PHYSICAL REVIEW E, VOLUME 65, 036614

Increased damping of irregular resonators

S. Russ1,2 and B. Sapoval2,3
1

Institut fur Theoretische Physik III, Universitat Giessen, D-35392 Giessen, Germany
Laboratoire de Physique de la Matie`re Condensee, CNRS, Ecole Polytechnique, 91128 Palaiseau, France
3
Centre de Mathematiques et de leurs Applications, Ecole Normale Superieure, 94140 Cachan, France
Received 5 February 2001; revised manuscript received 3 October 2001; published 28 February 2002

It is shown that fractal drums and jagged geometry resonators may be more damped than ordinary Euclidean
systems. Several damping mechanisms are examined and studied by numerical calculations. The results depend
on the dissipation mechanisms but globally they increase with localization, frequency, and the irregularity of
the resonator. The increased dissipation is due to the uneven spatial distribution of the vibrational amplitude in
two different ways. First, it is related to the partial confinement of the vibrational modes. Secondly, increased
dissipation may be due to singularities in the amplitude distribution. This is the case when a few points exist
where the vibration is pinned to zero inducing local logarithmic singularities. This last effect can be spectacular: a single defect can dominate the surface damping by viscous forces of a square drum.
DOI: 10.1103/PhysRevE.65.036614

PACS numbers: 43.20.g, 47.53.n, 46.40.Ff

I. INTRODUCTION

Objects with irregular geometry are ubiquitous in nature

and their vibrational properties are of general interest. How
do trees respond to wind? How do sea waves depend on the
topography or geometrical structure of the coasts and breakwaters? How to explain the vibrational properties of glasses?
All these questions remain largely unanswered. On the other
hand, current knowledge of waves and resonators indicates
that even small perturbations of a resonator geometry may
strongly increase the damping of specific modes. A small
change of the boundary is often used in musical instrument
manufacturing and microwave technology to prevent the existence of spurious modes which are effectively damped by
the choice of a suitable defect in the geometry.
This paper deals with the damping properties of resonators with irregular shapes or resonators with point defects.
Our general goal is to understand which of the vibrational
properties influence damping, what are the reasons, and how
they are related to the resonator geometries. For this purpose,
we examine several representative geometries and several
damping mechanisms. Between these geometries, special attention will be given to fractal drums, as the emergence of
fractal geometry has been a significant breakthrough in the
description of strong geometrical irregularity 1. For simplicity we decided to study scalar vibrations instead of vector
vibrations like in electromagnetic cavities. Fractals permit us
in many cases to describe approximately strong statistical
irregularity. They also permit the study of physical properties
of deterministic but very irregular objects 2. In this paper,
we deal with damping in fractal and nonfractal types of geometrical irregularities. The principal result is that damping is
directly influenced by localization and, if present, by strong
singularities of the amplitude distribution.
The three main properties of vibrating systems are their
spectrum, the amplitude distribution of the harmonic excitations, and their damping, the damping being related to the
first two quantities. The effect of the resonator geometry on
the damping has been rarely addressed, although there exists
a wide body of empirical knowledge on how to build good
1063-651X/2002/653/03661410/$20.00

resonators. A good resonator is a resonator that exhibits a

large quality factor. The quality factor Q n for the nth resonance mode at frequency n is the dimensionless ratio of the
reactive energy E n to the loss per cycle P n ,
Q n

2En
.
Pn

Its value characterizes the ability of the resonator, driven

by an oscillating force of frequency n , to accumulate reactive energy for a given power input P n . It also determines
the lifetime Q n / n of the oscillation when no power source
is present. The general question is: do geometrical irregularities play a role in the losses and if so, why?
It has been suggested that there may exist an increase in
the damping of fractal resonators due to the irregular distribution in space of the vibrational amplitude 3. There exist
two types of fractal resonators. The first are mass fractal
resonators 4 6, for which increased damping has been
shown numerically for the so-called fracton modes in percolation clusters 6. A second type is surface fractals such as,
e.g., fractal drums or fractal cavities. To prevent the use of
the term fractons in this case, their eigenmodes have been
named fractinos 7. The system is generally called a
drum but a real drum is a more complex physical system
which possesses two membranes and an air column between
them. Here, we consider a tambourine which is an instrument
with only one membrane. The increased damping of fractal
acoustical cavities has been predicted by numerical computations and recently been confirmed by experiments 8.
Mathematical aspects of the study of fractal drums have been
previously reported 912. To our knowledge, no study has
been made about the comparison of damping between fractal
and nonfractal but irregular drums. The drum geometries
are shown in Fig. 1. They range from systems with fractal
and nonfractal boundary roughness to square systems with
point defects. Fractal drums are created, starting from a
square membrane not shown in the figure. By applying the
fractal generator shown in Fig. 1a to each side of the
square, one obtains the fractal drum of first generation. By

65 036614-1

2002 The American Physical Society

S. RUSS AND B. SAPOVAL

PHYSICAL REVIEW E 65 036614

II. FUNCTIONAL DEPENDENCE OF DAMPING
MECHANISMS

This study is restricted to the limiting situation of weak

losses. In this case the eigenmodes and eigenvalues can be
reasonably approximated by those of the corresponding zero
loss system. The time-dependent vibrational elongation of a
given eigenstate is z n (x,y,t)A n n (x,y)cos(nt) where A n
is an arbitrary prefactor of the vibrational amplitude. The
eigenstate n (x,y) is the normalized amplitude of the eigenmode of frequency n which satisfies the Helmholtz eigenvalue equation:
n

2n
c2

n ,

where c is the sound velocity of the membrane.

Calling the surface density of the membrane, the values
of the maximum elastic energy or kinetic energy E n are
E n A 2n
FIG. 1. a Generator for the prefractal geometry. b and c
Fractal drum of first and second generation. The area is conserved
through the iteration process. The fractal dimension of the perimeter
is D f ln 8/ln 43/2. d Rough structure nonscaling surface irregularity. e Example of the square system with two pinned
points.

applying the fractal generator again to each straight line of

the boundary structure, one can create fractal drums of arbitrary generations. In this paper, fractal drums up to the third
generation are considered the last one not being shown.
Figures 1b and 1c show fractal drums of first and second
generation. Figure 1d shows a drum with a rough, but nonscaling, boundary. The comparison of this system with the
fractal drums permits us to understand the respective role of
mechanical screening in hierarchical and nonhierarchical geometries. Figure 1e shows an example of a square membrane with point defects. In these last systems, the boundary
is smooth, but the wave amplitude can develop strong singularities. In all cases we apply Dirichlet boundary conditions,
which means that the vibrational amplitude is maintained at
zero on the perimeter of the drum, as well as on the pinned
points, shown in Fig. 1e.
Several possible damping mechanisms are introduced in
Sec. II. In Sec. III the vibrational amplitude distribution of
the fundamental mode and of the fourth excited mode are
described. The damping of uniform membranes is studied in
Sec. IV. In Sec. V we investigate the role of the localization
for the same systems. In Sec. VI we examine the case of a
nonuniform membrane for which the damping force exists
only at the boundary of the resonator. This can be realized by
using an inhomogeneous membrane which presents large internal viscosity only at the drum periphery. Finally, in Sec.
VI we study the case of a square drum with a few point
defects that pin the amplitude. Contrary to the fractal drums,
it is the logarithmic singularity of the amplitude distribution
that creates efficient damping in this case.

A 2n 2n
2n 2
n x,y
.
2
2

dxdy

We use the normalization condition dxdy n (x,y) 2 1.

The loss per cycle is equal to the work F d (x,y)dz over one
period of the damping force F d (x,y)
P n

dt

period

drum

dz n
dF d .
dt

Different situations exist, depending on the nature of the

damping forces and hence on their spatial distributions. The
damping force on an element dxdy will be a function of
velocity and shape of the membrane under motion. The
shape dependence of the damping force can be generally
expressed as a function of various space derivatives of the
vibrational amplitude:

dF d dxdy K 1

K 3 d

d 2z n
dx 2

dz n
dz n
K 2
dt
dt

d 2z n
dy 2

/dt ].

The first term characterizes a damping force proportional

to the local velocity of the membrane. It describes an artificial situation where a massive membrane is linked to many
individual dashpots with negligible inertia. Additionally,
such a term would participate in radiation damping. A crude
and insufficient approximation for this is acoustic radiation
in a fluid, where each surface element dxdy behaves as a
piston. The problem of acoustic radiation in air is extremely
complex for two reasons. First, if the membrane is light, its
motion is strongly coupled to the air and the system does not
obey a simple Helmholtz equation 13. Second, even for a
massive membrane, one should consider not only the socalled radiation monopole, but also the various multipolar
radiations 14. This is possibly important for this study,
where the amplitudes exhibit singularities on every wedge of

036614-2

INCREASED DAMPING OF IRREGULAR RESONATORS

PHYSICAL REVIEW E 65 036614

the boundary. For simplicity we do not consider these complex effects, assuming that they can be neglected as compared to the other damping mechanisms. Then, the force
on an element dxdy takes the simple form dF 1
dxdyK 1 (dz n /dt) and the value of P n is P 1,n
A 2n K 1 n .
The second term is proportional to the gradient of the
local membrane velocity. It corresponds, for example, to
damping due to the viscosity of an embedding fluid 6. If
the fluid above and below the membrane is pushed by the
membranes displacement, its vertical velocity will depend
on x and y. There will exist a gradient in the fluid velocity
and an associated viscous force and dissipation. Here the
damping is due to a vertical viscous force of the form dF d
dxdyK 2 (dz n /dt) dxdyA n K 2 n sin(nt)n(x,y).
The amplitude gradient n (x,y) is the strain of the membrane. The energy loss per cycle in this case is P 2,n
A 2n K 2 n dxdy n n . The damping depends on the
spatial distribution of the vibration and strain and thus on the
geometry of the drum.
The third term describes damping due to internal viscosity
if the membrane is viscoelastic. Owing to its finite thickness,
the upper part of the membrane is slightly more stretched
than the lower part. In this case the upper layer moves relatively to the lower layer. As these layers slide, there exists a
viscous force which dissipates energy. The relative horizontal displacement is proportional to the curvature d 2 z n /dx 2
d 2 z n /dy 2 z . It is also proportional to the membrane
and thickness 15. The energy loss per cycle in this case is
P 3,n A 2n K 3 n

dxdy n

d 2 n
dx 2

d 2 n
dy 2

. 7

There is a fourth interesting situation where the damping

force is nonlinear. We consider here a damping force proportional to the square of the velocity gradient dF d
dxdyK NL (dz n /dt)2 . In that case the dissipated
power can be written P NL,n A 3n K NL 2n dxdy
n ( n ) 2 . It will be proportional to the third power of
the amplitude and the quality factor will decrease when the
amplitude increases. As shown below, this damping is directly related to localization as in the case of linear damping
in fractal acoustical cavities.
The quality factors corresponding to these different cases
can be written as with dVdx dy)
Q 1,n

Q 2,n

Q 3,n

K 1 n

dV 2n

K 2 n

dV n n
K 3 n

Q NL,n

11

III. NUMERICAL CALCULATIONS OF THE EIGENMODES

In the following sections we compare the quality factors

for the different geometries of Fig. 1 and for the different
damping mechanisms from Eqs. 811. To the latter we
refer to as case 1 until case 4damping. For the calculations, it is necessary to know the frequency spectrum n
and the spatial distribution of the modes amplitudes n .
These computations have been performed using the exact
correspondence between the Helmholtz and diffusion equations described in 7.
Wave form singularities are expected to appear at the irregular resonator boundary. The singularities create a local
augmentation of the spatial derivatives and consequently a
corresponding local increase of the energy losses along the
boundary 4,13. Modes may be singular near the wedges of
the boundary, i.e., their derivatives are infinite at particular
points on the surface geometry. Consider, for example, a region of the membrane around a salient corner, i.e., a corner with an opening angle of 3 /2 3. Close to the boundary, the amplitude of the vibration is very small and is
close to 0. In this case, using polar coordinates (r, ) around
the corner, the solution of the Laplacian with Dirichlet
boundary conditions is approximately of the form

r
r0

2/3

sin 2 /3 ,

12

where r 0 is of the order of the local small scale of the irregular geometry. For a fractal drum r 0 is of the order of the
small cut-off scale of the fractal. The gradient / r
r 1/3 tends to infinity when r tends to 0 3. This corresponds to a local infinite stress and strain of the membrane.
Such a property should be true around every salient point in
the structure.
Note that real physical objects only present rounded
wedges. Around these wedges, the derivative does not tend
to infinity but to a large finite value proportional to the inverse of the curvature radius of the contour.
Due to the existence of singularities, precise eigenfunctions are required. The method described in 7 was chosen
because it allows for the large spatial resolution required for
the study of fractal resonators. The computation is made on a
discretized square grid with lattice distance a. The 200300
lower modes have been computed and are studied here. The
states are given by their numerical values at sites i, j of the
square grid, normalized by the relation

, 10
2

dV n n

.
2

represent damping
The constants K 1 , K 2 , K 3 , and K NL
factors which depend on the specific system and its material
properties. For simplicity, they are assumed to be frequency
independent.

a2

dV n d n /dx d n /dy
2

A 1
K NL
n

036614-3

2n i, j 1.

i, j

13

S. RUSS AND B. SAPOVAL

PHYSICAL REVIEW E 65 036614

FIG. 2. Fundamental mode n1. Top: Distribution of the amplitude. Bottom: Distribution of the absolute value of the strain or
amplitude gradient. One observes strain peaks at the salient edges,
but the edges situated in the bays are partially screened and see
smaller strains. Courtesy of J. F. Colonna.

As indicated by Eqs. 811, the Q values depend directly on the spatial distribution of the eigenfunctions and
their gradients. The vibration distribution is shown for states
n1 and n4 in Figs. 2 and 3. One can observe that the
modes are singular near the wedges of the boundary, i.e.,
their derivatives are infinite at particular points on the surface. The fundamental state is localized in the large central
region of the drum. It decays very rapidly when entering
narrow regions 7. Second, the absolute value of the gradient is smaller in narrower regions. This is caused by the
decrease of the amplitude itself. The decrease of the gradient
corresponds to a screening effect analogous to Laplacian
screening. For Laplacian screening, the region with large Laplacian fields here n ) has a dimension equal to 1, inde-

FIG. 3. Mode n4. Top: amplitude distribution. Bottom: distribution of the absolute value of the strain or amplitude gradient. At
higher frequency, because of the smaller wavelength, the modes
penetrate more deeply into the narrower regions close to the boundary and the strain peaks are more uniformly distributed. Courtesy of
J. F. Colonna.

pendently of the geometry 16 18. This means that large

strains are distributed over a region of total size on the order
of the system size, independently of structure.
For state n4 the same properties are found. However,
due to the higher frequency and shorter wavelength, the
mode penetrates into narrower regions, allowing for a better
exploration of the geometry. If the damping really depends
on wedge singularities, it should increase at high frequency.
IV. UNIFORMLY DISTRIBUTED DAMPING: RESULTS
AND DISCUSSION

The quality factors are computed using the discretized

expressions of the integrals A1A3 given in the Appen-

036614-4

INCREASED DAMPING OF IRREGULAR RESONATORS

PHYSICAL REVIEW E 65 036614

FIG. 4. Q 2 in arbitrary units for homogenous linear damping of

type 2 in different geometries plotted versus the normalized frequencies / 0 . The symbols represent, respectively, squares for the
square initiator, empty triangles () for the fractal drum of first
generation, empty triangles () for the generation 2, filled triangles
() for generation 3, and circles for the rough structure. There
exists an effect of the geometry but this effect is weak and does not
change the order of magnitude of the Q factor.

dix. Herein, the numerical results correspond to the same

discretization grid and can thus be compared. Consider a
square lattice of side L, with fundamental mode frequency
0 2 c/L. The area of the membrane is conserved by
applying the fractal generator, which allows us to compare
fractal drums of different geometries but with the same surface L 2 . In the following, all frequencies are given by the
size-independent normalized frequencies n / 0 . This enables us to discuss the results on a single frequency-scale.
The different types of viscous damping behave in the following way.
Q 1,n is equal to n K 1 , independently of the spatial distribution of the vibration. In this case the quality factors depend on the frequency but are independent of the shape.
Therefore, Q 1,n is the same for vibrations in Euclidean and
fractal systems apart from the fact that the spectrum n is
different. Note that the same result would be obtained for
the quality factor of an irregular superconducting microwave
cavity in the case where damping is only due to dielectric
losses in the volume of the cavity.
The values Q 3,n are obtained from Eqs. 2, 10, and 13.
One obtains Q 3,n K 3 c 2 ( n dxdy 2n ) 1 K 3 c 2 / n ,
also irrespective to the drum shape. The fact that Q 1,n and
Q 3,n are independent of the shape has been verified numerically.
The cases of interest are therefore case 2 where the energy losses are due to the viscosity of the embedding fluid
and case 4 nonlinear damping. Case 2 is computed using
expressions 9 and A1 see the Appendix. The results are
shown in Fig. 4. One can see that the Q 2 factors are modified
by the irregularity of the drum but the effect remains small.
This is due to the fact that the local losses are proportional to
the product of the amplitude gradient with the amplitude

itself. Around a wedge, the losses behave as

(r, )(r, )r 2/3r 1/3r 1/3, which means that although the strain is singular, the losses are not. Therefore, the
existence of wedges in the structure does not influence the
quality factors Q 2 significantly. In order to increase the
damping it would be useful to prevent screening by keeping
singular wedges exposed to a large strain. It is for that purpose that the rough drum has been studied open circles in
Fig. 4. Indeed, one observes that a nonscaling or regular
irregular structure is more efficient in decreasing the Q 2
factor than the self-similar geometry. The effect, however,
remains small.
Therefore, it is necessary to check the numerical errors of
our calculations. This can be done by comparing the numerical with the theoretical Q factor which can be computed
analytically for a particular case. The fractal drum of generation is constituted by a collection of joint identical squares
of size L/4 . Therefore, there exists a number of trivial
eigenfunctions of the form (x,y)sin(22mx/L)
sin(22my/L) with m,m 1,2,3, . . . . For the state
(m,m )(1,1) of the first-generation fractal drum, we find
the numerical and analytical quality factors to be Q (num.)
17.755 86 and Q (anal.) 17.756 97. This corresponds to a
numerical error of only 0.006%, far smaller than the effects
discussed here, of the order of tens %. Note that, because of
the finite discretization of the mesh, the wedge strain is finite. In first approximation, it corresponds to the real boundary gradient if the wedges were rounded to the scale of the
mesh, here L/128, as explained in the preceding discussion.
The nonlinear viscous damping case 4 is computed using expressions 11 and A3 see Appendix and the results
are shown in Fig. 5. The factor Q NL is now strongly modified
by the contour irregularity note the logarithmic scale for
Q NL). Specifically, higher-order drums possess many modes
with very small Q NL values. One also observes that modes
with close frequencies may have very different quality factors spreading over nearly one order of magnitude. The differences between the second- and third-order drum are not
significant in the frequency range under study.
The fact that Q 2 and Q NL show very different frequency
dependencies as well as the dispersion of Q NL values can be
understood by considering the spatial dependence of the amplitude and the effect of mode localization. This is discussed
in the next section.
V. ROLE OF THE LOCALIZATION AND FREQUENCY
DEPENDENCE OF THE QUALITY FACTOR

Localization has already been found both numerically and

experimentally in fractal acoustical cavities and in fractal
drums 8,18. The localization of a mode n is usually characterized by the value of its localization volume defined by
V n

14

4n x,y dxdy

where n is normalized in the resonator volume according

036614-5

S. RUSS AND B. SAPOVAL

PHYSICAL REVIEW E 65 036614

FIG. 5. Q NL in arbitrary units for homogenous nonlinear damping in different geometries plotted versus the normalized frequencies / 0 . The symbols represent, respectively, squares for the
square initiator, empty triangles () for the fractal drum of first
generation, empty triangles () for the generation 2, filled triangles
() for generation 3, and circles for the rough structure. The same
symbols as in Fig. 4. There exists a strong effect of the frequency
and of the geometry. One observes that in the same frequency range
the fractal drums exhibit a large dispersion of the Q factors. This is
due to very different localization effects, as shown in Sec. IV.

to Eq. 13. The relative localization volume V n /L 2 or participation ratio measures the relative volume occupied by
this mode.
Figure 6 shows the different localization volumes V n /L 2
computed from the amplitude distributions. The states are

indexed by their normalized frequencies / 0 and the different symbols refer to different geometries. The cloud of
points in this figure are indicative of the wide dispersion of
the participation ratios, first, between different systems and
second, also for a given system. Only for the square drum,
where the eigenstates are delocalized sine functions, V n /L 2
4/90.44 for all states. The relative localization volumes
of the second triangles down and the third-order drum
filled triangles are distributed between 0.05 and 0.35. These
values reflect a very different behavior of the corresponding
modes. The upper values are not very distinct from the value
of a regular square drum and accordingly, the respective
wave functions are more or less distributed over the whole
system. The modes that are significantly more confined than
V n /L 2 0.3, are called localized. Most interesting are the
modes with values of less than 0.1. We can see in the figure
that there are several of them that occur in branches around
special frequencies. Examples are the modes around / 0
10 and 14 for the second-order drum. It is exactly those
modes which show the smallest Q NL factors in Fig. 5. A
different behavior is found for the rough drum circles. Its
localization volumes above 0.3) indicate that its vibrations
are not localized, as expected as this resonator possesses no
screened regions. The following discussion permits us to relate in a formal manner the observed dispersion in Fig. 5 to
the observed dispersion in Fig. 6 linking directly damping
and localization.
Using V n one can obtain a rough estimate of the quality
factors by distinguishing between regions of large and small
amplitude. We consider the regions of large amplitude and
define an average absolute amplitude by n
V 1
n dxdy n . For a localized mode, assuming that the
amplitude is approximately zero outside its localization volume, one can write the normalization condition as n 2
V n 1 or

n V n 1/2.

15

We now consider the different types of damping separately,

according to cases 2 and 4. At frequency n the characteristic distance for the amplitude space variation is a halfwavelength n /2 c/ n . The order of magnitude of the
n / c, which leads to a
gradient is 2 n / n V 1/2
n
Q 2,n factor of
Q 2,n n

FIG. 6. Dispersion of the relative localization volume V n /L 2 for

the resonators shown in Fig. 1. The states are indexed by their
normalized frequencies / 0 for all systems. The constant value
V n /L 2 4/9 for the square drum is indicated by the dashed line. The
symbols correspond respectively: empty triangles () to the fractal
drum of first generation, empty triangles () to the generation 2,
filled triangles () to generation 3, and circles to the rough structure.

dV V n 0 n 0 .

16

This semiquantitative prediction means that Q 2,n is essentially independent of frequency and localization, which is
compatible with the data in Fig. 2 where we found that the
effects of geometry are relatively small.
The situation is very different for the nonlinear damping,
where we find by the same estimations that Q NL,n is of the
form

036614-6

Q NL,n 1

dV n 2 V n 1/2 n 2 . 17

INCREASED DAMPING OF IRREGULAR RESONATORS

PHYSICAL REVIEW E 65 036614

FIG. 7. Q NL in arbitrary units versus (V n /L ) for all systems, including the square generator. Same symbols as in Fig. 4.
Note that the vertical scale is linear whereas the scale on Fig. 5 is
logarithmic.
2

2 1/2

The localization argument indicates that Q NL should be

markedly smaller for localized states and high frequencies.
This is verified in Fig. 7 where the relation 11 is compared
to the numerical results. The values of Q NL 2 are plotted
versus (V n /L 2 ) 1/2. The fact that for all drums and all states
the points fall onto the same universal curve illustrates that
damping is directly influenced by localization and that the
simple semiquantitative reasoning described above applies to
irregular drums, independent of their shape. Note the linear
scale in the figure.
For all these cases, the divergences of the strain at salient
points do not contribute significantly to the dissipation. Even
for the nonlinear case, the local strain diverges as / r
r 1/3 and the integrand is of the form r 1 which is regular.
Again, damping due to uniformly distributed mechanisms is
not dominated by strain divergences.

VI. BOUNDARY DAMPING

In the last two sections we study cases where the dissipation mechanism is not distributed uniformly over the membrane. In the first case of practical interest, only the periphery
of the drum presents internal viscosity case 3. The membrane is purely elastic in the drum interior and viscoelastic
along the boundary of the resonator. This situation is of practical interest when one wishes to dampen spurious high frequency resonances of membranes. It is linear damping described by Eq. A2 but the integration has to be performed
only over a small layer along the boundary. We call the quality factor Q b3 .
The power P 3,n dissipated along the boundary can be
written from Eq. 7 as

P 3,n A 2n K 3 n

boundary

n a n a ,

18

FIG. 8. The quality factor Q b3 in arbitrary units versus ( n / 0 ) 3

for the boundary linear damping of type 3 with the same symbols as
in Fig. 4. The line of slope 1 is a guide to the eye. For the
second-order drum, the states with frequencies around / 0 10
and / 0 14, showing strong localization and very small quality
factors, are indicated by arrows.

where n (a) and n (a) are the respective values of the

amplitude and of the Laplacian on mesh sites next to the
boundary, i.e., at a distance a of one lattice constant. The
data corresponding to the fractal drum of third generation are
not given due to the lack of resolution of the smaller cutoff in
the grid used for the computation only three vibrating sites.
The other results are shown in Fig. 8, where Q b3 is plotted
versus ( / 0 ) 3 . One observes i a strong decrease with frequency following a power law with exponent -3, ii a
gradual decrease of the quality factors as one moves from
regular to more and more irregular drums, and iii a strong
effect of the localization. So, the quality factors for the
square and the rough drums, which show no localization, lie
clearly above those of the fractal drums, which have many
localized states. Additionally, for the second generation,
those states with frequencies around / 0 10 and / 0
14, showing strong localization, have the smallest quality
factors indicated by arrows in Fig. 8.
The general behavior of the results may be understood as
follows: with n 2n n , one expects that
Q b3 1

boundary

2n a ,

19

where the sum runs over all boundary sites. The value of the
sum in Eq. 19 depends on the amplitude distribution and
the localization close to the boundary. Note that the square
and the rough generator show no localization. If the amplitudes behave approximately as sine waves, the amplitude
value next to the boundary is proportional to (a/) and thus
to n . It is also proportional to the amplitude factor
n (V n ) 1/2 of Eq. 15 which depends on localization.
n 2 ) 1 (V n )
This
means
that
Q b3 ( n n,b
3
1
( n n,b 1) , where the prime denotes the active bound runs over all boundary sites in the
ary and the sum n,b
localization regions of mode n.

036614-7

S. RUSS AND B. SAPOVAL

PHYSICAL REVIEW E 65 036614

FIG. 9. Fundamental modes of square systems with a one, b

two, and c 20 pinned points. The amplitudes are indicated by
different gray levels. The white regions stand for very large amplitudes. The darker tone stands for nearly zero amplitude. Node lines
do not exist in the fundamental mode.

This argument, which is valid for all the structures, explains the frequency power law. The interpretation of the
dependence of Q b3 on the localization, on the other hand, is
difficult, because Q b3 depends both on the localization volume and on the set of perimeter sites which are active in
the localization volume. Also, the local irregularity around
wedges could play a role as its relative contribution to damping effects could be larger than in the case of homogeneous
damping. To disentangle these various factors would require
the computation of the eigenmodes of fractal drums of higher
generations with sufficient resolution. This is currently not
possible.
In any case, if one wishes to increase the damping by the
singular behavior of the strain, one should search for a stronger divergence around salient points. This is found around
needles where behaves as r 1/2 and therefore as r 1/2.
Even in that extreme case the integrands for the various integrals mentioned above are regular and the singularities of
the geometry play only a small role. In contrast, the existence of pinned points inside a regular drum will create logarithmic singularities which might dominate the damping.
This is shown in the next section.
VII. DAMPING AROUND PINNED POINTS

We discuss here the damping of a regular square drum on

which a certain number of inner points are pinned. One example is shown in Fig. 1e. The vibrational amplitude vanishes at these points and the membrane is viscoelastic along
the boundary as well as around the pinned points. We consider three different cases, whose fundamental modes are
shown in Figs. 9a9c. It can be seen that around pinned
points, there are large gradients of the vibrational amplitudes.
The choice of the positions for the pinned points was guided
by the effort to avoid the natural nodelines caused by the
symmetry of the square.
In case a, only one point is kept fixed at a position close
to (x,y)(L/R,LL/R), where R is the golden mean. The
use of the golden mean is in order to place the pinned point
at the most antiharmonic location with regard to the horizontal and the vertical symmetry axes of the membrane. R is
most distant from a rational number and so no nodelines of
the sine functions sin(nx/L) can occur close to xRL or
x(1R)L. However, this point lies close to a diagonal

p
FIG. 10. Q b,
3 in arbitrary units for square membranes with different numbers of pinned points, plotted versus the normalized frequencies / 0 . The different symbols represent the normal square
without pinned points open squares and the systems of Fig. 10: a
full squares, b diamonds, and c open circles. Inset: The wave
function (r) is plotted against the horizontal distance r from a
pinned point filled circles and from the boundary open circles
for the system of Fig. 9a. It can be recognized that (r)log r for
r0 around the pinned point and (r)sin rr at small distances
from the boundary.

which is also a symmetry axis. Eigenstates must be either

symmetric or antisymmetric under reflexion on the diagonal.
The antisymmetric modes of the square have a zero value on
the diagonal. These modes are unperturbed by the existence
of a pinned point on this diagonal. In order to suppress any
symmetry in case b two pinned points were introduced at
the positions (x,y)(L/R,LL/R) and (x,y)L/4
L/(2R),LL/(2R). In case c 20 randomly chosen
points are pinned.
Now the stronger singularities in the amplitudes do
modify the damping. Very close to a fixed point, the solution
of the Laplace equation shows logarithmic singularities,
(r)log r, and one expects large contributions to damping
at small r. In that case the sum in Eq. 19 contains large
(log a)2 terms at the denominator. The numerical results are
shown in Fig. 10. The figure contains the data for the normal
square without pinned points and for the systems of Figs.
9a9c. In the inset, the behavior of the vibrationonal amplitudes (r) are shown along the horizontal distances r
from the pinned point of system 9a and from the boundary.
The singularity around the pinned point is observed. Close to
the regular boundary, on the other hand, the amplitude behaves linearly as sin rr. The value of (r) at a distance of
one lattice constant from a pinned point is about 2030 times
larger than at the same distance from the boundary. Consequently, in a square lattice of side length L256a, one
single pinned point is about five times more efficient for the
damping than all the approximately 1000 boundary points
altogether. This estimation contains the difference in 2 and
the number of neighbors around the pinned point and would

036614-8

INCREASED DAMPING OF IRREGULAR RESONATORS

PHYSICAL REVIEW E 65 036614

be even larger if we considered viscoelasticity up to secondnearest neighbors.

As one observes in Fig. 10, the quality factors decrease
drastically due to the existence of pinned points. As expected, for one pinned point on the diagonal Fig. 9a the
quality factor for the fundamental is decreased by a factor of
about 5. It is a most remarkable result that a single point
defect can dramatically increase the damping. It has to be put
in relation with a common technique in radar technology,
where a small wire is introduced into a microwave cavity at
a maximum of the electric field of an eigenmode in order to
kill that mode. Note that damping in microwave cavities is
also a surface damping due to the classical skin effect.
However, the modes that are antisymmetric with respect
to the diagonal are unperturbed and show the same quality
factors as the square. The symmetry effect is suppressed in
the systems with several pinned points. In case b or c
there are always several pinned points close to amplitudes
maximaat least at low frequencies cf. Fig. 10c. Therefore, in the entire low-frequency regime, the quality factors
are strongly diminished as compared to the boundary
Q b,p
3
damping in the normal square. With increasing frequency,
the relative effect becomes smaller, because modes become
more and more confined between defects. At the same time
the frequency spectrum shifts to higher values. Contrary to
the results of the fractal drums, the states of these membranes are not strongly localized. The participation ratios of
these modes are found to be V n /L 2 0.3, much larger than
for the fractal drums. Therefore the observed damping is, in
this case, a direct consequence of the singular behavior of the
vibration amplitudes. Not shown here, we have computed the
damping of the fundamental state for one single pinned point
with coordinates (x,L/2) as a function of x. A first order
perturbation theory correction would predict that the losses
go with sin2(x/L) attached to the unperturbed state. This is
approximately found.
VIII. CONCLUSION

In summary it has been shown that irregular shapes not

only drastically alter the spatial character of drum vibrations
but may also increase their damping. This has been investigated for several geometries with scaling and nonscaling
surface roughness and with pinned point defects and for
different damping mechanisms. The general conclusion is
that irregularity increases the effective damping of the vibration. It has been found that there are two different reasons for
these effects: localization and the existence of algebraic singularities of the spatial distribution of the modes amplitudes.
For drums without point defects the damping is related
directly to localization as a localized vibration presents larger
amplitude and velocity gradients and therefore increased dissipative viscous forces. On the other hand, the existence of a
few pinned points on an otherwise regular drum can increase
the damping dramatically. This is due to the existence of

logarithmic singularities in the viscoelastic behavior around

pinned points. For example, it has been found that only 2
pinned points suitably placed are significantly more efficient
in damping than some 1000 points along the smooth boundary. This last damping phenomena is not related to localization, but to the strong singularity of the vibrational amplitude. This is the first quantitative explanation of why point
defects, such as those used in radar technology, really permit
the kill of unwanted spurious modes.
ACKNOWLEDGMENTS

We would like to thank Jean-Francois Colonna for providing us with Figs. 2 and 3 and Markus Dejmek and Armin
Bunde for a careful reading of the manuscript and for interesting remarks. One of us S.R. has benefited from the
E.E.C. program Human Capital and Mobility. The computation was performed at the Institut du Developpement et
des Ressources en Informatique Scientifique IDRIS in Orsay, France. CNRS is unite mixte du CNRS No. 1254.
APPENDIX

With the notations of the discretized lattice the integrals in

Eq. 9 are substituted by the expression

dxdy n n

a 2 /2

n i, j n i1,j n i, j

i, j

n i1,j n i, j n i, j1
n i, j n i, j1 n i, j .

A1

In the same way the integrals in Eqs. 10 and 11 are expressed on the discretization grid as

dxdy n d 2 n /dx 2 d 2 n /dy 2

a 2

n i, j n i1,j n i1,j 2 n i, j

i, j

n i, j1 n i, j1 2 n i, j ,

036614-9

dxdy n n 2 a 2 /2

A2

n i, j n i1,j

i, j

n i, j 2 n i1,j
n i, j 2 n i, j1
n i, j 2 n i, j1
n i, j 2 .

A3

S. RUSS AND B. SAPOVAL

PHYSICAL REVIEW E 65 036614

1 B. B. Mandelbrot, The Fractal Geometry of Nature Freeman,

San Francisco, 1982.
2 B. Sapoval, Fractals Aditech, Paris, 1990; Universalites et
Fractales Flammarion, Paris, 1997.
3 B. Sapoval, Physica D 38, 296 1989.
4 S. Alexander and R. Orbach, J. Phys. France Lett. 43, L625
1982.
5 T. Nakayama, K. Yakubo, and R. Orbach, Rev. Mod. Phys. 66,
381 1994.
6 S. Russ and B. Sapoval, Phys. Rev. Lett. 73, 1570 1994; in
Dynamics in Small Confining Systems, edited by J. M. Drake,
S. M. Troian, J. Klafter, and R. Kopelman, MRS Symposia
Proceedings No. 366 Materials Research Society, Pittsburgh,
1995, p. 243; Nuovo Cimento D 16, 1103 1994.
7 B. Sapoval, T. Gobron, and A. Margolina, Phys. Rev. Lett. 67,
2974 1991; B. Sapoval and T. Gobron, Phys. Rev. E 47, 3013
1993; S. Russ, B. Sapoval, and O. Haeberle, ibid. 55, 1413
1997.
8 B. Sapoval, O. Haeberle, and S. Russ, J. Acoust. Soc. Am. 102,
2014 1997; B. Hebert, B. Sapoval, and S. Russ, ibid. 105,
1567 1999.

9 M. L. Lapidus and M. Pang, Commun. Math. Phys. 172, 359

1995.
10 M. Lapidus, J. W. Neuberger, R. Renka, and C. A. Griffith, Int.
J. Bifurcation Chaos Appl. Sci. Eng. 6, 1185 1996.
11 M. Levitin and D. Vassiliev, Proc. London Math. Soc. 72, 178
1996.
12 J. Fleckinger, M. Levitin, and D. Vassiliev, Proc. London
Math. Soc. 71, 372 1995.
13 G. A. Kriegmann and C. S. Scandrett, J. Acoust. Soc. Am. 86,
788 1989.
14 P. M. Morse and K. Uno Ingard, Theoretical Acoustics Princeton University Press, Princeton, N.J., 1968.
15 P. G. de Gennes and J. Prost, The Physics of Liquid Crystals
Clarendon Press, Oxford, 1995, p. 418.
16 N. G. Makarov, Proc. London Math. Soc. 51, 369 1985.
17 P. Jones and T. Wolff, Acta Math. 161, 131 1988.
18 B. Sapoval, M. Filoche, K. Karamanos, and R. Brizzi, Eur.
Phys. J. B 9, 739 1999.
19 C. Even, S. Russ, V. Repain, P. Pieranski, and B. Sapoval,
Phys. Rev. Lett. 83, 726 1999.

036614-10