Classical Damping, Non-Classical Damping and Complex Modes

CEE 541. Structural Dynamics

Department of Civil and Environmental Engineering
Duke University
Henri P. Gavin
Fall 2018

1 Classical Damping

The equations of motion of an un-forced N degree of freedom elastic structure with viscous
damping are
M r̈(t) + C ṙ(t) + Kr(t) = 0, (1)
with initial conditions r(0) = do and ṙ(0) = vo . If the system is un-damped (C = 0N ×N ), the
free response of the system will not decay with time, and a suitable trial solution to the differential
equation (1) is r(t) = r̄ sin(ωn t), where r̄ is a constant vector of dimension N . Differentiating r(t)
twice, r̈(t) = −ωn 2 r̄ sin(ωn t), and substituting the trial solution into equation (1) we obtain

− ωn 2 M r̄ sin(ωn t) + K r̄ sin(ωn t) = 0. (2)

For the assumed trial solution to be true for all time,

2
[K − ωnj M ]r̄j = 0, (3)

which is a general eigen-value problem, in which eigen-values are squared natural frequencies, ωn 2j ,
and the eigen-vectors are mode-shape vectors, r̄j . If the structure is modeled with N degrees of
freedom, then there will be N natural frequencies and N modal vectors. The modal matrix R̄ is
the column-wise concatenation of the N mode-shape vectors, R̄ = [r̄1 r̄2 · · · r̄N ]. The modal matrix
R̄ diagonalizes both the mass and stiffness matrices. The Rayleigh quotient is the ratio of the
diagonalized stiffness matrix to the diagonalized mass matrix.

k1∗ /m∗1 2
ωn1
   
R̄T K R̄ .. .. 2
= = =Ω . (4)
   
. .
R̄T M R̄ ∗ /m∗ 2
kN N ωnN

For mass-normalized modal vectors R̄T M R̄ = IN and R̄T K R̄ = Ω2 .

A damping matrix that is diagonalizeable by R̄ is called a classical damping matrix.

 ∗
c1 /m∗1 2ζ1 ωn1
  
R̄T C R̄ .. ..
= = . (5)
   
. .
R̄T M R̄
c∗N /m∗N 2ζN ωnN

where ζj is the damping ratio of the i-th mode, and ωni is the un-damped natural frequency of the
i-th mode. Systems with classical damping are triple diagonalizeable. The modal vectors of triple
diagonalizeable systems depend only on M and K, and are independent of C, regardless of how
heavily the system is damped. There are many ways to compute a classical damping matrix from
mass and stiffness matrices.
2 CEE 541. Structural Dynamics – Duke University – Fall 2018 – H.P. Gavin

A Rayleigh damping matrix is proportional to the mass and stiffness matrices [6],

C = αM + βK. (6)

where α and β are related to damping ratios and frequencies by

1 ωk
ζk = α +β (7)
2ωk 2
Mass proportional damping ratios decrease inversely with ω and stiffness proportional damping
ratios increase linearly with ω.

Rayleigh damping can be extended. It can be shown that the damping matrix

C = αM + βK + γM K −1 M + δKM −1 K (8)

is a classical damping matrix. An extended Rayleigh damping matrix, called Caughey damping
[1, 2], can be computed from
j=n
X2
C=M αj (M −1 K)j (9)
j=n1

where n1 and n2 can be positive or negative, as long as n1 < n2 . The coefficients αj are related to
the damping ratios, ζk , by
j=n
1 1 X2
ζk = αj ωk2j (10)
2 ωk j=n
1

Alternatively, a classical damping matrix can be computed for a specified set of modal damping
ratios ζj from the mass matrix and all N modal vectors and natural frequencies.

2ζ1 ωn1 /m∗1

 

C = M R̄ 
 ..  T
 R̄ M . (11)
.
2ζN ωnN /m∗N

The displacements r(t) of triple-diagonalizeable systems can always be expressed as a linear

combination of real-valued modal coordinates, q(t),

r(t) = r̄1 q1 (t) + r̄2 q2 (t) + · · · + r̄N qN (t) = R̄q(t). (12)

Substituting equation (12) into equation (1) and pre-multiplying by R̄T gives

R̄T M R̄q̈(t) + R̄T C R̄q̇(t) + R̄T K R̄q(t) = 0, (13)

or, for each mode, i, 1 ≤ i ≤ N ,

2
q̈j (t) + 2ζj ωnj q̇j (t) + ωnj qj (t) = 0, (14)

which are the N uncoupled equations of motion in modal coordinates. The damped free response
of each modal coordinate decays exponentially with time

qj (t) = e−ζj ωnj t (q̄cj cos ωdj t + q̄sj sin ωdj t), (15)

where ωdj is the j-th damped natural

q
frequency, is related to the j-th un-damped natural frequency
and damping ratio by ωdj = ωnj |1 − ζj2 |, and the coefficients q̄cj , q̄sj depend on the initial condi-
tions, the modal vectors, and the mass matrix.

CC BY-NC-ND H.P. Gavin October 23, 2018

Non-Classical Damping and Complex Modes 3

2 Non-Classical Damping
In general, the damping is not classical, R̄T C R̄ is not a diagonal matrix, and the natural
frequencies, damping ratios, and modal vectors depend on the mass, stiffness, and damping matrices
of the structural system. To determine the mode-shape vectors, natural frequencies, and damping
ratios from M , C, and K it is necessary to write the 2nd order differential equation (1) as two
sets of first order differential equations. Defining the velocity v(t) = ṙ(t), so that r̈(t) = v̇(t), and
solving equation (1) for r̈(t),
d
v(t) ≡ r̈(t) = −M −1 Kr(t) − M −1 C ṙ(t). (16)
dt
Re-writing these two sets of first order differential equations in matrix form,
( ) " #( )
d r(t) 0N ×N IN r(t)
= . (17)
dt v(t) −M −1 K −M −1 C v(t)

The 2N -by-2N matrix in the square brackets is called the dynamics matrix. Note that it is not
symmetric.

For any damped system (classically or non-classically damped) we must assume that the
free-vibration response decays with time,

r(t) = 2r̄r eσt cos(ωd t) − 2r̄i eσt sin(ωd t). (18)

All of the terms in equation (18) are real valued, however, it will be convenient to express this
equation in terms of complex values. We now introduce a complex mode shape vector r̄ = r̄r + ir̄i
and a complex modal coordinate.

q(t) = qr (t) + iqi (t) = eσt (cos(ωd t) + i sin(ωd t)), (19)

where r̄r and r̄i are the real and imaginary parts of r̄ and qr (t) and qi (t) are the real and imaginary
parts of q(t). With these new definitions, the trial function may be written compactly as

r(t) = r̄q(t) + r̄ ∗ q ∗ (t).

Note here that the subscripts “r” and “i” indicate real and imaginary and are not indices. Note also
that
eσt (cos(ωd t) + i sin(ωd t)) = eλt (20)
where λ = σ + iωd . So, the complex modal coordinate, q(t), can
p be written q(t) = eλt . The real
∗
part of λ equals −ζωn , the imaginary part of λ equals ωd = ωn |ζ − 1|, and λλ = ωn2 .
2

Re-writing and differentiating equation (18) to solve the first order differential equations (17),
∗t
r(t) = r̄eλt + r̄ ∗ eλ (21)
λt ∗ ∗ λ∗ t
v(t) = λr̄e + λ r̄ e , (22)

or ( ) " #( )
r(t) r̄ r̄ ∗ eλt
= ∗ , (23)
v(t) λr̄ λ∗ r̄ ∗ eλ t
and ( ) " #" #( )
d r(t) r̄ r̄ ∗ λ 0 eλt
= ∗ . (24)
dt v(t) λr̄ λ∗ r̄ ∗ 0 λ∗ eλ t

CC BY-NC-ND H.P. Gavin October 23, 2018

4 CEE 541. Structural Dynamics – Duke University – Fall 2018 – H.P. Gavin

Substituting equations (23) and (24) into the differential equations (17),
" #" #( ) " #" #( )
r̄ r̄ ∗ λ 0 eλt 0N ×N IN r̄ r̄ ∗ eλt
∗ = ∗ , (25)
λr̄ λ∗ r̄ ∗ 0 λ∗ eλ t −M −1 K −M −1 C λr̄ λ∗ r̄ ∗ eλ t

For this equation to be true for all time,

" #" # " #" #
r̄ r̄ ∗ λ 0 0N ×N IN r̄ r̄ ∗
= , (26)
λr̄ λ∗ r̄ ∗ 0 λ∗ −M K −M −1 C
−1 λr̄ λ∗ r̄ ∗

which represents a complex-conjugate pair of standard eigen-value problems:

" #( ) ( )
0N ×N IN r̄ r̄
= λ (27)
−M K −M −1 C
−1 λr̄ λr̄

and " #( ) ( )
0N ×N IN r̄ ∗ r̄ ∗
= λ∗ . (28)
−M −1 K −M −1 C λ∗ r̄ ∗ λ∗ r̄ ∗
The solution to one of these two standard eigen-value problems implies the solution to the other.

A relationship between the modal vectors found by solving the general eigen-value problem (3)
and the standard eigen-value problem (27) can be found by solving equation (27) for the un-damped
case (C = 0N ×N ):
" #!
−λIN IN  
det −1 = det λ2 IN + M −1 K = 0 (29)
−M K −λIN

Comparing this characteristic equation to the general eigen-value problem, it can be seen that
λ2 = −ωn 2 , or that λ = ±iωn . The eigen-vectors of this standard eigen-value problem for the
un-damped system, [r̄ T iωn r̄ T ]T , are directly related to the solution of the general eigen-value
problem. Recall that eigen-vectors may be arbitrarily scaled, and it is not uncommon for numerical
solutions to (27) to be scaled so that r̄ is imaginary and iωn r̄ is real. For the un-damped case, the
eigen-vectors can be more-intuitively scaled so that r̄ is purely real and iωn r̄ is purely imaginary.

The real modes arising from systems with zero or classical damping have nodes, which are
stationary points at which the structure has zero displacement. In contrast, for a complex modal
vector, r̄ = r̄r + ir̄i , there is not always a point on the structure at which the modal displacement
is zero at all times within a periodic cycle.

CC BY-NC-ND H.P. Gavin October 23, 2018

Non-Classical Damping and Complex Modes 5

3 Numerical Examples
The Matlab programs Cmodes3run.m, Cmodes3analysis.m, and N dof anim.m, may be used
to explore the modal characteristics of non-classically damped structures. These programs make
plots of the real and imaginary parts of the displacement modal vector, r̄, the modal phasors for
each degree of freedom, the real and imaginary parts of the displacement modal coordinates, q(t),
and the displacement responses of the coordinates of a three-degree-of-freedom building model, for
which,
     
m1 0 0 c1 + c2 −c2 0 k1 + k2 −k2 0
M =  0 m2 0  C =  −c2 c2 + c3 −c3  K =  −k2 k2 + k3 −k3 
     
0 0 m3 0 −c3 c3 0 −k3 k3
Values for the floor masses, mi , inter-story viscous damping rates, ci , inter-story stiffnesses, ki , and
displacement initial conditions, r(0), are specified in Cmodes3run.m. Running Cmodes3run.m results
in plots and an animation of the free response to the specified initial conditions.

In the .m-function Cmodes3analysis.m, each complex mode vector r̄j is scaled by a rotation
θj in the complex plane (via multiplication by the complex scalar e−iθj ) so that the real part of
the displacement the mode shape, Re(r̄), is maximized (and the imaginary part is minimized). For
this rotation, tan θj = Im(r̄jk )/Re(r̄jk ), where r̄jk = max |r̄j |. The magnitude of each mode is then
scaled so that the displacement parts of the modes are mass-normalized by dividing the real and
imaginary parts of r̄j and iωn r̄j by r̄jT ∗ M r̄j = IN .

When running Cmodes3run.m, you may try to:

1. Run a simulation with the as-provided default values for mi , ci , ki , and ro (mi = 1 tonne,
ci = [0, 3, 0] N/mm/s, ki = 1000 N/mm, roi = [1, −2, 3] mm). Observe how the real part
of mode j has j − 1 zero-crossings; how the free response of each modal displacement qj (t)
contains only a single frequency, the damped natural frequency, ωdj ; how all three modes are
damped even if there is damping in one story only; and how the free response of a higher-
frequency mode decays faster (in less time) than that of a lower-frequency mode, even if the
higher-frequency mode has slightly less damping.
2. Confirm that if C = 0 the modes are purely real (with the normalization implemented as
described above.)
3. Examine modal characteristics for systems with a Rayleigh damping matrix. For example by
setting ki = 1000 N/mm and ci = 2.0 N/mm/s, C is stiffness-proportional (C = 0.002K). Is
R̄ real or complex in this case?
4. Determine values of ci that will give approximately 5 percent damping in all three modes, for
mi = 1 tonne and ki = 1000 N/mm. This will involve some trial-and-error iteration on the
three values of ci . (hint: c1 > c2 > c3 ; 11 < c1 < 13 N/mm/s; and 2 < c2 < 4 kN/mm/s)
Are the resulting modes real or complex? Is there anything unusual or surprising about any
of the values of ci required to meet this goal? Does this finding imply a fallacy in the concept
of “damped real normal modes” with arbitrary modal damping ratios?
5. Set the initial displacement, ro = r(0), proportional to each of the three mode shape vectors,
and observe that the free response consists almost entirely of that mode. In Cmodes3run, if
you set roi = j, where j ∈ [1, 2, 3], ro will be set to r̄j . Next select some other set of initial
displacements and observe that the free response contains all three modes.

CC BY-NC-ND H.P. Gavin October 23, 2018

6 CEE 541. Structural Dynamics – Duke University – Fall 2018 – H.P. Gavin

6. The phasor matrix, Φ, of a complex modal matrix, R̄, is given by √ Φij = arctan(R̄iij /R̄rij )
(−π/2 < Φij < +π/2). How does multiplying a modal vector by −1 affect the associated
column of Φ? For a complex-valued mode, are values in the associated column of Φ equal to
one another? Why, or why not? The “complexity” of modal vector r̄j can be characterized by
Cj = maxi |Φij −Φ(i−1)j | Using the phasor plots generated by Cmodes3run.m with mi = 1 tonne
and ki = 1000 N/mm, find values of c1 , c2 , c3 that give a mode with a complexity greater than
about 30 degrees.

7. Explore the effects of changing the values of mass, damping, and stiffness. When changing a
value of mi , ci , ki , and roi , try to predict the effect of the change on the natural frequencies,
damping ratios, mode-shapes, modal responses, and floor responses; then use Cmodes3run.m
to check yourself.

(a) What happens if you increase a value of ci so that the damping of one of the modes
approaches 100 percent?
(b) What happens if a single value of ci is negative?
(c) What happens if a value of ci is so negative that one of the modal damping ratios becomes
slightly negative (≈ −0.50%)?
(d) What happens if one of the stiffness coefficients is much much larger than the other
coefficients?
(e) What happens if one of the stiffness coefficients is slightly negative?
(f) What happens if one of the mass coefficients is very negative?

References
[1] Caughey, T.K. “Classical Normal Modes in Damped Linear Dynamic Systems,” Journal of Applied Me-
chanics, 27(2)(1960): 269–271.
[2] Clough, Ray W., and Penzien, Joseph, Dynamics of Structures, 2nd ed. (revised), Computers and Struc-
tures, 2003.
[3] Lang, George Fox, “Demystifying Complex Modes,” Sound and Vibration, January 1989, pp 36-40.
[4] Liang Z., and Lee, G.C., “Damping of Structures: Part 1 - Theory of Complex Damping,” NCEER
technical report NCEER-91-0004, October 10, 1991.
[5] Plato, “The Allegory of the Cave,” Republic VII, 514 a,2 to 514 a,7 Translation by Thomas Sheehan
[6] Lord Rayleigh, Theory of Sound, Dover, 1945.
[7] Luco, Enrique J., “A note on classical damping matrices,” Earthquake Engineering and Structural Dy-
namics, 37 (2008): 615-626
[8] Tong, M., Liang Z., and Lee, G.C., “Physical Space Solutions of Non-Proportionally Damped Systems,”
NCEER technical report NCEER-91-0002, January 15, 1991.
[9] Woodhouse, J, “Linear Damping Models for Structural Vibration,” Journal of Sound and Vibration,
215(3) (1998): 547-569.

CC BY-NC-ND H.P. Gavin October 23, 2018