Similarity Transformations for Partial Differential Equations

Author(s): Mehmet Pakdemirli and Muhammet Yurusoy

Source: SIAM Review, Vol. 40, No. 1 (Mar., 1998), pp. 96-101
Published by: Society for Industrial and Applied Mathematics
Stable URL: http://www.jstor.org/stable/2653001
Accessed: 29-09-2015 14:17 UTC

REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/2653001?seq=1&cid=pdf-reference#references_tab_contents

You may need to log in to JSTOR to access the linked references.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

Society for Industrial and Applied Mathematics is collaborating with JSTOR to digitize, preserve and extend access to SIAM
Review.

http://www.jstor.org

This content downloaded from 111.68.98.168 on Tue, 29 Sep 2015 14:17:18 UTC
All use subject to JSTOR Terms and Conditions
SIAM REV. (a) 1998 Society for Industrial and Applied Mathematics
Vol. 40, No. 1, pp. 96-101, March 1998 006

SIMILARITY TRANSFORMATIONS FOR PARTIAL DIFFERENTIAL

EQUATIONS*
MEHMET PAKDEMIRLIt AND MUHAMMET YURUSOYt
Abstract. The importance of similaritytransformationsand their applications to partial differ-
ential equations is discussed. The theoryhas been presented in a simple manner so that it would be
beneficial at the undergraduate level. Special group transformationsuseful forproducing similarity
solutions are investigated. Scaling, translation, and the spiral group of transformationsare applied
to well-knownproblems in mathematical physics, such as the boundary layer equations, the wave
equation, and the heat conduction equation. Finally, a new transformationincludingthe mentioned
transformationsas its special cases is also proposed.

Key words. similaritytransformations,partial differentialequations, mechanics of solids and

fluids

AMS subject classifications. 35Q35, 35Q72, 76D10, 35K05, 35L05

PII. S003614459631001X

1. Introduction. In manyundergraduatecourseson the applicationsof math-

ematics,a studentis taughthow to solve certainwell-known problemsin mathemat-
ical modellingsuch as the wave equation, the heat equation, etc. Some techniques
forfindingsolutionsare presented,usually specificto the problemunderconsidera-
tion. When the studentis faced with a different type of problem,he or she realizes
that the knowledgeof a generaltreatmentof these problemsis lacking. When the
equations are nonlinear,the studentcannot use any of the techniquestaughtin the
courses.
Generalmethodsforfindingexact solutionsof linearand nonlinearpartialdiffer-
ential equations indeed exist. The classical approach is the Lie grouptheory,which
is discussedin detail in [1, 2]. This methodof employingspecial transformations for
findingexact solutionsof differential equationswas inventedby S. Lie approximately
one centuryago. It is onlyin the last halfofthis century,however,that the powerof
thesemethodsis realizedand used widely.An alternativeapproachusingtheso-called
exteriorcalculus was proposed in [3] a fewdecades ago. The detailed presentation
of the theorywith applicationsto physicalproblemscan be foundin [4]. Although
the methods presentedin [1-4] are very powerful,they are too complicatedto be
presentedat the undergraduatelevel. Yet, simplifiedversionsof them may proveto
be quite usefuland accessiblewithouttoo muchcomplicatedtheory.These simplified
techniquescan be added to certaincourses,e.g., differential equations,mechanics.
With this understanding, we will presenthere some simplifiedtechniqueswhich
worksuccessfullyfora large numberof problems. Instead of employinggeneralLie
grouptransformations, we will use special formsof themsuch as scaling,translation,
and the spiral group. The firsttwo workformost of the equations,and the firstone,
the scalingtransformation, producesnontrivialsolutionsin the majorityof cases.
Lie grouptransformations or theirspecial formscan be used in twodifferent
ways:
they can be used to produce a new solutionfroma knownsolutionor they can be

*Received by the editors June 22, 1996; accepted for publication (in revised form) August 23,
1996.
http://www.siam.org/journals/sirev/40-1/31001 .html
tDepartment of Mechanical Engineering,Celal Bayar University,TR-45140, Muradiye, Manisa,
Turkey (mpak@cbu.bayar.edu.tr).
96

This content downloaded from 111.68.98.168 on Tue, 29 Sep 2015 14:17:18 UTC
All use subject to JSTOR Terms and Conditions
 CLASSROOM NOTES 97

used to findthe so-called similaritysolutions (group-invariant solutions). We will

concentrateon the lattertype of solutionsin this article.
Similaritytransformations are the transformations by which an n-independent
variablepartial differential
systemcan be convertedto a systemwithn - 1 indepen-
dent variables. The situationis best when n = 2, since one deals with an ordinary
differential-
equation instead of a partial differential
equation. The above mentioned
transformations are some of the most effectivetools in producingsimilaritytransfor-
mations. Of course,these specifictransformations will not in generalgive all of the
similaritytransformations but may providea usefulpartial list.
We will apply these special transformations to a couple of basic equations in
mathematicalphysics,namely,thewaveequationin mechanicsand the boundarylayer
equation in fluidmechanics. At the end, we will combineall threetransformations
into one transformation and treat a nonlinearheat equation problem. For further
applicationsof these special transformations,see references[5-9].
2. Scaling transformation. The scalingtransformation is one ofthemostcom-
mon. We apply it to the well-known boundarylayerequationsin fluidmechanics.The
equations were derivedat the beginningof the centuryby Prandtl, opening a new
era in fluidmechanics. They are asymptoticapproximationsof the Navier-Stokes
equationsin the vicinityof the surface.The equations greatlyreducethe difficulty
of
solvingthe originalNavier-Stokesequations. For a two-dimensionalincompressible
laminarfluidflow,the equations are
Au Dv
x aDy
(1) au
ux
au 02u dU
ay Dy2 dx'

u (x, O) = O, v(x, O) = , u (x, oo) = U (x).

Prandtlhimselffounda similaritytransformation forthe above equationsby employ-

ing ad hoc methods. We will show that the equations admit scaling symmetryand
reducethe equationsto ordinarydifferential
equationsin a moresystematicway. The
scalingtransformationforthe equations can be writtenas

(2) x* eEax y * =C Eby u* =eECu v* eEdv U*= eeeU.

The above transformation is actually a point transformation

fromthe coordinates
(x, y,u, v, U) to the new coordinates(x*,y*,u*,v*,U*) [2]. The transformation pa-
rameteris e. a, b,c, d, e are arbitraryparametersto be determinedby the invariance
conditionon the equations. Substituting(2) in (1), dividingby the coefficients
of the
leadingtermsin each equation,we have the equations in termsof the new variables
?*+E(b+c-a-d) * -0
Dx* 9ay* d
u*
(3) U + E(b+c-a-d) v eDy*
* -3
* =
=E(2b+c-a) ?2+ eeE2(c-e)U* dU*
Dy*2
U*(X*) 0O)0) V* (X* O) = 0) u*(x*, o) = eE(C-e)U*(X*).

For equations (1), to admit scaling transformation,

the transformedequations (3)
should have exactlythe same formas the originalequation. This immediatelyleads

This content downloaded from 111.68.98.168 on Tue, 29 Sep 2015 14:17:18 UTC
All use subject to JSTOR Terms and Conditions
98 CLASSROOM NOTES

to the followinginvarianceconditions:

(4) b+ c-a-d = 0, 2b+ c-a = 0, c-e = 0.

Solving (4) in termsof the parametersa and c, we have

(5) b = (a -c)/2, d = (c -a)/2, e = c.

admittedby the boundarylayerequations has

Therefore,the scalingtransformation
the followingform:

(6) x* = e Ex y -e(a-c)/2 , U* = eecu V* = e(c-a)/2V U* ecu.

We have two arbitraryparametersa and c, each corresponding to a different

trans-
formation.This givesflexibilityin assigningspecificvalues to the parameters.
Expanding the exponentialsin (6) in a Taylor series, keepingtermsup to or-
der c, denotingthe differencesin transformed and originalvariables as differentials
and finallysolvingfore and equating each term,we obtain the followingequivalent
differential
system:

(7) dx dy du dv dU
ax by cu dv eU

Equation (7) will give us the similarityvariable and functions.Withoutloss of gen-

erality,we assume that c = ma, where m is anotherparameter. Using this new
parameterm and equations (.5), we rewriteequation (7) as follows:

dxa dy du dv dU
(8)
x ((1 - m)/2)y mu ((m - 1)/2)v mmU

Using the methodof characteristicsforsolving(8), we obtain the similarityvariable

and functions

(9) 4 = yx(m-l)/2 u = xmf(), v = X(m-l)/2g9(0) U = kxm,

wherek is a constant.Substituting(9) intothe originalequations (1), we reducethe

systemto an ordinarydifferential
partial differential system

mf + ((in -
1)/2)>f' + 9' = 0,
(10) mf2 + ((in - 1)/2)>ff' + gf' = f"/+ ink2,
f (0) =, g(0) = 0, f (oo) = k.

If therehad been anythingwrongin the similaritytransformation, the equationscould

not be expressedcompletelyin termsof the similarityvariables,hence providinga
checkforthe calculations. The dependentvariableg can be eliminatedbetweenthe
equations,yieldinga singleequationforf. Solvingforf requiresnumericaltechniques,
since the equations are highlynonlinear. In this case, we are unable to findclosed
formsolutions. However,a numericaltreatmentof (10) is easier comparedto the
originalequations (1). We notethat mosttextbookson fluidmechanicspresentad hoc
methodsforfindingthesimilarity transformations. In contrast,a moresystematicway
is presentedherewhichcan be applied to any problemadmittingscalingsymmetry.

This content downloaded from 111.68.98.168 on Tue, 29 Sep 2015 14:17:18 UTC
All use subject to JSTOR Terms and Conditions
 CLASSROOM NOTES 99

3. Translation transformation. Most equations admit translationsin one or

moreof theirvariables. These transformations, however,usuallylead to trivialsolu-
tionswhenused alone. If theyare combinedwithothertransformations, usefulresults
can be obtained. In this section,we will apply this transformation
to the nondimen-
sional wave equation and producethe well-known movingwave solution. For a single
spatial coordinate,the linearwave equation is
02u 02u

0t2 Ox2

The translationtransformation
for(11) is

(12) x x + ca, t* = t +Eb, u u + Ec

wheree is the transformation parameterand a, b, and c are to be determinedfrom

the equation. Substituting(12) into (11), we see that no restrictions
are imposedon
the parametersa, b, and c. Thereforethe equation allows arbitrarytranslationsin all
coordinates.For transformation (12), the equivalentdifferential
systemis
dx dt du
(13)
a b c
Choosing c = 0 and a = mb and solving (13) by the method of characteristics,
we
have the followingsimilarityvariableand function:

(14) r = x-mt, u = F(r)

Substitutinginto (11), we obtain

(15) F" (m2-1) = O.

We now have two choices;eitherF" = 0 or m = ?1. The firstchoiceleads to

(16) F = C1 + C2 or u = klx + k2t+ k3,

whichis thetrivialsolutionthatcan be founddirectlybyexamining(11). For m =1,

we obtain the well-known solution

(17) u = F1(x-t) + F2(x + t),

where F1 representsthe wave travellingto the rightand F2 representsthe wave

travellingto the left. We didn't considerany specificboundaryconditionsforthe
analysisand hence,the solutionsappear in theirgeneralform.
4. Spiral group transformation. In thissection,we again treatthe boundary
layerequations givenin (1). The spiral groupof transformations
forthese equations
are

(18) x x + ca, y* =e Eby u = eECu v = eEdv U* eEeu.

Substituting(18) into (1) and requiringthat the equations remaininvariantunder

the transformationyields

(19) b+c-d=0, 2b+c=0, c-e=0.

This content downloaded from 111.68.98.168 on Tue, 29 Sep 2015 14:17:18 UTC
All use subject to JSTOR Terms and Conditions
100 CLASSROOM NOTES

There is no restrictionon parametera. Withoutloss of generality,

we again assume
that c = ma. Solvingforall parametersin termsof a yieldsb =-(m/2)a, c = ma,
d = (m/2)a, and e = ma. The equations determining the similaritytransformations
are

dx_ dy _ du _ dv _ dU
(20)
(20) 1 -(m/2)y mu (m/2)v mU
The similarityvariableand functionsare

(21) ( = ye(m/2)x u = emxP(() v= e(m/2)xq((), U = kemx.

Substituting(21) into (1), we have

+ q' = 0,
mp+ (m/2)(p'
(22) mp2 + (m/2)(pp' + qp' = p" + nk2,
p(O) = 0, q(0) = 0, p(oo) = k.
These equations representanothersimilaritysolutionforthe boundarylayerequa-
to more involvedproblems,see refer-
tions. For applicationsof this transformation
ences [6, 8].
5. A more general transformation. We propose,in thissection,a moregen-
eral transformationincludingall the previousones as special cases. We apply the
transformationto the followingnonlinearheat conductionequation:
a
au emu
(23) at = ax
r0
O\ xy
whereu is the temperaturedistributionand m is a constant.The transformation
for
the equation is

(24) x* =eax + b, t* eect + Ed, u* e u + cf.

For b = d = f 0, the transformationreduces to the scaling transformation; for

a = c = e = 0, it reducesto the translationtransformation;and finally,fora = d =
f = 0, it reducesto the spiral grouptransformation.Substituting(24) into (23) and
requiringthe invarianceconditionyields

(25) e = 0, 2a-c-mf =0.

Solving forf in termsof the otherparameters,we see that the transformation

has
fourarbitraryparameters:

(26) x* =eax + b, t* eect + Ed, u* u + c((2a - c)/m).

Actually,transformation (26) representsthe fullgroupforthis heat equation [1]. For

a special choiceof a = c = m and b = d = 0, we have
dx dt du
mx mt 1
Solving (27), we findthe followingsimilarityvariableand function:

(28) , = x/t, u = (1/m) logt + r(A).

This content downloaded from 111.68.98.168 on Tue, 29 Sep 2015 14:17:18 UTC
All use subject to JSTOR Terms and Conditions
 CLASSROOM NOTES 101

Substitutionof the resultsinto the originalequation reducesit to an ordinarydiffer-

ential equation

(29) m(emrr+)? m,r' 1.

The above equation is easier to handle comparedto the originalpartial differential

equation. Note that we did not considerany boundaryconditionsforthe heat equa-
tion. Imposingboundaryconditionsmay furtherrestrictthe transformation (26).
In all the examples,we treated partial differential
equations with two indepen-
dentvariables. For equationshavingmorethan two independentvariables,similarity
transformations can be applied successivelyto achieve finalreductionsto ordinary
differential equations. An example with three independentvariables can be found
in [9].

REFERENCES

[1] G. W. BLUMAN AND S. KUMEI, Symmetriesand DifferentialEquations, Springer-Verlag,New

York, 1989.
[2] H. STEPHANI, DifferentialEquations: Their Solutions Using Symmetries,Cambridge Univer-
sity Press, Cambridge, 1989.
[3] B. K. HARRISON AND F. B. ESTABROOK, Geometric approach to invariance groups and solu-
tion of partial differentialequations, J. Math. Phys., 12 (1971), pp. 653-666.
[4) D. G. B. EDELEN, Applied Exterior Calculus, Wiley-Interscience,New York, 1985.
[5] A. G. HANSEN AND T. H. NA, Similaritysolutions of laminar, incompressibleboundarylayer
equations of non-Newtonianfluids,ASME J. Basic Engrg., 90 (1968), pp. 71-74.
[6] M. G. TIMOL AND N. L. KALTHIA, Similarity solutions of three dimensional boundary layer
equations of non-Newtonianfluids,Internat. J. Non-Linear Mech., 21 (1986), pp. 475-481.
[7] M. PAKDEMIRLI AND E. S. SUHUBI, Similaritysolutions of boundarylayer equations for second
orderfluids, Internat. J. Engrg. Sci., 30 (1992), pp. 611-629.
[8] M. PAKDEMIRLI, Similarity analysis of boundarylayer equations of a class of non-Newtonian
fluids,Internat. J. Non-Linear Mech., 29 (1994), pp. 187-196.
[9] M. YURUSOY AND M. PAKDEMIRLI, Symmetry reductions of unsteady three-dimensional
boundary layers of some non-Newtonian fluids, Internat. J. Engrg. Sci., 35 (1997),
pp. 731-740.

This content downloaded from 111.68.98.168 on Tue, 29 Sep 2015 14:17:18 UTC
All use subject to JSTOR Terms and Conditions