Appl. Maih. Leti. Vol. 4, No. 5, pp. 27-30, 1991 089Ht359/91 $3.00 + 0.

00
Printed in Great Britain. ‘AU rights reserved Copyright@ 1991 Pergamon Press plc

A Note on Symmetry Boundary Conditions

in Finite Element Methods

L. DEMKOWICZ

Texas Institute for Computational Mechanics

The University of Texas at Austin

(Received November 1990)

Abstract. A short discussion on two kinds of symmetry boundary conditions in the context of
variational formulations and fmite element methods is presented. Applications indudiug modeling
of open boundaries in computational fluid mechanics are discussed and a numerical example is
presented.

1. SYMMETRY BOUNDARY CONDITIONS OF THE FIRST AND SECOND KIND

We begin with a model elliptic system related, for example, to the deformation of a two-
dimensional domain depicted in Figure 1. The motion in the top figure amounts to what we
shall call the physical symmetry. As a result of a symmetry of load and kinematic boundary
conditions, the motion becomes symmetric with respect to the y axis. In terms of the solution
vector u = (u, v), we have

u(z, Y>= --u(--tY Y); Zl(t, Y) = v(--r, Y). (1.1)

The second condition implies

$x9Y)= -2(-x,y). (1.2)

Passing with z -+ 0 in the first of conditions (1.1) and in (1.2), we get the symmetry boundary
conditions of the first kind
BV
U= 0; z=o, for 2 = 0, (1.3)
or, equivalently,

UZ 0; -Y=11(~+~)=0, for x = 0, (14

The problem can be solved in half of the domain fl with the symmetry boundary conditions (1.3)
or (1.4) along the Y axis. The situation is different in the second example in Figure 1. As a result
of the enforced uniform motion along the top and bottom lines, another kind of symmetry can
be observed. In terms of the components, we have

4x9Y) = +-2, Y); v(x, Y) = +-z,Y). (1.5)

27
28 L. DEMKOWICIZ

This leads again to a boundary condition on the derivative of the solution

BU au 6v
-_=(I -_=-_=o , for 2 = 0.
82 ’ 8x ax
(1.6)
( >

We call (1.6) the symmetry boundary conditions of second kind.

I
I
I

m I- ’

__---_-_-g _-_- __---

I I I
$x,y)

ii2
g-KY)

I:I 1 I
L-

Figure 1. Symmetry boundary conditions of the first and second kind.

Figure 2.

Formally the elliptic system can be again solved in the half of the domain R, this time with
boundary conditions (1.6) along the x = 0 axis.
The interesting difference between the two types of symmetry boundary conditions appears
when constructing the variational formulation of the problems. Multiplying the equations of
equilibrium by a test function 6u and integrating by parts, we arrive at the usual boundary term
of the form
t * 6u = ts 6u+ t, h, (1.7)
where t is the stress vector. Decomposing both the stress vector and the virtual displacement bu
into normal and tangential components (to the boundary), we get

t*6u=u,6u,+u,~~,, (1.8)
 Symmetry boundary conditions 29

where u, and ur are the normal and tangential stresses and 6u, and &I, are the normal and
tangential virtual displacements. Eliminating the normal virtual displacement
6u, = 0, (1.9)
we arrive at the boundary term
UT 6% (1.10)
which disappears according to the second of conditions (1.4). In particular, the second of the
boundary conditions (1.4) is identified as the natural boundary condition corresponding to the
governing linear elliptic operator.
The situation is quite different in the second case. Rewriting the boundary term explicitly in
terms of derivatives of U, we get

uzr 6u + boy 6v = (2/J+ A) %u+A%u+p (Z+E) sv.

ax ay
(1.11)

Accounting now for boundary conditions (1.6), we reduce (1.11) to

av
x -&L+p%v.
aY BY
(1.12)

Note the fact that (1.12) still does include the contributions of the solution vector (u, v) and it will
contribute to the stiffness matrix. These remaining boundary terms indicate that the symmetry
boundary conditions of second kind cannot be classified as natuml boundary conditions.
REMARK. The boundary terms (1.12) are perfectly legitimate from the mathematical point of
view. It follows from the identity

I( --
n
~:~+~~)~~dY=Jsn(~(-n,)+~n.) v+, (1.13)

that the tangential derivative au/C+ is a well defined element of H-i(afi) for all functions u
from H’(0) (camp. the generalized Green’s formula in [l]).
What remains to be verified is the coerciveness of the resulting bilinear form. In the example
shown, for instance, substituting 6~ = u in (1.12) we get the boundary term of the form

In the special case )r = G this term reduces

I(Brl
Au~+G%
0~
to
aY >
ds. (1.14)

due to the kinematic boundary

x
conditions
J 8n $uv,ds = 0,
imposed on the top and bottom of the plate.
(1.15)

2. MODIFICATION OF THE BILINEAR FORM

WITHOUT CHANGING THE DIFFERENTIAL OPERATOR
Including additional boundary integrals of the type (1.12) in the definition of the bilinear form
may be equivalent to adding extra terms in the domain integral contribution to the bilinear form
in such a way that the corresponding differential operator remains unchanged. An example of
such a procedure is given in [l]. Defining the bilinear form as

B(u,v)= n ~;++~-+f!!+““) dx4/ , (24

I( BY aY
with c a constant we easily identify the corresponding operator as the Laplace operator Au and
the boundary operator in the form of the oblique derivative
8U
E+Ca* (2.2)

The differential operator is in particular independent of c, and the form is easily proved to be
coercive under regular assumptions.
ML .:,-c
30 L. DEMKOWICZ

3. AN APPLICATION IN FLUID MECHANICS

Symmetry boundary conditions of the second kind prove to be very useful in modeling open
boundary conditions in fluid mechanics. The exterior flow problems are usually formulated in
unbounded domains. For practical calculations a finite computational domain must be selected
and, as a result, an artificial open boundary is introduced. Consequently, relevant boundary
conditions must be specified on such boundaries.
An example for an application of the symmetry boundary conditions of second type would be
a flow in an infinite horizontal duct. Assuming that, at a far enough distance, flow stabilizes, the
symmetry boundary conditions may be implemented on the exit boundary. Notice that no extra
information is necessary to apply these conditions.
The symmetry boundary conditions have been successfully applied to model much more com-
plex open boundary conditions resulting in perfect “fully absorbing” treatment of boundary terms.
Figure 3 shows an example of a solution to the steady state Euler equations by means of the
Taylor-Galerkin method decribed in [2] (Aow over a wedge problem). The top line of the domain
is identified as a subsonic inflow/subsonic outflow boundary, gradually changing into supersonic
outflow, and the right-hand side is a supersonic outflow boundary. The symmetry boundary
conditions applied along both boundaries result in no artificial reflection of the shock from the
upper outflow boundary.

Figme 3. Flow over a wedge problem for Euler equations.

REFERENCES

1. R.E. Showalter, Hilbert Space Method8 for Partial Differential Equationr, Pitman Publishing Limited, London,
(1977).
2. L. Demkowicz, J.T. Oden, W. Racbowicz and 0. Hardy, An h-p Taylor-Gale&in finite element method for
compressible Eula equations, Computer Method8 in Applied Mechanica and Engineering (to appear).

Texas Institute for Computational Mechanics, The University of Texas at Austin, Austin, Texas 78712, U.S.A