Professional Documents
Culture Documents
Mathematical Theory
f!f Elastuzty
1. S. SOKOLNIKOFF
Profe88OT oj Mathematic8
University of California
L08 Angele8
SECOND EDlTlON
",Bomba,-N_ DeIhl
ACC
. i.. >.J.,\'P
tI()._~ ;3,7 98
CL.IIIO._ _.
....
,.,.
------.:::::::':'::-:&:-:.1 ~~~ /
T M H Edition
New York.
PREFACE
The theory of elasticity, in its broad aspects, deals with a study of the
behavior of those substances that possess the property of recovering their
size and shape when the forces producing deformations are removed.
In common with other branches of applied mathematics, the growth of
this theory proceeded from a synthesis of special ideas and techniques
devised to solve concrete problems. This resulted in a patchwork of
theories treating isolated classes of problems, determined largely by the
geometry of bodies under consideration. The embedding of such diverse
theories in a unified structure, and the construction of the analytical
tools for calculating stresses and deformations in a strained elastic body,
are among the dominant concerns of the mathematical theory of elasticity.
This book represents an attempt to present several aspects of the
heory of elasticity from a unified point of view and to indicate, along
with the familiar methods of solution of the field equations of elasticity,
30me newer general methods of solution of the two-dimensional problems.
The first' edition of this book, published in 1946, had its origin in a
course of lectures I gave in 1941 and 1942 in the Program of Advanced
Instruction and Research in Mechanics conducted by the Graduate
School of Brown University. In those lectures I stressed the contributions to the theory by the RuSsian school of elasticians and, in particular,
the relatively little--known work of great elegance and importance by
N. I. Muskhelishvili. I planned ~o supplement that book by a companion volume dealing with effective methods of attack on the twodimensional and anisotropic problems of elasticity. The developments
in the intervening years, however, were so rapid that I was urged to
puhlish instead a single volume containing an up-to-date treatment of
material presented in the. first edition and supplement it with new topics,
in order to give a rounded idea of the current state of the subject.
The present edition differs from its predecessor by extensive additions
ad ~ Most of the material appearing in the .last three chapters
had no counterpart in the first edition. Throughout r bve tried to give
" .clea1' indieation of the fronti.,rs of the developments, and 1 have constantlY kept in mind those readers whose principal conrero is with prallw.l appication of the theory. While nfo) volume of this size can lay
.... to an _ _ _ve list of referenoos to reeeateh lite~, I bave
vi
PREFACE
vii
PREFACE'
of the investigations and lectures I gave during my tenure as a Guggenheim Fellow during the academic year 1952-1953. I am pleased to
have this opportunity to acknowledge my gratitude to the Guggenheim
Memorial Foundation, whose grant enabled me to discuss this book with
my colleagues in England and on the Continent. I also wish to repeat an
acknowledgment, made in the Preface to the first edition, to the Wisconsin
Alumni Research Foundation for a grant-in-aid that facilitated the
publication of the predecessor of this volume.
I am indebted to Dr. George E. Forsythe, Research Mathematician at
the University of California at Los Arigeles, for material on Relaxation
Methods in Sec. 125, and to Robert K. Froyd, Research Assistant at the
University of California at Los Angeles, for his help in proofreading and
preparing the index matter.
. I.
S. SOKOLNIKOFF
CONTENTS
v
PBml'ACI!l .
HISTORICAL SKETCH.
CHApTI!lR
1.
2.
3.
4.
5.
6.
'1.
8.
9.
10.
11.
1.
ANALYSIS OF STRAIN
Deformation
Affine Transformations .
Infinitesimal Affine Deformations
A Geometrical Interpretation of the Components of Strain
Strain Quadric of Cauchy
Principal Strains. Invariants
General Infinitesimal Deformation .
Examples of Stmin
Notation
Equations of Compatibility .
Finite Deformations.
2. ANALYSIS OF STRESS
Body and Surface Forces.
Stress Tensor .
Note on Notation and Units.
Equations of Equilibrium
Transformation of Coordinates .
Stress Quadric or Cauchy
Maximum Normal and Shear Stresses.
Examples of Stress
35
36
39
40
42
45
Mohr's Diagram
CHAPTER
CHAPTIIl1l4.
6
9
12
14
16
20
23
25
25
29
35
CIlAPTER
12.
13.
14.
15.
16.
17.
18.
19.
5
5
19
63
56
56
5ll
66
67
71
80
81
86
89
91
91
96
97
100
107
is
CONTENTS
109
114
36. Torsion of Elliptical Cylinder
120
124
37. Simple Solutions of the Torsion Problem. Effect of Grooves
38. Torsion of a Rectangular Beam and of a Triangular Prism .
128
39. Complex Form of Fourier Series
134
137
40. Summary of Some Results of the Complex Variable Theory
41. Theorem of Harnack.
143
42. Formulas of Schwarz and Poisson
145
43. Conformal Mapping .
147
44. Solution of the Torsion Problem by Means of Conformal Mapping
151
45. Applications of Conformal Mapping
157
46. Membrane and Other Analogies.
165
47. Torsion of Hollow Beams
169
177
48. Curvilinear Coordinates.
49. Torsion of Shafts of Varying Circular Cross Section .
186
50. Local Effects .
190
51. Torsion of Anisotropic Beams .
193
52. Flexure of Beams by Terminal Loads
198
53. Center of Flexure .
204
54. Bending by a Load along a Principal Axis.
208
55. The Displacement in a Bent Beam.
209
56. Flexure of Circular and Elliptical Beams
213
57. Bending of Rectangular Beams.
. . 217
58. Conformal Mapping and the General Problem of Flexure; the Cll.rdioid
Section.
219
J\9. k-wii."G.<>.f c:.i.',<'.I.\\g.. p.i,<w.
223
60. Stress Functions and Analogies; Beams of Elliptical and Equilateral Triangular Sections .
230
237
61. Flexure of Semicircular Beams .
62. Multiply Connected Domains. Deformation of Nonbomogeneous Beams
231)
with Free Sidee. Other Developments
63. Deformation of Cylinders by Lateral Loads
243
64. Torsion of a Cylinder by Forces on the Lateral Surface .
244
CHAPTER 5.
65.
66.
67.
68.
69.
10.
71.
72.
73.
U.
75.
76.
77.
78.
7.
89.
Introductory Remarks
Plane Deformation
Plane Stress. Generalized Plane Stress
Plane Elastostatic Problems.
Airy's Stress Function
General Solution of the Biharmonic Equation.
Formulas for Stresses and Displacements .
The Structure of Functions 9'(z) and 1{I(z) .
First and Second Boundary-value Problems in Plane Elasticity
Remarks on the Existence and Uniqueness of Solutions .
.
The Role of Conformal Representation in Plane Problems of Elasti{\ity .
An Elementary Method of Solution of the Basic Problems for Simply Connected Domains .
Solution of Basic Problems for a Circular Region. .
Solution of Problems for the Infinite Region Bounded by a Circle
Infinite Region Bounded by an Ellipse.
Problems for the Interior of an Ellipse.
249
249
250
253
257
259
262
263
266
269
272
273
216
280
2S6
292
296
CONTENTS
81.
82.
83.
84.
85.
86.
87.
88.
89.
CHAPTER
90.
91.
92.
93.
94.
95.
96.
97.
98.
99.
100.
101.
102.
103.
104.
CHAPTER
105.
106.
107.
108.
109.
110.
111.
112.
113.
114.
115.
116.
117.
118.
119.
120.
121.
122.
123.
124.
125.
6.
THREE-DIMENSIONAL PROBLEMS
General Solutions .
Concentrated Forces .
Deformation of Elastic Half Space by Normal Loads
The Problem of Boussinesq
Spherical Shell under External and Internal Pressures
Spherical Harmonics .
Elastic Equilibrium of a Sphere and Other Problems
Betti's Method of Integration
Existence of Solutions
Thermoelastic Problems
Thermal Stresses in Spherical Bodies
Two-dimensional Thermoelastic Problems.
Vibration of Elastic Solids
Wave Propagation in Infinite Regions .
Surface Waves.
7.
VARIATIONAL METHODS
Introduction .
Variational Problems and Euler's Equations
Theorem of Minimum Potential Energy
Theorem of Minimum Complementary Energy
Theorems of Work and Reciprocity
Illustrative Examples.
Variational Problem Related to the Biharmonic Equation
The Ritz Method. One-dimensional Case
The Ritz Method. Two-dimensional Case
Literature on Direct Methods
The Galerkin Method
Applications to Torsion of Beams and Deformation of Plate.
The Method of Kan1 1rovich.
The Trefftz Method
An Application of the Trefftz Method .
The Rafalson Method 'or the Biharmonic Equation .
The Method of Least Uquares. Collocation
.
The Function Space M thods .
.
The Method of Finite Differences .
An IIlustration of the Method of Finite Differences
Relaxation Methods '
xi
297
303
304
307
311
313
318
322
326
328
328
336
339
341
343
345
350
354
358
358
362
364
367
370
372
377
377
377
382
387
390
397
402
404
409
412
413
416
421
425
430
432
435
440
442
445
454
46i
SuBJEcT INDEX
471
IUSTOBICAL SKETCH
The theory of eiasticity is concerned with the study of the response of
elastic bodies to the action of forces. A body is called elastic if it possesses
the property of recovering its original shape when the forces causing
deformations are removed. The elastic property of material media is
shared by alI substances provided that the deformations do not exceed
certain limits determined by the constitutive characteristics of the body.
The elastic property is characterized mathematically by certain functional relationships connecting forces and deformations. Among such
relationships a linear law stemming from a generalization of Hooke's law l
is of fundamental importance. Hooke's law states, in effect, that the
extensions of springlike bodies, produced by the tensile forces, are proportional to the forces. An identical law was discovered independently
by Mariotte' in 1680 and used by him to investigate the strength of
cantiliver beams. Mariotte concluded that a cantilever beam resists
flexure because some of its longitudinal fibers are extended and others are
contracted. Although Mariotte's assumption regarding force distribution in fibers was correct, his investigations did not include the study of
the shape assumed by the beam's axis. Such a study' was made in 1705
by Jacob Bernoulli, who combined elementary equilibrium considerations
with Hooke's law to obtain the differential equation of the elastica, that is,
the curve assumed by the deformed axis of the beam. His equation implies
that the curvature of the elastica at each point is proportional to the
bending moment acting in the section through the point. It readily
fonows from this result that the work done in bending the beam is proportional to the integral of the square of the curvature taken along the
elastica. Daniel Bernoulli (1700-1782), a strong proponent of the minimum principles that were in the process of formulation at that time, suggested in a letter' to Euler that the equation of elastica should emerge on
minimizing the integral representing the work done in bending the beam.
In this manner Euler deduced Jacob Bernoulli's equation and integrated
it for a number of special cases. 6
I Robert Hooke, De potentia restitutiva (1678).
E. MariQtte, Traite de mouvement des eaux (1686).
J. Bernoulli, Collected Works (1744), voI..2, p. 976.
'The twenty-sixth letter of D. Bernoulli to Euler, October, 1742, in P. H. Fuss'
Correspondence mathematique et physique (184.3), vol. 2.
L. Euler, the addendum to "De eurvis elasticis," in Metodus inveniendi lineas
curvas maximi minimive proprietate gaudentes (1744). This paper and Euler's
msromcAL SKETCH
CHAPTER
ANALYSIS OF STRAIN
1. Deformation.
(i ==.1, 2, 3).
Xi
(i = 1, 2, 3),
MATH~;MATlCAL
THEORY OF ELASTICITY
auXa,
+ (1 + all)x, +
"loX. +
a2aX3,
+
a"XI + (1 + a..):r. +
a.o +
""Xl +
a, .... + (1 + a.,)xI,
X; = a,o
x; = a.o
x; =
(2.1)
x.,
c-
of order n.
A repeated subscript indicates summation as the index thatis repeated takes thfl
valu. 1, 2, 3. Thus
The symbol lit;, the Krrineckrr delta, is defined to have the vah,e one if i equal!! j,
,.
where the coefficients 4,; are constants, are well known. Since it is
desirable to demand the existence of an inverse, Eqs. (2.1) must be solvable
for the variables :1:1, X2, X3 as functions of x~, x~, x;. 1t follows that the
determinant 18#
a;;1 of the coefficients of the unknowns entering into
the right-hand member of (2.1) must not vanish. It is obvious that, t.he
inverse transformation
(i, j = 1, 2, 3),
(2.2)
is likewise linear.
It is easy to see from (2.1) and (2.2) that an affine transformation
carries planes into planes, and hence a rectilinear segment joining the
FlO. 1
points rO(x~, xg, xg) and P(Xl, X2, x.) is transformed into a rectilinear I;legment joining the corresponding points pO'(xr, xg', xg') and P'(x;, x;, x~)
(Fig. 1). This follows from the fact that the rectilinear segment pop can
be thought of as joining two points po and P on the intersection of two
planCl;! 8 1 and 8 2 ; under the transformation (2.1) points po and P go over
into points po' and P', which lie on the intersection of the planes S~ and
S;, into which the planes Sl and S. are carried by the transformation.
We shall denote the unit base vectors, directed along the coordinate
axes Xl, x., and Xa, by el, e2, and e., respectively. Thus, a vector A whose
components along the coordinate axes are At, AI, As can be written as
(i = 1,2,3).
Since the vector A = e,A, is uniquely determined once its components
A. (i = 1, 2, 3) are prescribed, we can represent the vector A by the
symbol A;. Under the transformation (2.1) the vector A; = x; - x?,
joining the points PO(X O) and P(x), is carried into another vector
A~
x~
- xf' .
.zero if i durers {rom j. The reason for friting the coefficients of x" x" and x, in the
first, second, and third lines as 1 + "'", 1 + au, 1 + a .. will appear later.
+ z. + tJt;jej,
z: "" a.o
we have
+ +
A: = ~ - x2' = (<%i0
Xi
<%i;Xi) - (a.o + x1
= (x, - xf) + a,/(xi - ~) = A, + <%iiA;,
+ I%>;X~)
or
(i, j = 1, 2, 3).
(2.3)
It is clear from (2.3) that two vectors Ai and B, whose components are
and
whose components are again
equal transform into two vectors
equal. Also two parallel vectors obviously transform into parallel
vectors. Hence, two equal and similarly oriented rectilinear polygons
located in different parts of the region R will be transformed into two
equal and similarly oriented polygons in the transformed region R'.
Thus, the different parts of the body T, when the latter is subjected to the
transformation (2.1), experience the same deformation independently of
the position of the parts of the body. For this reason, the deformation
characterized by (2.1) is called a homogeneous def()7'mation.
Consider the transformation (2.1), and let the variables be SUbjected
to another affine transformation,
A:
B:
x:
x~' = 'YkO
(2.4)
(6w
+ 'Yo;)x:.
+ x~ + 'Y1.-3:;.
A~;
then
+ 'Y..xi)
xn =
Al
or
(i, k = 1, 2, 3).
(2.5)
The product of the two successive affine transformations (2.1) and (2.4)
is equivalent to the single transformation obtained by substituting in
(2.4) the values of in terms of Xi from (2.1). Thus one has
x:
x~' = 'YkO
=
a10
Now if the coefficients Ct;i and 'Yii are so small that one is justified in
neglecting their products in comparison with the coefficients themselves,
then
ANALYSIS OF STRAIN'
zt"
akO
x~'
At
vecto~
+ (ak} + 'Y.t,;)(Xj
- x~)
or
(i, k = 1, 2, 3).
(2.6)
Thus, if one neglects products of the ll;; and 'Y;;, then the Cl>efficients in
the resultant transformation (2.6) are obtained by adding the corresponding coefficients ll;; and 'Y;! in the component transformatiol\s (2.3) and
(2.5). In this event, it is said that the product transformation (2.6) is
obtained by superposition of the original transformations. It is clear ftom
the structure of the formulas (2.6) that the resultant transtormation is
independent of the order in which the transformations are performed.
One of the chief sources of the difficulty that confronts one ~n the study
of finite as distinguished from infinitesimal deformations arilses from the
fact that the principle of superposition of effects and the ir.dependence
of the order of transformations are no longer valid.
A transformation of the type (2.1), in which the coefficients are so small
that their products can be neglected in comparison with the linear terms,
is called an injiniteBimal affine transformation.
S. Infinitesimal Affine Deformations. In this section we flhall be concerned with. the problem of separating the infinitesimal affine transformation defined by Eq. (2.3);
(3.1)
6A i
...
A: - Ai = aijA j ,
(3.2)
AliA = AillAi
plus terms of higher order in IIA" which are neglected, since we are concerned with the infinitesimal affine transformation. When the expres-
10
sions for &A, given by (3.1) are inserted in (3.2), one finds that
A iA = avA,As,
A aA =
Since for a rigid body transformation IlA vaniRhes for all values of AI, A.,
As, we must have
(l11
au
(l"
(l ..
= 0,
+ au = a.3 + a32 = au + au =
O.
Hence a necessary and sufficient condition that the infinitesimal transformation (3.1) represent a rigid body motion is
(3.3)
lXii
(i,j = 1,2,3).
-ai',
+ a32A.
This transformation can be written as the vector product of the infinitesimal rotation vector Cd = e."Wj and the vector A, namely I
IlA=CdXA=
if we take
\
WI"" (l32
(3.4)
w. ==
W3
==
(l13
Mi = A: - A,
or
Ox;
ox~
+ &Ai =
xn
Ilx~
+ (Cd X A),.
I We reeall that when a rigid body rotates with the angular velocity 0, the linear
velocity ... is ... - 0 X A and 8A - 0 X A at - .. X A, where", _.Q Uis the infinitesimal angle of rotation.
11
ANALYSIS OF STRAIN
Then the rigid body portion of the infinitesimal affine transformation (2.1)
can be written as
(3.5)
8x, = 8xt
- ",.(Xt - x~) + "'2(X, - xl).,
6x. = 8xg + "'3(X, - xY)
- w,(xa - x~),
{
6X3 = 6x3 - "'.(X1 - xV + ",,(x. - xi).
The quantities oxf "" x2' - x? are the components of the displacement
vector representing the translation of the point PO(x) (see Fig. 1), while
t,he remaining terms of (3.5) represent rotation about the point po.
At the beginning of this section, we proposed the problem of separatirig
the infinitesimal affine transformation oA, = a;,;Aj into two component
transformations, one of which is to represent rigid body motion alone; we
have seen that this rigid body motion corresponds to a transformation in
which the coefficients are skew-symmetric; that is, lXij = -aj'. Now,any
set of quantities a'j may be decomposed into a symmetric and a SKewsymmetric set in one, and only one, way.1 We can thus write
aij
Yz(ai;
+ aj,) + Yz(ai;
- aji).
oA,
= a'jA; = [Yz(lXi;
+ aj,) + Yz(lXi;
- aji)]A j ,
or
(3.6)
where
eij =
Wij
eji
-Wji
==
==
Yz (ai,
aji) ,
%(aij - (Xji).
/JA, = evA;,
12
&A
evA,Ai
A=~'
OA
A = ell
(4.2)
Thus, the component ell of the strain tensor rep1'esents the extension, or
change in length per unit length, of a vector originally parallel to the Xl-axis.
~a
FIG. 2
Hence, if all components of the strain tensor with the exception of el1
vanish, then all unit vectors parallel to the Xl-axis will be extended by an
amount 611 if this strain component is positive and contracted by the same
amount if ell is negative. In this event, one has a homogeneous deformation of material in the direction of the :kaxis. A cube of material whose
edges before deformation are I units long will become a rectangular
parallelepiped whose dimensions in the x1-direction are 1(1 + ell) units
and whose dimensions in the directions of the Xr and x,-axes are unaltered.
A similar significance can be ascribed to the components e22 and e83.
In order to interpret geometrically such strain components as e23, consider two vectors A = etAs and B = eaR. (Fig. 2), initially directed along
the Xr and x,-axes respectively . Upon deformation, these vectors become
A' = el BAt
BI == el &B t
CII", ... ( ..
13
ANALYSIS OF STRAIN
'" aA I aBI
:, A2 liB.
cos
(4.3)
e=
A'B'
A'B'
+ (B. + liB,)'
As = BI = B. = 0,
,xa
oB 2
= eB.,
6B~B'
liAs
= e23A .
cos (~ - a )
.
= au
liB.
__ ~
_. (/
I
~ ~
I
I
I I
II
liA.
"P
,-;, IiAa
sin a ..
= If; + L =
2e
fI,
or
Hence a positive value of 2e23 represents a decrease in the right angle
between the vectors A and B, which were initially directed along the positive Xr and x.-axes.
Again, from (4.4) and Fig. 3 we see that
LPOP' :, tan POP' =
1i:
= 6.s,
, &B
Since the angles POP' a.nd ROR' a,reequa.l, it follows tha.t, by rota.ting the
parallelogram R'OP'Q' through an a.ngle 623 about the origin, one can
obtain the configuration shown in Fig. 4. Obviously it represents a. slide
or a shear of the elements parallel to the zlxrpiane, where the amount of
slide is proportional to the distance x, of the element from the xlxrplane.
14
A similar interpret.ation can obviously be made in regard to the components e12 and e31.
It is dear that the areas of the rectangle and the parallelogram in Fig. 4
are equal. Likewise an element of volume originally cubical is deformed
into a parallelepiped. and the volumes of the cube and parallelepiped are
equal if one disregards the products of the
changes in the linear elements. Such deformation is called pure shear.
The characterization of strain presented
in Secs. 3 and 4 is essentially due to Cauchy.
It should be noted that the strain components e'i refer to the chosen set of coordinate
.r2
axes; if the axes are changed, the eij will, in
general, assume different values.
FIG. 4
5. Strain Quadric of Cauchy. With eac):;
point PO(XO) of a continuous medium, we shall associate a quadric surface,
the quadric of deformation, which enables one to determine the elongation
of any vector
A
= e,(x,
- x?)
and constrain the end point P(x) of the vector A, as yet unspecified, to
lie on the quadric surface
(5.3)
= k',
where k is any real constant and the sign is chosen so as to make the surface real. Comparison of (5.3) with (5.1) leads to the relation
(5.4)
From (3.4) we see that the extension of anllline througk PO(XO) ill irwer8elg
along the line frem the
point (~O), at wkick the Btrain is being Bttldied, to a peint (x) on the qttadric
15
ANALYSIS OF STRAIN
x:
_ _x_' I~_:_
x~
111
x~
z.,
X~
131
112
113
122
I
133
1,2
in which l;j is the cosine of the angle between the x;- and the xraxes.
old and the new coordinates are related by the equations
x, = lllx~
x. = lux;
x, = lux;
The
+ 121X; + lux;,
+ lux; + 1,,,:1:;,
+ 1.,,:1:; + laax;,
Xi
in~'erse
l...z~.
x:
k2
e;tl...lf1iX~
1
= e~~:.xft,
16
(5.9)
6/1
= UJ~J'
aG
-a
x,
e;j%J
==
~A,.
But~ are the direction ratios of the normal v to the quadric ~rface (5.5)
"x.
at the point (x,), and it follows that the vector 6A is directed along the normal to the
2
Af"--~v
plane tangent to the surface eifX;%j = k
(see Fig. 5). This property of the 8t~aiU
quadric will prove useful in. the Ilext sectIon,
~\:..""~ ~"" ~~ ~"" 1>~",<6.1>~\. 'U:ea.. 'If. t,ru._
quadric surface and their signifieance for the
FIG. 6
deformation.
S. Principal Strains. Invariants. We seek now the direcfion ratios of
the lines through (X O) whose orientation is left unchanged by jlhe deformation SA, == e;;A;. If the direction of the vector A is not aH,ered by the
strain, then ~A and A are parallel and their components are pfoportional.l
Therefore
or
(6.2)
ThhI set (6.2) of three homogeneous equations in the u"mowns AI
ptlI!S6IiIIteI a nonvanishing solution if, and only if, th", determJ.nant of the
t In other WOrds, the directions we seek are those of the axes m the quadric (6.6);
that., we wek the directions yielding the extreme values at t-he.elQaptiOlIIlS
17
ANALYSIS OF 8'l'IlAIN
lev -
(6.3)
elql == 0,
or
en-e en
en
\
ell
en - e en
= O.
\ en
e"
en - e
We prove next that the three roots el, eo, ea of this cubic equation in the
elongation e are all real.
Let the three directions determined by the three numbers ej be given
i
by the vectors l A.
e1 A j
ej.A .
12
12
12
(6.5)
wherej and k have been interchanged and the symmetry of e;. exploited.
Comparsion of (6.4) and (6.5) shows that
1
(el - e.)AjA; = O.
(6.6)
Now if we assume tentatively that (6.3) has complex roots, then these
can be written
where 'El' E I , e. are real.
O-j -'
A;A; == (a;
= af
+ ibs)(a; - Wj)
+ iii + al + bf + bi + b: ~ O.
Hence it follows from (6.6) that e, - e, ... 2iEs = 0, or E. "" 0, and the
roots I!i are all real.
l The indexi over It. indicates not the jth compo_t but rather the jth vector and ita
depeude_ ul)OIl the root 'I of the determinantal equation (6.S).
18
From (6.6) it follows that, if the roots el and e, are distinct, then
1 1I
A,A, = A A = 0,
j
the extensions of the vectors A in the principal directions, are termed the
principal strains.
We have seen that at any point (XO) there are three mutually perpen-
.i
dicular directions (assuming, for the moment, that the eo are distinct)
that are left unaltered by the deformation; consequently the vectors
i
A, the deformed vectors A + liA, and 8A are collinear. But (5.11) shows
that liA is always normal to the quadric surface (5.5), and therefore the
principal directions of strain are also normal to the surface and must be
the three principal axes of the quadric ei.jXixi = eA 2. If some of the
principal strains Ii; are equal, then the associated directions become
indeterminate but one can always select three directions that are mutually
orthogonal. If the quadric surface is a surface of revolution, then one
1
direction A, say, will be directed along the axis of revolution and any two
I
where 6 1, "2, ". are the sums of the products of the roots taken one, two,
and three at a time:
(6.8)
"I
{
+ +
+ +
= e1
e2
e. "" 6,
U2 = ese.
esel
ele.,
" .... 'lete
,fty expanding the determinant (6.7), we see that these expressions "Can
be written as
19
+ 6.1 + 6 .,
+ eu6'u + ellell -
iJs = 61tBaa
(6.8)
I I
en enl
eu en
eu ell
= e.. e.. + eu en + e12 e22'
". = eueue,s + 2elle..e31 - en4. - ene:l - eaaef.,
eu
eu
e12
The expressions for '" and ". can be written compactly by introducing
the generalized Kronecker deUa, B''!o~:::, which we now define. If the subscripts p, q, r, . . . are distinct and if the superscripts i, j, k, . . . are the
same set of numbers as the subscripts, then the value of 0:/':::: is defined to
be + 1 or -1 according as the subscripts and superscripts differ by an even
or an odd permutation; the value is zero in all other cases. We can now
rewrite the formulas (6.9) in the form
{} =
(6.10)
(i = 1, 2, 3),
eii,
""
fl. =
21 B'Jqe",.eq;,
(i, j, p, q = 1, 2, 3),
". =
~ o~:repie ojerk,
(i, j, k, p, q, r = 1, 2, 3).
Since the principal strains, that is, the roots e" e., e. of (6.7), have a
geometrical meaning that is independent of the choice of coordinate
system, it is clear that ", "" and ". are invariant with respect to an
orthogonal transformation of coordinates. [Note that this invariance
could have been used to derive expressions (6.8) from (6.9).]
The quantity " has a simple geometrical meaning. Consider as a
volume element a rectangular parallelepiped whese edges are parallel to
the principal directions of strain, and let the length!! of these edges be h,
I" l.. Upon deformation, this element becomes again a" rectangular
parallelepiped but with edges of lengths ll(l + el), l2(1 + e.), l.(l + e.).
Hence the change oV in the volume V of the element is
OV = I,Z.za(l
= 1Il.l.(el
~l
+ ell + e.)
eo.
- hZ.z.
Thus
BY
and the first strain invariant" represents ;he expansion of;i unit ~olume
20
due to strain produced in the medium, For this reason , ill CAlled the
cubical dilatation or simply the dilatation,
PROBLEMS
1. Determine the principal directions by finding the extremlll values of
Note that the x,/ A - P, are the direction cosines so that e - 6;/PiPj, }'Iaximize this
Bubject to the constraining condition
1.
1I. Refer the quadric of deformation to a set of principal axes, and dismISS the nature
of deformation when the quadric is an ellipsoid and when it is a hyperbOloid, Draw
appropriate figures and note that if 61 > 0, e. > 0, e. < 0, then, depending on the
direction of the vector A from the origin of the quadric, one must consider the surfaces
e1~ + es4 - le,lz: - kt ,
.,V, -
6A,
",;-+2lX;i- + -a.;--21%;;)
= a.;Aj = (- Ai
= (e;j
+ "',j)A;;
the 6;J and "'ii were constants and so small that their produ()ts could be
neglected in comparison with their first powers, Now we <,onsider the
general functional transformation and its relation to the affine deformation,
Consider an arbitrary material point PO(x~, xi, xB) in a continuous
medium, and let the same material point assume after deformation the
position PO/(X~', xt',
(see Fig, 1). We denote the small d1splacement
of the point po by
xn
u,(x, xi,
~)
== x2' - xf.
21
ANAL'iSIS OF STRAIN
Ai
= xi -
+ At, xi + At, x~ + A
(au,)
ax;
3) -
u;(x,
4, x~)
Ai
0
plus the remainder in the Taylor's expansion of the function u,(xY + At,
xg + Aa). The subscript zero indicates that the derivative is
xi + A 2 ,
::; ==
U;,j,
and the subscript can be dropped without confusion, since we shall deal
only with vectors at po. If the region in the vicinity of po is chosen
sufficiently small, that is, if A is sufficiently small, then one has the
formulas analogous to (7.1),
<lA, = u,.;Aj.
(7.3)
U',j'
Now if we assume further that the displacements u" as well as their partial derivatives, are so small that their products can be neglected then
(7.3) defines an infinitesimal affine transformation of the neighborhood
of the point in question. Hence the considerations of the earlier sections
are immediately applicable; the transformation (7.3) can be decomposed
into pure deformation and rigid body motion,
(7.4)
6A = u;.A =
.1
1
<=
(u"j
where
(7.5)
#I;d = ~(U;,j
+ UJ,'),
C/J;j
= ~(u;'J -
UJ.).
41 .;. C&lJZJ -
t.oIJX!,
Ut
==
lis
f1ttSl,
t.oIJXl,
22
where c.), is the infinitesimal rotation vector about (0, 0, 0) and the Go are
constants representing a trsnslation. It is clear that the transformation
defined by (7.5) is in general no longer homogeneous, inasmuch as both
the strain components /3;; and the components of rotation c.)'J are functions
of the coordinates of the medium. The dilatation
" = ell
OUI
= OXI
+ Eiu + en
+ OU, + oU.
ox. = U;
OX2
== x,
X2 = y,
ell
= ea:s,
en = e"", etc.,
and denote the components of the displacement vector (Ul, Us, u.) by
(u, II, w). The components of the strain tensor become:
AU
8v
ew. = oy'
e... = ox'
eo. = !20y
(ow + OZ~),
=!20z
(au + oXow),
...
owoz'
e"" =
!20x
(811 + 8y
au),
1 Some of the important relations of vector analysis will now be written in tensor
notation. In cartesian coordinates the divergence, gradient, and Laplacian operators
can be written as follows:
div A .. v A _ aA. + aA, + "A, _ aA, _ A',l,
ax.
a.,
grad ." - V." -
ax,
ax,
a".,
at.,
ax,
ax, .. "",
a"."
as.,
tA.,.tltr - tA,r;du,
where du is an element of are..., dT is an element of volume, and " is the exterior normal
to the surface fT.
L"'." tltr... L
'1','" duo
23
ou av iJw
"=e,...+ew+ e.. =-+-+_
ax a1l az
'" =.!:2
(J)ij
read:
(aw _~),
OZ
i)y
",". =
~(:~ -
=.!:2
(au _axaw),
OZ
:;}
+ e.. + e..
= 3e.
e.., = el~l'
{ e.. = elul It,
e.. = elf!,
e = elul13,
eu = elf.,
en = el 13ltt.
Thus, a simple extension in the direction (Ill, l12, I13) may be specified in
any x, y, z coordinate system by means of the six strain components given
in (8.2).
c. Shearing Strain. Let the equation of the strain quadric when
referred to the x', 1/', z'-system of coordiuates be given by
(8.3)
the directions of the 3/- and y'-axes. This is the equation of a hyperbolie
eylinder l18ymptotic to the x'z'- a.nd y'z'-pl&nes. The equation of the
quadric (8.S) l181!umes the form
u' -
8Y'
k'
when the axes are rotated through an.angle of 45 about the z'-axis.- A
comparison of this equation with the general equation of the strain
quadric when the latter is referred to the principal axes of strain,
ez.;c'
+ e""y' + e.z' =
k',
+ e..y2 + 2e".xy
P,
corresponding to
in the x',
e:".. and
PROBLEMS
1. Verify the invanance of the functions
in the case of simple estension.
{J, {J"
and ".
lsee Eqs.
a. Find the dile.tation and the prineips.l strs.ins, and deserihe the strain quadric for
the case of simple extension.
S. Show that the examples of strain given in this seetion ce.n he described by the
fonowin( dispJa.eement components:
a. Uniform dilats.tion, v .. n, II = 61/, W - ez.
b. Simple extension, v' .. n', If' - Wi - O.
c.. Shee.ring strain, ",' -= 2Iy', v' - tJ1' .. O.
d. Plane strain, u - v(.:a:. 11), " .. 11(2:, 11), w = O.
'" Show th.at in the examples of strain given in this sect,ion the rotatioa components
are given by:
Uniform dih1.tation, Wn - . . . . . . . . . . - O.
0. 8iDlPle~, _:. - ~... w~ - (l~
$.
d. Plaueetr&in..... -
~ (~ -
o.
"':. -
ii)..... - - - 0.
9. Notation. The values of the shear components e,..., e...., e.,. of the
strain tensor e.;, defined in (7.5), differ from the quantities e.." e.., e.. used
hy Love, l who writes
e
..
au
ihI
-+_.
ax
ay
The factor ~ was inserted in the formulas (7.5) in order that the set 'of
quantities may transform accorqing to the tensor laws.
Trefftz2 writes for the components of his strain tensor
'Y... = 2
au
ax'
aw
'Y.. =
ay
'Y"" == 2
ihI
+ liZ'
'Y... =
au
ay'
au
'Y.. = 2
az'
av
aw
az + ax'
aw
'Y.. =
lJU
ax + ay'
==
/Ju
ax'
for the components of normal strain and agrees in notation with Trefftz
for the components of shearing strain.
llBF.BllENCBS FOR COLtATBRAL llBADING
A. E. H. Love: A Treatise on the Mathematical Theory of Elasticity, Cambridge
University Press, London, Sees. 1-14, pp. 32-46.
R. V. Southwell: Theory of Elasticity for Engineers and Physicists, Olcford University Press, New York, Sees. 292-307, pp. 285---297.
A. G. Webster' Dynamics of Particles and Rigid Bodies, Verlag von Julius Springer,
Berlin, See. 169.
(10.1)
will be looked upon in this section as a system of partial differential equations for the determination of the displacements u; when the strain components e;; are prescribed functions Of the coordinates. We shall discuss
first a necessary condition for the uniqueness of the solutions u. of Eqs.
(10.1). Thereupon we shall raise the question:
1 A. E. H. Love, A Treatise on the Mathematieal Theory of Elasticity.
IE. "I'retttI, HaI:Idbuch der Phyaik, vol. 6, Meehanik der ela.sti~.hen g6rper.
S. 'rimoBbeQko a.nd J. N. Goodier, Theory of l!laetieity.
26
Wha.t restrictions must be placed on the given funetions e;;(xl, Xt, %.)
to ensure the existence of single-valued continuous solutions U,(Xl'
x" XI) of Eqa. (10.1)?
it is clear ,rat of all that specification of the e;j does not determine the
displacementi u; uniquely, for the strain components characterize the pure
deformation ,'of the medium in the neighborhood of the point (x), while
the functions 1'< may involve rigid body motions which do not affect the ~-;.;.
In fact, if one obtains some solution
(10.2)
of the system (10.1), and if PO(x, x~, xg) is an arbitrary point of the body,
then the addition to the right-hand member of (10.2) of the terms'
(10.3)
u; = u~
+ ",MXk -
x),
representing the motion of the body as a whole, will not affect the values
of the prescribed components of strain entering into (10.1). It thus
becomes clear that the solution of the system (10.1) cannot be unique
unless one specifies the components of displacement uS and the components of rotation ",1j of some point po of the medium, and we shall suppose
in the following discussion that this has been done.
Inasmuch as there are six conditions imposed on the three functions
u; by Eqs. (10.1), one eannot expect in general that the system (10.1) will
possess a solution for an arbitrary choice of the functions e'j. We seek the
further conditions that must be imposed on the functions if the system
of Eqs. (10.1) is to possess a solution for the triplet of functions u,.
The fact that the strain components e,; cannot be prescribed arbitrarily
can be seen from the following rough geometrical considerations: Imagine
that a body .. is subdivided into small volume elements, which in the
interior of .. may be assumed to have the form of cubes. The strain components e;j are given on the faces of each cube, and the displacements u;
of those faces are to be calculated. If each individual cube is subjected
to a deformation so that it becomes a parallelepiped, then it may happen
that it is impossible to arrange the parallelepipeds to form a continuous
distorted body .. '. The points that were coincident on,the interfaces of
the eubes may no longer coincide on the interfaces of the parallelepipeds.
In fact, there may even be gaps between the pairs of initially coincident
points. The requirements of continuity and single-valuedness imposed
on the components of displacement place some restrictions on the choice
of the strain components e"j if the differential equations (10.1) are to
posseIIB solutions.
Let P"(zt, xl, 4) be some point of a simply connected region' .., at
1 Of. formulaa (3.5)
e"
A tegion of space is said to be simply connected if fWery eloaed curve drawn in the
be shrunk to a point, by continuous deformation, without paasing out of the
l'IIIion _
27
ANALYSIS OF STltAIN
which the displacements uJ(x!, x~, xl) and the components of rotation
We determine the displacements 'UJ at any other
point P'(x~, x~, x~) in terms of the known functions I eo; by means of a line
integral over a simple' continuous curve C joining the points po and pI:
r::
f:
d'UJ = u~ +
UI . dx.
+ jPfI
(P' elk dx. + jpo
(P' "'I. dx.,
x~)
+ f: (x~ -
and hence
(lOA)
uJ
+ (x~ -
An integration by
x2)"'7.
X.)"'ik,l dx"
+ f: [ejl + (x~ -
x.)w;. 1] dx,.
a 1
2 ('UJ.
==
aXI
2 ('UJ.kl
a 1
ax. 2 (UI,; + U;,I)
- u.;)
U.,;I)
+ 21 (UI,I. -
UI,;k)
a 1
ax; 2 (U.,I + Ul,k),
It follows
(10,5)
u~
6j;.1 -
8kl(I!;,.k -
+ (x~ -
ekU)
+ ekl.;;)
= O.
The first line of (10.8) vanishes identically, and since this equation must
be true for s.n s.rbitrll.ry choice of x~ - x~, it follows ths.t
(10.9).
eij,kl
+ ekl.ij -
Bik.;l -
Elil." = O.
(10.10)
ANALYSIS OF STRAIN
29
11. Finite Deformations. The preceding sections of this chapter contain all the principal results of the classical theory of infinitesimal strain.
It is clear from the general discussion of the affine transformation in Sec. 2
that the linearization of the equations appearing there led to a consideration of infinitesimal transformations that permits the application of the
principle of superposition of effects. Many technically important problems in elasticity, includipg those of buckling and stability, call for a consideration of finite deformations, that is, deformations in which the displacements u together with their derivatives are no longer small. This
section contains only a brief introduction to the theory of finite strains
and provides an admirable iilustration of the complications that appear
in the development of a theory when the fundamental equations bee orne
nonlinear.
There are two modes of description of the deformation of a continuous
medium, the Lagrangian and the Eulerian. The Lagrangian description
employs the coordinates a; of a typical particle in the initial state !1S the
independent variables, while in the case of Eulerian coordinates the independent variables are the coordinates Xi of a material point in the
deformed state.
In the preceding sections, we have used the Lagrangian viewpoint as
the natural means of describing the deformation of the neighborhood of
the point (al' a2, aa). When we come, in the next chapter, to the discussion of the stresses acting throughout the medium, we shall find that these
stresses must satisfy equilibrium conditions in the deformed body and
hence Enlerian coordinates are indicated. In this section, we shall
describe the deformation from both points of view, and we shall see that
when the deformation is infinitesimal (that is, when products of the
30
dB: = daf
+ dai + dai =
do.; do.;,
and
(11.2)
0.;
while ds' = dx, dx; = Ojk dxj dx.. It is evident that the equality of ds'
and ds: for all curves Co is the necessary and sufficient condition that the
transforma~ion 0.; = o.;(XI, x" x.) be one of rigid body motion; hence we
shall take the difference ds' - d8~ as the measure of the strain and write
(11.3)
From the expressions given above for ds' and dsg, we get
21);' =
~jk
o.;,,.a;,"
0,.. -
'Uj,k -
Uk,;
+ u.".l;,'
and hence
(11.4)
31
ANALYSIS OF STRAIN
daj da..
ajk
(11.5)
and since
x;
a;
+ U;, we have
(liij
X;,jX;,k =
+ U;,j)(liik + U;,k)
liik
(Uj,k
and
ds 2
d8~
with
(11.6)
2E;. = U;,k
+ Uk,j + U;,jU;,
In order to exhibit the fact that the differentiation in (11.4) is carried out
with respect to the variables Xi, while in (11.6) the a; are regarded as the
independent variables, we write out the typical expressions for '/Uj and E;j
in unabridged notation,
~ =
E04
:: -
(::Y
2
Eab
av ~
iJx iJy
+ oW aw),
+ (iJU iJu +
iJv ~
iJa iJb
+ ~l1! ?!!!),
au
iJy
av _ (au au
iJx
iJx iJy
iJu
iJb
iJv
iJa
'1..
iJa iJb
ax iJy
iJa iJb
It was shown in Sec, 4 that ell, e22, and e33 can be interpreted as extensions of vectors originally parallel to the coordinate axes, while en, e.a, and
eu represent shears or changes of angle between vectors originally at right
angles. When the strain components are large, however, it is no longer
possible to give simple geometrical interpretations of the functions E;j
and"lij.
Consider a line element with dso = da da. = daa = 0, and define the
extension E, of this element by E, = (ds" - dso)/ds., or
ds = (1
(11.7)
+ E,) ds.
or
(11.8)
E, ""
VI + 2eu -
1.
32
When
Ell,
d8 dB cos 9
X :x...1
If a23 = 1(/2 - fJ denotes the change in the right angle between the line
elements in the initial state, then we have
.
sm au =
and by (11.7) and (11.8)
.
(11.9)
sm a., =
Again, if the strains
Eij
da. dill
2ft3
d8 ds '
2. 23
---;===-:;:_===
VI + 2 VI + 2
t.b..~a
= Y2(U;.i
+ 'Uf).
The ratio of the volume element in the strained state to the corresponding element of volume in the unstrained state is equal to thll functional
determinant
'
(11.10)
I I"" I
ax.
au.;
a(a; + '14)
au.;
= 1&
+ u;,A.
+ e,. +
33
ANALYSIS OF STRAIN
tions u" where the components of strain 'Ii.; (or 6(;) are prescribed functions.
Since these equations are nonlinear, the problem of integration is much
more involved. While it is not difficult to formulate the conditions on
the function '1/;; (or fiJ) if the set of Eqs. (11.4) [or (11.6)] is to possess a
solution with suitable properties, this will not be pursued here. 1
In concluding this brief treatment of finite strains, it should be emphasized that the transformations of finite homogeneous strain are not in
general commutative and that the simple superposition of effects is no
longer applicable to finite deformations. These facts are responsible, in
the main, for the absence of satisfactory solutions for all but the simplest
problems, such as homogeneous strain, simple tension, and torsion of an
elliptical cylinder, which become trivial when the equations are linearized.
REFERENCES FOR COLLATERAL READING
L. Lecornu: Theorie mathtlmatique de l'elasticittl, Memorial des sciences mathematiques, Gauthiers-Villars & Cie, Paris, pp. 34-40.
A. -E. H. Love: A Treatise on the Mathematical Theory of Elasticity, Cambridge
University Press, London, Sees. 32-40, pp. 6~73.
E. Trefftz: Handbuch der Physik, Verlag von Julius Springer, Berlin, vol. 6, Secs. 8-9.
1. S. Sokolnikoff: Tensor Analysis, John Wiley & Sons, Inc., New York, pp. 200-305.
F. D. Murnaghan: Finite Deformations of an Elastic Solid, John Wiley & Sons, Inc.,_
New York, pp. 27-42.
PROBLEMS
1. Show that a tensor a'i can be decomposed into a symmetric tensor e'i = el' and
a. skew-symmetric tensor Wii = -Wi' in one, and only one, way. Hint: Assume that
the decomposition can be made in two ways:
2. From aA, = "'IA I , find aA and aA for a vector lying initially along the x-axis,
that is, A = lA, and justify the statement of Sec. 4 that in this case
at
en.
Does aA
x;
x;
e;1 = ~(U'.I
+ UI,')
34
that the equations of compatibility are necessary conditions for the exj8tenoe of continuons single-valued displacements. Hint:
e;f.loI -
}i ('It'.lil
+ 'ltt.til)
and, by interchange of i, k and likewise of i, l, e""1 = }i(U. lii + UI.",,) Add these,
.
interchange j and k, and show that the compatibility equations (to.9) result.
'1. ,Show that the shear strain en, for example, can be interpreted as th~ extensIOn
of the diagonal OQ of the rectangle OPQR (Fig. 4), provided the rectanPie 18 a square.
CHAPTER
ANALYSIS OF STRESS
F d'T, or
R. = iF; d'T.
(12.1)
The resultant moment M = e;M, due to the body force F can be written
as the integral over 'T of the vector product of the position vector r = e,z.
and the force vector F; that is, :M ""
M. =
(12.2)
1 The
alternating tensor
d".
i (r X F) dr, or
e;pa;Fk dr.
40'-+0
per unit area of the surface acting at the point (x.) and is called the stre88
vectQr.
T.
Let P(x) be any point in the medium and the stress vector acting on
an element of surface at P, with the normal v. Draw through P three
planar elements parallel to the coordinate planes, and pass the fourth
plane ABC (Fig. 7) normal to v and at a small distance h from P.
I
a.
}of, -
J. (x,F. -
x';") dT.
ANALYSIS OF BTBESS
t
P ABC and by T and T thtl stresses acting on the faces PAC and P AB,
i
.respectively.
e/A.
It will be convenient
(13.1)
so that
i
T = e,"1';;.
We shall show that the nine scalar quantities Tij are the components of a
tensor, the stres8 tensor, and that the T;; serve to determine completely the
c
o
FIG. 7
state of stress at the point P. The stress vector can then be calculated
from the Tij for any orientation v of the surface element at P. The meaning of the SUbscripts in the components Tij should be carefully noted.
Observe that in r23, for example, the first 8uhscript, 2, indicates the
2
coordinate axis normal to the element of area on which the stress T acts,
while the second subscript, 3, indicates the direction of the component of
this stress vector~
If the volume element is taken in the shape of a rectangular parallelepiped, with faces parallel to the coordinate planes, and
38
vector acting on .. face of the parallelepiped perpendicular to the ~,
the components 1'# are shown in Fig. 8. The convention in regard to the
signs of the scalar quantities is the foHowing: If one draws an exterior
normal to a given face of the parallelepiped, then the components Til are
reckoned positive if the corresponding components of force act in the
directions of increasing Xl, ;1::, X3 when the normal has the same sense as
the positive direction of the axis to which the face is perpendicular; if,
on the other hand, the exterior normal to a given face points in the direction opposite to that of the positive coordinate axis, then the positive
values of the components 1',; are associated with forces directed oppositely
to the positive directions of coordinate axes. The arrows in Fig. 8 indicate vectors representing forces which, for positive values of the 1'lj are
%2
FIG. S
i.
Let T"
Till
39
ANALYSIS OF STRESS
body force a.t the point P; then, on a.ecount of the assumed continuity of
the stress vector Ti , the x.-component of the force acting on the face ABC
of the tetrahedron is
(T,. + ;)(1',
where lim
,,__.o ;
= O.
The corresponding
because the exterior normals to the faces of areas (1'; are directed oppositely
to the direction of increasing x,-coordinate. Finally, the contribution
of the body force F; to the x,-component of the resultant force is (F, +
E;)73'htT, where 73'hu = /1'1' is the volume of the element P ABC and
lim = O. Thus, for equilibrium of the tetrahedron we must have
h-O
<
(T: + E,)U +
(13.2)
(-'I'ji
+ Ej,)Uj +
(F,
+ ED73'htT
o.
If in (13.2) we set U; = uV; = U cos (x;, v), divide through by (1', and pass
to the limit as h -> 0, we get
(13.3)
It is clear from (13.3) that, having specified the components of the
stress tensor '1'" at any point P(x) of the medium, one can calculate the
. .. ,
X.,
'1'38 ""
Z .
Most American writers (as well as many Russian and German authors)
write
d ll
Tn,
tI.
= 1'83
for the normal stresses and denote the remaining six tangential, or shear,
stresses '1'12, TU, etc., by '1'.." '1'.., etc.
The notation
'I'll =
xx,
.. " ,
'1'33
has been suggested by K. Pearson and is quite convenient when one contemplates using orthogonal curvilinear coordinates. When it appears
desirable to exhibit the dependence of the components of the stress tensor
on the x, 7/, z-system of coordinates, we shall write 'I'll = '1'.., '1'" = Till"
TU =< '1'.." Ttl = '1'"", etc. In this notation, formulas (13.3) read:
f" ...
'1'"" 008
(:1:, p) -I;
'1'". COllI
(1/, p)
p),
TI/ == '1'"" COS (:1:, .) + cos (11, p) + '1'.. COS (z, II),
T. = '1',.. COS (:1:, .) + '1'... COS (II, .) + '1'.. COS (z, ,,).
'1'l1li
From the definition of the stress vector, it follows that the stress vector
or
In the cgs system, the stress is measured in dynes per square centimeter,
while in English practical units it is measured in pounds per square inch
or in tons per square inch.
REFERENCES FOR COLLATERAL READING
A. E. H. Love: A Treatise on the Mathematical Theory of El""ticity, Cambridge
University Press, London, Sees. 41-48, pp. 74-80.
E. Trefftz: Hsndbueh der Physik, Verlag von Julius Springer, Berlin, vol. 6, Secs. 1-3.
R. V. Southwell: Theory of El""tieity for Engineers and Physicists, Oxford University
Press, New York, Sees. 258--268, pp. 259-264.
f. F, d1" + 1. T, au
0,
1";1
',,:1:.
!"i< are continuous
and single-valued in '1', then the Divergence
.
41
"ii.' =
(15.3)
-F"
or, when written out in full in the notation explained in Sec. 14,
(1.,._
+ (1.,. + (1.,..., =
-F
z,
(1.,...
+ (1.,.oy.. + (1.,.oz =
-F
(1.,.",
+ (1.,.,,<
+ OTOZ.. =
oy
-F
,.
OX
oX
ox
oy
oz
.,
80
e;;kTjk d.,. = 0,
e;;."'j. = O.
is arbitrary, we
42
or since
4-1,1'21 = Tn;
and
In short,
(15.6)
that is, the stress tensor is symmetric. The symmetry of the components
of the stress tensor allows us to write (15.3) as 1',;,; = -F. or, recalling the
definition (13.1),
(15.7)
TJ,; =
div T = -F,.
Since the nine stress components 1'# are bound by the threc relations
(15.6), we sec that the state of stress at any point is completely characterized by the six quantities 1'lI, 1"2, 1'33, 1'12 = Tn, 1'23 = 1'32, 1'31 = 1'13.
It follows from the foregoing that the six components of stress must
satisfy the three partial differential equations (15.3),
in the interior of the medium and that on the surface bounding the
medium they must satisfy the three boundary conditions (13.3),
+ Mi
-- O.
What can be said in this case a.hout the symmetry of the straM components? See in
this connection Eric ReislJner, "Note on the Theorem of the Symmetry of the Stress
T_r," J<iUNUJl oj MatItematics and Phyflc8, vol. 23 (1944), pp. 192-194.
18. Transformation of Coordinates. The symmetry of the shear components of the stress tensor (1';; = 1';;) established in Sec. 15 is but a special
ease of a general theorem that will prove useful in establishing the laws 0(
transformation of the components 0( the stress tensor under an orthogonal
transformation 0( coordinate axes. We prove the following theorem:
ANALYSIS OF STRESS
Let the BfJ,rface element8 &T and &T', with unit normal8l' and
thruugh the point P; then the component of the stress vector (acting
>'
on; &T) in the direction of v' is equal to the component of the stress vector T
(acting on AU') in the direction of the normal v.
In vector notation, the theorem reads:
THEOREM:
.', pa88
f. v = f. v'.
(1~.1)
The proof of the theorem employs only Eq. (13.3) and the symmetry of
the stress components. For
If'
,,'
T.v
TiP,
(TijVi)Vj
= TjiViV,
=
TiJli = T. v',
" v
T"
(16.2)
= TH";"i,
x:
(16.3)
x~-axis
x~-axis.
Thus,
Then
)I: =
II;
cos (x~,
Xi' ==
laiJ
and we get
(16.4)
(16.5)
The law of transformation (16.4) is identical with that deduced in Sec. 5
for the transformation of the strain tensor and exhibits the tensor character of the quantities 'Tv. Indeed, these equations represent the trans-
1'11
1'22
+ Til
.'.,
T = T
then the stress vector on any plane Q that contains T lies in the plane P.
I. Show that the symmetry of the stress components T<; = Tj' follows from (16.1),
.'
Ty = T'''',
P
Q
I. If T and Tare thc stress vectors at a point and acting across planes P and Q, find
R
,"" d
~w
121
d8
~;:dB
vv
T,. sin
de
or
mu
iii -
-20,
That is, the normal stress across a tmrf'ace element varies 88 the element is rotated and
at a rate which is twice the shear component (with sign changed) perpendicular to the
axis of rotation.
17. Stress Quadric of Cauchy. For the purpose of studying the nature
of the distribution of stresses throughout a continuous medium, we define
at each point P(x) a quadric surface, the stress quadric of Cauchy. The
discussion of this quadric will parallel closely that of the strain quadric
in Sees. 5 and 6.
Consider an element of area with normal " and containing a point
PO(X O), and let T be the stress vector acting on this surface element (Fig.
We introduce a local system of axes Xi with origin at po, and we
denote by A the vector, in the clirection of the normal ", from po to Bome
9).
N = T " =
or since
(17.1)
Xi
TiVi
T;jViV"
= A Vi,
N A2
Ti,.XiX,..
2F(Xl,
x" x.) =
FIG. 9
'('i;XiX;.
N =
k2
Ai'
46
x:,
= N A' ==
k.
But both N and A have values that rio not depend on the particular
coordinate system used, and hence
(17./j)
Thus, the quadratic form T;;XiX; has a value that is independent of the choice
of coordinate system. In other words, it is invariant with re!!pect to an
orthogonal transformation of coordinates.
The invariance of the form r,,:x,,xh shown by Eq. (17.5), affords an easy
means of calculating the equations of transformation (16.4). For (cf.
Sec. 5) if one substitutes in the right-hand member
of (17.5) the expressions for x~ in terms of the x;,
namely,
then the resulting expression
T iJX.,x~
= l;al;Pi;x.,x~
From 2F(Xl, X" x,) = T,;X;Xj and Eq. (13.3), it is seen that
(17.6)
Thus, the quadratic form F(Xl, x" x.) has some attributes of a potential
function, since its derivatives with respect to the variables Xi are proportional to the corresponding components of force.
Since the
gent to the quadric surface (17.3) at the point P(x), we see from (17.6)
that the stress vector f is also normal to this tangent pJ.a.ne. This gives
an easy means of constructing the stress vector f from the knowledge of
its normal component N. All that is necessary is to draw the quadric
aurface (17.3) and construct the tangent pJ.a.ne to the quadric through the
terminus P(z) of the vector A (Fig. 10). Then the vector is directed
along the perpendicular POQ to the tangent pJ.a.ne. If the magnitude of
47
ANALYSIS OF STRESS
If the direction" is taken along one of the axes of the quadric, then"
(and A) will be normal to the plane tangent to the surface at (Xi). But
of is perpendicular to
>
Ti.
= TV.. = r Oi/Vi
and = TV, the constant T denotes the magnitude of the stress vector of
that acts on an element normal to the axis of the surface. For any
direction" we have
(17.8)
T 6;;";, or
(Ti; - TOi;)"; = O.
of the SurfllOO element at the center of the sphere, the stress tlxperienced
by it will be purely normal.
We recall that Tl, T2, T, are the only stresses acting on the Ilurface elements perpendicular to the principal directions ~, ;, ~, while T11, Tn, Taa
are the normal strlll!8es on elements perpendicular to the coor(iinate axes.
If the coordinate axes are taken along the axes of the quadric, then the
shear stresses T12, T23, T31 disappear from the equation of the surface
T;f,:&J = k2, which now takes the form
(17.10)
8,;1 =
-T'
+ e1T' -
e2T
+ e.
= 0,
+ T2 + T, = Tn + T22 + T33 == e,
+ TtT3 + TaTI
T22 T231
1 T11 :31 1 + I T11 T12I'
T.3 + T31
T88
T12 T.2
I T23
T,TtT.
9. =
(17.11)
e.
Tl
TIT.
'Ttl
1'u.
T1...
A reference to formulas (16.5) shows that one can write down at once
the expressions for the components of the stress tensor T,; in terms of the
principal stresses. Thus, if the direction cosines of the principal axes of
strIlI!8 X; are given by the table
Xl
X.
X3
Xl
X2
III
l21
til
In
In
I12
In
In
lu
'
X.
The ehara.eter of the di8tribution, of stress -at the point 1'0($0) -depends
on the signs of the principal stresses. (Note the agreement above conI
49
(leming the choice of the sign of k!.) If the principal stresses are all
positive, then the equation of the stress quadric has the form
1"IX~
+ 1",xl + 1"aX: =
k",
and the surface is an ellipsoid. Equation (17.4) now reads N = k ' / A "
from which it follows that the force acting on every surface element passing through the point p. is tensile. If, on the other hand, all 1"0 are negative, then (17.10) takes the form
or
depending on the orientation of the surface element at PO(XO). The first
of these equations represents an unparted hyperboloid and the second a
biparted one (Fig. 11). If the normal to the
surface element at p. cuts the biparted hyperboloid, then N = -k 2/ A', so that the stress is
compressive, while if the normal cuts the
unparted hyperboloid, then N = k'/ A', and
the stress is tensile. Vectors A that lie on the
surface of the asymptotic cone
1"IXf
+ 1",x~ - hlxi = 0
of the stress vector in the direction ", normal to the surface element, is
inversely proportional to the square of the radius vector A\' to the stress
quadric. The extreme values of the radius vector lie along the axes of
50
T1, T2,
we then have
(18.1)
T.
= 'T.Va,
and since
we get
(18.2)
But from Fig. 9
8 2 = I'fl' - N',
and on substituting in this formula from (18.1) and (18.2) we obtain
(18.3)
It is clear from (18.3) that if the directions" are taken along the axes
of the stress quadric so that
V,
1,
1,
= 1,
V2
1', =
V3
= Va = 0,
V3
V,
PI
V.
= 0,
= 0,
then 8 = O. This merely verifies the known fact that the planar elements
normal to the principal directions of stress are free from shear. Thus the
minimum (zero) values of 181 are associated with the principal directions.
To determine the directions associated with the maximum values of 181,
we maximize the function in the right-hand member of (18.3), subject
to the constraining relation II;Vi = 1. The simplest way of doing this is to
use the method of La.grange multipliers a.nd seek the free extremum of
the function
F = 8 1 - X... I',.
This leads to the three equations,
iJF
iJv;
51
ANALYSIS OF STRESS
in X and "i which, together with the relation Vi"i = 1, serve to determine
the desired directions.
We dispense with the elementa.ty computa.tions and record the final
results in the accompanying table, the last column of which gives the
values of INI associated with the extreme values of lSI.
TABLE OF EXTREMAL VALUES OF
0
0
:1:1
0
V2
2
0
0
V2
~/2
V2
2
181""
Va
"
VI
-~
0
0
0
INI
IT.I
hi
[TIl
~~lr2
- Ta!
~~[T2
+ TIl
V2
~ih
- Ttl
Hlra
+ Ttl
~~ITI
- T,I
~21Tl
+ ral
~
2
If 1'3 < T. < 1'1, SO that 1'1 is the maximum value of Nand 1'3 is its minimum value, then the maximum value of lSI is,
We see from the table that the maximum shearing stress acts on the surface element containing the X2 principal axis and bisecting the angle
between the X,- and xa-axes. If 1'2 = 1'3, there will be infinitely many
directions associated with the surface elements that are subjected to a
maximum shearing stress. We summarize the main results of this section
in the following theorem:
THEOREM: The maximum shearing stress is equal to one-half the difference
between the greatest and least normal stresses and acts on the plane that
bisects the angle between the directions of the largest and smallest principal
stresses.
The results of this .section can be further illuminated by constructing a
diagram proposed' by o. Mohr.
If we rewrite Eqs. (18.2) and (18.3) in the form
N =
8'
+ N2
". Ti"~
lOtto Mohr, ZiviJ."iew- (1882), 1>. 113. See also his book Technische
Mecbanik, 2d ed. (1914).
52
recaJI that
vI + vi + vi
VI, we obtain
(18,4)
S'
+ (N
We consider now the space of the variables (S, N) and plot in the cartesian
SN-plane (Fig. 12) the values of S as ordinates and those of N as abscissas.
FIG. 12
The equation
(18.6)
S' + (N - 'I",)(N -
'1".)
= 0
represents a circle 0, with center on the N-axis and passing through the
points ('1"2, 0), (T3, 0). Hence the region defined by (18.5) is ~rior to
the circle (18.6) and includes its boundary. Further, '1". - '1", > 0,
TI - '1"1 < 0, and we conclude from the second of Eqs. (18.4) that
(18.7)
o.
Thus the region defined by (18.7) is a closed region, interior to the circle
C, (Fig. 12), whose equation is
S'
+ (N -
'I".)(N - "1) -
o.
ANALYSIS OF STRESS
S2
+ (N -
Tl)(N - ,.,)
0,
s _ Tl
mu:: -
- Ta.
2
+ ,..,
2
and the substitution of this value and Smax = ~h - 7",) in (18.4) yields
vi = vi = ~, vi = O. These coincide with the values appearing in the
table on page 51.
PROBUMS
Discusa the Mohr circle diagram for the case where T, = n, and detennine the orientation of surface elements experiencing extreme shearing stresses. Consider also the
case where 1"1 = 1"2 = Ta.
and
T"" = T .. = ,."" = 0,
The stress quadric in this case is a sphere whose equation is
a;1
+ y2 + Zl.=
k2
_
_
T_
Any set of orthogonal axes that pass through the point po may be taken .
as principal axes of the quadric. This case corresponds to hydrostatic
pressurei,."" is negative.
b. Simple Tension Dr CompresBioo. A state of simple tension or compression is characterized by the fact that the stress vector for one plane
through the point is normal to that plane and the stress vector for any
plane perpendicular to this one vanishes. Hence if the x'-, y'-, and
z'-axes coincide with the principal axes of streM, then the stress quadric
(17.3) has the equation
Transforming to any other orthogonal coordinate system x, y, z with the
aid of (17.12), we obtain the following stress components:
'T.. = 'T,lflJ
'T.. = 'T,llll21,
'TIIJI = 'T,l~"
'T.. = 'T ,l"l",
= 'T1Z111
'T = 'T,l31l11,
where lll' l", l31 are the direction cosines of the x'-axis relative to the axes
x, y, z. A positive value of 'TI represents tension, and a negative represents compression.
c. Shearing Stress. Consider a streM
B
quadric
(19.1)
which is a hyperbolic cylinder whose elements are parallel to the z'-axis and
which represents a shearing stress of magnitude'T. Equation (19.1) takes the form
'Txt -
FIG. 13
ryl
kl ,
when the axes are rotated through an angle of 45 about the z'-axis. A
comparison of this equation with the general equation of the streM quadric
(19.2)
when the latter is referred to the principal axes of stress shows that we
must have
'Tu = 0,
Tu
==
- TW
= 1".
Thus, the shearing streM is equivalent to tension acr088 one plane and
compression of equal magnitude acr088 a perpendicular plane. This can
also be shown geometrically by considering the equilibrium of the element
PBO (Fig. 13). Hence the stress on the face BO is a pure shear of magnitude l' - -'T"" = +'T... This type of shearing stress would tend to slide
planes of the material originally perPendicular to the y'-axis in a direction
parallel to the :r!-axis and planes of the material originally perpendicular
to the :r!-axis in a direction parallel to the y'-axis.
d. Plane St.r_. If one of the principal stresses vauishes, then the
1ItNI8II quadric becomes a cylinder whose base is a eonic, the stress conic.
55
A state of stress, in this case, is said to be plcM. The base of the cylinder
lies in a plane containing the directions of the non vanishing principal
stresses. For example, if this plane is perpendicular to the ,-axis, the
equation of the quadric is
T..x
+ T..Y + 2T.,xy
k.
For simple tension in the x-direction, the stress conic reduces to the pair
of lines
x ==
kt
-_.
T ...
kl
_.
2T..,
CHAPTER
EQUATIONS OF ELASTICITY
20. Hooke's Law. It has already been noted that the treatment contained in Chaps. 1 and 2 is applicable to all material media that can be
represented with sufficient accuracy as continuous bodies; this chapter
will be concerned with the characterization of elastic solids.
The first attempt at a scientific description of the strength of solids was
made by Galileo. He treated bodies as inextensible, however, since at
that time there existed neither experimental data nor physical hypotheses
that would yield a relation between the deformation of a solid body and
the forces responsible for the deformation. It was Robert Hooke who,
some forty years after the appearance of Galileo's Discourses (1638), gave
the first rough law of proportionality between the forces and displacements.
Hooke published his law first in the form of an anagram" ceiiinosssttuu"
in 1676, and two years later gave the solution of the anagram: "ut tensio
sic vis," which can be translated freely as "the extension is proportional
to the force." To study this statement further, we discuss the deformation of a thin rod subjected to a tensile stress.
Consider a thin rod (of a low-carbon steel, for example), of initial cr088sectional area ao, which is subjected to a variable tensile force F. If the
stress is assumed to be distributed uniformly over the area of the cross
section, then the nominal8tre8s T = F I ao can be calculated for any applied
load F. The actual stress is obtained, under the assumption of a uniform
stress distribution, by dividing the load at any stage of the test by the
actual area of the cross section of the rod at that stage. The difference
between the nominal and the actual stress is negligible, however, throughout the elastic range of the material.
If the nominal stress T is plotted as a function of the extension e (change
in length per unit length of the specimen), then for some ductile metals
a graph like that in Fig. 14 is secured. The graph is very nearly a straight
line with the equation
(20.1)
T = Ee
until the stress reaches the proportional limit (point P in Fig. 14). The
position of this point, however, depends to a considerable extent upon the
sensitivity of the testing apparatus. The constant of proportionality E is
known as Young's modulus.
EQUATIONS OF ELASTICITY
57
In most metals, especially in soft and ductile materials, careful observation will reveal very small permanent elongations which are the results
of very small tensile forces. In many metals, however (steel and
wrought iron, for example), if these very small permanent elongations are
neglected (less than 1/100,000 of the length of a bar under tension), then
the graph of stress against extension is a straight line, as noted above, and
practically all the deformation disappears after the force has been
removed. The greatest stress that can be applied without producing a
permanent deformation is called the elastic limit of the material. When
the applied force is increased beyond this fairly sharply defined limit, the
material exhibits both elastic and plastic properties. The determination
of this limit requires successive loading
and unloading by ever larger forces
U
until a permanent set is recorded. For
many materials the proportional limit
is very nearly equal to the elastic limit,
and the distinction between the two is
sometimes dropped, particularly since
the former is more easily obtained.
When the stress increases beyond the
elastic limit, a point is reached (Yon
the graph) at which the rod suddenly
Strain
stretches with little or no increase in
the load. The stress at point Y is called
FIG. 14
the yield-point stress.
The nominal stress T may be increased beyond the yield point until the
ultimate stress (point U) is reached. The corresponding force F = Tao is
the greatest load that the rod will bear. When the ultimate stress is
reached, a brittle material (such as a high-carbon steel) breaks suddenly,
while a rod of some ductile metal begins to "neck"; that is, its crosssectional area is greatly reduced over a small portion of the length of the
rod. Further elongation is accompanied by an increase in actual stress
but by a decrease in total load, in cross-sectional area, and in nominal
stress until the rod breaks (point B).
The elastic limit of low-carbon steels is about 35,000 lb per sq in.; the
ultimate stress is about 60,000 lb per sq in. .Hard steels may be prepared
with an ultimate strength greater than 200,000 lb per sq in.
We shall consider only the behavior of elastic materials subjected to
stresses below the proportional limit ; that is, we shall be concerned only
with those materials and situations in which Hooke's law, expressed by
Eq. (20.1), or a generalization of it, is valid.'
I In order to give the reader Ilome feeling regarding the magnitude of deformations
with which the theory of elasticity de&ls, note that a l-in.-long rod of iron with proportionallimit of 25,000 Ib per sq in., a yield point of 30,000 lb per sq in., and Young'swod-
58
T = I(e),
where' is a single-valued nonlinear function. More frequently, however,
the loading-unloading diagram has the appearance shown in Fig. 15b.
In this diagram the curve 0 A is associated with the loading of the specimen and AB with the unloading. In this instance there is a residual
T
CI
lal
Ibl
FIG. 15
59
EQUATIONS OF ELA8T1CITY
= Fij(en,
en, .. ,en),
(i, j
= 1, 2, 3)
between the Tij and eoj and that the Tij vanish when the strains eo; are all
zero. This last assumption implies that in the initial unstrained state
the body is unstressed. Now, if the functions Fij are expanded in the
power series in e;; and only the linear terms retained in the expansions,
we get
(i, j, k, I = 1, 2, 3).
(21.1)
The coefficients Cijkl, in the linear forms (21.1), in general will vary from
point to point of the medium. If, however, the C,jkl are independent of
the position of the point, the medium is called elastically homogeneous.
Henceforth we confine our attention to those media in which the (;,;kl do
not vary throughout the region under consideration. The law (21.1) is
a natural generalization of Hooke's law, and it is used in all developments
of the linear theory of elasticity.'
Inasmuch as the components Tij are symmetric, an interchange of the
indices i and j in (21.1) does not alter these formulas, so that
Moreover, we can assume, without 1088 of generality, that the C'j'" are also
symmetric with respect to the last two indices. For if the constants
C:jkl and C:;kl are defined by the formulas
c:;>1
72(Cijkl
+ Ci;a),
C~;.l:l = 72(Cijld -
O:jlk
and
C:;kl =
Cij1cl
CijlJ:),
-C:;'k'
= C~ikl
ThuB
Cijkl
can be written as
+ C~;kh
in which the C:j11 are symmetric and the C;;kl are skew-symmetric with
respect to k and 1. Accordingly, the law (21.1) can always be written
in the form
Tij
C:;kleH
+ C:;kle",.
However, the double sum in the second term of this expression vanishes
inasmuch as ekl = eU. and C:;kl =
Thus,
-C:;11'
Tij
==
C~.i.l:,eA:l,
where the C#kl are symmetric with respect to the first two and the last two
indices.
, It is Important to note that the generaliJled Hooke'alaw (21.1) is akIo uaed in some
in'\"e8tiptionB. where the Btraina are finite, in the llen&e of Sec. 11. For many materiall
a linear relationship (21.1) holds for an appreciable range of values of the 8</. The
linear theory of elasticity, however, is baaed on the UBe of the infinitesimal straina,
dsfined in Sec. 7, and on the linear law (21.1).
60
We shall consider henceforth tha.t the CV'I in (21.1) have been symmetrized, so that there are at most 36 independent constants in the general stress-strain law (21.1).
To avoid dealing with double sums, we can introduce the notation
1"22
elt =
and write
t~e
T2t
733 ::e
e2,
e33.=
Ta,
ea,
723 =
T",
2e., = e.,
Tn
Ti,
2e .. = e"
Tn = ra,
= eo,
2eu
'1'1 = Cllel
'1'6 = CUel
(21.2)
The relations between the '1', and e; must be reversible, hence Ic,;I ;;c 0, and
we can write
e; = C,fTi'
The constants 1:<, in (21.2) are called the elastic constants, or moduli, of the
material. Inasmuch as the strains e, are dimensionless, the c;, have the
same dimensions as the stress components}
We have just remarked that the maximum number of the independent
elastic constants is 36, but this number reduces to 21 whenever there
exists a function
(21.3)
with the property,
(21.4)
oW
Oe; =
Ti
For one can always suppose that the quadratic form (21.3) is symmetric,
and it then follows from (21.4) that
'1', = CVil;,
JlQUATlONS OF 1IlL.UlTICITY
61
62
x~,
==
%a
-z~.
o
o
o
o
-1
e~
-T4,
-e4,
or
1"1
ClSe. -
= Cu = Cu =
Cu
1"1
givlln by (21.2)
= Cu = Cn = 0,
~=~=~=~=~=~=~=~=~
ott
ctl
law
(4)
If the
ttl are invariant (so that c'ft - etl) under a given coordinate tr-ansform.ation,
then the tranafo1'lllation Characteril!eB the nature of elastic symmetry. The ~ IigurIJ:c
63
EQUATIONS OF ELASTICITY
(~: ~ ~
(21.5)
Cn
Co.
c..
u. ~)
Cu c" 0
0
0
coo
Such materials as wood, for example, have three mutually orthogonal
planes of elastic symmetry and are said to be or.tftoiJ:J!,pic. In considering
such materials, we shall choose the axes '(;r co()rdinates so that the
coordinate planes coincide with the planes of elastic symmetry. Tn this
case, some of the coefficients Co; exhibited in the array (21.5) vanish,
Besides the symmetry with respect to the xlx.-plane, expressed by (21.5),
the elastic constants co; must also be invariant under the transformation
of coordinates defined by the following table of direction cosines,
Xl
X.
X.
-1
- -- - - - - --I
Xl
1
0
0
x.,
0
0
x.
This change of coordinates is a reflection in the x.x.-plane and leaves the
and unchanged with the following exceptions:
r,
e,
e~
e~
= -e5,
= -e.
CUel
This becomes
or
(21.6)
Caa 0
0
0 0 C44 0
OOOOc"
o 0 0 0 0
Cu
C..
Note that elastic symmetry in the ztzrplane and in the "'o2'rplane intplies elastic
~ in the """'rplane.
1
64
Or.; = Or,;
ax.
ax:
ax; aXk
x.
Ora':,
Xk
an; (a k! + au!)
axi
ax.
aTij
ax~
+ Or;; aU!.
axl ax.
are small compared with unity. In what follows, it will be assumed that
both the components of strain eoj and the components of stress Ti; are functions of the initial coordinates (Xl, X" xa).
REFERENCES FOR COLLATERAL .READING
A. E. H. Love: A Treatise on the Mathematical Theory of Elasticity, Cambridge
University Press, London, Sees. 60-65, pp. 92-100.
Chap. VI of Love's treatise is given to a discussion of the eqUilibrium Of nonisotropic elastic solids and contains further references on the subject. Voigt's Lehrbuch der Kristallphysik is a standard treatise on the subject.
L. Lecornu: Theorie mathematique de ]'eiastieite, Memorial des sciences math6matiques, Gauthier&-Villars & Cie, Paris, pp. 12-1S.
Contains a discussion of the theory of Poincare regarding the number of elastic
constants in the generalized Hooke's law.
PROBLEMS
1. Are the principal axes of strain coincident with those of stress for an anisotropic
medium with Hooke's law expressed by Eq. (21.2)? For a medil1m with one plane of
elastic symmetry? For an orthotropie mediull1 ? Him: Take the eoordinate axes
along the principa.l "as of strain @() that e. - e. _ .... - o.
55
EQUATIONS OF ELASTICITY
I. Show directly from the generalized Hooke's law (Eq. (21.2)) that in an isotropio
body the principal axes of strain coincide with those of stress. Hint: Take the coordinat<, axes along the principal axes of strain (e. - e. - e. = 0), and consider the effect
on Til and Tn of a rotation of axes by 180 about the x.-axis.
r:,
eu = Cu,
Similarly, a rotation of axes through a right angle about the x3-axis leads
to the relations
en
= C12,
C23 =
C13,
We introduce, finally, the coordinate system x;, x~, x;, got from the x" x.,
xa-system by rotating the latter through an angle of 45 0 about the x.-axis.
In this case, we have
1';.
+ 721'22,
-721'11
= -721', + 721'2,
C66
T6
x~,
e~
= -e, + e.
c,', we have
C.ue6.
r~ =
c,.e; or
1'2 =
cue,
cue,
+ c"e, + cue.,
+ C.06. + c.06.,
Many cast metals are notable exceptions. The processes of rolling and drawing
frequently produce a definite orientation of crystals, so that many rolled and drawn
metals are anisotropic.
1
66
wege!;
-~T'
CII
Cn ... Ctl
+ ~"'t == ~(Cll -
==
c,,)(-e,
CII,
+ e,).
= A.
CUe11
= c,,(e11
= Af)
C,te%2
C,te"
+ e" + e ) + (ell
+ 2p.eu.
- cU)e11
= (3).
+ 2p.)iJ.
Equations (22.3) can now be solved easily for the strains ei; in terIDS of the
stresses Tij. We have
or
(22.5)
->'8,;
eo; = 2/J(3)..
+ 2p,) e + 2;
Tij.
p, "F 0
L&m~
Lame constants.
We .have shown that the stress-strain law for isotropic media involves
no more than two elastic constants. The fact that no further reduction
is posiible is physically obvious from the simple tensile tests, but an
67
EQUATIONS OF ELASTICITY
Tll
= const,
= Ta. =
T12
= To.
= T31 = O.
Since the body forces are not present, the state of stress determined by
(23.1) satisfies the equilibrium equations (15.3) in the interior of the
cylinder, and equations (13.3) show that the lateral surface of the cylinder
is free of tractions.
The substitution from (23.1) in (22.5) yields the appropriate values of
strains, namely,'
(23.2)
+ p.)T
+ 2p.)'
-XT
(X
p.(3X
ell
e12
= e23
= e31 =
0,
+ p.)'
u"
is carried out in
(23.4)
E T,
-(T
"If" T
6u = 6aa =
= -.,.6u,
<f
e11
produced
33
= I e611 1 = I 6611 1;
thus .,. denotes the ratio of the contraction of the linear elements perpendicular to the axis of the cylinder to the longitudinal extension of the rod.
The quantity E is known as Young'8 modulu8, and the number 11 is called
the Poisson ratio.
It is easy to verify that one can express the constants A and p. in terms
of Young's modulus and Poisson's ratio as
E.,.
E
(23.5)
A = (1
+ (1)(1
- 211)'
p.
---
= 2(1
+ (1)"
= T =
CODBt,
Tn = 1'22 = Tn
Til
= Tn
= 0.
e" = 21' T,
Thus the number.p. represents the ratio of the shearing streBII T to the
change in angle au produced by the shearing stress. For this reason the
69
T; = -pv;,
(23.7)
'1'22
1"'33
= -p,
712
e = Tn + 1'22 + 1'33 =
1'23
to the surface.
=
-3p,
satisfies the equilibrium equation in the interior of
From (22.5) we deduce the expressions!
en = e
(23.8)
= e..
= -
3A
2,..'
en
eta
== 0,
Tn
'I'
eu = 0,
+ e.. + e31
" = -
t'
= - A . :%,..'
k = -~,
or
"
+ %,...
Since Ui.1 +
equations,
-p(l - 20-)
'
6'1"
0 for i '#- j.
where aH
the integration constants. These integration conIf we fix the point Xi .. 0 (assumed
to be in the body) and impose the condition that the rotation vector "" (Sec. 7) vanishes, We get
-
-aji
and the
ao; are
til -
-liz..
10
Thus, the quantity k represents the ratio of the compreesi.v~ "ress to the
cubical compression, and for this reason it is called the modul'UI oj compre8Bion. Since for all physical substances a hydrostatic pressure tends
to diminish the bulk, it is clear that k is positive. Substituting in (28.9)
the expreesi.ons for A and p. from (28.5) gives
E
k = 3(1 - 20-}
Since k is positive for all physical SUbstances, it foUows that IT is less than
one-half, and hence [see (23.5)] A is positive. For most structural materials, the value of IT does not deviate much from one-third. If the material is highly incompressible (rubber, for example), IT is nearly one-half and
p. == E13.
The stress-strain relations (22.5), when written by making the Bub. stitutions from (23.5), assume the simple form
(23.10)
where e = T". If we recall the notation of Sec. 14, these relations caa
also be given in the following form:
en
IT(r,,,,
+ T..)],
E [T - IT(r
+ r;.),
1
E
[T.. 1+1T
+ T",,),
= E [Tu
e"" =
(23.11)
e =
ell_ =
.,.(T
-r T
1Iz,
e.z =
1+.,. 6Z,
-r'T
The following table gives average values of E, p., and IT for several elastic
materials; the moduli E and p. are given in millions of pounds per square
inch.!
a
a-'!'-l
2"
11.5
11.0
6.5
5.6
4.9
0.29
0.28
0.25
0.33
0.33
0.25
0.283
0.273
0.269
0.339
0.327
0.250
(experimental)
Carbon steels.. . . . . . . . . . . . .
Wrought iron.............
Cast iron................
...
Copper (hot-rolled)..............
Braaa, 2: 1 (<lOld-drawn)... .. .. . ..
29 .5
28.0
16.'5
15.0
13.0
Glass... .......................
8.0
3.2
1. 5
O.OS
lIn. the engineering literature, the modulus of sheAr is often denoted by G, and the
reciproeal of Poillson's ~tio t1 is denoted by m; that is, m - l/a.
7I
BOUATIONS OF ELASTICITY
external load).
,I
(16.1),
T.
T.'
O.
4. Derive the following relations between the Lame coefficients :I. and 1', Poisson's
ratio '" Young's modulus E. and the bulk modulus k:
A _ ~ = I'(E - 21') = k _ ~ I' =
1 - 2<T
31' - E
3k<T
3k (3k - E)
-1+,,=
&r
(1
+ ,,)(1
- 2<T)
9k-E'
3kE
- 9k - E'
I =
3k - 21'
2(3k
+ 1')
-~'
E _ 1'(3)-
+ 21')
= A(l
+ ..)(1
cr
:1.+1'
- 21'(1
Ie _ :I. +
+ ..) = 3k
~
+1'
~
= A(1
31'
= 3k(1 - 2<T)
'
pE
H. Equilibrium Equations for an Isotropic Elastic Solid. The complete system of equations of equilibrium of a homogeneous isotropic
elastic solid is made up of the following equations:
a. Equa.tio1Ul of Equilibrium. From (15.3)
(24.1)
1';;.1
+ F, =
b. StreIB-Strain Relations.
(24.2)
0,
(i,j = 1,2,3);
From (22.3)
72
where
{J = 11;;,
Il;;
~(U;.;
+ 1.4.,).
The systems of Eqs. (24.1) and (24.2) must be satisfied at every interior
point of the body T, and on the surface 2: of the body T the stresses must
fulfill the equilibrium conditions (13.3)
(24.4)
where the v, are the direction cosines of the exterior normal v to the sur-
face 2:, and is the stre3S vector acting on the surface element with normal Y. To these equations one must adjoin the equations of compatibility [from (1O.9)J
e;j.kl + ekl.'j - e;'.il - ejl." = O.
It will be shown in Sec. 27 that the system of Eqs. (24.1) and (24.2),
subject to the conditions of equilibrium on the surface (24.4), is complete
in the sense that, if there exists a solution of the system, then that solution
is unique. There are nine equations in the system on the set of nine
unknown functions Ti;, u, (i, j = 1, 2, 3). Once the displacements u; are
determined, the strain components eij entering into (24.2) are readily
calculated with the aid of the formulas (24.3). We have assumed that
the displacements Ui are continuous functions of class C' throughout the
region T, and a reference to (24.2) shows that the components of stress
Tij are continuous of class C2 in the same region.
The equations of
equilibrium (24.1) contain the components Fi of the body force F, and
they are assumed to be prescribed functions of the coordinates Xi of the
undeformed body. Typical examples of the body forces F, occurring in
practical applications, are centrifugal forces and forces of gravitation.
(24.5)
Ti
T,
be such that
73
EQUA'l'IONS OF ELASTICITY
T;; =
>.
6ijUU
+ !'(u,.j + Uj.i).
or
(24.7)
where
!'U;.jj
p.V2u;
+ (X + !')Uj.;; + F. =
0,
0" + F. =
+ (X + !') -;"Xi
74
Note that we need not adjoin the compatibility equati01l8 (10.9), for
the only purpose of the latter is to impose restrictions on the strain components that shall ensure that the 6;; yield single-valued continuous displacements u., when the region 'T is simply connected.
It is clear that Prob. 2 is completely solved if one obtains the solution
of the I!ystem (24.7) subject to the boundary conditions
(i = 1,2,3),
+ eli.;; -
~,jl ,- ejl.ik =
0,
we obtain
(24.8)
'T;,'.'"
+ 'Tid.;; -
'Til,jl -
'TJI.ik
1 + u (6.,B,ll
+ &ktB,i; -
Oile,;1 - O;tB.il)'
Since the indices i, j, k, l assume values 1, 2, 3, there are 3' = 81 equations in the system (24.5), but not all these are independent, for an interchange of i and j or of k and l obviously does not yield new equations.
Also for certain values of the indices (such as i = j = k = l), Eqs. (24.5)
are identically satisfied, and, as already noted in Sec. 10, the set of Eqs.
(24.5) contains only six independent equations obtained by setting
k = l = 1,
k = I = 2,
k = l = 3,
k = l = 1,
k=I=2,
k = I = 3,
i
i
i
i
i
i
= j = 2;
= j = 3;
= j =
= 2,
3,
= 1,
=
1;
j = 3;
j = 1;
j = 2.
75
+ Ta,li -
Tu.,i~
Ti~,a. = 1 ~
IT
(Oii9 ,a
+ ou9,,; -
oa.9,tt - ap9.a.).
Vs,,
"
+ 8.s, -
To.
'.J""0.
IT"
0. =
(ooV28
1 +IT
'J
1' ..
15,'
e.
T".k!
Vs",;
Tik.;k,
x'" we get
= -F,.;
+1+
IT
8.,; - 1
IT
IT
o,;V28 = -(Fo.;
+ F;,,),
This set of 6 independent equations can be further simplified by expressing an invariant V28 in terms of the derivatives of the body force F. This
may be done as follows:
If we set k = i and l = j in (24.8) and sum with respect to the common
indices, we get
But
Tn
Tfi
S,
ai,9,ii =
e,.. =
V"e,
and
O;.fJ,H
= 0;;9... = 3V"e.
V!9 =
_ f I-
1+0"
(vta - 3V29)
76
or
1-(1
...
.,.t,, = --vte
1
+"
(24.12)
T'i,ij
and inserting this in the left-hand member of (24.12) yields the formula
(24.13)
Substituting from (24.13) in (24.11) gives the final form of the compatibility equation in terms of stresses,
(24.14)
V2,-,;
+ 1 ~ " e,;; =
~ ~;; div F
iT
- (Fi .;
+ FM)'
o~
+ 1 + "OX'
V2,- + _1_ 0'9
2,-
(24.15)
+ " oy'
1 8'9
+ 1 + " oz'
__
iT_
= _ _"_
div F _ 2 of .,
1- "
OX
1- "
"d' F
= - 1_ "
of.,
oy
2 8F.
Tz'
div F _ 2
lV
1 + II oy oz
8z
oy
V2,- + _1_ 8'9 = _ (8F. + of_),
1 + II OZ 8x
OX
OZ
V2,- + _1_ o'e = _ (oF. + 8F.).
V2,-
""
1+118xoy
8y
OX
Equations (24.15) were obtained by Michell in 1900 and, for the case
when the body forces are absent, by Beltrami in 1892. They are known
as the Beltrami-Michell compatibility equations. Thus, in orJer to determine the state of stress in the interior of an elastic body, onll must solve
the system of equations consisting of (24.1) and (24.15) subject to the
boundary conditions (24.4).
The system of Eqs, (24.1) and (24.15) is equivalent to the system consisting of Eqs. (24.1), (24.2), and (24.5).
If the field of body force F is conservative, so that
F=
v"
77
EQUATIONS OF ELASTICITY
or
then
div F "" Fi .; =
V'<p,
<P.;; ""
and
Pi,;
fP.ij,
"';'i;
+ 1 +1
t1
t1
e.# = - 1 _
t1
8;;'Vf", - 2'1'.#.
We shall consider two particular cases of body forces, namely, the case
in which F is a constant vector and that in which the potential function 'I'
is harmonic (that is, div F = V2rp = 0).
If F is constant, then <p is a linear function. In this case the right-hand
member of (24.16) vanishes, and we obtain the equations of Beltrami,
(24.17)
"';'i;
+1+
(f
e.;;
o.
vte
0,
and since the strain components e;; are linear functions of the'Ti;, we have
V'ei; = O.
A function V of class C., and satisfying the equation V'V = 0, is
called a biharmonic function.
If the body force F is derived from a harmonic potential' function, so
that
div F =
"'trp
= 0,
""f}
;=
0.
78
which satisfy the systems (24.1) and (24.2) with prescribed body forces
(24.18)
(i = j = 1, 2, 3)
will correspond to the choice of the body force whose components are
+ Fl2). If the set of functions (24.18) represents a solution of the
homogeneous system, that is, when Fl') = 0, then the expressions (24.19)
represent a solution of the problem corresponding to the choice of the
body force with components Fl!).
Fl!)
PROBLEMS
1. Show that the following stress components are not the 8Olution of a problem in
elasticity, even though they satisfy the equations of equilibrium with zero body forces:
T.. -
T. . T. Ta: .. T' -
C[ZI
y")I,
zl)I,
c ... 0,
-2t:c:ny,
1"' -
o.
}. -0,
p.-~E,
A:-~E,
T" -
Ee" - ~E(.... ;
+ ",,,),
79
I!lQUATlONS OF ELASTICITY
\no sum on repeated subscripts). That is, the six stl"ell8 components art! connected,
in this cue, by the three equilibrium equations
+F, -
T'M
iJx iJy
iJx'
iJy'
and two similar equations obtained by cyclic interchange of x, y,'. J)erive these
compatibility conditions from Eq. (24.8) by setting .. - 0, Ie = i, I - j.
A. and L. Foppl have discussed' the simplification of the equations of elasticity
obtained by choosing for Poisson'. ratio" = 0 or fT = ~2. Westergaard' has treated
the problem of obtaining the general solution from a solution for a partiCUlar choice
of Poisson's ratio.
S. Define the stress function S by
and consider the case of zero body force. Show that, if Poisson's ratio fT is assumed to
vanish, then the equilibrium and compatibility equations given in the preceding problem reduce to
V'S = const .
.. Show that, if Poisson's ratio
I'
HE,
X ..
fT
00,
~2,
then
OOJ
+ H a,;6
+ 'tiM) + H a,;6.
2"",1
- 1'(....1
T'I -
'til ,. _
.,
at) _
aX.
and that the equilibrium equations (24.1) can be written in the form
'tI. -
0,
v'u
V"v
0,
0,
+H~~
1(13az
ae +F. )
V'tn +;
a.. iJv ilw
8Z + iJy + cJz
This cue (. -
~)
'tI, 0,
tn,
0,
o.
80
21. Dynua.ical Equations of an Isotropic Elastic Solid. The differential equations of motion of an elastic solid can be obtained at once from
the equations of equilibrium (24.1) by invoking the Principle of D'Alembert and adding the forces of inertia to the components Fi of the body
force. If P(X" X2, x.) is the density of the medium, then the components
of the force of inertia acting on the mass contained within the volume
element d.,. are' - p aa~' d.,..
body force F in (24.1) the components of the force of inertia per unit
volume gives the system of equations
(25.1)
..
"'ii.i
a'Ui
+ F.
pii,;,
..
I'V'U;
pii,;.
To these equations it is necessary to adjoin the initial and the boundary conditions. Thus, at each point of the surface X of the undeformed
medium, the surface forces T, or the displacements U; must be prescribed. The functions U; prescribed on the surface X, in general, are
functions of the space coordinates x, and of the time t. If the surface
forces T, are prescribed as functions of Xi and t, then the components of
stress must satisfy the usual equilibrium conditions (24.4) on the surface
X and in addition one must know the initial conditions on the displacements U; and on their tiIl<e derivatives. We set forth these conditions
explicitly for the fundamental boundary-value problems of dynamical
elasticity that correspond to the problems of equilibrium in Sec. 24.
Problem 1. Determine the displacements U;(X" X2, X3, t) that satisfy
in T the By8tem of Egs. (25.2) and satisfy the cvnditivns
1M t = to thrQU(Joout T,
a.nd that satUfy on the 8urface 2: of the region.,. the boundary cvnditiona
1M t
t
If is a function of I, we write _ ~
(p ~.) d-r.
to.
81
EQUATIONS OF ELASTICITY
in
for t
2:
to.
which the surface forces T, are prescribed functions of Xi and t over part
of the surface and the displacements u, are given functions of Xi and t over
the rest of the surface. As an example of such a problem, consider an
elastic plate clamped at the edges. Let the plate, initially at rest, be
subjected to a normal load varying with time. In this case, the displacements are known on the edges of the plate (for t 2: to), while the surface
forces are given functions of Xl, X" Xa, t.
REFERENCES FOR COLLATERAL READING
A. E. H. Love: A Treatise on the Mathematical Theory of Elasticity, Cambridge
University Press, London, Sees. 85, 86, 91, 92.
E. Trefftz: Handbueh der Physik, Verlag von Julius Springer, Berlin, vol. 6, Sees.
13-15.
26. The Strain-energy Function and Its Connection with Hooke's Law.
We introduce the definition of the unstrained, or natural, state of a body
as a standard state of uniform temperature and zero displacement, with
reference to which all strains will be specified.
If the body is in the natural state at the instant of time t = 0, and if it
is subjected to the action of external forces, then the,latter may produce
a deformation of the body and hence will do work. We shall be concerned with the rate at which work is done by the external body and
surface forces. If (Xl, X" X3) denote the coordinates of an arbitrary
material point P of the body in the unstrained state, then at any time t
the coordinates of the same material point P will be x, + U,(XI, X2, X" t).
Since the displacement of the point P in the interval of time (t, t + dt) is
given by
aU,
at
dt.., ti.; dt,
it follows that the work done in dt sec by the body forces acting on the
volume element dT located at P is F.11.; dT dt. The work performed by the
external surface forces in the same interval of time is 1',11.; du dt, where
du is the element of surface. Denoting by 8 the total work done by the
body and surface forces, we have the following expression for the rate of
doing work on the matter originally occupying some region T,
(26.1)
~=
F(iJ, dT
+ fJl TtiJ.@;
82
l T,it.; 1
1.
1
dn
'T'I,,11; d'T
==
r.i,JONt
'T'I,;U, d'T
4'"
dr
1
1.
+1
+
+
'Ti,11;,; d'T
.,.
TtJ
(U.'i +
2 llj,i
('Ti,.ei;
+ U.,i -2llj'i)
?d'
+ 'TiPi;) d'T.
When this is inserted in Eq. (26.3) and the resulting expression used in
(26.1), one obtains
(26.4)
J. pU,iUi d'T,
di ""
.'d
J.( pu.,'Uo
'T.
dB
dK
at ... di
+
1 at
'Tii
oeii
d'T.
'Ta = 'Til,
ea == ell,
". = "Ia,
e. "" 2tu,
,... == 1"n,
_ == 2eu,
It Is .-umed here that the variation of the denaity p with time Is nesJjgible.
83
EQUATIONS OF ELASTICITY
used in Sec. 21, and suppose that there exists a function W(el, el,
of the independent variables ei such that'
e.)
aw
(26.6)
ae. =
= at
+ raw ?_e; dT
)T ae, at
d
+ di)T WdT.
e=
U ==
IT W dT,
0 and
U,
where
(26.8)
since both K and e vanish in the natural state. The function W is called
the volume density of strain energy, or the elastic potential, and U is the
strain energy of the body.
Equation (26.7) has a simple physical interpretation. The work e
done by the external forces in altering the configuration of the natural
state to the state at the time t is equal to the sum of the kinetic energy K
and the strain energy U. The strain energy U may be conceived as the
energy stored in the body when it is brought from the configuration of the
natural state to the state at the time t. If at the time t the body is in
equilibrium, then K = 0 and e = U.
We assume now that the strain-energy density function Wee"~ e., .. ,
e,) can be expanded in a power series
2W = co
and discard all terms of order 3 and higher in the strains; the constant
term Co can be disregarded since we are interested only in the derivatives
of W. Thus, we have, from (26.6),
To = C,
e;,
(26.9)
We shall exhibit such a function for an isotropic elastic medium in formula (26.16)
It is cleal'ly associated with the initially prestreseed state.
84
Hence
(26.10)
T;
oW
oe;
2 (C;i + CJi)8j.
It is thus seen that the coefficients in the generalized Hooke's law are
symmetric if the strain-energy density function, with the properties
stated above, exists.
If the quadratic form (26.9) is symmetrized in advance, we can write
(26.10) in the form
(26.11)
T,
e;;e,;,
W = J,2T,,-e;;
(i, j = 1, 2, 3).
oW = C.,.,..
a.,.,
" ,
= e;.
W = 72C;,-e,-e;,
(i, j = 1, . . . , 6),
solides.
t Alti dfll4
85
constants
form'
ev.
(26.15)
and it follows from (26.12) that
(26.16)
+ P.eije,;
+ p.(e;, + ei. + ei. + 2ei2 + 2e~. + 2er.).
W = 72A6e"
= 72 M2
A quadratic form that takes only positive values for every set of values
of the independent variables, not all zero, is said to be positive definite.
Equation (20.16) shows that the strain-energy density W is a positive definite
form in the strains eij, since both h and J.l are positive constants. This
important property of function W will be used in Sec. 27 to establish the
uniqueness of solution of the fundamental boundary-value problems in
the linear theory of elasticity.
As a consequence of the linear character of the stress-strain law (26.15),
the function W is expressible as a positive definite quadratic form in the
stress components r;;. Thus, on substituting from (23.10) in (26.12) we
get,
IT
1+1T
(26.17)
W = - 2E (j'
2E rifT;;,
or
IT e'
lV = - 2E
+~
1+
IT (r.
11
IT (r .
12
+ r. + rn2),
3
where 13 = r11
r22
r33.
It is easily checked that
ilW
1+
r
IT
ilr;; =
IT
r;j -
E eo,;
= e;;.
It is clear that lV, the energy of deformation per unit volume, has a
physical meaning that is independent of the ehoice of coordinate axes, and
hence it is invariant relative to all transformations of cartesian axes. It
is also known that every invariant of a tensor e,; can be expressed as a
function of the principal invariants' 6, 6., 6 3 Inasmuch as W is a
quadratic form in the e;/t it cannot depend on 6 3, and hence it must involve
only
6 = e,
eu
em = e;,
and
6 2 = euem + ellIel + elen = 72 8..,/e",-e.;.
+ +
1 It is worth recalling that this law was deduced in Sec. 22 without invoking the
&.lI8umpt,ion that eli - eil in the generalized Hooke's law (21.2).
See (6.10).
86
We have, in fact,!
(26.18)
2'1. Uniqueness of Solution. Remarks on Existence of Solution.
Before proceeding to the proof of the uniqueness of solution of the funda
mental boundary-value problems of the linear theory of elasticity, we
establish an important theorem concerning the strain-energy function.
CLAPEYRON'S THEOREM: If a body i8 in equilibrium under a gillen BY8Um
of body force8 F; and BUr/ace force8 To, then the 8train energy of deformation
is equal to one-half the work that would be done by the external force8 (of the
equilibrium 8tate) acting through the diBplacementB U; from the unstre8sed
state
f~ T.1/,;
Then
cJq =
cJq
(i,j = 1,2,3),
and
(27.3)
(i, j = 1, 2, 3).
87
EQUATIONS OF ELASTICITY
ul2),
ujl) -
will satisfy Eqs. (24.1) with Fi = O. Thus, for the "difference" u;, 'TiJ
of the two solutions, we have from the formula (27.1)
Wd'T = O.
u., 'T;;,
(i, j = 1, 2, 3)
satisfies the differential equations when body forces are set equal to zero.
We have in all cases the condition that
(27.4)
on l:,
proble~,
0 for all t
T. =
to.
0; . .
0 on
88
= to.
Hence
K+ U = 0,
and since both the kinetic energy K and the function U are essentially
positive, one has
for all t ;:::: to.
K= U =0
It follows from these equations that
aU,
at
0,
and
(!;;
0,
for all values of t ;:::: to. The first of the above-written relations states
tnai we are deaung wiin a siatic case, and tne second means tnM deformation of the body is not present, so that the solution (UI, Us, us} represents
a rigid body motion. But the displacements (Ul, us, us) vaniE;b at t = to,
and hence rigid body motion cannot be present in our solution, or
Us = Us = 0
for all t ;:::: to.
Thus, the two assumed solutions (27.2) and (27.3) are identi<lal.
The proof of uniqueness given here is essentially due to Kirchhoff.l
It should be noted that the crucial point in the argument is the positive
definite character of the strain-energy density function W. In nonlinear
theory, where large strains may be present, W need not bEl a positive
definite quadratic form in the strains, and the proof br()aks down.
Indeed, problems concerned with elastic stability and bUckling contemplate large deflections, and it is well known that Boluti()ns of such
problems need not be unique. The reader may be familil\,r with the
situation in the theory of Euler's columns where a column Subjected to
end thrusts may assume several distinct equilibrium configurations.
We conclude this section with a few remarks on the existenCE! of solution
of the fundamental boundary-value problems in linear elasticity. Because
of the resemblance in the formulation of such problems tt:J the basic
problems of Potential Theory, it is natural that the early ~ttempts to
establish the existence of solution center.ed on methods simil.ar to those
UI =
EQUATIONS OF ELASTICITY
89
CHAPTER
91
92
ax + iff""
ay + iff""
az
iff%%
F
-z,
aT. + aT +
F
ax
Ty Tz - - .
aT + aT.. _ F
ax + Ty
Tz- -"
iff _
(29.1)
iff. z
au 1
ax "" E [T%<
lJv
cr(T
+ T),
(29.2)
(29.4)
T , T , T ,
The functions Tij, naturally, must satisfy the Beltrami-Michell compatibility equations (24.15).
The problem, formulated with this degree of generality, presents
formidable complications because of the difficulty of fulfilling the boundary conditions (29.3). In fact, the generality of formulation of the
b.oundary conditions (29.3) is quite unnecessary from the practical point
of view, since the actual distribution of applied stresses on the ends of
the cylinder is rarely, if ever, known. A designer knows, more or less
accurately, the resultant force T and the resultant moment M acting on
the ends of the beam, and quite often the nature of the distribution of
stresses over the ends of the beam, which give rise to the force T and the
moment M, is a matter of indifference. On the other hand, if one accepts
the principle of Saint-Venant and considers a beam whose length is large
in comparison with the linear dimensions of its cross section, then the
actual distribution of stresses over the ends has no appreciable influence
on the character of the solution in portions of the beam sufficiently far
removed from the ends. That is, one is free to prescribe any distribution
93
of stresses at the end of the cylinder so long as the resultant forces and
moments reduce to those given in the formulation of the problem. This
principle will be applied throughout our discussion of beams. This
means that the mathematical solution obtained will give, near the ends
of the beam, either (1) the exact solution of the physical problem in which
the applied stresses are distributed in the way specified by the solution or
(2) the approximate description of the physical situation in which the
~vstem of external forces and moments is statically equivalent to that
a~sumed by the solution but is distributed in some different manner.
One need be concerned with one of the bases only, for the specification
of the resultant force T and of the resultant moment M on the base z = 1
requires that the resultant force acting on the base z = 0 be - T and that
the resultant moment acting on the same base be so chosen as to satisfy
the condition of static equilibrium. Let the point 0' of intersection of the
z-axis with the base z = 1 be the center of gravity of the base, and suppose
that a force T and a couple M are applied at 0'. The force T can be
resolved into two components, one in the direction of the z-axis and the
other in the plane of the base z = T. The component of force T, in the
direction of the z-axis will be responsible for tension or compression, while
the other component T B, lying in the plane of the base, will produce bending of the beam. The couple M, acting on the end of the beam, can likewise be decomposed into two couples, the moment of one of which is
directed along the z-axis and hence will be responsible for twisting of the
cylinder, while the moment of the other lies in the plane z = 1 and will
produce bending.
Thus, our problem can be solved, by utilizing the principle of superposition, if we succeed in solving the following four elementary problems:
1. Extension of a cylinder by longitudinal forces applied at the ends.
2. Bending of a cylinder by couples whose moments lie in the planes
of the bases of the cylinder.
3. Torsion of a cylinder by couples whose moments are normal to the
bases of the cylinder.
4. Flexure of a cylinder by a transverse force applied at one end of the
cylinder, while on the other end there act a force equal in magnitude but
oppositely directed to the transverse force and also a couple of such magnitude as to equilibrate the moment produced by the transverse forces.
Our general plan of attack upon the four elementary problems listed
above is that of the Saint-Venant semi-inverse method of solution. This
consists in making certain assumptions about the components of stress,
strain, or displacement and yet leaving enough freedom in the quantities
involved to satisfy the conditions of equilibrium and compatibility. In
applying the semi-inverse method to problems on beams, we shall make
one general assumption about the stress distribution in any beam; further
assumptions regarding the st,resses or the displa('ements will be introduced
TJlII
"IIZ
= 0,
1. Show that if the stress components Tu, Tz., T . . and the body forces F, vanish, then
.J'T..
.J'T.. .. a'T.. = 0 ; t h at IS,
. t h e stress component 'T.., IS
. I'Inear In
. x, In
. y, and In
. z.
ax - iijjI
t
azr
Write out the most general form of the function in this case.
2. Integrate the dilierential equations of equilibrium T,;.; + F, = 0 throughout
the volume of an elastic solid, apply the Divergence Theorem, and show that the equation8 of static equilibrium
are satisfied and hence that the resultant force on the body vanishes.
3. Show with the help of the Divergence Theorem that if the following dilierential
equations of equilibrium
TJlioi
+F" ==
0,
1'#.;
+F, == 0,
(j - x, y, .)
1 One may equally proceed by assuming that the distribution on the cross section
of the stress constituting each component of the resultant force and couple is the same
at all sections. This is equivalent to assuming that for Probs. 1 to 3, stated on p. 93,
we have
iJru
az
== iJT. ,1 == Ch.. == 0
a.
.Jz
'
95
are satisfied, then the following equation of static equilibrium also holds:
ti'1115
"l>pr..ssing t,h.. vanishing of the x-component of the resultant mompnt on the body.
t,
~ =_ p (a
T'II
on z
COllSO, }
I.
0,
If we assume
Tu =
p,
Txx =
Tyy
=-
Try
'Ty,;
= Tz%
= 0,
throughout the cylinder, then the equilibrium eqnations (29.1) and (29.4)
are obviously satisfied.' The Beltrami-Michell compatibility equations
are also satisfied, since the components of the stress tensor are constants.
The displacements u, can be readily calculated. Thus, from (29.2),
au
ax =
av
ay
~+~
ax
ay
E p,
=0
It
aw+~=o
aJ!
aw p
az = E'
au+aw=o
az ax
'
p,
az
'
and since the right-hand members of these equations are constants, one
is justified in assuming that the solutions are linear functions of x, y, and z.
A simple calculation gives
u = -
up
EX,
up
= - FfY'
if one neglects the terms representing rigid motion of the beam as a whole.
Of ('ourse one could obtain the displacements by making use of the general
formula (10.6). This problem has already been discussed in Sec. 23.
PR.OBLEMS
1. Consider a bar of length I in., area of cross section a sq in., Young's modulus
Use both Eq.
(26.12), W = ~T;;e;;, and Eq. (27.1),
W dT =
1. F,u,
dT
LT,u,
du,
96
to show that the strain-energy density Wand the tota.! strain energy U -
I.
W dr
W - 2a'E
T'1
U _ 2aE
. lb
m.-.
I. Find the greatest amount of strain energy per unit volume that can be stored in
a steel bar under tensile forces T without producing permanent set. Take the elastic
limit to be 30 X 10' lb per sq in. and Young's modulus as 30 X 10'lb per sq in.
3. In Prob. 1 take I = 10 in., a = 2 sq in., T = 50 X 10' lb, E = 30 X to' lb per
sq in. Find the strain-energy density W, and show numerica.!ly that the total strain
energy U is one-ha.lf the product of the force T by the elongation of the rod.
" Two gage marks 1 in. apart are made along the axis of a steel bar 10 in. long and
of 2 sq in. cross-sectional area. The bar is then subjected to a tensile force of 50,000
lb. Find the stress, strain, elongation between gage .marks, and total elongation of
the bar. What is the total change of volume of the bar? What is the change in the
cross-sectiona.! area of the bar? Take E = 30 X 10' lb per sq in., " = 0.3.
Ii. Consider a beam stretched by a tensile force T applied at each end. The magnin
tude of the stress vector acting on a section with normal n is T = T /(a sec "')
{,fOS8
Tu COS! (/I,
section.
7,
Tn
fP
cos
fP.
Derive these 'results also from the formulas of Sec. 19b. Show that the maximum
normal stress is r.. (at", = 0) and the maximum shear stress is hr.. (at 'I' = 45).
Compare this with the theorem of Sec. 18. What are the inclinations of the cross sections on which the shear and normal stresses are equa.! in magnitude?
6. Find the maximum shear stress in the beam of Prob. 4. What is the normal
stress on the planes on which the shear stress is a maximum?
7. Consider a rod under uniform longitudinal stress r .. = p. Let the rod be so constrained that there is no lateral contraction in the x-direetion (e u = 0), while the rod
is free to contract laterally in the y-direction. Define the effective Young's modulus
by E' == Ttl.le .. and the effective Poisson's ratio by (T' == -evlI/el." and show that,
owing to the lateral constraint, one has
E' == ___,
1 - ,,'
,
fT
" =--.
1-"
I-fT
E' = (1 _ 20-)(1 + ..) E.
What is the effective Poisson's ratio?
97
pgz,
Txr
Tuv
= TII~ =
Tzu
0,
'fez
Jt
l_t
which obviously satisfies the equations of equiI
I
librium and the compatibility equations (24.15).
I
I
I
The conditions (29.4) that no forces are applied to I
I
I
I
the lateral surface of the beam are likewise fulfilled. J
I
J
There are no tractions applied at the lower e'1d; I
I 1
I
J
hence all components of stress vanish there, while
I
at tht' upper end we have T" = p(Jl, which is directed
vertically upward. Thus, the assumed distribution
x
of stress requires that the upper end of the cylinder
be supported in such a way as to yield a uniform disFIG. 16
tribution of stress.
In order to determine the displacements Ui, we note the relatiolls (29.2),
which yield,
(31.1 )
(31.2)
au
av
ax
ay
aw
<1PYz
- -R'
~1J)
all
~!'+~=O
ax
ay
.
+ az
~
'!.~.
'
az
(:~I.I)
gives
p(JZ2
+ wo(x,
w - 2E
p(JZ
="E'
az
+ aw
ax
o.
y),
wherc w" is a fUlJetioll of ,r and !J alonc, and it follows from the last two
of ';'18. (:l1.2) that,
au
a;
awo
fii'
av
and
a;
and
awo
-ay'
Hence
aw.
u= -z 71:
+ 110(2',
y),
= -z
awo
a!I
+ I',,(X, u),
98
where u.o and PD involve x and y only, Substituting the values of u and v
just found in the first two of Eqs. (31.1) gives
(31.3)
(JUo =
ax
avo = 0
ay
'
a2wo
oy'
upg
=
7!['
while the substitution of the same values in the first of Eqs. (31.2) yields
_o"w o = 0
ax oy
,
(31.4)
OUo
ay
+ avo = o.
ax
It is clear from the first. two of the differential equations (31.3) that
Uo =
F(y),
and
= G(x),
Po
reqUires
.
and t hIS
t hat dF
dy
Uo
= F(y) = ay
+ dG(x)
=0
dx
'
dG = -a, were
ha'IS a constant.
a, dx
+ b,
and
Po
= G(x) = -ax
Thus,
+ c.
1",
+ ay + b,
b' z-ax + 1',
u;: zx - a'z
upg
v=-1fzy-
w = ;~ (Z2
about the axes through (0, 0, 1) that are parallel to the x- and y-axes, we
1 See Prob. 4, at the end of this section.
We demand, in effect, that the rotation
ilw ilw
components ( -a. - Uu
-, -Uu - , - - - vanish at (0 0 I).
110)
liz
/Jtjh
ilrlJII
lJz
"
au
/)z =
0,
99
These six
conditions enable us to eliminate the six constants a, b, c and a', b', c' .
An elementary calculation shows that the displacement, in this case, is
given by
ru
,(31.5)
= -
q;:
ZX,
I)
= - ";} zy,
I').
It is seen from this solution that points on the z-axis are displaced
vertically according to the law
All other points of the beam have both vertical and horizontal displacements on account of the contraction in the transverse direction. The
shape of the beam, after deformation, is indicated by the dotted lines in
Fig. 16. Any cross section of the beam is shrunk laterally by an amount
proportional to the distance from the lower end and is distorted into a
paraboloid of revolution. This can be seen by noting that, for a cross
section z = c,
z'
+ w = c + Pfl(_c;i
l22
The upper ba.'le of the cylinder is warped upward (see Fig. 16) because
of the assumed uniform distrihution of the stress component T" over that
face and the fixing of the point (0, 0, l).
REFERENCES FOR COLLATERAL READING
A. E. H. Love: A Treatise on the Mathematical Theory of Elasticity, Cambridg~
University Press, London, Sec. 86.
S. Timoshenko and J. N. Goodier: Theory of Elasticity, McGraw-Hill Book Company, Jne., New York, Sro. 86.
PROBLEMS
1. Discuss the solution of the "'"",tostat.ic probl~m for the cas" wher..
T:z
T'II"
Tn
= -p
+ pgz,
This state of stress corresponds to that found in a body immersed in a fluid whOlle
density is the same as that of the body, where p is the pressure of the fluid at the level
of the origin of coordinates.
ll. Determine the displacements in a cylinder of length 21 and of density p when suspended in a fluid of density p'. Let the pressure of the fluid at the lev..l of the center
of gravity of the cylinder be p. Choose the origin of the coordinate system at the
100
==
'1''1/11
== -p
+ p'UZ,
T,. - -p
+ (p
- p')gl
+ pgz,
S. Obtain tbe solution given in (31.5) from the general solution (10.6).
4. Show from Eq. (7.5) that the displacement components
u =
v=
w =
- ry
rx
-qx
+ py
+ qz + a,
- pz
+ b,
+c
e=
it
Wi
In this relation, d is the distance of the filament from the central plane
drawn through the central line at right angles to the plane of the couple
(the xz-plane in Fig. 17), and R is the radius of curvature of the central
line. Now the length ds o of the portion of central filament subtended by
an angle dO is ds o = R dO, while the length of the element ds subtended
by the same angle dO and at a distance d from the central plane is
ds = (R
+ d) dO.
ds - ds o
~-dS;-
(R
+ d)RdOdfT----R dO
d
= Ir
~--
= lid.
fA
J:
du =
fA Y du = fA XY du =
O.
It follows that the distribution of stress in any section will be characterized by the formula
E
Tzz=-/iX,
where the negative sign arises from our convention in regard to the signs
of tensile and compressi ve stresses.
We shall now verify that the boundary conditions on the ends of the
beam are satisfied, namely, that the resultant force and moment acting
on the bases (or on any other cross section of the ('ylinder) reduce to It
moment about the y-axis. The resultant force T acting on any section A
has the components
T.=
fAT du=o,
T. =
Ty
.,,,du =
-.~
J.4r !fT" du
-l!R-
f.{
fA T.ydu = 0,
xdu
xydu
0.
= 0,
102
M. =
-l
x-r du =
xtdu =
E~.,
where I. is the moment of inertia of the cross section about the y-axis.
Thus, the curvature of the central line of a beam bent by a couple c
magnitude M is l
EI
(32.1)
R = M'
Formula (32.1), connecting the curvature of the central line with the
bending moment, is called the Bernoulli-Euler law. It will recur when we
come to consider this problem in a rigorous way.
It appears from the foregoing discussion that the stress in a beam giving
rise to a couple M is a longitudinal stress of magnitude
r = -
x = -
Ii x.
Under the action of the tensile stress r." the cross section of the beam
will be deformed, and the amount of the transverse contraction (or extension), from the definition of Poisson's
R
S
ratio tT (see Sec. 23), is
.i\
fS
tTX
tTMx
If =ET'
----- ----Q~.
"
1 The subscript y on I and M has been dropped, since no confusion is likely to arise
here.
t Those interested in the experimental determination of u are referred to S. TimoPenko and .T. N. Goodier, Theory of Elasticity, p. 254, ",he..., further references OJ)
this subject will be found.
103
bend only slightly under the action of the couple 114, and hence the magnitude of E I provides a measure of the rigidity of the beam. For this
reason the constant EI is called the modulU8 oj flexural rigidity. In order
to increase the flexural rigidity of a beam, one must design it so as t.o make
, the moment of inertia I as large as possible. This is one of the reasons for
making beams with cross sections in the shape of the letters I, T, Z, etc.
We are now rAady to considAf rigoTOuRly the prohlem of ~nding of a
heam.
Assume the system of Rtreslles
(32.2)
(32.3)
Then, as
iJv
17M
iJy = El x,
iJw+~=O
iJy
iJz
'
+ wo(x, y),
104
Hence
_ M
aWo
aWo
+ vo(x " y)
u - 2EI z - z
(32.4)
= - zay
ax + Uo ( x, y),
Substituting these
(32.5)
a2wo + aVo
-zW
ay
(1M
= EI
x.
Since these equations are true for all values of z, it appears that
a'Wo = 0
(32.6)
ox'
- 2EI
tTM x 2
Uo -
tTM
+ I 1(Y),
Vo = EI xy
+ /2(x).
EI Y
ax
=0
Since the last three terIllS in this equation are independent of z, it follows
that
a2w
o
- = 0,
(32.7)
ax iJy
and hence
Thus,
df,
(1M
dy + EI
and
(32.8)
Y = a,
(32~7)
+ 'YY + C;
I.
= -ax
+ b,
tTM
il = - 2EI y2
+ a1J + a,
that Wo is a linear
3!_ (zt
u =
2EI
M
EI ITX'g
[
(32.9)
1V=
+ ITxt
105
- try2)
+ay-,8z+a,
- ax
llO=-~XZ
+ ,8x
-- 'YZ
+ 'YY
+ b,
+ c.
10
= au =. ~ =
az
az
av
ax
at (0, 0, 0).
(32.10)
2EI (z
+ ITx'
- try'),
v=EI ITxy ,
10
= -
EI
XZ.
It is clear from (32.10) that the filaments lying in the central plane x = 0
suffer no extensions; that is, the plane x = 0 is the neutral plane of the
beam. The longitudinal filaments on one side of the central plane
(x > 0) are contracted, whereas those on the other side (x < 0) are
extended. Points which, prior to deformation, had the coordinates x; go
into points with coordinates x:, where x; = Xi + Ui. Hence the points
on the z-axis (that is, the points on the central line of the beam) go into
. points
(32.11)
Y' = 0,
Z'
= z.
106
d
dz7
'
2
dx'
if dz' is small.
which is the Bernoulli-Euler law, discovered earlier from rough geometrical considerations. This formula states that the magnitude of the
bending moment M is proportional to the curvature of the central line
of the beam. The Bernoulli-Euler law forms the point of departure for
all considerations in the technical theory of beams.
Consider a cross section of the beam made by the plane z = c. Mter
aeiormation, points in tillS cross section mIl lie in tile plane
z' = c
+w
c -
~ xc = c (1
- Ii}
b (1 + i x')-
The points
a of the
<==
107
VAhll'~
~-
+ 0'(a
1/2)J.
of 1/ R, we have
+n
-
+- 2f(
1 f,'-' + (T(a'
- 1/2)}
-,
,
See in this connection Sees. 52-61, dealing with the ftexure problem.
lOS
planes normal to the z-axis will remain plane after deformation and that
the action of the couple will merely rotate each section through some
angle 0. The &mount of rotation will clearly depend on the distance
of the section from the base z = 0, and since the deformations are small,
it is sensible to assume that the amount of rotation 0 is proportional to the
distance of the section from the fixed hase. Thus,
0= az,
where a is (,he I,wist, per unit length, that is, the relative angular displacement of a pair of crOSR sections that are
y
unit distance apart.
If the cross sections of the cylinder
remain plane after deformation, then
'I'
the displacement w, along the z-axis, is
zero. The displacements u and v are
readily calculated. Thus, consider any
P
point P(x, y) in the circular cross section, which, before deformation, occupied the position shown in Fig. 20.
After deformation, the point P will oc}>'IQ,2O
cupy a new position P'(x + u, y + v).
In terms of the angular displacement IJ of the point P, we have
u
V
where {3 is the angle between the radius \'ector r and the x-axis so that
x = T cos {3, y = T sin {3. If thc angle 0 is small, we can write
u = -8y,
Since ()
x, y, z
8x.
(33.1)
V =
azX,
O.
1'
cos (x, p)
1'
cos (y, p) =
-P.OlY
If) "" 0,
since, for a circle of radius a, cos (x, If) = x/a and cos (y, If) = Yla.
The only non vanishing component of the couple M produced by the
109
distribution of stresses (33.2) over the end of the cylinder is M., which is
easily cruculated. Thus,
M.
ff (X1"," - d:.c dy
= pa II (x' + y') d:.c dy
?/'T)
= palo,
= ir
+ j-r,y + kr"
= pa( -iy
+ jx),
acting at a point (x, y) on any cross section z-constant, lies in the plane
of the section and is normal to the radius vector r joining the point (x, y)
with the origin (0, 0). The magnitude of T is
(33.3)
From this we see that the maximum stress is a tangential stress that acts
on the boundary of the cylinder and has the magnitude paa, where a is
the radius of the cylinder.
34. Torsion of Cylindrical Bars. Consider a cylindrical bar subjected
to no body forces and free from external forces on its lateral surface. One
end of the bar is fixed in the plane z = 0, while the other end, in the plane
z = l, is twisted by a couple of magnitude M whose moment is directed
along the axis of the bar.
Navier, being guided by Coulomb's solution of the torsion problem for
a circular shaft, assumed that, in the general case of torsion of noncircular bars, the sections of the bar perpendicular to the z-axis will
remain plane. This assumption led him to erroneous conclusions. The
fact that the displacements characterized by formulas (33.1) cannot be
valid for bars whose sections are not circular can be seen from the boundary conditions (29.4). A substitution of the stresses (33.2) in the third
of the boundary conditions yields
(34.1)
~)
+ pax cO"
(11,
v)
= 0,
where 1', as always, denotes the exterior normal to the boundary C of the
cross section R of the beam. But from Fig. 21 it is seen that
1 We denote the unit base vectors along the :1:'-,
tively.
110
::I
(34.2)
:: = sin
80
= cos
(z, 8)
COS
(1/, "),
(x, ..),
that, upon dividing out the nonvanishing factor 1-'01, Eq. (34.'.) becomes
x d:t;
+ y dy
= O.
-+--------.-_:1&
'II =
-azy,
.V
azx,
= a'l'(x, y),
{ :: : ::
1-'01 ( : : -
Y).
(34.5)
throughout the cross section of the cylinder. Furthermore, jf the system
of stresses is to satisfy the boundary conditions (29.4) on the lateral
surface of the cylinder, we see that
(y, v) = 0
on C,
where C is the boundary of the cross section R of the cylinder (Fig. 21).
III
But
ax COB (x '
0",
80
J')
+ ay
IJ", cos (y v)
,
l5
tLp,
d.
(34.6)
d",
dll = Y cos (x, v) - X cos (y, v)
onC.
It follows from (34.5) that ",(x, y) must be a harmonic function throughout the region R bounded by the curve C and that on the boundary C the
normal derivative of ",(x, y) must assume the value given by (34.6).
Since the displacements are single-valued functions, it follows from (34.3)
that ",(x, y) must also be a single-valued function. Thus, the lJroblem of
determining the torsion function ",(x, y) is a special case of the second
boundary-value problem of Potential Theory. This latter lJroblem is
associated with the name of Neumann and consists in determining a function <p that is harmonic in a given region and whose normal derivative is
prescribed on the boundary of the region. We shall meet this problem
again in our study of several problems of elasticity. At this time we
shall simply remark that the harmonic function", is determined by the
boundary condition (34.6) to within an arbitrary constant.' The substitution of '" + constant in formulas (34.4) obviously does n()t alter the
stresses, and it is clear from (34.3) that the addition of a constant to <p
means a shift of the cylinder as a whole in the direction of the z-axis.
Thus, the additive constant in the solution of the problem of Neumann
is immaterial in our case.
The condition for the existence of a solution <P of the Droblem of
Neumann is that the integral of the normal derivative of the function <P,
calculated over the entire boundary C, vanish. This follows from the
identity
i ~: U
ds
div (V<p)
Ie ::
ds =
Ie
Ie
au =
[I
V'<P
au
(y dy
+ xdx)
= 0,
since the integrand is the exact differential of the function 72(:1;2 + y2) +
constant.
It is easy to show that the distribution of stresses given by Eqs. (34.4)
, See Sec. 42.
112
",ex
!c
[~
- y cos (x, v)
+ X cos (y,
v)] ds,
where the line integral is evaluated over the boundary C of the region R.
The integral (34.8) vanishes on account of the boundary condition (34.6).
It is shown in a similar way that
r dxdy = 0,
so that the resultant force acting on the end of the cylinder vanishes.
It rema.i.ns to show that the system of stresses defined by Eqs. (34.4) is
statically equivalent to a torsional couple. The resultant moment of the
exterllal forces applied to the end of the beam is
(34.9)
M. =
ff
R
= ",ex
(XT .. - yr.:) dx dy
ff (x. +
y2
+ X~ -
~;) dx dg.
D =
rr
}}
(X2
+ y2 + X ~
_
iJy
y iJrp) dx dy
(Ix
'
we have
(34.11)
M = Dex.
113
It follows from the foregoing that the torsion problem for a beam of any
cross section is completely solved once the function 'P(x, y) is determined.
For the torsional rigidity D is determined by 'P from (34.10), and the
moment M required to produce the angle 0: of twist per unit length can
be calculated from (34.11) .
. In carrying out the foregoing calculations, no assumptions "'ere made
regarding the location of the origin 0 or concerning the orientation of the
axes x, y. Inasmuch as the first two of the formulas (34.3) represent the
infinitesimal rotation of any cross section of the beam as a whole about
the z-axis, it may seem at first glance that a different choice of the axis of
rotation parallel to the axis of z may yield a different solution of the
problem. For instance, if the axis z' is chosen parallel to the z-axis, and
if it intersects the xy-plane at some point (XI, YI), then the displacements
Ut, VI, and Wt will be
ttl = -o:z(y - YI),
o:z(x -
VI =
y),
WI = O:'PI(X,
XI),
and there is no a priori reason why the functions 'PI(X, y) and ",(x, y)
should be identical.
Calculating stresses that correspond to displacements (UI, VI, WI) yields
r T,v
= p.o: (";'
1'
= p.o:
(~:I
'TZJI
(34.12)
'Tn
+ X- X}
"f,rll
+ YI),
= 'T.a =
0,
2
2
'P1 + a 'P1 =
aax2
ay2
'PI -
a'Pl
ax
+ Yl)
COB
X, v)
+ (a'Pl
iJy
Xl
COB
(y, v)
=
Y cos (x, v) -
cos (y, v)
or
d
dv
('PI
+ YIX
- xlY) = Y
COB
But the function 'PI + YIX - XIY is harmonic, and since it satisfies the
same boundary condition as the function [see (34.6)], it folloWE! from the
uniqueness of solution of the problem of Neumann l that the two can
'P
114
Thus,
"'1 =
1(aa",
lJt/I)
lP(z,.)
(a",
al")
a
dx + "dy =
- ! l dx +
dy ,
c x
uy
P,(z. ,)
uy
X
where the integral is evaluated over an arbitrary path joining some point
Po(Xo, Yo) with an arbitrary point P(x, y) belonging to the region R. If
the region R is simply connected, the function ",(x, y) will be singlevalued; in a multiply connected region, ",(x, y) may turn out to be
multiple-valued. For the time being, we shall be concerned with simply
connected regions, and the discussion in this section will be confined to
such regions.
It is not difficult to phrase the torsion problem in terms of the conjugate
function ",(x, y). Thus, noting the relations (34.2), one can write the
(35.2)
",(x, y) =
1 Some basic results of the thoocy of analytic functions of a complex variable may
be {ouad in 1. S. and E. S. Sokolnikoff, Higher Mathematics for Engineers and Physicists, Chap. X, pp. 440-491.
115
expreesion for the normal derivative :: with the aid of the tangential
derivatives: and
~!, 80 that
dtp
otp
dv = oX
C08
(x, v)
otp
+ ay C08 (y,
v)
otp dy
otp dx
=oxds-oyds'
+ 01/1 dy
"" dl/l.
dB
oy dB
= y
= x ds
+ y dy
dB
1 d
= 2 ds (x'
+ yO).
Hence
dl/l=!~(x2+ 2)
d82d8
y,
so that
(35.3)
1/1
= Y2(x'
+ y2) + const
on C.
0'1/1+0'1/1 =0
ox'
oy'
. R
In
116
= ",(x, y) -
72(X'
+ y).
We have
ow =01/1
--x
ox ox
'
and, upon recalling the formulas (34.4) and (35.1), it follows that
(35.6)
Tn
= p.a
oW
oy'
Since the stress components T .. and T,. are obtained from the function
w(x, y) by differentiation, the latter is called the stress function. It is
readily checked that the stress function w satisfies Poisson's equation
"",T, ""
(35
7)
v '"
o'w
+ o'w
ox'
oy'
. R
In ,
= -2
and on the boundary C of the region R [cf. (35.3) and (35.5)J assumes the
value
w = const.
Consider a family of curves, in the plane of the cross section of the
beam, obtained by setting
w(x, y) = const.
(35.8)
The slope
oW
oX
+ oW dy
oydx
'
dy
ax
T""
T ..,"
Thus, at each point of the curve w(x, y) = const, the stress vector
117
we see that in this case the maximum shearing stress occurs on the
boundary of the section. It is not difficult to prove that in the general case the points at which maximum shearing stress occurs lie on the
boundary C of the section, so that elastic failure of material in shear is
to be expected on the lateral surface of the beam. In order to prove
the assertion, we refer to a theorem.
THEOREM: Let a function <I> of class C2 and not identically equal to a
constant satisfy the inequality V'<I> ~ 0 in the region R; then this function
attains its maximum on the boundary C of the region R.
The proof of this theorem follows at once from the well-known property
of subharmonic functions. It will be recalled that a function <I>(x, y) is
called subharmonic in the region R if at every point (Xl, YI) of the region
(1)
where the integral is evaluated over the circle 'Y of sufficiently small
radius T, with center at (Xl, YI). N ow, if it be assumed that the maximum
value M of a subharmonic function <I>(x, y) ~ const is attained, not on
the boundary C, but at some interior points of R, we arrive at a contradiction. For if S is a set of such interior points and Q is a frontier point
of S, we have from (1)
(2)
= <I>(Q) 5
z!.r
<I>(x, y) ds,
where 'Y is a circle with center at Q and of radius r so small that 'Y is interior
to R. But since 'Y is partly outside S, the mean value of <I> over 'Y is less
than M, that is,
1
-2
rr ,. <I>(x, y) ds
<M
118
V'T' =
Thus V'T2 is nonnegative, and therefore T' is subharmonic in R. AccordiI).gly, T attains its maximum on the boundary of R.
Since the strength of the beam to resist torsion depends on the maximum shearing stress, practical rules for the design of beams carrying
torsional loads are expressed in terms of th'" safe maximum shearing
stress T.
The formula (34.10) for the torsional rigidity D can be phrased in
terms of the stress function' it. The resUlting expression is of great
interest in deducing approximate solutions of the torsion problems by the
membrane analogy discussed in Sec. 46.
We first recall the formula (34.11),
M = Dot,
where
M =
ff (XT,. -
y-r ) dx dy.
Since"
ait
T =~aoy'
we have
M =
-~a 11rr (x oW
dx dy
ax + y OW)
oy
,
R
so that
(35.9)
II (x :: +
-~ ff [o~:} + il~:2]
-~
y :;) dx dy'
dx dy
+ 2p
ff
it dx dy.
+ y cos (y,
v)J ch
+ 2~ JJ it dx dy.
B
on C,
That ~ attains its minimum values on the boundary follows from (35.7). For if ~
were to take on its minimum at some interior point P, then ~. - ~. - 0, ~ ~ 0,
+.. ~ 0 at P. But this is impossible, since ~ +~..... -2 at P.
.
1
119
2p.
JJ 'It dx dy.
R
120
PROBLEMS
1. Consider a circular shaft of length I, radius 11, and shear modulus 1', twisted by
a couple M. Show that the greatest angle of twiRt e and thr maximum shear stress
7' ..
T!~ + T~" are given by
v'
I. A steel shaft of circular cross section 2 in. in diameter and 5 ft long is twisted by
end couples. Find the maximum twisting moment and angle of twist if the greatest
shear stress is not to exceed 10,000 lb per sq in. Take E = 30 X 10' Ib per sq in.,
,,- 0.3.
I. The .haft of the preceding problem is not to be twisted more than 1. What is
the corresponding maximum shear stress?
,. Derive the expression
M.
63,000 H
n
for the torque M. on a solid cirenlar shaft transmitting H hp at a speed n rpm. Hint:
Let the radius of the shaft (or pulley) be T in., and let T = M.IT be the tension in the
belt. Calculate the work done in each minute against M.. (lhp = 33,000 ft-Ib
per min.)
6. Derive the expressions
for the twist per inch length a (radians) in a solid circular shaft of diameter din.,
transmitting H hp at n rpm against a torque of M, in.-lb.
6. How much torque can be transmitted by a solid circular shaft 3 in. in diameter if
the allowable shear stress is 10,000 lb per sq in.? What is the angle of twist per foot
of length? Use I' = 12 X 10' lb per sq in.
36. Torsion of Elliptical Cylinder. It was shown above that the solution of the torsion problem for a solid cylinder of arbitrary cross section is
completely determined if one obtains the harmonic function", that on the
boundary C of the cross section assumes the value
(36.1)
Consider the harmonic function
(36.2)
'" =
C2(X2 -
y2)
+ k 2,
where C and k are constants. The function defined by (36.2) will enable
us to solve the torsion problem for some region R on the boundary of
which (36.2) reduces to ,(36.1). Hence points of the boundary C of the
+ k'
~(X2
121
ThuB,
+ 11'),
or
(36.3)
Ii'
<
if we choose c'
y'
+ b2 ""
I,
and
k
b=
a = .----,
V%' - c
k
.
V~ +c2
Then
C' -
a'b'
k' - a' + bi
1 a' - b'
- 'la'
+ b"
pO! ( -
~ + x).
Hence
-2paa'y
a' + b"
(36.5)
ff
(XT,. -
yr.~) dx dy
y'dXdY)
122
we have
1I"pcxa'b'
M = at
+ bT
n=
'lrpa 3b'
Of
+ b'
Jt was shown in the preceding section that the maximum shearing stress
on any cross section occurs on the boundary of the section. The location
of the points on the boundary at which the greatest stress Tm .. occurs can
be determined 1 by maximizing the expression for T that has been obtained
as a function of a single variable by utilizing the equation of the boundary
C. In the case of an elliptical cylinder, the points of greatest shearing
stress can be found easily from some simple geometrical considerations.
FIG. 22
Consider an elliptical section, shown in Fig. 22, and draw from the
center of the ellipse a semidiamet,er op to an arbitrary point P(x, y) of the
boundary. Since the diameter of the ellipse conjugate to the diameter
through P is parallel to the tangent line' at P(x, y), it follows that the
conjugate semidiameter op' intersects the curve at the point P'(x', y'),
where
,
bx
x' = _ ay,
11 = u
b
When the stresses at P(x, y) are written in terms of the coordinates x',
y' of the point P', we have
stress at P(x, y) is
T -
I. =
+Y
_~
2IUXab _ lx"
T" T Tii. = a' + bt V
123
2~b.J
+ b' T ,
a'
= a'
+ b'
Thus, the maximum stress occurs at the extremities of the minor axis of
the ellipse, contrary to an intuitive expectation that the maximum stress
would be at the points of maximum curvature.
It is easy to verify that the conjugate harmonic function ip, apart from
a nonessential constant, is 1
a' - b2
(36.6)
ip = - a' + b2 xy.
This function determines the warping of the cross sections of the cylinder.
for the displacement along the z-axis is
given by w = aip(x, y). The contour
lines, obtained by setting ip = const,
are the hyperbolas shown in Fig. 23.
The dotted lines indicate the portions
of the section that become concave,
and the solid those that become convex, when the cylinder is twisted by
a couple in the directions shown in the
FIG.2:l
figure by arrows.
The lines of shearing stress are determined by drawing the contour lines
for the surface z = 'I'(x, y). Setting 'I'(x, y) = const gives,' in this case,
a family of concentric ellipses,
x'
-c'(a'
y'
az + /j2 -
1 =
+b
a'b'
'
124
From this relation it follows that the m""imum shearing stress occurs at the ends of
the minor ""is of the ellipse.
2. Derive t,hp expression (36.6) from
<p(z, y)
J:' (~dx - ~
dy )
+ const,
and evaluate the line integral over the path consisting of the straight-line segments
from P.(z .. Y.) to Q(z, Y.) and from Q(x, Yo) to P(z, y).
I. Show that the stress function for an ellipticsl section can be written as
y'l
+ b' (XI
ll' + lJi
-a'b'
,.. = as
_. 1
and is thus proportional to the function appearing in the equation of the boundary 01
the section. The problem of determining the sections for which this proportionality
holds has been treated by Leibenson. 1
V';ft = 0,
and determined the equation of the boundary of the cross section of the
cylinder on which the function", reduces to ~(X2 + y2). Inasmuch as
the real and imaginary parts of every analytic function of a complex
variable x + iy satisfy Eq. (37.1), we can build up a list of functions '"
and, by working, so to speak, backward, can determine the equations of
the contours for which these functions", represent the solution of the
torsion problem. For example, if we consider the function (x + iy)",
then by choosing n = 2, we get two solutions, x 2 - y2 and 2xy, of Eq.
(37.1). The first of these solutions was utilized in the preceding section
to solve the torsion problem for an elliptical cylinder. If n is set equal to
3, we obtain the harmonic functions x. - 3xy2 and 3x'y - y.. Now
consider the harmnnic function
(37.3)
c(x 3 - 3xy 2)
+k
= ~(X2
+ y2).
-ry).
125
(x - a)(x - y
= 0,
p.ot
a y(x
T,.
- a),
~~ (x'
+ 2a.;r
y2).
We see from these formulas that the ;t-cOmponeIlt of the shearinI!', stress.
~: x(x
+ 2a).
The
-2d
t---3a~
FIG. 24
FIG. 25
+ y2)
The direction
const
M = %p.otlo,
where 10 = 3 0 a< is the polar moment of inertia of th{) triangle. The
nature of the distortion of the initially plane sections is indicated in Fig.
126
24, where the contour lines of the surface <p(x, y) "" (3x'y - y3)/6a = const
are shown.
It appears from this example, and from that of the preceding section,
that a circular shaft of the aame cross-sectional area as an elliptical beam
or a triangular prism has the greatest
y
torsional rigidity. lOne can also prove
that, if the region is simply connected,
then, for a given moment ill and for a
given cross-sectional area, the smallest
-+\~""'~!-,-.L-:......J.:"'__ _-+_... x maximum stress will be found in a cir'"
cular beam. This is discussed further
in Sec. 47 in connection with the torsion of beams with multiply 00nnected
cross sections.
The effect of grooves or slots in the
FIG. 26
beam on the maximum shearing stress
can be discussed in an elementary way by studying an example due to
C. Weber.'
Consider a pair of harmonie functions,
x
and
x'
+ y"
a(x ~ b',x'__+]/.
x__ ) + !2 b' = a(r cos (J _ b' cosr (J) + 2
! b!
'
= Yz(x 2
+ y')
= Yzr',
so that the equation of the boundary for which the function if; solves the
torsion problem is
b' cos (J)
f1 ( r cos (J - ~~'r~+ 2'1 b" = 2'1 r',
or
cos (J
r' - b' ~ 2a(r' - b') -r- = O.
Factoring this expression gives
(T' - b2)
(1 _2a CrOSIJ) = O.
127
= b
and
r = 2a cos 8,
T...
PROBLEMS
1. Let Do be the torsional rigidity of a circular cylinder, D. that of an elliptical
cylinder, and D, that of a beam whose cross section is an equilatpral trumgle. Show
that for cross sections of equal areas
D. = kD o,
where
2ab
k = (i.
+ b'
~ 1,
T,.
T= vir;. + T;.
T.,
=- ~
(b l
4a
4a
pa
4a'
COB'
(a - iisecl 8).
9)
COB
28,
cos' 8
128
Find the magnitude of the shearing stress at the point (see Fig. 26) where the groove
enters the shaft.
38. Torsion of a Rectangular Beam and of a Triangular Prism. Consider a beam of rectangular cross section, and let one side of the cross
section, of length a, be parallel to the x-axis and that of length b be parallel
to the y-axis. It will be supposed that b ~ a and that the z-axis passes
through the center of the cross section.
The torsion problem will be solved if we succeed in det(\rmining the
function ",(x, y) that is harmonic in the region bounded by x = a/2,
y = b/2 and that assumes on the boundary of the region the values
}~(x' + Y'). In this case, the boundary conditions can be written as
(38.1)
( a) ="8a' + 2'
'" 2'
( , -+ -2b) b8 + -.x'2
2
Y'
.1.
"
= -
lex,
y) =
a',y
ax' + 1.
av;
lex, y) = - ay'
+ 1.
ao",
-= 1
ay'
onx =
2'
av;
ony =
2'
ax'
= 1
and from (38.2) and (38.3) it follows that the boundary values of the
harmonic function lex, y) are
lex, y)
a
onx=2'
I(x, y)
= 2
onY=:r
(3804)
ax' + ay'
== 0,
129
..
f(x, y)
CnX,.(x)y,.(y),
.. -0
where each term of the series satisfies the differential equatior. (38.5), and
where X.(x) and Y.(y) are, respectively, function3 of x alone and of y
alone. Substituting X.(x)Y.(y) in (38.5), and denoting the derivatives
by primes, we 'get
X::(x)Y.(y)
+ X,.(x)Y::(y)
0,
or
X::(x) _
Y::(y)
X.(x) - - y,.(y)'
Since our solutions must satisfy the boundary conditions (38.4), we reject
the terms involving the odd functions sin knx and sinh k.y, and choose the
product X.Y,. of the form
cos k ..x cosh k"y,
where
kyo = (2n
+ 1)....
a
f(x, y) =
.. -0
130
from which it follows that the coefficients e" can be formally determined
by utilizing the scheme used in expanding functions in Fourier series. If
we multiply both members of (38.7) by cos (2m + l)1rx/a and integrate
term by term with respect to x between the limits - a/2 and a/2, then
because of the orthogonal property of trigonometric functions, namely,
0/2
{O
~
0/2
we get
O/2
-0/2
2 cos k..x dx
if m F n,
if m = n,
a
k b
-2 e.. cosh 2"'
,,-0
and since
a2,f
ax. = f(x, y) - 1,
and
a2t/!
ay. = -f(x, y)
+ 1,
at/!
(-1)"
cosh k"y
ax = -x
+ i2 ,,-0
1.. (2n + 1)2 cosh (k"b/2) sm k"x,
at/! _
8a ~
(38.10)
ay - y
(-1)"
sinh k"y
''I'lu!se are the Fourier coefficients for I(z) _ 2, -a/2 < z < a/2, and /(x) _ 0,
-a < x < -a/2 and a/2 < x < a.
131
T ..
-8aJJ.Ci ~
( -1)sinh k.y
= ~ '"' (2n
1)2 cosh (k.b/2) cos k.x,
.-0
(38.11)
Ta. =
JJ.Ci
(-1)"
cosh kny
.
]
Sa ~
1)' cosh (k.b/2) sm k.x .
[ 2x - ,...' '"' (2n
-0
The solutions (38.11) are formal, but the series converge so rapidly that
there is no serious difficulty in justifying the term-by-term differentiation
to show that the equilibrium equations are satisfied. The x-component
of shear obviously vanishes when y = 0, while the y-component at the
mid-point of the longer side is equal to
(38.12)
Toy
\x-an
= JJ.Cia
.-0
8 ~
1
k.b]
[ 1 - ;0 '"' (2n + 1); sech 2 .
n-O
It is not difficult to prove that (38.12) gives the maximum value of the
shearing stress, by taking note of the fact that the term 2x in the brackets
of (38.11) dominates the series. Now in the most unfavorable case (for
convergence) of a square beam (b = a),
(38.13)
But
~
L
n-l
(2n
1 ~
,...
2e-('n+l)(r/2)
1 + e (2n+l)~
n-l
<
-2 L~
9
3r/2
.-1
e-('n+l) (r/2)
= -2 -e-- _r = 0.002.
!ll - e-
Since sech (11'/2) =0 0.4, it follows that the first term in the brackets in
(38.13) gives the value of all the terms in the brackets with the accuracy
(If 72 per cent. Hence, for practical calculations, the value of Twa. can be
assumed to be given by the formula
T ....
==
JJ.OIa
(1 - !
11"
sech
11'_~).
2a
f- /2 /./2
-b/2
-./2
(XT.. - yr ) dx dy
132
calculations is
M = jU%ba '
6
+ 16jU%a' b ~
i..
,..'
", .. 0
+ 1)'
(2n
and since
M = I'abal
3
/:0
..
~ tanh (k.b/2)
.-1
L.,
(2n
+ 1)6
'
1
(2n
+ 1)' =
0.0046,
.. - I
while tanh (1I"b/2a) ~ 0.917. Thus, the first term of the series gives the
value of the sum to within ~ per cent, and one can use, for prllctical purposes, the approximate formula
M
== l'aba 3
3
2a
(38.10),
Noting
Sa' ~ ( -1)"
sinh k ..y
(38.15)
-/I(x, 1/)
at
= 4"
+ 2 (1/"
133
- x")
Sa! ~
cosh k..y
(-1)"
- -;;:a ,,-0
1.. (2n + 1)" cosh (k nb/2) cos k"x.
The solution of the torsion problem for a prism whose cross section is
an isosceles right triangle (Fig. 28) can be obtained from the foregoing
;y
;y
FIG. 27
FIG. 28
~, = ~ + ~ (y' -
x (x - ;) (y 2
) -
;)
Sa' ~
-;a 1..
.-0
and
~" = ~ + ~ (x' -
y2) -
( _1)n
sinh k,.y
(2n + 1)3 sinh (k na/2) cos knx,
(x - ~) (Y - ;)
8a' ~
( -I)"
sinh knx
(k"a/2) cos k"y .
A comparison with the expression (3S.15) for 1f; shows that the function~"
reduces to ~'2(x"
y 2) on the sides x = a/2 and y = a/2, That~" also
satisfies the boundary conditions on these sides can be shown either from
considerations of symmetry or by direct calculation of the boundary
values and by noting the expansion
~
4
2 _ Sa' ~
(-I)"
(2n + 1)1rx,
x - ,..' 1.. (2n + 1)3 cos
a
.. -0
See B. G. Galerkin, Bulletin de l'academu de> sciences de RU8Bie (1919), p. 111, and
G. Koloesoff, CompteB rendu8 hebdomadaires de> "Mnce. de l'acadtmie de> 8Ciences, Paris,
vol. 178 (1924), p. 2057,
1
134
"'I
+ "'2)
a
-XV + 2 (x + y)
4a2 ~
ra
L. (2n
.. -0
Thus,
(-1)"
~+
(39.1)
1:
+ b. sin k~)
i-I
with
(39.2)
cos nt dt,
135
+ f(lIo- )J.
The 8ymbolsf(lIo+) andf(lIo-) stand for the right- and left-hand limits
of f( II) as II --+ 110.
The restrictions imposed UpOll the function f( II) in this theorem are
known as the Dirichlet conditions. 1 We assume that the reader is familiar
with this theorem.
If f(lI) not only satisfies the conditions of Dirichlet but is continuous
in the closed interval (0, 21r),' then one can show that the Fourier series
converges uniformly in the closed interval (0, 211-).
We also have the following theorem concerning the bounds on the
coefficients in Fourier series:
THEOREM: If the function f( 6) is periodic and is such that its pth derivative
satisfies the conditions of Dirichlet in the interval (0, 2... ), then the Fourier
coefficients for f( 9) satisfy the inequalities
M
and
< n,,';'l'
la.
cos nil
!a..!
+ !b.! < ~,
n
converges, it follows from the Weierstrass M test that the Fourier series
for a function whose first derivative satisfies the conditions of Dirichlet
is absolutely and uniformly convergent and hence can be integrated term
by term.3
Since the coefficients of the series obtained by differentiating the series
term by term are of the order na.. and nb., it is clear that, in order to
ensure the convergence of the derived series, it is sufficient to demand
1 The restrictions imposed on the function 1(9) can be relaxed, and it is sufficient to
demand that 1(9) be a function of bounded variation.
In this case, the requirement of periodicity imposes the condition 1(0) - f(2r).
I As a matter of fact, ('!Very Fourier series can be integrated term by term.
136
that the second derivative f"(J) fulfill the conditions of Dirichlet in the
interval (0, 211").
The Fourier series (39.1)- can be written in an equivalent form
(39.3)
I( (J) = c.
..
..
k-I
k-l
Lc~' + L
C_.e-ikl
where
(39.4)
c..
1 (2r
= 211"
Jo
0, 1, 2, ...).
(n =
f(t)c'n'dt,
e' = cos u
U
+ i sin u,
>
(39.5)
Co
ao
:f
Then
. c.e""
L
i--..
=~
=~
i (~ + ~) (C~8
..
i-I
+b
sin k8).
k-I
Let h(8) and !t(8) be a pair of real functions, each of which can be
expanded in Fourier series in the interval (0, 2r), and form the complex
function ft(8) + if,(O). Then
(39.6)
f1(0)
+ ifz(O)
..
L c.e
ik
',
, _ _ flO
where
c.. =
i..102r
[!tel)
+ if2(t)]e-int dt,
If we set
c..
= "Y.
(n =
+ i3..,
0, 1, 2, ...).
137
+ ifJ(D)
..
l
.. --. .
L
(')'k
(')'t
1:-- -
+i
=
')'0
..
L
k--
110
[(lit
k)
COB kIJ
+ ('Y' -
+ ilio
-1
Hence
where
!1(fJ) = %ao
%ao = 'Yo,
Y2a~
lio,
a. = 'Y'
+ 'Y-.,
a~ = li.
L.,
(k = 1,2,3, .
b. = -li. + li .....,
bk = 'Y. - 'Y .....,
.).
It follows from these formulas that the representation of a complex function f1(IJ) + i!2(fJ) in a series of the type (39.3) is unique, since the representation of the functions !1(IJ) and f2(IJ) in series of the type (39.1) is
unique.
REFERENCES FOR COLLATERAL READING
D. Jackson: Fourier Series and Orthogonal Polynomials, Mathematical Association of
America, Chicago, Chap. 1.
I. S. Sokolnikoff: Advanced Calculus, McGraw-Hill Book Company, Inc., New York,
Chap. XI.
E. T. Whittaker and G. N. Watson: Modem Analysie, Cambridge University Press,
London, p. 170.
K. Knopp: Theory and Application of Infinite Series, Blackie &: Son, Ltd., Glasgow.
p.356.
H. S. Carslaw: Fourier Series and Integrals, The Macmillan Company, New York.
R. V. Churchill: Fourier Series and Boundary Value Problems, MoGraw-Hill Book
Company, Inc., New York.
138
+ w(x, y)
(40.1)
av
au
ax
ay
-= --,
V'v =
o.
fe f(A) dA =
O.
139
=2ri.
Jc( ~.
~ - a
This latter formula, in conjunction with Cauchy's Integral Theorem,
can be used to establish Cauchy's Integral Formula.
CAUCHY'S IN'l'EGRAL FORMULA: If a = a is an interior point of the regicm
R bounded by a contour C, then
(40.2)
J.
(fW da
21rtjca- a
f(a),
f(a)
= _!_. ( fer)
dr.
z...t}cr-a
!'Ca)
_!_ [ fCr) dr ,
z...i Jc (r -
a)2
and, in general,
j<n)(a) = n! [ fer) dr .
21ri c (.\ - 5)n+1
The Integral Formula of Cauchy can be used to establish the fact that
an analytic function 1(i!) can be expanded in Taylor'S series, so that
f(a) = f(a)
+ !'(a)(! -
a)
(a)
+ ... + -,(! n.
I(n)
a)n
+ ....
This series converges to f(IJ) at every point IJ interior to any circle 'Y that
iies within the region R and whose center is at a. Moreover, the representation of f(a) in Taylor'S fleries is unique.
Consider now the region II bounded by two concentric circles C, and
CI, s.nd let 6 = a be the center of the circles. If I(a) is continuous in the
closed annular region formed by C , and C 2 , and if it is analytic at every
140
interior point of the ring, then one can represent I(a) by the. Laurent series
L- br.(a -
I(a) =
a)lo,
JI-- ..
where
(k
= 0, 1, 2, ...),
I(a) =
(a
~~) .. +
a)
a)2
+
a = a and
+ . . . + b_.
+ b_t
+ bo + btl" + btl' + . . .) dj
Jcr I(a) da = Jcr(b_
l'"
j2
j
=
b- t
i ~=
21rib_t.
g(a)
I(a = (a - a)h(a)
has a 8imple pole at a = a (m = 1), then the residue at a = a ie g(a)/h(a).
In general, when C encloses n poles at a = a1, a = a2, . . , 8 = an, the
last equation is replaced by
/c/(a) da
= 2ri
(SUll'
of residues at poles).
JcrI(a) da
1
-,
a-al
1
--,
a-a,
14i
where F(r) is any continuous function defined on the simple closed boundary C and ~ is interior to R, then this integral defines some function of a
and it is easy to verify that 4>1(3) has the derivative 4>;(3), which is given
by the formula
,
( F(r) dr
<PI
21ri
F(r) dr .
Jc (r -
3),,+1
Thus, the function <PI(a) is analytic for every value of a that is interior to
the region R bounded by C. If a is some point exterior to the region R,
then the integral
(40.5)
defines some function 4>2(a), and it is easy to see that 4>2(a) likewise has
derivatives of all orders and hence is analytic. Thus, the integrals (40.4)
and (40.5) of Cauchy's type define two analytic functions that in general
will be distinct. The situation here is the same even when FCr) represents
the boundary values of some analytic functionfCa). For, by (40.3), if Ais
interior to the contour, the value of the integral is precisely equal to f(3),
and if a is outside the contour, then the integral defines the function 0,
since the integrand f(t)/(r - a) is an analytic function of r throughout
R. It should be observed that, as a tends to some definite point r on the
contour from the interior of R and from the exterior, the difference
between the two limiting values is fer) - 0 = fCr).
One can raise a similar question regarding the connection of the limiting
values of the functions 4>1(a) and 4>2(a) with the density function F(r). If
we place no restrictions on the function F(!;) beyond continuity on the
contour, then the problem becomes an exceedingly difficult one. If,
however, some further restrictions on Fer) are imposed, then it is possible
to establish a definite connection of the dellllity function F(r) with
lim 4>l(i) and lim 4>2U).
l-+t
l-+t
142
sal:Uijie8 the Holder condition on a smooth closed contour C, then the limit8
4>+(t) and 4>-(t) a8 a approache8 an arbitrary point t on C from the interior
and exterior of C, respectively, are:
4>+(t) ...
(40.7)
4>-(t)
! F(t) + ~
F(r) dr,
Z,...z}cj-t
2
-
.! F(t) + ~
F(r) ds.
Z?rt}cs-t
integral (
Jc-!' - t
pat value of (
Jcr -t
....0
f.
F{t) dr.
C-Ll'-t
4>1(~)
143
55 O.
dr
= 2,..i jo W - 4) = 0
4>1(3)
lor every position of the point a in the region R.l Hence if we add this
integral to an integral of Cauchy's type that defines an analytic function
>1'(3), we shall obtain another integral of Cauchy's type that defines the
same analytic function >1'(3). It follows from these remarks that no conclusion can be drawn concerning the equality of the density functions
F1(t} and F.(r} from the equality of the two integrals
..!.. (
F,(f) dr =
27rzjor-a
J.... (
F.(r) dr
27rzjor-3
for all values of 3 in the interior of C. We shall see, however, that if some
additional restrictions are imposed on the density functions and on the
contour C, then the equality will obtain. This is the subject of the next
section, which contains a discussion of the Theorem of Harnack.
41. Theorem of Harnack.' In considering the applications of the
theory of functions of a complex variable to problems in elasticity, we
shall most frequently deal with the region bounded by the unit circle,
that is, the region 131 ~ 1. In order to avoid a possible misinterpretation
of the formulas, we shall draw the unit circle in the complex t-pJane,
where r = ~ + i." (~ and." being real). The boundary of the unit circle
111 ~ 1 will be denoted by the letter 'Y, and the points on the boundary 'Y
by 0" = eit . 3 All functions of the argument 8 will be assumed to be
periodic, so that 1(8 + 211") = f(J).
THEOREM: Let f(J) and !p(8) be continuous real functions of the argument
8 (defined on the unit circle 'Y); if
(41.1)
for
all,'al~8
=..!.
r 1"(8) du
2rij..,u-r
==
1"(8).
If the point r i8 exterior to oy, and if the equality (41.1) is true for all values
'F
or
r(r -
i) -
'hi
1, 1 h
Hr _ II - if; ence
dr
lor<r -
1
i) - 2.rii
dr
lor -
r T - i1 - i1 - o.
1
dt
i - 'hii}O
144
of i', then
f(fJ) = 'I'(fJ)
+ Canst.
We consider first the case when the point i' is inside 'Y.
equality (41.1) that
J..,
2ri ).,
It follows from
where F(fJ) "'" f(fJ) - '1'(8), and we shall prove that F(8) "" O.
Now since Ii'I < 1, we have
I I i ' i'2
"--=-+-+-+
- i' "
,,2
,,3
and
(41.2)
L
.-0
~
=
where [see (39.4) and (39.5)]
Un -
ibn
1
= -.
J,
2,".,
(an - ibn)i'n,
F(8),,--1 d"
= -1
2"'0
. . . ).
"-i'=
1"
,,'
-r-P-P-
and
(41.3)
where
Un
+ ibn = ~
F(8)"n-l dcr
21r
'
(n = 1, 2, 3, ..).
145
+ const.
It follows from this proof that if the point l" is outside")" and if in addition
to the equality (41.1) we have the equality
11
-.
21ri7
11
fee) du
- = -.
IT
2lN7
du
'P(e) -,
IT
thenf(e) = 'P(e).
An important corollary follows from this theorem.
COROLLARY: If we have four real continuou8 ,functions f" f.,
the foUowing aimultaneous equalities for all values of l":
_!_ /, fl
2ri
+ if. du
IT-l"
_!_ /, 'P,
2ri
2ri
IT-l"
2ri
+ icp. du
IT-l"
'PI -
'PI,
'P' and
'
i'P'du
IT-l"
'
then
if Il"l
and
cp, =
it + const,
'P. = f2
+ const,
<
1,
if Il"l
>
1.
This corollary follows at once from Harnack's Theorem when we consider the results of adding and subtracting the equalities in question.
REFERENCES FOR COLLATERAL READING
146
I" = f(9),
va,
+ iv(~, ,,)
(42.2)
+ P(If)
= 2/(8)
on 'Y.
2~
du ,.' where I
n u -,
is any
point interior to 'Y, and integrate over the circle 'Y, we obtain the formula
au
au
where M and '" are positive constants. This condition is less restrictiw than tbe
requirement of the existence of a bounded derivative.
Since F(i) - F(O)
'(11) - '(0)
1
p'(O) ;;
if n > 0,
if 11
O.
147
P(I) =
('12.4)
?N,
f(O),. du - ao
~u-,
+ ibo.
+ t'b
1
= ---:;
'In
and hence
2a o =
(42.5)
~
1I"t
1
~
f(
9) du
-
f(9) du
'Y
=~
11"
ao
+' t'b
0,
The quantity bo is left undetermined, as one would expect, since the funcis determined to within an arbitrary real constant.
tion
Inserting the value of ao from (42.5) in (42.4), we have
va, ., )
(42.6)
F(I) =
7rZ
f(O) dO' -
~u-I
J:_.1
27rt
<RF(r)
f(8) du
u
+ ibo
~lf(J)
+ Ir!!: + ibo'
~
u-Iu
27rt
0'
1
ua,.,,) = 27r
{2T
Jo
1 -
eiB
(1 - p2)f(O) dO
2p cos (0 - ,y)
+ po"
This is the integral of Poisson, which gives the solution of the problem
of Dirichlet. It is possible to prove that (42.7) represents the solution
of the problem of Dirichlet under the assumption that f(O) is merely a
piecewise continuous function.'
The discussion of this section was confined to the general solution of the
first boundary-value problems of Potential Theory, when the boundary
curve is a circle. It is possible to generalize the formulas obtained above
so as to make them apply to any simply connected region. This is done
by introducing a mapping function, and we proceed next to an outf.ne of
some basic notions that underlie the idea of conformal mapping of simply
connected domains.
43. Conformal Mapping. Let the functional relationship A = w(l) set
up a correspondence between the points I = E + i." of the complex
I-plane and a = x + iy of the complex a-plane. If a = w(l) is analytic
in some region R- of the I-plane, then the totality of values A belongs to
some region R' of the a-plane and it is said that the region R is mapped
1 See O. D. Kellogg, Foundations of Potential Theory, and G. C. Evans, The
Logarithmic Potential, Cha.p. IV, f01" It discussion of the problem of N eulOann
148
into the region R' by the mapping function w(r). If C is some curve
drawn in the region R and the point r is allowed to move along C, then the
corresponding point i will trace a curve C' in the i-plane and C' is called
the map of C (Fig. 29).
The relationship between the curves C and C' is interesting. Consider
a pair of points t and r + At on C, and let the arc length between them
be A8 = FQ. The corresponding points in the region R' are denoted by
FIG. 29
aand i + Ai, and the distance between them, measured along the curve
C', is As' = FQ'. Since the ratio of the lengths of arc elements has the
same limit as the ratio of the lengths of the corresponding chords,
4t-O
As
.>.r-o
IAr
Ai 1= Idi I
dt
pendent of the manner in which Ai -+ 0 and it follows that the transformation causes elements of arc passing through P in any direction to
experience a change in length whose magnitude is determined by the
modulus of ~ calculated at P.
It will be shown next that the argument of ~ determines the orientation
of the element of arc A8' relative to A8. The argument of the complex
number At is measured by the angle 8 made by the chord PQ with the
f-axis, while the argument of Ai is measured by the corresponding a.ngle 8'
between the x-axis and the chord P' Q'. Hence the difference between
the angles 8' and 8 is equal to
A~
149
at P and to 0' at P', respectively, and hence l arg ~ is the angle of rotation
of the element of are A8' relative to A8. It follows immediately from this
statement that, if C, and C2 are two curves in the I-plane that intersect
at an angle 'F, then the corresponding curves C~ and C~ in the rplane also
intersect at an angle 'F, since the tangents to these curves are rotated
through the Bame angle. A transformation that preserves angles is
called conlcmnal, and thus one can state the following theorem.
THEOREM: The mapping performed by an analytic lunction "'(!) is conIcmnal at all points 01 the I-plane where ",'(!) ;>6 o.
We shall be concerned, for the most part, with the mapping of simply
connected regions, where the mapping is one-to-one and hence ",'(!) ;>6 o.
The regions Rand R' may, however, be finite or infinite. It should be
noted that if the region R is finite and R' is infinite, then the function
"'(!) must become infinite at some point a of the region R; otherwise we
could not have a point in the region R that corresponds to the point at
infinity in the region R'. It is possible to show that at such points the
function "'(!) has a simple pole, so that its structure in the neighborhood
of the point is
"'(!) = ! ~ a
+/(!),
ae+~+~+ +~+
...
Let there be given two arbitrary simply connected regions Rand R',
each of which is bounded by a simple closed contour that can be represented parametrically by'
~
= W),
"
= ,,(t),
".pIe
The term
150
maps the region R' on an infinite plane with a circular hole. In general,
it will be found convenient to map finite, simply connected regions on
the unit circle 1.11 ~ 1 and infinite regions on the portion of the .I-plane
defined by the equation It I ~ 1.
In regard to the mapping of multiply connected regions, we shall make
a few general remarks. It can be shown that a doubly connected region
R' can be mapped on a circular ring but that the radii of the circles
making up the ring cannot be chosen arbitrarily. It is obvious that in
general one can map in one-to-one manner only regions of like connectivity. The condition of like connectivity, however, is not sufficient for
the existence of a mapping function.
The affirmative answer to the question of the existence of a function
.I = wet) that maps conformally a given region in the .I-plane on another
given region in the t-plane helps little in the matter of the actual construction of mapping functions for specified regions.' However. there
1 Mathematiscn.e Annalen, vol. 73, pp. 305-320.
This theorem is associated with the name of Riemann, who proved the existence
of .,(l') under conditions that are less general than those enunciated above .
II the regions R, and R, in the planes " and A" respectively, are mapped on the
unit circle in the l'-plane by functions" = w,(i) and .' = w,(,), then the region R, is
mapped on the region R. by a transformation .' = O(b), obtained by eliminating
i from .' - ",,(i) and b - .,,(i).
'A systematic account of several practical methods for constructing conformal
maps is contained in L. V. Kantorovich and V. 1. Krylov, Approximate Methods of
Higher Analyais (in Russian) (1952), pp. 374-563. See also Zeev Nehari, Conformal
Mapping (1952). A bibliography of numerical methods in conformal mapping was
compiled by W. Seidel, in Construction and Application of Conformal Maps (1952)
pp. 26~280. A method for approximate conformal mapping of polygonal regions on
a unit circle was proposed by 1. S. Hara; Dopovidi Akademii Nauk UlcraimM RSR
(1953), pp. 289-293 (in Ukrainian).
An excellent account of the underlying theory is given by G. M. Goluzin, Geometric
Theory of Functions of a Complex Variable (1952) (in Russian).
A brief cataiogue of useful conformal maps was compiled by H. Kober, Dictionary of
Coufonnal Rl'presentation .(1952).
151
are some explicit formulas that permit one to construct mapping functions for certain classes of regions. If, for example, the region R' is that
bounded by a rectilinear polygon of n sides, then the function w(r) that
maps the interior of the polygon on the unit circle Irl ::; 1 has the form
(43.1)
A= A
f: (I -
+ B,
where t, are the points on the boundary 'Y of the unit circle that correspond to the vertices of the poly!;;on in the a-plane, and the nllmhers 01(11'
are the interior angles at the vertices of the polygon. I
The formula (43.1) was derived by Schwarz and Christoffel' and is
known as the Schwarz-ChristojJel transformation.
REFERENCES FOR COLLATERAL READING
L. Bieberbach: Einfiihrung in die konforme Abbildung, Walter De Gruyter & Co.,
Berlin.
C. Caratheodory: Conformal Representation, Cambridge University Press, London.
A. R. Forsyth: Theory of Functions of a Complex Variable, Cambridge University
Press, London.
W. F. Osgood: Lehrbuch der Funktionentheorie, vol. 1, Teubner-Verlagsgesellschl!.ft,
Leipzig.
R. Rothe, F. Ollendorff, and K. Pohlhausen: Theory of Functions Applied te Engineering Problems, Technology Press, Massachusetts Institute of Technology, Cambridge, Mass.
M. Walker: Conjugate Functions for Engineers, Oxford University Press, London.
Burkhardt-Rasor: Theory of Functions of a Complex Variable, D. C. Heath and
Company, Bosten.
A = wet)
be the function that maps the region R on the unit circle It I ::; 1.
function F(a) can be expressed in terms of the variable r, so that.
(44.2)
The
It I
1.
The formula is usually phrased in terms of mapping of the polygon on the half
plane, but the transformation that maps the unit circle on the half plane does not
alter the form of (43.1).
t For derivation of this formula see H. A. Schwarz, G.sammelte Abhandlungen, vol. 2,
pp. 65-83; E. B. Christeffel, Annali di matematica pura ed appiicata, vol. 1 (1867), pp.
95-103, vol. 4 (1871), pp. 1-9. For a detailed discU88ion of the Schwarz-Christeffel
transformation and of the Schwarz reflection principle, see Zeev Nehari, Conformal
Mapping (19S2), pp. 173-198.
I
152
== ~(Xl
+ y2)
satisfies on the
= ~U;
hence the imaginary part of the function 1(1) defined by (44.2) must
satisfy the condition
(44.3)
on
'Y,
~ 1(1) = .p -
Up,
III
'" = >-2w(l)w(f).
But thU! is a special case of the problem treated in Sec. 42, amI a reference
to (42.4) shows that
f!
If
~ I(r) = .!,
,
or
I(t) =
(44.4)
w(er)w(lt) dcr - ao
.,2 er-t
w(u)w(lt) dcr
21r ., er - t
+ ib o
+ const.
Noting that, on the boundary 'Y of the unit circle It I = 1, er = flo' and
hence It == COl == l/er, one sees that the integral (44.4) can be written as
(44.5)
I(J) ==
'fJ
+ i'#t =
;..
J., C4("}w:.Yer)
dcr
+ const.
The formula (44.5) gives us at once the torsion function 'fJ alld its conjugate.p, so that the solution of the torsion problem is reducllli to quadratures. H the numerator C4(er)w(ljer), of the integrand, hapJ)eD8 to be a
rational function of er, then the integral can be evaluated with the aid of
the theorems on residues.
It is not difficult to express' the torsional rigidity D directly in terms
of the functionJ(t). From (34.10)
(44.6)
==".
IS
ff
".10
(Xl
+ liS) dx dy +".
ff (x ~ - ::)dxey
II
+ ".Do,
'The oa.lculatioDe leading to formulaa (44.7), (44.8), III1If (44.10) alii due to N. L
Muakheilillhvili. Bee, for example, hill paper "Sur Ie probl~ de tGmiou. dee c,.Jiadns
6Iaatiqu.. isotropea," AID ~ I'fNIk ~ ~ . . lMtai, _. II, vol 9
(l929), pp. 295-300,
153
[f (x ~: -
1/
~=) dx dy =
ff [~
+ y2.
= _ ( .pd!.r2 ,
Je
r' = Ai = ",(.,.)'-;'(It)
and
'" = %(f(.,.)
+ 1(1t)];
hence
D. = -~~
(44.7)
Also
I. =
ff
(x'
+ y2) dx dy =
ff
[:y (x"1l)
a:
(xyl)] dx dy
= - !cXY(Xdx - ydy).
But
x = A + A,
2
and we find that
But
Ie AidA
0,
and
!.e I' d~
fe A"a da = fe AI d(%a')
= -
Ie = -
i fe
A2adA = -
ii
'
fe alA dA,
HenCe we can write the
[';'(11)]2",(.,.) dc,,(fT).
If "'(.,.) is a rational function, then the integrands of (44.7) and (44.8) can
be easily evaluated with the aid of theorems on residues and the expression
for the torsional rigidity D Can be obtained in closed form.
We may note that the shear components T.. and T.. of the stress tensor
can likewise be exp,_J directly in terms of the functions F(A) and f(I).
154
Tn -- IT..
/let
(orp.
orp
iJx - , oy
orp + l.-0'"
= /let [-iJx
aX
y -
l X -
.)
tX
.(
. )]
lY ..
But.
and we get
(44.9)
Since F(3)
F[w(.I)
f(.I), we have
= 1'(.1) t!l
F'(3)
d3
f'(.I) _1_.
w'(.I)
1".. -
l1",.
[f'(.I)
= 1t0l w' (.I) -
._(;:)]
lW,
L
.-0
~
5 = w(.I) =
an.ln,
w(u)w
0) i i
Lb.u + Lb.u-,
.
amu m
m=O
anu-'
n=O
n-O
n-l
where
(44.12)
b. =
Lan+,.a,.
,-0
or
(44.13)
f(.I) = rp
+ it/! =
.-0Lb.l"",
155
where Cauchy's Integral Formula and Eq. (41.2) have been used. The
expression (44.13) for the complex stress function '(J + i", was derived by
R. M. Morris! by a different method and was used to obtain formal solutions of the problem of torsion for those cases in which the complex constants an are known.
A formal expression involving the constants an can be given for the torsional rigidity D = !l(Io + Do) [see (44.6). Equation (44.8) for the
moment of inertia 10 can be written as
10 = -
[w(o-)w
G) ]
w (;) dw(o-).
.-0
wG) dw(er)
1:
~
cner n dl),
,, _ _ GO
where
1:
1:
~
c,
(n
+ r)a,+rdr,
r- 0
(44.14)
C_n =
rdn+..a"
(n = 0, 1, 2, . . . ).
r-O
(I)
10
= 31
/:7 (1:
bmo- m +
".-0
00
1: bmo---m) (1:
".-1
110
C.O-'
n-O
c_.cr-n) dl).
n-l
Since 0- = eiB , we see that the integral of every term involving 0-" (n
vanishes and we are left with
10 =
Similarly we can write
~ [boro +
0)
i (bnc_" + bnc,,) l
.-1
156
D=
~ [boCo +
.-1I
+ c._,.b. -
(c.Ii.
2OO.Ii.. )
lu!, an illustration of the application of the foregoing procedure, we consider a beam whose crOl!8 section is the cardioid
r = 2c(1
(r' = Xl + y2, tan a = y/x).
suitable mapping function is
~t
a=
+ cos a)
aJ = -2c,
c_, = -2c',
b, =
6c',
Co
Cft
-4c',
c,
b. =
-6c',
c',
c,
2c'.
f{t) = 'P
+ i+ =
I b..s.-0
ic'{6 - 41
+ to),
while
(44.17)
151
x'
y'
(ii+l)i=1
is inverted with respect to ita center, the point (x, y) is carried to the
point (x', y'), which is such that
r'(r')'
==
+ y2)[(X')2 + (y'}']
(x'
= 1.
x
.-+
= -c1 cosh k cos u,
x
y
Y 2 = -1 Sl'nh k sm
. u,
-'----+
x
y
c
or
(45.1)
+ iy
= C
Bec (u
+ ik)
A=
sec (w
+ ik),
= u + iv.
with A = x + iy, W
the resulting function
(45.3)
If we put
2cekr
a = wen == f'
FIG. 30
0,
+e
2k '
r=
i ..
> 0,
> 0,
mapa the cross section R of the cylinder upon the interior of the unit
circle Irl :::; 1.
Tn the preceding section, expressions were derived that give a formal
solution of the torsion problem when the mapping function is expanded
in an infinite series
a = wet) =
..
..
2: a..t- = 2: a.e
ft.-O
'''''.
Jl-O
"The Torsion of a Prism with Cross Section the Inverse of an Ellipse," Journal DJ
Phpictf, vol. 13 (1942), pp. 457-459.
A~
158
a" ==
0,
{ 2c( -1)(a-l)!le-"",
if n == 0, 2, 4, . . ,
if n = 1, 3, 5, . . .
The infinite series entering into (44.13) and (44.15) were then summed
and the torsion function and twisting moment obtained in closed form.
However, since ",(t) is a rational function of r, it is simpler to proceed
directlyl from (44.5), (44.7), and (44.8).
When w(r) from (45.3) is inserted in (44.5), it is seen that
(45.4)
where R, and R. are the residues of the integrand at". = iek and". =
respectively. We have
+ iek)(".2 ~ e
Rl = [(".
R, = 4(r
-ie>
iek) sinh 2k'
.<)(". -
r)L. .
-ie',
and hence
f(r) =
!P
+ i,p
(w
+ ik),
where the constant in (45.4) has been taken equal to -ic' csch 2k.
The shearing stresses may be found, either from the relations
V =
,p - 2 (x'
+ y2),
[CBCh 2k cosh (v + k)
sinh (v + k)
]
cos 2u
cosh 2(" + k) + cos 2u + cosh 2(v + k) ,
cach 2k sinh (v + k)
cosh (0 + k)
]
cos,. [ cos 2,. _ cosh 2(v + k) - COB 2u + cosh 2(0 + k)
.
1' -
pac
T -
2jl.CtC
SIn U
.r
".1(".' - e2k)
10 = 4e'i J7 (".2 + e2k)I(cr2 + e
= -&rc'(R. + R.),
2k)!
dO"
159
R,
d [ -,-;;--;---';;;:",-;-'-:;--;r.;
u'(u' - ell)
]
(u'.+ e21)'(u + ie--l)2 ........
-each' 2k(2 + cosh 4k) _ R
16
"
= -
du
and therefore
10 = 'll"c'(2
u(l - u)
R~
+ e-
2k
)B du
Ri = :2 du2 (u'
u(l - ( 4 )'
+ e2k )3(u + ie
-7S cscha 2k
We
k)' ..;._.
R.,
and hence
M = p.a(Io
+ Do)
p.Cl'II"c 4 (2
csch' 2k
+ eseh
2k}.
The curve resulting from inversion of the ellipse with respect to its
focus is called an elliptic limaCon. The torsion problem for a cylinder
with elliptic-limaQon cross section was treated by Stevenson and Holl
and Rock,l and the corresponding problem for a hyperbolic limRQon by
Lin, Whitehead, and Yang.' Methods of solution used by these investigators differ somewhat from those presented here.
As an example of the type of calculations required when the mapping
function is not rational, consider the map of the unit circle obtained with
the aid of
(45.5)
a=
w(t) "" a
v'fTI,
where a
> O.
v'fTI
160
FIG. 31
When
(-,.. ~ II ~ "'),
= YilII,
and
r
= 2 C08 YilII.
Hence
(-,..~II~,..).
Let R and
Re''''
y2 cos Yil9 e
iB14
Hence
a 0-;;-08 2"'.
Thus, the map of the unit circle is one loop of the lemniscate shown in
Fig. 31.
Substituting from (45.5) in the formula (44.5), we have
(4;).6)
fm
1
= --
1,
2,.."
at
yl-F~ ~~-+-~
------ du =
u-s
a2
-.
2...
f'Y..._;;u-s
+IT
1
-~-
du
-_.
Since the sign of the square root must be chosen positive, we can write
~ "" e"".
If we cut the negative axis as shown in Fig. 32, then the integrand of
(45.6) \\;11 he a single-valued funct.ion in the simply connected region
161
FIG. 32
indicated in the figure and the only singularity of the integrand is the
pole at fT = I. Hence
2~ [i g(fT, I) dtT +
fl
g(fT, I) dtr
g(u, I) dtr
lo-I
g(fT,)) dtr]
R,
where
g(fT, I)
l+fT '
== _ /:
vu (fT -I)
R = 1
Thus,
1 /, g(fT, I)
--;
2n l'
+ I.
dtT = -
1
2--;
n
[fO
-1
g(fT, I) dtr
1-1
0
g(u, I) dfT
+
__ !
-
1r
1 - t _!i:!_
Jo Vi t + .I
g(fT, I)
dtT] + R
+ 1 + I,
0
where we have dropped the integral over the small circle 6, since it
vanishes when the radius of fj tends to zero, and where the integrals
over the portion of the real axis between 0 and -1 are combined hy
making an obvious change of variable and by noting the difference in
sign of the function y.: on the upper and lower banks of the cut. Integrating and dropping the nonessential additive constant, we have finally
fer)
~ 1 +J 10
1r
VI
+ 0,
0
1
i
g 1 - i
where
1 +i0
log - ._- = 21. VI ( 1 - -r
l-~Vr
3
+ r'
-5 - ..
162
The function f(t) solves the torsion problem for a beal1l whose cross section is one loop of the lemniscate. The calculation of stresses presents
no serious difficulty and is left as an exercise to the reader.
As a third example of this general method of attack upon problems of
torsion, consider the case of a cylinder whose cross section is bounded by
two circular arcs.'
Consider a region R of the complex A-plane bounded by two circular
arcs C, and Ct making an angle a " 0 at their points of intersection
a = h and a = a. (Fig. 33a). It is obvious that the transformation
y
i~~
C.
ex
h
:t&
Ia)
(b)
FIG.
(e)
33
163
and
_ (3
1 - (3
+ 1)2 - i(3
+ 1)2 + i(3
- 1)2
- 1)2
The solution of the torsion problem for the region described in (1) above
is given .llext.
From
23
;- = A2 + l'
it follows that, on the boundary of R, 3 = [1 - (1 - 1T2)~JI<T, where the
appropriate branch of the square root is determined from the observation
that the imaginary part of 3 is positive whenever 0 in <T = ei6 lies between
o and T. Then the numerator of the integrand in (44.5) is
W(IT)"
0)
= [1 -
(1 - 1T2)~i] [1 - (1 - ~Y1
Substituting this expression in (44.5) and evaluating the resulting integrals, we get
i(l - ;-2)
1 - ;f(;-) = -i(1 - 12)~ log - - + const
1 +1
"'1
'
F(3)
~ ~'
+1-
",(x, y) = {T(X'
+ y(x' + y2 - 1)[(x 2 +
where
S = [,_,_(1::__-",x'-,--_y<,;;2:,-)'.+--'--,.4.::;y~2),-~
(1
+ X)2 + y'
. [(x'
and
tan- 1
+ y2)' -2y
l-x'-y2
1)
x2
'" = ~(x'
as it should.
+ y') + const,
164
(i + r)'
+ _4_
,..(i + r)
~-plane
1I'2a
r) +
r
nst
co
= _!_ [ . 2 + 2(~2 + 1)
2....
+ (a' -a"
1)21
+ a~ + co
1 -
og 1
nst] .
165
.,(,1
!'
X!
+ a"
" >0,
a>
1,
anddeduee
in J ..!
f(r) = (a' - 1)(,'
T..
and
T .,
+ a')
166
== (T dy
:;t + fx (
fJ2 z
= T fJx,dxdy,
T dy
~) p ax -
(T dy
:;t
fJ'z
T fJy' ax dy.
1 A detailed discUllSion of the procedure employed in deriving aorne approximate
formulas for sections whose components are rectangles (as well as for aome tubular
sections) is given in the National Advi$ory Committee for Aeronautics Report 334, by
G. W. Trayer and H. W. March, entitled "The Torsion of Members Having Sections
Common in Aircraft Construction." This report CO.!ltains an extensive bibliography
and a compariaon of their formulas with those obtained by other investige.tol'll. A
description of the experimental procedure used in studying torsion of beams with the
aid of soap films is given by Trayer and March and also by A. A. Griffith and G. I.
Taylor in the AdrMory Committee 011 Aeronautics Technical Report, Great Britain
(1917-1918). A bri..! account of the procedure employed by Griffith and Taylor is
found in S. Timoshenko and J. N. Goodier, Theory of Elasticity, See. 00.
167
P rk dy
02
+ T 0~2Z rk dy + T oy'Z rk dy =
and we have
b'z
02Z
0,
-p
-+-=-.
ox'
oy' T
(46.1)
z = lTV'
on the boundary.
Equation (46.3) is identical with that obtained in Sec. 35 for the stress
function V, and it is clear from the discussion there that the slope of the
membrane at any point is proportional to the magnitude of the shearing
stress
or
J1.a
+ (av)2
VI(av)'
ax
oy
= J1.a
dvdv
ff
R
it becomes clear that the volume between the plane of the opening z = 0
and the surface of the membrane is proportional to the torsional rigidity
D of the section. Since the contour lines of the membrane can be mapped
out, and the slope at each point and the volume under the membrane can
be determined, one can secure the desired information concerning the lines
of shearing stress and the torsional rigidity of the beam from experimental
measurements.
A consideration of the equation of the unloaded membrane,
02Z
iJ~2
02Z
+ oy'
= 0,
168
heights of the membrane over the contour of the section have the values
given by (46.5).
The membrane analogy has been used in an interesting way by Timomenko to discuss an approximate behavior of a beam of narrow, rectangular crOBB section and in analyzing the stress concentration near
fillets in channel sections and I beams. It is interesting to note that the
maximum shearing stress in a narrow beam of thickness c is twice as
great as in a circular shaft of diameter c and subjected to the same twist.
The details of the calculations and furt.her discussion will be found in
Timoshenko and Goodier's Theory of Elasticity, Sees. 93 and 94.
The technique of measuring the ordinates of the membrane has been
discussed by Thiel,l who used stereoscopic photography, while Reichenbieher! has decsribed an optical device for the automatic plotting of the
contour lines of the membrane. The soap film has been replaced by a
paraffin surface by Kopf and Weber,' and by the interface between two
immiscible liquids by Piccard and Baas' and by Sunatani, Matuyama,
and Hatamura. L. Foppl' and Deutler7 have discussed the form of the
membrane analogy in which the film is under zero resultant pressure and
its boundary has variable height.
The boundary-value problems of torsion can also be interpreted in
terms of various hydrodynamical analogies. These are discussed briefly
in Timoshenko and Goodier's Theory of Elasticity, Sec. 100, where several
references are given. To these may be added a paper by Den Hartog
and McGivern 8 in which experimental technique is described.
The analogy between the torsion of a cylinder and the potential of a
plane electric field affords another way of obt.aining experimental 801uI A. Thiel, "Pbotogrammetrisches Verfahren zur vf'rstlchsmii.ssigen I.osung von
Torsionaaufgaben (nach einem Seif,mhautgJeichnis von L. Foppl)," Ingenio/tr Archiv,
vol. 5 (1934), pp. 417-429.
H. Reichenbacher, "Selbsttatige Auamessung "on Seifenhautmodellen (Anwendung auf daa Tormonsproblem)," Ingenieur Archiv, vol. i (1936), pp. 257-272.
E. Kopt and E. Weber, "Verfahren zuy Ermittlung der Torsionsbeanspruchung
mittels Membranmodell," Zeit8dorijl de8 V.,..ine3 deulJlc1aer IngenUuTe, vol. 78 (1934),
pp.911HH4.
A. Piccard and L. B_, "Mode experimental nouveau relatif a I'application des
surfaces a courbure constante a la solution du probleme de la torsion des barres prism&tiques," Proceedings of Ike Second InJernational Ccmgres. for Applied Meclttmw,
ZUrich (1927), pp. 195-199.
Chid6 Sunatani, Tokuzo Matuyama, and Motomune Hatamura, "The Solution of
Torsion Problems by Means of a Liquid Surface," TeehniaAl Reports of Ike TMoh
Imperial Utaiversity, vol. 12 (1937), pp. 374-396
L. Foppl, "Eine Ergiinzung des Pra.ndtlBchen Seifenhaut-Gleiehnisees JIll)' Torsion," ZeiJBcltriftfi.ir angewandle Mathematik mul Mechanik, vol. 15 (1935), pp.37-40.
T H. Dentler, "Zur versuchsmii.ssigen Liisung von Torsionsaufgaben mit Hilfe des
Seifenbautgleiebnisses," Ingeniftw Archiv, vol. 9 (1938), pp. 280-282.
"I. P. Den Hartog and J. G. McGivern, "On the Hydrodynamic Analogy of Tal'&ion," Joumol of A'JIPlietl M~, vol. 2 (1935), pp. A46-A,48.
169
tiOll8 of the torsion problem. This is described in See. '1, Chap. III, of
Teehnische Dynamik: by C. B. Biezeno and R. Grammel and in a paper
by H. Cranz. 1
The equation for current flow in a conductor of variable thickness is
identical with that describing the torsion of a shaft of varying circular
section.' This analogy, which yields a practical method for studying
stress concentration in the neighborhood of fillets or grooves in shafts
under torsion, is described in Sec. 104 of Tim08henko and Goodier's
Theory of Elasticity and in papers by Thurn and Bautz,' .Jacobsen,' and
8&1et.'
A discussion of several an&1ogic methods of approximate solution of
Saint-Venant's torsion problem, including extensive bibliographical
references, is contained in two papers by T. J. Higgins in PrQCeeding8 of
the Society for Experimental Stress Analysi8, vol. 2 (1945), pp. 17-27, vol.
3 (1945), pp. 94-101.
4.7. Torsion of Hollow Beams. The discussion of the torsion problem
has been confined thus far to solid beams, so that the region of the cross
section has been simply connected. Hollow or tubular beams are of considerable technical importance, and it is necessary to extend the formulation of the torsion problem so as to include multiply connected regions.
Let it be assumed that a beam has several
longitudinal cavities so that the boundary of the
cross section of the beam is made up of several
simple closed contours. Denote the exterior
contour by Co, and let C 1, C., .. ,C" be the
simple closed contours lying entirely within
the contour Co (Fig. 35). The contours C 1,
C., ... , C" correspond to the cavities of the
FIG. 35
beam. The discussion in Sec. 34 that led to the
formulation of the differential equation (34.5) is valid in this case, and we
have the differential equa.tion
(47.1)
in R,
1 H. <::ran .., "Experimentelle LOsung von Torsion ..... uigaben," Ingenieur Arcbiv,
vol. 4 (1933), pp. 500-509.
See See. 49.
A. Thum and W. Bautz, "Die Ermittlung von Spannungsspitzen in verdrehbean!<pruchten Wellen dureh ein elektrischea Modell," ZeitBchrijt des Vereine8 deui8cher
l . . .ieure, vol. 78 (1934), pp. 17-HI.
L. S. Jacobsen, "Torsional Stresses in Shafts Having Grooves Or Fillets," Journal
of Applied. Mec1umica, vol. 2 (1935), pp. Al54--AI55.
G. Salet, "Determination des paintes de tension dans lea arbres de revolution
dOumia A torsion au moyen d'un modele 6Iectrique," BuUetin d. l'a.sOli4tion teclmiqm
maritime d abontJtiqm, vol. 40 (1936), pp. 341-350, 351.
170
~=
= y cos (x, v) -
X cos (y,
on Ci,
p)
(i = 0, 1, 2, . . . ,n).
on C,'
0, 1, 2, . . . , n),
'" = >2(x 2
+ y2) + k,
(i = 0, 1, 2, . . . ,n),
on Ci ,
where the k, are the integration constants. The value of one of these
constants, say ko, can be specified arbitrarily,' but the remaining n constants k. must be determined so that the function
tp(x, y) =
tp(x, y) =
l
l
(X, Y)
P,(XG.IIO)
(x, v)
P.(z vo)
(a/
dx
X
+ a)
,,'I' dy ,
uy
(lJift
lJift)
- ax - - dy ,
ay
ax
F(x, y) =
P(Z,Y)
Po(:tO,'II0)
[M(x, y) dx
+ N(x,
y) dy]
(4.7.6)
8jj
aN
=
ax'
But if F(x, y), defined by (47,5), is to be single-valued in a mUltiply connected domain, then in addition to (47.6) we must demand that the
integrals
1.
0,
(M dx
+ N dy)
vanish when evaluated over each interior contour forming the boundary
of R. Since '" in (47.4) is a harmonic function, the condition (47.6) is.
t
t
171
clearly, satisfied and ",(x, y) will be single-valued in R if, and only if,
(47.7)
(i
1, 2, . . . ,n).
in R,
on C,
V'", = 0
{
'" = ,%(x'
+ y2) + k,
(i = 0, 1,2, . . . ,n)
w "'" ",(x,
y) - ,%(x'
+ yO),
V'w
= -2
w=
in R,
on Ci,
k,
(i = 0, 1, 2, . . . ,n)
+ !C, (y dx
- x dy) = O.
Jei oy ds
A.. dw ds = -2A,
(47.10)
'j'e, dv
(i = 1,2, . . . ,n).
p.
(x>
+ y' + x ~ -
Y ::) dx dy
=p.
U-(x: +y~:)dXdY,
172
B",= _(X+B't'),
ax
and
ay
ff r
= 2p.
2it -
+" dxdy
+ p.ic it(ydx -
xdy),
where we make use of Green's Theorem, and the subscript C on the line
integral means that the integration is to be performed in appropriate
directions over all the contours C. (i = 0, 1, 2, . . . , n). Now if we
chOO8e the value of it over the contour Co to be zero (that is, ko = 0)
and note the boundary conditions in (47.9), we have
D = 2p.
ifl
n
itdxdy
+ p.
But
~Ci (ydx
- xdy)
=2
kl
fCi (ydx -
JJ dxdy
xdy).
2A.,
Ai
fl
JI
A
it dx dy
2p.k,A
M = 2p.a
(If 't' dx dy +
R
k.A.).
i-I
It will be recalled that the curves it(x, y) = const determine the lines of
shearing stress (sec Sec. 35), and it follows from the boundary conditions in
(47.9) that one can obtain a solution for the torsion problem of a hollow
shaft from the solution of the torsion problem of a solid shaft by deleting the portion of material contained within the curve it(x, y) = const.
Thus, in the discussion of the problem of torsion for an elliptic cylinder
in Sec. 36, it was shown that the lines of shearing stress are similar ellipses,
concentric with the ellipse
x2
y!
iii + Ii' = 1,
representing the cross section of the cylinder.
Accordingly, if we delete
173
as + 11'b' =
x'
(47.13)
(1 - k).
(0
<
<
1).
then the stress function " for an elliptical beam of semi-axes a and b will
have a constant value over the curve (47.13) and the sa.me function"
will thus solve the torsion problem for a hollow beam bounded by similar
elliptic.al cylinders.
The lines of shearing stress for a beam of circular cross section are
circles concentric with the outer boundary. and it follows at once that
the formulas contained in Sec. 33 are applicable to hollow cirCll.lar shafts.
In particular, the torsional rigidity D is
(47.14)
= p:r
2 (a' - a4)
o.
qw
dx dy ==
~kl
dx dy =
Hk1A.
Thus, approximately,
T.
= ~l.
le. ~t =
11 {
2A"
174
80
that
M .. 4,.,.tA~,
l
and
pp.62-68.
175
176
177
I. If CI coeh r, where , - :t
iI/ and r - ~
i., show that the family of
curves ~ - conat defines a set of confocal ellipees. Hence verify that the function
J.
.. =
4"
~l)
2
cos .r,
solves the torsion problem for a tube whose cross section is bounded by confocal
ellipees E - E. and E = El. This result was deduced by A. G. G"""nhill, Quarterly
Jmunal of M~, Oxford Series, vol. 16 (1879), pp. 227-256.
S. Compute the torsional rigidity of a hollow shaft whose cross section is bounded
by two similar ellipses.
'- Use Green's formula to show that
dJ: dy,
ff
(x'
+ 1/') dJ: dy -
po
If
po
11 ("'! + ..;1
do; dy.
(x'
Ii. Use results in the preceding problem and the fact that the Dirichlet integral
'P
(i = 1,2,3).
1 The calcul&tions in this section are far less general and more laborioU8 than the,.
would haft ~ if the apparatus of tensor calculus were at our disposal. A oone.isP
general tensoria1 derivation of the basic equations of linear and nonlinear mechanics of
_tin\K'\l8 media ill eontainl'd in I. S. SokolJH1<of('. T_r Analysis, pp. 290-319
178
(i = 1, 2, 3)
(48.2)
ds'
gii da;,
i-I
where gi' are the metric coefficients that can be calculated' from (48.1).
Let Po and P be two neighboring points in an unstrained medium, and
let these points take the positions P~ and P' after deformation. We shall
confine our discussion to infinitesimal deformations' and shall represent
the displacements in the directions normal to the coordinate surfaces
ai, a2, a, by UI, u" U3, respectively. The curvilinear coordinates of the
points Po and P are a; and ai + dai, respectively. The coordinates of the
points P~ and P' will be denoted by a; + ~. and a, + ~. + da, + d~i'
Then it follows from (48.2) that
Now the length of the element of arc ds joining the points Po and P is
given by
3
(48.3)
ds' =
while the length of the same element in the deformed state is given by
3
(48.4)
(ds')' =
But
(da;
+ dE,)' =
(da,)"
(da;)'
+2
179.
(48.5)
(da')' =
i-I j - l
where
3
(48.6)
G'j
6ij ( II;;
\' 011")
O~,
O~i
+ k-l
L oak ~k + IIi< oa. + IIi; oa.'
1
1
and where we neglect the terms involving the products of ~; and ~~i.
The
Uaj
symbol 6,;, as usual, denotes the Kronecker delta. The expression for
Gi; has been symmetrized by replacing G;; by Y2(G ij + Gji ).
It is clear from (48.2) and (48.5) that the elongations of linear elements
and shears are characterized by the coefficients llii and G,j. Thus, consider a linear element ds, directed along one of the coordinate lines a,.
From (48.2), its length is
.ya;. dai.
~,and
their derivatives.
Noting
(48.7)
180
by the formula
(hi =
..-2 -~;
9,;
sin
then
COB
;.
<Xi;.
The shear components of the strain tensor are defined by the relation
= 2e.;. Substituting in (48.8), we geP
~=.!.__0t_ ;..!.~
2~
2~'
e"j
(Of + gn oa,o~.)
1 - g.i ___!
-2 v'ii:ih,
oaf
_!
if i j.
= - 1- - [ g 0 ( - 14 ) +U 0 ( - 14 ) ]
2 ~ .. oaf
"oa; v'Y;; ,
y'u:.
(48.10)
>." + 21'e,.
'T..
'Ti;
= 2peij
or
Or'Tii
EfT
(1
+ fT)(l
E
'T'j
= 1
+ fT~'
+ e.. + en.
_ 2fT) "
if i
+ 1 + IT eo,
j,
Note that
terms involving produeta of & and ita derivativM, terms that were neP>eted pre
vioualy.
Cf. Sees. 22 "nd 2:1.
181
~"..,,)
aa;
_ ! \'
2 if-I
+ \'
g..;; ag"
g;; oCti
if-I
j_ ( ggi,Ti;)
v'ii:ik
+ Fig yg.; =
aCt;
0,
(i
1, 2, 3),
where g "" vigllg2'I:(Jaa, and the Fi are the components, in the directions of
the coordinate axes, of the body force F.
A complete set of the field equations of linear theory of elasticity, valid in all coordinate systems, is recorded here for the benefit of readers familiar with tensor calculus.
In these formulas a comma followed by the subscripts i, j, . . . denotes the covariant
derivatives with respect to the variables Xi, X;, ,and a repeated index is summed
from 1 to 3. The g'l and gi; are, respectively, the components of covariant and contravariant metric tensors. The meaning of alI other symbols is identical with that
used previously.
a. Hooke'. Law
'rH = )\f}gij + 2~ei;,
{} ~ g,leH,
e'l = 7}(U',1
+ u;,,).
b, Equilibrium Equation.
git.,.'i.k
+ Fi =
Tij.,i
==
0,
Ti ,
in 'T
on l:.
c. Navier'. Equations
= 0,
in T,
We write out the expressions for the strain components (48.7) and
(48.9) and the equations of equilibrium (48,12) for three important special
cases of curvilinear coordinates.
a. Plane Polar Coordinate8. In this case, the index i assumes the
values 1, 2, and according to the usual notation
The coordinate surfaces in this case are circular cylinders perpendicular to the xy-plane (r = const) and radial planes through the origin
(8 = const). The element of arc is given by
dIJ' = dr'
so that
fill
== I,
+ r'd8',
gu = 0,
182
....
I
I
TOg I
FIG. 37
Noting the formulas (48.7) and (48.9), we see that the strain components
in this case are
OUr
(48.13)
err
= TT'
egg
= --
1 aUg
r 0(J
er , =
+-r'r
1 (OUr
2r 7i8 -
Ug
au,) '
+ r 7ir
(48.14)
or
+ ~ Orrg + T
r
0(J
rr _ - T98
+F
= 0
r,
+ ~r OT99
+ ~r rr' + F, = O.
or
a(J
b. Cylindrical Coordinates. The variables involved here are
al = r,
0<, = 8,
a, = z,
and the element of arc in cylindrical coordinates is given by
dB" = dr" + r2 dtl" + dz',
so that
g11 = 1,
gn = 1.
922 = r',
OTT'
183
The surfaces r = const and 8 = const are circular cylinders and radial
planes as in case a above, while the surfaces z = const are planes parallel
FIG. 38
to the xy-plane (Fig. 37). Substituting the values of the metric coefficients in (48.7), (48.9), and (48.12) gives the expressions for the strain
components
iiu.
eM" =
(48.15)
Tr'
1 iiu"
iilJ
ell
=r
e..
= Tz'
iiu.
u.
+ r'
184
(48.16)
= dr'
ds 2
we have
gu = 1,
go, = r" sin" (J,
The strain components, in this case, are
iiu,
-aT'
err =
Ien
e....
(48.17)
! iius + ~,
1 iiu..
r sin (J iia
r at!
+ U.
cot (J
r + Uo -r-'
!. (_~_ iiu,
e,.. =
2 rsm(Jiia
~
r
+ iiU..),
iir
B) ,
ih"
iff
(48.18)
UT
SIn
a",
a.
T .... )
'
185
I. S. Sokolnikoff: TetlllOr Analysis. Theory and Applications, John Wiley '" Sons,
Inc., New York, Sees. 103-113.
F. Odqvist: "Equations de compatibilite pour un systeme de coordonnees triplex
orthogonaux quelconques," Comptu rendus hebdomedair des .tances de l'academie
des 8ciences, Paris, vol. 205 (1937), pp. 202-'204.
F. K. G. Odqvist: "Kompatibilitatsgleichungen bei Zylinderkoordinater," Zeilachrift
fur angewandte Mathematik und Mechanik, vol. 14 (1934), pp. 123-124.
L. Brillouin: "Lea lois de l'elll8ticite en coordonnees quelconques," Congre. international de mathematique, Toronto, 1924, A nnales de physique, vol. 3 (1925),
pp.251-298.
T. N. Blinchikov: "Differential Equations of the Equilibrium of the Theory of
Elasticity in the Curvilinear Coordinate System," Applied Mathematics and
Mechanics, New Series, vol. 2 (1939), pp. 407-413 (in RU88ian with an English
summary).
E. Volterra: "Questioni di elasticita vincolata I. Componenti di deformazione e
potenziale elastico in coordinate qualsivogliano," Atli della reale accademia
nazionale dei Lincei, ser. 6, vol. 20 (1934), pp. 424-428.
H. Thirring: "On the Tensor Analytical Representation of the Theory of Elasticity,"
PhY8ikali8che Zeilachrifl, vol. 26 (1925), pp. 518-522.
PROBLEMS
1. Show that the metric coefficients g<; can be calculated by observing that
3
(dx.)'
dXI; aXI;
,...--,...-va,Uaj
da, da j
i,j-l
It follows that
3
(dx.)' =
k,i.i-l
or
3
ds" =
i.i ....
gij
da. daj,
where
!5i
r, a, .. 6
and
3
'\'
g'j -
ax.~.
i-I
186
VOu
-va;; do,
+ 0 .. d,,~
va;;a;.
e,g =
!(~
2 ar
- ,:),
T
T,..,
= Tee =
To
T ..
= 0,
T,9
= J.I
(~
aT -
,:),
- .. ~.
'-"oz
T9
or!
+ !r ~
_ .':. + aiv =
or r' oz'
187
of = r 8 !_ (~),
or
OZ r
so that
iJ (v)
1 of
- aT;: =;:a
oz .
and
F(r, z).
~ (~)
= ?!:,
az
r
az
= r!_ (~) = ?!: _ ~,
!_ of = r
r2
ar
_ !_ aF
r2 az
ar r
ar
and comparison of these expressions with the last two of the formulas
(49.2) shows that the nonvanishing components of stress are given in
terms of the function F by the formulas
J.l of
J.l iJF
(49.5)
T,6 = -;:a OZ .
T60 = ;:a ar'
Since the lateral surface of the shaft is free from external loads, it follows that the resultant shearing stress must be directed along the tangent
to the boundary of the aXIal section. Accordingly, the component of the
resultant stress in the direction" normal to this boundary must vanish,
and we have the boundary condition l
Tu.
cos (z, v)
+ T"
dr
cos (r, v) =
dz
o.
.
IS
along the boundary of the axial section (Fig. 39), and we have
dr
-Th ds
dz
+ Tr&dB
=
on the boundary.
iJF~
iJr ds
+ iJF~ =
iJz dB
0
'
188
or
dF
ds = 0
on the boundary.
Thus, the condition that the lateral surface be free from extemalloads
demands that the function F(r, z) assume a constant value on the boundary of the axial section, determined by (J = const.
r
FIG. 39
(2- (a
Jo Jo
(a
Jo
rOre. dr = 2...~
(a aF
Jo ar dr
The solution of Eq. (49.4) is quite simple for the case of a conical shaft,
shown in Fig. 40. It is easily checked that the function
(49.7)
F(r, z)
= C {era
; z2)fi -
~ [(r>; zf']}
z')f'
The value of the constant c can be determined from (49.6) when the
twisting couple in the terminal section is known.
Indeoo, from (49.6),
AI
%),
since cos a
= z/(r + Z2)1i,
2
189
and hence
M
= 211"1'(cos
__ a
~
-------.~--~--~~
- ~~ cos 8 a -
%)
We can also show that the maximum value of the shearing stress r. occurs
at the narrow end of the shaft. For the principal stresses T, are,
T3
and
(T')~1
!~~h
T.)
(T~.
= 0,
+ T;O)~ = T,.
Reference to formulas for Teo and TrO shows that '"' has the maximum
value at the narrow end.
Since
1
1 ilv
eo, = 21' To, = 2
az'
ere
_!_Tre
2Jl.
= 2!(~
aT
- ~),
r
M
v
= 67rJl.k
(z'
+ r')% + wr,
190
(T:.
191
(50.1)
F(r, z)
>4ar' +
L A.r r
.-1
2
'J.(k.r) ,
where the prime denotes the derivative with respect to the argument k.r.
But'
(50.3)
where .f,Ch-r) stands for the Bessel function of order I, which is known to
slltisfy the equation
(50A)
d2
( dr'
1d + k,
+ Tar
-,21)
J,(kr) = O.
Substituting (50.3) in (50.2) and setting z = 0 gives the expression for the
distribution of stresses To. over the end z = 0 of the cylinder,
(50.5)
h.).-o '"
IJ. [
ar
L Ankn.l l(k.r) ].
.-1
192
dr
..
l
A.k.
10 r J 1(k"r) dr l
2
.. -1
10r2J (k r)dr =
1
O.
J1(kr) = - - 2 k dr'
so that
+ frJ1(kr),
and if the numbers k. are the roots of the equation J.(ka) = 0, then
J{(k.a) - (1/k n a)J ,(k.a) = O. Thus, the integrals in (50.6) involving
Bessel's functions vanish, and we get
1I"a 4
M=21'<"
which is the same expression for the moment M as previously obtained in
Sec. 33.
Since a suitably restricted function of r defined in the interval (0, a) can
..
we can obtain the solution of the torsion problem that corresponds to distribution of stress (r,:)o over the end z = 0, where (rez)o is a prescribed
function of r. It is obvious from (50.2) that the effect of the terms
involving the factors e-k diminishes with an increase in z, and hence the
distribution of stresses in a long cylinder, over the far end, is sensibly
equal to a couple of moment M.
1 See Hankel-SchliUIi expansion on p. 577 or G. N. Watson's Bessel's Functions,
2d ed. (1948).
193
The results of this section are essentially due to Dougall. They have
been extended by Synge, who proposed a more general approach to the
Saint--Venant torsion and flexure problems.!
Instead of prescribing the distribution of stress over one end of the
cylinder, one may impose a requirement that one of the sections of the
twisted cylinder remain plane. 2
51. Torsion of Anisotropic Beams. We saw that the deformation of
long isotropic cylinders twisted by end couples is the same in every cross
section. The corresponding deformation of anisotropic cylinders is more
complicated. The anisotropy of the medium ordinarily gives rise to
bending moments which deform the planes containing the axis of the
cylinder. If, however, the medium is such that the planes normal to the
axis of the cylinder coincide with the planes of elastic symmetry, then
the twisting couples produce no bending. This fact was first established
by Voigt3 and, for the special case of an orthotropic medium, by SaintVenant. As a consequence of this, the torsion problem for such cylinders
can be reduced to the solution of the torsion problem for certain isotropic
cylinders. We proceed to show how this is done when the material is
orthotropic. The corresponding solution for the case when the medium
1 J. Dougall, Tra1l8acti<ms of the Royal Sociely of Ed;nburgh, vol. 49 (1913), pp
895-978.
J. L. Synge, Quarterly of Applied Mathematics, vol. 2 (1945), pp. 307-317.
I A solution by energy methods of such a torsion problem for a beam of elliptical
section was found by A. Foppl (1920) and is given in A. and L. Foppl, Drang and
Zwang, vol. 2, Sec. 77. The beam of rectangular section was considered by S. Timoshenko, Proceedings of Ihe London M atkematical Sociely, vol. 20 (1922), p. 389 and by
J. Nowinski, Arch. Mech. SIo8., vol. 5 (1953), pp. 47-66 (in Polish). B. P. Netrebko,
V eatnik, Moscow University, No. 12 (1954), pp. 15-26 (in Russian), used energy methods to investigate the torsion of a rectangular parallelepiped by arbitrarily specified
distributions of shearing stresses on the bases. Energy methods are also used by
N. V. Zvolinskij in "Angenaherte Losung der Torsionsaufgabe ffir einen elastischen
zylindrischen Stab mit einem nicht verwolbten Querschnitt," Bulletin de l'academie
des sciences de I' U RSS, elMse des sciences mathhnatiques et ...zturelle8, No.8 (1939),
pp. 91-100 (in Russian',. The problem of flexure of such a beam has been treated by
R. Sonntag in "tlber Biegung bei verhinderter Querschnittskrummung," Ingenieur
.1rchiv, vol. 4 (1944), pp. 415-420.
The effect of local stresses corresponding to different modes of applying torsional
couples to a circular cylinder has been discussed by Wolf and by Deimel IK. Wolf,
Sitztlung8berichte der A,kademie der Wi8sen8c/w.ften in Wien, vol. 125 (1916), p. 1149;
R. F. Deimel, "The Torsion of a Circular Cylinder," Proceedings of the National
Academy of Sciences afthe United StaIRs of America, vol. 21 (1935), pp. 637-642\.
W. Voigt, I,ehrbuch der Kristallphysik, p. 648. In Chap. VII, Sece. 315-324, of
this work Voigt discusses the torsion problem for cylinders with the most general kind
of anisotropy. See also S. G. Lekhnitsky, Theory of Elasticity of an Anisotropic Body
(1950), pp. 141-172 (in Russian) .
B. Saint-Venaut, Memoirea presemes par divers IOIIants a l'acadbnie des ~,
Scienr.etl mathlmtatiques el phy.iqltU, vol. 14 (1856).
194
where
(51.2)
e.'.
(au + aw)
ax' ,
2" az
u = -azy',
= azx',
= a",(x',
y'),
where a is the angle of twist per unit length of the bar and ",(x', 1/') is the
torsion function associated with this problem.
H we let 9 denote the angle between the axes x and x', then the expressions for the nonvanishing components of stress T .. ami T y , are related to
the components T.', and Ty', hy the formulas'
T ..
Tve
= Ty'.
195
fT..
(51.4)
= all.
sin 8
T = all,
cos 8
U; +
U; +
Xl)
+ all, cos 8
U; - y'),
sin 8
UJ, - y').
XI) -
alll
The partial derivatives of the torsion function l(J(x ' , y/) appearing in the
right-hand. members of these expressions can be calculated in terms of
OI(J
oX and
0'1'.
ay'
SInce
X =
We have
r OI(J
i OX'
I
(51.5)
OI(J
= -- ('OR
oX
; 0'1' =
oy'
e - OI(J
-- SIn
(J
oy'
'oX
!! sin (J + oy
~'" ('os iI.
Inserting t,he "alues from (51.5) in (,'i1.4) and introducing the abbreviations
A = Il' Hin' (J + III COS' (J,
B = (Il' - Ji') sin (J cos e,
C = Ji2 cos' (J + Jil sin' (J,
we get
T"
(51.6)
I
a(A
OI(J
0'1'
ax
+ B iii;
+ Bx -
0'1'
y, = a ( B -,;vX
0'1',
+ C!l
+ ex vy
Ay ,
By) .
Hin('e T" = Tyy = T" = Tzy = 0 and Tn and T., are independent of z,
the first two of thp equilibrium equations (15.:l) arp identif'ally satisfi('ri
and the third one gives the equation
0 2 '1'
(51. 7)
A !12
vX
(J2rp
(J'I(J
+ 2B vI
~. + C ,..
vy
vir
O.
Thus, in this case, the torsion function 'I' no longer satisfies Laplace's
equation.
Let the boundary C of the cross section have the equationf(x, y) = 0;
then the components cos (x, v) and cos (y, v) of the normal v to the
boundary C are proportional to :~ and
cos (x, p)
p) = 0
on ('
196
in the form
(51.8)
+ (BiJ'"
+ C aay",) Of
( A~+BiJ",)af
ax
iJy iJx
iJx
By
=
(Ay - Bx) af
ax
+ (By
_ Cx) af
iJy
onC.
(51.9)
., = y -
:Ax.
(51.11)
1/)
o.
then
(51.13)
?",' aF + acp' aF
il~ il~
il., il.,
aF _
=
'1
a~
~iJF
a"
on C/,
or
(51.14)
onC',
197
ff
(r,.x - rz.Y) dx dy
A(t'
+ "')] dt d",
where the integration now extends over the region R' bounded by the
curve C' . Recalling that ",' "" (VAAl A) "', we have
M --
aA' {r (U",I
V~1~2 {J t u" -
7/
u",'
at + ~
+ .) d
7/
d
7/,
and since
M=aD,
where D is the torsional rigidity, we see that the torsional rigidity of a
nonisotropic cylinder can be deduced from the torsional rigidity of the
isotropic cylinder obtained from the nonisotropic one by a homogeneous
deformation (51.9).
We conclude this formulation of the torsion problem for a ncmisotropic
prism by remarking that the transformation (51.9) changes the boundary
of an ellipse
x'
y'
(i2+p=1
into another ellipse, and since the solution of the torsion problem for an
isotropic cylinder is known, we can write down at once the solution of the
corresponding problem for a nonisotropic elliptical cylinder.
The transformation (51.9), in general, carries a rectangle into a parallelogram, and hence the solution of the torsion problem for a nonisotropic
rectangular beam is not covered by the discussion contained in Sec. 38,
unless the x'-axis coincides with the x-axis. If these axes coincide, then
B = 0, and the rectangle will be transformed into another rectangle of
different length. The solution corresponding to this case is written out
in Love's Treatise, on page 325.
REFERENCES FOR COLLATERAL READING
198
is
:
i.-l-z ...
--
W
FlO. 42
T:tu:
= r zy =
71/1/
o.
The functions T", T. y , and T" will be so chosen that the equations of
equilibrium and compatihility, as well as the boundary conditions, are
satisfied.
In writing an expression for T ... we shall be guided by an expression for
the bending mOlllent M. that would be produced by the load W. acting
alone. In any cross section z units distant from the fixed end, one would
have
M. = W.(l -- z).
(52.2)
(bending by couples),
with
I" -=
x'dxdy,
199
T"
= _- E(l
+ K.y),
- z)(K"x
ff
(52 . .5)
T"
Jf T," dx
dx dy = lV"
1/
('ondition~
rl!1 = Jr.,
1/
,
I
aT.. = 0
Or
=0
az
'
az
'
~;. + a;;. + E(K.x + K.y)
(52.6)
= O.
It follows from the first two of Eqs. (.52.6) that the shear p,omponents
or" and T,. have the same value in all cross sections of the beam, while the
third equation can be rearranged to read
:x
o.
fx (~~) + :y (- ~~) = 0,
it is evident that there exists a function F(x, y) such that
_a - 2
aF
!T,. -- T.. -
(52.7)
1 EK
"x,
aF
ax
- '12 EK.y2 .
The conditions to be satisfied by the function F(x, y) can be determined from the Beltrami-Michell compatibility equations (24.15), which
reduce in this case to
V
+ 1EK.
+
T..
_ 0
<1 -
._
V,.
+ 1EK.
+
<1
Substituting from (52'.7) in these equations, we see that the latter will be
fulfilled if
V'F(x, y) = -2JJdK.x
+ 2JJdK.V -
2/-101.
200
The physical significance of the constant of integration -21'Cl will be discovered presently. It is not difficult to obtain a particular integral of
(52.8) in the form of a polynomial, and it is readily verified that the
solution of (52.8) is
'
(52.9) F(x, y) = I(x, y) - J,i"",(KvX' - K.y) - Y2"a(x' + y'),
where !(x, y) is a harmonic function.
It will prove advantageous to write the stresses T", T z" not in terms of
I(x, y) but in terms of its harmonic conjugate g(x, y), where
a!
ax
(52.10)
ag
=
ay'
ax
ag
= - ay
- "ay
+ "",K.y' - !2
EK.x'
'
+ jJ.ax + "",KvX' -
'2 EK"y'.
! (av
2 ax
_au).
ay
(Or.. Or..)
-~ax-ay'
Substituting the values of the shear stresses from (52.11), one gets
aw
i)z =
Cl
+ u(KvX -
K.,y).
The mean value of the local twist over the section (or, equally well, the
value of the local twist at the centroid of the section) is Just the constant
a. Thus, we see that the terms in (52.11) that involve a represent a
twist of the beam, and, indeed, the terms - jJ.ay and jJ.ClX in these expressions also appear in the solution of the torsion problem (see Sec. 34). In
the latter problem, one has
(pure torsion).
We are thus led to introduce the torsion function <p(x, y) into the flexure
problem by writing
(52.12)
201
where ",(x, 1/), "'l(X, 1/), and ",:(x, 1/) are hannonic functions.
We can now
write
'1'
(52.13)
{ T..
pa
pa
a-
( i l ' l 'Y)
x
(~+
oy
x)
il'l', + I'K. [ a
x
+ I'K. [a'oyl" -
cos (x, v)
<p,
a'l'
+ ,.K. -2,
ax
+ I'K'ay
0'1',.
v) = 0,
(52.15)
d<P1
a; + K. d<p,
dv =
['
K. (1
v)
on C,
and this will be satisfied if the functions <P1 and <p, are subject to the
conditions
(52.16)
7,,'
d;:.
= [(1
[(1
+ <1)x' + <1)y2 -
on C,
on C.
The flexure problem has thus been reduced to the task of finding three
functions, hannonic within the region R of the cross section, whose normal
derivatives are prescribed on the boundary C; that is, we have been led
to the problem of Neumann. In order to see that the condition of the
existence of a solution of this problem is fulfilled, we observe that
Ie 7,,'
ds
Ie
[(1
+ <1)x' -
= 2(1 + <1)
11
<1Y'] dy
x dx dy
= 0,
( d<p'ds
Je dv
= _ ( [(1
Je
= 2(1 + <1)
+ <1)y' -
<1X'] dx
ydxd1/ = 0,
on C,
'" =
+ y2) + const
on C.
The torsion fUllctions 'I' and 1/1 are harmonic conjugates, so that 'I' + N is
an analytic function of x + iy. The flexure problem may also be reduced
to a problem of Dirichlet by introducing the harmonic functions 1/1" "'"
conjugate to '1', and '1'2, respectively. Then
(i
1, 2),
+ O')x'
dy
- O'Y'] di'
d""
-
dB
-[(1
+ O')Y'
dx
- O'x']-,
ds
or
1/1,
(52.17)
- ~O'y3
{"'2 = %O'x
+ (1 + 0') f
(1 + 0')
f
~0
(XO,tlD)
(%'Y)
(zo,lIo-)
+ const
x 2 dy
y'dx
+ const
on C,
on C,
w.
r dxdy,
W. = pOt
rr 0'1',
JJrr orp
oX dxdy + pK. JJ OX dxdy
+ p.K. ~:2 dx dy
+ p.K.[ - (1 + 0')1, + 0'1.],
whers
I. ""
y'dxdy,
1. =
x'dxdy.
203
With the aid of this identity and t.he houndar., ('onditions (52. I!;) lind
(.";2.16), Eq. (52.18) becomes
IV.
+ tT),
W.
E(K.I.
where
I.. =
+ K.I ),
JJ xy dx dy
R
Simifarfy, from
[J T,. dx dy,
it follows that
W.
(52.20)
E(K.I.
+ K.I.,,).
EK = I.W. - I."W.
(52.21)
1.1. - I;'
For
f/:
dXdy
JJ[fx(x~n +~(X~)]dXdY
- dy = Jdofl
X-d.
J( -x -ilofl dx + x ilofI)
ilx
dv
.
c
iJy
204
Jf yrudxdy =
M. ""
-(l - z)W..
-XT"
dx dy = (l - z)W.,
which are precisely the bending moments produced in the section z = const
hy the forces W. and W .
The stress distribution over any cross section is easily seen to be
statically equivalent to the load (W., W., 0);
f!r dxdy=W.,
R
o.
The first two equations are satisfied by virtue of our choice of the constants K. and K., while the third follows from our assumption that the
z-axis passes through centroids of cross sections. The constant a in
formulas (52.13), for shear stresses, is determined by the condition that
the twisting moment M. be such that
(52.22)
(xr - yr.. ) dx dy
xoW. - yoW.
In (52.22), (xo, Yo) are the coordinates of the load point relative to any set
of axes intersecting at the centroid of the section.
We see that the solution of the general Saint-Venant flexure problem
is reduced to the determination of harmonic functions 'P, 'P" and 'P2 that
satisfy the boundary conditions (52.15) and (52.16). The boundary
conditions (52.16) are somewhat unwieldy, and we shall show in Sec. 53
how the formulation of the problem can be simplified by introducing the
idea of center of flexure.'
63. Center of Flexure. The formulas (52.13) for shear stresses suggest
a resolution of the general flexure problem into the following simpler
problems:
I The treatment of the flexure problem given here is influenced by L. S. Leibenson,
Central Aero-hydrodynamical Institute Technical Note. 45, Moscow (1933), and A. C.
Stevenson, Philosophical Tramactions of tke Royal Society (London) (A), vol. 237
(1938-1939), pp. 161-229. These authors have departed from. Saint-Vena.nt's
formulation by supposing that the load acts not at the centroid of the end section
but at an arbitrary load point (xo, Yo, I). Also, they abandoned Saint-Venant's
assumption that the x- and y-axes are the principal axes of inertia of the eross section.
Freedom in the choice of axes is of importance for asymmetric cross sections because,
(or such sections, the principal axes seldom provide the most convenient mathematical
description of the boundary. Leibenson obtained formulas, equivalent to those given
here, by a transformation of coordinates in the classical Saint-Venant's solution.
205
(XT.,. -
vr...) dx dy
= !W. - tiW.,
which must hold for an arbitrary choice of W. and W.. The load point
(x, ti, 1), corresponding to a = 0 is called the center of flexure and is
denoted by (x."
l).
2. A torsion problem with the twist a due to a couple of moment
y."
W.(xo -
XcI) -
W.(yo - y</),
and with shear stresses determined by (52.13), with K. and K. set equal
to zero.
We can thus think of the load W at the point (xo, Yo, l) as being replaced
by an equal load at the center of flexure and by a couple producing the
twist a. The solution of the general flexure problem is then got by superposing the solutions of these two simpler problems. The decomposition
of the general flexure problem into problems 1 and 2 amounts to resolving
the twisting moment
M.
xoW. - yoW.,
ff (xr,. -
(53.1)
V(xr.. -
vr.. ) dx dy
= J(I,/J, - 1..8,),
Yc/ = J(I,.8, - 1.8,),
Xc,
{
(53.3)
where
8, =
If 7v' - +
ff [x 7v' - a;; [x
try.] dx dy,
y 0:X'
(1
+ l1)x'y -
(1
Ii
8, =
Ii
J-' = 2(1
+ 11)(1.1. -
I!,,).
206
(53.4)
~ x., =
2(1
~ rr)I.
(y., = 0,
J-l
0,
and
[f
= 2(1 + rr)I.Iv,
0;,' - YO:: -
[x
(1
which state that the center of flexure lies on the axis of 8jJmmetry of the section.
Accordingly, if the cr088 8ection has two perpendicular axe8 of symmetry,
then the center of flexure ccn'ncides with the centroid of the section.
In general, the center of flexure does not lie on either of the principal
axes and may even be outside the cross section of the beam. 1
The solution of the simple flexure problem is given by the harmonic
functions and 'P., which satisfy the conditioWl (52.16) on the boundary.
Simpler boundary conditions can be realized by subdividing the problem
once more. We define the harmonic functions "'u, "'12, 'Pn, "' by the
relations
"'1
(1
'PI =
(53.5)
"'. = (1
+ 0')"'11 - rr",u,
+ 0')"'21 + 0''''21.
{ (1
(1
""I
- a
1 There is some confusion in the literature concerning the relation of the flexural
center to the cenUr 01 twiat, the latter being defined as the point at rest in every Cl'088
aection of the beam fixed at one end and twisted at the other by a oouple. The center
of flexure is sometimes vaguely defined as the point in the end section of a cantilever
beam such that the load applied at that point produces "tomionless bending," There
are different definitions of toreionless bending [E. Treffu, ZeitBckri/t 11l.r ~
MatMm4tiJ: und Meckanilc, vol. 15 (1935), pp. 226-225; J. N. Goodier, JuumqJ, oj tile
A_utiml &isntJe, vol. 11 (1944), pp. 272-280), and the confusion in the identification of the two oenters generally stems from the .failure to define toreionlese hendina
and to specify the mode of fixing the beam. It is poIIBl"ble to define the center of
flexure (aleo oaIled the cenUr oj 41h<1ar) and the center of twist 80 that both oenten
coincide. See A. Weinstein, ~ oj Applied M~, voL I; (1947), pp.
97-llII. The _ten of flexure for JeVeI'IIl beams with polygonlll _
.IOOtiona have
been ealeuIated by N. Kh. Arutiunyan and N. O. Gulkanyan, PriIcl. Mat. Me1:1.;
A~
207
We are at liberty to prescribe arbitrary boundary conditions on the individual functions <Pi;, subject only to the restriction that the relations
(53.6) be sa.tisfied on C. Boundary conditions that are simple in form
and independent of the elastic constants of the material will be realized
if it is required that the functions '1'.; satisfy conditions
(! XI),
d'l'22
) = iE.
d (31
Tv = y' cos (
y,"
y') ,
(53.7)
d'l'lt
Tv '=
d'l'21 =
dv
y2
(13 y'),
= .!!_ (! x,).
'ds 3
)
d
cos (
X, V = dB
-x' cos
(y ,,)
ax
0'1'21
ax
0"'12
0'1'12
0"'21
0'1'21
= -
ax'
Ty = -
ax'
Ty'
-oy ,
0"'12
Ty
Ot/t21
and in terms of these functioI " the last two boundary conditions can be
written as
or
on C.
"'21 = ~X' + const
(53.8)
"'12 = ~yl + const,
The solution of the simple flexure problem in which the applied load
(W., W.,O) acts at the center of flexure (with a = 0) is thus given by the
stresses
'Tn
= -Tzw =
T ..
= -E(l - z)(K.x
T ..
T.. =
1'fIII
0,
~K. [ (1 + u)
+ K.y),
e:;1 - x.) -
(at;, - y.) ]
+ ~K [(1 +
where
EK
I.W. _ I..,W.
I.I. _ I:_ ,
u) 0'1'22
+ IT ~.!],
+ o;;l}
ax
oy
IT
IT
208
and where the harmonic functions 11'11, 11'22, -/tu, -/t21 satisfy the boundary
conditions
c;;'
d'Tv
l'tJ =
+ const,
"'" = %,x'
"'" = %,y.
+ const
on C.
The connection between the functions employed in this seotion and the
classical Saint-Venant flexure function is discussed in the next section.
M. Bending by a Load along a Principal Axis. The general problem
considered in the last two sections will now be specialized to an important
partiCUlar case, namely, that in which the axes are taken to be the
principal axes of the section and the load (W., 0, 0) is directed parallel
to one of these axes. In this case, I ... = 0, and Eqs. (52.21) yield
K. = W.
E1.
21'(1
W.
+ q)1.
K. = 0,
_ I'a (a'l')
l" ax - y + 2(1 W.
+ q)1. [a'ax
" -
Tar =
y.a
0'1')
ay
(1
+ q)x I+ qy,
2]
+ 2(1 W.
+ q)1. 0'1"
ay
The flexure function 'I"(x, y) is not of the same form as the classical
Saint-Venant flexure function <I>(x, y) used by most writers; the two functions (together with their harmonic conjugates 1/1, and if) are related, in
fact, by the e"pression
'P,
+ i"',
'I', =
(54.1)
= - (<I>
or
"', =
In terms of the harmonic function <I>(x, y), the stresses can be written as
{ ..,.~ =
T'Q
,," = -
(54.2)
,"" =
Till
TIN
!f
0,
(l - z)x,
(!; + x) -
2(1 !"fT)I" [ :
+ (2 + q)xy
From (54.1) and from the boundary condition (52.16) on the function 11'1,
it follows that the harmonic function <I> must satisfy the condition
209
(54.3) : : '" -
,,)
on C.
This special case of the general flexure problem has been formulated in
terms of the torsion function 'P and the flexure function if> as a problem of
Neumann. It may be rephrased as a problem of Dirichlet by writing the
stresses in terms of the conjugate harmonic functions '" and w. The
appropriate boundary condition on w is seen from (52.17) and (54.1) to be
(54.4)
w = -U(l -
721T)Y'
+ (1 + 72u)X2y -
(1
+ 2(1 + u)
+ IT)
r(~,v) x'dy
} (Zo.f't>
+ const
r(~'v) xydx
} (:1:0,110)
+ const
on C,
where the last step makes use of integration by parts, and where the line
integral is to be evaluated along the contour C.
The y-coordinate of the center of flexure is found from (53.3) and
(54.1) to be given by
(54.5)
1101 = 2(1
~ ,,)1
"R
+2(1
- 2(1
~ ,,)1.
- 2(1
~ ,,)1.
II
[(1
210
-ayz
r
(55.1)
+ Kc{72a(l -
+ +
+
+
The linear terms that arise in deriving Eqs. (55.1) represent a rigid body
displacement and can be made to vanish by imposing suitable llonditions
of fixity.
When the flexure problem is specialized to the case of bending by a load
(W., 0, 0) along a principal axis (Sec. 54), then Eqs. (55.1) take the form
+ ;'j. [~er(l -
u = -ayz
(55.2)
v =
ctXZ
+ ~: er(l -
w = a'P(x, y) -
z)(:l: 2
~ za + ~ lz,}
y2) -
z)xy,
;;J~(x,
y)
+ XV' +
(lz - ~ zt)
x].
~u
"x
~:
= O.
These give
b = c = O.
that is, the deformed central line of the beam lies in the plane of bending
K.
y=-x=
(55.4)
I.WlI - I ..W. x
K.
I.W. - I ..W.
The greatest deflection of the central line of the beam occurs at the loaded
end z == 1, where
'
1 KJ'
1 I.W. - loW. fa
U ..
1K
II
1 I.,W. - loW.
=- 3 ..... i E(I.I. - 1:'> II .
11
211
while, for bending by a load W. along a principal axis, the end deflection is
1 =
W' ll
3 E1v
The plane of the load (the plane containing the z-axis and the line in the
direction of the load) does not, in general, coincide with the plane of
bending, since the equation of the former is
y
Wv
= W. x.
The neutral plane is defined as that plane whose filaments are not altered
in length; that is, it is characterized by the equation en = O. Since
en =
(lw
(lz
z)y,
K.
(55.5)
y = - Kv x.
The planes defined by (55.4) and (55.5) are orthogonal, and hence the
neutral plane is perpendicular to the plane of bending.
In the case of bending by a load (W., 0, 0) along a principal axis (Sec.
54), the xz-plane contains the deformed central line, while the :!Iz-plane is
the neutral plane.
Consider now the curvature of the deformed central line of the beam.
Taking coordinates r = y' X/2 + y'2 and z in the plane of bending, we
have from (55.3)
r
= y'K;
+ K; (-%z + Y2lz2).
If the displacements and their derivatives are smail, one can write
r = y'K:
+ K: (-%Z'l + Y2lz'J),
from which it follows that the curvature of the central line is given
approximately by
it .. :; . . viE: + K: (I -
Z/).
That the curvature is proportional to the bending moments M., Mil ill
easily seen by referring to Sec. 52, where it was found that
212
M. =
g-
and hence
M
w.
- R v'K~
+ K:
For the case of bending by a load W. along a principal axis (Sec. 54),
t.hese tela.tiotla Uee.om.e
M. = EI.
R
M. = 0,
z' = c
+ w = c + dlp(X, y)
- (Ie - ~C2)X - 713(2 + cr)x' + ~crxy2)
+ K.[Ip.(x, y) - (lc - ~C2)y - 713(2 + cr)y3 + ~crx2y).
+ K.[Ip,(x, y)
For the special case consideted in Sec. 54, this takes the form
(55.7)
z' =
+ (lip (x, y)
- ;;.: [
xl
dy =
Ik
T"",
T
{ K.
+ K. ~I} "
[= - :t' -
,,(:t' - II') ]
+ K. =} d1J -
O.
213
For the special ease of bending by a load ~ong a principal axis, this
becomes
[2: + O'x' +- (2 -
0')y2] dy =
O.
This equation will be used to dete~mine the distribution of lines of shearing stress in a bent circular beam.
66. Flexure of Circular and Elliptical Beams. Let the equat.ion of the
boundary of cross section of a beam of length I be
x'
+ y2
= as,
and let the terminal load W be applied at the centroid of the end section
and directed along the x-axis. The form of the boundary suggests the
use of polar coordinates (r, (J). In terms of these coordinates, the equation of the boundary assumes the simple form r = a, and the boundary
condition (54.4) becomes
'If
= -711(1 - 2(1
or
'If = -(%
72O')a l sin l
(J -
72O'a 3 cos'
(J
sin
(J
+ 72O')a
sin
(J
+ ;!,ia
sin 311,
on r
a.
Since the function 'If is harmonic in the interior of the circle r = a, the
appropriate particular solutions of the equation V2'1f = 0 in polar coordinates are of the form r" sin nil. Hence we must have
'If =
(%
4>(x, y) = -
(J,
(% + 72O')a'x +
Y.t(x - 3xy).
From the symmetry of the cross section, it is seen that the center of
flexure coincides with the centroid of the end section, and as the load
point has also been taken at the centroid, it follows that in this example
a = O. The stress components are found from (54.2) to be
_ (3
+ 2u)W (
+ 0') a
(1 + 2u)W
2 _
r - 2ra'(1
(56.2)
1'
= - ra'(l
+ 0') xy,
4W
rp = - ra' (I - z);t.
1 - 20' .)
+ 2u Y ,
214
FIG. 43
T."
_ (3 + 2o-)W ( t
2ra4(1 + cr) a
= 0,
T.., -
2)
1 - 203 + 2cr Y ,
T.%
3+20- W
+ <1) ra'
1+2<1W
=- --.
1 + <1 ra'
iiX
T ..
T.;
+ 2cr)xy dx
- (3
dy .. 0,
215
+ y. =
(f
a2
3+20-
+ cyl+20-,
= 0.3.
z' - c = - - E [- (3
Tat
+ 2rr)a + 2c(21 2
c)Jx - - - (x'
Ta'E
+ y)x.
- 7ra'E (x 2
+ y2)X =
const.
"
Flo. 44
f(x, y) "'"
x'
y'
iii + l)i
- 1=
o.
2Hi
We assume, as above, that the load acts in the direction of the x-axis and
is applied at the centroid of the end section.
Now the direction cosines cos (x, v) a.nd cos (y, v) of the normal to the
boundary of the ellipse are proportiona.l to
:~
and
:t,
respectively.
Hence
cos (x, v)
x/a'
cos(y;-;j = y/b 2
On the boundary C of the section, the flexure function 4> satisfies the condition (54.3); hence
(565)
.
a4'b x+ a4'
ax
ay a y=
2
2
2X-(2+u)a 2xy '
-[!uX
+(I-!u)y2]b
2
2
onC.
+ iit
= el(x
that is,
(56.6)
and substituting in the boundary condition (56.5) gives
[el
+ 3c.(x'
+ (1
- Yzu)y'lb 2
- (2
+ u)a 2y2.
-a'
Cl
C2
(2
+ Yzu)a + (1 9a" + 3b
2
Vzu)b 2
+ ..
+ .. +
'.l_.2)
"""If
217
It is obvious from these formulas that the y-component, T , of the shearing stress vanishes on the hotizontal axis (x = 0) of the cross section and
that
_ 2W 2(1 + <,:~~_j-_b~_ [ _
(1 - 2o-)a'
(T )_O - ;(i3b (1+oT(3a' + bt ) a
2(1 + u)a' + b' y .
t]
Hence
2W 2(1
+ a)a' + b'
if 0+ a)(3a 2 + btl'
and A = 'll'ab is the area of the cross section. Evidently Tc. reaches its
maximum on the boundary, ~nd if we put x = a cos (J, y = b sin (J, then
(XY)mn = (7fab sin 2(J)_. = 7fab, and it is seen that.
h.)m&>
2W b (1 + a)a' + ab'
a(1 + u)(3a' + b')'
=A
If b a, then the shape of the beam approaches that of a thin rectangular plank loaded parallel to its longer aide. In this case, neglecting
terms of order b'ja', we get
(
)
'T zx
InIlloX
.4W
= 3A'
. 4W b
("Z,",) max = aA 2ll'
b a,
a b.
57. Bending of Rectangu16I' Beams. The problems in the preceding
section illustrate the solution of the boundary-value problems of flexure
by forming those combinations of particular solutions of thedifferential
equation
or
V'w = 0
V~ = 0
that satisfy the boundary conditions on the function cI> or '1'. The
flexure Qroblem of circular beams was treated by inspection of the
boundary values of 'I' in polat coordinates and by utilizing the particular
solutions of the form rn sin nO. Beams of elliptical cross section were
handled by observing that tlle boundary condition on
:~ involved only
homogeneous polynomials in x and y, and this fact suggested that a solution for the complex flexure fllnction cI>
iw be sought as a sum of terms
of the form c.. (x + iy)". III this section, the solution of the flexure
problem for a beam of rectaJlgUlar cross section is given as an infinite
aeries of particular solutions A .. sinh a% cos fJy, the coefficients A .. being
SO chosen as to ensure the satisfaction of the boundary conditions. The
218
next two sectiollll will illustrate the use of analytic functions in solving
the flexure problems.
Let ilIe equation of the boundary of the cross section of the beam be
(x' - a2)(y2 - b' ) = 0,
and let the terminalloa.d be directed along the positive x-axis and applied
at the origin.
A reference to the boundary conditions (54.3) shows that, on the sides
x = a, we must have
a4> =
ax
_.! ua'
2
(1 _.!2
u) yS
'
-b
<
< b,
ay = +(2
+ u)bx,
< x < a.
-a
In order to simplify the boundary conditions, we define a harmonic function !(x, y) by the relation
7il(2 + u)(x'
!(x, y) = 4>(x, y) -
- 3xy');
+ u)a + cry',
on x = a,
a! = 0 ,
ay
on y = b.
It follows from the discussion in Sec. 38 that one can build up the
desired solution by forming an infinite series of particular solutions
!(x, y) = Ax
'\'
n1lx
nry
..
"\'
-b
b.
osh~
n1r'g
+ ...L..1 ~A
b "c
b cos b
== -(1
+ u)a
i
+u [ -b
(-1)" cosnry],
+ -4bt 2:" -_
3,..2
nS
b
!
219
and equating the coefficients o.f cos (nry/b) leads to. the result
f( z y) -
sinh n...z
'\' (-1)-
n-l
n'
b cos!!!1lb
ftra
coshT
The flexure functio.n <I>(x, y) can now be fo.und from the relatio.n
= f(x, y)
+ ~(2 + (f)(x
_,3 xy2).
We shall dispense with the calculatiQn Qf shear stresses. An elabQrate
discussio.n o.f the bending pro.blem fo.r a rectangular beam is given by
TimoshenkQ,' who., ho.wever, appro.aches the pro.blem in an entirely different way, by using an analo.gy between a certain stress functio.n and
the deflectio.n o.f a stretched membrane under no.nunifo.rm pressure. We
shall discuss this analo.gy in Sec. 60.
58. Conformal Mapping and the General Problem of Flexure; the
Cardioid Section. The examples co.nsidered in the preceding sections
give illustrations of the specialized pro.blem of flexure by a lo.ad (W., 0, 0)
directed along a principal axis (which was also one o.f the two. axes o.f
symmetry) o.f the cro.ss sectio.n o.f the beam. The analysis was also. simplified by taking the centro.id o.f the sectio.n as the Po.int o.f applicatio.n
Qf the lQad. We cQnsider nQW, as an illustratiQn Qf the general prQblem Qf
flexure, the prQblem Qf bending Qf a beam whQse crQSS sectiQn is bQunded
by a cardiQid and thus has Qnly Qne axis Qf symmetry. The lQad (W.,
W.,O) will be co.nsidered to. act at so.me Po.int (xo, Yo, 1) Qf the end sectiQn,
and the Qrigin Qf cQo.rdinates will be taken in the fixed end at the centrQid
of the cardiQid. UpQn this pro.blem we shall bring to. bear the PQwerful
weapQn o.f analytic functio.n theQry, which was used earlier in the case Qf
to.rsio.n o.f a beam. 2
In Sec. 53, it was seen that the general prQblem o.f flexure by a lo.ad
(W., W., 0) acting at any po.int (xo, Yo, l) can be reso.lved into. (1) a
simpler flexure pro.blem with IX, the mean lo.cal twist, set equal to. zero.,
and with the lQad applied at the center Qf flexure (xc" Ye/, l), and (2)
a to.rsiQn pro.blem with a twist IX due to. a co.uple o.f mQment W.(xo - xc,)
- W.(Yo - Yell. The pro.blem o.f to.rsio.n o.f a cylinder was reduced in
Sec. 35 to the bo.undary-value pro.blem o.f finding the analytic functiQn
,,(x, y) + il{t(x, y) with
I{t = 72(X' + y2)
o.n the bo.undary C.
4>(x, y)
220
d"pu
d--
'/1.1
= ~X3
~2
+ const,
on C.
~yl + const
The boundary conditions 011 the normal derivatives of 'll and '22 can
he replaced by conditions on the houndary values of the I'olljugate function!! .pll and .pt! by noting that
1.'!..lJ
dv
O.!~l
ax
<!.: + o'!._lI ~Y
oy
dv
dv
i1rh_1 <!JL
oy
d8
+ 01/111 ~
= d'/ttL
OX d8
ds
on C,
or
.pll
=/
x2
~: dB =
x' dy
on C.
Similarly, we have
"'.2 =
- /
y' dx
on C.
in R,
onC,
by mapping the region R on the interior of the unit circle ItI :<; 1 ami
applying the formula of Schwau [Eq. (42.4)]. The same procedure can
be used, of course, to write down the solution, in the form of an integral,
for any problem of Dirichlet for any region that can be mapped conformally on the interior of a unit circle. In particular, the Schwarz integral affords solutions for the boundary-value problems of flexure.
It is not difficult to express the boundary conditions imposed on the
functions r/tu in Sec. 53 in terms of the complex variables a = x + iy and
! == x - iy or, when the mapping function a = "'(t) is known, in terms of
t and f. The application of the Schwarz formula (42.6) would then yield
the complex flexure functions 'Pij + ir/t.. in a form analogous to (44.5).
Because of the complicated form of the boundary conditions which the
!/Iii satisfy, the resulting general formulas are of doubtful value in specific
221
5c
"3
+ r cos t,
11 = r sin t,
x =
{y
5e
= 2c
cos t) =
t) = 2c
5e
"3
- e + 2e cos t - c cos 21,
sin
t - ('
sin 21.
Then
x
5c
+ ~y. = "3
-
(
c 1 - 2e"
c
+ e'" ) 5
="3
-
c(l - eit)',
with
w(l)
5c
== "3 -
c(l - I)',
5c + .,
= x + ty. =:3'
re' ,
maps the interior of the eardioid on the unit circle 151 .:::; 1, with () = t.
The inverse transformation is
(58.3)
0':::;
t.:::;
2".
1 See S. Gosh, Bulletin ()f the Calcutta Mathematical Society, vol. 39 (1947), pp. 1-14.
It should be noted, however, that a suitll.ble choice of curvilinear coordinates may
yield simpler solutions. Thus the ftexure problem for a cylinder whose cross section
is formed by the arcs of two intersecting circles of different radii, which includes the
segment of the circle as a special case, is more easi,,- .olved in bipolar coordinates.
See Ya. S. Uflyand, D()klady Akademii Nauk SSSR, N~lv Series, vol. 69 (1949), pp.
'751-754.
222
5c
'3
-
and we have
(1)
w(O')w -;
where
ll(O')
-6 - 60'
3 (2 + &
- 30'%).
90" 1,(0'),
+ 490"
- 6u a - &4.
<p
c
+ t'." = 1871'
1,(0')
0'2(0' _ r) dO'
ic'
=9
(R ,
+ R.),
where R, and R. are the residues of the integrand at 0' = 0 and 0'
respectively, and where 'Y denotes the contour Irl = 1. We have
= r,
= ~ + ~,
dtrO'-r ...O!
!'
I,(t)
6
6
R. = -= - - - - + 49 - 6" - 6'"
"
!'
1"
= 9ie' (49
- 6! - 6!')
and
'P
+ i1{l
. , 37 _ lSi ~
= u;
9
e
or
<p
'" =
t - :32 rc SIll
. t,
lHi
2r e cos '2
Io
_!:_ 2c'
j,(0') d
4i 27 y 0"
0',
where
1.(0') = 18 - 360' - 1770'
+ 4950"
- 2200" - 12&'
and hence
71'c"
10 = 27 Ra,
where
1 d'
Ra = 31 dtr 8 [12(0')].-0 = 495
+ 360'" + 120'7,
223
Thus,
551rc
10 = -3-
(58.4)
_! i4c
Do =
4 9
r!3(U)
du
u
,
J'Y
with
Then
Do
2".c
R, =
2".c' 1 d'
4".c'
4i du' [f3(U)]._0 = - 3 '
,..(10
+ Do)
= 17,..".c'.
The shearing stresses may be fonnd either from Eq. (44.10) or from the
relation
T zz:
JJa.
a'l'
iJy ,
"'.iJl
a'l'
+ 1,/2,\
on C.
~ (J +~}
3"c [5
c( -3
3_,!:'2
6u 2
and
with
f.(u) = -27
+ 162<1 - 21611
+ 82Bu7 2
224
.1.
_
'1'21 -
on 'Y,
.
+ uhl)
1 c3
-0 -
7rl
or
648
1
y
/.Crr) drr
----- rr 6 (rr - n
'
1"21
and
Rs
he!).
IS
Then
I"u
+ i.p21
ic
= 324 (248
+ 828r -
4411' - 54r 3
216r 4
+ 162r6 -
27r6).
3~4
(5OOC'
1 (
36
48r;2c~
t .
COB 2 - 76rc' SIn t
3t
+ 18r~'c" cos '2
- 21r'c sin 2t - 3r' sin 3t).
Similarly, the function l{t12 can be determined by noting that (58.1) and
the relations
\lost
~(rr +~).
sin t =
_!, (rr
~),
2~
rr
yield
ie 3
1/' = 80-' (1 - &
+ 12rr' -
2rr' - 27rr4
+ 3&5 -
+ 27rr 8 + 2rr'
12rrlO + &11 _
3&7
-
rrn)
225
"'12 is
"'12 = !3 ic8 li(a)
as
on 'Y.
+ 11/;"
'1'12
ie'
li(fT)
where
R7 = ~1 5do
;). dfT
1.(fT)
----- ]
,
.fT - ! 0_0
= -
fls
Hence
+ i"'J2
'P12
c3
12 (36!
- 271' - 2,i'
c3
12 (R, +
R.),
=Mp.
!
12,i4 - 6r
+ r 6),
"'' -i2 (
{ 12
6re' sin t -
(58.6)
1 (
"'" =
14c'
6r%c~' cos ~
3t
)
3r'c cos 2t - r' cos 3t .
"'II
= 30fT'
_!!._ (9 - 36fT
dy =
ic
;;a
(1
- fT - fT"
3w' + 9fJ'8) ,
+ fT4) dfT,
and
with
,f.(fT)
= -3
+ 18fT -
24fT'
'Pll
+ ~"'II
x'dy =
"'11 II~
+ 6fT' -
ic" 1.(fT)
= -72 -fTS
-,
139fT'
ie (fs(fT)
-c'
72';' J~ ;;S(fT _ !) do- = 36 (R,
+ RIO),
where
and hence
'PII
+ i1/tll
and
(58.7)
'PII
+ 18r6 -
3!8),
and
Y' dx
4;1 (6J.1 -
12J.11
where
MIT) = 1 - 6tT
y 2 dx = 1{Iu
I
'Y
+ I2IF' + 2IF1 -
J.I = cos t,
+ 8J.11 + aJ.l' -
57rr'
The boundary
6J.16
+ 2J.1').
= - -c h(lT)
--,
24 IT'
.
+ t1{lu
= -
c
241r
'Y
herr)
rr6(1T -.I) dtT
ic'
= - 12 (Rll
+R
12 ),
with
5'
_ 1 d [herr) )
Rll d---' q
~-,_
,
,fT
-)I1_O
or
and
(58.8)
'1'22 =
1 ( 48r~cJi cos 2t
.SIR t - IOr~'c% cos 2
at
12
+ 24rc'
- 3r'c sin
2t -
rl sin at)'
Before the stresses can be found, the constants K. and Kg must first
From (52.21) we get, since I = 0,
be evaluated.
K. = : ; ;
Now
I.
y' dx dy .,. -
Je yl dx,
I = 21-lI'c'.
227
ff 8:;
x
dx dy =
Ie
dx
-X'I'2
11
= h-X[(l
+ 6 cos t -
= 3" (5
6 cos' t),
<
. t
r" =
2c" sm 2'
0 S t _ 211".
g ~;2
X
dx dy '" (1
== (9
+ u)".c& + uz;c'
+ llu)".c'.
9
Similarly,
Jf
~. dx dy
Ic
= (1
II
ff
xy'dxdy =
+ u)'I' + U'I'.tl dy
,,+ <T1I"C'
_ (6 + 7u)rc'
)2rc
33
'
y[(!
-u [x y3 dX = rc',
11
x 8 dxdy =
~ !eX<dY = -~c',
Xci
+ 40")
+ u) C,
= - 63(1
Y</
o.
228
= a + b(l + cos 8)
and
r' = a'
+ iw
.
L
n--
(a,.
+ ~oft)4'.
-I-- iw =
.
A
rft(a"
+ ~'b,,)e''''
r"(a"
.
L
<I> =
L T"(a. cos 18 - ..
..
(59.1)
w=
T(a. sin nO
229
The Constants (1", b" may be determined either from the bounda.ry condition on the norma.l derivative of the function cI> CEq. (54.3)J or from
the boundary values of the function ~ [Eq. (54.4)]. That ia, we may
solve either a problem of Neumann for the flexure function cI> or a Dirichlet
problem for the conjugate function~. The latter course will be followed, since the boundary condition on ~.ha.s already been given for a
circular boundary [see Eq. (56.1)].
Rewriting Eq. (56.1) for the boundaries with radii r, and ro and using
Eqs. (59.1), we get
I rrCa. sin
-....
I.. r~Ca. sin
nil
+ b. cos nli)
n8
+b
- (%
cos nli) = - (%
Comparing the coefficients of sin n() and cos nil gives a system of equations for the determination of the constants an and bR We have
-r; la_l + rial = -(% + ~1T)rl,
-r;-'a_. + r~a3 = ~rl,
-rola_1 + roal = - C% + ~1T)r3,
-ro'a-3 + r~a. = ~:l:rg,
ifn = 1, 2, 3, ... ,
b. = 0
an = 0
if n = 2, 4, 5, 6, . . . .
Solving these equations, we get
al = - (%
a.
7:1,
+ ~1T)(rl + r3),
a-I
= - (%
a_3 = 0,
+ 721T)rlr:,
cI> = -
~=
II
+ ~ r'sin311 + const.
The expressions for the stresses can be easily calculated with the aid of
Eqs. (54.2).
If ri is set equal to zero, we get the flexure function for the solid circular
beam discussed in Sec. 56.
PROBLEM
Calculate the stresses in a circular pipe of thickness I, fixed at one end and subjected
to bending by an end load W, and show that the following approximate f()rmulas are
valid:
230
-W(l - .):10
"'.. -
-rr_:'
_1!.,
(T .. )...,.
rrtl.
60. Stress Functions and 1na!ogies; Beams of Elliptical and Equilateral Triangular Sections. We recall that in Sec. 52 the equilibrium
equations (52.6) led to the definition of the stress function F(x, y), in
terms of which the stresses T.e and T were determined by (52.7);
aF
1
T
= aF _ !EKeX'
T .. = ox - 2" EK~y .
..,
oy
2
'
The function F(x, y) was seen to be determined by the differential equation (52.8)
V2F(x, y) = -2,..uK~ + 2,..uK.y - 2,..a,
and by the boundary condition (52.14)
Tn COS (X, v) + T COS (y, v) = O.
In the course of the solution of the general flexure problem in Sec. 53,
it was found convenient to phrase it not as a boundary-value problem
for the determination of the function F(x, y) but rather in terms of the
torsion function '/) and the flexure functions '/)1, '/)2 or '/)u, '/)12, '/)21, l"22.
In this section, the flexure problem will be stated in terms of a new stress
function T(x, y), which will be seen to be of value in certain problems.
We introduce the stress function T(x, y) by defining
(60.1)
The functions R(x) and S(y) may be so chosen as to yield either a simple
boundary condition or a simple differential equation for T(x, y). The
stresses can be written in terms of T(x, y), with the aid of (52.7), as
aT
ay + 8(y)
'2 EKeX 2,
To.
T ..
= - ax - R(x) -
(60.2)
aT
2 EK.y,
while (52.8) yields the following differential equation for T(x, y):
(60.3)
V2T(x, y)
== -2puK~ - d~x)
+ 2puKsY _ d~~)
- 2,..a.
aT dy
aydB
+ aT dz
aXd8
$!I
Ij =
dB
[i2 EKeX2 _
B(Y)] dy
ds
231
dx
R(x)
= -%EK.y',
S(y) = %EK.x',
We choose
on Cj
on C.
Thus, the function T(x, y) is constant along the contour C, and since the
choice of this constant cannot affect the stresses, we shall take it equal to
zero. With this choice of the functions R(x) and S(y), the stress function T(x, y) is determined by the differential equation (60.3) and by the
condition
T=O
(60.6)
onC.
It is to be noted that the function R(x) [or S(y) may take any value
~: (
~) vanishes.
or
,...
dx
+ 2,,"K
y
,..%.
_ dS(y) - 2"a
dy
,.. .
= W. =
W.
,
K. = O.
EI.
2jJ(1 + IT)I.
The stress function T(x, y) is determined by the conditions
(60.7)
V'T(x y)
,
+ IT
=0
I.
dy
on C,
232
(60.8)
W~
21.
on 0,
X2
except that 8(y) may take any boundary value along a portion of the
contour where
~! is zero.
_ aT
r - ay
(60.9)
r .. = -
+ 8()
y
W.
21. x ,
aT
ax
The position of the center of flexure can be found in terms of the function T(x, y) by applying the definition given in Eq. (53.1) and using
Eqs. (60.2). We have
xc,W. - Yc'W. =
[f
{x [ - : ; - R(x) -
- y
ff
{2T(X, y)
~ EKyy2]
+ :x [ -xT(x, y)
- xy8(y)
+ ~ EK.x1y]
- :y [YT(X, y)
= 2
ff
+ xyR(x) + ~ EKyXY']}
dx dy
T(x, y) dxdy
Ie
[YT(X, y)
+ xyR(x) + ~ EK.x y3 ] dx
Ie [
-xT(x, y) - xy8(y)
X.,W. - Y.,W. = 2
~he
T(x, y) dx dy
~E (K.
fexy, dx + K. feX'ydy),
",herein we are to set in the function T(x, y) the constant a equal to zero.
The coordinates Xc"
of the center of flexure are then found by comparing the coefficient!! of W. and W.. For the special case of bending
y.,
233
As an illustration of the use of the stress function T(x, y) in the solution of the flexure problem, we consider the bending of a beam of elliptical cross section with a contour given by the equation x'/a' + Il'/b' = 1.
We suppose that the load W. is applied at the centroid of the end section. Since on the boundary one has
x2 =
a'
bi (b 2 -
y2),
Wa' 2
21:
bi (b -
y2).
From Eq. (60.7) it is seen that the function T(x, y) is subject to the
conditions
"21'(
Wz ( 1 +
u u
v
x, y ) -_ T.
T(x, y) = G
I>n
x'
Q2
+ abi2)
y I nR,
y2
+ b2
-
1 = G.
(f. + f. - 1).
T(x, y)
+ u)a 2 + ub 2J W.
+ u)(3a' + b2) T. y
a 2[(1
=
2(1
(x.Q2 + b2 y2
)
1 .
The stresses 1'"., 1'". can now be found from Eqs. (60.9). This method of
solution should be compared with that applied in Sec. 56 to the oame
problem.
The stress function T(x, y) will now be used to solve a special ease' of
the flexure problem for a beam whose cross section is an equilateral
triangle (Fig. 46).
The boundary of the triangular section can be written as
2a+
) = 0,
y'3
V'3 Y
( + 2a+Y)(
(y - a) x
1
x -
The flexure function for a beam with an arbitrary triangular cross section is not
known. Some special triangUlar cross sections have been considered by B. R. Seth,
Prt)Cfledings of the London Mathematical Society, ser. 2, vol. 41 (1936), pp. 323-331.
234
where the origin has been taken at the centroid of the section.
side y == a we have
Along the
boundary values of the function B(y) ruonlt this side, whereas we require
that
B(y) = W. x, = W. (2a + y)~
x = 2a + y.
on
21.
21.
3
va
Therefore we tako.
B(y)
== ::;: (2a
+ y)2,
V T(x, y) =
W. [2(<1 - .J1)
T.
3(<1 + 1) y
32a] ,
where we have set a = 0 and are, accordingly, solving the problem of pure
flexure by a load applied at the center of flexure.
FIG. 46
on
~ ~. a
t!.:
We try
T(x, y)
0=
k[x 2 - U(2a
in R,
aU (2a
+ y)2J(y
+ y)!.
- a)
T(x, y} = ::;:
[X2 - ~ (2a +
y)l] (y -
a).
235
[f
3.y3 a'
2
T(x, y) dx dy = 3
~I~a' = a:~,
Jc
Jo
~.Y3_a,
5
and therefore
= O. Since the cross section is symmetrical about the
y-axis, we see that the center of flexure is at the origin, and hence at the
centroid of the section. Thus, the function T(x, y), given above, furnishes the solution of the flexure problem for an incompressible beam of
equilateral triangular section when the load W. is applied at the centroid.
The flexure function T(x, y) was introduced, for the case of bending by
a load along a principal axis, by Timoshenko' and was used by him to
solve the flexure problem for a number of cross sections.
It will be recalled that, when the functions R(x) and S(y) were introduced, it was remarked that they might be so chosen as to yield either a
simple boundary condition or a simple differential equation for the function T(x, y). The first course led to Timoshenko's stress function
T(x, y), discussed above, which can be interpreted physically as representing the deflection of an elastic membrane stretched over an opening
of boundary C in a plane plate and subjected to a nonuniform load. We
follow now the alternative course and choose R(x) and S(y) to make
T(x, y) a harmonic function.
Let us define
R(x) = -w,Ky x 2 - J1.OiX,
(60.13)
{ S(y) = p.uK.y2 - J1.Oiy.
y.,
We shall designate the function 7'(x, y) defined by Eqs. (60.3) and (60.4)
with this choice of R(x) and S(y) by lI1(x, 'y). Then these ~quations
become
\l211f(X, y) = 0
(60.14)
and
ddM
= [J1.(1
8
+ rr)K.x' -
J1.rrKrl/'
.
+ J1.OiV] dt!!l8
- [J1.(1
+ rr)K.y' -
J1.rrK.x 2
J1.OiX) ::
on C.
(1922), p. 398.
236
+ y2) + I'K. [ -
M ... 72l'a(x 2
+ I'K. []icrx'
]iuyl
+ (I + cr) l<,,,.w)
Xl dy]
....w.)
+ cr) j(%O.lIo)
((z.) Y'dx] + const
- (I
on C.
The stPesseS 1',.,1',. can be written from Eqs. (60.2) and (60.13) as
aM
Ty
T,.
= -
+ Jl.K.[cry'
a;~ +
- (1
I'K.[crx 2
+ cr)x']
(1
- I'ay,
+ cr)y'] + I'ax.
aM
By
1',.
= -
w.(1 +
+ 21.
cr
.)
cr y - x
- I'ay,
aM
ax
+ I'ax,
+ Y 2)
- 6(1
W..
+ cr) Yo
y
W f("W)
+ --=
21.
cr
(z)
x'dy
on C.
One can interpret the determination of the harmonic function M(x, y),
subject to the condition (60.15) or (60.16) on C, in terms of a membrane
analogy, as was done in connection with the torsion problem in Sec. 46.
Thus, the solution of the flexure problem by means of the membrane function M(x, y) is mathematically identical with the determination of the
deflection of an unloaded elastic membrane stretched across a closed space
curve whose projection on the xv-plane is the contour C and whose
variable height above the plane is given by the boundary values of
M(x, y) [(60.15) or (60.16). This analogy, for the case in which the
boundary values are given by Eq. (60.16), has been used by Griffith and
Taylor,l among others, to obtain experimental solutions of the flexure
problem for beams whose cross sections are such that the problem does
not yield readily to mathematical treatment. N emenyi2 has derived both
M(x, y) and another flexure function Fl(x, y) as special cases of a more
1 A. A. Griffith and G. 1. Taylor, National AdvUory Committee on Aeronautic.
'l'ecJmical Report, Great Britain, vol. 3 (1917-1918), p. 950.
Bee S. Timoshenko and J. N. Goodier, Theory of Elasticity, 800. 113, for a descrip.
tion of the experimental proeedure.
P. Nem6nyi, ""Ober die Bereehnung der Sehubspannungen Un gebogenen Balken,"
ZeitBchriftjfJ.r G~ Mathematik und Meclum.ik, voL 1 (192t), pp. 8!Hl6.
237
general flexure function and has discussed these two formulations of the
flexure problem in the light of the membrane analogy and of numerical
solutions, respectively.
There exists a very close connection between the membrane function
M(x, y) and the canonical flexure functions 'Pll, 'P12, 'P21, 'P22, discussed in
Sec. 53. A comparison of the boundary condition (60.15) with the
boundary values taken by the torsion function 1{1 [Eq. (35.4)] and by the
flexure functions 1{1, and 1{12 [Eqs. (52.17)] shows that
(60.17)
M(x, y) = p.at/t(x, Y)
+ pKm1{l,(X, y) + pK",y.(x, y)
on C.
:(
X
FIG. 47
+ iM
= I'a(rp
y)
so -that
Equations (53.5) now furnish the following relation between the complex
membrane function L + iM, on the one hand, and the torsion function
'P + i1{l and the canonical flexure functions CPll + i1{ll1, . . ,on the other
hand:
L
+ iM
I'a(rp
238
EL.~STICITY
MATHEMATICAL THEORY OF
by
(61.1)
where tl = 4a/31T is the distance of the centroid 0 from the point 0' on
the diameter of the semicircle of radius a.
Making this change of variables in Eq. (60.7), we get
(612)
.
V2T(")
x ,y = 1
W. (y' + T.
(f
11
y-)
+ T.
WE y
- 2
1'0<,
where we take S(y) in accordance with (60.8), and note th~t along the
circular part of the boundary x" = a2 - V"~, and along the diameter
dy'jd8 = O. The corresponding boundary conditions are:
(61.3)
T=O
{ T=O
on y' = 0,
on y' =
va
2 -
/2
(J
B=
(61.5)
(J,
!4 (W.
I. 1 +
_(f_
+ 21'0<)
tl
(f
(61.6)
,,=0
The coefficients Aft, Bft must be chosen so that the conditi0I1S (61.3) are
fulfilled; that is,
(61.7)
T(r,
~) = 0,
T{a,8) = 0,
+ Crt cos 28
+ A I a+J1''''+l ow, (2n + 1)',
.. -0
239
and choose B = C.
A 25+1 so that
Aa' cos 8
+ Ba 2 (1 + cos 28)
... -
+ 1)8,
.. -0
".
- 2<
".
< 2'
Al = - 16a B - a 2 A
37r
'
a- 2' H I6a 2( -1)-B
A'''+l
".(2m
1)(2m
m=
1)2 - 4)'
1,2, ....
T(r, 8) = A r 3 cos 9
+ Br (1 + cos 29)
+ A.n+lr' n+
2
L
~
cos (2n
+ 1)8.
n=O
The center of flexure obviously lies on the y-axis, and, by using (60.11),
it ('an be shown that'
y:,= 15(18:1T)1([3+1T(~~-1)l
Leibenson' used a similar method to obtain an approximate solution
of the flexure problem for a semicircular tube of small thicknefls.
The flexure problem for a cylindrical beam whose cross section is a
segment of the circle was solved in bipolar coordinates with the aid of
Fourier integrals by Uflyand. 3
62. Multiply Connected Domains. Deformation of NonhoDiogeneous
Beams with Free Sides. Other Developments. Although the mathematical formulation of the flexure problem for beams with multiply
connected cross sections is quite straightforward, an explicit determina1
\'
1..
m-a
(2m
+ 1)2(2m
- 1)2(2111 .~
ai - 8 -
a".s
128'
240
FIG. 48
the cross section bounded by the exterior contour Co and several interior
contours Ck (k = 1, 2, . . . , m), shown in Fig. 48. The regions R.
(k = 1, 2, . . . , m), enclosed by the C., are filled with materials whose
1
The general flexure problem for a hollow beam bounded by two eccentric circles
241
u,
Ti
Tij'; = 0
(a)
(b) (T,;.;)O = (Ti;';).
{
(c)
(Ui)O = (Ui).
on Co,
011 C. (k = 1, 2, . . . ,m),
on Ck (k ;", 1, 2, . . . ,m).
The subscripts 0 and k outside the parentheses in these expressions indicate that the values of affected quantities are computed along the interior
contours C. for the regions R. and R., respectively. The unit normal
vectors ("i)O and (Pi). along such contours point int,o the regions Ro and
R. as shown in Fig. 48.
The satisfaction of boundary conditions (62.1), in the instance of the
Saint-Venant torsion problem, when it is patterned along the lines of
Sec. 34, presents no logical difficulties. However, in the problems of
extension, pure bending, and flexure by the transverse force, the Saint-..
Venant assumption that Tn, T 1III , and T"" vanish does not lead to continuous
displacements along the contours C. unless the media on either side of the
contours have the same Poisson's ratios. The reason why no difficulty
of this sort arises in the torsion problem is that the displacements there
do not depend on Poisson's ratios.
To remove the discontinuities in displacements, it is necessary to
superimpose un displacements resulting from the hypothesis
the displacements present in certain two-dimensional elastostatic problems. Such problems are discussed in the next chapter. In this manner
valid solutions have been deduced for several interesting problems on
deformation of compound beams. Besides solutions of the torsion and
flexure problems for circular beams reinforced by circular cores (in
general, eccentric), solutions are available for the torsion of a composite
rectangular beam formed by gluing two rectangular beams along their
sides' and for the torsion of an elliptical cylinder reinforced by a circular
rod whose axis coincides with the axis of the cylinder. s
The Saint-Venant torsion and flexure problems for cylinders having a
small initial twist in the natural state has been considered in a series of
1 See papers in the preceding footnote and a paper by L. E. Payne, "Torsion of
Composite Sections," lqwa Stau College JOU'I"It4l qf SciImce, vol. 23 (1949), pp. 381-395.
D. I. Sherman, lnzlienernyi SOOrnik, vol. 10 (1951), PI'. 81-108 (in RWlllian).
242
= 0,
+ kz
) = 0,
with k, again, a small parameter. This cylinder has a slightly curved
axis. The torsion and bending of such cylinders were studied by Riz
and Rukhadze. 3
f(x, y
1 P. M. Hiz, D()/dady Akademii Nauk SSSR, New Series, vol. 23 (1939), pp. 17-20,
441-444, 765-i67.
A. I. Lourie, Prikl. Mat. Mekh., Akademiya Nauk SSSR, New Series, vol. :.I (1938),
pp.55-68.
A. 1. Lourie and G. Dzhanelidze, Doklady Akademii Nauk SSSR, New Series, vol. 24
(1939), pp. 24-27, 227-228; vol. 25 (1939), pp. 577-579; vol. 27 (1940), pp. 436-439.
A. Gorgidze and A. Rukhadze, Soobshcheniya Akademii Nauk Gruzinskol SSR,
vol. 5 (1944), pp. 253-21.'2.
A. K. Rukhadze, Soob8hcheniya Akademii Nauk Gruzinskol SSR, vol. 5 (1944),
pp. 483-492; Prikl. Mat. Mekh., Akademiya Nauk SSSR, vol. 11 (1947), pp. 533-542.
This paper contains an explicit solution of the flexure problem for an elliptical rod
with a small initial twist.
All these paIWrs are in Russian.
'D. Y. Panov, "Concerning the Torsion of Nearly Prismatic Rods," Prikl. Mal.
Meleh., Aleademiya Nauk SSSR, New Series, vol. 2 (1938), pp. 159-180; Dolelady
Aleademii Naule SSSR, New Series, vol. 20 (1938), pp. 251-253.
A K. Rukhadze, "The Problem of Bending Nearly Prismatic Beams," Soobshchenilla
Akademii Nault Gnmmleoi SSR, vol. 1 (1940), pp. 577-582; Prikl. Mal. Mek".,
Aleademiya Nauk SSSR, vol. 6 (1942), pp. 123-138.
Ail these are in Russian.
P. Riz, Dokladll Akademii Naule SSSR, New Series, vol. 24 (1939), pp. 110-113,
229-232 (in Russian) .
.A. K. Rukhadze, Boobsheh."iyu Akademii Nauk GruzifIBkoi SSE, vol. 2 (1941), pp
.3&-42 (in Russian).
243
Equilibrium equationa
7'''.i
= 0,
(i = 1,2,3),
V'rOj
+ 1 + t1 9.;1 = 0,
(i,j = 1,2,3),
c. Btnmdary conditiona'
7'11"1
7'21"1
7'11'"
+ 7'U"2 =
+ 7'.'''' =
+ 7'''''' =
T 1(x" X2),
T.(x" Xii),
T 3(x" X.),
This problem was first considered by Almansi and Michell,' who demonstrated in effect that it is possible to reduce it to the determination of
two stress functions, one of which is harmonic and the other biharmonic.
Some broad clasBes of boundary-value problems in the biharmonic equation are discussed in Chap. 5, but, as an introouction to them, we consider in the following section the problem of torsion of a long cylinder
by tractions suitably distributed over its surface.
64. Torsion of a Cylinder by Forces on the Lateral Surface. Let a
cylinder with an arbitrary cross section R be twisted by tractions T
applied to the lateral surface. We suppose that the cylinder is of length
1 and that one of its ends is fixed in the plane XJ = 0, while the end
x, = 1 is free. The surface tractions T, assumed independent of the
xrCoordinate, are directed parallel to the x,x.-plane and proouce a twisting moment in the cross section Xa = const. We take the magnitude of
this moment to be Ml, so that M is the torque per unit length of the
, If the system of stresses T, is not self-<lquilibrating, it is necessary to apply a suitable distribution of forces on one of the ends of the cylinder in order to maintain the
cylinder 88 a whole in equilibrium. This can be done by supposing that the end
$0 ... 0 is fixed and the other end is free.
t E. AIma.nsi, Nota II, Alti della Accademia naziOMle lUi Lin~ Rmciieomi, Rome,
aer. 5, voL 10 (1901).
J. H. MicheD, Qwrterlll Journal of M~, vol. 32 (1901). Almansi also conthe _
when the external stresses T. are polynomials in
A summary of
these eontn'butions is contained in Love's Treatise (1921), Sees. 239-241. The eorreIIJIOndiDc problem for compound beams when PoiBm'll ratios are identical thtoughout
the .en. section, but Young's moduli are diff_t for each component of the beam,
WM treated by O. M. Hatiasi:tvili, SoobBheAeniva Akademii Nault ~ 88R.
vOl. 13 (1902), pJ). 335--341; vol. 14 (1953), pp. 197-204 (in RU8I!ian).
ei_
z..
cylinder.
245
Thus,
Ie T(xI,
T. - 0,
fe [xIT.(x., xz) -
x.) d8 - 0,
x.TI(XI, x.)] ds = M,
T13
II
II
II
1'13
the
= H
'J
(XIT32 -
X2T 31)
XIT,.
du =
II
l,
du
= 0,
must yield:
1'.,
for x,
0,
T33
Ti;
du
(64.2)
T23
du
II
1"3
du = MI,
X.T'3
du = O.
The conditions in the first line in (64.2) demand that the resultant force
vanish on the end x, = 0, the second line states that the twisting moment
in the section X3 = 0 is MI, and the last line requires that there be no
bending by couples.
The conditions on the lateral surface of the cylinder, clearly, are
1'11",
(64.3)
{
1'21'.
1'31"1
+
= T , (x" x.),
+ 1"'''' = T .(x., x.),
+
= 0,
on C.
TU'"
1'321'2
The problem thus consists in determining the set of functions Ti; satisfying the equilibrium and Beltrami's equations and the boundary conditions (64.1) to (64.3).
A solution of the problem stated with this degree of generality presents complications because the third of the boundary conditions in (64.1)
is .difficult to satisfy. If, however, we relax the condition 1'3' = 0 for
x, = l, by merely requiring that the resultant force in the direction of
the xraxis vanish, that is,
(64.4)
for x, = l,
Tas du = 0
XITaa
du =
XiTaa
du = 0,
246
One can argue on the basis of Saint-Venant's principle that the solutions of the original and relaxed problems can differ significantly only
near the end $, .. l of the cylinder.!
The fact that the forces assigned on the lateral surface produce torsion
of the cylinder suggests that the expressions for the Tlj in the relaxed
problem have, in part, an appearance similar to the stresses in SaintVenant's torsion problem. Taking cognizance of the linear variation of
the twisting moment along the length of the beam, it is reasonable to
consider [cf. Eqs. (34.4)] the shearing stresses in the form
(64.6)
rW
{ TlV
= pa(l - X3)(cp .
XI),
= pa(/ - X3)("".1 - xo).
TW =
Til> =
Ti\' =
-2pOl(p,
T~V. = ,.,.0:<;,
%pa(xi - xi),
we find that cp satisfies Laplace's equation, and the stress system Tl}' is
an admissible system. Inserting from (64.6) in the third of the boundary
conditions (64.3) yields
d""
dv =
XOVI -
XIV.,
on C,
rir
r'W
rW
rW = TW =
BXI
+ Cx.,
r~W
= 0,
u. = 0
(a = 1,2),
247
Ti'r
T~'i = -aU. ll ,
TW = aU. lt ,
T.f = -auV'U,
where U(XI, X2) satisfies the biharmonic equation
T~f
= -aU. lt,
= TW = 0,
VU
in R.
V,l!
a
T}3
-;; =
(64.9)
+ /J'P,
+ ~ (xl
- xV,
+ ,..,,)
"'~
1'23
1'83
= -U. 11
Tn
a.
+ x,),
= -2/J'P - uV'U
BXl
+ Cx .
If we now insert from (64.9) in the first two boundary conditions (64.3)
P, =
dx,
ds'
P2
= -
= To/a -
/J'PP2 -
:2 /J(xl
Ttla -
/J'PP, -
:2 /J(xl
(64.10)
=
dx,
. fin d
da' we easlly
xnp"
on C,
X~)V2'
on C.
Thus, if the torsion function 'P is known for the region R, the values of
the derivatives U. G of V can be calculated on the contour C. The problem of determining the biharmonic function U from prescribed values of
the partial derivatives of U on the contour C is known as the fundamental
boundary-value problem in the biharmonic equation. We shall see in the
next chapter that there are effective methods for solving it.
It remains to show that the constants a, A, B, and C can be chosen in
such a way that the end conditions (64.2), (64.4), and (64.5) are fulfilled.
The fact that 1'13 = 1'2. = 0 on x. = l is obvious from (64.9). The verification that the resultant forces ffT13 dtr and ffT" dtr vanish over the
end x. = 0 is, in every detail, identical with that given in Sec. 34. The
condition
(x IT.,
X2T_U)
= D'
1 See Eqa. (69.4) and (69.5). The constant -a was introduced here for convenience. It can clee.rly be absorbed in U(.,., .,.).
248
MATHEM.~TICAI,
THEORY OF ELASTICITY
12(a2
+ b')
where a and b are the semiaxes. The torsion function '" for this section is given by
formula (36.6).
L. N. G. Filon, Phiu",ophical Transactions of the Royal Society (London) (A), vol.
198 (1902), p. 147. This problem was reconsidered by A. Timpe, MatheTTW.tisc1te
Annalen, vol. 71 (1912), p. 480.
N. V. Zvolinsky and P. M. Ri~, IZ!Jestiyu, Akademiya Nauk SSSR, No. 10 (1939),
pp.21-26.
P. M. Biz, Prikl. Mat. Mekh., Akademiya Nauk SSSR, New Series, vol. 4 (1940),
pp. 121-122.
These authors also consider the case where T varies linearly along the length of the
cylinder. The paper hy Zvolinsky and Biz contains explicit formulSl! for the displacements and stresses in a circular cylinder twisted hy tangential tractions.
S. G. Lekhnitzky, Prikl. Mat. Mekh., Akademiya Nauk S8SR, vol. 2 (1939),
pp. 345-368; vol. 6 (1942), pp. 3-18. See also this author's monograph Theory of
Elasticity of an Anisotropic Body (1950) (in Russian).
H. Luxenberg, Journal oJ ResearcA oJ the National Bureau of Standards, vol. 50
(1953), pp. 263-276.
V. K. Prokopov, Pri1cl.. MGt. Melch., Akademiya Nauk 888R, vol. 13 (1949),
pp. 13&-144 (in Russian).
B. L. Abramyan and M. M. DzerbSl!hyan, Pri/d. MGt. Mekk., Akademiya Nauk
888R, voL 15 (1951), pp. 451-472 (in Russian). See also B. A. Kostandyan, IZf1estiya
AkademU Nauk A~ 8SR, Physics and Math Series 7, No.4 (1954), pp. 2353 (in Russian),
'G. S. Shapiro, Prikl. Mai. Mekh.., Akademiya Nauk 88BR, vol. 8 (1944); vol. 17
(1953), pp. 243-252 (in Russian).
CHAPTER
ticity (1953).
249
250
u.. = u"(x,,
{ u. = o.
X2),
It follows from (66.1) and from the definitions (7.5) of the strain and
rotp,tion tensors that the non vanishing components of these tensors are
given by the formulas,
(66.2)
X,.
{22.S1
Some writeM define t])e .we plane atrain by requiring that v, - conat and
and 'lit 1M! the ~ '!;tand "'
VI
251
we get
(66.3)
(66.4)
TafJ
= AIH~
Tn
+ P,(ua.~ + u~.),
Ar'J,
TI3
= T = 0,
T22
(M
2(A
+ T.~'
Indeed,
T33
+ p,) +
= a(Tl1 + T),
= 2(A
(TU
T2.)
and three equations (66.3), The substitution from (66.3) in (66.6) yields
the appropriate Navier equations:
(66.7)
where
82
V' ""_
8"
+-.
<lxfax~
252
253
l'
If
dx~
dx.
The distribution of stress Taa "" U(TU + Tn) over the ends of the cylinder
may also produce a bending couple whose moment lies in the planes of
the ends. Indeed, the longitudinal stresses Tn are necessary to maintain
the cylinder in the state of plane deformation; without their presence
the displacement u. will not, in general, vanish. If, however, in the
given physical problem the ends of the cylinder are free, the desired solution can be got by superposing, on the Ilolution of the plane problem just.
FIG. 49
considered, the solution of an auxiliary problem. This auxiliary problem concerns the deformation of a cylinder with free lateral surface by
the end loads equal and opposite to those given by equation (66.5). If
the cylinder is long, the auxiliary problem is in the category of simple
Saint-Veuant's problems' fully considered in Chap. 4. After superposing the solution of an auxiliary problem on the solution of the planedeformation problem, the resulting deformation will not, in general, be
plane.
67. Plane Stress. Generalized Plane Stress. A body is in the stare
of plane 8tres8 parallel to the x.x.-plane when the st.ress components T.,
TU, Ta. vanish.
If we wrire the stress-strain relations (22.3) in the form
(67.1)
T'j
MO,j
",(u..j
Uj ),
iJ
11,
(67.2)
UI. .... -
"
+ 2# (ULI + Ut,t).
254
Substituting this ill (67.1) jields the following expressions for the
vanishing components 'Tall:
(67.3)
2>""
+ 2" (11,1.1
'TIl
= >..
'T'2
= >.. 2>""
+ 2" (11,1"
TU
= ,,(11,1,2
IlOJl~
+ 11,2,,) + 2,,11,1,1,
+ 11,.,,) + 2"11,,,2,
+ 11,.,1).
'T:aII,P
+ F.
0,
where {h ...
( >..
11,1.1
2>"p.
) 0"1
2_
11,2,2
and Vi""
0'
"2
"Xl
-Fa,
o
+ "X,
"2'
255
ponent Fa of the body force vanishes and the ,components F. are symmetrical with respect to the middle plane. Under these hypotheses, the
points of the middle plane will undergo no displacement in the directiOll
of the xraxis, and 'if the plate is thin, the displacement '1, will be small.
Indeed, the symmetry of distribution of external forces implies that the
mean value of '13 with respec~ to the thickness of the plate is precisely
zero. For thin plates the mean values Uo of the displacements Ui give as
FIG. 50
U.(XI, X2) e
u..
T13(XI, X2,
h) =
T23(XI, X2,
h) =
T33(X1,
X2, h)
= 0,
1"13.1
1"23.2
+ T33.3 =
0,
demand' that 1".8 (X" X2, h) = O. The fact that T3. and its derivative
with respect to x. vanish on the faces of the plate suggests that T33 can
differ from zero but slightly throughout the plate if h is small. This
justifies us in assuming that T33 e 0.
The remaining equilibrium equations
T
1 From
al.l
Td.1
1"aI.3
+ Fa = 0,
256
upon integration with respect tGXI between the limits -It and
(67.9)
1
2ii
fA~Io
(T.. l.1
1'... 1
1'01.1
+1&, yield
+ F ..) dxa = u.
Since
1'.. ,.,
+ 1'..... + P.. = 0,
where
T..#(X" x.) ""
(67.11)
P,,(XI, X2)
E!
J:.
1'v"1l..
a;; +
+ (A- + 1') .,,--"x..
f!,
l' ..
(XI, x.) = 0,
from which the average displacements 11.. can be determined when the
values of the 11.. are specified on the contour.
The system of equations involving the average stresses 1'..~ can be got
by deducing the corresponding Beltrami-Michell compatibility equations.
It turns out to bel
(67.14)
where 9 1 = 1'11 + Ttt.
This equation, together with the equilibrium equations (67.10), suffices
to determine the mean stresses 1'# when the boundary conditions on the
edge are given in the form,
For integrating these equations with respect to X3 between the limits -It
I
257
nnC,
where '/'",(8) dB are the components of applied force acting OJ{ the element
of.arc dB of the contour C.
The two-dimensional boundary-value problem consisting I)f the lIystem
of Eqs. (67.10), (67.14), and (67.15) is known as the problem in generalized
plane stress. 1
68. Plane Elastostatic Problems. The discussion of the plane-deformation problem in Sec. 66, and of the generalized plane-strell!s problem in
Sec. 67, shows that their mathematical formulations are idtlntical. The
relevant differential equations and
%2
boundary conditions in Sec. 67 differ
from those in Sec. 66 only in the
appearance of the barred symbols:
iz"" T",~, X, '/'a, etc. Henceforth we
shall refer to problems of these two
types as plane elastostatic problems.
In the formulation of these plane
problems no restrictions on the connectivity of the region R was introduced. If the region R is multiply
connected and finite, we shall supFIG. 51
pose that its boundary C consists of
m + 1 simple closed contours Co, such that the exterior cont<lUf Cm +1 contains within it m contours C; (Fig. 51). We shall BUPP()8e that the
contours C" with the possible exception of a finite number 4f points, are
smooth. This means that a smooth contour C. can be represented
parametrically by equations of the form Xa = X,,(8), where the functions
X,,(8) have continuous derivatives that do not vanish for th() same value
of the are parameter 8. We shan agree that the positive direction of
description of the contours is such that the region R remain.s on the left
in the course of tracing the contour Co. The positive direlltion for the
tangent vector t along C; is that of the positive direction of description
of C; and the positive unit normal y at any point of the contOln is directed
outward relative to the region R.
In the boundary conditions,
"
(68.1)
on C,
on C,
I This terminology was introduced by A. E. H. Love, but such 8ta~ of stress were
first investigated by Filon in the study of bending of a beam with rect~lar cross section. See L. N. G. Filon, Philo8ophical Transactions of the Royal SociUII (London) (A),
vol. 201 (1903), pp. 63-155.
o.
~'O-DIMENSIONAL
259
ELASTOSTATIC PROBLEMS
(68.3)
T,.~
I
.
2X + IJ.
4(X + 2p,)
U~I = - 8(X
pul
+ X22) u~
!(
2
X,
2(X
P,
+
2/,) pw XaX~,
2
pul2
+ T~1J,
+ U~l),
T~'JV~
T~l),
on C.
U~l) = f~l),
(69.1)
'Ta~.~
= 0,
in which the 'Ta~, as follows from Eqs. (66.8) and (67.14), satisfy in the
region R the compatibility condition
(69.2)
On the boundary C of R, the functions
Ta~
(69.3)
where the Ta(s) are known functions of the arc parameter son C.
The equilibrium equations (69.1), as first noted by a British astronomer
G. B. Airy, imply the existence of a function U(X1, X2) such that
(69.4)
T12
= - U,l2,
T11
= U,22'
Indeed, the 'Ta~ thus related to the Airy function U(X1, X2), satisfy Eqs.
(69.1) identically, while the compatibility equation (69.2) demands that
U(X1, X2) satisfies the biharmonic equation
V'"V2U = 0,
or
(69.5)
V'U e U. lll1
0,
in the region R.
Every solution of this equation of class C' is termed a biharmonic function, but inasmuch as we are interested in those states of stress for which
260
o
FIG. 52
(69.6)
but, from Fig. 52,
(69.7)
V2
= cos (Xt,
v)
=-
COS (Xl, 8)
:2,
= _ :"
~ (U. 2)
= T,(s),
- ds (U. I )
= T.(s).
(69.8)
(69.9)
( U .(8)
J...
I.:
T.(s) ds
T,(s) ds
== /,(s)
== "(s)
80
to a vari-
+ CI,
+ Ct.
It is clear from (69.9) that the derivatives of U along C are not determined uniquely. Moreover, if the region is multiply connected, the
261
integration has to be performed over each contour C. forming the boundary of R and the resulting functions /,.(B) need not be single-v&lued. Some
degree of arbitrariness in the choice of U and its derivatIves is to be
expected, however, inasmuch as the stresses rail are determined by the
second derivatives of U.
We see that the boundary-value problem characterized by the system
of Eqs. (69.1), (69.2), (69.3) is intimately related to the boundary-value
problem of the type:
in R,
V'U = 0
(69.10)
{ U.", = 1",(8)
on C,
wherein the 1",(8) are certain known functions. The problem (69.10) was
the subject of numerous investigations that have led to developments of
cardinal importance in the theory of differential and integr&l equations,
in the calculus of variations, and in several other branches of analysis. 1
It is known as the fundamental biharmonic boundary-value problem.
This problem can be phrased in a somewhat different form, by observing
that the knowledge of the U. a (8) on C permits one to compute the value
of U(8) and of its normal derivative
~~ on C.
Indeed
and since
dU = U. a dx a = la dx""
U(s)
Conversely, if U and
~~
f",
(69.11)
V'U = 0
in R,
U = f(8) + const }
dU
dv = g(s)
on C,
Fa = -V,,,,,
1 A apecia1 case of this problem first arose in the study of the transverse defleetions
of elsmped elastic plates.
This ill the ease with the gravitational and centrifugal forces discussed in Sec. 68.
262
U. tl
+ V,
TU -
-U.1>,
Tn'" U,lI
+ V.
21' V'V.
V'U =->'+21'
If V is harmonie, then, as above, U is a biharmonic function.
we are led to consider equations of the form
Otherwise,
in R.
in R,
and if we let ViU "" PI(XI, x.), the function PI is, clearly, harmonic in R.
Consequently' we can construct an analytic function
F(z) "'" PI
of a complex variable z = XI
+ iP.
J:'
Then
+ P dx.)
.Is independent of the path joining an arbitrary point M ,(x!., x:), with the point M (X"
P.(x., x,) -
(-P."dx.
z.), II' PI is harmonie. 'It follows that P.(Xl, x,) is determined to within an arbitrary
constant C and, hence, F(z) .. P. + iP to within a pure imaginary constant Ci. If
" x.), and hence F(z), is single-valued. In a
the region R is simply connected, P,(x.,
multiplY connected region, F(z) is, in general, multipll'-valued, and we can confine our
attention to some single-valued branch of F(z). The same considerations apply' to
.,('1") ... ~~ JF(z) liz.
263
(70.2)
= PI
+ ip.,
+ l.op.
- = -1 (P + t'P)
.
'( ) op.
fOZ=ox.
ox.
PI.I
PI.I
PI .
= p'.' = ~~p ..
= -P2.l = -%p .
Using these results and the fact that PI and p. are harmonic in R, we
readily verify that
v'( U -
in R.
(70.4)
+ x(z)],
where z == XI - ix. and mdenotes the real part of the bracketed expression.
Since fO(Z) and x(z) are analytic functions, it follows from (70.4) that
U(XI, x,) is of class C" in R. The important representation (70.4) was
first deduced by Goursat by different means.
If we denote the conjugate complex values by bars, so that, for example,
~(z) == PI - ipo, then (70.4) can be written as
(70.5)
2[1 = ZfO(Z)
+ z~ + x(z) + Xtz).
1"11 ...
U. n ,
1"11
= -U,lt,
1""
= U. ll,
2[1 =
~(z)
'1"11
{
+ ~'I"12
STU
'1"22 -
U,l
~U'12
= U,22 -
-i(U,l
7" iU,2)."
+ iV,. =
I"(z)
We easily
+ Zl"'(z) + ~,
(71.3)
+ iT12
'1"22 -
iTI2
I"'(z)
= I"'(z)
T11
+ Tn
1'11
'1" . . -
= 2[1"'(z)
+ 2iTn
U, = 'Ar'J
21'Ul,l,
= U,lI = 'Ar'J + 2I'u".,
= - U,I' = I'(UI,2 + U',l).
T11'=
(71.5)
{
1'22
1'12
Solving the first two of these for Ul,l and U,," we get
2",Ul,l
X + 2}o!
+ 2(X
+ "') V U,
X + 21'
U,2' + 2(X +-~ V U,
= - U,lI
2",U.,2 = -
U . = aU at
a. ilx,
+ au ~
iI' ax,
265
2"u.,! = - U. 22
2(X + 2,,)
+ X +" Pl.l,
2(X + 2,,)
+ X +" P.,10
- U. 1
(71.6)
where f(x.) and g(Xl) are, as yet, arbitrary functions. The third of Eqs.
(71.5) serves to determine! and g. Since P,.' = -Pt." we easily find that
f'(x,)
+ g'(x,)
0,
and hence
!(x,) = ax, + fl,
g(x,) = -ax, + 'Y,
where a, p, and'Y are constants. The forms of! and g indicate that they
represent a rigid displacement and can thus be disregarded in the ana.lysis
of deformation.
If we set! = g = 0 in (71.6), recall that <p = p, + iP2, and make use
of (71.2), we easily deduce the compact formula
(71. 7)
where
(71.8)
being the Poisson ratio.
The formula (71.8) for" is that corresponding to the state of pla.ne
strain. In the generalized plane-stress problems, X must be replaced
by ;; = (2X,,)/(X + 2,,), and if the corresponding value of " in (71.7) is
denoted by X, we find
0-
X + 3"
5X
+ 6"
3 -
+" = 3X + 211- = 1 +
We note that both x and x are greater than 1.
X
;;
00-'
Inasmuch as the functions ",(z) and I/I(z) are analytic in the interior of
the region R, it follows from formulas (71.4) and (71.7) that the T~ and
u. are analytic functions of the real variables x, and x, throughout the
interior of the region occupied by the body.
T..
= u.. = 0,
then the '1'~ vanish throughout the region R. This result is due to
Almansi, l who proved this theorem for the three-dimensional case.
12. The Structure of Functions l"{z) and I{I(z). The consideratiolls of
Sec. 70 indicate that there is some freedom in the choice of fUllctions
appearhig in the representation (70.5) of the general solution of the
biharmonic equation. This implies some arbitrariness in the selection
of functivns rp(z) and "'(z) in the representation of stresses and displacements by formulas (71.4) and (71.7). In this section we discuss the
precise extent of this arbitrariness and record the structures of l"(z) and
!/t(z) for several domains of interest in applications.
We begin with a finite simply connected domain R bounded by a contour C and raise the question: What is the difference in the forms of two
sets of functions (rp, "') and (l"o, !/to) that correspond to the same stress
distribution in R?
If the stress distribution specified by l" and", is to be identical with that
given by rpo and !/to, the formulas (71.4) demand that
<R[l"'(z)] = <R[rp~(z)]
(72.1)
and
(72.2)
il""(z)
+ ",'(z)
= zrp~'(z)
l"~(z)
rpo(z) = rp{z)
= l"'(z)
+ "'~(z).
+ ci,
+ eiz + a,
so that
(72.4)
"'o(z) = ",(z)
+ fl,
267
c=o
(72.5)
and
xC<
p.
Hence, if the displacements u; are known in R, the function 'P is determined to within a complex constant c< and the specification of this constant completely determines the constant (3 in !/-o(z).
If the origin of coordinates is taken within R, the functions ~(z) and
!/-(z) will he determined uniquely for the given state of stress, if c, c<, and
(3 ar,e chosen so that
~(O) =
(72.6)
0,
g~'(O)
= 0,
!/-(O) =
o.
L" a.z",
L" bnz
I/I\Z) =
fl
in R.
n=O
n=O
L'"
A. log (z - Ir.)
+ fez),
k-I
(a)
,,(z) -
t-I
B. log (z - '.)
k-I
+ ".(z),
268
I(z) dz -
B. log (z - z.)
+ 9'.(z),
k-l
where ".(z) is an&1ytic and single-valued in R. It is clear from (a) that 9'''(z) is a single-valued function, and since the left-hand members of {71.4) are single-valued, it
follows that >f'(z) is &180 single-valued. Therefore
L
".
(b)
>fez) -
C.log (z - z.)
+ >f.(z),
k-1
where >f.(z) is analytic and single-valued in R. If we further suppose that the displacements u" are single-valued functions in R, then the increment acquired by
21'(ft, + iu.) in describing the contour C. is zero. Using this condition in (71.7),
with 9' and >f in the forms (a) and (b), we find
21ri[("
o.
(72.7)
!/t(z) = 2".(1 ~ x)
(Xlk) -
+ ..yo(z),
k-l
where (Xik\ X~) is the resultant vector of external forces applied to the
contour C. and is an arbitrary point in the simply connected region R.
bounded by C.. The functions 'Po(z) and !/to(z) are single-valued analytic
functions in R.
If R is an infinite region, bounded by several simple closed contours'
C. (k = 1, 2, . . . , m), and if the stress components T'j are bounded in
the neighborhood of the point at infinity, then' it is not difficult to prove
that for sufficiently large Izl,
z.
, The region R in this case can be thought to be obtained from the region R of
Fig. 51, by making the contour Cm +, expand to infinity. It corresponds to an infinite
plate with m holes bounded by the C ..
(72.8)
'P(Z
I)
t/t(z )
Xl + iX,
= - 211"(1 + ,,) log z
=
K(XI - iX t )
211"(1 + x) log z
.
(B + 1.C)Z
269
+ rpo(z),
provided the origin of coordinates is taken outside R, that is, within one
of the contours Ck. The Xl and X. are the components of the resultant
vector of all external forces acting on the boundary C l + ... + Cm, so
that
m
Xl
+ iX.
(X\k)
+ iX~';
k-l
(72.9)
2B
+ B'
= T'2( <Xl),
where Ta~( <Xl) represents the limiting value of Ta~(X) as the point x recedes
to infinity. The constant C has no effect on the state of stress and is
related to the rigid rotation", "" lim H(U2.1 - u, .) at infinity by the
1-1--
formula
C = 1
2"
+ K""
..
".(.)
-,,-0" ~,
L., .'
>/I.(z) =
L-~.
.. -0
270
U. 1
+ iU. t
where
It + i/t
+ i!.(s) + cOnSt
= ft(8}
= i
[TJ(s)
on C,
+ iT.(s)] ds.
But if we recall the formula (71.2), we can write the condition (73.1) as
(73.2)
on C.
",(z) + z7{z} + "f(z'j = fl + if. + const
The constant in the right-hand member of this formula is, in general,
complex and has a different value on each of the contours forming the
boundary C. However, we saw in the preceding section that there is some
freedom in the choice of ",(z) and >/I(z) corresponding to the same state of
stress. This freedom can be utilized to fix the values of some constants
in (73.2).
Thus, if the region R is finite and simply connected, the replacement of
'" by '" + ciz + a and of >/I by >/I + fJ does not change the state of stress in
R. Accordingly, the left-hand member in (73.2) can be replaced by
",(z) + z9lfz) + "f(z'j + a + p, and hence, by choosing suitably the value
of a + p, the constant in (73.2) can be fixed in an arbitrary manner. If,
for example, we set this constant equal to zero, we can no longer take l
~) = 0,1",'(0) == 0, and >/1(0) = 0, but we can still choose ",(0) = 0 and
11",'(0) = O. The condition ",(0) = 0 fixes a, and g",'(O) = 0 determines
c. But if a is known, the value of a + P fixing the constant in (73.2)
determines If and hence we no longer have control over the choice of p.
The situation with an infinite region bounded by one contour C is
similar. In this case '" and >/I have the structures shown in (72.8). If the
constant in (73.2) is fixed, we can consider ",o( <I = 0, C ... O. This
choice, together with the choice of the COnstant in (73.2), determines "'.
and >/1o completely.
If the region is multiply eonnected (finite or infinite), the constant in
(73.2) may have a different value on each contour C. forming the boundary of R and only on one of these contours can it be fixed arbitrarily. On
the remaining contours the integration constants are determined from
the requirement of continuity and 8inil~uedness of displacements and
stresses. ,
The determination of the corresponding. boundary conditions in the
1
See {7:l.6}.
See an analog()t18 problem in Bee. 4.T.
271
This time "e have the
where the g,,(8) are known functions. We recall t.he formula (71.7) and
deduce at once the boundary condition
(73.3)
1<!O(Z) - z!O'(z) -
'f(Z'j = 2,.[g,(S)
+ ig.(s)
on C.
+ T2.
T11
(T .. -
+ 2iT12)e''',
where 01 is the angle measured from the positive directi~n of the xl-axis
to the normal ...
The substitution in (73.4) from (71.4) yields the desired boltndary conI See (72.5).
This fonnula is easily cheeked by using the transformation fonnula; (16.4) upon
taking the direction of N along the Z:-axis and that of T along the x':-axis. Then
N - ,.~., T - ..:.. and the transfonnations connecting the coordfuate sY~tem8 are:
a:. ..
Zl CO$
,
Q: -
2;,
am. a,
a:.-z:sinQ+~~C08..
1'11
+ TO,
{Tn - 1"11
+ 2trll)s....
T_
is an invariant.
272
+~ -
es;..{!<p,,(z)
+ 1//(zl] =
N - iT
on C,
I = It + il. + const,
while, in the second, a = - 1 { and 1= -21'(g, + ig.).
The existence of solutions of the first and second boundary-value problems in plane elasticity can thus be made to depend on the demonstration
of the existence of functions ",,(z) and I/t(z) which satisfy on the boundary
C the conditions of the form (74.1). We shall see in Secs. 83 and 86 that
the boundary conditions (74.1) can be used to construct a Fredholm
integral equation of the second kind for the determination of these functions, and their existence would then follow directly from the fact that the
associated homogeneous integral equation has no solution other than the
trivial (zero) solution.'
As regards the uniqueness of solution, it should be noted that the
Kirchhoff proof (Sec. 27), for finite three-dimensional domains, is clearly
valid for finite two-dimensional regions. Instead of formula (27.1) we
now have the equation
(74.2)
1 The first boundary-value problem, as was shown in Sec. 69, is equivalent to the
fundamental biharmonic boundary-value problem. The existence of the solution of
it, for finite simply connected domains, was established by: J. Hadamard [Memoir68
468 llllilnla eJ.ranger8, vol. 33, No.4 (1908)], G. Lauricella (Alti della reals accademio
fIIUionale dei Lincei, vol. 15 (1906), pp. 426-432), T. Boggio (Alti della aceodemia
~ di Torino, vol. 35 (1900), pp. 219-239; Atti del reals institute Veneto di
~, latIN. ed arti, vol. 61 (1901-1902), pp. 619-636]. and A.. Korn (Annal68 de
t'lools I'IONIIiJle BUpbieuN. vol. 25 (1908), pp. 529--583).
The matter of the existence of solutions of the first, second, alia mi:!ced problema for
finite and infinite multiply connected domains (including anisotropic media) was
eettled principally by: N. I. Muakhelishvili [Malhem4li8che Annahn, vol. 107 (1932),
pp. 282-3121, S. G. Mikhlin IMtJtematichuki Sbornik, vol. 41 (1934), pp. 284-291, ~
420; TnMlt/ &i&mological InstUule Akademii Nauk SBSR, No. 65 (1935), No. 66
(1935), No. 76 (1936)1, and D. I. Sherman (Trudy &iBmological lnetitute Akadilmii
Nouk SS8ll, No. 54 (1935), No. 86 (1938), No. 88 (1938), No. 100 (1940); DlJlclady
Ak<Id.emii Nouk SSSR, vol. ?:1 (1940), pp. 911-1113; vol. 28 (1940), pp. 29-32; vol. 32
(1941), pp. 31H15; PriId. Mal. Mula., AkademiJIG Nauk SSSR, vol. 7 (1943), pp.
34J~, 4lH2O}.
aeu.
where
w=
3-2).(e11
273
= C1
+ ... + C...
and by the circle C. with center at III = 0 and with radius p so large that
Cp contains C within it. Then if the integral
as
(74.3)
p~
J,
the argument in Sec. 27 establishes the uniqueness of Solufi?n in the twodimensional infinite region. The fact that (74.3) is, ind, true follows
from (71.4) and (71.7) if we recall' that, for sufficiently large izi,
For, in this case, the integrand Taua is at least of the order lip', and hence
the integral tends to zero as p becomes infinite.
76. The Role of Conformal Representation in Plane Problems of
Elasticity. We have indicated in the preceding chapter how effectively
conformal mapping can be used in solving the Dirichlet problem for
simply connected domains. Techniques, similar to those used in calculating the complex torsion and flexure functions, can be applied to the
boundary-value problems in plane elasticity. We suppose that the given
region R (finite or infinite) is simply connected and map it conformally
on the unit circle Irl :$ 1 by the analytic function
(75.1)
III
= w(r).
Z ...
",(l") =
.
l
s",
k ..
lsI
:$ 1,
.. -1
See (72.11), where we have set X, _ X. - B - C - B' - C' - 0, since we supthat the Uiapl.&eementa remain bounded at infinity.
This is a apeeiaI. case of a theorem due to V. Smimoft', M ~ Annalen,
vol. 107 (1932), pp. 313-323.
.
l
]lOBe
274
"'(.I) =
.
f + l k"r",
1.11 ~ 1,
.-0
Since
'( )
I" z
dil'l dr
= dr dz
'()
1"1 .I w'(r) ,
lJ,1
+ ill"
21'(Ul
= 1"1(.1)
+ iU2)
+ ~ 1";(.1) + "'I(n,
1.11 ~ 1,
w \il
= 1<1",(.1) -
~ 1";(.1)
w \.1)
- ",,(.I),
1.11 ~ 1.
1",(.1)
+ ~ 1";(.1) + ",,(n
= F(.J),
on 1.11 = 1,
on 1.11 = 1,
'" \.1)
(75.8)
1<1",(.1) -
w(n
=
w \.1)
-rr.:\
....-T.:\
where F(") and G(") are uniquely determined by (75.1) on the boundary
of the unit circle from known values, I, + i/2 + const and 21'(U' + ig2),
specified' in the contour C of R.
The structure of the left-hand members in (75.7) and (75.8) suggests
that we impose on Fl(.r), F:<r), and .pdi) the requirement of continuity
in the closed region Jil ~ 1. Moreover, if the domain is bounded, I"I(r)
'Y
1 Occasionally it proves oonvenient to map an infinite region R on'the region 1.1 <:: 1
and make the point at infinity in the z-plane correspond to the point - .. . The
appropriate mapping function is obtained then from (75.3) by repla.cing by llr.
We 8UppnBe that the value of the integration conatant in (73.2) is fixed in some
definite way, aay, by setting it equal to zero. This can be done by utilizing the available freedom in the choice of fez). The transform of !t(8) + il.(8), which we denoted
by F("}, is then a known function 1.(6) + il.(") of the angular variable" in the
.-plane. The functional forme of 1M) + if.(") will, in general, differ from /t(.)
i/.(s), but the values of these functions at the corresponding pointe on ')' and (J are
the_.
275
and ",(!) are analytic in the region Irl < 1. In the first boundary-value
problem for such domains, we are free to assign arbitrary values to ,,(0)
and to 11[,,'(O)J. Hence ' the values of "1(0) and s["i(O)/w'(O)J can be
specified arbitrarily.
In the second boundary-value problem we can assign the value to
either ,,(z) or ,,(z) at some point of R, and in the sequel we shall choose
to assign an arbitrary vah"" to ",(r) at some point r = ro of the transformed region.
If R. is a bounded region, ",(r) and ",(!) have the representations
.,
.,
iCI(r)
/'
aftr",
'her)
b..t",
ItI ~
1.
n-O
71-0
----:0::+-1- - - , . l : l
FIG. 54
276
It is not difficult to express e-i in terms of the mapping function z ... (aI(l).
For, if df represents a displacement of .I == peif along the radius, the
eorresponding displacement dz in the z-plane will be along the line
iJ == const. Hence
dz = e'"
Idzl
and
and we find
(75.9)
e-> ==
f;rm
p1",'(.1)1"
Up
. == f;rm(
+ tu"
pIw' (n I Ul +.)
tu .
4>(0')
+ 4>(0')
~ ~4>'(O') + ",'(O')v(O')]
'" (0')
== N - iT
on 1.11 == 1,
where
fI(.I) == ,y; (.I)
(ai' (.I) ,
If we let Th == 1'.., 1';. = Til, 1';. == Tp~ in the formulas in the footnote
on page 271 and recall formulas (71.4), we get the useful expressions,l
Tpp
(75.12)
1'"" -
Tpp
+ T~"
== 2[4>(.1)
+ iW],
+ 2&", ==.plw~
Mn4>'(.I) + (aI'(.I)v(.I)]
\.1)
+=
where IT ... ell is the value of .I on the boundary of the unit circle. In the
first boundary-value problem a == 1 and HelT) "'" ft(iJ)
i!.(iJ), and in
the second problem a == -7t: and H(IT) == -2I'[gl(iJ)
ig.(t1)1. If the
region R is finite and simply connected, the functions 'l(f) and '{tl(r) can
be represented in the power series
+
+
277
(76.2)
lI'l(r)
U.re,
~l(t) =
k-O
bktk,
k-O
(76.3)
H(a)
);=_
Cke"'~'"
Coer",
GO
and write the complex Fourier series for the known function
(76.4)
The insertion from (76.2), (76.3), and (76.4) in (76.1) yields the equation
110
(76.5)
00
a.uk
1:=1
00
coer"
k=-110
GO
kU.u-H1
k-l
IlO
b.a- =
k=O
C.cr",
k=~oo
CIa
aoer"
1:-1
k-l
00
CIQ
CIO
l (L mamCmH-I) cr" + l (l
",-1
k-O
m-l
mamCm_;_I) a-'
b.u-k =
k-O
L Coer",
k __ oo
(76.6)
aUk
l
l
mUmCm+k_1
= Ck,
(k = 1,2, . . . ),
",-1
e
(76.7)
6.
ma..c.,_;-1 = C_;,
(k = 0, 1, 2, ... ).
",-1
If the system of Eqs. (76.6) can be solved for the ak, the b. are determined at once from formula (76.7). In the first boundary-value problem
the system (76.6) cannot be expected to yield a unique solution if the
imaginary part of at is left unspecified, since the function 't(r) is not
determined uniquely unless the value of g[,~(O)/c.l(O)J "" g[at/",'(O)J is
assigned. No such supplementary condition is needed for the second
boundary-value problem, inasmuch as the condition !"1(0) = 0 completely determines both lI'l(t) and tIm
278
w'(tT)
2:
n
-- =
(76.8)
ckuk
k-O
2:
k
C_kU-
k-1
+ il,(8)
fe
= i
(T,
1.'..
+ 1 in
+ iT,) do
(T,
= O.
+ iT,) do,
and hence the increment in I, + if. as the contour C is traversed is zero. This is
another way of qaying that H(tf) = /1(r'J) + ij.(r'J) is single-valued on 1,1 - 1. The
vanishing of the resultant moment requires that
fe (x,T. - x,T,) do
fe (x, d/,
+ x, d/t)
= O.
+ X".(8)]C
fe [/,(8) dx,
+ 1,(8) dx,]
= 0,
and since the function in the brackets is single valued, the bracketed term vanishes.
fe [/,(8)
+ if,(8)] az =
!lEI
+ if.(8)] az -
<ll
G\
fe [/'(8)
f,(r'J)
O.
",'(,) d"
f. H(u);;T(;)i'l::: - O.
The last of these equalities inlplies a restriction on the choice of the C~ in (76.3).
"The left-hand member of (76.8), viewed as .. function of .. complex variable ..,
has a pole of order n at inlinity and no other singularities in the region 1..1 2: 1. Hence
it has a Laurent series representation in the region 101 2: 1, whieh, for ",j - 1. has the
form (7G.8).
279
aliI
(76.9)
+ alC..
aa,.
.. +k+l
(76.10)
2:
bk = -
+ 1.
mdmCm-IH + C-k,
k ~ n
aak = Ck,
(k = 0, 1, 2, . . . ) .
.. -1
It follows from these formulas that, if ICkl < M/k', then the series
(76.2) define the analytic functions "",(I), ",,~(r), and ",,(I) in the region
1rl < 1, which satisfy the boundary condition (76.1). The Fourier
"", ()
r =
X,
2,..(1
(76.11)
",,(r) = -
where ""O(r) and ",O(r) are analytic and single-valued for'lrl < 1.
The constants B, B', and C' are related to the stress distribution at
infinity. They are,
B
= T11(OO)
+ T22(oo),
T11 ( 00 )
--~-2--'
C'
= T12( 00 ).
As noted in Sec. 72, the constant C can be set equal to zero. To obtain
the boundary conditions for ""O(r) and "'O(l) in the first boundary-value
problem, we substitute from (76.11) in (76.1) and find
(76.12)
".O(rr)
PO(u),
where
(76.13)
~ iX210g u
~c
280
of the resultant foree acting on C vanish. In this case we have for the
determination of <pO (i') and fO{S) the boundary condition similar to (76.1).
If the resultant external force does not vanish, F(u) acquires an increment
i(X I
iX 2) as the point u traverses the circle Ii'I = 1 once in the counterclockwise direction. I But as u traverses the circle, log u acquires an
increment 211'i and thus the contribution from the second term in the
riilht-hand member of (76.13) annuls the contribution from the first
tern. It follows that FO(u) is single-valued in this case also. Thus, in
either circumstance, we have a problem of the same type as for the finite
domain.
The treatment of the second boundary-value problem for an infinite
domain v!hen the values of Xl, X 2, B" C" B', and C' are specified in
advance obviously leads to the boundary condition
(76.14)
= GO(u),
where <pm and fO(S) are single-valued and analytic functions for Ii'I
and GO(u) is a known single-valued function.
<1
PROBLEM
Show that whenever the function "(il, mapping an infinite simply connected domain
on a unit circle 1,1 - 1, has the form
n
.,(il -
f + L -r..r,
i-I
then
..
L aor.,
from
k-l
(16.12), is:
a,
a.
(16.15)
a_I
+ 41Cft._1
a. - A:'
- 2)a._tC._. = A~,
- 3)4._ae._. = A~,
k ~ n - 1,
..
l
k--.
A:".
+ ./.(11)
(T,
+ iT.) tis,
281
(77.1)
sothat"'(cr)~
Izl
R.
Rr,
ti~n c, = 1, vanish.
If we represent the function
+ if.(tJ),
f,(")
characterizin~
the stress
F(,,) =
(77.2)
A~,
k,--" .,
where
k = 0, 1, 2, .. ,
a, + a, = A,
at = At,
b. =
.iLt
(k
IX
= 1, yield:
k;':: 2,
k = 0, 1,2, .
+ 2)Ak+2,
..
(77.4)
2: a.r i
..
.
"'I(!) == L b.r = L
"",(r) ""
'-1
and
(77.5)
1-0
..
L Atrt ,
r+
k-2
[A-k -
(k
+ 2)A t +2Jrt
1-0
+ iu,) = e-i8[x",,(z)
+ 'Til = 4<R[",,'(z)],
21'(ur
(77.6)
'Trr
- z",,'(z) - ~J
..
0(") =
L B~
t - - .,
282
..
",at 41 = B l ,
",a, == B"
5. == - R... - (k
-It
k> 1,
k ~ O.
+ 2)4HI,
+ if2(8)
= i
(T1
+ iT.) dB
- i JD PeiDR d8
-PRe".
Thus,
and hence all
Al = -PR,
vanish.
The substitution of A., = -PR and Ak = 0, k
(77.5) then gives
I, in (77.4) and
,her) = 0,
so that
<p(z)
pz
= - 2'
!/I(z)
O.
P(l - x)
4",
= Tee =
r,
-P,
Us
= 0,
TrS
= O.
-UQ
cos 8,
and we find
2",(g1
+ ig,)
... -2I'UetT.
B. == 0,
k"" 1.
283
Thus,
all = 0, Ie;;if: 1,
and, hence,
'h(1") = O.
Using
I'(z)
2J,1u o
- x-I
"'(21) = 0
R'
ur=-l['
c. Concentrated Loads.
U,
= O.
(0, - P) act at the point 210 = Re' of the circular boundary and an equal
and opposite force with components (0, PA_act at zo = Re-'<l (Fig. 55).
IO,-Pj
IO,P)
FIG. 55
+ if.{s}
= i
(Tl
+ iT.) ds
will be Constant &long the part of the boundary where no load is applied,
but as the variable point passes the point of application of the load,
h + ifa suffers a discontinuity of magnitude P.
284
In our case,
/1 + if,
F(II) ""
a<8<21<-a,
< a,
~ 8
= 0,
=P,
~ 0,
Hence,
1
A. = 21<
2,.. -
IX
<8~
ikl
211' "
211".
dB
k = 0,
Pi(1k
=21Tk
e a-e-u)
",
,c O.
.-1
rk
Pi
-
(e a - e-ia)r ,
where
Since for
Ixl
xjk
.<::: 1,
log (1 - x) and
k-l
.-0
the formulas for V'1(!) and .,h(!) can be written in closed forJIlS:
V'l(t)
oa
(
e - I:
e" - e- )
+ Pi
2,.. log e-ia _ I: - --2--- i: ,
pa + bo + ~ (log e~ - i: + .ez,..
e---i: ea-i:
pa
= -;
h(S) =
ia
1(
Zo
_j:__).
e-,"-i:
Pi (
Zo
= z;:
log fa
tf;(z) =
- Zo )
ZZ - Zo
----zR2
z ,
21r
where we dropped the nonessential constants that do not affect the stress
distribution.
The computation of stresses and displacements presents no serious
difficulties. 1
1 These are recorded in N. I. Muskhelishvili's Some Basic Problems of the Mathematical Theory of Elasticity (l953J. pp. 327-328, where the functions '('(z) and I/t(.)
are obtained in a different way. See also TimOllhenko And Goodier, 1'beory of Elas-
285
r~~
o.
Ta -
=0 -
""
0 and the
T~OJlI{J
(r~Ol
Ta
(l)
Thus,
T,
+ iT,
-[rIT cos
Substituting for r~J from (68.3) and noting that, on the boundary of the
circle of radius R,
X, = R sin 8,
XI =. R cos 8,
we find,
T1
Hence
fI
+ tf 2 =
+ t'T" --
j'
2>'
4(>.
T t'T ,
tI
+
+ 3"
2,,) pw 'R'"
e' .
d8
2>'
4(X
+
3" pw"R'"
+ 2,,)
e' .
It is clear from this that the problem of determining the stress distribution r~'J is identical with the uniform-pressure problem considered in (a)
2>'
a b ove, wh ele we must set P = - 4(>.
+ 3"
+
2,,) pw "R"
ticity (1951), pp. 107-111, where this problem is solved by indirect means. This
problem was originally treated by H. Hertz, Zeitschrift fur Mathematik und Physik,
vol. 28 (1883), and later by J. H. Michell, Proceedings of the London Mathematical
Society, vol. 32 (1900), pp. 35-61, vol. 34 (1902), pp. 134 142, who solved severa)
similar problems by ingenious devices. A unified and systematic trep.tment of this
category of problems was first given by G. V. Kolosoff and N. I. Muskhelishvili in
Itvestiya Petrograd Electrotechnical Institute, vol. 12 (1915), pp. 39-55 (in Russian).
1 For different solutions of this problem see Love's Treatise, Sec. 102, and Timoshenko and Goodier's Theory of Elasticity, Sees. 30 and 119. The problem of the
disk rotating about an axis normal to the disk at an arbitrary point of the disk was
solved by Ya. K. I1'yn, Doklady Akademii Nauk SSSR, vol. 67 (1949), pp. 803--806
(in Russian). A solution of the problem of rotating disk with att!).ched concentrated
m _ is outlined in Sec. 80 of Muskhelishvili's Some Ba.sic Problems of the Mathematic!).l Theory of Elp-sticity (1953).
z = we!) =
R
T
the functions 1I',(r) "" lI'[w(r)] and "',(!) "" ,p[w(r)] assume the forms [see
(76.11)],
If
+ (B + iC) + 1I'(r),
",SX, - iX.) log r + (B' + iC')!!_ + ",O(r).
211'(1 + x)
.I
(78.2)
,p,(r)
(78.3)
For the determination of the analytic functions 11'0(.1) and ",oCI) we thus
have the boundary condition
1
(78.4)
11'0(0') - -11'0'(0')
0"
+ "'0(0')
= FO(O') ,
where
78. 5)
(
BR
F O()
0'= F()
0'- Xl +
211' iX, IogO'---;-
+ fa [;:"(1-;:) a -
BRu 2
J-
(B' - iC')Ra,
and F(u) '"" f,(O) + i}2(O), determined by the specified stress distribution
on the circular boundary.
Setting
.yom
Lbk5\
A-O
in (78.4) a.nd writing in the right-hand member the Fourier series representation for the single-valued function,
287
we obtain
..
(78.6)
..
a~ -
k-1
-
aD
kii.".-k-!
k-l
k-O
X,
+ iX. [ . + ~L. k1 (-k
-~ 1I't
q
-
..
fj.".-k
uk)]
A"uk
k--"
-
2BR
-q-
(B'
t'C')Rq
k-l
Xl - iX.
+ 2... (1 + x) U-
The comparison of like powers of u then yields:
al = Al
ak=k
0=
(78.7)
61
= A_1
Xl
iX
- 2BR,
iX.!
b-2 -- A -2 _ Xl +
211'
2
-k
A
(
)b =
-k -
k - 2
+ X, -
ak_2 -
XI
iX.
+ x)'
+
iX 1
211'
ii'
211'(1
3.
f.
+ if.
+i
-i
(Tl
+ iT.) ds
Thus,
F(u)
= -PRq-'.
1 The negative sign is introduced in the integral because the positive direction of
integration along the circle is clockwise, inasmuch as the normal v is directed toward
its cent-er.
288
"1m
O(n,
A_, = -PR,
vanish in tne expansion for F(O'), we conclude from (78.7) that
at
b,
= 0,
k
= - P R,
1,
bk
0,
k",cl.
Hence
1/11 = -PRj,
and, therefore,
Hz)
.. (z) '" 0,
PR2
Z
-T"
= - 7'
'Tr 8 =-
0,
T2=~'
2'11'R
(78.8)
acting on the boundary of the circle JzJ = R. The resultant force produced by the stress distribution T, + iT, is, clearly, X, + iX 2. If we
assume that the stresses at infinity vanish, FO(u) defined by (78.5)
becomes,
iX, I
_!_ Xl - iX,.
(78.9)
F O(".) = F(U ) _ X, +
2'11'
og U + u' 2'11'(1 + x)
But
;~
J1 + ~n
80
that
'J'(T
'T)d 8 =
= t
I l .
i(Xl
iX,) r'J
F()
". =
211'
X,
!!5
2
.X1 +iX
21r
-l
..
v,
211'
Inserting this in (78.~j ..-e see that the right-hand member in (78.6)
reduces to the single term
X, - iX 2 -2
F O(" ) -_ 211-(1
+ x) 0'
289
= 0,
Xl
> 0,
+ iX.
= 211"(1
k F- 2,
+ x)'
and hence
",O(r)
""O(r) = 0,
++iX.x) s' .
= X,
211"(1
X,
(78.10)
+ iX. R
+ x) log ""i'
.1'(2)
..
-x(X , - iX.) 10 ~
2... (1 + x)
g z
(78.11)
Izl 2:
'( ) __ X, + iX.!
'" z 2... (1 + x) z'
""(z)
= x(X , - iX,)
2... (1
+ x)
+ X, + iX. R'.
211"(1 + x) z'
R is determined by (77.6),
"( ) _ X,
'I'
+ iX.!_
+ x) Z2'
2... (1
! _ X, + iX. R'.
z
...(1
+ x)
z'
(78.12)
"r"
orr'
= -
1C
+ 3 X, cos (J + X, sin (J
x - I X I cos
+1
21fr
(J
-,
+ X. sin (J
211'1'
'
x - I X I sin (J - X, cos
X + 1
211'1'
(J
M.
T 1 = - 2... R' sm
(J,
T,
M
2... R' cos 8,
290
Binoe
fl
+ if.
if'
(TI
+ iT.) ds = ~
2rR
f'
eil dB - -
Ali elf
2rR
'
Mi
F(u) = - 2rR u- I
Thus, the only nonvanishing At in (78.7) is A_I = - Mi/2rR.
stresses at infinity vanish, the system (78.7) yields,
ak = 0,
k = 1, 2, .
Mi
b, = 2rR'
inasmuch as Xl
+ iX
= O.
If the
.,
k"" 1,
bk = 0,
Hence,
and
<pCz) == 0,
"{I(z)
Mil
= 2r Z
'TIS
- 2,""2
-M,
= 0,
where M = -2rTR2.
d. Uniaxial and Biaxial Tension. Pure Shear. We consider next the
effect of the stress concentration in the neighborhood of the hole Izl = R,
located in a plane subjected to the action of constant loads at a great
distance from the hole.
Let us suppose first that the plane is stretched by the tensile forces
acting in the x,-<iirection. We take
.
'Tl2( 00) =
'Tu(
00)
o.
Since the hole is free of stress, XI + iX. = 0 and F(u) = O. The constants B, B', C' in (78.7) are determined by (78.3), and we find,
B =
~quations
!1.P1,
C' = O.
bo = 0,
Itt
b. = 0,
k> 3.
= _
P~R,
Ie> 1,
b: = 0,
291
Thus
(1 )
P,R (1f + r - r").
2P,R 2r +! '
= - -2-
Hence
<p(Z) =
",(z) =
~' (~ + ~}
- ~' [z +
R'G - ~82)J
U, =
(x - l)r'
PI [r2
41'r
+ R2(X _
1)
~2'J cos 2+
+r
R'] sin 28
2'
r<r
T66
T 8
= -
I
(78.13)
~~
~2
(1 + 2R'r' -
=P
~ [( 1 - -R') - ( 1 - 4R2
- r2
2
r2
T
1'1'
(78.14)
+ 3R')
-r cos 29 'J
4
. '
p. [( 1 + -;:0
R2) + ( 1 + 3R')
"2
--:;:< cos 29 ] ,
T"
Tr8
P. (
="2
1
+ 2R2
--;:0 -
3R<).
--;:< sm 26.
Trr
P(1 - ~!}
T"
P(1 + ~:}
Tr ,
O.
On the boundary of the hole in all these cases Trr =' T r 6 = 0, !J.I! it should,
but for the case of the uniaxial tension the hoop stress 1" is,
'Til
= P t (1 - 2 cos 29).
292
I
I.-R
>
o~
0,
1,
It I ~
I,
It I =
x, = R(1
+ m) cos D,
x.
~o(.,-)
+ wC"-)
",'(a)
rpOI(a)
+ 1/-0(a)
= FO(a).
293
In our problem,
so that the coefficients en, in the expansion (76.4), vanish for all n ;:: O.
It follows then from (76.6) that
k;:: 1.
so that
(79.5)
'l'0(S) =
.'l.r',
.=1
where
(79.6)
or
(79.7)
(I) -- 2n'
r
'I'
PO(u)
~ ueu _ r) dO'.
1"(0)
&
1"0'(0')
+ "'O(u)
= FO(o),
294
contour -y.
We get,
But,l
1
",o(u)
2ri i.,u _ t du = ~O(O)
0,
and we have,
(79.9)
Since in our problem
w(u)
u(u 2
w'(u) = - 1 -
+ m)
mIT! '
<
1,
we see that the first integral in the right-hand member of (79.9) can be
evaluated by Cauchy's Integral Formula to yield,
(79.10)
>f0(r) = r(i
+ m)
I-mr'
O'(r)
+ __!_.
FO(u) du.
211'1i.,u-t
where
PI = cos (Xl, V) = cos (X2, 8)
pa
Thus,
I!O
that
= cos (X2, v) = -
it
(TI
+ iT.) ds
:2,
cos (Xl, 8) = _
= -
~"
P dz = -Pe,
+ mIT),
295
~(r)
= -PR
[r + mr(r
+ m)],
1 - mrs
2
T22( 00)
= P sin' a,
+ iC'
~ e- ia
= -
and hence
FO(I) = _ PR
4
P
B =4'
'
The substitution of this in (79.7) and (79.10) yields, after simple calculations,
>O(t) = PRt (2e 2ia
4
PRt
.!.O(I-) _
'Y
Thus
p:r (2e
[2 _ 1 - e""'(r'
2(mt2 _ 1) m
2ia - m
",(I)
~(t)
= _ PR [e-2ia!
m),
+ m)].
+.p}
+ e2iar
m
_ (1
+ m2)(e
m
2ia
m) __
r_],
1 - mrs
1. Compute the displacements and stresses in the problem treated in the illustration
of Sec. 79a, for the case when m - o.
I. Solve the problem of deformation of an infinite plate with an elliptical hole. when
a constant tangential force acts on the boundary of the hole.
1 The solution of this problem was first obtained by C. E. Inglis, Transactions of 1M
l'Mtitute of Naoal Artihitec18. London, vol. 55 (1913). pp. 219-230. The solution given
FIG. 56
We consider
(80.1)
z = ",(I) = R
(I + T).
R > 0,
m ~ 0,
+ ix., find
Pl
x, = R
X2
(80.2)
p = p.
297
R>O
(80.3)
maps the entire z-plane, slit along the real axis between Xl =- -2ft and
= 2R, onto the region III 2 1. As the point I = e'l) traverses the
circle once, the corresponding point z traverses the slit twice, so that the
points tT = e'/I, and tT = e-ill , correspond to one and the same point Po
on the slit. The ring bounded by the circles P = Po > 1 and P = 1 then
corresponds to the interior of the ellipse Co, cut along the real axis between
the points (-2R, 0) and (2R, 0).
If either the displacements or the stresses are specified on the boundary Co of the uncut ellipse, the functions ~,(I) and ",(t) are determined by
the condition of the form
Xl
(80.4)
"''1'1(1)
w(t) -:rr.:\
-;:-r.:'\
+ wl(t)
~1\1) + >/11\1)
= H(I),
(80.5)
'l'1(t)
k--
akr,
>/1,(1) =
L M.
k--
110
they ca.n be
10
'l'1(tT) = '{>,(It) ,
1/I,(tT)
1/I,(It).
k = 0, 1,2, . . . .
298
cular ring. Although a doubly connected domain can be mapped conformally on a circular ring, a generalization of the formulas of Sec. 76 to
doublY" connected domains usually leads ~o intractable systems of equations for the coefficients in the series representations of '1'(21) and '{t(z).
The treatment of the first and second boundary-value problems for
the circular ring is identical, and we confine our discussion to the first
problem.
Let the ring be formed by" a pair of concentric circles C.., a = 1, 2, of
radii R", where R, < R.. To simplify" calculations, we shall suppose
that the external stresses applied to each boundary- are such that the
resultant force and moment vanish for each bound8.ry. In this event,
the logarithmic terms do not appear in the representations (72.7), and
the functions tp(z) and '{t(z) will be analytic in the ring R, < Izl < R .
Accordingly, we can write
LakZ",
~
(81.1)
'P(z) =
The coefficients
(81.2)
tp(z)
where
(81.3)
at
R, <
+ ztp'(z) + f(zj'
[T\")(s)
= f\")(s)
+ if~")(8) + const
on 0,.,
F.. (J) =
.
L
Ak")e
-
ik ',
and, recalling (81.1), we can write the boundary conditions (81.2) in the
form
(81.4)
..
.
...
L
akR!uk + R.." L GkkR:-'u-(k-ll + L6.R:U-ll '" LA~")uk,
-.
-..
-..-.
where u == eO'.
The system of equatiollS for the unknown coefficients <It a.nd b. is then
299
got by comparing the coefficients of like powers of fT in (81.4). The solution of this system presents no difficulties, and the resulting series (81.1)
can be easily shown to correspond to the desired solution if the derivative of F,,(fJ) is of bounded variation.'
We limit ourselves to the study of the case in which the boundaries of
the ring are subjected to constant pressures. 2
Let the pressures acting on contours Ca. be P". Then, on taking
account of the convention for the positive direction of the normal to Ca.,
we have,
TiO) = -Po cos fJ,
T~2) = -Po sin fJ,
T~') =< P, sin fJ.
T~') = P, cos II,
Thus,
F,(II)
==
(81.5)
F.(II) "" i
t
t
(Tl"
+ iT~l)
ds
= -i
(Ti2J
+ iT~2)
ds = -i
J' P,eiBR. de
=<
CIt
-P.R,e",
where the integration constant in F.(II) has been set equal to zero.
Substituting in the right-hand member of (81.4) from (81.5), we get,
on equating the coefficients of like powers of fT, the following systems:
+ 2Ria. + 50 = 0,
+ R.a, + fi_,Ri' = -P,R.,
akR~ + (2 - k)a2-kR~-k + 5_kR-;:k
ao + 2R1a. + bo = c"
a,R, + R,a, + b_,Rt ' = -P,R"
akRt + (2 - k)a._kRr-k + 5_kRt
ao
a,R.
(81.6)
= 0,
for k ;c 0, 1.
= 0,
for k ;c 0, 1.
_ RiRf(P, - P.)
-1 -
Ri - Ri
300
Thus
(81.7)
Using formulas (77.6) we find that,
P2R~
T..,
- P,Rf
Ri - Ri
_ P2R~ - P,Rf
(81.8)
T88 Tr8
Rf - Ri
P2
- Rl -
+P
2 -
P, RiRf
R~
ro'
P, RiRf
Rf - m----;T'
= O.
and we see that the radial stress Trr is compressive, while the hoop stress
It is interesting to note that (r88)", .. > P, regardless of
how thick the ring is.
The procedure indicated above, when applied to the problem of the
ring deformed by two oppositely directed concentrated forces on the
exterior boundary, yields very slowly converging series (81.1) when the
ring is narrow. The concentrated forces acting on the boundary C2 give
rise to the singularities in <p(z) and 1/I(z) , and it is advisable to modify the
problem by making use of the solution of the corresponding problem for
the solid disk found in Sec. 77c.
If the radius of the solid disk is R 2 , the concentrated forces produce in
it certain known shearing and normal stresses along the circle Izl = R,.
On subtracting the known solution of the problem of the solid disk of
radius R 2 , under the action of the same concentrated forces, from the
desired solution of the ring problem, one is led to consider the following
auxiliary problem: Find the state of stress in a ring whose exterior boundary jzl = R2 is free of stress and whose interior boundary Izl = R, is subjected to continuously distributed shearing and normal stresses equal
and QPposite to the stresses present in the solid disk along the circle
Izl = R I The superposition of the solution of this auxiliary problem
on the known solution of the problem for the solid disk yields the desired
solution.'
Te8 is tensile.
1 This familiar device has been used by S. Timoshenko and J. N. Goodier [Theory of
Elasticity (l!l51), pp. 116-123] and K. Wieghardt [Sitzungsberichte der Akademie der
Wiemchafte:t in Wien, vol. 124 (1915), p. 11191 to solve the problem in Fourier series.
However, the series converge slowly when R,jR, is near unity. An application of the
alternating method, discussed in Sec. 88, enabled M. Z. Narodetzkv [Izveatiya Akademii Na"k i:;SSR, Technical Series, No.1 (1948), pp. 7-18] to deduce a solution
that converg"" more rtlpidly. The stress distribution in a circular ring under the
action of twc "aual and oppositely directed concentrated forces applied at the neare&t
301
The function
r---1,
z=wW =ar
-1
a> 1.
maps the region between two eccentric circles onto a circular ring. The
reader may find it instructive to formulate the first boundary-value problem for the region bounded by two eccentric circles with the R,id of this
mapping function and deduce from the boundary conditions the appropriate systems of equations for the coefficients ak and bk in the expansions for <PI(r) and ",,(r). The solution of the resulting systems presents
difficulties, and it is simpler to treat the equilibrium problems for eccentric rings in bipolar coordinates. I
The function
R > 0,
> 0,
as we saw in Sec. 80, maps the region bounded by two confocal ellipses
onto a circular ring of radii P = pa, ex = 1, 2. If the external stresses
acting on the elliptical boundaries are such that the resultant force and
moment acting on each boundary vanish, the functions <p(z) and ",(z) will
be single-valued and analytic in the elliptical ring. Consequently their
points of the boundaries of the ring has been studied by D. V. Weinberg lPrikl. Mat.
Mekh., Akademiya Nauk SSSR, vol. 13 (1949), pp. 151-158J. By increasing the radii
of the circles, Weinberg deduces the known solution for an infinite strip subjected to
the action of two oppositely directed concentrated forces. See also L. N. G. Filon's
paper, entitled "The Stresses in a Circular Ring," Institution of Civil Engineers, London, Selected Engineering Papers, 12 (1924).
1 See G. B. Jeffery, Philosophical Transactions of the Royal Soeiety (London) (A),
vol. 221 (1921), pp. 265-293, and Ya. S. Uflyand, Bipolar Coordinates in the Theory
of Elasticity (1950), pp. 193-210 (in Russian). The equilibrium problems for a semiinfinite plane with a circular hole are also in this category. Bipolar coordinate. have
been used by Ya. S. Podstrigach, D"VOvidi Akademii Nauk Ukrain'sko'i RSR (1953),
pp. 456--460, to study the stress cc,Ltcentration in an infinite elastic plste weakened
by two unequal circular holes, when the boundary of each hole is subjected to uniform
pressures. The case of uniformly stretched plate weakened by two unequal circul8J>
holes is also considered in this paper.
As an illustration of the "alternating method," the equilibrium of an eccentric ring
is discussed in See. 88.
The state of stress in a heavy semi-infinite sheet with one circular hole was investigated by R D. Mindlin, "Stress Distribution around a Tunnel," Proceedings of the
American Soeiety of Civil Engineers, vol. 65 (1939), pp. 619....642.
Stress distribution in a heavy semi-infinite sheet with two circulsr holes was studied
in detail by D. I. Sherman, Prikl. Mat. Mekh., A1w.demiya Nauk SSSR, vol. 15 (1951),
pp. 297-316, 751-761.
An investigation of the stress concentration in a heavy semi-infinite sheet, near
arch-shaped and trape~oidal openings stiffened by absolutely rigid rings, was made by
I. S. Ham, DoptMdi Akademii Nauk Ukrmn'.koi RSR (1953), pp. 299-303.
302
transfofms
(81.9)
<Pl(t)
<Pl(.i) =
PI ~
IfI
~ PI,
~..air,
where 0'" = p"e'6 and the c" are constants. The functionsf,,(O',,) are completely determined from the assigned stresses on the boundaries C" of
the elliptical ring. If these functions are expa.nded in Fourier series and
the series (81.9) are inserted in (81.10), it is possible to write down the
system of equations for the determination of the Gk and bk Although
this procedure is quite straightforward in principle, the calculations are
quite involved and we shall not pursue them here.'
PROBLEMS
(~99) ....x
P,(Ri+Rl)
(R. + R,)t '
where t - R. - R., and that for thln rings (or long pipes)
(
) UlU =. P.R.
T69
-,-'
2. Consider the problem of the stress distribution in a hollow shaft, of inner
radius R, and outer radius R" rotating with constant angular velocity (d. Take
~Il - 1'l:j + 1';;6, where T~4 is given by (68.3). Show that the formula
003
yieldll
(T!"
1
+ iT.{UlIC. _ ~
-'ll', 4{>'+2,.)
2>' + 3,. ."
(T{l)
1
+ iTCl) c,
_ _
til' 2>.
4(>.
pw,
+ 3,.
"
+
2,.) e
on the boundary C 1.
Henoe conclude tb&t the solution of the equilibrium problem of rotating shaft with
free lateral surface is deducible from the results of Sec. 81. Show tb&t the maximum
! : ~:::
1 t-+1
at and'mtegrating the result over the contours
by 21I'i
Note
that the coefficients in the Laurent series (81.1) for ",(z) snd f(z) are given by
a. '" _2_
2ri
r ",(t) dt
JC
and
b. _
t,,+l
"''I'(er)
w(IT) - -+~
'I"(er) + >/t(er)
w (er)
H(er),
where H(IT) == h,(,'}) + ih,(,'}) is a single-valued function having continuous derivatives with respect to i} satisfying Holder's condition.'
The boundary condition (82.1) can be reduced to an integrodifferential equation for the determination of 'I'(l) and >/t(l) by a technique similar to that used in deriving Schwarz's formula in Sec. 42.
1 der
If we multiply both members of (82.1) by 2ri er _ l' where Irl < 1,
and integrate over 1', we get
(82.2)
with
1 We omit the subscript 1 and the superscript 0 on", and f in the formulas (76.11),
(76.12), (76.14) and in sll expressions of this snd the following three sections.
I See Sec. 40.
2n
~O'-f
a<p(l)
+ 2n
~
~O'-f
1
~
.ilO')
"'(0')
dO'
",1(0') 0' - r
+ ,yeO)
= A(l).
;~~ 1"'(0')
- a;;(u).
If we mUltiply this by 21. du ,.' integrate over ,/, and note that
nO'-,
305
w(r~
""'(0') ~
J~w'(u) 0' -
211"1
w(r)
= ",,'(0)
",'(0)
a",,(r)
+ __!_,
211"1
J~ w
= A(r),
where
k = ",,1(0).
w'(O)
(83.2)
(83.3)
",,(r) = - ;; w{l)
a""o(t)
+ J_
r w(u) -
we!")
+ ""o(!"),
On substituting (83.3) in (83.1)
",,~(u) dO'
+ 1f(0)
A(r).
a '(t)
<Po
+~
riat
211"1 J~
A'(t).
Since
lim w(u) - ~(t)
...... t
w'(t),
0'-
the kernel
K(u, t)
55
IS continuous for all 0' and t in the closed circle 'Y (except for 0' = 0, t = 0
in the case of the infinite domain) so long as the contour C is such that
"," (r) is continuous in ItI ~ 1.
Thus (83.5) is of the standard type.l
1 By separating (83.5) into real and imaginary parts, this equation can be reduced
to a pair of standard :real equations, but such reduction is not neceM&rY for our
\)IJl1)OIIe8
306
The existence of '" continUOll8 solution of (83.5) follows from the fact
that the related homogeneous eqUIIJtion [in which A'(t) ... 0] can bve no
solution other tbn the trivial solution 1P~(t) ... O. For the homogeneous
integral equation corresponds to the physical situation in which either
the displacements or the stresses vanish on the boundary 0, and the
assumption tbt a nonvanishing solution exists in such cases violates the
uniqueness theorem.
Let us suppose that by some means we have obtained a solution 'P~(t)
of (83.5). Inserting it in the integral of (83.4), we obtain 1P0(r) and fix
it so that "0(0) = 0 [see (82.3)]. We then construct lP(r), defined by
(83.3), and choose k in accord with (83.2).
From (83.3),
,,'(0) = -
~IX ",'(0)
+ IP~(O),
and hence
and, therefore,
(83.6)
+ ~ = ,,~(O).
IX
",/(0)
In the second boundary-value problem this equation completely determines k. In the first problem a = 1, and (83.6) demands that
/; + k =
,,~(O)
",'(0)
be real.
k + k = 2ill ,,/(0),
(T(O)
80
(83.7)
",m = fc + ",oen,
where "'o(r) is analytic in Irl < 1. It is easy to verify' that the function
11'01(.1), introduced in (76.H), satisfies the integral equation given in the
problem at the end of this section. Thus the problems for the simply
connected infinite domains differ in no essential particulars from the
problems for finite domains. This fact has already been noted in Sec. 76.
1 Detailed ealeulatiollS will be found in N. 1. Muakhelishvili's Some Basic Problema
of the Mathematical Theory of Elasticity (1953), Sec. 79.
307
K(rr, t)
.-1L
a.(t)b.(IT) ,
and hence Eq. (83.5) ia solvable in the closed form. This remarkable
result, first established by Muskhelishvili, can also be deduced in the
manner of Sec. 84, where two special forms of rational mapping functions are considered.
PROBLEM
Show that, if the domain is infinite and the mapping function has the form (83.7),
the function ",'(1) satisfies the equation
where
K(" I) = _1_~ [wo(O') -wo(I)].
,
w'(u)iJl.
0'-1
w(s)
'Yt!
'Yt ;;c 0,
'Yn ;;!.
0,
'=
11S
+ 1.s + .. + 1.s.
2
... + 'Y.s
.. + n1ns- +
+ 'YnS n... + n1n
n
(84.2)
308
Moreover, since ",'(I:) F- 0 for ItI S 1, ';"(I/t) 'F 0 for ItI ~ 1, and
thus (84.2) represents an analytic function for all Il"l ? 1, except for
t = QO, where it has a pole of order n. It follows, then, that for Il"l ~ 1,
(84.3)
",(t)
';"(l/r)
= cnr n +
C-kr
k-O
The fact that this expansion has a finite number of positive integral
powers of r will enable us to evaluate the integral in (82.3) in finite terms.
Since
",et) = all" + a.r' +
Il"l :$ 1,
nan.
(84.4) 'P'
Ih +
+ . + rn - I + .
Il"l ~ 1,
2;.
G)=
(!) -- K
~:AI t
';"(l/rl"
+ K n-I,>-n-l + ...
>-n
n,
L K~l
~
+ Klr + Ko +
...,
m-~
where
K"
K n-
(84.6)
== (heft.,
+ 2a.c.,
== litc. + 2a.c. + ... + (n - l)an_lCn,
= alcl + 2d.c. + . . . + nanc.
iitCn_1
~ "'(0-) ",'(0-) du =
2," ~ ",/(U) U -::- r
.. -0
r-,
K ..
..
since
f . L K_,..r-
1
2ri.,
1m
(84.7)
+ ~
a<p(r)
-1
u-r
U=.
'/:'0
K..r"
+~
_!_
B(IT) d#
2ri .,IT -
r. '
309
&8
_1_ = .!
u - t
u
1
2ri
(84.8)
+ i + . . . + u+
J:._ + -...
J,
US
.,
where
1
C~ "" 2ri
H(u)
~
u _ ,. du = L. C~t",
k-O
k = 0, 1, . . .
The substitution of the power series (84.8) in (84.7) and the comparison
of coefficients of like powers of t then give,
(84.9)
f(O) + Ko = Co,
aa.. + K .. = Cm ,
;,:
la",
= 1,2, . . . ,n,
m> n.
e".,
~~' ~. ~'C~
aa.
-: :-
+ ale.
~~n~". ~ ~~'
= Cn ,
ItI ::;;
1,
and hence
L.
m-Q
0-,...
A."
w(I/t) '(,.) _ ~
",'(t) ""
L.
,..-1
g ~
.
t(O =
~i uH~u~ tJg -
310
If the domain is infinite and the mapping tunction has the form
(84.12)
wet) ==
~ + k-O 'Ykrt,
and for
Irl ==
I c-nr",
.-0
~
~ _
&'(l/n -
,..n-2+
-i- cil
n
C -2>
1 we find
,,' (IT)
W'W(IT)
(IT) '-0
K(O) .".,-.
+ KjO)1T + K'o) +
n-T'
K'!!;"IT- m ,
",-1
where
K~O_:2 = alC n_2,
(84.13)
K~.'.3
= alCn _3
KiO)
aiel
+ 2a"cn-2,
+ 2a,c. + . . . + (n
- 2)a n-"cn_2,
"<Po (i)
-2
K::)r
+~
(Fo(lT)
= Zri J~ IT _ i dlT,
m-O
+ "'(0)
IT
211'Jo
and the values of the remaining K~) are determined from (84.13) upon
solving for the a.. the system of n - 2 equations
"al
(84.15)
"a,
2)an_,cn-' =
AY,
3)a._scn-. =- A\,
with
k = 1, . . . ,n - 2.
311
The ideas leading to the calculation of <p(t) [or <po(t)], in principle, are
identical tE> those of Sec. 76, but generally formulas given in this section
enable one to compute <p(t) and Ht) with less effort.
Similar results can be obtained by considering w(t) in the form of the
quotient of two polynomials.
85. llinstrative Examples. For comparison purposes we apply the
formulas of Sec. 84 to problems which have already been solved by the
series method in Secs. 77 to 79.
When the region is a circle of radius R, the mapping function is
= w(l') = Rl',
w(l)
w'(lIll
Thus all Ck, with the exception of CI
(84.6) that
K .. = 0,
I.
m = 2, . . .
,n.
a<p(l) = -1.
2".t
---
,u-l'
+ ",(0)].
Ko
+ -",(0)
Co
==
1 iH(u)
- - du,
2-'
7rt
>
1,
(85.3)
The value of
(84.10), is determined by
(85.4)
inasmuch as c, = I, Ck = 0, k > 1.
In the first boundary-value problem a = I, and the imaginary part of
a, can be set .,qual to zero. Equation (85.4) then yields
a,
+ a, '= 2a,
= Qi = _!__
2
41ri
= C,
1
>
H(u)
,,'
.L
au.
312
The substitution in (85.1) then yields .p(r), for the first boundaryvalue problem, in the form
(85.5)
(0
.p
..!_, (
211'1 }~IT -
j..,
fT
tIter) =
(85.6)
~ iITH~~ du -
'P0m
IT -
_..
If we insert
"'(IT)
",'(tT)
rFO(fT~ duo
}ytT -
+ ma-'
= tT(m
- fT')
0(1)
1
ma-~ 'Pi)I(tT) dq
2ri}ytT(m - tT')fT - r
+ _!_.
+ "'0(0)
_!_.J,
FO(IT) duo
211'1 ytT - r
Since,
lui ;:::
1,
But ,.,(0)
~i:::-~du -
"'(0).
0, so tha.t
313
and, hence,
(85.9)
The calculation of
(85.10)
.;O(t) =
_!_,
2ri
FO(u) du
"yU -
+ r(r'
+ m) 'l'0'(t)
1 - mr t
'I'~(t)
.! A'(t).
'"
+ p,
314
tage of being dependent on the solution of an auxiliary Dirichlet's problem for Green's function. It is clearly desirable to formulate the relevant equations so that they depend only on the assigned boundary values.
The fs.ct that this can be done was demonstrated by Lauricella l in a
rather involved paper concerned with the integration of the equilibrium
equations for the clamped elastic plate. This particular problem, as we
have already observed in Sec. 69, is closely related to the first boundaryvalue problem in plane elasticity. The Lauricella equations have been
rediscovered by Sherman, 2 who deduced them in a very simple way and
made use of them in solving the standard boundary-value problems, and
certain important new types, in plane elasticity.
A detailed account of Sherman's work would consume more space than
we have at our disposal, and we give only a sketch of the essential ideas.
We recall that in a finite simply connected domain ",,(z) and I/I(z) are analytic in the interior, and on the boundary C they satisfy the condition,
(86.1)
a",,(t)
= J(t)
on C.
2" Jes - z
(86.2)
!{t(z)
r
2"}es-z
= ~
'
W(S) ds _
w(s)
ds ,
c (--)2
s- z
315
(86.3)
'"
But on integration by
r W'(S) ds.
2n}os-z
()t = wet)
2
W(8) d
2ri } 0 s _ t 8,
+~.
1{;(t) = aW{t)
(86.4)
r w(s)
ds _ tw'(t) _
21N}os-t
'(t) = w'(t)
<p
+ _!_
+ _!_
r w'(s) d
2ri } 0 s _ t 8.
r sw'(s) ds
_!_.
2n}os-t'
a
+ -2'
w(s)d
rto
(8-t)
log --_ + 11=(8-t) ds =
8-t
no
s-t
-2'
w (s)
-_- _
f(t).
aw(t)
af
+ 2-'
w(s)d
n
0
(8-t)
log -_-_ 8-t
11
2-'
n
-t
W(sjd 8
-_-_
8-t
= f(t).
If we set
8 -
t = re"
aw(t)
+!r }o [aWes)
ag(t)
-! }o
r [P(s) sin 20 -
(86.7)
'If
+ cos 29)] dO
= f,(t),
=
/2(e),
Where II
ifI .,. I
The simultaneous integral equations (86.7) are of the Fredholm type.
and by a familiar device they can be reduced to a single real Fredholm's
equation.
816
if1 Z -
Zj
lie within the interior contours Cj The constants b; are then defined
so that the resulting equation for w(t) is free of unknown constants. 3
As a simple illustration of the use of Eq. (86.6) conFider the determination of 'P(z) and "'(z) for the problem of the solid disk of radius R comprp,ssed by a uniform pressure P on its boundary.
Zj
1 For example, if we take n points s., St, , 8. on the boundary C and apply
some formula of mechanical quadrature. to the integrals in (86.7), we get a system of
2n algebraic equations in 2n unknowns,
"p,
+~
[PI(a -
cos 28'1) -
ql
+ (t,),
;-1
n
aq, -
[Pi
- I.(t,)
+ ..(1,),
;-1
where the '" are the errors made in the process of replacing integrals by finite terms.
The solution of this system would enable us to compute pet) and get) approximately.
The functions <I'(z) and ,,(z) can be regarded a8 single-valued in all cases, since
the multiple-valued terms in (72.7) can be incorporated in I(t).
I An account of this is contained in the first two of Sherman's Doklady papers, cited
on p. 314. These papers are reproduced practically without clu1nge in Muskhelishvili'. hook Some Basic Problems of the Mathematical Theory of Elasticity (1953),
pp.412-420.
An illustration of the use of F.q. (86.5) in the solution of the first boundary-value
problem for the interior of the region bounded by an ellipse is contained in D. I.
Sherman's paper in Doklady Alcademii Nauk 88SR, voL 31 (1941), pp. 301HJ10, and
in S. O. Mikhlin, Integra! Equations (1949), pp. 292-294. For applications of the
Sherman method to doubJy and triply connected domains Bee D. I. Sherman, "On
the Stresses in a Heavy Balf-plane Weakened by Two Circular Openings," Pri/r;l. Mat.
Mekk., Akademiya Nauk SSSR, vol. 15 (1951), pp. 297-316, 751-762; "On the State
of Stress in Some Shrink-fitted Members," r-.tiya Akademii Nauk 888R, 'fechniea.l
Series (1948), pp. 1371-1388. See a.lso M. F. Gur'ev, "Distnbution of Stresses in a
Stretched Isotropio Rectangular Plate Weakened by a Circular Hole," Dopavi.di
Akademii Na ..k Ukrdin'lIkdi MIt (1953), pp. 133-139.
317
fn this ease, as shown in Sec. 77a, f(t) = -Pt, so that we seek the
solution of
wet)
(86.8)
Since 8
+!.r Jc
( [WeB)
and hence
e'" = e"(e,(I-,.) - 1).
2 sin ~(Ii - 80 )
Thus, as the point P describes the contour C, cp varies between the limits
FlO. 57
lio
7r/2 and 80
311/2. The structure of the right.-hand member in
(86.8) suggests that we seek the solution in the form wet) = at + b,
where a and b are constants.
The substitution of the assumed solution in (86.8) yields [since
W(8) = wet + rei,,) = at + are'" + b]
at
11"+(3r!2)
+ b + -r
'.+(r!2)
(at
+ are''' + b - ate'" -
P8
2"'
Thus
318
+ uX,
:'.; in other problems,
L(tp)
L(tp) = 0
tp = F(s)
in R, + R 2,
on C~' + C;',
= F(s)
"" f(8)
on Ci',
on Ci,
where f(8) is assigned arbitrarily on C;. Having determined U1, construct in the region R. the function II, which satisfies the equation
1
319
= F(8)
= Ul(S)
on C',f
on C~.
= u.(s)
on C',f,
on C~.
= F(s)
= Vn_I(S)
on
on
C~',
C~,
and
v. = F(s)
= un(s)
onC;'
on C;.
FIG. 59
+ z",'(z) + ~
The proof of this is contained in many books. See, for eu.mple, ~. GoUl'Sat,
CoUts d'an,uy$, 5th ed. (1942), vol. 3, pp. 207-210.
1
+ C as
L[~(t), .y(t)] ... ;p(t) + ~'(t) + W'5
(~(I),
.y(1 to
= f(t).
(~,
.y), we determine
and 80 on.
The use of this procedure in constructing the approximate solutions of
special elastostatic problems in doubly connected domains is presented
in detail in Sec. 88.
The proof of convergence of the Schwarz algorithm in the solution of
the second elastostatio boundary value problems for a doubly connected
domain R12 (Fig. 59) for the case when the contours C 1 and C 2 bounding
this domain are sufficiently far apart has been supplied by Mikhlin.l
In essence Mikhlin's proof is based on Neumann's modification' of the
Schwarz procedure for solving the Dirichlet problem in Laplace's equation
for the domain Rl + R,.
A more general proof of the Schwarz algorithm for the second boundary
value problem of elasticity in three dimensions was sketched out by
Soboleff. This proof reduces the consideration of convergence of
sequences of approximate solutions for the sum Rl + R, of the overlapping domains Rl and R" and for their product domain R 12 , to a study
IS. G. Mikhlin, Trudy, SeiBmDlogicallns!itute of the Academy of Science8, USSR
vol. 39 (1934), pp. 1-14.
'
J C. Neumann, Leipziger Berichte, vol. 22 (1870), pp. 264-321.
A detailed and
careful presentation of the Schwan-Neumann method of solution of the Dirichlet
problem for a class of elliptic partial differential equations in two dimensions and in
solving certain systems of integral equations will be found in L. V. Kantorovich and
V. I. Krylov, Approximate Methods of Higher Analysis, 4th ed. (19&2), pp. 637-695.
The trea.tment given in this book is also applicable to three-dimensional problems.
S. Sobolelf, "L'algorithme de Schwan dans la thoorie de l'e1a.sticite," C_ptea
Rend"", (/)()klady) de L' Academie deB Sci61lCe8 de I' URBS, vol. IV (XUI), No. (I9S.),
pp. 243-246.
321
(A
+ ",)1. .... = 0,
(i
= 1,2,3),
subject to
u.. .
(87.4)
j01" = F1,
1..[
... as" = G... ,
+-
(87.5)
1..\
... Ct' = G... .
We shall see in Sec. 107 that the solution of the boundary value problem
(87.3), (87.4) is equivalent to obtaining the vector 1.; which minimizes
the energy integral
(87.6)
U(u.) =
)Rl+RJ
322
(88.1)
'I'(t)
+ t'l"(t) + W>
= const
= -pt
on C"
on Co.
323
{88.2)
L
,,-0
cp(.)(z),
t[t(z)
t[t(,,)(z),
.-0
where the functions '1'('.), >/;".) are single-valued and analytic in the finite
region Izl < Rand <f('.+1) , 1/1(2.+1) are single-valued and analytic in the
region Iz - al > r, including the point at infinity.
FIG. 60
+ t<p(O)I(t) + >/;(O)(t)
-pt
Izl < R,
on Co.
These functions, clearly, will not satisfy the conditions (88.1) on the
boundary C I We next obtain the solution '1'(1), >/;(1) in the region Iz - al
> r, corresponding to the zero stresses at infinity, such that
(cp(l)(t)
+ tcp(l)'(t) + I/I(I)(t)
-l~[",'<)(t),
1/1(0)
(t)l
on C"
where I
(88.3)
L(<p,
.y) ""
",(z)
+ z?(zi + ~_
Then the functions ",(0) + cpCI), .y(0) + >/;(1) will be such that L(",(O) + '1'(1),
+ .p(l) vanishes on C but it does not reduce to -pt on Co"
In general, ",(.) (z), +(")(z)
will be determined from the boundary
conditions,
!/teo)
I It is not difficult to show from the uniqueness theorem that, if L[",(I),.,(I)J - const
on 0" then LI",(z), .,(z)] - eoo.st throughout the region.
324
(88.4)
L[,,(2..)(t),
t/r(2 ..)(t)]
on C1> n '"' 0, 1, 2, . . .
... -L[,,(2..-1)(t), .y(t-"(t}]
on Co, n == 1,2, . . . .
The general solutions of the exterior and interior boundarY-value problems for the circular region are known. Thus, if
,,(t)
+ i7(l5 + "f(l)
(88.5)
fez)
= F oCt)
on C"
1e. t - z
= _!__. r Fo(t) dt _
2111 le. t - z
t'
R" <pf(Z)
+!!!.,.
4riz
r Fo(t}
dt
t2
,
le.
On the other hand, the functions <p(z} and !/t(z), analytic in the regior.
Iz - al ~ r and satisfying the condition
!pet)
+ t!pl(t) + "f(l)
= F ,et)
!p(o)(z} = - ~pz,
",<O)(z} == 0,
Izi <
R.
and, therefore, the functions <P(1)(z), "'(1)(z) are determined in the region
It - al > r, from the boundary conditions,
tp(l)(t)
+ ~ + ",U)(t)
'"' pea
1
I
Ie.
+ t).
TheIle follow from Eq. (85.S) and (85.6) upon setting, - Rr.
TheIle follow from (85.9) and (85.10) upon setting m - 0 and , .. ,./. where
- . -0.
~ Ia verifyinc
SettingFl
..
pea
ri
80
that
(88.8)
",(I)(z)
r 2p
",(I)(z) '" - -
= 0,
z-a
+ pa'
Iz - al > r.
We next form
I
Izl < R
Ie. =
+ pa'
-L(rp(l), ",(1lc.
= -
pr'
r=a
- pa.
Making use of the formulas (88.5) with Fo(t) given by the right-hand
member of the expression just found, we obtain,
+ az)
2R'(R' - az)'
pr2a
",(2)(Z) = -pa +
(2)( ) '" _
(88.9)
'"
pr2z (R2
(R2 - az)2
(2R2 - az).
",(t)
+ t7m + ~ = const
= J(t)
J(t) = 0,
for t = Re",
=P,
for t = Re",
on C1 ,
on Co,
1 See, for example, Ya. S. Uflyand, Bipolar Coordinates in the Theory of Elasticity
(1950), pp. 204-210 (in RWl8ian).
See Sec. 77c
326
We again seek a solution in the form (88.2), where '1'(2.) , >/I<k) are analytic
for 1:1 < R and '1'( ..... 1), 1/1(2..+1) are analytic for Izl > r. AI> our first
approximation '1'(0) (z), I/I(O)(z) we take the known solution, dedy.ced in Sec.
77, for the solid circle of radius R, under the action of concentrated
forces. It is,
'1'(0)(:)
!2 (log: - ~R + ~),
2r
.1.(0)(:) = Pi
"
2.".
(10
z + ~R
z - ~R
g Z + ,R
+ _!!!__,_ _ ~).
z - ,R z + $R
The subsequent approximations are determined from the boundary conditions (88.4), with the aid of formulas (88.5) and (88.6).
Although the process indicated here leads to convergent series (88.2),
the convergence is slow. However, because of the special character of
loading, it proves possible to deduce the general expressions for '1'(1,,),
t(l,,) and sum the dominant terms in the resulting series.
Narodetzky'
obtained in this manner an approximate solution, valid to any specified
degree of accuracy.
Variants of the Schwarz method have been used by Mikhlin and
Sherman to solve certain integral equations furnishing solutions of the
first elastostatic boundary-value problem for the semi-infinite plate with
an elliptical hole. 2
89. Concluding Remarks. The principal object of this chapter has
been to introduce the reader to certain powerful general methods of solution of the two-dimensional problems in elasticity. These methods have
recently been extended to plane problems in anisotropic elastic media
and modified to include the problems of transverse deflection of thin
plates and several categories of contact problems s in elasticity. Among
the more comprehensive contributions of this type are: 4
S. G. Lekhnitzky, Anisotropic Plates (1947).
1. N. Vekua, New Methods of Solution of Elliptio Equations (1948).
1. Ya. Shtaerman, The Contact Problem of Elasticity (1949).
S. G. Lekhnitzky, Theory of Elasticity of an Anisotropic Elastic Body
(1950).
.
1 M. Z. Narodetzky, IZ1Je$tiya Akademii Nauk SSSR, Technical Serie8, No.1
(11148), pp. 7-18 (in RUllI!ian) .
S. G. Mikhlin, Trudy Seisnwlogical Institute, Academy of Science of the USSR,
No. 391 (11134) (in RllBSian).
D. 1. Sherman, Trudy &iBmological InstitutB, Acade>/ty of S~ of the USSR,
Nos. 53 and 54 (1935) (in RUIlSian).
The contact problems are treated in Chap. 13 of N. I. Muskhelishvili's Singular
Integral Equations (11153), as well as in his monograph Some Basic Problema of the
Mathematical Theory of Elasticity (1953).
With the exception of the book by Green and Zerna all these monographs are in
the RUIlSian language.
327
CHAPTER
THREE-DIMENSIONAL PROBLEMS
90. General Solutions. The key to effective treatment of the twodimensional boundary-value problems, discussed in Chap. 5, is in the
special representation of solutions of appropriate field equations with the
aid of certain arbitrary functions. Although several attempts have been
made to construct analogous "general solutions" of the three-dUnensional field equations of elasticity, such solutions have not been exploited
in a systematic way. The so-called general solutions are but particular
forms of solutions of the field equations involving arbitrary functions of
special types. Thus one can construct a solution of Navier's equations,
containing srbitrary harmonic functions that enter in particular combinations with certain known functions. The choice of known functions
and the form of solution are determined, in part, by the differential equations and, in part, by the topology of the region. Another" general
solution" of Navier's equations can be constructed with the aid of the
biharmonic functions, and there is no a priori reason why one form of
general solution should be readily transformable into another. The criterion of the generality of a given form of solution lies in the possibility
of determining the arbitrary functions 80 that the boundary conditions
are fulfilled.
Thus, in dealing with the two-dimensional elastostatic problems in simply connected domains, the general solution of the homogeneous Navier's
equations was obtained in the form 1
(00.1)
where lP(z) and 1/I(z) are single-valued analytic functions. This solutioll
is general in the sense that the unknown functions II' and", can be determined, essentially uniquely, when suitable boundary conditions are
Unposed. If one relaxes restrictions on the connectivity of the region,
or on the behavior of displacements on the boundary, the representation
(OO.I) may cease to be valid.
An equivalent form of the general solution involving four arbitrary
plane harmonic functions can be deduced from (00.1) by setting,
1 See.
71.
328
TBllE1!l-DDlENSIONAL PROBLEMS
329
+ i'l'S(:l:I, X2),
+ i"'2(XI, X,).
We readily find
(a,
(90.2)
P = 1,2),
"'2
(90.3)
p.V~
+ (X + p,)".i =
(i = 1,2, 3),
0,
" = 14. ,
X + p.
+ ! curl A
fJ
'
330
or
(90.4)
'Ui ..
>. +
".....
.. = A+j.I
_1_ V'>t "" " ,
(OO.5~
+ w,,) = o.
Hence
(90.6)
j.lU;
+ >t"
= '11"
+ V'>t = 4>",
(90.7)
A particular integral of this equation is ~ : : ;j.I x,.4>" and hence' the ~en
eral solution can be written in 'the form
(90.8)
j.lU;
= 4>, - 4>0,; -
+21'I' (X,.4>,.)",
'12 >..>.. +
>.. + 3j.1
= 2(>" + 2j.1) 4>, -
1 >.. + JI.
A + 2j.1 xlIIi,' - 4>0.,.
'2
But
>.. + 3j.1
3 - 4a
2(>" + 2j.1) = 4(1 - a)'
>.. +2j.1
- - = 2(1 - 0')
>"+,u
and hence
3 - 4a
1
,uU; "" 4(1 _ 0') 4>, - 4(1 _ 0') Xlpi,' 1
+G.l,
THB.l!lE-DIMENSIONAL PROBLEMS
331
(90.10)
21lu;
X"" -
"'0.'.
Xj<Pj., -
(90.11)
+ 1)/po
+ ipO.h
Xj/PO.h
in (90.10) yields
(90.12)
21lu;
(90.13)
ip2,
and
=
/P3.
xip. -
Likewise a substitution
+ ip3.a,
(O!
<Po
= (x
Xj/Pj.i,
+ 1)/p,
X,/P,.i
+ ipo,
= 1, 2),
(i = 1, 2, 3),
results in the expression of the form (90.10) involving only ipo, /PI, and ip
To establish the validity of the assertion, it is necessary to show that
the systems of Eqs. (90.11) and (90.13) possess harmonic solutions ip for
the arbitrarily specified harmonic functions <p. If we suppose that the
same set of displacements u, can be represented in either of the forms
1 To make the formal analogy complete, set 1"'.1 = "'" 1"'" = "' .
P. F. Papkovich, Comple8 rendus hebdomadair des seances de l'acadhnie de.
8ciences, paris, vol. 195 (1932), pp. 513--515, 754-756; !zvestiya Akademii Nauk
888R, Physics-Mathematics Series (1932), pp. 1425-1435.
H. Neuber, Zeit8chrift fiir anyewandte Mathematik und Mechanik, vol. 14 (1934),
p. 203, or his book TheorY of Notch Stresses (1946), pp. 21-25.
These usually stem from misconceptions about the meaning of the term "general
solution" and from inadequate recognition of the fact that the form of such solutions
depends on the topology of the domain.
'The proof of this in Neuber's book is lacking. Indeed, as We shall see presently,
the statement is not always true.
332
~; -
'P
XR;.; = 'Po.,
where
~; ... '1'. -~.
(x
+ 1)~; -
(x#;) .
(x
1)~;,k -
(:c,~;),;t,
=
Xk,
'Po,.
that
= 'Po,...
Interchanging the indices i and k and subtracting the result from (90.16)
yields
=
~;,.
~.,.,
(90.17)
= F,;.
(90.18)
O.
+ I)F,; -
(x;F,;) . =
'1'0,
so that
(90.19)
(x
+ l)F -
xjF,j =
'1'0
+c
'Po =
~ r"Y..(IJ, '1'),
0
A aummary
TBBEl!l-DIMEN8IONAL PBOBLBIIS
(x
of
~
r or = L.. r"Y.
+ I)F -
+ c,
.. -0
~ x + 11 _
+c 1 + ..L..
-0
nr
"
Y...
'Po
r(n+1) Y ..,
n-O
F
.
=x+l
+ L..1<+n+2
~
1
r
n-O
(+1)Y
"
ft
Since 1< > 0, this solution is valid and hence a representation of the form
(90.12) may prove possible in an infinite simply connected domain.
The possibility of representing every solution of Navier's equations in
the form (90.10), wherein one of the functions 'Pi is set equal to zero,
say 'Pa = 0, hinges on the construction of the harmonic function la from
the specified values of its derivatives. It is clear from (90.13) that the
harmonic functions II, l2, and lo will be uniquely determined once the
harmonic function Is is obtained from the equation
(90.20)
la,a
'Pa
If the region is a sphere, the function 'Pa can be represented in the series of
spherical harmonics as
e
'Pa =
rny..
.. -0
""
,,-0
m[A..P..{cos 8)
.-1
I The scalar product :r;;F.; of the vector r with the gradient vF of F is clea.rly equal
to r '!!..
8fo
It is easy to check' that the solution of (90.20) can be taken. in the form
-. '" i
.. -0
,.-+1
[n ~ 1
P,*1(COS9)
-!:
(90.21) becomes
which is precisely the formula (90.6). This connection was first noted
apparently by Mindlin. 5 We shall see that in a finite simply connected
domain every biharmonic fUnction can be expressed in terIlU! of two harmonic functions. It follows from this, and from the representation
(90.10), that at least two Qf the six harmonic functions entering in the
Galerkin solution are not independent.
1 In verifying it is advisable to use the integral representation of 8Illid harmonics
such as is recorded in Sec. 18.31 of Whittaker and Watson's Modern imalysis.
M. G. S10b0dyanski, Prikl. Mat. Mekh., AkademiJlO NaUk :888B, \rot 18 (19M),
pp.54--78.
Such formulas have been ,..,corded by several authors: G. S. SW.piro, Compla
ren.dua (Doklady) de l'academi. des scienC68 de l'UB88, vol. 55 (1947), pp. 693--695;
W. Freiberger, AU8tra/ian JouT'fWol of Scientific &search (A), vol. 2 (11149), pp. 483492; G. Yu. DzhanelidJe, Doklady Akademii Nauk 888B, New Series, vol. 88 (1953),
pp. 423-425; M. Brdi~.ka, CzechQslmJak Journal of Phyllia, vol.. 3 (1953), pp. 36-52 B. G; GaIerkin, Compta rt1id... Mbtiomadaire8 du .knee. de l'~ des BCienceB,
Paris, vol. 190 (1930), p. 1047; Compla randua (Dokladll) de l'~ des BCienceB
de rUR88, aer. A, vol. 14 (1931), p. 353, vol. 10 (1931), p. 281; PriIt(. Mat. Mekh.,
Akademiya Naulc 888B, vol. 6 (1942), p. 487.
R. D. Mindlin, B-ullmn of 1M Ameri.:an MatMmtIIical 8oeietv, "ol. 42 (1936),
pp. 373-376.
THREE-DIMENSIONAL PROIlLEMS
335
(90.22)
4>0
+ X;if>i,
(i = 1, 2, 3),
where the 4>'s are harmonic. 1 Any two of the functions <Pi can be aut
equal to zero without loss of generality, so that every biharmonic function (in a finite simply connected domain) is expre~sible in terms of two
harmonic functions in one of the forms:
F = <Po
+ XI4>,
= <Po
+ x.4>,
= <Po + X84>.
(90.23)
Let F be an arbitrary biharmonic function. The functions 4>0 and <P can
then be constructed as follows: On forming the Laplacian of (90.23), we
get
(90.24)
and since V'F is known, we can construct the harmonic function <P satisfying this equation. Having determined <P, we insert it in (90.23) and
get
4>0 = F - XI4>,
Til.!
= 0,
+ <P33.2' + <Pll.a> '1"3 = '1'11.22 + 'l'22.ll = CP3'.12 + '1'12.13 = '1'12.23 + <P23.2' 'I'll = <P 33
(90.26)
<P33,11
2CP.>.23
2<p31.31
2'1'12.12
'I'll
<P1I.23 -
'1'23.11
'1'31
'l'U.31 -
<P31.2'
'1'33.12 -
<P12.>3,
'1'12
<P23.31
'1'''.32 -
where the <Pi; = 'l'fi are of class C'. On setting <P12 = '1'23 = '1'31 = 0, we
obtain solutions proposed by Maxwell, and on taking <P1I = <Pn '" <P13 = 0,
I This is identical in form with the Goursat representation of the plane biharm'onic
motion U ("'" "'.) - 'PQ + "'o'P_ deduced in Sec. 70.
.
we get solutions due to Morera. 1 The functions <Ai are further restricted
by Beltrami's compatibility equations. These restrictions have been formulated implicitly (in tensor form) by Schaefer,' who aJBo indicates
a connection of relations (90.26) with the formulas for stresses deduced
from the Neuber-Papkovich solution (90.10).
91. Concentrated Forces. The general solution of the nonhomogeneous Navier's equations,
(91.1)
p.'V'u,
(l\
+ p.)""
= -F,
in
1',
u,(x) = A
[B
F,;~)
where
l\
31l
"" l\+p.'
and r = [(Xl - h)2 + (X2 - ~2)' + (xa - ~3)21~ is the distance from the
field point (Xl, XI, Xa) to the variable point (~1, ~., ~3) in T. The functions
F;(~) are the components of the body force F, expressed in terms of the
variables of integration E,.
The fact that (91.2) is indeed an integral of (91.1) can be verified by
direct substitution.'
A solution of Eqs. (91.1), appropriate to the deformation of an elastic
body by the concentrated force F2 applied at some point ~;, can be easily
deduced from (91.2). We suppose that the body forcesF. are distributed
over some subregion 1'1 of T, including the point ~;, and vanish over the
rest of the region. The resultant of the body forces acting on 1'1 is
F1
!." F, dT.
THREE-DDIENSIONAL PROBLEMII
If we now let F. incJ:.ea.se in such a way that this integr&l. has a finite
limit 11 as 1'1"'" 0, we arrive at the notion of the concentrated force 11
acting at the point ~i'
The displacements ",(x) produced at the point Xi ;C E. by the force F~
APplied a.t ~., as follows from (91.2), are
(91.4)
where
(91.5)
we find,
(91.6)
338
The tractions T;, produced by these stresses over the sphere S of radius
We get,
T = _ 6pCx a x 3,
a
a'
(91.7)
= a,
we find,
(91.8)
(91.9)
= rer
DXa
+ X3)'
u,
= -;'
1, 2),
(91.10)
x~
+ xi
xl
++ x: ) - 2( X~]
X3
r r + X3 )2'
T11
T.2
2pD [ r'( r
'Ta3
= -2pD~,
r8
712 =
T13 '"
+Xf]
X3)' '
X1
-2pD"ii'
'Til
+ 2r).
+ x.)'
-2 D X1 X.(X,
p
r'(r
T ..
=0
-2pD r 2(r ~
X3
)'
T.
1
-2pD;:o'
r == a.
Rc =
fs T .. dtr = 0,
Ra =
fs T3 dtr =
-SlruD.
'We note that, when, - 0, the BOlutiPn of Navier'. equations reduces to the
familiar problem in potential theory.
339
THREE-DIMENSIONAL PROBLEM6
We shall see in the following section that a superposition of the elementary solutions (91.4) a.nd (91.9) ca.n be made to yield the state of deformation present in an elastic half space whose plane boundary is under the
action of the concentrated normal force. 1
92. Deformation of Elastic Half Space by Normal Loads. Let the
semi-infinite region Xa ~ 0 be occupied by an elastic medium, and assume
that the concentrated force P, applied at the origin, acts in the positive
direction of the x.-axis. Since the point of application of the load is a
singular point in the solution of Navier's equations, we delete it from the
region x. ~ 0 by describing a hemisphere of small radius a and confine
our attention to the semi-infinite region bounded by the hemisphere and
the xlx2-plane,
We shall construct a solution such that the resultant of all exterIl81
stresses acting on the hemisphere is P, and
(92.1)
T13
723
= T33
= 0
over the rest of the boundary. To this end we form the sum of displacements in (91.4) and (91.9) and get
x"x.
DXa
(92.2)
Ua
= C -r , + r (+
r
x, )'
(92.3)
U,
= C
(a
1, 2),
(xX+l'r
+ 31' ! + ~) + !:?,
r'
r
The distribution of tractions over the surface of the hemisphere, corresponding to the displacements (92.2) and (92.3), can be got by adding the
tractions in (91.7) and (91.11). From computations leading to formu4ul
(91.8) and (91.12), it is obvious that the resultant force on the surface of
the hemisphere acts in the x,-direction and has the magnitude'
++
P = 4rjlC(X
21')
X
jl
+ 4rI' D ,
involving two unknown COIl8tants C and D. Another equation involving these constants is got by imposing the conditions (92.1),
1 Several problems in this category have been worked out by J. Boussineeq, ApplicatioIlll dee potentieJs A l'etude de l'6quilibre et du mouvement dee 80lidee 6laatiquee
(1885).
~ This is one-half the BUm of the valuee given by (91.S) and (IH.12), wherein the
integration was performed over the entire sphere,
340
2"C:I:."
2 D:I:
(a = 1,2),
so tha
2,,2(1
riC>' + II)
(92.5)
2"D
+7
o.
C=~,
(92.6)
Xa:I:..
'U ..
= 41r"
'Us
= ~~
41r1' r'
- 41r(>' +
x.
p)
rex. + r)'
(O!
= 1,2),
+ P(>. + 21') !.
41rp(>. + II) r
It is worth noting ' that at a great distance from the origin the displacements vanish as 1/r, and hence the stresses vanish as l/r2. In this connection it should also be observed that the concept of the concentrated
load is a mathematical abstraction resulting from specific assumptions
concerning the behavior of continuous distributiOlis of loads when a definite limiting process is followed. It is not surprising, therefore, that different limiting processes might yield singular solutions different from
(92.6). A decision about the practical validity of any given singqlar
solution should rest on physical rather than mathematical grounds. The
definition of the concentrated load in the instance of curved surfaces
obviously involves an even greater degree of arbitrariness. Because of
the usefulness of the solution in the,form (92.6) it is natural to use it as
a criterion for an acceptable definition of the concentrated load acting
on a curved surface. s
The solutions (92.6) can be generalized, in an obvious way, to yield
the displacements produced in an infinite region Xa ~ 0 by suitably
restricted continuous distributions of normal loads.
1 See remarks in See. 74 regarding the behavior of displacements and streMea in the
two-dimensional ease and their bearing on the uniqueness of solution.
See in this connection:
E. Sternberg and F. Rosenthal, "The Elaatic Sphere under Concentrated Loade,"
Juumal t1/ Applied Mechanics, vol. 19, No.4 (1952), pp. 413-421.
E. Sternberg and R. A. Eubanks, "On the Singularity at a Concentrated Load
Applied to a Curved Surface," A Technical Report to ONR, Department of Mechaniea,
mina JDstitute of Technology (1953).
A. Huber, "The Elaatic Sphere under Concentrated Torques," QuarlerJw of A~
lIt111wtNJtica, vol. 13 (1965), P.iI. IllH02.
341
THREE-DIMENSIONAL PROBLEMS
p(~, 'II) be
If we let
'U
a
(92.7)
x,xa
4...", ))
Xa
4".(X
+ "'))
-~
r(r
x.)
-~
i[
",V'Uj
+ (X + ",).,J.; = 0,
342
14J
(93.2)
UJ ... 'Pi
+ XI"'.i,
where the 'Pj and " are harmonic functions. These functions, as noted
in Sec. 90, are not independent since the 14 satisfy Navier's equations.
Indeed, on substituting (93.2) in (93.1) we easily find that
).+1'
>/t = -
).
+ 31' 'Pl.l.
'Pi(Xl,
X2,
XI)
..
II
gi(
-.
01,
..
(93.6)
//(Xl, x.) =
But it is well known 1 that properly restricted functions /;(Xl, x,) can be
represented by the Fourier integral, in the form (93.6), where
(93.7)
gj(OI, 13) ==
special_>
343
The substitution from (93.7) in (93.5) thel1 yields the desired functions <pj.
""j'
..
to Xa,
(93.8)
'"
= - : .:
..
u, =
Jf
\g,lOl, fJ) -
..
r(
{g.(Ol /3) JJ
'
- ..
..
u, = r( {g8(a (J) JJ
'
- .
u. =
A
A
344
Ui = ",(r)x"
(94.2)
d'", +! d",
dr'rdr
0
'
)l.uuo'i
+ I'(Ui.i + 1.4,;),
where iJ = 3A 1
given by
'Tij
= Mo,;
P,
= xilr is
Tr = (3)1.
+ 2p.)AI -
4p.A.
--,:s'
Also, the stress T, acting on the planar element with the unit normal
to p; can be easily computed from
n; orthogonal
T, = 'T,j1t.1I;,
= (M
that
T, .., AD
==
+ 2!i<P
t
(3~ + 2!i)A 1 + 2p.1
THREE-DIMENSIONAL PROBLEMS
345
for r = ai,
for r = as.
Al =
A
2
(3X
Plaf - paa~
afai PI - p.
-----,
a~ - af
r' ai - af
PIal - p,.ai
a~ai PI - p'.
ai - af
2r 8 ai - af
T.=
(94.7)
T,
T =
r
T =
plat
(1 - r~)3<-0'
Plat
(1 + 2r'
a~) > O.
a~-a~
ai - at
Thus, the maximum tension (T,)""", is at the inner surface of the shell.
We have,
(T )
fJmax
~ 2al + ai
2 a~-af'
The maximum extension ell obviously will occur on the inner surface of
the shell, so that the yielding will begin on the inner surface.
Most of the results recorded above have been deduced I by Lame.
Spherical Harmonics. The considerations of Sec. 90 indicate the
great usefulness of harmonic functions in elasticity. in solving the prot>-leIDS of equilibrium of an elastic sphere, one special class of harmonic
functions, known .as spherical harmonics, is particularly useful. The
essential facts about these functions are summarized in this section.
We first determine a class of particular solutions of Laplace's equation
(95.1)
.... = 0,
(i = 1,2,3),
<l>n =
ap",xfxlx ',
p+q+r-n
where the sum is extended over all positive integral values p, q, r such
that p + q + r = n.
It is obvious that the polynomial of degree 0, satisfying (95.1), is
<1>0 = ao, where ao is a constant. The polynomial
<1>, =
(i = 1, 2, 3),
G;Xi,
clearly satisfies (95.1) for an arbitrary choice of the constants as. In this
case we have three linearly independent solutions of (95.1), namely,
X" X2, x..
The linear combination of these solutions is the most general
solution of Laplace's equation in the form of the homogeneous polynomial of degree 1.
If we take the homogeneous polynomial of degree 2, namely,
(95.3)
<1>.
(i, j
G;;X,xj,
= 1, 2, 3),
+ a .. + a33
= 0
connecting six distinct constants in (95.3). Hence there are five linearly
independent homogeneous polynomials of degree 2 that satisfy (95.1) .
. These polynomials can be determined explicitly by setting
an = - (all
in (95.3).
+ au)
We thus get,
+ aS2(x~ -
xi)
xf - xl,
xl - x:,
of degree 3, substituting it in (95.1), and equating in the resulting expression the coefficients of the x. to zero, we obtain three relations among
ten tI.;ii in (95.4). Accordingly there are seven linearly independent homogeneous polynomials of degree 3 that satisfy Laplaoe's equation.
;'47
THREE-DIMENSIONAL PROBLlilMS
Xl = r
X, =
(95.5)
x.
= r
cos 8,
~.(Xl,
x., x.),
(95.6)
where Y.(9, <p) is a polynomial in sin 9, cos 9, sin <p, and cos <po ,The
function Y.(8, <p) is termed surface, or zonal, harmonic and rn Y.(8, '() is
the solid spherical harmonic. Inasmuch as the number of linearly independent integral harmonics of degree 11. is 211. + 1, the number of linearly
independent surface harmonics Y.(8, '() is also 2n + 1.
We deduce next an explicit representation of the surface harmonic
Y.(8, <p), by investigating the solutions of Laplace's equation in spherical coordinates in the form
~ =
(95.7)
)
1 02~_
aro ( r or~) + sin1 8 ii80 ( sm. 8 o~
08
+ sin' 90<p' - O.
0 (.
SinS aD
OY)
sm 8 iii
1 o'Y
+ SEi'8 8ift'
348
+ 2rf'(r) - kf(r) = 0,
1 o'Y
+ sin' IJ 0",' + kY = 0,
rtf"(r)
1 0 (.
oY)
sin' iii \SIn "011
(95.10)
where k is a constant.
We are interested only in continuous solutions of Eq. (95.10) since, as
observed above, the surface harmonics are trigonometric polynomials.
Equation (95.10) will have such (nontrivial) solutions only for certain
values of the parameter k, and our problem is to determine these characteristic values and construct the corresponding functions Y(II, ,,). On
comparing (95.7) with (95.6) we see that fer) satisfying Eq. (95.9) is r\
and on inserting this in (95.9) we get an infinite number of the values
of k, namely
k
n(n
+ 1),
11.
= 0,1,2, . . . .
&!
fJ
=0
for the surface harmonic Y.(fJ, ",). The considerations pertaining to the
number of linearly independent spherical harmonics leads us to expect
that, corresponding to each characteristic number k = 11.(11. + 1), there
will be 211. + 1 linearly independent solutions Y.(II, "') of (95.11). We
can obtain these solutions by taking .
(95.12)
+ 1) sin' fJ = -m',
R.are
R( )
'"
where m = 0, 1, 2, . .
= {Sin
m""
cos m",
,n.
The equation for Q.(I) can be cast in the standard form by introducing a new independent variable x == COB 8. On making this change
TBBEJII-DIMlIlNSIONAL PllOBLBMS
! [(1 -
Xl)
P~-I(Z)
dP~(X) J+ [n(n + 1) -
satisfies Legendre'.
1 ~. X2 ) Plrl(x) == 0,
(m = 0, 1,2, ... ,n).
There are two linearly independent solutions of this equation, only one
of which is continuous in the interval Ixl ~ 1, that is, for 0 ~ II ~ .
This solution is
"
L
.. -1
j.
Moreover, this set is complete in the sense that every function f(lI, I{J),
specified on the unit sphere l:, whose square is integrable over l:, can be
represented in the series of spherical harmonics which converges in the
1
mean to f(6, rp). If f(6, rp) is of class C' on :2:, the series converges to
f(fJ, rp) uniformly. In fact, we have the following representation:
L {(I,~ft)P.(cos
e
f(6, rp) =
(I,~O)
.. _1
8)
.. -1
where
(1,(.)
..
= (2n
l)(n - m)!
2r&..(n+m)!
l)(n - m)'.
be,,) = (2n
..
2r8.. (n + m)!
II
,T
f(1I
,~)P(oo)
,T
with 0.. = 2 if m = 0, Om = 1 if m > 0, and ]X.!') (cos II) "'" Pft(cos II).
This representation permits us to solve the problem of Dirichlet for
the sphere of radius a in the series of solid harmonics. We first represent th~ function f(8, ",), s~ified on the surface of the sphere in the
series
f(fJ, "') =
Y ft (lI, "'),
.. -0
n ... O
;)
L
e
Y .. (8, "')
< a.
~..+1,
()
r> a.
.. -0
pV~
+ (A + ",)u.,.; ... 0,
which are such that on the surface of the sphere of radius a they reduee
to a oomplete set of surface harmonics Y.(fJ, rp).
351
THREE-DIMENSIONAL PROBLEMS
(i, k
1,2,3)
c = -
(96.3)
2{;>.(n - 1)
+ (2n
- 3)1..]
L ("'In, + c(n)r'<p~~li)'
,,-0
u; =
But <pl?J1i =
<Pl~k =
0, and hence
L (",lnl + c("+)r2<pk~t.2l).
~
(96.4)
U;
,,=0
If we go over into spherical coordinates with the aid of (95.5) and set
r = a, each term in the series (96.4) becomes a surface harmonic of
degree n.
Now if the displacements u; = f,(6, "'), specified on the surface of the
sphere r = a, are represented in the series of surface harmonics as
c
(96.5)
u;
L AI"'ell, "'),
,,=0
352
~ (<Pl
(96.7)
ft
+ cCft+2)a'<p~~~J)
n-O
must converge for r = a to the same function/.(e, "") as the series (96.6).
Making use of the theorem on uniqueness of representation in the
series of solid harmonics, we can write
for r = a,
and deduce
(96.8)
(!)"]
[ A "I") a
and since
",,1,~t2)
""I"')
, i l
+ c C"+2)a ",,j,"H)
,Ain,
2
is harmonic, we get
(96.9)
Thus <PI::' is completely determined since the AI") are known functions.
The substitution from (96.9) in (96.8) determines the <PIn) in the form,
(96.10)
Hence the solution (96.4) can be written entirely in terms of the AI")
determined in (96.5). It is
(96.11)
u; =
.t
{Aln)
(~r + cC.H)(r' -
2
a ) (
A1,-+ 2) (~)"+tJ
TB.JUIE-DUlENBIONAL PROBLlIIIfS
353
meth9d described here to solve the first bound.acy~value problem for the
sphere. Although, conceptually, this problem is not any more difficult
than the second boundary~value problem, the necessary ea.lcuia.tiOnB are
considerably heavier. 1
.An obvious generalization of the method, making use of particular
solutions in the form <P(")r-1_ 1, where the <P(n) are surface harmonics, has
enabled Kelvin to treat the problems of equilibrium of an elastic shell.
These problems have recently been reconsidered by Lourje, who, in following the Kelvin mode of attack, uses the Neuber-Papkovich expressions for displacements, involving four harmonic functions instead of
three used by Kelvin. This results in some simplifications, enabling
Lourje to carry out the computations more fully.s
When the radius of the inner shell is small, the solutions indicate the
nature of stress concentration in a large body near a spherical cavity.
A brief survey of the technically important problems on stress concentration is contained 3 in Timoshenko and Goodier's Theory of Elasticity.
Among the three~imensional problems for which explicit solutions are
available are several problems in the category of contact problems of
elasticity. The problem of deformation of an elastic half space by a
rigid circular punch, first treated by Boussinesq, was developed in some
detail by Shtaerman and Lourje and, more recently, by Leonov.' A systematic treatment of this and related contact problems of elasticity will
be found in 1. Ya. Shtaerman's'monograph entitled The Contact Proble~
of the Theory of Elasticity (1949).
Since exact solutions of the three-dimensional problems pose serious
mathematical difficulties,6 recourse is made to approximate solutions
1 See, for example, A. E. H. Love, A Treatise on the Mathematical Theory of Elasticity (1929), pp. 267-270.
A. I. Lourje, Prikl. Mat. Melch., Akademiya Nauk SSSR, vol. 17 (1953), pp.
311-332.
'See also H. M. Westergaard, Theory of Elasticity and Plasticity (1952), pp. 154157 and R. A. Eubanks, "Stress Concentration Due to a Hemispherical Pit at a Free
Surface," Journal 01 Applied Mechanica, vol. 21 (1954), pp. 57-62.
I. Ya. Shtaerroan and A. I. Lourje, Prikl. Mat. Mekh., Akademiya Nauk SSSR,
vol. 5 (1941); M. Ya. Leonov, Prikl. Mat. Mekh., Akademiya Nauk SSSR, vol. 17
(1953), pp. 87-98. See also N. A. Rostovcev, "On the Problem of Torsion of an
Elastic Half-space," Prikl. Mat. Mekh, Akademiya Nauk SSSR, vol. 19 (1955), pp.
55-60.
Interesting recent contributions to exact solutions of the axially symmetric pro~
lems of elasticity are contained in two papers by E. Sternberg, R. A. Eubanks, and
M. A. Sadowsky, Joumal of Applied Physica, vol. 22 (1951), p. 1121, Pror.eedingB of
the First United State. National Congress of Applied Mechanics (1952), and in a brief
paper by G. S. Shapiro, Doklady Akademii Nauk SSSR, vol. 58 (1947), pp. 13091312, in which the equilibrium of an ellipsoid of revolution is considered.
The equilibrium of an elastic parallelepiped was considered by M. M. FilonenkoBorodieb, Prikl. Mat. Mekh., Akademilla Nauk SSSR, vol. 15 (1951), pp. 136-148,
562-574, with the aid of Papkovich-Neuber str_ functions.
72(U;.; -
Uj.;)
write
T, =
Tv".
= (M&ij
= Mil.
+ 2j.1ll;j)"i
+ 2j.1U;,jllj + j.I{u". au
A-8". + 2j.1 a,: + 2j.1 ii"i
u;'i)lIj
duo
WIi"i'
THREE-DIMENSIONAL PROBLEMS
355
action of two systema of surface tractions T. and T, producing the displacements "" and "'~' respectively, then
/::: T."" dtT = /::: T:u; do-.
(91.4)
In deriving this theorem it is assumed that the functions u" u~ and their
derivatives are continuous throughout the region r.
If we now consider a solution of Navier's equations in the form
(97.5)
a,.-'
ax.
",9=-,
u:
(97.6)
But.
Is
T;u?
au
Is
r;;vjuf
au
=
=
Since {} and
IJ;j
Js (Ma + 2/JlJ;j) ~
i;
or- 1
ox, du
Is + X:l au
Is (x;, + ;:. XiX~;;) au.
(Maii
2/Jeii)
/s
XiX;
au
= 'Ys.".R'a;;,
/s au = 4rR2,
/s T;u~ do-
-->
+ 2/Je;;(0)'Ys.".a;;
+ %/J){}(O).
4rM(O)
= 4r().
~o /s T~u; du
_1%1r/J{}(0).
This formula. ena.bles one to compute the dila.tation a.t a.ny point (which
we have chosen. to be the origin) whenever both the displa.cement u. and
tra.ction Ti are known over 1:. If only the u. are known, we can eliminate the T. by 801vi~ the following auxilia.ry problem: Find a solution
u: of Navier'lJ equation8 in the region T such that u: = u2 on 2:.
For if such u~ are known, then, by the theorem (97.4),
f:z:
T;uidu =
fz
Tiu; du =
f:z:
T.uf M,
where both the Tf and T; can be computed since the corresponding displacements uf and u~ are known.
We note that the determination of the u; is equivalent to finding a set
of functions
satisfying the following conditions:
1. v. satisfy Navier's equations, except at the origin.
2. v. = 0 on :t.
3. become infinite at the origin in the manner (97.5).
It is thus clear that the Vi are analogous to Green's functions.
If the surface tractions Ti are specified, we can compute {J by finding
the solutions u;' of Navier's equation corresponding to the tractions
Ti' = T2 on:t. Then, from the Reciprocal Theorem,
"i
I" T~U;M
J:z: T.-u,' M,
+ 21'){J(0)
Jz
Ti(u;' - uf) M.
UO
with components,
arl)
ar-l
0
- , -OXI
-,
, aXa
<
THREE-DIMENSIONAL PROBLEMS
357
f T.-u? dO'
0,
lL~ 18 T?u, du
41rJ.'(Ua.2 - U2,Ifo
lim
R-0
= 81rJ.'W'(O).
Hence,
(97.11)
In order to express this formula in terms of the surface displacement8
alone, we consider a regular solution u! of Navier's equations such that
u; = uf on l: and find, as we did for the dilatation,
&rJ.'W,(O) =
Ix (T; -
T?)Ui dO',
aT-'
- , -at-')
-,
(0
, aX3
ax.
where the origin of i' is at 0.
Let the tractions corresponding to uf - u? be T!', and let u;' be the
regular solution of Navier's equations in l' such that on t it yields the
tractions T;'. Then the Reciprocal Theorem and (97.11) yield
S"'J.'[w,(O) - w,(O) =
=
358
'The integral equati0D8 for " and .. have recently been derived by
Arzhanykh. 1
88. EDstence of Solutions. We saw in the preceding chapter that the
existence of solutions of the fundamental two-ditnensionall,roblems follows directly from the existence of soluti0D8 of certain wellknOWJ.I. int&gml equations. 'The demonstration of existence of solutions of the threedimensional problems can also be made to depend on the exjlJl;ence of the
solution of integral equations of the Fredholm type or, altefIllltively, on
the construction of Betti's auxiliary functions. We shall not pursue this
subject here and shall merely remark tltat the matter of existence of
solutions has been satisfactorily resolved for domains of grej\t generality
by Fredholm, Lauricella, Kom, Weyl, Lichtenstein, and Sh~man.!
The caliber of mathematicians who have concerned tltexnaelves with
the problem is indicative of its complexity.
An extension of the uniqueness theorems in the linear tltoory of elasticity to problems involving concentrated loads is provided. in a report
by E. Sternberg and R. A. Eubank.'
99. Thermoelastic Problems. We have assumed in preceJing chapters
that the elastic bodies undergoing deformations were maints.ined at oonstant temperatures. Thermal changes in a body are -accompanied by
shifts in the relative positions of particles composing the f>ody. Such
shifts, in general, cannot proceed freely, and tlte body beeoxne& stressed.
lJ'\\d~1: {ree i;.t>.~t'&.\Il ~~<\'\\ c;,i ~~~\.~ \\<li~ %. 'f<\t'M'lM\~1;. to.. tM.
shape of a rectangular parallelepiped with edges It parallel t4J coordinate
axes deforms into a similar parallelepiped with edges l;. For small temperature changes T(Xl, Xs, x.) the relationship between l'J &jld l; has the
1 Prikl. Mat. MeJcA., Akademiva Nauk 8S8R, vol. 15 (1951), pp. 381-391.
For proofs relating to the first boundary-value problem see:
A. Korn, Annales de l4/aeuJte deB IICitmca de Toulouse, aero 2, vol. 10 U908), pp. 1115-
269'
Ii.
Weyl, ReM.ictrati del cirotJlo 1R/lI6matio9 di Palermo, vol. 39 (1915), pp. 1-49
For the second:
I. Fredholm, Arkiv lOr MaIematik, .tBtronomi oc/& F1JBi/I;, vol 2 (1906), pp. 3-8.
G. Lauricella, Atti del14 rtlIIle acaJdemia nazionale !lei Lincei, aero 5, ~ol. 15 (1906),
pp. 426-432, vol. 16 (1907), p. 373; It Nuorocimento, ser.li, vol. 13 (1907), pp. I04-U8,
155-174,237-262,501-518.
A. Korn, Annales de fkok fIONIUlk IlUphieure, aero 3, vol 24 (1907), pp. 9-75;
RImdiconti del circtJlo mtJIematiI di Palermo, vol. 30 (1910), pp. 138, 336; MathematiKIte
Annalen, vol. 75 (1914), pp. 497-544;
L. Lichtell8tein, M~ Zeitachrift, vol. 20 (1924), pp. 21-2!l; vol. 24 (1925),
p.640;
D. I. Sherman, PriJd. MaL. Mel:1&... Akadomitl'l Nauk SSSR. vol. 7 (1943), pp. 341360.
A Technical Report to the 0Biee of Naval Reeearch, Departmenf of the ?iavy,
from the Department of Mechanica, DIinoiB Institute of Technology, June 15 (1954).
359
TJIIUIl]!l-DDlEN810NAL PBOBLltVS
form
~ = l1(1
+ aT),
e:'; = aT8u.
where
eo; = .),2(u;.J
+ 'Uj.,}.
Tij
= >.."8i;
+ 2,.e;; -
a{3>..
+ 2,.)T&;;,
+ Fi = pu.,
#,V'u; + (>.. + ,.)"., + Fi -- {JT.,
TiJ.;
yields
==
~,
".....ti_.
where
fJ ... (aX
+ 2p)a,
,..V'1t;
+ (X + "')""
= - (F, - pT.,I.
Then
and
V'u; = rp,;a.
On substituting in (99.6), we get
j.lrp.;kk
+ (X + p)rp.kki
= -ft'i
+ {JT."
and if there exists a potential <I> such that Fi = -<1>,;, we can write,
(A
+ 2p)rp,Ui =
(<I>
+ PT),;.
'I'.u
L
:=
X + 2,.. (<I>
X;
yields
+ (3T) + const.
We recall thai this is "lways the case with the gravitstionalsnd centrifugal forces.
THREE-'&IMENSIONAL PROBLEl\IS
361
>.
(x) = _
If'
_!_
41r J.
r(x, x')
=h
In the solution (99.9), rex, x') is the distance from the point (x) with
coordinates x, to the point (x') with coordinates x:, and the integration is
performed with respect to the primed variables. Once If' is determined
from (99.9), the desired particular integral is given by the formulas
(99.8). We note that, when the body forces vanish, the function p in
(99.9) is simply
fJ
_
p - h + 21' T.
Borchardt' has made use of integrals of the form (99.9) in the general
discussion of the thermoelastic problems and in solving certain special
problems for spheres and circular plates subjected to asymmetric temperature distributions. A method of integration of the thermoelastic
equations, with the aid of integrals similar to those of Betti and Somigliana, was outlined by Rosenblatt. 2 Goodier, Mindlin, Cheng, 'and
Mykelstad used integrals of the type (99.9) to study the effect of special
temperature distributions in the infinite and semi-infinite elastic solids.'
Instead of dealing with Eqs. (99.6) we can start with Cauchy's equations,
1'.;.; + Fi = 0,
where the 1'ij satisfy appropriate compatibility conditions. The latter
can be written down at once from (24.14) by replacing the Fi in (24.14)
by the "effective body force components," Fi - /3T. i Another way of
1 C. W. Borchardt, M()~ der Akt:ldemi. der WiB8emckaft, Berlin (1873),
pp.9-56.
A. Rosenblatt, Rendiconti del ciroolo matematioo di Pakrmo, vol. 29 (1910), pp.
324-328. See also W. Nowacki, Arch. Meek. Sto8., vol. 6 (1964), pp. 481-492
,in Ioliah).
'J. N. Goodier, Philoaophical Magazine, vol. 23 (1927), pp. 1017-1032; R. D.
Mindlin and D. H. Cheng, J(J!J.mal ()f Applied Physica, vol. 21 (1950), pp. 926, 931;
No O. MykJestad, J(J!J.mal ()f Applied McchtmicB (1942), p. A-13l.
362
deducing such equations is to insert from (99.4) in the Saint-Venant compatibility equations (10.9).
The thermoelastic problem is further complicated by the fact that in
many instances it proves necessary to determine first the temperature T
from the Fourier heat-conduction equation. The available exact solutions
of the heat-conduction problems are limited to spheres and cylinders and
to a few problems involving plates and rods subjected to special temperature distributior.cs. t We shall consider some of these in the following
sections.
100. Thermal Stresses in Spherical Bodies. The deformation of a
spherical shell subjected to a centrally symmetric distribution of temperature can be determined2 in the manner of Sec. 94.
We take the temperature function in the form T(r), where r is measured
from the center of the sphere, and seek a solution of the system (99.6)
with Fi = 0 in the form
(100.1)
Ui = xiCP(r),
!!.r T' =
r.
(100.2)
(100.3)
(X
cp(r) = At
where
(100.4)
+ ~: + )\ :
11'
cpo(r) = -.
r
r,
'
2p. cpo{r),
7'(r)r' dr.
The lower limit rl in the particular integral (100.4) can be chosen ill any
convenient, but definite, manner.
On noting (100.1), we get
u..;
= Oi,'P
+ xi<P'(r) ~,
r
This problem and the corresponding problem for the circular cylinder were first
IIOI.ved by Duhamel in the mllmoir cited in See. 99. An independent solution was
alIo given by F. Neumann in 1841. There are numerous papers on these problems
rediscovering the DuhlUllel-Neumann solution; some of these contain e1aborate
eaJeulations.
Cf. (114.3).
363
THREJl-DlMENSlONAL PBOBLJilMS
J = 3" + 1',,'.
Hence, i,he t'tress T. = Ti;lIi"; in the radial direction "i =
+ 21'[" + 1',,'(1')] -
T. = M
(100.6)
x./r ill
PT,
T.
(3X
+ 2j.1)AI
41'
1"
4j.1P
As - A + 21' 'Po(r).
T. = 0
for r = aI, l' = as,
and, on solving these equations, we get
(3A
+ 2)A
I'
= 4j.1p[a~'PO(a2) - ai'PO(al)]
(X
2j.1)(a~ - an '
A. = paia~['Po(a.) - 'Po(a,)1.
(i\
2,,)(a: - aV
On setting a, = 0, we get the solution for the solid sphere. In this case
it is convenient to take the lower limit of the integral (100.4) !t.S zero.
If the How of heat in the shell is produced by maintaining the outer
surface of the shell at zero temperature and the inner surface at a constant temperature To, then, for steady heat flow,
Hence,
'PO=-1
r3
o
J.' -Toal
- - (a
- - 1 ) r dr
2
at
a2 -
al
T. =
aETo a,a,
1---'--'
- qa, - a,
[+1
as - -r (al + a,a, +
Ih
af)
+ alai]
-a,. '
in
364
Since T. > 0 for To > 0, the hoop stress is a monotone increasing function, 80 that the largest value of T, is on the outer surface of the shell.
The extreme value of T. occurs when 1 r' = 3aiai/(ai + ala. + a:).
101. Two-dimensional Thermoelastic Problems. The state of stress
induced in a long cylindrical body by the distribution of temperature,
which does not vary along the length of the cylinder, can be determined
by solving the familiar problem in plane elasticity. For if we take the
xraxis along the length of the cylinder and assume that the temperature
T(X1, x.) is independent of the x.-coordinate, then the stress components
T,.; will not depend on x,.
These stress components can be balanced by
the application of suitable longitudinal forces and bending couples applied
to the ends of the cylinder, so that its cross sections remain plane. If,
now, the solution of the plane-deformation problem is superimposed on
the solution of the simple problems of tension and pure bending, the
result will represent a valid solution of the original problem not too near
the ends of the cylinder.
We thus need consider only the plane-deformation problem wherein
Ua
= 0,
(a = 1,2).
Since
e'j =
e13 = en = ea.
Yz (14,j + Uj,') ,
(i, j = 1, 2, 3),
(101.1)
{
1'33
+ Il(U.,~ + u#,.)
- kT0a#,
= Xu", - kT.
U"r
+ 2kT
+ Il) ,
= 2(A
so that
(101.2)
T ..
X
2(X
+ Il)
kll
7'0. -
+ Il T.
7' a#,~
= 0,
365
THREE-DIMENSIONAL PROBLEMS
+ e".ll
ell."
2e U lt
and making use of (9904) and (lOlA), we find that W satisfies the equation
(101.5)
V"V'W
+ cV'T =
0,
where I
C
21'k
= ---.
A
21'
If we set
w=
(101.6)
U- V,
V'V
cT,
V'U
o.
V(XI, x.) =
where r' = (XI - XI)' + (x, - x~)' and R is the cross section of the
cylinder.
When the tractions T a are specified on the boundary C of the cross
section R, we have
(10LlO)
U. ll
V. u ,
1'11
= U .2
V ...,
1'12
V. lS
U. lI ,
we find that l
(101.11)
dU. _ T ( )
(Ii' 18
+ Ts'
dV. t
dU. I = -T.(8)
ds
+ dV.
ds
1,
866
U,l
+ iU"
i/.'..
"" 11(8)
(T,
+ il,(s) + const
on C.
0,
THREE-lHMENSIONAL PROBLEMS
367
== .;
a
roe
a
rre = O.
rT(r) dr,
368
same for each particle. Thus, the general motion of the system of n
degrees of freedom can be represented by a linear combination of n simple
harmonic motions with n distinct frequencies. These frequencies are
determined by solving the secular equation which is completely determined by the quadratic forms representing kinetic and potential energies
of the system.'
When the system is continuous, the corresponding secular equation has
infinitely many real roots and hence infinitely many characteristic Juneti(}1l.8 representing normal modes of vibration. These characteristic
functions are solutions of the differential equations of motion with
appropriate boundary conditions. Thus, in dealing with small free
vibrations of an elastic solid, the characteristic functions, satisfy the
equations
in r,
(102.1)
Ti/Pj
for all t,
on 2:,
Ui(X, t)
+ E).
with ~' = uti' The solution of Eqs. (102.4) satisfying the boundary
conditions (102.2) is known to exist only for a denumerable set Wk (k = 1,
2, . . . ) of values of 101, all of which are real. Thus, the characteristic
functions are
(102.5)
ulk)(x, t) = u: Ck ) cos (Wkt + e),
(k = 1,2, . . .),
where the u:(t, are the appropriate solutions of (102.4). One then concludes that every oscillation of the body can be represented in the series
1);;
L Akul
),
k-I
where the Ak are suitable constants, whenever the ulk ) are orthogonal
with respect to the region under consideration. 2
1 See, for example, E. T. Whittaker, Analytiea.I Dynamics, Chap. 7, or H. Goldetein,
Quaieal Mecha1!ics, Chap. 10.
The mode of 8Olution described here is precisely that UIIed in aoIviq the probllml
of lIQl&Il traoaYer8e 'Vibrations of an elastic string by the Fourier method. The A. are
determined from condition. ebaracterisinc the initial diaturbance.
369
THREE-DIMENSIONAL PROBLEMS
(102.6)
where r' =
+ .)
Then
u: = zd(r).
(102.7)
f"
(102.8)
+ ~ f' + kif = 0,
where
k2=~.
(102.9)
+ 21'
f =
(102.10)
~ (kr cos kr
- sin kr),
z,
u.=u,r
= rf(r)
u,.
= (31
cos (",t
+ .),
so that
(102.11)
r" = [(3),
(). + ~)((2
- sin ka) = O.
370
MATHEMATICAL
Ta~RY
OF EL.\STlClTY
values of the six lowest roote of thi8 transcendental equation for ~ = p"
which corresponds to IT = ~, are recorded' in Sec. 196, p. 285, of Love's
Treatise.
It ill clear from this simple example that the determiuation of characteristic functions in the more general case ill extremely difficult.'
103. Wave Propagation in Infinite Regions. If a region ill so large
that the effects of the boundaries can be dillregarded, it ill possible to
represent the disturbance as a sum of two waves propagated with velocities that depend only on the density and elastic constante of the medium.
Indeed, the displacement vector u can be represented as a sum of two
vectors, one of which is solenoidal and the other irrotational. This leads
to a consideration of two special types of disturbance for one of which
div u = 0 and for the other curl u = O.
N ow if div u = 0, u;" = " = 0 and the general equations
(103.1)
,.~'u;
+ (X + ,.)"., =
pu.,
reduce to
(103.2)
Tt2
(103.3)
ctv2u"
c=
(103.4)
(X
+ 2,.)V2u, =
,nl;.
THREE-DIMENSIONAL PROBLEMS
371
a!(ru)
---;)iI =
C'
as
ars (rIu).
ct),
tIi = Fi(X;,,; -
where the Fi are arbitrary functions and the Vi are the direction cosines
of the normal to the plane of the wave. If we insert (103.7) in (103.1)
and eliminate the Fi from the resulting equations, we get the equation
for c in the form
(pc' - p)(pe' - ~ - 2p) = 0,
which shows that plane waves travel with the equivoluminal and irrotational velocities.
PROBLEMS
1. ShOW that. when the displacement vector u is written in the form
u = v ..
+ curl ,,",
II'"
/It' -
C:VS...
Ct '"'
.Ji+
2,.
"V~--.
p
and
Cl-=
~.
p
.. _ 04......+4<...->"1.
",_~_-..oI,
372
8. Referring to Prob. 1, show that when the gradient and curl are exprellll!d in
cylindrical coordinates (r, '. #), then. in axially symmetric problems,
u, - 0,
Md
a",
(i'",
iJr
r iJr
a'", 1 a'",
at' =--,
cl at'
at,y lat,y
~+--+t
at,y
lay,
x. ;;:: 0, be set in motion by forces applied at some distance from the free
boundary (x, = 0) of the solid. We shall suppose that the resulting
waves are propagated in the positive direction of the x.-axis and that the
deformation is plane with Us = o. Weare then led to consider solutions
of the two-dimensional equations
(a = 1,2),
(104.1)
in the region
X2
(104.2)
x, = o.
U ..
Va
+ W.,
p.V'v.. = pD..,
(>.
+ 2p.)V~.. =
piD...
vol. 2, p. 441.
THREE-DIMENSIONAL PROBLEMS
373
Thus the II. are associated with an equivolumina.l wave moving with
the velocity
(104.6)
while the
Wa
(104.7)
Ct
~). ~ 21'.
(104.8)
= Aa~rH(.S-l-JI'I),
{ Wei" =
a> 0,
b > O.
Bcce-h::-HCl-"l),
The period of the waves in (104.8) is 2.../p, and the wave length is 2.../8;
hence the velocity of propagation is
(104.9)
C.
1!..
8
(104.10)
S2
+ k2
+ h' =
0,
0,
where
k'
(104.11)
Further, since
Va
= PP',
I'
Hence
and we can take the vectors in (104.3) to be
(104.12)
v, = Aac"'" sin (8X, v. = A8e--' cos (axl WI = Bse--l>z sin (ax, w. = Bbe--l>z cos (ax, -
pt),
pt),
pt),
pt),
cr.
1
Prob. 2, Bee. 103. In accordance with the wrual practice, the desired vectors
v. and w. are determined. by the real parta of expreesion.a in (104.8).
374
1'22
1'lt
=;
= 0 on
law,
(104.13)
+ 2"ul,l =
+ = 0,
{ Xu.....
U1.2
0,
U2.1
for x.
O.
(104.14)
which have nonvarushing solutions for A and B if, and only if,
[X(b 2
(104.15)
8')
+ 2I'b'](a' + 8')
- 41''tWs'
= O.
(2 - 82k,)4
(104.16)
(1 - 82h
2
= 16
1 -
k')
82 .
But
~ =~,
8'
C{
(104.17)
where
(104.18)
'"
cf
ci
== c'
..!,
cf
I'
x=-=--'
X + 2"
On simplifying we get,
",' - 8",'
(104.19)
If we take X =
1
1',
+ (24
then x =
- 16",)", - 16(1 - x) = O.
3",1 - 24",'
+ 56", -
or
('" - 4)(3",' - 12w
32 = 0,
+ 8)
= O.
= 4,
"" =
2 + 2/0,
"'3
= 2 -
2/0.
It is easy to check that the first two of these do not give positive values
Thus the only suitable root is '" = 2 - 2 Vi,
and hence, from (104.18),
to a and b in (104.10).
Ca
Cl
V;;;
= 0.9194
~~.
375
'J.'IHtEE-DI1IENSIONAL PROBL1DI8
&>2
+ 24w -
0,
(f
== ~, x = 0),
16 = 0,
and one finds, as above, that the velocity of the surface wave is
c.
= 0.9553
~.
}..u....,
= !'(U1,2
+ 2!'u2,2
+ U2.l)
=
=
P cos
pt),
(SXl -
o.
P,
O.
These can be solved for A and B. The corresponding expressions for u..,
with x. = 0, give the displacements on the free surface of the solid under
the action of forces P cos (Bl: - pt) distributed along the xa-axis. These
can be generalized l in a familiar way with the aid of Fourier integrals to
provide formulas for the displacements on the free surface due to the
forced vibrations of a more general sort. The analysis of resulting formulas shows that the steady-atate disturbance of tpe free surface consists of three waves one of which moves with velocity Cl, the other with
velocity C., and the third with the Rayleigh velocity Ca.
A similar analysis has been carried out by Lamb for the semi-infinite
1 H. Lamb, PhiloaophictJl T~ of the Royal 8ociet1l (London) (A), vol. 203
376
half space a portion of whose boundary is subjected to the axially symmetric forced vibrations. This problem w~ treated by a different
method by Smirnoff, Soboleff, and Narychkina. 1
1 V. Smirnoff and S. Soboleff, "Sur une methode nouvelle dans Ie probleme plan dee
Vibrations 6lastiquee," Trudy, &is~ lnatitute of the Academy of Scimcei of the
USSR, No. 24 (1932); No. 29 (1933).
S. Soboleff, Matematichuki Sbornik, vol. 40 (1933).
E. Narychkina., "Sur lee vibrations d'un demi-espace aux conditions initial"" arbitraires,"Trudy, 8eisnwlogicallnatitute of the Academ!l of Scienceo of the USSR. No. 45
(1934); No. 90 (1940).
See also closely related papers by:
Ya. A. Mindlin, Dokladll Akademii Na"" SSSR, vol. 15 (1937), vol. 24 (1939),
vol. 26 (1940), vol. 27 (1940), vol. 42 (1944); Prild. Mat. Mekh., Akademiya Nault
88SR, vol. 10 (1946), pp. 229-240.
CHAPTER
VARIATIONAL METHODS
[(y) =
1(II"
... F(x,
y, yf) dx,
where F(x, y, yf) is a known real function F of the real arguments x, y and
yf "'" dy/dx. The value of the integral (106.1) depends on the choice of
y =0 y(x), hence the notation [(y). We shall use the term functioru'Jl to
describe functions defined by integrals whose arguments themselves are
functions.
To Illake the symbol I(II) meaningful, it is clearly necessary to impose
some restrictions on the choice of the argument y(x) and on the pre377
378
scribed function F appearing in the integrand of (106.1). We shall suppose that the admissible arguments y(x) belong to a class 0' and 8B8Ume
at the end points of the interval (Xo, Xl) the specified values yo and Y1.
Thus,
y(xo) = yo,
{ y(X1) = Y1,
(106.2)
+ E1I(X),
fi = Vex)
(106.3)
==
E71(X).
(106.4)
I(y),
+ .,,)
I(y).
or
(106.5)
I(y
The left-hand member in the inequality (106.5) is a continuously differentiable function of E, and therefore a necessary condition that 1I(x)
minimize (106.1) is
I_.0 "".
O.
1/'
dI(y + ..,)
dE
(106.6)
But.
l(y
+ ..,') dx,
379
VARIATIVNAL METHODS
dI{lI: e7/)
e-O
= /.""
('qFv
:h
+ Fv"{) dx = O.
AI!
%0
/.""
::to
'"''
/.z....
f/ _,:_' dx,
u;c.;
dF
.. , (
dF)
"" F. - ~ ,,(x) dx =- O.
/.
Thus the integral (106.8) vanishes for every flex) of class C' satisfying the
condition (106.4), and we conclude thatl
F - dF" = 0
(106.9)
dx
'
d'y
F v'.' dx 2
,dy
+ ]i.,.
dx + F". -
Fv = 0
p(X)(y')'
+ q(x)y' + 2f(x)y,
:x (w') - qy -
f .. 0,
whose solution satisfying the end conditions 1/(Xo) = 1/0, 1/(Xl) = 1/1 actually minimizes the integral
1(1/) = j",.
r'" [P1/'2
(106.12)
whenever! p(x)
(106.13)
1(1/) =
~ -
Xo
A functional of the form (106.12) arises in the one-dimensional elastastatic problems concerned with the study of deflection of bars and strings.
Similar calculations performed on the functional
Xl -
jz.
F~ -
dxF,;
~
+ dx2F,;'
...
+- (-I)" dxRF~R)
= 0,
leu) =
F(x, '/I, u, u z ,
~)
dx d'/l
on the set lu(x, 1/) I of functions of class C', where each u(x, '/I) in the set
takes on the boundary C of the region R specified continuous values
U = <p(8).
We suppose that F, viewed as a function of X, '/I, u, Uz , ~ is
of class C' in the appropriate domain of definition of these arguments.
Let us suppose that a certain function u(x, '/I) in the admissible set
actually minimizes (106.15) and that every function 11(x, '/I) in this set
is included in the formula
u(x, 1/) = u(x, '/I)
+ t7J(x, '/I),
+ E7J) I
dt
.-0
= 0,
+ E7J)
61
F{x, 1/,
so that
(106.16)
(F.7J
2'}3.
381
VARIATIONAL METHODS
We rewrite (106.16) as
&I =
II
(F.. -
a:;. - a:;.)
'1 d.x
dy
It
II [tx
(FUzfl)
+ ;Y (F...fI)] dx dy
It
H =
a:;, - a:;.)
(Fu -
fI
We get
dx dy
"R
fI(Fu.dy - Fu.dx).
F" _ of
(lOtU8)
ax
fI
in R,
iJF". = 0
iJy
leu)
U z .,
u ..) dx dy,
(106.21)
leu)
[(us)'
+ (u.)t + 2f(x,
~ ~(8)
on C.
V'u = f(x, y)
inR.
382
We can easily show that the solution of this equation, satisfying the
boundary condition (106.22), makes the functional (106.21) an absolute
minimum on the set of all functions of class C2 taking on C the values
11'(8). Indeed, let l1(x, y) be any function in such set, and we define
,,(x, 11) by writing
u(x, y)
u(x, y)
+ 'I(x, y),
[Cu.
+ '1.)2 +
(uv
-g
[(U.)2
= 2
(u.'1z
+ 1t.'Iv + f'1) dx dy +
[('I.)'
+ (Uy)' + 2fu) dx dy
('1.)2) dx dy.
so that
AI = -2
=
ff
'I(V'u - f) dx dy
[('I.)'
[('I.)'
+ ('I.)'] dx dy,
+ (".)'] dx dy,
011
C.
Moreover, the equality sign in (106.24) holds if, and only if, 11 = u.
For AI = 0 if, and only if, 'I. = 'I. = 0, that is, when 'I = const. But
" = 0 on C, and hence" "" 0 ill R.
101. Theorem of Minimum Potential Energy. In this section we
introduce an important functional, cal1ed the potential energy of deformation, and prove that it attains an absolute minimum when the displacements of the body are those of the equilibrium configuration. This theorem lies at the basis of several direct variational methods of solution of
elastostatic problems.
Let a body T' be in equilibrium under the aetion of specified body and
turface fomes. The surface forces T; may be prescribed only over a por-
VARIATIONAL METHODS
the surface ,;, and we shall suppose that over the remaining
part ,;.. the displacements are known. Denote the displacements of the
equilibrium state by Ut, and consider a class of arbitrary displacements
u; + 6u;, consistent with constraints imposed on the body. This means
that, over the portion 2:.. of 2: where the displacements are assigned, the
functions 6u; vanish, but, over the part 2:T, the 6U; are arbitrary save for
the condition that they belong to class C' and are of the order of magnitude of displacements admissible in linear elasticity. We shall term such
arbitrary displacements 6u; the virtual displacements. The virtual work
au performed by the external forces F, and T; in a virtual displacement
6u; is defined by the equation
tion
';T of
(107.1)
and we recall from Sec. 26 that the strain energy U is given by the
formula
[26.8]
==
I. W d...,
where'
[26.16]
The strain energy U is equal to the work done by the external forces on
the body in the process of bringing the body from the natural state to
the equilibrium state characterized by the displacements u;.
Since the volume ... is fixed and the T; and F, do not vary when the
arbitrary variations 6u; of the displacements are considered, (107.1) call
be written in the form
(107.2)
au
au =
(107.3)
"I. W d....
a (I.
W d... -
o.
This formula states that the expre88ion in the parentheses has a stationary
value in a class of admissible I'ariations 6u; of the displacements u. of the
equilibrium state.
1 The considerations of this and following sections need not be restricted to isotropic
bodies. If the body is anisotropic, we use W - Hc,;e;ei (i, j - 1, 2, . . . , 6), or
W - H,.,e;, where
c,;e;. The essential point in the proof is that W is a positive
definite form.
T, -
384
(107.5)
V =
J. W dT - I]; T.1/,;.
dtr -
J. F,u; dT,
o.
oV =
But
Hence
and, therefore,
.6.W =
- p.e,;eij.
.6.W = M(oU;)"
+ 2",e,i(ou,),j + P,
where
(107.8)
A
== '2l(ou;U2
r
+ 4:p. \(&Ui),;
+ (llu;),,1 2 ~ O.
oe;;
= 72[(OU;),j
+ (OUj)"J
ADo,;
+ 2",e'h
Since
Tij
= O.
+P
VARIATIONAL METHODS
+ J. P dT
J.
J 'E T"V'
'11 ou; du J.,. T-""1'
(T,;
J.
au;).; dT -
T'j.;
+Q
otl; dT + Q
au, dT
where
(107.10)
Q""
J. P dT ~ O.
= -F,
= T,
in T,
on ~,
and, therefore,
(107.11)
l::.U =
l::.V = aU - J % To,,dO"
- JFoudT
>0
<Nt.
T"'-
on this set.
Wc recall formula (107.9) for l::.U, insert it in (107.12), and obtain
- J. (T",;
+ Fi )
oU; dT
+ J:E (TijV;
+ Q~
O.
The contribution from the surface integral in this inequality is zero, since
on the part ~T of ~, where the T, are assigned, Ti;Vj = To and over the
remaining part, ~.., the au. vanish. This is so because of our hypothesis
concerning the character of admissible functions U; + ou;. Consequently,
(107.13)
'riM
+ F ... 0,
for every interior point in T. For suppose that one of the equations
(107.14) is not satisfied at some point P of 'r, and for definitcn('.Rs let
at P.
(107.15)
(k 2
> 0),
+ F,) ou,; dT
= k 2M,
~k'M
Ac-
+ Q;::: o.
over the entire surface l:, a'lld the body forces F, vanish throughout T, then
the formula (107.12) demands that AU;::: O. In other words, the strain
energy U has, in thls case, all absolute minimum on the set of arbitrary
junctions u; satisfying the boundary conditions and subject only to the differentiability restrictions ensuring the satisfaction oj Saint-Venant's compatibility equations.
PROBLEM
Consider the ease in which there are no body forees and the displacements u; are
preeeribed _ r the entire surface 2:. Show from Eq. (107.11) that the increment AU
in attain energy is positive in this ease And that the equilibrium displacements yield.
1\ minimum value for thp strain enel1O' U.
387
VARIATIONAL METHODS
(108.1)
+ Fi
'TijVj
Ti
Ui
=f,
T:j
in T,
on 2:T,
on 2:".
of class C2 in
T,
(108.2)
with the following properties:
l. in the interior of r, the T:j satisfy the equations
(108.3)
M-ij
satisfy the
It should be observed that the To; are associated with the equilibrium
state of the body and hence they satisfy the Beltrami-Michell compatibility equations, but we do not assume that the M-i; satisfy any such
conditions.
We consider now the strain-energy density W in the form- (26.17),
[26.17J
where 9 =
T",
+ tl.W
where 89 ""
"'ii.
T:j
= Tij
IT
+ M-ij.
IT
- 2E (9
We have
09)1,
388
l+cr
l+cr
= ~ ri/(Ori;) -
cr
- 2E [26 3a
+ (89)S)
cr
E e 89 + W(Ori;) ,
where
cr
E ea.,
= I1;j
+ W(Or.;)
+ 'l4.) Ori; + W(Orij)
(108.7)
AW = to; Or;;
= ~(u,.;
= u,.; Or.;
= (u, Or,;).;
+ W(Orij)
- u,(Or.;).; 1; W(Or'j).
The displacements u, figuring in (108.7) are those of the actual equilibrium state of the body, since the ro; were assumed to satisfy the BeltramiMichell compatibility equations.
The increment AU in the strain energy U is, therefore,
AU = iAWdr
+ W(Or,j)] dT
u.(6r'i).; dr + J. W(6r,,) dr,
W(Or,j) dr
~ O.
389
VARIATIONAL METHODS
tJ.U -
V* = U -
T,,dtr
'-"l
1
;E. . .
r;;,;
+ Fo = 0
TijJlj
in r,
on 2:.
T,
1.
To; -
i e8
i;)
81'0;
d1'
1.
W(81';;) dr,
J. B;;Fo;dr + Q ;::: 0,
I
Bee (108.6).
390
where Q is essentially positive and the Bij are expressions formed from
the derivatives of the T,;. Next a special choice of the F,; is made which
violates this inequality if one supposes that Bij F- 0 at some point of T.
The conclusion is that
J. TijJU' dT
J. (T.:,'lLD.; dT - f. T'i,,'lL,
=
dr,
'See also V. D. Kliuahnikov, PriI:l. Mat. MMh., AJcademilf4 Noule 8IJlJR, vol. 18
(1954), pp. 250-252.
t See, for example, E. Reiasner, JOV1'II4l oj M~ lind PApke, vol. 27 (IlKS),
pp.159-1GO.
VABL\TlONAL METHODS
391
(109.1)
With the exception of the last term, the integrand is obviously symmetric
in the primed and JInprimed variables. But the last term can be written,
by interchange of i, j, as
and we see that the integral, and hence the original expression, is symmetric with respect to the primed and unprimed states. That is, we can
write
(109.2)
This theorem can be expressed in words.
RECIPROCAL THEOREM OF BETTI AND RAYLEIGH: If an elaatic body ill
lubjected to two aystems of body and aurface fO'Tce8, then the work that would
be done by the fir8t ay8tem T" F, in acting through the displacements u; due
to the second ayatem of fO'Tce8 i8 equal to the WO'Tk that would be done by the
second aystem TI, F; in acting through the displacements u, due to the first
aystem of force8.
The Reciprocal Theorem can be written in terms of the stresses and
strains by modifying Eq. (109.1). We observe that the integrand on the
right-hand side of (109.1) can be written as
+ ~(U:.i -
ui.i)]
But the last term vanishes l on account of the skew symmetry of the rots.tion components, "':, = - "'Ii' Hence we have merely
1'iJUtl = l',,-e~,
I
or
This can be _
by interehanging i, j.
Thus,
with
+ U,., + UII.'.
But the final state of the body is independent of the order in which the
loads are applied.! Hence we must have
U", +
UIl,II
U"II = U II ,II
U",
UII"
or
that is,
+ IT Fiu; d."
The value of the Reciprocal Theorem lies in the fact that, for every
choice of values for the variables of the primed state T I, F I, ui, one
obtains a theorem relating the applied forces T i , F" and displacements
U; in an elastic body. This is exemplified in the problems at the end of
this section and in the theorems we now proceed to derive as special cases
of the Reciprocal Theorem.
We write out now an important specialization of the Reciprocal
Theol'flD, to the case of a body deformed by concentrated forces. For the
1 It sJ;tould be noted that the Reciprocal Theorem depends only on the linearity of
the equations of equilibrium and hence on the principle of superposition. We assume
thAt the forees Pi, T, do not depend on the displa.eements ..; and that the displacements u, do not affect the forees P:, T;.
VARIATIONAL METHODS
393
(109.4)
Let the beam be loaded only by concentrated forces PI, P., ... applied
at the points XI, X" ,and denote by a'j the transverse displacement at
Xi due to a unit transverse force applied at Xj.
We choose for the load
system p the concentrated force PI; then the corresponding displacements
y at XI and
are
x.
Similarly, for the load p' in (109.4), we take the force Po; the associated
displacements y' at the points Xl, X. are
The quantity IX;; is called the influence coefficient (designated so by Maxwell) for transverse deflection at Xi due to a force applied at Xj. The
symmetry of the influence coefficients
F~
., F.,
T.
= { T.
2:',
on }; on };',
+ AT,
and get1
(109.6)
Now, by (27.1), the strain energy U associated with the original load
system T.. F. is
2U =
J. F~u~ dT + f
J. F.u: dT + f
t'
'
't
.%
T~u~ dtr
%:
T,u: dtr
S'l
T:, F: is
%'
(AT.)u: dtr = 2U
or
2(U' - U) "" 2 AU
+ u:)2:'.
Then
dU
d!l; ..
U;.
It is lI8III1IXled that the body is rigidly supported and henee that the aupportinc
eo- do DO work.
C. B. Biueno and R. 0 - .
395
VARIATIONA.L METHODS
PllOBLEMS
1. Consider a beam loaded by concentrated forces P r at Zr and P. at z" and let 1(1
be the deflection at XI' Calculate the additional dellections d'llj and the change dU
in the strain energy "corresponding to a change dP I in the force PI. Show that
au ==
o'U
BP! aPt'
U=
where 6 is the end deflection '11(1).
Z'
3EI
6EI P ' = 21,6',
Show that
dU(6) _ P
da
.
:=; .,.;,
1-20-'
= 3, e'/" -E--
a'h
and
,
Uj
T;
T';i"i =-
Jli)
1-20-
(109.2), and derive the following expression for the change in volume AVo in an elastic
body under the action of surface forces T, and body forces F,:
40. Fill in the details of the following direct calculation of the change in volume AV.
of an elastic body T under the action of surface forces T, and body forces F,:
396
.. Show that the average value of a strain component eu, say, througbout an elaRie
body subjected to surface forces T, and body forces F" can be found fron>. the Reciproca.! Theorem in the form (109.3) by putting T:, - 1, other Tij - O. Derive the
formula.
J.
eu dT -
dT.
II
(<TXT
~_()
+ "I/T ) d:t dy +
II (-..",... -
.-1
"YT
+ /T,,) d:t dy
- -E
.-l
.DO
which is valid for the stress system in any beam free of body forces and loaded at the
ends. Verify the Reciprocal Theorem by taking the longitudinal displacement wand
the stresses T'j to be those of the problem of bending by end couples.
8. Define the strain deviations
by
e;i
,Ma,j
eij -
thu.t is,
~-~+~,
+ e;l;
~-~t~
e::IJJI = e. lI ,
ell' = e".,
e'::IJ
~~~+~
== e ,
and hence that the strain-deviation ten.or represents a change in shape without a
change in volume.
Show that the principal strains eI and the principal strain deviations e; are connected by the relations
Hint: The principal strain deviations are the roots of the determinant&! equation
1.;1 - .'B'II - O.
Define the "tress deviation
by
"':1
"'I a=
or
--~+~
T;II)W ==. 'T" IP
J,~e8'i
+ r:f
~-~t~
Ttf_
==
Til.'
-;~+~
T'/kZ
==
T.*,
397
VAIiUATlONAL METHODS
1"; .. 1'l
!oie,
1'111 -
~e.
+ e + e..)'
+ e:. + e:.) -
". ... (e
= 3(e!.
(e .. - e )' - (e -
e..)' -
(e" - )',
and show that the strain-energy density can be written in the form
W= W,+W,
where
W, = 7lk".,
W. =
and where .;; is the strain-deviation tensor. Show that W, depends only on the change
of volume, while W. is that part of the strain-energy density arising from a change of
shape. We call W, the strain-energy density of dilatation and W, the stram-energy
density of distortion.
= 21";; (see Prob. 9), show that the strain-energy density
11. From e = 3kD and
W can be written as the sum of the strain-energy density of dilatation W, and the
strain-energy density of distortion W" where
r:;
W, - __!_ e'
lSk '
W. =
1~1' [(T..
- Tn)'
-
r ..)'
The strain-energy density of distortion has been used as a criterion for failure of the
material. See S. Timoshenko, Strength of Materials, vol. 2 (or Theory of Elasticity,
p. 1(9), and A. and L. Foppl, Drang und Zwang, vol. 1, See. 6.
lS. Consider a beam stretched by a longitudinal stress p uniformly distributed over
the end sections. Show that the strain-energy density of dilatation W, and that of
distortion W. are given by
p'
W, - lSk'
J!.!. _
W. _
61'
+ ..)
2(1
W,.
1 - 2..
W,-o,
W!
--
r'
1(
2; .. 2;
T.., + , )
1". . . .
l3. Show that the strain ~!lergy stored in a beam of length I (parallel to the %-&Xis)
hent by end couples M. can be written as
U -
J. Wdr - It
T
W.
M'J
--cia:
- -.
281.
281.
UO. lllustrative Eumples. As an illustration of the use of the minimum principles in deriving the equilibrium and compatibility equations,
we consider several specific problems.
J~J(x)y!lx,
V = U -
where the strain energy U is equal to the product of the tensile force T by
the total stretch e of the string. But
e=
J~ (ds
- !lx) =
J~ [VI
+ (y')2
- 11 !lx,
and since we are dealing with the linear theory, (y')' 1, and we can
write
e =
Consequently,
E
Tll
U = 2
and, finally,
V =
(y')2!lx.
(y')'!lx
fa [~T(y')2
'
- J(x)y] !lx.
This functional has the form (106.12), and we see from (106.11) that the
appropriate Euler's equation is
Ty"
+ f(x)
O.
This is the familiar equation for the transverse .deflection of the string
under the loadf(x).
b. Dejlectibn of the Central Line of a Beam. Let the axis of a beam of
constant crass section coincide with the x-axis, and suppose that the beam
is bent by a transverse load p = J(x) estimated per unit length of the
beam. As is customary in the technical theory of beams, we suppose
that the shearing stresses are negligible in comparison with the tensile
stress
My
1"-1
399
VAlUATIONAL lUCTBOJl6
2 'rue..
W ==
M'y'
= 2EIs'
The strain energy per unit length of the beam is found by integrating
over the cross 8et'tion of the beam, and we get
JR W da =
%EI(y")2.
The total strain energy U is got by integrating this expression over the
length of the beam, and we find
U =
f~ %EI(y")2 dx.
We suppose that the ends of the beam are clamped, hinged, or free so
that the supporting forces do no work and hence contribute nothing to
potential energy V. If we neglect the weight of the beam, the only
external load is p = I(x) and (107.5) then yields
J; %EI(y")' dx - J: f(x)y dx
= f; [%EI(y")l - f(x)yJ dx.
V =
This functional has the form (106.13), and the Euler equation, therefore,
is
d'
dx' (Ely") - f(x) = o.
c. Deflection oj an Elastic Membrane. Let the membrane, with fixed
edges, occupy some region in the xy-plane. We suppose that the membrane is stretched so that the tension T is uniform and that T is so great
that it is not appreciably altered when the membrane is deflected by a
distributed normal load of intensity f(x,y). The situation here is
analogous to that considered in the first problem of this section. As in
that problem we first compute the strain energy! U.
Now, the total stretch e of the surface z == ,,(x, y) is
e=
(dtr - dxdy) =
(y'u;
+ u: + 1 -
1) dxdy,
I StrietIy ~ing we Are calculating th4! ine_ in strain energy OYer the coosta.nt
T. Since U. is coosta.nt and we a.re'eoncemed with the variation of the total energy, we can disregard U..
~ U.produced by thestretehing forces
400
v:
== ~
and hence
u =
Therefore
V =
U + u:)
g[~ +
(u!
(u!
rk dy.
-lex,
u;)
y)u] rk dy.
rv'u + lex, y)
0, and we
o.
='
-azy,
azx.
V* =
u-
Z.
Till; d0-
'
the surface integral must be evaluated only over the ends of the cylinder,
since the external forces are known over its law.ral surface, We recall
that
u=
w=
we find
1. W d,.,
+ T,..e.,,).
401
VAlUATIONAL lOlTIIODS
and hence
i,.l
u=
(T:c
+ T:.) dT.
fa T."U; dtr
since u = " = 0, for z = O.
On the end z = I, we have
fa T."U; dtr
(T U
= 0,
+ T'IIIi) dx dy
Thus,
(110.2)
V = ;,.
U+
(T:.
T:.) dx dy - al
[f
(XT.., - YT ) dx dy.
(110.3)
cos (x, v)
T ..
on C.
cos (y, v) = 0
av
(110.5)
T ..
=-p,C<ax
av dy
( ay
lis
+ av dx)
oX dB
'
or
dv =0
lis
V = ,,;"
[(''t.)'
= -2
inJl,
which is precisely the equation for the Prandtl stress function we got by
(ri".
ax dy -
2[f ~ ax
dy.
But
ax dy
80
that
V* ... ,,;11
[(V~)2
cos (x, v)
V =
,,)] d8,
4~] ax dy
(110.7)
+ y cos (y,
,,;21 ff [(V'it)2 -
,,)
~ =
0 on C and, for
4'itJ ax dy.
I(u) =
where the admissible functions u(x, y) belong to class C' and satisfy on
the boundary C m R the conditions
(111.2)
du
= g(8),
dll == 1(11).
VARIATIONAL METHODS
u(x, y)
o;(x, y) = u(x, y)
+ ''1(x, y).
61 = al(u
+ Efj)
a.
and find
01 = 2
(111.3)
ff (VuV", -
I.-0.
1'1) dx dy.
The first two integrals in the right-hand member of this identity are in
the form to which the Divergence Theorem is applicable, and we get,
[f
V'uV'l dx dy =
fe
V'u
~ ds -
fa'! a~:u ds +
'lV'V'U dx dy.
H = 2
[g
(V-4u - f)'l dx dy
Ic
V'u
~ ds -
Ic
'1
a~:u ds
~
vanish
on C, we conclude that
(111.5)
Vu = I(x, y)
in R.
(V'u -
n" dz dy ... O.
r V'u~da _ r
Jc
0"
Jc
'I
o(V'u) ds = 0
0"
:~ ;4 0 on C,
it
V'u = 0
onC.
o(vtu) = 0
0"
onC.
VARIATIONAL METHODS
405
(112.1)
to be minimized on the set of all functions y(x) of class CI, such that
y(xo) = Yo,
(112.2)
Y(XI) = YI.
Let the exact solution of the problem I = min be y = y*(x), and let
= m be the minimum of I(y) on this set. If, for some admissible
function y(x), the value I(y) = iii is close to m, one would suspect that
y(x) is a good approximation to the minimizing function y*(x). Moreover, if a convergent sequence of functions ii .. (x) (n = 1, 2, . . .) is
determined for which
lim I(ii .. ) = m,
l(y*)
n-
y*(x).
Guided by such considerations, Ritz proposed to construct iitx) by choosing a family of functions
(112.3)
depending on k real parameters a.., where '{' is such that for all values of the
a.. the end conditions (112.2) are satisfied. Then, on inserting (112.3) in
(112.1), one obtains I (ai, a., .. , ak), which can be minimized by
. determining the values of the ai from the system of equations
01 = 0
oa,
'
(r
= 1, 2, . . . ,k).
If the solutions of this system minimizing ley) are ai, then the minimizing
fun"tion ii(x) is
ii(x) = '('(x, ai, a., . . . , ak).
(112.4)
1
1~..(x; ~. ;,,(x: ~1: ~2"
.. ,a.),
where the family 1/.... ".(x, aI, as, . . . ,a.)' includes in it ali functions
in the families with subscripts less than k. The parameters (I; in each 1/can be determined so as to minimize the integral 1(1;). We denote them
by a. and the minimizing functions by
(112.5)
g.(x) = ",,,(x,
(n = 1,2, . . .).
Since each family Vk(X) includes the families Y._1(X) for special values of
parameters (1;, the successive minima I(,U.) are nonincreasing and we can
write
I{ih)
I(ys)
....
Since the infinite sequence II (Y ..) J is bounded below by m, it is convergent, but it need not converge to the minimum I(y*) = m; In
order to ensure the convergence of the sequence I I (Yn) I to 1(1;*), one
must impose Bome restrictions on the choice of functions '(>; in the set
(112.4). These restrictions pertain to a special character of approximation of every admissible function y(x) by functions of the set (112.4).
DEJl'INITlON: Let Vex), of class O' in (xo, x,) 8atisfy the end conditions
1/(xo) == 1/0, y(x.) = y.. If, for every E > 0, there exist8 in the family
(112.4) a function y:(x) = y .. (x, at, a~, . . . ,a:) such that Iv! - 111 < E
and Iv:' - 11'1 < Efor all x in (xo, x.), then the set of junctiofi.s (112.4) i88aid
to be relatively complete.
We prove next that when the Bet (112.4) is relatively complete the
sequence 19,,1, defined by (112.5), is such that lim I(Y .. ) = I(y*) = m.
Indeed, when the set (112.4) is relatively complete, there exists a function 1/:(x, a~, a~, . . . , a::) that approximates arbitrarily closely both
the exact solution y(x) of the problem I(y) = min and the derivative of
1/* (x). That is,
(112.6)
E,
Iy:' -
y*'1
< E.
<E
for Xo :S x :S x.
Consequently
I(y:) - I(y) =
%'
[
."
and therefore,
(112.7)
I(y:)
407
VARIATIONAL :METHODS
<
l(y")
+ E',
nomials
f a~(l
k-l
- x).
k-l
complete follows almost directly from Weierstrass' Theorem on Approximation of Continuous Functions.'
PROBLEMS
1. Show that the system of equations iHjJ(Yn) = 0, j - 1, 2, . . . , n, for the coeftii
n
2:
i-I
1(y) = / ; (P1/"
+ q1/' + 2/1/) th
= min,
1/(0) - 1/(1'1 - 0,
{!
"
10 (PY.'Pi
+ qy.'P; + I'P;) th
- 0,
(j - 1, 2, . . . , ..).
<..
~ "'COS (h-z/ll.
"-I
y'(z}l < .. buta.lao 1P.(z) - ,,(z)1
b.+
!n~hod
to the problem
10 ley')' -
yt - 2z1l1 dz - min,
11(0) - 11(1) - 0,
+ aoX + . . . + 1JI.x~').
Determine the a. in fh and ii., and compare the approximate solutions with the exact
solution 1/ .. (sin. x/sin I) - x of the Euler equation.
I. The deflection of a cantilever beam of length I, carrying a uniform load p per
unit length of the beam {see example Sec. nObJ i8 given by
EIII(x) = H.&p(x' - 4Zx'
+ 61':c'),
provided that 11(0) - 11'(0) = 0, 11"(1) = 11"'(1) - O. Note that the end condition
1/"'(l) - 0 is satisfied by choosing Ely'''(x) = c cos ("",/21). Integrate this, and
obtain
Ely (x) .. - 2cl
- - -2Z ( x - 2Z.
- SlU "."')]
... 2 . .
,,21.
[x.
Compute the potential energy V corresponding to this choice of deflection, and determine c by minimizing V. Show that the maximum deflection found from the formula
80 obtained is 1/(1) - 0.12603pl'/EI and the maximum bending moment
M(O) = -Ely"(O) = -O.469pI'.
y. -
cd,ex),
i-I
where the functions /lex) I!&tisfy all the boundary conditions. Show tht.t, if the ends
of the beam are clamped, hinged, or free, then the minimizing condition can be written
iDthef_
a~
..
to
[(HI'll::)" -
pl!, dz - O.
,. In the problem of & ealitilever bent by an end 1* P, tab for the .-umed do&c"OQrftthe~
HIy(lII) - _.
+ ..,
409
VARlATIONAL METHODS
o.
113. The Ritz Method. Two-dimensional Case. The special technique of constructing minimizing sequences, discussed in the preceding
section, can be easily extended to apply to functionals in the form
(113.1)
J(u)
Un(x, y)
(n
:= 1, 2, . . . ),
with parameters a;, and suppose that each family Uk(X, y) includes in it
families with subscripts less than k. We further suppose that the set
(113.2) is relatively complete, 80 that for every admissible u(x, y) there
exists a function u!(x, y) = CPn(X, y, ai, .. ,a!) belonging to the set
(113.2) such that
(113.3)
lu! -
u: <
E,
ax - ~~
ax 1< '
Iau!
E
au~
\ ay
_ au
ay
I<
410
leu) =
[(VIu)2 - 2fu)
ax dy
u = 0,
(113.6)
QU
QP
on C,
='
closely every admissible u(x, y) together with its first and second derivatives.
If we form l(u.) and determine the at so that
(JI~(Uft)
Vaj
= 0 (j = 1,2,
fJ (VIu.V'tpJ B
q!
iI
ftpj)
ax dy
(a;V2""V'tpj - f'Pi)
ax dy =
0,
(j = 1, 2, . , n).
i-I
L" a.tpt(X,
y),
.1:-1
and from the assumed rEtlative completeness of our set of coordinate functions it is easy to show that 1('Ii.) ...... l(u), where u minimizes (113.4).
Moreover, it turns out l that in this problem the hypothesis of relative
This
completeness suffices to establish convergence of !1i..} to
may seem odd because the corresponding hypothesis does not suffice
to establish the convergence of {u.1 to u for the simpler functional
11.
UM + u;) ax
411
VARIATIONAL METHODS
[10(\.21)
ff (u; + u! + 2uf) dx
dy,
OIle
considers the
f)') dx dy,
ff (u~ +
u!
.-1
u = U
un L',
ia
_0,
(j - 1, 2, . . . n).
"
412
(114.1)
o.
1,." Iy.(x) -
y(x) I'
ax,
and when 0....... 0 as n -+ 00, Yn(X ) is said to converge to VeX) in the mean.
This is a weaker type of convergence than the uniform convergence,
based on the criterion lim En = O.
Practically all recent investigations of convergence of various direct
methods in variational calculus and numerical analysis are based on some
such criterion as (114.1). The most effective tool in such investigations
proved to be the theory of operators in Hilbert spaces.'
The difficult problem of estimating errors in successive approximations
by the Ritz method was treated quite fully by Krylov' in connection
with those variational problems whose Euler's equations are linear and
ordinary. The corresponding problem, relating to multiple integrals, is
vastly more involved. Some important results were obtained by
L. V. Kantorovich and are recorded in his Uspekhi paper. No recital of
references on approximate methods in applied mathematics can fail to
include a superb monograph by L. V. Kantorovich and V. 1. Krylov,
1 See, for example, papers by K. Friedrichs in M aU!.ematuche Annalen, vol. 1O'J
(1934); American JOUT1llJI of Mathematics, vol. 68 (1946); Annals of MaUi.ematics,
vol. 68 (1946), which contain fundamental contributions to the problem of convergence of the Ritz a.nd related methods. An excellent account of the role assumed by
functional analysis in applied mathematics is presented in a long paper, in RIl88ian,
by L. V. Kantorovich in vol. 3, No.6 (28), of Uapekhi MatMnatichukikh NauJr. (1948).
Two monographs by S. G. Mikhlin, entitled Direct Methods in Mathematical Physics
(1950) and The Problem of the Minimum of a Quadratic Functional (1952) (both in
Russian), provide a systematic treatment of the sub}ects indicated by the titles and
include extensive bibliographical references. These books presuppoae .)lne familiarity with Lebesgue integration and linear function apacea.
N. M. Krylov, Lea Mllthod... de solution approach lie des probl~mes de 1& physique
matMmatique, Mbnorial du ,~ matMm4liquu, Paris, No. 49 (1931).
413
VARIATIONAL METHODS
L(u)
=0
in R,
i- 1
where the 'l'i are suitable coordinate functions and the ai are constants.
In regard to the 'l'i we shall suppose that they satisfy the same boundary
conditions as the exact solution u(x, y) and that t he set l'I'i} is complete
in the sense .that every piecewise continuous function f(x, y) can be
L
N
approximated in
by the sum
CJ'I'i
j - l
(115.3)
can be made as small as we wish.
The finite sum (115.2) ordinarily wi llllot satisfy Eq. (115.1), and th..
substitution of u. will yield
L(u.) = . ,(x, y),
,.(x, y) . 0
iu k.
If max E.(X, y) is small, u.(x, y) can be considered a satisfactory approximation to u(x, y). Thus, E.(X, y ) can be viewed as an error function, and
the task thel! is to select the a; so as to minimize E.(X, y).
'B. G. GaIerkin, Vutnik lnz.henerov, vol. 1 (1915), pp. 897-908.
'The GaIerkin method can also be applied to nonlinear problems. See D. V.
Panov, Prikl. Mat. Mekh., Akademilla Nauk SSSR, vol. 3 (1939), pp. 139-142.
As an illustration see system (116.10) .
414
i-I
ft
'=
0...",
then the
i-I
orthogonality condition
If
(115.4)
asn->oo
L(u.)'P'(x, y) ax dy = 0
(i = 1, 2, . . . , n),
L(u.)'P'(x, y) ax dy = 0,
g (1
L
(i = 1, 2, . . . ,n),
a;'Pi) 'Pi ax dy = 0,
aa;
We shall verify this for the problems associa.ted with the functionals
(106.21) and (113.4). It is important to note that in Galerkin's formulation there is no reference to any connection of Eq. (115.1) with a vari,
ational problem. Indeed, the Galerkin method can be applied to a broad
class of problems phrased in terms of integral and other types of functional equations. a
1
~(x,
1/) -
~(x, y),
i-I
all . - .. ,
",<p;,
If
But the
415
VARaTlONAL METHODS
leu) ==
(u!
u = (I
on C,
(j = 1,2, . . . ,n).
But
g(~~~~ + ~~. ~)
d:cdy =
-11
dxdy
VU.IP; dx dy.
dy
Je ~.(au.
ax
since
~i =
0 on C.
JJ
au.
dx)
ay
= 0
(V'u. - f)IP; dx dy
(j
0,
= 1, 2, . . .
,n).
But this is precisely in the form (115.5), since the Euler equation associated with this problem is L(u) == V'u - I = O.
In a similar way it follows that the system of Ritz's equations
[113.7]
[f (Vu.V~j
- lIP;) dx dy
0,
(j = 1, 2, . . . ,n),
connected with the functional (113.4), can be cast in the Galerkin form'
(Vu. - J)IP; dx dy = 0,
(j = 1, 2, . . . ,n).
The Galerkin method has been used widely in Russia. A bibliography of its numerous applications to problems ill dasticity is contained in a synoptic paper by Perelman.'
I See problem at the end of See. 113.
Note the identity,
- r v~.
II({ VuV.., th d7l" ij(f vV'u dz d71 + Jc( V'u!!!!.d8
Ib
Jc 8
Ya. I. Perelman, Priid. Mot. MekA. Akademiua Nalik 888R. vol. 5 (1941),
pp. 345-358.
416
1. Recast the system of equations in Prob. 1, Sec. 112, in the GaJerkin form.
t. Solve Prob. 2, Sec. 112, by the Galerkin method.
ff
I(ir) =
If R is a rectangle
Ixl
A,
Iyl
the form
(116.3)
J'0 fA
-0
BS)a,
-A
+ 2)'P' dx dy = 0,
A ')a, + 2](x' - A ')(y2 -
(V2irl
+ 2(x'
B2) dx dy
= O.
+ B2)a, 5
al =
'3%A 3B' = 0,
4A' + B'
We discarded the odd powers of x and 1/ because 'I'(x, 1/) is obviously an even
VARIATIONAL METHODS
417
D =
21'V ~dxdy
and the maximum shear stress T occurs at the mid-points of the longer
sides. If B > A, then (Sec. 38)
(116.6)
On inserting
~,
.-0
z-A
-1")(
(()~)
ax x=A
.-0
1+
+ a.x' + a.y')
+ a.(x' + y2) j,
1'a
'418
O.675pQCl.
. ...
.~i-l.3,5
2:
~i-l.3.5.
, . ,
which is a known expression for the Prandtl function in the double series.
This series, however, converges rather slowly. 1
We saw that the Ritz method provides an upper bound to the exact
minimum l(ti*) = m. A lower bound for the functional (116.2) was
obtained by Friedrichs,! who proposed a device of setting up an auxiliary
variational problem J(w) = max such that max J(w) = min l(ti). The
determination of upper and lower bounds in the solution of the Dirichlet
and Neumann problems (for simply and multiply connected domains)
was made, among others, by Diaz, Greenberg, and Weinstein.'
As another illustration of the application of the Ritz-Galerkin method,
we consider the problem of defortnation of an elastic plate by a parabolic
distribution of tensile forces over its opposite sides.'
1 Tables of numerical values of D, and ...., obtained by using the approximations it.
in the form of trigonometric polynomia.Is, appear on pp. 22(}-221 of S, G. Mikhlin's
Ditect Methods of Mathematical Physics (1950).
I K. Friedrichs, N ac/arichten von der Gesell8chaft der W iJluMch.aJIen zu GlJUingen
M~P"lIBikoliBcM Klaue (1929), pp. 13-20.
"J. B. Dia.z and A. Weinstein, American Jou7'1llll oJ M~, vol. 70 (1948),
pp. 107-116; H. J. Greenberg, Jou7'1llll of Mat4efltQtic$ and PAlIM, vol. Z1 (1948),
pp. 161-182.
A systema.tic procedure for construoting monotone sequences of upper and lower
bounds for quadratic functionals is developed in a memoir by J, B. Dia.z, &mioorio
M~ de ~ Coll:to_ Mat1>ematka, vol. 4 (1951). This memoir contains extensive bibliographical referenoea.
See also H. F. Weinbe -er, Jou7'1IIIl of M~ and PAyBia, vol. 32 (1936),
pp.64-62.
This problem wall first considered by S. Timoahenko, P~ M~
vol. 47 (1924), p. 1095. See also pp. 167-170 in Timoahenko and Goodier'. ThOOl.7
of Elaetioity (1961). A modified treatment of it, presented here, is contairu!d in 1.. V.
KAntorovich and V. I. Krylov'. Approximate Method8 01 Numerica.l Ana1,ysia (1960),
lIP- ~ This book _tains many Ulut.rationl of ~he _ of dheot methods in,
419
VARlATIO::>!AL METHODS
11 =
o
y--b
FIG. 61
to the law -r= = 8(1 - y'/b') (Fig. 61). The state of stress in the plate
is determined by the Airy function U(x, y), which satisfies the equation
V'U =
(116.7)
o.
-r..
= U,JlJI'
= -U...,
-ro. = U. H
0,
{ U.%JI
U... - 0,
y.)
b' U . = 8 ( -b-'-
on
U ... = 0
on y = b.
x = a,
u=
Uo
+ U,
where
(116.9)
420
u,.,,:: u,w::
In order to fit the problem (116.10) in the pattern of Sec. 113 {Eqs,
(113,5), (113.6)J, we co,nsider the system
(
28
l'
U,% = 0
Vu =
(116,11)
U =
U
= U,. =
on
on
Ixl =
Iyl =
a,
b,
and observe that, because of the naturo of the region, the function
satisfying the system (116.11) also satisfies the system (116.10).
Accordingly we are led to the problem
leu) =
fb f.
[CV'U)2 -
11,
~~ 11,] dx dy = min,
The coefficient
fb
0.1
is determined by
t.
L(uI)(x' - a')2(y' - b
O)' dx dy
"" 0,
UI =
11,0
a')'
421
VARIATIONAL .I4ETHOl>S
leu) =
II F(x, y, u, U., u
y)
ax dy
= min,
Un =
I.-1
akl"k(X, y),
where the 1". satisfy the same boundary conditions as those imposed on
u(x, y). We then determined the coefficients a. so as to minimize l(u n ).
If we now suppose that the a. in (117.2) are no longer constants, but are
unknown functions of x, such that the product a.(x)I"'(x, y) satisfies t.he
same boundary conditions as u, we are led to minimize
n
l(u.)
I [
.-1l
a.(.'t)I"'(x, y) ].
Since the 1". (x, y) are known functions, we can perform integration
with respect to y and obtain a functional
(117.3)
l(u.) =
f. "'
>,
Ct[a.(x), aL(x),
xl dx.
leu)
1"(8)
on C.
422
Let
(117.6)
u ..(x, y) = "O(X, y)
'i ak(X)"~(Z,
y),
i-I
"0 Ie = ,,(8),
and the 'I'.(x, y) vanish at least over II. part of the boundary C. On the
part of C where the ~ do not vanish we shall require that the a.(x)
vanish. Thus,
(117.7)
onC.
Unless the ".(x, y) vanish over the entire bounaary C, we shall be obliged
to impose certain restrictions on the shape of the region R in order to be
able to satisfy the boundary conditions imposed on the a.(x) in (117.7).
We introduce next, in the manner of Sec. 106, a. fa.mily of varied
functions
U.. (E, x, y) "" ,/,o(x, y)
(117.8)
L"
+ ~lI.(X)".(x, y),
[ak(x)
1:-1
where the
(117.9)
I(EI, ... ,
E..) =
1[,,0
(a.
+ E.lI.)I".],
i=1
(117.10)
211 [~:. a:
(9'kll.)
O.
But
ff [~:. a:
=
(l"kllk)
+ ~~"
a:
(l"kll.) ] dx dy
if[!(iJa:l"lllk)+~(iJ;;'''I''l:lIk)1dxdY- i f ~.lI.dxdy
. -LI9'l:lIk ( -
~';" dx + ~~ dY)] -
[f
VAlUATIONAL ltETRODS
188
= 1,2, . . .
that
,n.
N ow if the region R bas the form shown in Fig. 62, we can write the double
I
I
I
---r--------r---________
I
I
~I~--.~
I~-b
I
I
I
r
FIG. 62
".(x) tk
[.V"h(Z)
II-,I
()
(V'un
Z
!)rp.(x, y) dy = 0,
h(")
[.I(Z
Inasmuch as
k = 1,2, . . .
,n.
424
f~B IV'[(yt -
+ 2}(y2 -
Bt) dy = 0,
+ 2al(X) + 21(y' -
B') dy = O.
B')al(x)]
or
f B [(y2 -
.8')a;'(x)
-8
a~'(x)
2~2 alex)
kx
.
kx
+ c, smh
B- 1,
2~'
= 0,
whose solution is
a,eX)
= c, cosh B
= cosh (kAIB)'
Thus,
(117.15)
'Y
= ( -
1]
I116.41
al
= %(A 2 + B~)
got by the Ritz method in Sec, 116. In formu1a (116.4) the term
al(x' - A') replaces the expression in the brackets of (117J5), and it is
essentially the first term in the power series expansion of the function
in those brackets.
If we use (117.15) to compute an approximate value Dl with the aid of
(;{S.lO), we find
Dl
fr bOa!, (1 - :k tanh
k:),
(2k
1)1JT
'Yn = '-' a.(x) cos -2~
.1:=1
VARIATIONAL METHODS
and obtain the exact solution of the type (3S.15) by a.llowing n to increase
indefinitely. In this case computations simplify considerably because
of the orthogonality of cosines.
The method of Kantorovich has successfully been applied to numerous
problems in elasticity, including some three-dimensional problllms. 1
118. The Trefftz 'Method. We have noted in Sec. 113 that an application of the Ritz method to the problem 1(1.1) = min, where 1(1.1) is a quadratic functional, yields a minimizing sequence {I (1.1,,) I, which approaches
the minimum 1(1.1*) from a.bove. Should we succeed in constructing a
sequence I I(v,,) J approaching 1(1.1*) from below, then
(l1S.1)
and we may get a good estimate of the minimum 1(1.1*), if the bounds iLl
(l1S.1) are sufficiently close.
In 1925 Trefftz2 suggested a mode of constructing a minimizing
sequence of lower bounds I(v.) (without proving convergence) for the
Dirichlet integral
(l1S.2)
1(1.1) =
(1.1:
+ u~) dx dy,
V'u
1.1
=
=
in R,
on C.
0
'I'(s)
dv
(1.1* - v) - ds
dv
1)
1.1* = v
0..
by the formula.
+ 1),
A bibliography of such work will be found on pp. 321-373 of the Kantllrovieh and
Krylov monograph cited in this aection. This monograph also includfle a ca.reful
study of convergence of several direct methods.
E. Treft'tz, Mathemati8cJle An1l4len, vol. 100 (1928), pp. 503-521.
See footnote 3 on page 429.
426
and then
(118.5)
l(u) =
[(0.
,. U +
(v:
v;) dx dy
If (,,! + ,,:)
+
dx dy
2!!
(0.".
If (v.". + v.".)
If
dx dy = -
'IV"v dx dy
+ fe 'I ~ ds.
}e Cu
(118.7)
dv
v) (bd8 =
80S
l(u) = lev)
o.
+ [('I),
(118.8)
Moreover, the equality sign in (118.8) holds if, and only if, 1(,,) = 0,
that is, when " = const. In this case u * = v + const.
We are now in a position to construct a sequence of lower bounds l(u,.)
for l(u"). Let {v.1 be a sequence of harmonic functions v. dl'lfined in the
closed region R, and construct a sum
(118.9)
u"
===
La,Vi,
i-I
""
u* - u. on the boundary
w, 80 that
427
VAlUATIONAL )lETHOns
L. f
i-I
a;[rp(s) - u..]
idvv ds =
5: 1(1.1*).
0,
!c
[rp(s) - u.]
or
( [rp(s) -
(118.11)
Jc
~; ds
i a.v,] ~:
= 0,
ds = 0,
(j
1, 2, . . . ,n),
(j = 1, 2, . . . ,n).
i-I
(118.12)
where l
_
(118.13)
dv;
a,; = Je Vi dv ds,
fl,
55
Je '1'(8)
dv
d: ds.
TIle system (118.12) has a unique solution whenever 1a.;1 0;& O. Suppose the contrary, and assume that 1a.;I = O. Then the homogeneous
system
1"
(118.14)
ai,-aj
=- 0,
(i = 1, 2, . . . ,n),
i-I
has solutions
Note that
Clil
=-
ail,
V"'s .. 0.
428
where
of OIiJ ==
OIj.
1L"
d;v;
dv;
d. ds == 0,
(1j=1
or
r - dUn d
(118.15)
JeUnTv
0
== .
Un
dd~n ds
ff
(VUn)' dx
dy.
Un
i=l
ff {V(u -
Un)2 dx dy
R
n
where Un =
Sl.rmma,'hle sn:U6.Tt!
429
VARIATIONAL :METHODS
4.;
oJ = 0 ,
oa;
(.J
= 1, 2, . . . ,n).
Ol(u - un)
((
a
au.;
= 2
V(u - u.) . au; V(u - Un) dx dy
Jl
-2[IV(U-U
211
dxdY
(u - u.)V'Vj dx dy - 2 fa (u - u.)
II
= -
n )''(/Vi
(u - Un)
~~j ds,
1(a (u
dv
- u.) d: ds = 0,
~; ds
Vi
are harmonic
(j = 1, 2, . . . , n),
480
'Y'y, = 0
{ y, = }.i(x2
(119.2)
in R,
on C.
+ y2)
J(y,)
J! (V.p)2 tb dy = min,
R
(116.2]
ff [(V'lft)' -
4'1ft] dx dy
D = 2"
[f 'l'* dx dy,
if
R
==
(V'Ift)2 dx dy
V'Ift vex'
+ y2) dx dy +
(x'
+ y2) dx dy.
{(V'l')2 - 4'l'] dx dy
U+
(x'
y2) dx d1/.
VARIATIONAL METHODS
431
ThUB
(119.4)"
where I. ==
(Xl
J(t) = 1(+)
+ II') tk dy
+1
0,
section.
The application of the same form of Green's Theorem also yields
80
(119.5)
D ""
I'
ff (Vi'*)' dx dy.
R
Trefft~
method.
+ a,(x -
6x'y'
+ y<).
get
j.,. 1,2.
432
Since v. = 1, we have in fact only one equation ' and we can obtain
another equation by requiring, for example, that the mean error vanish
on Ixl = A, Iyl = A. This gives
Ie [~(X2 + yS) -
.y,J d8
== 0,
10A '.
+ y2)
7
= 90 A' - 72A' (x' - 6x2y'
Thus
D, "" 21'
'.If,dxdy '"
+ y4)
1%35~'
'"
2 (x 2 + II').
O.1407I'a4,
(120.1)
in H,
u == 0,
iJu
ill'
=0
on C,
I(u) ...
HV'u)2 - 2fuJ dx dy
= min.
433
VARlATIONAL METHODS
in R
V'v = J(x, y)
(120.4)
oj claS8 Cl 1'n R
+ C.
Then
I(w) 2: -
(120.5)
If v' dx dy.
R
[f
wf dx dy =
=
=
ff
wV'v dx dy
vV'w dx dy
ff
We have
+ Ie (w ~~ - v~~) ds
vV'w dx dy,
by virtue of (120.2).
I(w)
But
v' dx dy
II [(V2W)' -
2Jw
+ V2] dx dy
It
[(V'w)' - 2vV'w
Jf (V'w -
+ v'] dx dy
v)' dx dy 2: 0,
It
l(u*) 2: -
fj v' ax dy.
Thus, if one obtains any solution of the Poisson equation (12004), the
right-hand member of (120.6) provides a lower bound for I(u*). To
1 z_
(120.7)
'Pi'I'1
dx dy =
and constructs
(120.8)
VA '"" V -
6v,
a;'Pi,
i_]
where
(120.9)
a;
ff V'P' dx dy
R
But,
-g
~ -
[f v; dx dy.
g I r
ft
v! dx dy = -
(v -
- ff
a;'P'
dx dy
It
dx dy
+ ff (2V
R
a;'P') dx all
i-I
it
1ar,
(2V ita,'P') dx dy = 2
[f (.t1a.'P'Y
dx dy
so that
(120.10)
al,
VARIATIONAL METHODS
485
&nO.
(120.12)
u*(x, y) = -
II [v(~,
'I) -
of equation L(u)
= 0 was defined as
(121.2)
(x, y)
= L(u.).
ff L(u.)<p;dx dy
0,
(i = 1, 2, . . . , n),
IS
[f .:(x,
y) dx dy == min,
least~.
436
(12L4)
ff .~:~
This lead!;
(i = 1, 2, .. , n).
= 0,
To ensure that this system have a solution and the resulting sequence
(unl converge to the solution u(x, y) of L(u) = 0, one must impose some
. restrictions on the operator L and on the choice of coordinate functions
Such matters have been studied by Krylov ' and his followers. In
constructing the approximate solutions for linear equations, Krylov
selects a complete set of coordinate functions, with suitable differentiability properties, such that each 'Pi satisfies assigned boundary conditions. Recently, Mikhlin2 developed a least-squares method in which
the coordinate functions 'Pi satisfy the equation L(u) = O. The coefficients a; in (121.1) are then so selected that they minimize the integral
of the square of the error _.(x, y) in the boundary conditions. This
approach, suggestive of the procedure in the TrefItz method, often leads
to more rapidly convergent sequences {u~ I than in the usual leastsquares method.
As an illustration of procedures followed in the standard least-squares
method, we sketch the determination of torsional rigidity of a square
beam. As our first approximation, we consider the function
'Pi.
(121.5)
fleX,
y)
= V'WI
=
+2
+ 2al[(X'
A')
(y' - A2)J
1",
turn out to be
a"'" 2.4,
while the exact values recorded in the table of Sec. 116, are,
D =- 0.14061Ul\
1"
= O.675IU>a.
437
VARIATIONAL lIlE'l'1l0DS
Among sundry formal devices proposed for minimizing the error function is the method of coZlccatwn. This method requires that the error
function vanish at n specified points in the region R. In this procedure
the error En(X, y) is "collocated," or assigned, at n points of the region,
n
Il;
in
Un
ll;<Pi
are obtained
i=l
A ')(y' - A ')[a,
= (x' -
= '2(A/2,
A/2)
25
a,
+ a.(x + y')],
O.
Then
2
a, = 21A<
42A"
D.
= 0.1407pa"
'1'2
0.690paa,
a"'" 2A.
= -2
'=O
V2Q;
in R,
on C,
while the membrane deflection z(x, y) is found from the relations (46.1)
V'z
=- ;
z= 0
in R,
on C.
-.}in
by writing
't',,'"
ll;'I'i(X, y),
;,-1
with
<Pi'"
on C.
The approximate stress function 't'.. will not, in general, satisfy the
438
differential equation (46.3), and Vi'.. will equal not -2 but sorne function l P.(x, y).
The function P.. (x. y) can be interpreted, in terms of the membrane
analogy, as the nonuniform pressure necessary to constrain the membrane
to take the form defined by the function v,,(x, y).
If the collocation method were applied to this problem, one would
demand that the "approximate loading function" P.. (x, y) equal the
"true load" of -2 at n points of the region R. Instead of such local
conditions, one may impose a different set of conditions. The region R
may be divided into n regions R., over each of which it is required that the
total approximate load equal the total true load. That is, the n coefficients a; are to be determined from the n conditions
[[ V2i'. dx dy
or
V2i' dx dy,
V'v n dx dy
(121. 7)
(i = I, 2, .. ,r.),
B2)
f: foA (x' -
A2
+ y. - B2) dx dy
Equa-
= -2AB,
E.
"" -",-01.
V......
P.(z,".
489
VAlUA'l'IONAL ME'l'HODS
and
!I' D1 = !3 1 +(~!G)f
'b - 01667 4
(bla)' G - .
a,
(square section),
1
3 (bla)!
(bla)' a = 0.700,
I'a (r ....h = 2 1
(square section),
R lI :
0::; x::;"2'
R 12 :
"2::;
R .. :
2::; x::; A,
0::; y::;"2'
A
x::; A,
"2::;
y::; A.
ff
V'w.dxdy
It'i
-2
ff
dxdy
= _
~2
Ril
ff
it. dx dy = 0.1413a'
and is 0.5 per cent greater than the exact value, 0.140&'.
shearing stress is given approximately by
-
p.a
The maximum
(r....). - 0.6716,
a vahle ~t is 0.6 per cent less than the exact result, 0.6700.
440
The method of this section was introduced by Biezeno and Koch, l who
have applied it to problems of thin plates and elastically supported
beams. Biezeno 2 has observed that this procedure may be applied to the
general problem of elastic equilibrium.
It is clear that various techniques of obtaining approximate solutions
presented above are but special cases of the general procedure of minimizing the error function by making it orthogonal with respect to some
function M('Pi), that is, by imposing the conditions
(12l.8)
II L(u.)M('Pi)
dx dy = 0,
(i = 1, 2, . . . ,n).
{~
in R i ,
elsewhere in R.
As noted in Sec. 116, when the M('Pi) are such that an arbitrary function
1)(x, y), satisfying the required boundary conditions, can be represented
in the series of M('Pi) , with suitable properties, then the condition (12l.8)
may imply that L(u,,) -> L(u) = 0 and one can then justify calling
uA(x, y) an "approximate solution."
122. The Function Space Methods. The theory of diverse methods
of approximate solution of problems in mathematical physics was presented from a unified point of view by Kantorovich 3 in a paper which
shows clearly the growing importance of the Banach and Hilbert space
theories in the solution of concrete problems. Not only are the results
obtained from a general point of view more complete and incisive, but
they also lead to more effective methods of approximate solution and to
sharper estimates of errors than the special techniques.
The analytical concepts of functional analysis were cast in a geometric
I C. B. Biezeno and J. J. Koch, "Over een nieuwe Methode ter Berekening van
:"lakke Platen met Toepassing op Enkele voor de Techniek beIangrijke belastingsgevanen," De lngenwur, vol. 38 (1923), pp. 25--36.
C. B. Biezeno, "Graphical and Numerical Methods for Solving Stress Problems,"
Procuding6 of tM Firat lrdernaiional Congr_ of Applied Mechanica, Delft (1924),
pp.3--17.
C. B. Biezeno and R. Grammel, Technizche Dynamik, Chap. III, Sec. 9.
C. B. Bie6eno, "Over een Vereenvoudiging en over een Uitbreidina van de Metilode
VIID ruts," CItri8tiaan HU1/gen8 International MlJtIiemotiBch Tijd8cAriJt.. voL 3 (1923),
pp. 69-75.
L. V. Kantorovich, "Functional Analysis and Applied Mathemat.ica," U~
M~ Nauk, vol. 3, No.6 (1948), pp. 89-185. An Englishtranel&tio1tof
tbitI important pe.per was ~ for the National Bureau of Standatda, Rcpon
J509 (1952).
VARIATIONAL METHODS
441
form by Prager and Synge! and effectively used by them to construct
bounds for exact solutions of elastostatic problems. These authors deal
with a real linear function space F, each point of which is a function To;
representing the state of stress in an elastic body T. The set of functions
Tij can be thought to define a point P in F or, alternatively, a position
vector T from the origin 0 (T,; = 0) to P. The vectors T are required to
satisfy the familiar laws of addition, multiplication by scalars, and scalar
multiplication of ordinary vector algebra. The square of length of T
is defined by the formula
(122.1)
T .T
[f
Tijei; dT,
e,;
= - E Di,e
l+u
+ -e
Tij,
1.
!J;,'T:; dT,
satisfying the reciprocity relation (Sec. 109) T T' = T' . T. Consequently, the notions of directions, angle, orthogonality of vectors, etc.,
can be defined in a manner familiar from Riemannian geometry, and one
can use the suggestive language and ideas of elementary geometry as an
aid in visualizing relationships among different states of stress.
Thus, if one considers a set S of all functions Ti; of class C', this set
contains a subset 8 1 of those T,; which satisfy the equilibrium equations
in
(122.2)
T.
eN.;; -
eik,jl -
ejl,i'
=0
T;j
in T.
Likewise there are subsets Sk containing the T,; that satisfy various types
of boundary conditions on the surface of T. The exact solution t,; of a
given boundary-value problem consists only of those Ti; which are found
1 W. }'racer and J. L. Synge, Quarterl1l of Applied Mathematics, vol. 5 (1947), pp.
241-269.
J. L. Synge, Q1uJrter11l of Applied Mathematics, vol. 6 (1948), pp. 15-19; Proceedi.
fI/' the Boyal Irish Academy (A) vol. 53 (1950), pp. 41-64; Rendictmti di 1fIQ.temIJtU:4,
Home, aero 5, vol. 10 (1951), pp. 1-21.
t Although we write this law for an isotropie body, it is not essential to restrict thct
theory to iaotropic elastic media. Any law of the fonn e'l - "'11iIT.. can be UlIed 80 long
1III'r,lf<i is & positive definite form.
442
(1.1).
I
.Diu CI1ll...,..
&1ld the
443
VARIATIONAL METHODS
oped princip&lly because of the need for coping with problems of this
sort..
The main disadvantage of all numerical techniques is that they give
numerical values for unknown functions at a set of discrete points instead
of the analytical expressions defined over the initial region R. Of course,
when the boundary-value data are determined by measurements at a
finite set of points of R, the difference-equations methods may be the
best mode of attack on the problem. Any analytic technique would
require fitting curves to the discontinuous data.
We proceed to the outline of the general procedure followed in reducing
the given analytical boundary-value problem to a problem in difference
equations. For definiteness let the region R be bounded by a simple
closed curve C, and we seek to determine the function u(x, y) satisfying
some differential equation in R. From the definition of partial derivatives it follows that'
(123.1)
au
ax
a2u
ax'
+ h, y) h
lim u(x + h, y) lim u(x
u(x, y),
h~O
~ =
axay
h_O
1:->0
and so on.
For small values of hand k the partial derivatives are nearly equal to
the difference quotients appearing in the right-hand members of formulas
(123.1). If one replaces derivatives in the given differential equation by
difference quotients, there results a difference equation which is a good
approximation to the given equation when hand k are small.
Thus, to Laplace's equation
(123.2)
(123.3)
+ u
= 0,
where
u"., =<
1lw
!lIE
1
Ii} [u(x
+ h, y)
- 2u(x, y)
+ u(x
+ h)
- 2u(x, y)
+ u(x, y
Ii} {u(x, y
- h, y)],
- h)].
444
C'
Fw.63
445
VARIATIONAL METHODS
(124.1)
'" =
g(s)
inR,
on C,
We lay down a square net over the region R, assign known values to the
net points on the boundary, and approximate (guessed) values at interior
points. The differential equation of the system (124.1).is replaced by a
difference equation, which we proceed
y
to derive.
In the neighborhood of any interior
point of R (taken, for the moment, as
the origin of coordinates), we can write
",(x, y) = "'0 + aloX + aOlY
+ a2oX' + a02Y' + auxy + . . .
(124.2)
= "'0 =
FIG. 64
aOO,
while at the neighboring net points to the right and left one has (Fig. 64)
1 L. A. Lusternik, "tlber einige Anwendungen der direkten Methoden in Variationarechnung," Matematicheski Sbornik, vol. 23 (1926), pp. 173-201.
R. Courant, K. Friedrichs, and H. Levy, Mathemati8che Annalen, vol. 100 (1928),
pp.32-74.
1. G. Petrovsky, U8pekhi M<Jtematiche.kikh Nauk, No.8 (1941), p. 161. See also D.
Panov, Matematicheski Sbornik, vol. 40 (1933), pp. 373 393; U8pekhi Matelllatiche8kikh
Nauk, No.4 (1937), pp. 23-33.
S. E. Mikeladze, Izvestiya AkademH Nauk SSSR, Mathematical Series, vol. 5
(1941), pp. 57-74.
P. P. Yushkov, Prikl. Mat. Mekh., Akademiya Nauk SSSR, vol. 12 (1948), pp. 223226.
See a.lsQ Chap. V by W. Feller in J. D. Tamarkin and W. Feller, Partial Dilierential Equations (1941), pp. 160--196.
This section is taken, virtually without changes, from the first edition of this book
It was prepared for that edition by Dr. R. D. Specht.
'-0
+ a,eII' -
I!.nd
Similany,
1/12
+ 1/1.
== 2"'0
+ 2(l02h2 + 2ao.h' +
+ 2a02,
1/11
4"'0 = (V"'f)o
+ terms in h'.
A.s the choice of the origin of coordinates is not essential to the argument
above, the foregoing expression relates V'I/I at any point to the value of 1/1
at that point and to the neighboring values. We drop the terms in h'
and replace the Laplace ditIerential equation V"", ... 0 by the Laplace
difference equation
"'(x
+ h, y) + I/I(x -
h, y)
+ '4-(x, y + h) + I(X, y -
h) - 41/1(x, y) = 0,
or
(124.4)
Expressed in words, the value of ",(x, y) at any point is the mean of its values
at the four immediate neighboring points. This difference equation is
equivalent to a set of linear equations for the values of '" at interior poin~
in terms of the prescribed boundary values. The number of variables is
usually so large, however, that direct solution is out of the question.
Instead, we may resort to an iterative procedure.
The simplest (but most laborious) way of solving the Laplace difference
equation is to guess at the proper values for the interior points of the network; this guess is then corrected by traversing the net, replacing each
interior value by the mean of its four immediate neighbors. Repeated
traverses of the net will give interior values that converge to the values
of the solution function I/I(x, y). The convergence, however, is so slow
as to require almost unlimited manpower in order to secure sufij.ciently
accurate results. Fortunately, va.rious procedures are a.va.ila.ble .for
impr()ving the rapidity 0{ convergence.
Instead of expressing the v'alue of '" as the mean of the four immediat.e
VARIATIONAL KETHODS
'ltl
llla)
be used' (Fig.
'It,
- 4",0.
(124.5)
If both the immediate and the diagonal neighbors are used, then the
formula
(124.6)
ZOtPo
= 4(""
"'0
"'ii
(124.7)
"'00 = M6[D,
+ D, + D. + D. + D. + D. + D, + De
+ 2(M, + M. + M. + M.)].
The corner values, such as "'", are obtained from the diagonal neighbors,
so that
"'10
"',. =
448
This same procedure may be applied, of course, to find the value of '" on
any block of 9 points in terms of the surrounding values Mi, Di, Ci,
whether the latter lie on the boundary of the region R or not.
The method of finite differences will now be applied to the torsion
problem for a beam of square cross section with side length 2A. The
conjugate torsion function
y) is defiMd by
!/t(x,
V"'_
to
"'ex.
'" = 7i(x'
+ y')
inR,
onC.
~I
jlO
(124.8)
Q = 10'
Ds
10,000
2500
2500
2812
1718
2812
2812
1718
2812
1 1
.-
(1,!/t Q
I}
is subject
,,0
the
in R,
onX = 1,
onY=l.
A coarse net is now laid down over the square section (Fig. 67). Equation (124.7) gives .1)(0, 0) = 1250, while !1(.%,.%) = 2812 is derived from
its diagonal neighbors by (124.5) and U(7i, 0) = 1718 from its immediate
2500
10,000
10,000
Fm.67
449
VARIATIONAL METHODS
from a further application of Eq. (124.6) to the net values in the same
order as before.
.
The values found above can be checked against those given by the
10,000
2812
2500
2250
2214
1250
1708
1750
1718
1572
1634
FIG. 68
= 104 [1 + y. -
where N = (2n
are found to be
+ 1)11/2.
X' - 8
Jo (-;}t c~~~t:
11(0, 0) = 1787.4,
cos NX
and it is seen that the approximate values are in error at most by 2 per
cent.
The torsional rigidity is given by
; D
=2
U
ff
it dx dy
= A'
(1
=2
If ['" - ~
+ 1O-'{!) dx dy
(x'
+ y2)] dx dy
~ A.,
or
1
p.O'
= 12
10+ 16
1 11
1
-1
-I
ndX dY.
Th find the approximate value of the shear stress at this point, we pass II.
450
parabola through the points 1l(1, 0), 1'" 0, 0(1 - h, 0) ... 01, and 0(1 2h,0) .., 01, where It. is the mesh of the -net in the X, Y coo.mnatea_ The
slope of the parabola at (1, 0) is given by
(:~)~:6 = ih (Ill. -
4[1.),
[10-'
2h
(124.9)
(!l, - 40 1)
h =
72,
and
1
-2 T . . . -= 0.6196A,
pa
/1/1
-I
and
-I
The numerical
0 dX dY = 80,246,
9
1
-. D = 0.1391,
pa
n +"11(0),
where flO) is the error at the net point in the'lJolution of the difference (not
the differential) equation. A single traV'e1:&e oh:lle net yields an improved
value 0(1) and error f<l):
0(1)
+ 'Ell).
451
VARIATIONAl METHODS
\
other words, the differences 6(1) are improved in the sa.me way [for example, by Eq. (124.4)] as were the original values n, and successive traverses
yield the 'iml?roved differences 6(2), 8(1), . The final values of n are
obtained by adding to n(l) the improved differences:
For
and, in general,
Then
{lW
_ .(.
+ ... + (.(n)
_ .(0- 1
10,000
Q
2214
2241
2253
2258
Q
1750
1776
1788
1794
118(0)
[1799.8J
0
42
26
12
6
0
-36
'1:l
12
5
2500
(2263)
[2263.61
1634
1660
1672
1678
62
26
12
6
(1684)
(1663.81
FIG. 69
In Fig. G9, these values of n are shown in the left-hand columns, while
in the columns 011 the right are entered the differences 8(1), <1 (2) , 61' ) , 81f)
obtained by applying (124.6) to the differences ill the order (),2 , ),2) .
();2, 0), (0,0) . Two advantages of working with the difference function
6 rather than with the original function n are immediately seen. Both
the zero boundary values and the smaller number of significant figures
in the differences 6 make for easier computation. A third and more
powerful advantage is that use of the difference function 8 makes possible
the estimation of the effect of infinitely many traverses of the network.
That is. inspection of the successive differences 6 (2), 6 (3) , 6 1') suggests that
452
these and succeeding values may form a geometric progression of ratio Yil.
We hazard a guess that this is indeed the case and sum each infinite series
of differences, getting, in this case,
0(5)
+ 0(6) + 0(7) + . . .
20(5)
0(4).
obtain an estimate of 1]<") (see the fifth entry in each column on the left
in Fig. 69). The assumption that the successive differences 8<n) form a
geometric progression can now be tested by using (124.6) to improve the
values of I] just obtained (see the last entry in the left-hand columns of
Fig. 69).
Shortley, Weller, and Fried, in an investigation of the convergence of
the method of finite differences,' have shown that this extrapolation to the
limiting net value by summing the infinite geometric series of differences
is possible in general.
The final net values in Fig. 69 satisfy the difference equation (to within
one unit) but not the differential equation. This is shown by comparison
with the exact ordinates given above, and the disagreement arises from
the fact that, in setting up the difference equation, terms of higher order
in the net mesh were neglected [see (124.3)J. We proceed, therefore, to
decrease the mesh of the net to one-half its original size.
In the process of interpolation leading to the values of n at the new
net points, the difference equations (124.4) and (124.5) are used to ensure
that the interpolated values satisfy, at least approximately, the differential equation 'V"I] = O. The mean of the diagonal neighbors [Eq.
(124.5)J furnishes the values of 1](%, %), 1](%, ~~), and fl(~~, ~~). The
immediate neighbors are used [Eq. (124.4)J to get I](~~, 0), 0(%, 0),
OC).-2, ~~), and OC%, ),2). The resulting values are shown in Fig.
70. Without any further improvement, these values give 0, = 1226,
0, = 1683 [see Eq. (124.9)J, and
1
-2- Tonu
IJ.CX
= 0.601A,
which is 2 per cent below the exact value 0.675A. The integral of fl is
found by Simpson's rule to be approximately 9292, and the approximate
torsional rigidity is found to be D/(lJ.a 4) = 0.1414, a value 0.6 per cent
above the exact result of 0.1406.
Instead of traversing the lattice points in a fixed order and extrapolating to the limiting net value, as above, one can correct the lattice values
in any way at all. Indeed, all that is required is that one arrive at values
n for which the difference equation is satisfied-or, alternatively, for
1 G. H. Shortley, R. Weller, and B. Fried. "Numerica.l Solution of Laplace's and
Poisson's Eqnatiolls," Ohio Slate Fnit'er.ity Studies, Engineerinll: Ser. (1940), p. 18.
453
YAIUATIOK AL METHODS
which the diffprcnces 8 are zero. In this way, the experience and physical
intuition of the computer can be used to good advantage.
Another variation in the finite-difference method consists in replacing
the lattice with square mesh , used above, by a lattice formed of regular
polygons.'
The torsion problem of a beam of square cross section, considered
ahove, is a particularly simple one in that the square cross sect.ion
10,000
1799
4316
5625
2263
2672
2500
1857
1853
1611
625
1799
1683
1226
F,n.70
v'n
Q
= 0
given
in R,
on C,
454
a'v
ax'
a'v
(125.1)
+ 2u,
'U,
== r, = 0
'Ua
= r. = 0
u, + 2u. - u.
= r. = 0
- Ua + 2u. - u. = r. = 0
- u, u, = r. = 0
u,
+ 2u.
:"1
~U2
:"3
+ ..
FIG. 71
455
VARIATIONAL METHODS
It. . .
8,
!La
= 12,
1.1"
TI
= 14,
T.
!L6
= 15.
From (125.1)
we
find that
== - 1, T, = 1, r, = r, = O.
Since these residuals ri are not all zero, we shall improve the trial solution
by reducing them. We pick one of the numerically largest ri, say ra, and
bring it to zero by an appropriate change .Au. in Ua alone. It is clear from
(125.1) that a unit increase in u. (Au, = + 1) would cause changes only in
r.,r.,andr"andthesechangeswouldbeAr. = -l,Ar, = +2,Ar, = -1.
To make r, = 0 ralls for Ar, = -I, which we bring about by selecting
Au. == -0.5.
As by-products we have AT, = AT, = +0.5. Accumulating the rand
Ar, we find the residuals
= -1.0, T. = -0.5, r, = 0.0, r, = 0.5, '.
== 0.0. There is now a single numerically largest residual, r" and we proceed to "liquidate" it by selecting Au, = +0.5. Next time Au. = +0.5,
etc. Eight steps of this process are summarized in Table 1. An
experienced computer goes very rapidly, calculating mentally and recording a residual only when it changes.
'I
TABLE
l.
RELAXATION SoLUTION OF
i.
First guess of u, . .
4 .0
-1.0
Residuals r i
i
EQS. (125.1)
8 .0
12 . 0
14.0
15 .0
- 1.0
1.0
0.0
0 .0
Residuals
All,
- --
-0.5
+0 . 5
+0 . 5
-0 . 3
+0 . 3
+02
+U . 2
-0 .3
~' .
~nelit sOlution
-I.U
-0 . 5
0.0
0 .5
0 .0
0 .0
-1.0
0 .0
0 .5
0.0
-0 . 5
0.0
-0 . 5
0 .5
0.0
-05
0 .0
-0 . 2
-0 . 1
0.3
U. J
-0 . 3
-0 . 2
-0 . J
0.3
- U. J
U. J
-U . 4
-0 . 1
o.a
-0 . 1
-0 . 1
0 .0
-0 .3
0 .3
-0 . 1
-0 . 1
0 .0
0 .0
0 .0
456
457
VARIATIONAL METHODS
recommended the basic method and many of the standard tricks. Seidel'
proved it would converge for linear systems with positive definite matrices.
In the thirties Southwell' rediscovered Gauss' method and named it. He
and his school have developed the method and brought its wide appli cahility to the attention of engineers and scientists everywhere. The
method has proved especially suited to the analysis of complicated
redundant pin-jointed frameworks . These have equations like (125.1),
with more involved coefficients.
As a digression it should be noted
1
that one-dimensional problems like
2
3
4
5
that of Fig. 71 should not in practice
7
6
8
9
be solved by relaxation, for they are
10
easily solved by a fast trial-andrecursion scheme, even for variable
11
12
13
14
15
masses. For example, in (125.1)
change the last equation to read
16
17
- 1 - u,
+ 2u.
- u.
r.
o.
18
19
Fix r, = . .
= r. = O. By symmetry we should have 1/,. - U. = O.
20
21
Try up' = 3, From the modified
equatiQl1s (125.1 ), we find successively
/L~" = 5, U&" = 6, ?lll, = 6, U~" = 5,
FlO. 72
U~" = 3, whence u~" - U~" = -2,
Now try u\" = 6, We find suceessively u~2) = 11 , u~2l = 15, ul" = 18,
u~" = 20, u~" = 21, tI~2) - u~2) = + I.
Interpolating linearly betw~en
u\" and u\" to make U6 - u. zero, we get u, = 5, whence the true solution
is obtained rec ursively.
The point of Table 1 was to show the technique of relaxation in a simple
setting. The practical applications of the method begin with twodimensional problems- like trusses. Or, in closer relation to Fig. 71 ,
suppose one had an L-shaped network of light strings with 21 weights
on it ill a horizontal plane (Fig. 72), How might one calculate the
equilibrium position of the weights under large tension and under
gravity? With the same assumptions as above, we find 21 equations like
(125.2)
-1 - U2 - U6
+ 4U7
Us -
UI2 = r7 =
O.
1 Ludwig Seidel, "Ueber ein Verfahren, die Gleichungen, auf welche die Methode
der kleinsten Quadrate flihrt, sowie line"re Gleichungen liberhaupt, durch successive
Anniiherung aufzulosen," .4.bhandlungen mathematisch. physi8chen Klass., Bayrisrhe
Akademie der Wi.,.en8chaften (Mllnchen) , vol. 11 (1874), pp. 81-108.
'R. V. Southwell, Relaxation Methods in Engineering Science (1940) and &Iaxation Methods in Theoretical Physics (1946).
458
O!X
l
xxx
Oxx :
______ x_xJ _______ _
:xxx 0
'0
,
(125.3)
{l;j
0,
for i in B, j in B (i
459
VARIATIONAL METHOD>!
(f25.5)
a,j1(j
+b
ri
(i = I, . . . ,n).
j=1
For the moment we do not assume that the matrix ~atisfies (125.3) or
(125.4), but it is essential that, no aii = O. Iterative methods for solving
(125.5) have been popular since Gauss' time, if not longer. One pro('ess,
called the Seidel or Gauss-Seidel method, is the following: One solves t.h('
first equation (l25.5) for Ul, using the ('urrent values of U" , 11
Then the second equation is solved for 1l., using the latest known \'alues
of UI, U3, U., . ,1tn. And so on. All the equations (125.5) are solved
in cyclic order for U" ,un, always with the latest values of the other
unknowns. In other words, supp0se 'ult , , . , u~' are known. One
gets u\k+ I " , u~+l) by suecessively solving these n equations:
i
(125.6)
a,ju)'+!l
J",,-1
aijut'
+ b,
0,
(i = I, 2, . . . ,n).
j='+1
One hopes that the l;lk, will converge as k ...... <Xl to the Ui which solve the
system (125.5).
, The reader will note that the Seidel process is closely related to the
relaxation process described earlier in this spction. The difference is the
order in which Eqs. (125.5) are solved. In the relaxation mpthod we do
the equations in an order determined by the size of the ri' III the Seidel
process the order is fixed and eyeli(.
Let us analyze the behavior of the Seidel process. In our applieations
the following theorem, apparently first completely proved by Schmeidler'
in 1949, is essential:
THEOREM: 1f the matrix A is symmetric and positit'e definite, then in the
Seidel process the U)k' co/werge as k ...... ac to limits 1Ii (i = I, . . . , TI),
solving the system (125 ..5).
What happens when A is not positive definite? Or, in any ease, how
does the vector u(" approach u? If the cOIl\'ergenee is slow, how ean it.
be speeded up? The answers to these three qUCtitiOIlR ('all be obtaillf~d
through some use of matrix theory, as follows:
Let the matrix of coefficients A in (125.5), a:s~umed non~ingular, uc
written as the sum of three matrices: A = D + E + F. Here D has the
ma.in dia.gonal of A but is zero elsewhere; E has the below-diagonal
elements of A but is zero elsewhere; and F has the above-diagonal ele, Werner Schmeidler, Vortriige tiber Determinanten und Matrizen mit Anwendungen in Physik llnd Terhnik, .'\kademie-Verlag, Berlin (1949). Professor A. Ostrowski
has traced incomplet. proofs back to p, Pizzetti, .. ltti della rrale arradentia de; Un"i,
Re1tdiconti (4), vol. 32 (1887), pp. 231}2:~fi, 288293.
460
Thus, schematically,
(D
+ E)u"+1l + Fu,k) + b =
O.
(D
+ E)u + Fu + b =
O.
elk' =
u(k' -
denote the
+ E)e(k+1) + Fe(k) = O.
Now since no
= 0, the matrix D + E has an inverse (D + E)-I.
Letting H denote the matrix - (D + E)-IF, we find from (125.9) that
(D
(125.9)
a;;
(125.10)
whence
(125.11)
IH - pII = 0
and
I(D
+ E)p + FI
= 0
have the I:!allle /I real or ('omplex foots PI, . . . ,p". If alllp,1 < 1, thfn
in the Seidel process utkJ -. U. If any 11-';1 2 1, the Seidel process
diverges. In principle this settles the question of convergence.
For most matrices A, to each of the roots I-'i of (125.12) there corresponds a unique vector y(i) su~h that H y(i) = I-'iy(i). That is, the transformation H leaves the vector yli) unchanged in direction but stretches it
(11',1 > 1) or shrinks it (11-',1 < 1) to the fraction I'i of its previous length.
All these vectors y(i) form an oblique coordinate system, in terms of
which we can resolve the initial error vector e(O):
e(O\
L"
.. -1
c;y(i)
461
VARIATIONAL METHODS
<.
Yz(1 + u~k) ,
Yz (1 + U\'H)
= Yz (1 + U~k+lJ
= Yz(1 + u~'+))
= 1
u'f+o
U\k+ll =
U~k+ll =
U~k+l)
ulk+ ll
U~k+l)
+ u~'),
+ ulk),
+ u~k),
Several rounds of this are shown in Table 2. The residuals are not
shown, and one line of the table amounts to a full cycle of the above
algorithm.
..,\
w~:.t
errors
0 .50
0 .24
0.75
0 . 96
TABLE
2.
SEIDEL SOLUTION OF
.-~-
EQs. (125.1 )
12
14
15
- --
12
14
15
u (O
4.50
8.75
II .88
13.94
14 .94
I u(S)
4.88
888
11 .88
13 . 91
14.91
4 .94
8 . 91
11 .91
13 .91
14.91
u ( 4)
4.955
8 .932
11. 921
13.916
14.916
U ( 6)
4 . 966
8 . 944
11.930
13 .923
14 .923
u(1}
4 .9 72
8.951
1I .937
13930
14 .930
u('J )
4.9755
8 .9562
11 .9431
13 .93G6
14 .9366
uCo)
u(1)
0 . 92
0.909
0 .906
462
In the first column of Table 2 is given the ratio of the worst error of the
u, in the preceding and following rows of the table. By (125.13) this
ratio converges to 141. fA solution of either of Eqs. (125.12) gives
}Jo, = 0.9045, }Jo. = 0.3455, }Jo , = }Jo4 = O.J
The approach of u(k) to u is
one-sided and very regular. It will take about 22 eycles to gain one
decimal point in accuracy. Making educated guesses at u in such
problems is easy in desk work , if one knows (125.13) and its analog when
1}Jo,i = 1}Jo,1
The present availability of electronic digital computing machines
makes it possible to solve much larger problems than have been previously feasible. Such machines carry out arithmetic operations at all
effective speed on the order of 10' times faster than a human being with
a desk calculator. Something like 10' numbers of desk calculator precision can be held in a fast-access "memory" and made available as
rapidly as the arithmetic organ can operate. Something like 10' more
numbers can be held in an intermediate storage and transferred to the
high-speed memory in a few milliseconds. Moreover, current developments will probably have made the figures given here obsolete before this
book is published.
Because of the speed and capacity of such computers, many persons
want to solve their problems on them. It is pertinent to ask : What
methods will prove most feasible for the computers? While definitive
answers must await investigations as yet not made, certain indications
are now possible.
A first observation is that for large problemR of the type of (125.1) or
(125.2), iterative methods are relatively attractive, for much the same
reasons as for pencil-and-paper calculation. But the relaxlltion method
as outlined in connection with (125.1) has one considerable disadvantage.
The scanning of all the residuals r, in a search for max, hi is comparatively time-consuming. In fact, while computing r, it would take almost,
no extra till\e to solve the ith equation for u,. But if one solves the ith
equation for u, (i = J I . ,n), one is actually carrying out the Seidel
process, which is accordingly preferred in machine calculation to conventional relaxation.
A second observation is that solving a large system (125.5) by the
Seidel-method is likely to be slow. To speed UD the solution, acceleration
methods are needed, as indicated above. But acceleratiflns involve new
routines, new coding, and the mundane but important problems of storing
or reading in new codes. It is important with machines to reduce coding
and operating to the utmost in simplicity.
It is the remarkable discovery of Youngl that for certain problems a
1 David Young, " I terativ p Methods for Solving Partial Difference Equations of
Elliptic Type." Transadions of the A ....,.,:can Mathematical Society. vol. 76 (1954).
pp. 02 J J 1. The Bame 81lgl(rstion WIU! put forward a little earljer in a special case by
VARIATIO~AL
4f>3
METHODS
EU(HI)
and
U(Hl)
= u(')
We now eliminate
I)(k)
+ Dv(k) + Fu(k) + b =
+ fJ(v(k)
- u(k
from (125.14).
DU(Hl)
(1 - fJ)u(k)
+ fJv().
Since
(1 - fJ) Du 1k )
+ (3 Dvlk"
we have
(125.15)
Ett IH1 )
+ ~ D11 1H +
I)
(1 - ~) DU
1k
+ F'1t 1 + b =
k)
O.
I<1E + ~ D + (1 -~) D + 1= O.
F'
For matrices satisfying (125.3) and (125.4) and for a certain ordering of
the equations, it can be shown that the maximum of the /<T,/ is least when
~,)-I,wherei'listhelargestrootof(125.12).
wechoosefJ = 2(1
Hence this fJ defines the optimal a.mount to overrelax. Moreover, for
this {j a.ll 1<1,1 are equal.
Stanley P. Frankel, "Convergence Rates of Iterative Treatments of Partial Differential Equations," Mathematical Tables and Other Aids to Computatiun, vol. 4 (1950),
pp.65-75.
464
~{'\THE~L\T I C .\ L
THEORY OF EI,.\STICITY
3.
SOLUTIO" OF
(125.1)
BY SY:;TEMATIC OVERRELAXATIO"
Ratio
of
worst
errOrs
12
14
15
u(O)
12
14
15
U(l)
4. 76
9.34
12. 26
14.20
15.::H
U l2 }
5.39
9.32
12 .26
14 .33
15.34
u(l)
5. 039
9 .060
12 . 161
14.209
15.140
u( ' )
5. 025
9. 110
12. 159
14.119
15.1 08
U< l )
5 .071
9 . 118
12. 097
14. 093
15.085
U( I)
5. 053
9 . 052
12 . 059
14 . 061
15.048
u('J)
5. 012
9 .027
12. 036
14.032
15.024
uea)
5 .014
9 . 024
12 . 024
14.020
15.018
1.1,(')
5. 011
9 . 014
12 . 013
14 . 013
15 . 010
u (lO)
5 . 0049
9 . 0063
12 . 0079
14 . 0068
15 .0051
0.34
0.51
0 .54
0.76
0. 74
0.52
0 .59
0. 67
0 .58
0 .56
u (x
+ h, y) + u (x, y + h) + u(x -
h, y)
+ u(x,
y - h)
- 4u(x, y) = 0,
VARlATIOX.\L METHODS
165
of relaxation takes 8.5 sec. For the Seidel procpss the dominant eigenvalue 1-'1 = 0.99416. To reduce the error to 10-' times its initial value
by the Seidel process (fJ = 1) would require about 2300 cycles, because
approximately one has (0.99416)2300 = 10-'. This would take over 5 hr
on SWAC. If, however, fJ is taken at its optimal value of
2[1
+ (l
- 0.99416;]-' = 1.85802,
then the dominant eigenvalue is 0', = 0.85802. For this 0'" it requires
about 90 iterations, accomplished in only 13 min, to reduce the error from
I to 10-'.
Some practice with SWAC convinces people it is not difficult to estimate {J well enough in a few minutes. Actual running time to reduce the
error by a factor of 10-' is on the order of 20 to 30 min, including the time
necessary to determine {3 approximately. Similar experience is reported
by Young and Lerch.'
On SW AC the progress of the calculation can be monitored by observ-
I
n
E(k)
Ir;"1
When
i-I
U. A.S. BANGALOItE
UNIVERSITY UBRARY.
4 JAN i374
Ace
. NO..........
53~~~
we
CL. ,.,0.........................
PI-=
._
NAME INDEX
Abramyan, B. L., 165, 175,248
Aleksandryan, E. A., 164
Aimansi, E., 244, 266
Arutyunyan, N. K., 164, 175,206
Arzhanykh, 1. S., 358
Avazashvili, D. Z., 228
c"
,,~'::~-.
468
Lagrange, J. L., 2, 3
Lamb, H., 370, 375
Lame, G., 84, 299, 345
Lauricella, G., 89, 272, 314, 358
Lebedev, N. N., 327, 367
Lecornu, L., 33, 64
Leibenson, L. S., 124, 204, 239, 419
Lekhnitsky, S. G., 193, 194, 248, 326
Leonov, M. Ya., 353
Lerch, F., 465
Levin, Eo, 327
Levy, H., 445
Lichtenstein, L., 89, 358
Likhachev, V. A., 327
Lin, T. C., 159
Ling, C. B., 175, 190
Lourie, A. I., 242, 343, 353
Love, A. E. H., 6, 25, 29, 33, 40, 44, 61,
64,71,81,99, 107, 114, 119, 127, 134,
164,184,197,243,244,285,329,336,
341, 353, 357, 370
Lusternik, L. A., 445
Luxenberg, H., 248
Macdonald, H. M., 174
McGivern, J. G., 168
MacMillan, W. D., 119
MacRobeTt, J. M., 1~1
Maguerre, K., 367
Mandzhavidze, G. F., 327
March, H. W., 166, 197
Mariotte, E., I
Mason, M., 329, 336
Matuyama, T., 168
Maxwell, J., 336
Mecugov, V. H., 243
Michell, J. H., 187,244,285
Mikeladze, S. E., 445
Mikhlin, S. G., 272, 302, 313, 316, 326,
412, 414, 418, 419, 428, 436
Milne, W. E., 444
Milne-Thomson, L. M., 444
Minasyan, R. S., 243
Mindlin, R. D., 301, 334, 341, 361
Mindlin, Va. A., 376
Mises, R. von, 90, 130
Mises-Frank, 134
Mohr, 0., 51
Morera, G., 336
Morris, R. M., 155, 219
Moskovitz, D., 454
Murnaghan, F. D., 33
Mushtari, N. M., 228
Muskhelishvili, N. I., 4, 6, 91, 142, 143,
152,175,240,249,262,264,268,269,
272, 284, 285, 295, 296, 299, 306, 307,
314, 316, 326, 366
Myklestad, N. 0., 361
NAME INDEX
~adai,
A., 367
:\'ahari, A., 150, 151
",arodetsky, M. Z., 175, 300, 302, 326
:><arychkina, E., 376
Navier, C. L. M. H., 2
Nemenyi, P., 176,236
Netrebko, B. P., 193
:\"euber, H., 331
Neumann, F., 61, 359
Novozhilov, V. V., 4, 33
Nowacki, W., 165, 176,361
Nowinski, J., 193
469
470
SlJBJECT INDEX
Affine transformation. 6
iu6.uite~im.8.1,
472
SUB.TECT INDEX
Gradient, 22
Green-Gauss theorem, 22
Grooves, effect of, 12&-127
H&lf space (.ee Three-dimensional problems)
Harmonics, spherical, 345-350
Harnack's theorem, 143
Holder condition, 142 ,
Homogeneous media, 59, 65-67
Hooke's law, 1, 56
$..M-.I\..~.t.w 1P.~k .n....a.~~,..cJ' lunction, 81-85
generalized, 58--67
for homogeneous isotropic bodi~s, 66,
180
Hypercircle method, 440-442
Infinite regions, half space, 339--343,
372-376
(See 0130 Plane ela.stostatic pro\:'lems)
Influence coefficients, 393
Integral equations, 304-307, 313
Integral harmonics, 34&-347
Integrodifferential equations, 303~304,
307-313
Invariants, of strain, 18--20
of stress, 44, 48
Irrotational wave, 371
Isotropic media, 61 , 65-67
Kantorovich method, 421-425
Kronecker delta, 6".
generalized, 19
Lagrangian coordinates, 29-30
Lagrangian strain components, 31
Lame constants, 66
Laplace difference equation, 443
Laplacian, 22
Laurent series, 140
Least-squares method, 435-436
Legendre polynomials', 349-350
Limit, elastic, 57
.proportional, 56
Lipschitz condition, 142
Load, plane of, 211
point, Hl8
Local effects, 190-193
Maxwell's equations, 335-336
Minimum principles, complement..ry
energy, 387-390
potential energy, 382-386
strain energy, 386, 389-390
473
Moduli of elasticity, 60
bulk,70-71
compression, 70-71
for isotropic media, 65-71
of Lame, 66
number of, 3, 60-67
POisson's, 68
relations among, 71
rigidity, 68--69
shear, 69
values of, 70
Young's, 56, 68
Modu{us o{ 6'exura{ rigidity, [(/3
Mohr circle, 51-53
Morrera'. equations, 335-336
Multiconstant theory, 3, 60-61
Multiply connected domains, 297- 302,
313-314, 316, 318--326
Natural state, 81
Navier's equations, 3, 73
in curvilinear coordinates, 181
for plane problems, 251, 256
solutions of (8ee Plane elastostatic
problems; Three-di-nensional
problems)
Neumann problem, III
Nondeformable body, 5
Nonlinear elasticity, 29-33
references, 4
Notation, for strain, 22, 25
for stress, 37-39
Numerical methods, finite differcnces,
442-454
application of, to Dirichlet problem
445-448
to torsion of a beam, 448--453
Laplace difference equation, 443
references, 444-445
relaxation, 454-465
Gauss-Seidel process, 459-462
Seidel process, 459-462
Young process, 463-465
Orthotropic media, 63
Papkovich-Neuber solution of Navier's
equations, 331
Plane, of bending, 106, 210
of the load, 211
neutral, 211
Plane elastostatic problems, 257-327
boundary conditions, 270-272, 274276
boundary-value problems, 269-273
existence of solution, 272
uniqueness of solution, 272-273
474
SUBJECT INDEX
Shearing strain, 23
Shearing stress, 38
lines of, 116-117
maximum, 51
Sherman's method, 314-317
Simply connected region, 26
Solid harmonics, 347-348
Solution, existence of, 4, 88--89, 272, 358
uniqueness of, 25-26, 86-88, 272-273
Sphere, hollow, 343-345, 353, 362-364
solid, 350-353
vibration of, 369-370
SphericaJ coordinates, 184
Spherical harmonics, 345-350
Stability of columns, 2
Strain, analysis of, 5-34
components of, 11, 21
in curvilinear coordinates, 181
cylindrical, 179-180
orthogonal, 179-180
plane polar, 182
spherical, 184
Eulerian, 30-31
geometrical interpretation of, 12-14
Lagrangian, 31
transformation of, 15-16
examples of, 23-24
invariants, 18--20
notation for, 22, 25
plane, 24--25, 250-253
boundary-value problems, 251-252
equilibrium equations, 251
Navier equations, 251
(See also Plane el.."tostatic problems)
principal, 16-18
principal directions of, 18, 67
quadric of Cauchy, 14--16
examples, 23-24
shearing, 23
Strain energy, 83-86
and compatibility equations, 390
minimum, 386, 389
Strain energy-density function, 60, 83-86
Stress, actual, 56
analysis of, 35-55
characterized by Cauchy, 1
components of, normal, 38
shearing, 38
tangential, 38
examples of, 53-55
Invariants, 44
maximum shear, 61
nominal,56
normal, 53
notations for, 37-39
plane,54-55,~53-254
generalized, 254-257
(See also Plane elastostatic problems)
principal, 47
473
4.76
CI. No:;
~ ,.
?>WSO k
'1 0 OC1198(1
1 9 Jir~~~1-
~~\~
Ac.
No.5371<t
THE-ORY
OF ELASTICITY
IVAN S. SOKOLNIKOFF
','
A unified treatmen! of seemingly unrelated
great value to engin .;ers an{j mathematicians ccncr'~
with engin.:':"ring pro blems of d'? fo rmaille bodie~
des a chapter on ane ~vtical and nume~kal me
approximate solutioi l (Jf problems in e!::;stidty
tional and related technique,s.
NEW DELHI