SIMILARITY SOLUTIONS FOR PARTIAL
DIFFERENTIAL EQUATIONS GENERATED BY
FINITE AND INFINITESIMAL GROUPS
by
Henry S. Woodard and William F. Ames
Sponsored by Office of Naval Research Contract No. N00014-68-A-0196-0004
and
Project Themis
Contract No. DAAFO3-69-C-0014
uHR Report No. 132
Iowa Institute of Hydraulic Research
The University of Iowa
Iowa City, Iowa
July 1971
Approved for Public Release; Distribution Unlimited
a
ABSTRACT
The problem of developing systematic methods for obtaining
similarity variables is considered for partial differential equations. Similarity variables are a set of transformations which reduce a
partial differential equation to an ordinary differential equation. This paper considers two methods of generating similarity vari-ables. The first method uses a group of finite transformations and
the second uses a group of infinitesimal transformations. The
math-ematical theory for both techniques is described and illustrated. The two methods of obtaining similarity variables are applied to the Burgers' equation u + uu u and to the laminar boundary
y X XX
layer equations with a pressure gradient. In all cases considered, new types of similarity variables are found. In addition, the
aux-iliary conditions are discussed in the light of the new similarity
Chapter Page LIST OF FIGURES
LIST OF SYMBOLS
vi
I INTRODUCTION a.
II
SIMILARITY BY FINITE GROUP METHODS 9Continuous Groups g
Invariant Transformations u.
Absolute Invariants
Similarity 16
Boundary Conditions . 20
III
SIMILARITY BY INFINITESIMAL GROUPS . 22Infinitesimal Transformations ...
22Invariants 25
Canonical Variables . 26
Reduction of the Number of Independent Variables
in a Partial Differential Equation . . . .. 28
Similarity Solutions in Practice 31
IV APPLICATION OF THE FINITE GROUP METHOD 35
Burgers' Equation . 35
Laminar Boundary Layer
TABLE OF CONTENTS CONT'D.
Chapter Page
V APPLICATION OF THE INFINITESIMAL GROUP METHOD. . . 61
Burgers' Equation 61
Boundary Layer Equations 7
VI
CONCLUSION 88BIBLIOGRAPHY 91
LIST OF FIGURES
Figure Page
1.1 Heat Conduction in Circular Cylinders 2
1.2 Prcfile for Velocity in Boundary Layer 8S
£ = infinitesimal parameter
a1,a = finite continuous parameter = free stream velocity
= kinematic viscosity
u = characteristic velcitv
C
-u' component of velocity in the x-direction v* = component of velocity in the y-direction
r, 2, , X., g. = absolute
invariants
f. = functions which define a finite transformation
x, Y, U, V, X functior which define an
infinitesimal
t ran s forma t ion.
= a differential form
7, Q differential operators
I = invariant solutions
i = Bessel function of the first kind
/3
1
= Bessel function of the second kind
/3
P unknown function in the boundary layer
equations
J = Jacobian
a0 = identity element
CHAPTER I
INTRODUCTION
This manuscript concerns itself with the problem of developing systematic methods for finding new similarity variables for partial
differential equations in engineering. Similarity variables are a set
of transformations which, when applied, reduce the number of
indepen-dent variables in a partial differential equation. The advantage of
such a transformation is the simplification of the problem at hand.
Similarity is often associated with the physical nature of the
problem. For example, steady state heat conduction in a hollow
dr-cular cylinder is described by Laplace's equation Txx + Tyy =
where T is the temperature and x and y are cartesian coordinates. If
the temperature is constant on the inner and outer surface (see Fig. 1.1).
the physical symmetry suggests that the isotherms are concentric circu-lar cylinders and the temperature distribution could be described by a single variable whose constants corresponded to the radii of the
circular isotherms. Hence, we may postulate from a physical standpoint that a reduction in the number of independent variables is plausible
and a transformation such as
T(x, y) - T(R)
2
2-R Cx
+ y
)2variables for Laplace's equation.
Boltzmann[1]* considered the problem of nonlinear diffusion which is expressed by the equation
3
=
[(T)
and found the similarity variables
T(x, y) -T(n).
Later Blasius [2] postulated the form of similarity variables by
reasoningabout the physical nature of the problem. Although this method worked adequately for some problems, many other problems were too complicated to attack without a more systematic approach. One answer to this shortcoming was a method which will be referred to as
the separation of variables. As an illustration, suppose we have a
partial differential equation with one dependent variable T and two
independent variables x and y. Separation of variables entails search-ing for a transformation of the form
n = y g (x)
(1.4)
T = h(x, y) F(n)
such that the resulting application yields an ordinary differential
equation in F and n. In practice, severe restrictions of the functional
form of g(x) and h(x, y) are often necessitated and many writers refer (1.2)
(1.3)
similarity variables.
Though restrictive, the method of separation of variables proved to be suessful for many problems in boundary layer theory. Herzig
and Hansen [3] considered three dimensional flows over a flat plate and assumed that the leading edge was under main flow streamlines which are translates and representable by polynomial expressions. In a later paper 1-lerzig and Hansen [1+] used separation of variables to attack
three -dimensional boundary layer flows in polar coordinates and found that similarity could be achieved by restricting the functional form
of the free stream velocity. Cohen and Reshotko [5], when considering compressible boundary layer flow with heat transfer and pressure grad-ients, found that the necessary condition for the existence of similar-ity placed restrictions on the combination of Prandtl number and
external Mach number together with requirements on the free stream
velocity. A more general attack on the boundary layer problem in curvilinear coordinates given by Hansen [6] revealed that again simil-arity using separation of variables placed restrictions on the free
stream velocity for various curvilinear systems. These and other examples show that the method of separation of variables is effective; but because of the nature of applying the method, definite restrictions on the generality of the similarity variables occur.
Chapter II of this report describes a group theoretic technique
for generating similarity variables. The groups which will be referred to here are continuous transformations with one or more parameters and have the following form:
f1 (x, y, a)
(1.5)
= f2 (x y, a)
In order that the above transformation be a group, three properties
must be satisfied. First, if composition' is the operator of the group and distinct but arbitrary values of the parameter a determine the
nbers of the group, then the composite is also a member of the group. This requirement is called the closure property. Second, the inverse
of each transformation must exist and be a member of the group. The
existence of an inverse (see [7]) is guaranteed if
f1(x, y, a) and f 2(x, y, a) are continuously differentiable at every point.
the Jacobian J )/3(x, y) is nonzero at every point.
Third, there must exist an identity element a0 such that f1(x,y,a°) = x
and f2(x,y,a°) y.
The general theory of group techniques is placed on firm mathema-tical ground by Morgan [83 and Michal [9]. Morgan [lo] gives a
dis-cussion of the application of his theory to the problem considered by Hansen [6] and finds that a simple groupt essentially produces the same results as the method of separation of variables. For problems in
engineering, detailed descriptions of the application of group techniques
are given by Birkhof f [il], Hansen [12], and Ames [13]. Application of
'The composition of two transformations is the successive perfor-mance of them.
tA group of continuous transformations with one parameter will be
the group method for non-Newtonian boundary layer flows may be found in [iLe] and [15].
Perhaps the first effort to find and use a more general group was
given by Krzywoblocki and Roth [16], [17] and [18]. For Laplace's
equation in two dimensions they found the rotation group was applicable:
= x cosa - y sina
= x sina + y cosa (1.6)
= u.
Because Laplace's equation has the property that V2 ü = w(x,y,a) V2 u,
it will be called conf ormally invariant. As will be seen later, con-formal invariance is an essential property of the group technique.
Krzywoblocki and Roth considered the Laplace's equation in higher dim-ensions and some problems in viscous fluid mechanics, but only found
simple transformation groups for the latter. Their failure to find new similarity variables for fluid problems was probably due to the restrictive form
of
their transformation group.The works of Morgan and Michal lacked a formal mathematical proce-dure to find the invariants of a group. A function is an invariant of a group if it has the same form in the transformed variables as in
the untransformed variables. For example, in the rotation group
x2 + y2 is an invariant of the group because 2 + 2 = x2 + y2.
Gaggioli and Moran ([19] through [23]) found a theorem in Cohen [2LeJ
for the necessary and sufficient conditions for the existence of
in-variants of a group. Combining this theorem with the theory of Morgan and Michal, Gaggioli and Moran considered the problem of finding
similarity variables with a general class of finite groups, but in
their illustrations only simple groups were used. In addition these
authors considered two parameter groups of the type
X. X.
1
x. = C (a, b) x.+ k 1 (a, b)
1 1
and applied the result to a three-dimensional boundary layer problem. Finally, Gaggioli and Moran developed a method for handling auxiliary conditions and demonstrated it with an illustration.
Chapter III of this report describes a relatively new method which will be referred to as the infinitesimal group method. An infinitesimal group of transformations assznes the form
i. = x. + c X.
(x
...
x )1 1 1
l
'
flwhere c is an infinitesimal parameter. Infinitesimal groups were
originated by Lie [25] in order to unify the methods of solution for
ordinary differential equations. In addition, Lie developed a theory
of one parameter groups including both the finite and infinitesimal
cases. Some authors refer to one parameter groups of transformations
as Lie groups. A discussion of the application of Lie groups to ordinary differential equations and partial differential equations of the first order is given by Cohen [2L1]. Ovsjannikov [26] was perhaps the first one to use infinitesimal groups to generate similarity variables of partial differential equations; he considered the non-linear diffusion problem. Later Cole and Bluman [27] published a
des-cription of the infinitesima], group method and an application to the linear heat conduction equation. Bluman [28] gives a more complete discussion of the infinitesimal method and several examples.
Chapters IV and V of this report are devoted to expansion of both the theoretic group technique and the infinitesimal group method and to the application of these developments to a number of cases.
The main contribution of this work is threefold. First, extanding
upon the ideas of Krzywoblocki and oth, a practical method for obtaining generalized groups is presented together with the development of
appropriate similarity variables. In addition a systematic method for carrying an unknown function through the extended analysis is presented together with a procedure for determining restrictions on the nature of that function so that similarity is preserved. Boundary conditions are examined in the light of a set of necessary and suff
i-cient conditions to determine whether they are compatible with the
generalized group. Second, following the ideas suggested at the end of Cole and Bluman [27], this report uses the infinitesimal group to
generate nonclassical similarity variables and extends the method to handle simultaneous partial differential equations. The treatment of the boundary conditions and unknown functions is illustrated. A physical interpretation of some new similarity variables is presented and some numerical results for problems using such variables are given. Third, a possible pitfall in the method of infinitesimal groups is noted with an explanation of its occurrence and supporting examples are
CHAPTER II
SII4ILARITY BY FINITE GROUP METHODS
Continuous Groups
In Chapter I continuous transformation groups have been defined. Because the properties of such groups are extremely important for the discussion which follows, they will be reviewed here in greater
detail.
An r parameter continuous group of transformations has the form
= f(x1,..., x; a1,..., ar), i
It will be assumed that the f. are continuously differentiable func-tions of the variables xi, i l,...,n, and of the parameters
j = l,...,r. Further, the parameters a. must be essential in the
sense that a small variation in each of them produces a definite change in the functions f. For example, suppose that each parameter
a is incremented by an arbitrary small quantity .; then the
para-meters are not essential if
(x1,..., x; ai,...ar) f (x1,..., x; a1 + c1,..., ar + tr) An example of a group with a non-essential parameter is
= x + (a1 + a2)
y + a1.
unchanged. A necessary and sufficient condition (see Eisenhart [29]) for r parameters to be essential is that the functions f. do not satisfy an equation of the form
af. i
a ) - = O.
kL.
Xk (a1,..., rIn addition we require that the inverse set of transformations exist and that we can find parameters a., which are functions of the
parameters a., such that
xi = f1 (x1,..., a1,..., a)
and that
= fX1,...,
; ä1,, ¡) = f1(f1,..., f; ä,.., ä) = x.
The inverse property is guaranteed if the Jacobian J f )
n
J=
is not zero at any point.
The closure property as stated in the introduction requires that
f1
1,
,xn;1,s, a
ii = f(x1,..., "n' b1,..., br) That is 3(x1,...,x)
= f.(x1,...X;
c1,...c)
where the parameters c. are functions of the parameters a. and
b1.
ifci = i1(a1,..., ari b1,..., b).
Next, we require that there exists a set of parameters a called the identity element such that
i. f.(x1,..., x; a10,..., ar°) = x.
Finally, the operation of composition must be associative.
Continuous r parameter groups of transformations which satisfy the inverse property, the closure property, and have an identity element are called r parameter Lie groups of transformations.
As an example consider the group al
xa x
a2 y a y.
The identity element is a = 1. The closure property implies that if
al ci1
b
=(ab) x
al
then there exists a Cc) = (a b) such that
a =
X =
(c) Xbelongs to the group. The inverse transformation is
i
xa
5-a2
y=a
yInvariant Trans formations
is denoted by
A differential form of the kth order and m independent variables
11
y
i n
(x1,..., xm; yl,...,
()k '''
A continuous group of transformations G1 represented by
= f.(x1,..., xm; a)
- G1 (2.2)
y1 v1(y1,..., y; a)
may be enlarged by appending the partial derivatives of y. with
respect to X. up to and including the kth derivative. The resulting
set of transformations forms a group Gk. Such a group will be called the kth enlargement of the group G1. Denoting the arguments
of by z1,..., z, we say that the differential form is conformally invariant under the enlargement Gk if
= E(z1,..., z; a) (z1,..., z). (2.3)
If E is a function of a only, is said to be constant conforinaily
invariant; and if E is a constant, then is said to be absolutely
invariant.
Now consider a path curve h(x1, x2) = O and a one parameter
group
= f1(x1, x2, a)
g2 = f2(x1, x2, a)
with the requirement that the identity element is a° O. The curve h(x1, x2) = O is an invariant curve if h(1, whenever
h(x1, x2) O. Expanding h(1, x2) in a Taylor series we obtain
h(1, X2) h(x1, x2) + Qha + Q2ha2/2 +
where
II
I ( )
a0o
+2j
;;ao
and Q2 QQ. A necessary condition (see Cohen [2+]) for h(x1, x2) to be an invariant curve is that Qh O whenever h(x1, x2) O; that is
Qh = w(x1, x2) h(x1, x2). In that case we find that Q2h = QQh = Qwh + wQh = (Qw + w2)h and similarly if Q% = 8(x1, x2)h then QQh = (Qe + 8w)h. Hence the vanishing of Qh, whenever h(x1, x2) does, is both a necessary and sufficient condition for h(x1, x2) O to be an invariant curve. This argument extends to the differential
form and leads to the following theorem (see Morgan [8]):
* (1)
Theorem 2.1. If is at least in the class C with respect to each of its arguments, then a necessary and sufficient condition for
to be conformally invariant under a one-parameter group of transfor-mations is that V = w1(z1,..., z) 41(z1,..., z ) p where V
z)
( ) i-...
+ (z1,..., z )ì_(
) p z p 3f. and =-i.
3az;
'). (2.k) (2.5)To give an example: the group given by equations (1.6) may be applied to Laplace's equation with the result that V2ü = V2u and we conclude that Laplace's equation is absolutely invariant with respect
to that group. Suppose we consider the Poisson equation of the form
*
The class C1 is the class of continuous functions with one
2 2
3u
;:_
;--H(x, y, u)
y
and inquire as to the necessary conditions for this equation to be
absolutely invariant under the group (1.6). In that case we have
2
au
au
+ - H(, , )
=
4
+-
H(x, y, u)a ay ay ax
which requires that H(, , ü) = H(x, y, u). Now differentiating with respect to a and setting a e0 we see that with group (1.6)
ah
- .y + .-37x O.
(2.6)
(2.7)
(2.8) In this case the w1 in (2.5) is zero and the above equation is the necessary and sufficient condition for H(x, y, u) to be absolutely
conformally invariant. The solution of (2.8) is
1-i
H(
2
= x
i-y,u)
where H1 is an arbitrary function. Of course, H1 can be just a
con-stant.
Absolute Invariants
Let us inquire under what condition a function D(x, y) is an absolute invariant of the one parameter group
f1(x, y, a)
= y, a).
Without loss of generality, we may assume that a0= O is the identity
element. In order for D(x, y) to be an absolute invariant we require
D(,
) = D(x, y).Expanding D(, ) in a Taylor Series we have
D(,
) = D(x, y) + QDa + Q2D a2/2j
(29)
Now the oper:tor Q has the property that Q2D = QQD, Q3D = QQ2D,...,
Qfl
= QQD, and if QD = O, then au. the higher order terms in the
series given by (2.9) are zero. As a result we have the theorem given by Cohen [21e]:
Theorem 2.2. The necessary and sufficient condition for
D(,
) D(x, y) is that QD = O.The above result extends to two parameter groups. For the group = x C.(a, a2) + K(a1, a2), we have the following theorem (see
Gaggioli and Moran [23]):
Theorem 2.3. The necessary and sufficient conditions for to be an invariant are =j aa1 ax1 a1 = a1 o a2 = a2
!.
.L._
(n)
= O 1:l 3a2 i i a1 = a1° o a2 = a2 (2.10) (2 11)A group with m transformations and r parameters has only the limited number of functionally independent invariants expressed by the
following theorem (see Gaggioli and Moran [23]):
where Q
-
a aa a -oxai
+ -i-aa a i -ayTheorem 2»+. Let a group of transformations be expressed by i f1(x1..., Xm a1,..., ar) and i 1,...,m. Also define
3f. = o (2 12) j o a -r r
The number of
functionally
independent invariants is equal to m -where o is the rank of the matrix whose elements are given by (2.12).As an example consider the group given by (1.6). According to the theory presented above, the invariant n must satisfy the equation
-y-an an
(2.13) A general solution to (2.13) is n h1(x2 s y2) where h1 is an
arbi-trary function.
Similarity
When a differential equation is conformally
invariant
under agroup, then the solutions of the transformed differential equation will have precisely the same form as that of the original equation. For example, if u = u(x, y) is a solution to Laplace's equation, then
= ) is a solution of that equation under group (1.6) and
is the same
function
of and as u(x, y) is of x and y. A solution which has this property is called an invariant solution with the understanding that a particular group of transformations isassociated with it.
which may be expressed by
tn
(2.1L)
g1(Y1,. .., y; x1,. .., g(y1,. ..,
y.
X1. .. X ) (2.15) where the x's may be thought of as independent variables and the y's may be thought of as dependent variables. According to Morgan [8] the invariants have the property that the Jacobian Jg1,..., g) J
= y1,..., y)
is not zero and the rank of the Jacobian matrix
is equal to m - 1.
For Laplace's equation under the group (1.6) the invariants
are-2 2
n = h1(x + y ) g = u.
Suppose that an invariant solution of Laplace's equation is known arid is denoted by u = I(x, y). By definition of invariant solutions
I(,
). Substitution of the invariant solution into the invariantg = u implies that = I(, ) = u I(x, y) and that I is an absolute
invariant. The necessary and sufficient condition for I to be an
invariant is 3x 3a
aa
31aa°
3 -(2.16) (2.17) (2.18)The solution to (2.19) is
I h (x2 + y ) =2
where h2 is an arbitrary function. Hence, the similarity variables
for the problem are
u h2(n)
2 2
n = h1(x +y ).
Letting h1(x2 + y2) = X2 + y2 and transforming Laplace's equation with the similarity variables, we find
ah 2 an (2.19) (2.20) (2.21) (2.22) (2.23)
Equation (2.23) can be integrated by elementary methods and has the
solution
2 2
h1=b1ln(x +y)+b2
(2 2)where b1 and b2 are constants.
Before stating the general theory of similarity variables, let's summarize the procedure illustrated for Laplace's equation in more
general form. First, referring to group (2.2) take an invariant
g1 g.(y1,..., y;
X1,..., x)
and substitute in it an invariantsolution y. I.(x ,...,
x )
and observe that g. becomes an invariant2.
ii
m iof the subgroup i = 1,..., m. Second, apply the necessary and sufficient condition expressed by theorem 2.2 and obtain the result
where the functions g. may be determined by substitution into the differential equation under consideration,
The above procedure is a general one and is placed on firm math-ematical grounds by the following two theorems due to Morgan [8].
Theorem 2.5. Suppose we consider y6 and to be implicitly defined as functions of the x. and iL by the equations
z6(x1,..., x) g6(y1,..., Y; x1,...,
x )
in
6l'"'
=.''
where g6 are absolute invariants of a group. A necessary and
suffi-cient condition for y6 to be exactly the saine
functions
of x1,.,.,x
as the are of the iZ is that
and
where n1,..., nm_i are the invariants of the subgroup:
X),
i = 1,..., m.Theorem 2.6. If each of the differential forms . (see (2.1)) in i
a system of partial differential equations is conformally invariant
under the kth enlargement of a group, then the
invariant
solution of that system can be expressed in terms of the solutions of a system ofassume the form
n. = n.(x1,...,
i
i
x)
g
g(n1,...,
(2.25) (2.26)
the form akF n A6(n1,..., cm-1 F1,..., F;
''
k an1 Boundary ConditionsAs pointed out by Gaggioli and Moran, the auxiliary conditions must be compatible with invariant solutions. For an illustration of the concept consider Laplace's equation and denote the invariant solutions by u = I(x, y) and ü = I(, ). Suppose that the auxiliary
conditions are
A(u, x, y) = O when r(x, y) = 0. (2.27)
Transforming (2.27) under the group (1.6), we have
A(ü, E, ) = 0 when r(, ) o. (2.28)
Because in the group (1.6) = u, the above expression can be written
as
A(I(x, y), x cosa - y sina, x sina + y cosa) 0 (2.29)
when
r(x cosa - y sina, x sina + y cosa) 0. (2.30)
The auxiliary conditions are said to be compatible if (2.27) agrees with (2.29) and (2.30) for all values of the parameter a.
As is easily seen by comparing (2.27)with (2.28), the boundary conditions are compatible under a group of transformations if they
are invariant curves. Consequently, we may require that A(u, x, y) O
and F(x, y) = 0 are invariant curves and according to theorem 2.1 we
and I ai'
.2J
ai' aaj
y) r(x, y) I aA a aA aüi-
o + oV +
= -a a=a a=a (x, y, (2.31) u)A (2.32)CHAPTER III
SIMILARITY BY INFINITESIMAL GROUPS
Infinitesimal Transformations
A set of infinitesimal transformations can be generated from a
set of continuous transformations. For example, consider the con-t inuous con-transformacon-tion
(x, y, a)
(3.1) = (x, y, a)
and let a0 be the identity element. Replacing the parameter a with
a° plus an infinitesimal óa and expanding in a Taylor series gives g (x, y, a0
t
óa)x + -
óa + O(6a2)a°
(3.2) (x, y, a° + 6a) = y + - óa t O(6a2).
a0
The system (3.2) can be thought of as an infinitesimal transformation.
By again using Taylor series, the effect of an infinitesimal
transfor-mation on a function (x, y) is seen to be
- - aw w(x, y) = w(x,
y) t
a° + a0 ]ôa t... Let X = and Y -a°-.-
; then the second term in the above seriescan be written as Qwóa where Q + ,.
Once the infinitesimal ax
transformation is known the term Qw can be written and for this reason
an infinitesimal transformation with the understanding that it represents one only in the above sense.
Using the notation introduced above and letting óa = e the system
(3.2) becomes
= x + Xc + O(e2)
(3.3) = y + ye + O(e2).
Let us investigate whether the transformation (3.3) is a group to
order 2. The identity element is e O and the Jacobian of the
trans-formation is 3x ay =
:i-i
3x 3y ax ay 2 = i + c(X + Y ) + O(e ) O. X yHence the inverse transformation always exists. Letting i = x + Xe1 and = y + Ye2 and substituting into x + Xe2, we obtain
X2(X1 y1) x + Xe1 X(x + e1X, y c1Y)e2 x + Xe1 + Xe2 + O(c1 e2).
Because e and e are arbitrarily small infinitesimals, their sum is an infinitesimal and successive transformations lead to a member of
the group. All properties are satisfied and the system (3.3) is
a group.
As has
been
noted above an infinitesimal transformation can becontinuous group of transformations. For example, consider the group - a a a x = e X + e (1 - e )y
-y=e
y.The corresponding infinitesimal transformation is
This question arises: Once the system (3.5) is
known,
can the system(3.4) be found? There are two procedures illustrated by Cohen [24]
and one will be demonstrated here. To find the system (3L4) from
(3.5) solve the following set of equations:
(3.6) 2 i
As can be easily checked by the method presented in Chapter II, the
constant n of
integration
for the first pair of equations is aninvar-iant of the systems (3.4) and (3.5) and is found to be
or
x+v x+
n = ---t-
-
--y y
Integrating the second pair of equations (3.6), we have
in 2a
(3.7)
(3.8)
where c2 is an integration constant. Because at a = O = y, then = x + (x - y)c
(3.5)
= y. Equations (3.7) and (3.8) are simultaneous equations representing the equivalent finite continuous transformation which
is generated by the infinitesimal group and their solution is easily found to be the system (3M).
Invariants
The necessary and sufficient condition for a function
n(x,
y) tobe invariant under an infinitesimal transformation is the same as for continuous transformations and is given by
Qn = 0. (3.9)
Also, as in the case of continuous transformations, the necessary and sufficient condition for a curve H(x, y) = O to be an invariant curve under an infinitesimal transformation is
QH w(x, y)H (3.10)
where w(x, y) is an unspecified function. Considering x and y as independent variables and u as a dependent variable, an infinitesimal transformation can be written as
= x + Xe + O(e2)
y + Ye + O(e2) (3.11)
ü u + Uc + O(e2).
If we append to (3.11) the partial derivatives of ü with respect to
and up to and including the kth derivative, then we obtain another
th
k
3u 3u (x, y,
;,
ax1ayi
=
where i + j k. Letting z1 through z denote the arguments of (3.12)
th
and z1 through Zn be the corresponding coefficients of £ in the k enlargement of the group (3.11), then (3.10) can be written as
a a
- +
. . . + z - = w (z , . . . , z )4 .Z1 1 Z2
ai n 3z 1 .i. n
1 2 n
Equation (3.13) is then a necessary and sufficient condition for the
to be an invariant curve.
Canonical Variables
As was mentioned earlier an infinitesimal transformation may be represented as Qw and referring to the system (3.3)
a a
Q = X ( ) + Y ( )
With (3.l) we see that
Qx = X
Qy = Y
Consider a change of variable which is denoted by
X1 = x1(x, y) y1 = y1(x, y). Then (3.12) (3.13) (3.15) (3.16) sufficient enlargement of the group (3.11) the conditions expressed by (3.9) and (3.10) can be written for a differential equation of the
ax1 ax1
Qx X +Y-X
1 3x ayi
ay1 ay1
Q y1 X - + Y._.... - Y1
and the new infinitesimal transformation under the change of variables
becomes
= + 1 (3.18)
where is a function of X1 and y1. This procedure allows us to make
a change of variables and find the representation of the new
infini-tesimal transformation by solving the system of equations (3.17). As an illustration consider the rotation group (1.6) and the change of variables defined by
/2
2 p=\jX
+y
(3.19) -1 8 = tan (y/x). a ( ) a , we find thatSince for the rotation group Q - -
y
ay
Qp = - y + x
ax ay
2
2-½
22t½
=-yx(x +y)
+yx(x +y)
=0
and ae ae
Qe = - y
;.
+ x 2 2 -y
X 2 2 2 2X-y
x-y
hence the new infinitesimal transformation is represented by
Qw(p,
e)
=
and the new infinitesimal transformation is
U= e + £
(3.20)
It is interesting to note that the invariant of the group (3.20) is p. The variables which reduce an infinitesimal transformation to the
3w1
form Qw1 =
r-
or-r-
are called canonical variables and theY.L xl
resulting infinitesimal group is said to be in canonical form.
Following this notation the transformation (3.19) is a set of
canon-leal variables for the rotation group.
Reduction of the Number of Independent Variables
in a Partial Differential Ecuation.
Z.Z.
+Z.c i1,...,n
i i iThe infinitesimal representation of (3.22) is Q = Z1 .- + ... + Z
n
Letting z1 through z denote the arguments of the differential equation (3.12) and rewriting, we have
4(z1, Z2I
z)
= O. (3.21)The infinitesimal transformations for the z' s may be written symbol-ically as
(3.22)
(3.23)
As shown in the previous section of this chapter we may find a set of canonical variables which will reduce the transformation (3.22) to
canonical form. Let V. i = 1,..., n denote the appropriate canonical variables and let the
canonical
form of the group (3.22) be denoted by= V1 +
(3, 21+) = y. i = 2,..., n.
i i
If the differential equation (3.21) is conformaily invariant (see Chapter II) to the order when transformed by the group (3.22), then according to theorem (2.1) of chapter II we have
z
±+
...+ z
i z1 n az w(z1,..., z) 4(z1,...,
za).
n
Introducing the canonical variables y1 we find that (3.25) becomes
a
-w
-
2(v1,..., y) (y1,..., v)where is the differential equation when transformed by the
canon-ical variables. Also since v, i 2,..., n are absolute invariants
of the group (3.21+), we may apply the necessary and sufficient
condi-ditions given by theorem (2.2) to find that
3v. i
= O i = 2,..., n. (3.27)
Vi
If
l is absolutely invariant wider the group (3.21+), then (3.26) become s 1 = o. 3v1 (3.25) (3.26) (3.28)
Equations (3.27) and (3.28) guarantee that the variable y1 does not appear either implicitly or explicitly in
.
Consequently, the number of variables has been reduced by one. If z1 were an
This result is precisely the one needed to obtain similarity solu-tions.
Example
The infinitesimal transformations corresponding to the rotation
group (1.6) are
= X - yE
= y + x (3.29)
Laplace's equation in two dimensions is absolutely invariant under the group (3.29). We seek canonical variables s1, s2, and q1
such that
) = s1(s, y) + c
S2( ) = s2(x, y) (3.30)
q1,
) = q1(x, y).An appropriate choice for
l and s2 is given by letting s1 e and s2 p in (3.19). Inspection of (3.29) shows that u is an appropriate
choice for q1. With these choices (3.30) becomes
= etc
p=p
(3.31)In group (3.31) u is an invariant and accordingly = o. Treating 6 and p as new independent variables, Laplace's equation
transforms to
i
The similarity variables generated by the group
(3.29)
are/2
2 px +y
u = u(p)
As is seen from this example, the process of
introducing
polar coordinates and requiring that = U is the same as searching for similarity variables under the group of rotations.Similarity Solutions in Practice
In practice we may construct similarity solutions by a procedure
described by Bluman and Cole [27J. The method will be described for
the case of two independent variables and two dependent variables; however, the ideas are applicable to other cases.
The infinitesimal transformation is assumed to have the form
= x + X(x, y)c + 0(c2)
= y + Y(x, y)c
0(c2)
ü = u + U(u,
V, X,
y)c +
(2)
= y -
V(u, v
x
y)
*0(c2)
and the independent variables are x and y while the dependent variables
are u and y. The differential equation may be represented by
k k 3u av au
au
av
il'
nl
axy
axay
(3.33) (3 34) (3.35)where i + j k and n s i = k. We may append to the group (3.3Le) the
transformations of the derivatives of u and y which appear in $. As the determination of the transform of a derivative requires some careful
au
manipulations, the transform for
wil1 be presented here. By thechain rule
ìL
ax ax a
aya
To apply (3.36) we must determine and remenbering that terms
of order
e2 and higher are dropped.
Differentiating the first of
equations
(331) with respect to Z ,we bave
Differentiating the second of (3.3L1) with respect to and using
(3.38), we have O = Z. + (Y) +
O(e2)
ax 3 or ax a i = . + - (X)e +O(e2).
dXax
From (3.36) we see that
a
()
aax
aax
= -
(Y 1- Y*
+O(e2).
a2
x
The above equation tells us that is
at most of order e
and eis at most of order (e2). Hence (3.37) becomes ax O(e2) i =
=-+ eX
ax Xor
axax
(e2).
-
i - cX + 0
a xThe above eçpression tells us that .j. i + O(e) and that we may
further simplify to obtain
(3.36)
(3.37)
(3.38)
Similarly
- c + 0(c2). (3.42)
Substituting (3.141) and (3.42) into (3.36) gives
and = 1 - c(X ) + 0(c2). ax x au (1 - £ X ) + ( - c Y ) + 0(c2). -ax ax x ay x From (3.34)
au_au
-+ Uc-+ O(c)
aüau
Cc2).=---+cU +0
ay ay ySubstituting (3.44) and (3.45) into (3.43) gives
3ü
= - + cL -
au
Xxx
u + U - Y u J + 0(c2). x X y (3,143) (3.45) (3.46)Having calculated ail derivatives in the barred variables and
letting denote the differential equation (3.35) in the barred
var-i ables, we requvar-ire that X, U, Y, and V be chosen such that
= (i + £ (x, y, u, v)). (3,47)
Equation (3.47) is the necessary
condition
that be conformallyin-variant under the grout (3.314). If u = 01(x, y) and y = 82(x, y) is a solution to , then ) and 82(, ) is a solution to by
virtue of (3.47). Using (3.34) we see that
Expanding (3.48) in a Taylor series we have
e1(x, y) + (X u + Y u) 61(x, y) + tic + 0(c2)
and we conclude that
Xu +Yu
U. (3.'ig)X y
Similarly we find that
X
sYv =V
(3.50)X y
Once U, V, X, and Y have been determined to satisfy (3.47), then (3.49) and (3.50) define the similarity variables for the differential
equation 4. Actually (3.49) and (3.50) can be used when satisfying
the invariance condition expressed by (3.47). According to the term-inology of Bluman and Cole [27] the similarity variables obtained are cafled classical if (3.49) and (3.50) are not used in (3.47) and
non-classical if they are used. All similarity variables generated in this thesis by infinitesimal methods will be the non-classical type.
CHAPTER IV
APPLICATION OF THE FINITE GROUP METHOD
Burgerst Equation
The Burgers' equation u + uu = u has been used as a
rnathe-y
X XXmatical model of turbulence [30] and in the approximate theory for
weak nonstationary shock waves in a rea]. fluid [31]. We seek groups
of transformations of the form
= f1(x, y, a)
y, a)
(L1.l)
= fu(a) u + f5(x, y, u, a)
such that the Burgers' equation is conformally invariant. To accom-plish this task, we append to (4.1) the relevant partial derivatives
which are 3f 3f 5 5 3u 3x +
)
-= (f + 3x43x
3x (14.2) 3u 5 3u.. 3v- + - +
)
-Lê3y 3y u 3y 3x 3f 3f 3u 5 5 3u 3x - = (f ay14 -j +
+-
) .-3f 3f 3u 553u
+ (f14-s
+ .-
.._.) (14.3)2_ 2 3f 3f
3u
3x
5)
53u
3u 3uf -+- +
43
ay au 35E y 2 32f 2 3f3x2
3u
5+--+ 2
3u
5 5 3u + (-w)(f
-+ 2 2 au ia 3x 4 2 2 32f 3f 2 32f3u
+ .!.
(2 f4axay
+ 2 3x3y + 2 3 u + 2 5 3u 3x 3x 3uaxay
au3y 3x+2
5 3u 3u3x 3y 2 2 32f5__ 3u2
3f 2 5 3u()
+ 2au
3u3y 14) a ayay
32f5In (i.L) it must be assumed that O; otherwise 2
the square of the derivatives of u would appear on the right-hand side of (4.4) and would force this assumption in the present example. In problems where the squares of derivatives appear in the differen-tial equation, such terms should be retained. Using (4.l),(4.2), (4.3) and (4.4) we form the Burgers' equation in the barred variables and
equate it to zero. Since we cannot place any restrictions on u or
its derivatives, the coefficients of such terms must vanish if they do not appear in the differential equation. Inspection of (4.4)
2
reveals that the coefficient of is equal to
3x3 y
-2f
-2--.-
ìL
43x 3x
3u 35E 35E3f
and for this to vanish, either
O or -. =f
The latter37
alternative is rejected since
itwould cause ü to vanish identically.
The coefficient of u, which arises from the term ü
ti_,
is3f 5 3x
I- - -
43x
3x3f
5
which forces -
dx = O. With the above observation a search for termswhich do not have u or its derivatives as coefficients reveals that
3f
ay 3y
which forces f5 to be a constant. The coefficient of is
a2
+ f4 f5
* -
4-
o.
3x
The final, requirement is that the coefficients of the terms which
appear in the differential equation be equated; so that
2
3X_f
,=
3x 3 143y From (4.6) we have 3x 3f4
(.7)
2 ay 4Substituting the results of (4.7) into (4.5) gives
3x
5 4 (4.8)
The inverse transformation of (4.9) is
i
9' -
y
f4
The transformed Burgers1 equation is
+uu -u
)y x xx 4 y x xx
which satisfies the conformal invariance requirement. For convenience
let f4 and i-i(a) = f5(a) f2(a). Using (4.10) and (4.1) the transformation group becomes
= f x + Hy - 2
y=fy
- i H u = - U + -s-f'in Chapter II. Individual transformations can be denoted by
f(a.)x I- H(a.)y i i
- _2
y1 = f (a1)y H(a.)-w
u+
2 - f(a.,) f2(a ) i i (4.10) (4.11)The transformation (4.11) must satisfy the closure property defined
(4.12)
A constant multiplied by a function of the parameter a could
be added to obtain a more general group.
x f4 - 5
(4.9)
where a. is any real number.
The closure
propertyrequires
that ifand x2 are transformations then there exists an x3 such that
x2(x1, a) = E3(x, y, a3).
S irnilarly,
'2"l' S1 a2) = 3(x, y, a3)
U2(U1 a2) = ü3(u, a3).
Substituting (4.12) into (4.13), (4.14) and (4.15)
X3 f(a2) [f(a1) x + Fi(a1) y] * Fi(a2) f2(a1)y
(4.16)
= f(a3) x + H(a3)y
- f(a2) (f2(a1) x) f2(a3)y (4.17)
H(a)
H(a2) = f(a) f(a1) + f2 +f2()
u Fi(a3) + = f(a3) f2(a3)The above three equations require that
H(a3) f(a2) Fi(a1) + Fi(a2) f2(a1)
f(a3) = f(a1) f(a2)
(4.13)
(4.18)
Equation (4.19b) implies that interchanging a1 and a2 does not change
a3 because the real numbers are commutative. Using this result in
(4.19a) gives
Fi(a3) = f(a2) Fi(a1) + Fi(a2) f2(a1) f(a1) Fi(a2) + Fi(a1) f2(a2)
which may be rearranged as
(L 14) (4.15)
The subgroup iZ and possesses an
invariant.
separation of variables) which forms the subsystem of equations
H(a1) f(a2) [1 - f2(a2)] H(a2)[f(a1) (1 - f2(a)]. (4.20)
Equation (4.20) is satisfied if H(a) f(a) (1 - f(a)) and (4.11)
becomes
= f(a) x + f(a) (1 - f(a))y
= f2(a)y (4.21)
= f(a) + (1 - f(a)) / f(a).
Satisfying the closure property has produced a result which immediately satisfies the existence of an identity element a if f(a ) 1. Since
o o
equation (4.21) is seen to possess an inverse, it satisfies all the
requirements of a group. The invariants solutions of of (4.12) ax. -3v +
-
-'-are described ax . -i + au byfunctionally
independent ax.io.
(4.22) 3x 3aaa
o a =a ay 3a o a=aoSubstituting (4.21) into (4.22) gives
ax. ax. ax.
f'(a ) (x-y)
o -.- + 2 fT(a )yax o 3y - f'(ao) (u+l)
-.
0. (4.23)Assuming that f'(a) does not vanish and that A1 = A1(x,
y)*
ax ax
we have
(x - y) ._! + 2y = 0. (4.24)
or
ax2 ax2 ax
(x - y) - + 2
ax
y_-(u+i)i-o.
auAgain using the method of Lagrange the solution of (4.27) is found to be
(X+Y
A2 , (u +
l)V½)
Equations
(4.26) and (4.28) implicitly define the soughtsimilar-ity variables. A special case is
= n = (4.29)
= (u + l)Y½ F(n). (4.30)
Substituting (4.29) and ('+.30) into the ßurgers' equation gives
F"-FF'+½nF'-s-½F=O
(4.31)[F' - ½ F2 + ½ nF] O.
One integration results in
F' - ½ F2 -s- ½ nF const. (4. 32) (4.27) (4.28) x-y 2y O (4.25) The solution to (4.25) is x+ (4.26)
y-A second invariant of (4.21) is found by letting y-A2 = y-A2(x, y, u) and (4.23) becomes
A =
r b(b+l) ... (bi-r-1)r
Laminar Boundary Layer
The boundary layer equations for steady, incompressible two-dimensional flow (see Schlichting [35]) are
* dU 2
*3u
*au
au
U --
t V = U* 2
ax
ay
which is a form of the Riccati equation (see ref. [33]) and it may be transformed into a linear second order differential equation by the
trans format ion
F - 2 Pfl
(L33)
Applying (L.33) to (.32) results in 2+ n
+ C 4 O. (43l) fin ni
The solution to (L3 J4) is found by standard methods (see [3k]) and is
C C = C10 1Fj ( ; ½; z1) C11 (½ + ; /2 z1)(Ll.35) where 4 (ri) =
(zr)
2z1 = - ¼ n
1F1 (c2, b, z1) =A (z )"
o r i ('4. 37) c2(c2+1) .., (ctx-1) ('4.38a)au
av
r
+ r =
O.3x 3y
A set of dimensionless coordinates is defined by
y L
lu
LVC
V * * x =Lx
u U u Cwhere U is a characteristic velocity, L is a characteristic length, and y is the kinematic viscosity. Introducing equation (4.39) into
(4.38) we have U dU
au
au_
32uuf V
3x +ay
C 3uav_
axtay
O
yv=
U Clu
L C'4V
(4.38b) V (L1.39) ('+.40a) (4.4Gb)We want to determine continuous one parameter groups of tran8-formations which will keep (4.40a) and (4.4Gb) conformally invariant.
U dU
For reasons which will become clear later, the term will be assumed to be a function of x and y denoted by P(x y). A general
group
= f(a)u + f5(u, y, x, y, a)
f6(a)v + f7(u, y, x, y, a)
= f1(x, y, a)
is assumed. The relevant derivatives are
3f
3f
3f
3u 5 5 au 53v3x
--
¼f +t
4 a,
t -e-w
ax
3v 3x 3x3f
3f
3u 5 3u 5t
(f
4tt
3y 3us-
3y 3v 3y 3x3f
3f
3f
3v_
!!-+...2+
7 3u 7 3v 3x a y-
3x 3x5t
(4.43)
+ (f
3f7
3f7
3f
7 3v6 3y
T
3y57
ay
3f
3f
3f
5 5 3u 5 3v 3xs
s
+ ciy4 3x
T
(Li.44)
3u3f
5 3v(f4
3y 3u 3yV
2-
3f
3f
3f
2 3 U_ 3u 5 5 3u 5 3v3 x
7
3x 3x 3u3x +
'
7
ay
ay
3f
3t
3f
2 5 5 3u 5 3v 3(f
4 ++ .
+
.-232f
23f
a2f
3f
232f
23u
53u
5 3u 5 53v
53v
3x+ s +2 + +2
)(--)
4 3x2 3x2 3x2 3u 3x3u3x
3v 3x23v3x 3x
3y 232f
32f
2f
+(2f
3u
+2
+2
3u3u
35
2 5 3v4 3x3y
3x3y
3u3y+ 2
3x3y
3u3vay
3f
232f5
3u 3v(3x) (!L)
3v
+2
+2
) 3v 3x3y3u3x
3v3x
3y 3y 3 + 232f
23f5
32f
53u
3f
53v
232f
5 + 24
+ 2
3u3y . + 2+ 2
3v3 y ayay
ay
ay(4.42)
3v)()2
(4.45)
Using (Li.1U) through
(L45),
the system of equations(L.iO)
is written in the barred variables. The first step is to search for terms with coefficients which do not appear in the original system.2 Since the coefficient of
2f
39 39 39 39
i'
3x
must be equal to zero, we choose - = O. Actually we could, at this
ay
point, choose O. However, such an assumption uld cause the
Y2
coefficient of --to be zero. To achieve conformal
invariance
all ayother coefficients of terms in the differential equation uld be
re-quired to be identically zero, but this requirement uld cause
2 3f
2
unnecessary restriction. Since the coefficient of 4
,
(p.)
,3f5 ay Y
must be zero, we choose = O. With these assumptions the coefficient of in the momentum equation becomes
ay
3f
f
5 ay3f
which
requires that O. Finally, observing that the terms withoutany coefficient reduce to
3f
r 3f
5 + u f
-e- is taken to be zero to avoid f
5-
ecualing - u fIf f5 were equal to a function of the parameter a multiplied by
u, nothing would be added to the group (L1.tll). Consequently, f5 is
taken to be zero. With 'the above assumptions the transformed
rtmentun1
equation becomes au 2 ax-{uf2
t+f
f fu {f --}+
ax
14 3x ay 14 3x 14 n 3 6 14ay
V 2 2 )} - PCx, ) = O.f
}aU{f (!Z
1 ..2 y ayay
Before proceeding further with the nmentum equation, the continuity equation will be considered to see if there are any additional
res-trictions. The transform of the continuity equation is
au
axau
.L}
. f14 - } .$. ax ay4ax
au ayaf
av
+ {
f '. + 7ay
=ay6
The last term requires = O and the second term requires
af
-+--2.
z.=
. 14 ax au ay ax a f-IIl-=f
14 ax 6 ay Then (11.147) becomes au av C 6i
ay (4.146) (14.148)To satisfy the condition of conformal invariance for the continuity equation, it is necessary to require
(11.119)
Because 0,
cannot
be a function of if(4.49)
is to bey y
a2
satisfied. Hence, we assume that --= 0.
Returning to the momentum equation (4.46), we assume that
and
uf2
f¿4 ax 4 7
ay
in order to eliminate terms which do not appear in the original momen-tum equation. To satisfy (4.51) we set f7 p1(a) h1(x) u.
Substitu-ting into (4.51) furnishes
a p1(a) h1(x)
O. (4. 52)
Finally, to satisfy the
condition
of conformal invariance we let2 ax f f4 = 4 a a = f9(a) (4.51) (4.53) P(g, ) f9(a) P(x, y). (4.54)
Thus carrying an unknown function can be accomplished by simply forcing it to satisfy the condition of conformal invariance for the equation
as a whole and accepting whatever functional form arises.
To satisfy the
condition
of conformal invariance for the systemof equations (4.40), we have the following restrictions on (4.41): ax av f5 0 (4.55a) (4.55b) 47
3f 7 ay f f h1(x)
f= o
+ ax 6 Case I. 0.From (Lf.55e) 0, whereupon the group (L14l) becomes
= f1(a) u
may be found by forming simple ratios and are
(4.55c)
(L.55d)
(1.55e)
(L455f)
The system of (L55) does not entirely determine the transformation (LLLl). Equation (L 55f) implies that = f6 and (L.55d) implies that
-. Several cases must be examined. 3x f
As long as there exists an a° such that f(a°) = f 5(a°) = 1, then (L.56) meets the requirements of a group. To generate the
Al=
Upon integrating we find that f6 y = f - p1(a) .-. x.
-Since x = --x, equation (4.59) is easily inverted and we have
f6
y=
ip1(a)
y.'.
X.f6(a) f62(a)
To satisfy the closure property there must exist an a3 such that
a2) = y, a3). With (4.60) we require that
p1(a1) p1(a2) f4(a1)
f5(a2) f6(a1) + 2
X] +
f5 (a1) f62(a2)[2]x
p1(a3) (4.61) + X. f6(a3) f62(a3) (4.59) (4.60) A2 u (4.57) A3= y V
If in = O the invariants reduce to the classical ones used for the case of a flat plate.
Case IL h1 = 1.
Equation (4.55)e is solved to give
As in the case of the Burgers' equation we notice that
f6(a3) = f6(a2) f6(a1) and that a3 is not affected by interchanging
a1 and a2. Hence, we write
p1(a3) p1(a1)f6(a2) p1(a2)f(a1)
f62(a3) f62(a1)f62(a2) f62(a2)f62(a1)
p1(a2)f5(a1) p1(a1)f(a2)
f62(a2)f6 (a2) f62(a2)f6 (a1)
or
p1(a1) [f6(a2) - f4(a2)] = p1(a2) [f6(a1) - f(a1)]. The above relation implies that p1(a) = [f6(a) - f4(a)] and the group of continuous transformations becomes
X -
x
6_1
X y= çy +
(f6 -u = f u = f6 y + (f6 - f14)u (L,62) oLetting a denote the identity element for the group, the
invar-iants are found by solving
(4,63) Since there exists one invariant of the subgroup and , we choose
= X1(x, y) and obtain ax.
-+
ax.
iau
- ax. i.-
-av
+ -
aaax.
i
= 0.0ax
aaax ax
(fk'(a°) - 2f6t(a0))x._. + [-f6a°)y + (ft(aO)_ft(aO))xj_ o.
(L
64)
The solution to (14.64) can be found by the method of Lagrange and is
f(a°)
f1(a°)-f(a°)
(f1(a°)-2f(a°))
=
(4.65)
To check whether X1 is an invariant, form (4.65) in the barred
varia-bles and use the group (4.62) to obtain
r_
LJ
+Lxj
f'
6f'-f'
46
(f1-2f)
f-2f
i
'
11
4ri
L>'i +__TLXJwhere f4' and f6' are evaluated at a
a0.
Expanding (4.66) we find
f'
6f'
6f
f
+1
(f-2f)
f1-2f
(f'-2f6t)
f-2f
[f4 i
f6-f4
[f4
1y[x]
Li
[xl
f'-f'
46
ft-f'
46
f-2f
f,-2f
[xJ (14.66) 51(4.67)
Requiring equality between (4.67) and (4.65), forces
f,
6 i and (f'-2f6t)=1
f t_f 6 4 6 'LI-(f '-2f') f,'-2f6' f6-fI+ f 62 L .1: - - r-c ta I IiL+-.I
+I2t
l.f6J
L6J
(14.68) (4.69)Equations (4.68) and (4.69) are satisfied if f4 (f6)m where m is an
arbitrary constant. Our earlier choice is thus justified.
A second invariant may be found by choosing A2 X2(x, y, u) and solving equation (4.63). The result is
m
-7
u x ) = O
in i m-i
--
--ii--H2(A1,ux
,vu
+u
)=O
(4.70)
where H1 is an arbitrary function.
The third and last invariant of the group (4.62) is found by
cosing A3
A3(u, y, x, y) and again solving (4.63). The result is(4.71)
where h2 is an arbitrary function. The functions H1 and H2 implicitly define a set of similarity variables for the boundary layer equations. A special choice of these functions is
i in-1 m-2
A1yx
(L1.72) In m-2 A2x2(x1)
= u x (L473) In In A3 = X3(A1) = y u + u . ('1.74)In addition to the above transformation, a transformation consistent with the condition of conformal invariance must be
devel-oped for F(x, y). Differentiating (4.54) with respect to the parameter
a produces
a a a a a
- + -.
=
-ax aa ay aa 3a p(x, y).
m+2
Since f (a) = 2 = , the above equation becomes
9
46
a a a
Z-=
(in + 2) f m+3. p(x, y)-._.+-aa ax a aa
and setting a = a we have o aa 3V
-
+ -'-ax 3a a=a o a-
(in + 2) f '(a ) P(x, y) (4.75) ay 6 o a=a oAccording to threm (2.1) of Chapter II, (4.75) represents a necessary and sufficient condition for P(x, y) to be conformafly
in-variant under the group (4.62). Substituting the appropriate deriva-tives from (4.62), (4.75) yields
a
f '(a ) (in - 2) x + f '(a ) [-y+(l-rn)xJ (rn+2) f '(a ) P(x, y)
or
(in- 2)x+ [ y + (l-m) x]
(in + 2) P(x, y)ax ay
The solution of ('4.76) is found by the method of Lagrange to be
m+2
H3(X1, P x ) = O (.77)
where FI3 is an arbitrary function. Equation ('4.77) implicitly
speci-fies the functional form of P(x, y) which can be retained in the momen-tum equation without destroying the similarity generated by group
(L4.61).
A special choice of ('4.77) is
m+2
P = xH'4(A1)
('4.78)If it is desired that P be a function of x only, we just need to
choose H'4 to be constant.
substituting (L1.72), (i473), (4.7') and ('4.73) into (4.'4.Oa)
and (L4.L40b) gives
AA'
L
22
in 2 H (A ) s X " ______ (X ) + A + A ' X A 4 1 2 = in-2 i m-2 2 2 3 2 i A 'X 2 1 inin[1A x_ix
f +A3]-O
in-2 in 3 2 2 (1475) (4.79) (4.80)Equations (4.79) and (4.80) are ordinary differential equations with
as the independent variable and A2 and X3 as the dependent varia-bles.
Case III. h1(x) x.
Substituting h1(x) into (11.55e), we obtain
..=-xp1
plr:
-xv-
f6. 3 f6()2
y = f6 y - p1 4Inverting (4.82), the set of transformations gives
-X =
x
f6 = f u f6 V 1- X U p1 - i l 2f62f2
XThe procedure for obtaining p1, such that (4.83) satisfies the closure property, is the same as in Case I, whence p1 is found to be
f 2(a)
(a) = f (a) (4.84)
f6 (a)
Substituting (11.811) into (4.83) yields
(4.81)
(4.82)
(4.83) f
As in Cases I and II, - -x, and (11.81) may be integrated to
obtain
or - f4
X = i- X
f6 2 2)x
S:;
y+(_
2f64 = f4 u.The invariants of (4.85) must satisfy r6 12 f4 -f6 y + x u [-f6 - J
ax.
-1 auaa
T
+ax.
i
av
. +
-ou aaaa
o axi
+ 6' y + x u(3f6' - 2f4' )] =o.
Letting A A1(x, y) we find thatf6' 2f4t-3f6' ) (f '-2f6') f -2f
A1_y[x]
+½[x] 4 6 w-a w-a o (4.85)ax.
1=0
ax. ax.
ax.
1
(f4t
2fb') x -+ [-f6ty + ½
x2 (3ft 2f4t)]u-
1ax 4 3u
(4.86)
(4.87)
where f4' and f6' are evaluated at the identity element a. As in Case II the requirement that A1 be absolutely invariant under (4.85)
forces f4(a) [f6(a)]m and ('+.87) becomes
i 2m-3 m-2
X1yx
+½x
(4.88) axax.
-3aax
aaaa0
Following the same procedures as in Case II, two other
partic-ular invariants are found to be
_l
i-
1-.-m In A3X(A1)
y u + x u In A2 = A2(A1) = u XAgain as in Case II the functional foru of P(x, y) is determined by solUtionS of
ap 2
(m - 2) x - t
L
y + ½ x (3 - 2m)] - (m + 2) P(x, y)ax
ayand a special form of P(x, y) is mi-2
H5(A),
where H5 is an arbitrary function of A1. If P is required to be a
function of x only, then H5 may be taken to be a constant. Substitution of (4.89), (4.90), and (4.92) into equations (L#.1+Oa) and (4.40b) furnishes
1
AA'
22
m 25 1 2 m-2 i m-2 2 2' A3
rn
H(A)+A"
A i--A
i-A1 A2tA1 A3 A -1
i-A L
m-2 m-2 2 2 m 2 2+A3]=0
(4.89) (4.90) (4.91) (4.92) (4.93) (4.94)It is interesting to note that equations (4.93) and (4.94) are
identi-cal to (4.79) and (4.80). Also interesting is that A1 = O in Case II
Boundary Conditions
The boundary conditions for (14.L1Oa) and (L1.4Ob) for the case
of a zero pressure gradient are
u y O on the surface (4.95)
and
liTn U 1 (4.96)
Is the group (4.62) consistent with the above conditions? The necessary and sufficient conditions for u = O and y O to be invar-iant curves are
where 1(x, y) and 2(x, y) are arbitrary functions of their argu-ments. Choosing w1 = mf6'(a°) and recalling that f4 = f, (i.i.97)
integrates to
In
-H6(u x , À1) O (4.99)
where H6 is an arbitrary function and À1 is defined by (4.72). A particular form of (i99) is
in
in-2
u x H7 (À1). (4.100)
If the surface is described by À1 0, then choosing H7(0) = O we conclude that the condition u 0 on the surfaces is consistent with
3a 3a a 3a
aa
a=a 3u 3v -w(x, y) uw(x,
y) V
(4.97) (4.98)03x
aa
av
+
03x
a =athe group (4.62). A similar argument holds for the condition y = o
on the surface.
The condition (4.96) requires that u(x, y) approach a constant
for large y. We address our attention to finding the nature of the
family of invariant curves u(x, y) = 1. According to Cohen [24] the necessary and sufficient condition for u = i to be an invariant curve is ax 3a
cx
aaa
au-= F (u)
aa0
3y (4.101)where F1(u) is an arbitrary function. Following the method of Lagrange the solution to (4.101) is
1(u) = x H8(À1) (4.102)
where 1(u)
ÍU(U)
and H8 is an arbitrary function. If ü is to'1
be constant when u = 1, then the right hand side of (4.102) must be a
constant under the group (4.62). Unless x is taken to be constant this condition is not satisfied and we are forced to search for a
restriction. If F1(u) is taken to be zero, the preceding analysis is
invalid. In that case a particular solution to (4.101) becomes
u = (4.103)
where H9 is an arbitrary function. Equation (4.103) implies that
u and that = i when u 1; hence, the boundary condition (4.96) may be satisfied if u in the group (4.62). Consequently, we
may take f4 1 and derive a new set of invariants which will be com-patible with the boundary conditions (4.95) and (4.96).
On the other hand if the boundary condition (.96) is in m-2
limux
=1,
U dU m-2 u 2 dx C=x
in inm-2.
(L.lOO) confirms that is constant whenever u x is and
no modification of the group is necessary. Equation (L1.lOLi) implies
that the free stream velocity is
Equation (Le.106) meets the requirement set forth in (14.78).
(L4.104)
(.lO6)
m
U =U
X m-2 (.1o5)C
condition
CFLAPTER V
APPLICATION OF THE INFINITESIMAL GROUP METHOD
Burgers Equation
For the Burgers' equation u + uu = u we shall seek an
t X XX
infinitesimal transformation of the form
+ X(x, t, u) + 0(c2) = t s c
Q(2)
= u + U(x, t, u) + 0(c2) such that - 2- 2 au-au
au
)[3U
auau
u - --- w3(x, t, at ax u ax axThe relevant derivatives required are
--= u
ax x + c[Ux + (u - X )u - x u 2 + 0(c2) u X x u x2-au
= uxx t c[Uxx + (2Uxu - Xxx)u + (U - 2X )u x uu xu X dXx
uuu3+(U
x u-2X)u
x xx-3X u
u xxu]+O(e2)
x)u x u x u u]+O(c2)
ut
t X UXt
2 (5.1) (5.2) (5.3) (5.5)Xu
fu
=U
X t
u
=u +uu
U+(u-X)u.
XX t X X
Using (5.3) through (5.7), we write
3ü
au_
- - c{U s UU + u (-XU - X - X u) + (u )2 (X X)) + 0(c2) t t u x t u X U(5.8)
u -
u U + u (U + uU - UX ) + (u )2(-X u)} t 0(c2) 3x x x u X x u (5.9) 2_ 2 - - ì.) c { - U - U(u - 2X ) t u [2U _2 2 xx u x x xu x x - X - (u - X) (u - 2X ) + 3X U] xx u X U+ (u)2 [-U t 2x t 3X(u - X)]
+
(u)3 [X] }
+ 0(c2). (5.10)To satisfy (5.2), the sum of the right-hand sides of (5.8) (5.9),
and (5.10) must vanish. As or powers of u are not identically
0 2 3
zero, we require that the coefficients of (u) , u, (u) , and (u)
in the sum mentioned above vanish.
Setting the sum of the coefficients of (u)3 equal to zero gives
X 0. (5 li)
uu
Similarly, from the coefficients of (u)2, (us), and (u)° we find
t hat
(5.6)
determine the similarity variables and also use it in satisfying (5.2). Using (5.6) and the Burgers' equation we see that