FINITE ELEPIENTS

IN ANALYSIS
A N D DESIGN
ELSEVIER Finite Elements in Analysis and Design 18 (1995) 379-388

Smooth boundary integration method in a linear two-dimensional

problem of incompressible fluid and solid*
V. Rubenchik
Naval Air Engineering Center, Lakehurst, NJ 08733, USA
Received June 1989; revised March 1994

Abstract

This paper presents a new application of a theoretical and computational method of smooth boundary integration
which belongs to the methods of boundary integral equations. Smooth integration is not a method of approximation. In
its final analytical form, a smooth-kernel integral equation is computerized easily and accurately.
Smooth integration is associated with a "pressure-vorticity" formulation which covers linear problems in elasticity and
fluid mechanics. The solution presented herein is essentially the same as that reported in an earlier paper for regular
elasticity. The constraint of incompressibility does not cause difficulties in the pressure-vorticity formulation.
The linear fluid mechanics problem formulated and solved in this paper covers Stokes' problem of a slow viscous flow,
and has a wider interpretation. The translational inertia forces are incorporated in the linear problem, as in Euler's
dynamic theory of inviscid flow. The centrifugal inertia forces are left for the non-linear problem. The linear problem is
perceived as a step in solution of the non-linear problems.

1. Introduction

Smooth boundary integration, a theoretical and computational method, presented in this and
preceding papers El, 2] is n o t a method of approximation. The method only solves basic linear
problems, and solves them without computational difficulty. The method in its final analytical
form, is a smooth-kernel boundary integral equation, and is computerized easily and accurately.
Smooth integration belongs to the methods of boundary integral equations which also unite more
universal approximate methods of boundary elements.
There is a cost associated with smoothness; there is a question as to the uniqueness of the
solution in multi-connected regions. This is the only question that deserves attention as soon as the
problem and method are recognized in general. The solution presented in this paper follows
Ref. 1-2].

* This paper is dedicated to the memory of Paul R. Trumpler.

0168-874X/95/$09.50 ~:) 1995 Elsevier Science B.V. All rights reserved

SSDI 0 1 6 8 - 8 7 4 X ( 9 4 ) 0 0 0 6 4 - 6
380 V. Rubenchik / Finite Elements in Analysis and Design 18 (1995) 379-388

The Stokes problem in fluid mechanics is identical to the statical problem of linear elasticity with
a Poisson ratio v = 0.5. This c o m m o n problem is considered in this paper. We use the terminology
of two branches of mechanics interchangeably: velocity and displacement, vorticity and double
angular displacement, minus pressure and mean normal stress, viscosity and shear modulus.
The linear fluid mechanics problem considered in this paper has, however, a wider interpretation
than the Stokes problem. It combines Euler's dynamic theory of inviscid flow and Stoke's static
theory of viscous flow. Translational inertia forces are incorporated in the linear problem. Only
rotatory, centrifugal inertia forces are left for the corresponding non-linear problem. The linear
problem can be considered a step in the solution of non-linear problems. A relation to the stream
function vorticity formulation (a recent reference is [3]) can be traced.
Smooth integration is associated with a "pressure-vorticity" formulation. The choice of the
primary unknown functions is important. Incompressibility does not cause difficulties in the
pressure-vorticity formulation in contrast to the displacement formulation in elasticity. Similarly,
a difficulty with the infinite regions known in the two-dimensional Stokes problem [4] is not an
issue in the pressure-vorticity formulation; questions depend on the formulation.

2. Pressure-vorticity formulation

This is still an introductory section. The pressure-vorticity formulation for a steady incompress-
ible flow in two dimensions gives the following equations:

8p* 8~ 8p* t3o9

+ -Sfy = p ouy, ay ax - po)ux, (,)

where p* = p + ½plul 2 is a modified pressure function, co = Ouy/Ox - Oux/Sy is vorticity, p denotes

the pressure, ux, u r the velocities,/~ the viscosity, and p the mass density. The definition of vorticity
and the equation of incompressibility, 8ux/Ox + 8uy/Oy = 0, complete the system of four equations
with four unknowns.
The term with absolute velocity squared lul z = u 2 + urz in the pressure function defines the
translational inertia forces and does not cause a non-linearity, similar to Euler's dynamic problem.
The centrifugal inertia terms are placed on the right-hand side, implying an iterative procedure
based on a linear problem solution. The stream function-vorticity formulation follows if the
modified pressure p* is eliminated from equations ( , ) and the velocities are replaced with the
stream function ~k as ux = 8~O/Oy, uy = - O~k/Sx.
The linear problem is formulated with zeros on the right-hand side of equations ( • ). The two
homogeneous equations are recognized as the Cauchy-Riemann conditions.
In the three-dimensional problem, the formulation is as follows:

O---~ + # \ ~?y dz j = P(e)zur -- e)ruz) • (**)

The primary unknowns include a scalar function of the modified pressure p* and three
components of the vorticity vector. Two more equations like ( • • ) are implied. The fourth equation,
 V. Rubenchik / Finite Elements in Analysis and Design 18 (1995) 379-388 381

which completes the set of equations for the four unknown functions, is the following:

+-S)-y+-SYz=°" (***)
The system of equations (**) with zeros on the right-hand side and (***) represents a 3D
generalization of Cauchy-Riemann conditions. Possibilities of the pressure-vorticity formulation
in the three-dimensional problem are not yet explored. We return to the two-dimensional problem.

3. Field functions

The vector function of displacement or velocity, w = ux + iuy, is presented as a complex function

of complex conjugate variables z = x + iy, 5 = x - iy, in a Cartesian coordinate system x, y.
Derivatives Ow/Oz and 8w/c~i are of special interest:

8w 1 (Sux c~ur"] i (Ouy ~ux'] = ½(V.u + ko), (1)

+ ay]+ kax ay]
-2kax ayj \ax +ayj= (2)
Here V. u is the divergence of the vector u (zero for incompressible material), ~o is the vorticity
(double angular displacement or velocity), a~,, as, zxs are the components of the stress tensor, and
/~ is viscosity in fluid mechanics and shear modulus in elasticity. The combination
z = try, - o-s + 2iZxy is termed the stress deviator, to accompany the stress invariant a = ax + o s,
which is minus the double pressure. For a compressible elastic material, a is proportional to V.u.
Comparison of (1) and (2) yields the following equation of stress-vorticity compatibility:
8 "c
= 2 L ( V - U + ico). (3)
8z # c3i

4. Derivation in fluid mechanics

The equation of dynamic equilibrium in complex variables has the form

(Ow 8w Ow) t3a Oz
P -~-+W~z +~-~ =~-i+8---~' (4)

where p is the density and 8w/St is the partial derivative of w with respect to time. Vorticity, as
io9 = 8w/Sz - 8~/i~i, is used in the following modification:
(~w c~ ) ~a ~?~
p ~ - + icow + ~-~ Iwl 2 -- ~ + ~ . (5)

By this transformation, the translational and centrifugal inertia forces are separated: the term with
Iwl 2 = w~ = u ~ + u y2 (square of the absolute value of velocity) accounts for translational inertia
382 V. Rubenchik / Finite Elements in Analysis and Design 18 (1995) 379-388

and the term with cow for centrifugal inertia. Next, the compatibility equation (3) is used to replace
the stress deviator ~ (if p = const.), and the result is rearranged as follows:

8 a 8 __3 ko) ( S w )
8e P - ~ l w l 2 + 2 # 8 ~ , ( V ' u + =p --~+i(ow . (6)

If the density p is constant, the pressure and linear inertia forces are united into a single term, as
in Euler's theory of an ideal fluid. If the viscosity p is constant, then the divergence V. u becomes
a part of the pressure term, which is applied in elasticity. However, in fluid mechanics, if V. u # 0,
then p is variable. In the following, incompressibility ( V. u = 0) is assumed. The primary unknown
complex function couples the pressure function and vorticity:
= ~ - plwl z + 2pito. (7)
The differential equation for the non-linear problem is

8~- = p + icow . (8)

and the equation for the linear problem is

84
-8Z
- : O. (9)

5. Iterative solution to the non-linear time-dependent problem

Eq. (8) may be integrated using the technique of integral identities, with the following result:

2~l fn 4~eia [[Sw~t_+iogw

cPl. . . . = ds - p dx dy, (10)
0] ~ - z0 jj z - Zo
where the result is defined at a point Zo depending on its location: in the domain occupied by the
fluid (2r0, on the boundary (r0 or outside the domain (0). In the boundary integral, B denotes the
full boundary of the domain, which may consist of a number of inner boundaries, and one or none
outer boundary, ~ stands for z on boundary, a current point of integration;/~ is the outward normal
vector angle counter-clockwise from the x-direction; ds is the boundary arc length differential. The
area integral in (10) represents the difference from the linear steady-state problem.
The general problem can be solved iteratively with approximations for the nonlinear and
time-dependent terms, as usual in numerical methods.

6. Integrations in linear problem

In the following, we omit the area integrals given in (10). In the linear time-dependent problem
considered here, • = ~(z) is an analytic function of complex variable z = x + iy, according to (9).
 V. Rubenchik / Finite Elements in Analysis and Design 18 (1995) 379-388 383

The simplified equation (10) is recognized as the Cauchy theorem of complex analysis:

• (z) = e ia ds. (11)

OJ ~'-z
A pattern for the following derivation may be found in [5, 6]. The imaginary part of ~(z),
2yio) = ½[~(z) - ~(z)], is used in (3), with V . u = 0, to find

t3---~= 2/.ti 0:? - ½q¢(z), z = ~(z) - ½zCb'(z) (12)

introducing one more analytic function, ~(z). The Cauchy theorem is applied to it, with substitu-
tion from (12) and integration by parts, which yields

2~ fnl ( ( - - z ) l e-iP
OjZ= r + ½ ~ e2ip+~-Z--~ ~ - _ i d s . (13)

The result is defined at point z depending on its location, as noted for Eq. (10). For the displacement
or velocity function w, one finds

4 ~ w = O(z) -- ½ z ~ ( z ) + ½~o(z) (14)

where O(z), q~(z) are integral functions of 7'(z), ~(z), respectively. Integration proceeds further:

2~t t'l- we -it~ ~--z ]

04uw= + c_ e-i - ei lnl -zl ds. (15)

7. B o u n d a r y functions and conditions

Displacement or velocity w is prescribed on the boundaries. Relation (15) presents a valid

logarithmic-singular boundary integral equation for this problem. But our intention is to derive
an integral equation with a regular and smooth kernel, as most desirable in computations.
We operate with the boundary derivative of w, which is expressed according to (1), with V. u = 0,
and (2):

w' Ow Ow dw o) r i
- Os - &: ieia - ~3-~ie-ip - 2 ei¢ - 4--~ie-ia = 4y [½(~ - ff)e ia - ze-ia]. (16)

Vorticity o) has been replaced as the imaginary part of the complex primary function (7). Relation
(16) presents the boundary condition to be satisfied. The boundary data prescribed are partly lost
in the differentiation of w.
384 V. Rubenchik / Finite Elements in Analysis and Design 18 (1995) 379-388

The boundary stress (traction) vector is expressed, by using (7) and (16), in terms of the known
velocity and unknown primary functions:

q = qx + iqr = ½(o'eip + z e - i # ) = 2yiw' + ½(qb + p IwtZ)e i//. (17)

8. Integral forces and moments

Derivative w', being integrated along a closed boundary contour, must yield zero, for a single-
valued primitive function. With this, relation (17) is integrated over an internal boundary Bi:

e, = f. q ds = ½fn (~ + plwl2)eit~ ds, (18)

i i

where Qi is the total (drag) force acting on the internal body. This gives a uniqueness condition in
the solution to be used later.
Next, the force moment on the internal body is expressed as

M, = fB (xqr - yqx)ds = Im fB q ds = Im f, ' [2#iw' + ½(q~ + plwl2)eia]~ds, (19)

i i i

where Im represents the imaginary part of the complex number.

9. Boundary integral equation

Integral relations (11) and (13) are used with z = (o being a point on the boundary. Stress
deviator z is replaced, using boundary condition (16), on both sides of(13). A combination with (11)
is then formed in order to obtain the following second Fredholm integral equation:

7t~bo - f n ( 1 + ~( - ~ 0 - 2ei t ~ o ) Re ~~eiP ds = 4#i I - rtw~ + e - 2iaofn ~w'_d_s]

- (o J" (20)

Variables at parametric point (o are denoted with a zero subscript and Re stands for the real part of
a complex number. The kernel in the equation is regular and smooth. In a numerical solution,
quadrature formulas are applied to the entire integrand, with no local element approximations
involved [1, 2].
Eq. (20) in its homogeneous part is the same (accurately to multiplier i at the unknown function)
as the equation of elasticity with the traction boundary condition [2]. For problems with
non-rotating rigid bodies, the right-hand side of (20) vanishes, thus indicating ill-conditioning. The
analysis of eigenfunctions is presented in the appendix of [2], in which a coefficient, x, is to be set
equal to 1, for incompressible material. Modes of ill-conditioning include an arbitrary constant
additive to the pressure function, ~ = real const., as an eigenfunction (vanishing for infinite
 V. Rubenchik / Finite Elements in Analysis and Design 18 (1995) 379-388 385
domain), and the following zero integrations applied to the boundary Bi of an internal body:

f ... eit~°ds0 =0, Re~ .-.i~-oeit~°ds0 =0, (21)

,) Bi Jni

where " . . . " stands for either side in (20). The latter mode in (21) is caused by the incompressibility,
and is not present in the displacement boundary problem of compressible elastic material.

10. Additive pressure constant

For incompressible material, an arbitrary external pressure may be applied with no effect. We set
the pressure function, real part of 4, equal to zero at an arbitrary point zl in the domain. The
Cauchy theorem (l 1) is used, in its real part, for Re ~(zl) = 0, and once more to form the difference

2 ~ t • (z) = f'B q~eia ds - Re fB qbeia ds. (22)

0J ~-z ~-zl
The integral equation is now derived based on this statement to bring about the following result:

LHS + Re fn ( ~eit~
- z------~ds = RHS, (23)

where LHS and RHS stand for the left- and right-hand sides of (20). If the domain is infinite, then
the integral equatJion (20) is free from this sort of eigenfunction, and no amendment is needed (the
solution yields zero pressure function at infinity).

11. Integral conditions of uniqueness in solution

For the first mode in (21), the conditions of uniqueness are provided by Eq. (18). This is the
condition of continuity of vector function w upon integration of its derivative along the internal
body boundaries.
One more condition of uniqueness was found in [2], based on Cauchy theorem (11) and
numerical experinaentation:

f~ ~eip ds = 0, (24)
Im ~ - zi

where Im means imaginary part and zl is an arbitrary point inside the internal body.

12. Well-conditioned boundary integral equation

Integral conditions of uniqueness (Eqs. (18) and (24)) are used to modify the integral equations
(20) and (23) in t]ae way that precludes interference of additional terms in integrations (21), as
386 V. Rubenchik / Finite Elements in Analysis and Design 18 (1995) 379-388

explained in [2]:

Oei oei
r~o- 1 + ~ 2 ~ o e- Re~_--~ds + Re ~ - z----~

+ e-iS° fB 4~eis ds + (~o - za)e -is° Im fB ~~e~S

- zi ds
i

=4//i[--/~w~-Fe-2iSo~ _w-'--ds-1-Fe-iSo[2Qi-pfl Iwl2eiSds], (25)

J~ ~ - ~oJ

where zc~ is the centroid of the ith boundary, i.e. Sn,(~- zc~)ds = 0. The additional terms for
a particular B~ boundary are included when the parametric point ~o lies on the same boundary. If
there is an outer boundary, the term with Zl in the first line is included, otherwise not.
The system is still not complete since a set of unknowns the integral forces Q~, appear in (25).
Additional equations are formed based on relation (15) to recover the information lost in the
differentiation of w:

fn [ ~ - zi7 . (" we -is

4~eiS l n l ~ - z i l - ½ 4 5 ~ j as = 8r~Woo + 4~J.~-2-_ N ds. (26)

This is applied to each of the internal bodies, with z~ being a point inside it. Velocity Woo is included
forthe case of infinite region. This was calculated as the integral over an arbitrary outer boundary,
provided the velocity w on it is constant. One can see that there is no irregularity in the solution as
z - ~ , no Stokes paradox, in the vorticity-pressure formulation. In a numerical solution, the
integral equation (25) and conditions (26) are replaced by a system of linear algebraic equations.
The problem is solved, for the general displacement-type boundary conditions.

13. Interacting with rigid bodies

For a rigid body, the velocity and displacement of a boundary point ~, and its derivative are

W = Wi + iQi(( -- zi), w' = -- Qie i#, (27)

where Wi is the velocity of centroid point Zi of the rigid body area, and f2i is the angular velocity of
the body. Now, the right-hand side of (25) is expressed as

RHS(25)=4#i xQiei#°-e -2i#° Qj ~(_(oj+2e-iS°(Qi+ipAig2iWi), (28)

where A~ is the area of an internal rigid body. In the calculations we set

fB CeiS ds = f (y dx - x dy) = - 2A,.

i i
 V. Rubenchik/ Finite Elements in Analysis and Design 18 (1995) 379-388 387

Other integrals, :related to [w[ 2 after substitution from (27), are zero directly or as static area
moments about the centroid. The summation in (28), over index j, includes an outer rigid boundary,
if it is present.
The right-hand side of condition (26) is rewritten next as

RHS(26) = 8n#(Woo - W~) + 4#i . I2j fBJ ~ - zj

7-- ~i e - iads. (29)

For a finite regien, Woo may be understood as the velocity of a device containing the fluid. The
forces Q~ in (28) are to be found together with the primary unknown function 4). Upon solution, the
moments on rigid bodies are calculated according to (19):

M~ = l l m ~ q~-e i# ds - 4#Agf2~. (30)

d Bi

This completes the solution.

14. A problem wiith given forces and a stiffness matrix

Consider an arbitrarily shaped piece of an incompressible solid (or viscous fluid with no inertia
forces) in a rigid fixed frame and with rigid internal bodies. The forces Qi and moments Mi are
prescribed on the internal bodies. Angular displacements f2i present a set of unknowns in (28).
Equations additional to the algebraic system of integral equation, are formed using relation (30),
with I2i unknown and Mi known. The solution is obtained.
Consider the same problem, but for an inertial fluid. There is a non-linearity due to the product
f2~Wi in (28). It is better to use the formulation with the boundary velocities prescribed, and remove
this non-linearity from the level of an integral equation solution to the level of non-linear finite
element analysis.
The algebraic system of the boundary integral equation is solved simultaneously for each of the
factors of influence: a component of velocity vector Wi, or an angular velocity f2~, or a component
of the product f2~ W~, each one being equal to unity. The result, forces and moments due to the unit
velocity values, ferm a stiffness matrix, which is non-quadratic because of the terms of non-linearity
f2i W~. This forms a large non-linear finite or infinite element to be used in all types of analysis.
A more general discussion of the large elements formulated with use of the non-element boundary
integral method Jis found in [1].

15. Closing discussion

An equation may have no solution, one solution, or many solutions. Derivations in this paper
were aimed at a smooth-kernel boundary integral equation, (25), which possesses only one solution
and is, therefore, computational. The interested reader can understand the topic better by studying
the reference given at the end of this paper.
388 V. Rubenchik/ Finite Elements in Analysis and Desion 18 (1995) 379-388

The reader is invited to imagine the smooth integration method in a pleasant computer graphics
environment and compare it with Roark's tables. Smooth integration is intended to replace Roark
and other tables on a much larger scale, covering arbitrary cross-sections and regions.
Smooth integration can be attributed to and compared with other techniques for a finite element
formulation. The section above shows a formulation of a large arbitrarily shaped element (stiffness
matrix). Generally, a large element is possible for a homogeneous piece of material in linear statical
problem. The question of incompressibility is not critical in the smooth integration method, and
a comparison with the penalty method in finite elements would hardly be appropriate.
Finally, the author holds that approximations should be avoided if possible. An opposing
viewpoint is that approximations can solve a problem satisfactorily. This debate depends on
computer implementations and practice. To this end - computer implementation - this paper
recommends the smooth boundary integration method and shows its potential in a new
application.

Acknowledgements

Professors G. C. Hsiao, University of Delaware, and S. Mukherjee, Cornell University, reviewed

earlier notes on this subject, with valuable comments. The work has been completed at Exton
Center for Advanced Studies, a non-profit organization created by Dr. Paul R. Trumpler to
promote engineering and scientific research.

References

[1] V. Rub•nchik• "B•undary integral equati•n meth•d •f higher c•mputati•nal accuracy•È Finite Elements Anal. Des. 4•
pp. 133-145, 1988.
[2] V. Rubenchik, "Accurate numerical solution in two-dimensionalelasticity using smooth-kernel boundary integral
equations," Finite Elements Anal. Des. 6, pp. 197-205, 1990.
[3] B. Fornberg, "Steady incompressibleflow past a row of circular cylinders,"J. Fluid Mech. 225, pp. 655-671, 1991.
[4] F.M. White, Viscous Fluid Flow, McGraw-Hill, New York, 1974.
[5] N.I. Muskhelishvili,Some Basic Problems of Mathematical Theory of Elasticity, Noordhoff Ltd., Holland, 1953.
[6] V.Z. Parton and P.I. Perlin, Mathematical Methods in the Theory of Elasticity, Mir Publishers, Moscow, 1984.