + au = du
for x and y on the problems 2-D domain .
The PDE coefficients can be entered in a variety of ways. (See Scalar
PDE Coefficients on page 2-14 and Coefficients for Systems of PDEs
on page 2-42 and PDE Menu on page 4-19.)
Model File
The Model file contains the MATLAB commands necessary to create a
CSG model. It can also contain additional commands to set boundary
conditions, define the PDE, create the mesh, solve the pde, and plot the
solution. This type of file can be saved and opened from the File menu.
The Model file is a MATLAB function and not a script. This way
name clashes between variables used in the function and in the main
workspace are avoided. The name of the file must coincide with the
model name. The beginning of the file always looks similar to the
following code fragment:
function pdemodel
pdeinit;
pde_fig=gcf;
ax=gca;
pdetool('appl_cb',1);
setappdata(pde_fig,'currparam',...
char('1.0','0.0','10.0','1.0'));
pdetool('snapon');
set(ax,'XLim',[-1.5 1.5]);
set(ax,'YLim',[-1 1]);
set(ax,'XTickMode','auto');
set(ax,'YTickMode','auto');
grid on;
6-106
pdetool
The pdeinit command starts up the PDE app. If the PDE app has
already been started, the current model is cleared. The following
commands set up the scaling and tick marks of the axis of the PDE
app and other user parameters.
Then a sequence of drawing commands is issued. The commands that
can be used are named pdecirc, pdeellip, pdepoly, and pderect.
The following command sequence creates the L-shaped membrane as
the union of three squares. The solid objects are given names SQ1,
SQ2, SQ3, etc.
% Geometry description:
pderect([-1 0 0 -1],'SQ1');
pderect([0 1 0 -1],'SQ2');
pderect([0 1 1 0],'SQ3');
We do not intend to fully document the format of the Model file. It can
be used to change the geometry of the drawn objects, since the pdecirc,
pdeellip, pdepoly, and pderect commands are documented.
See Also assempde | initmesh | parabolic | pdecont | pdeeig | pdesurf
6-107
pdetrg
Purpose Triangle geometry data
Syntax [ar,a1,a2,a3]=pdetrg(p,t)
[ar,g1x,g1y,g2x,g2y,g3x,g3y]=pdetrg(p,t)
Description [ar,a1,a2,a3]=pdetrg(p,t) returns the area of each triangle in ar
and half of the negative cotangent of each angle in a1, a2, and a3.
[ar,g1x,g1y,g2x,g2y,g3x,g3y]=pdetrg(p,t) returns the area and
the gradient components of the triangle base functions.
The triangular mesh of the PDE problem is given by the mesh data p
and t. For details on the mesh data representation, see initmesh.
6-108
pdetriq
Purpose Triangle quality measure
Syntax q=pdetriq(p,t)
Description q=pdetriq(p,t) returns a triangle quality measure given mesh data.
The triangular mesh is given by the mesh data p, e, and t. For details
on the mesh data representation, see initmesh.
The triangle quality is given by the formula
q
a
h h h
=
+ +
4 3
1
2
2
2
3
2
where a is the area and h
1
, h
2
, and h
3
the side lengths of the triangle.
If q > 0.6 the triangle is of acceptable quality. q = 1 when h
1
= h
2
= h
3
.
References Bank, Randolph E., PLTMG: A Software Package for Solving Elliptic
Partial Differential Equations, Users Guide 6.0, Society for Industrial
and Applied Mathematics, Philadelphia, PA, 1990.
See Also initmesh | jigglemesh | refinemesh
6-109
poiasma
Purpose Boundary point matrix contributions for fast solvers of Poissons
equation
Syntax K=poiasma(n1,n2,h1,h2)
K=poiasma(n1,n2)
K=poiasma(n)
Description K=poiasma(n1,n2,h1,h2) assembles the contributions to the stiffness
matrix from boundary points. n1 and n2 are the numbers of points in
the first and second directions, and h1 and h2 are the mesh spacings.
K is a sparse n1*n2-by-n1*n2 matrix. The point numbering is the
canonical numbering for a rectangular mesh.
K=poiasma(n1,n2) uses h1=h2.
K=poiasma(n) uses n1=n2=n.
See Also poiindex | poisolv
6-110
poicalc
Purpose Fast solver for Poissons equation on rectangular grid
Syntax u=poicalc(f,h1,h2,n1,n2)
u=poicalc(f,h1,h2)
u=poicalc(f)
Description u=poicalc(f,h1,h2,n1,n2) calculates the solution of Poissons
equation for the interior points of an evenly spaced rectangular grid.
The columns of u contain the solutions corresponding to the columns of
the right-hand side f. h1 and h2 are the spacings in the first and second
direction, and n1 and n2 are the number of points.
The number of rows in f must be n1*n2. If n1 and n2 are not given, the
square root of the number of rows of f is assumed. If h1 and h2 are not
given, they are assumed to be equal.
The ordering of the rows in u and f is the canonical ordering of interior
points, as returned by poiindex.
The solution is obtained by sine transforms in the first direction and
tridiagonal matrix solution in the second direction. n1 should be 1 less
than a power of 2 for best performance.
See Also dst | idst | poiasma | poiindex | poisolv
6-111
poiindex
Purpose Indices of points in canonical ordering for rectangular grid
Syntax [n1,n2,h1,h2,i,c,ii,cc]=poiindex(p,e,t,sd)
Description [n1,n2,h1,h2,i,c,ii,cc]=poiindex(p,e,t,sd) identifies a given
grid p, e, t in the subdomain sd as an evenly spaced rectangular grid. If
the grid is not rectangular, n1 is 0 on return. Otherwise n1 and n2 are
the number of points in the first and second directions, h1 and h2 are
the spacings. i and ii are of length (n1-2)*(n2-2) and contain indices
of interior points. i contains indices of the original mesh, whereas
ii contains indices of the canonical ordering. c and cc are of length
n1*n2-(n1-2)*(n2-2) and contain indices of border points. ii and cc
are increasing.
In the canonical ordering, points are numbered from left to right and
then from bottom to top. Thus if n1=3 and n2=5, then ii=[5 8 11] and
cc=[1 2 3 4 6 7 9 10 12 13 14 15].
See Also poiasma | poisolv
6-112
poimesh
Purpose Make regular mesh on rectangular geometry
Syntax [p,e,t]=poimesh(g,nx,ny)
[p,e,t]=poimesh(g,n)
[p,e,t]=poimesh(g)
Description [p,e,t]=poimesh(g,nx,ny) constructs a regular mesh on the
rectangular geometry specified by g, by dividing the x edge into nx
pieces and the y edge into ny pieces, and placing (nx+1)*(ny+1)
points at the intersections.
The x edge is the one that makes the smallest angle with the x-axis.
[p,e,t]=poimesh(g,n) uses nx=ny=n, and [p,e,t]=poimesh(g) uses
nx=ny=1.
The triangular mesh is described by the mesh data p, e, and t. For
details on the mesh data representation, see initmesh.
For best performance with poisolv, the larger of nx and ny should
be a power of 2.
If g does not seem to describe a rectangle, p is zero on return.
Examples Try the command pdedemo8. The solution of Poissons equation over a
rectangular grid with boundary condition given by the file squareb4 is
returned. The solution time is compared to the usual Finite Element
Method (FEM) approach.
See Also initmesh | poisolv
6-113
poisolv
Purpose Fast solution of Poissons equation on rectangular grid
Syntax u=poisolv(b,p,e,t,f)
Description u=poisolv(b,p,e,t,f) solves Poissons equation with Dirichlet
boundary conditions on a regular rectangular grid. A combination
of sine transforms and tridiagonal solutions is used for increased
performance.
The boundary conditions b must specify Dirichlet conditions for all
boundary points.
The mesh p, e, and t must be a regular rectangular grid. For details on
the mesh data representation, see initmesh.
f gives the right-hand side of Poissons equation.
Apart from roundoff errors, the result should be the same as
u=assempde(b,p,e,t,1,0,f).
References Strang, Gilbert, Introduction to Applied Mathematics,
Wellesley-Cambridge Press, Cambridge, MA, 1986, pp. 453458.
See Also poicalc | poimesh
6-114
refinemesh
Purpose Refine triangular mesh
Syntax [p1,e1,t1]=refinemesh(g,p,e,t)
[p1,e1,t1]=refinemesh(g,p,e,t,'regular')
[p1,e1,t1]=refinemesh(g,p,e,t,'longest')
[p1,e1,t1]=refinemesh(g,p,e,t,it)
[p1,e1,t1]=refinemesh(g,p,e,t,it,'regular')
[p1,e1,t1]=refinemesh(g,p,e,t,it,'longest')
[p1,e1,t1,u1]=refinemesh(g,p,e,t,u)
[p1,e1,t1,u1]=refinemesh(g,p,e,t,u,'regular')
[p1,e1,t1,u1]=refinemesh(g,p,e,t,u,'longest')
[p1,e1,t1,u1]=refinemesh(g,p,e,t,u,it)
[p1,e1,t1,u1]=refinemesh(g,p,e,t,u,it,'regular')
[p1,e1,t1,u1]=refinemesh(g,p,e,t,u,it,'longest')
Description [p1,e1,t1]=refinemesh(g,p,e,t) returns a refined version of the
triangular mesh specified by the geometry g, Point matrix p, Edge
matrix e, and Triangle matrix t.
The triangular mesh is given by the mesh data p, e, and t. For details
on the mesh data representation, see Mesh Data on page 2-85.
[p1,e1,t1,u1]=refinemesh(g,p,e,t,u) refines the mesh and also
extends the function u to the new mesh by linear interpolation. The
number of rows in u should correspond to the number of columns in p,
and u1 has as many rows as there are points in p1. Each column of u
is interpolated separately.
An extra input argument it is interpreted as a list of subdomains to
refine, if it is a row vector, or a list of triangles to refine, if it is a column
vector.
The default refinement method is regular refinement, where all of the
specified triangles are divided into four triangles of the same shape.
Longest edge refinement, where the longest edge of each specified
triangle is bisected, can be demanded by giving longest as a final
parameter. Using regular as a final parameter results in regular
refinement. Some triangles outside of the specified set may also be
refined to preserve the triangulation and its quality.
6-115
refinemesh
Examples Refine the mesh of the L-shaped membrane several times. Plot the
mesh for the geometry of the L-shaped membrane.
[p,e,t]=initmesh('lshapeg','hmax',inf);
subplot(2,2,1), pdemesh(p,e,t)
[p,e,t]=refinemesh('lshapeg',p,e,t);
subplot(2,2,2), pdemesh(p,e,t)
[p,e,t]=refinemesh('lshapeg',p,e,t);
subplot(2,2,3), pdemesh(p,e,t)
[p,e,t]=refinemesh('lshapeg',p,e,t);
subplot(2,2,4), pdemesh(p,e,t)
subplot
Algorithms The algorithm is described by the following steps:
1 Pick the initial set of triangles to be refined.
2 Either divide all edges of the selected triangles in half (regular
refinement), or divide the longest edge in half (longest edge
refinement).
3 Divide the longest edge of any triangle that has a divided edge.
4 Repeat step 3 until no further edges are divided.
5 Introduce new points of all divided edges, and replace all divided
entries in e by two new entries.
6 Form the new triangles. If all three sides are divided, new triangles
are formed by joining the side midpoints. If two sides are divided,
the midpoint of the longest edge is joined with the opposing corner
and with the other midpoint. If only the longest edge is divided, its
midpoint is joined with the opposing corner.
See Also initmesh | pdeent | pdegeom | pdesdt
How To Mesh Data on page 2-85
6-116
sptarn
Purpose Solve generalized sparse eigenvalue problem
Syntax [xv,lmb,iresult] = sptarn(A,B,lb,ub)
[xv,lmb,iresult] = sptarn(A,B,lb,ub,spd)
[xv,lmb,iresult] = sptarn(A,B,lb,ub,spd,tolconv)
[xv,lmb,iresult] = sptarn(A,B,lb,ub,spd,tolconv,jmax)
[xv,lmb,iresult] = sptarn(A,B,lb,ub,spd,tolconv,jmax,maxmul)
Description [xv,lmb,iresult] =
sptarn(A,B,lb,ub,spd,tolconv,jmax,maxmul) finds eigenvalues of
the pencil (A B)x = 0 in interval [lb,ub]. (A matrix of linear
polynomials A
ij
B
ij
, A B, is called a pencil.)
A and B are sparse matrices. lb and ub are lower and upper bounds for
eigenvalues to be sought. We may have lb=-inf if all eigenvalues to
the left of ub are sought, and rb=inf if all eigenvalues to the right of
lb are sought. One of lb and ub must be finite. A narrower interval
makes the algorithm faster. In the complex case, the real parts of lmb
are compared to lb and ub.
xv are eigenvectors, ordered so that norm(a*xv-b*xv*diag(lmb)) is
small. lmb is the sorted eigenvalues. If iresult>=0 the algorithm
succeeded, and all eigenvalues in the intervals have been found. If
iresult<0 the algorithm has not yet been successful, there may be
more eigenvaluestry with a smaller interval.
spd is 1 if the pencil is known to be symmetric positive definite (default
0).
tolconv is the expected relative accuracy. Default is 100*eps, where
eps is the machine precision.
jmax is the maximum number of basis vectors. The algorithm needs
jmax*n working space so a small value may be justified on a small
computer, otherwise let it be the default value jmax=100. Normally the
algorithm stops earlier when enough eigenvalues have converged.
maxmul is the number of Arnoldi runs tried. Must at least be as large
as maximum multiplicity of any eigenvalue. If a small value of jmax is
6-117
sptarn
given, many Arnoldi runs are necessary. The default value is maxmul=n,
which is needed when all the eigenvalues of the unit matrix are sought.
Algorithms The Arnoldi algorithm with spectral transformation is used. The shift
is chosen at ub, lb, or at a random point in interval (lb,ub) when both
bounds are finite. The number of steps j in the Arnoldi run depends
on how many eigenvalues there are in the interval, but it stops at
j=min(jmax,n). After a stop, the algorithm restarts to find more Schur
vectors in orthogonal complement to all those already found. When no
more eigenvalues are found in lb < lmb <= ub, the algorithm stops.
For small values of jmax, several restarts may be needed before a
certain eigenvalue has converged. The algorithm works when jmax
is at least one larger than the number of eigenvalues in the interval,
but then many restarts are needed. For large values of jmax, which
is the preferred choice, mul+1 runs are needed. mul is the maximum
multiplicity of an eigenvalue in the interval.
6-118
sptarn
Note The algorithm works on nonsymmetric as well as symmetric
pencils, but then accuracy is approximately tol times the Henrici
departure from normality. The parameter spd is used only to choose
between symamd and colamd when factorizing, the former being
marginally better for symmetric matrices close to the lower end of the
spectrum.
In case of trouble,
If convergence is too slow, try (in this order of priority):
a smaller interval lb, ub
a larger jmax
a larger maxmul
If factorization fails, try again with lb or ub finite. Then shift is chosen
at random and hopefully not at an eigenvalue. If it fails again, check
whether pencil may be singular.
If it goes on forever, there may be too many eigenvalues in the strip.
Try with a small value maxmul=2 and see which eigenvalues you get.
Those you get are some of the eigenvalues, but a negative iresult tells
you that you have not gotten them all.
If memory overflow, try smaller jmax.
The algorithm is designed for eigenvalues close to the real axis. If you
want those close to the imaginary axis, try A=i*A.
When spd=1, the shift is at lb so that advantage is taken of the faster
factorization for symmetric positive definite matrices. No harm is done,
but the execution is slower if lb is above the lowest eigenvalue.
6-119
sptarn
References [1] Golub, Gene H., and Charles F. Van Loan, Matrix Computations,
2nd edition, Johns Hopkins University Press, Baltimore, MD, 1989.
[2] Saad, Yousef, Variations on Arnoldis Method for Computing
Eigenelements of Large Unsymmetric Matrices, Linear Algebra and
its Applications, Vol. 34, 1980, pp. 269295.
See Also pdeeig
6-120
tri2grid
Purpose Interpolate from PDE triangular mesh to rectangular grid
Syntax uxy=tri2grid(p,t,u,x,y)
[uxy,tn,a2,a3]=tri2grid(p,t,u,x,y)
uxy=tri2grid(p,t,u,tn,a2,a3)
Description uxy=tri2grid(p,t,u,x,y) computes the function values uxy over the
grid defined by the vectors x and y, from the function u with values on
the triangular mesh defined by p and t. Values are computed using
linear interpolation in the triangle containing the grid point. The
vectors x and y must be increasing.
[uxy,tn,a2,a3]=tri2grid(p,t,u,x,y) additionally lists the index tn
of the triangle containing each grid point, and interpolation coefficients
a2 and a3.
uxy=tri2grid(p,t,u,tn,a2,a3) with tn, a2, and a3 computed in an
earlier call to tri2grid, interpolates using the same grid as in the
earlier call. This variant is, however, much faster if several functions
have to be interpolated using the same grid.
For grid points outside of the triangular mesh, NaN is returned in uxy,
tn, a2, and a3.
See Also assempde | initmesh | refinemesh
6-121
wbound
Purpose Write boundary condition specification file
Syntax fid=wbound(bl,mn)
Description fid=wbound(bl,mn) writes a Boundary file with the name [mn,'.m'].
The Boundary file is equivalent to the Boundary Condition matrix bl.
The output fid is -1 if the file could not be written.
bl describes the boundary conditions of the PDE problem. bl is a
Boundary Condition matrix. For details, see assemb.
The output file [mn,'.m'] is the name of a Boundary file. (See
pdebound.)
See Also decsg | pdebound | pdegeom | wgeom
6-122
wgeom
Purpose Write geometry specification function
Syntax fid=wgeom(dl,mn)
Description fid=wgeom(dl,mn) writes a Geometry file with the name [mn,'.m'].
The Geometry file is equivalent to the Decomposed Geometry matrix
dl. fid returns -1 if the file could not be written.
dl is a Decomposed Geometry matrix. For a description of the format of
the Decomposed Geometry matrix, see decsg.
The output file [mn,'.m'] is the name of a Geometry file. For a
description of the Geometry file format, see pdegeom.
See Also decsg | pdegeom | wbound
How To Boundary Conditions for Scalar PDE on page 2-72
Boundary Conditions for PDE Systems on page 2-77
6-123
wgeom
6-124
Index
Index A
AC power electromagnetics 3-44
adaptive mesh refinement 1-36
adaptmesh function 6-2
animation 3-90
Application command 4-11
application modes 1-7
Armijo-Goldstein line search 5-26
Arnoldi algorithm 6-118
assema function 6-9
assemb function 6-10
assembling 5-6
assempde function 6-18
Axes Limits command 4-11
B
Boolean table 6-25
border segment 6-27
Boundary Condition matrix 3-116
boundary conditions 4-16
Boundary file 3-116
Boundary menu 4-15
boundary mode 1-28
boundary segment 6-27
C
circle solid 6-26
Coefficient file 3-116
Coefficient matrix 3-116
command-line functions 3-113
conductive media DC 3-50
Constructive Solid Geometry model 2-5
CSG model 2-5
csgchk function 6-22
csgdel function 6-24
cylindrical problem 3-92
D
decomposed geometry 2-5
Decomposed Geometry matrix 3-116
decsg function 6-25
discrete sine transform 3-119
domain decomposition 3-83
Draw menu 4-13
draw mode 1-28
E
Edge matrix 3-117
Edit menu 4-7
eigenmodes 3-106
eigenvalue equation 5-19
eigenvalue problems 1-36
eigenvector matrix 6-56
electrostatics 3-34
ellipse solid 6-27
elliptic equation 1-4
elliptic problems 3-72
elliptic system 5-10
energy norm 5-6
F
fast solver 3-119
FEM 1-32
File menu 4-4
Finite Element Method 5-1
G
Gauss-Newton method 5-26
Geometry Description matrix 3-116
Geometry file 3-116
graphical user interface 1-12
Greens formula 1-34
Grid Spacing command 4-10
Index-1
Index
H
heat distribution in radioactive rod 3-92
heat equation 3-88
heat transfer 3-57
Helmholtzs equation 3-77
Help menu 4-39
hyperbolic equation 1-4
hyperbolic function 6-32
hyperbolic problems 3-95
I
idst function 6-30
initmesh function 6-35
J
jigglemesh function 6-40
L
L-shaped membrane 3-100
Laplace equation 3-35
M
magnetostatics 3-37
mass matrix 5-5
Maxwells equations 3-34
mesh data 3-117
Mesh menu 4-21
mesh mode 1-28
mesh parameters 4-23
method of lines 5-14
minimal region 3-116
minimal surface problem 3-81
N
Name Space matrix 3-116
node data 6-70
nonlinear
equation 5-24
problem 3-81
solver 1-36
O
Open command 4-5
Options menu 4-9
P
parabolic equation 1-4
parabolic function 6-42
parabolic problems 3-88
Paste command 4-8
PDE coefficients 4-20
PDE menu 4-19
PDE mode 1-28
PDE Specification 4-20
pdeadgsc function 6-45
pdeadworst function 6-46
pdearcl function 6-47
pdebound function 6-48
pdecgrad function 6-51
pdecirc function 6-53
pdecont function 6-54
pdeeig function 6-56
pdeellip function 6-59
pdeent function 6-60
pdegeom function 6-61
pdegrad function 6-69
pdeintrp function 6-70
pdejmps function 6-71
pdemdlcv function 6-72
pdemesh function 6-73
pdenonlin function 6-75
pdepoly function 6-95
pdeprtni function 6-96
pderect function 6-97
Index-2
Index
pdesde function 6-98
pdesdp function 6-98
pdesdt function 6-98
pdesmech function 6-99
pdesurf function 6-102
pdetool function 6-103
pdetrg function 6-108
pdetriq function 6-109
pencil 6-117
plane strain 3-13
plane stress 3-6
Plot menu 4-31
plot mode 1-28
Plot parameters 4-31
poiasma function 6-110
poicalc function 6-111
poiindex function 6-112
poimesh function 6-113
Point matrix 3-117
poisolv function 6-114
Poissons equation 1-12
polygon solid 6-27
Print command 4-6
R
rectangle solid 6-27
refinemesh function 6-115
Robin boundary condition 5-2
Rotate command 4-14
S
Save As command 4-5
scattering problem 3-76
Schur vector 5-22
skin effect 3-46
solid object 1-14
solution vector 3-117
Solve menu 4-25
solve mode 1-28
solve parameters 4-26
sptarn function 6-117
stiff springs 5-11
structural mechanics 3-6
T
test function 1-34
toolbar 1-28
tri2grid function 6-121
Triangle matrix 3-117
triangle quality 4-21
V
von Mises effective stress 3-9
W
wave equation 3-95
wbound function 6-122
weak form 5-3
wgeom function 6-123
Window menu 4-38
Index-3