January 29, 2009 14:50 WSPC/INSTRUCTION FILE HeRitz 9 He’s amazing calculations with the Ritz method 0 0 2 n a PaoloAmore∗ J 9 Facultad de Ciencias, Universidad de Colima, Bernal D´ıaz del Castillo 340, Colima, Colima, 2 Mexico ] h p FranciscoM.Ferna´ndez† - h t INIFTA (UNLP, CCT LaPlata-CONICET), Divisi´onQu´ımica Te´orica, Diag. 113 y 64 (S/N), a m Sucursal 4, Casilla de Correo 16, 1900 La Plata, Argentina [ 1 v We discuss an earlier application of the Ritz variational method for strongly nonlinear 0 problems.Weclearlyprovethattheresultsderivedforseveralextremelysimpleproblems 6 6 ofsupposedlyphysicalandmathematical interestdonotprovideanyclueontheutility 4 oftheapproach. . 1 0 9 1. Introduction 0 : v In a recent review in this journal, He [1] analyzed several asymptotic methods for i X stronglynonlinearequations.Onesuchapproach,thesocalledRitzmethod,consists r a mainly in converting the nonlinear differential equation into a Newton equation of motion. Thus, by minimization of the action integral for the Lagrangian function oneobtainsanapproximatesolutiontothenonlinearequationthatisexpectedtobe optimalfromavariationalpointofview.Obviously,theaccuracyoftheapproximate solution depends heavily on the chosen variational ansatz or trial “trajectory”. The purposeofthispaperistoanalyzethe Ritz methodproposedbyHe[1]and determine if it is already useful for solving actual nonlinear problems. For the sake ∗[email protected] †[email protected] 1 January 29, 2009 14:50 WSPC/INSTRUCTION FILE HeRitz 2 P. Amore and F. M. Fern´andez of clarity in what follows we devote a section to each of the problems discussed. 2. Anharmonic oscillator First, He [1] transforms the equation of motion for the Duffing oscillator u¨(t) u(t)+ǫu(t)3 =0 (1) − into the variational integral 1 1 ǫ J(u)= u˙2 u2+ u4 dt (2) −2 − 2 4 Z (cid:18) (cid:19) and concludes that “it requires that the potential V(u) = u2/2+ǫu4/4 must be − positive for all t > 0, so an oscillation about the origin will occur only if ǫA2 > 2, where A is the amplitude of the oscillation”. Unfortunately, He [1] does not show the trial function from which he draws that conclusion. The Duffing oscillator has been widely studied and, consequently, its properties are well known [2]. For example, from straightforward inspection of the potential V(u) we already know that there is an unstable equilibrium point at u = 0 when ǫ<0.Onthe otherhand,whenǫ>0the potentialV(u)exhibits a localmaximum V =0atu=0andtwominima ofdepth 1/(4ǫ)symmetricallylocatedat 1/√ǫ. − ± IftheinitialconditionsaresuchthatV(A)>0thentheoscillationwillcertainlybe abouttheorigin;otherwisetherewillbeoscillationsaboutoneofthetwominima.It isclearthatbysimpleinspectionofthepotentialweobtainmuchmoreinformation that the one derivedby He [1] fromthe actionintegral.Therefore,He’s application of the Ritz method to this model is of no relevance whatsoever. 3. A chemical reaction He [1] proposed the application of the Ritz method to the chemical reaction nA C+D (3) → IfN (t),N (t),andN (t)arethenumberofmoleculesofthespeciesA,B,andC, A B C respectively,attimetthenHe[1]assumedthatN (0)=a,andN (0)=N (0)=0. A B C January 29, 2009 14:50 WSPC/INSTRUCTION FILE HeRitz Ritzmethod for nonlinear problems 3 If we call x = N (t) = N (t), then we conclude that N (t) = a nx, where x is B C A − knownasthe extent ofreaction[3,4].The unique rate ofreactioncanbe defined in terms of the extent of reaction as v = dx/dt. He [1] further assumed that the rate law is given by dx =k(a x)n (4) dt − At this point we stress the fact that this expression is correct only if the chemical reaction (3) is elementary, otherwise the rate law may be more complicated. Most chemicalreactionsarenotelementaryandthereforethe reactionorderandmolecu- laritydonotnecessarilyagree,asdiscussedinanybookonphysicalchemistry[3]or chemicalkinetics[4].Whatismore,theorderofreactionmaynotevenbeapositive integer [3,4]. For concreteness here we assume that the rate law (4) is correct. He[1]obtainedanapproximatesolutiontothedifferentialequation(4)bymeans of the action integral ∞ 2 1 dx J = +k2(a x)2n dt (5) 2Z0 "(cid:18)dt(cid:19) − # and the variational ansatz xvar =a 1 e−ηt (6) − (cid:0) (cid:1) where η is a variational parameter. Notice that xvar(t) satisfies the boundary con- ditions at t = 0 and t . He [1] found that the optimal value of the effective → ∞ first–order rate constant η was given by kan−1 η = (7) √n Furthermore,He[1]arguedthatchemistsandtechnologistsalwayswanttoknowthe half–time t =t(x=a/2)(whichhe calledhalfwaytime). Accordingto equations 1/2 (6) and (7) the half–time is given approximately by √nln(1/2) tvar = (8) 1/2 − kan−1 According to He [1] the exact reaction extent for n=2 is 1 xexact(n=2)=a 1 (9) He − 1 kat (cid:18) − (cid:19) January 29, 2009 14:50 WSPC/INSTRUCTION FILE HeRitz 4 P. Amore and F. M. Fern´andez This resultis obviouslywrongbecause itexhibits anunphysicalpole att=1/(ka). From this incorrect expression He [1] derived a meaningless negative half–time 1 tHe(n=2)= (10) 1/2 −ka Inordertoobtainareasonableagreementwiththevariationalresult(8)He[1]then carried out the following wrong calculation √2ln(1/2) 0.98 tvar(n=2)= = (11) 1/2 − kan−1 − ka InthiswayHe[1]managedtoobtaintwounphysicalnegativehalf–timesthatagreed 98%. Disregarding the mistakes outlined above we may ask ourselves whether the approximate variational result may be of any utility to a chemist. Any textbook on physical chemistry [3] or chemical kinetics [4] shows that the exact solution to equation (4) is 1 xexact =a 1 , n=1 (12) ( − [1+k(n 1)an−1t]1/(n−1)) 6 − and that the exact half–time is given by 2n−1 1 texact = − (13) 1/2 k(n 1)an−1 − Itiscommonpracticeinchemistrytoestimatethehalf–timefromexperimentaldata inordertodetermine theorderofthe reaction.Obviously,aninaccurateexpression would lead to an inexact order of reaction. Thevariationalhalf–time(8)isreasonablyaccurateforn=2becauseitisexact for n = 1. The reason is that the variational ansatz (6) is the exact solution for a first–order reaction when η = k. Notice that equation (7) leads to such a result when n = 1. We can easily verify that the ratio tvar/texact increasingly deviates 1/2 1/2 from unity as n increases.Therefore, n=2 (the only case selected by He [4]) is the most favorable case if n is restricted to positive integers greater than unity. The half–time (or half–life) is a particular case of partial reaction times. We may, for example, calculate the time t = t that has to elapse for the number of 1/4 January 29, 2009 14:50 WSPC/INSTRUCTION FILE HeRitz Ritzmethod for nonlinear problems 5 A molecules to reduce to a/4 (x=3a/4). It is not difficult to verify that texact 1/4 =2n−1+1 (14) texact 1/2 From the experimental measure of t and t chemists are able to obtain the 1/2 1/4 reaction order n. However, if they used He’s variational expression (6) they would obtain tvar 1/4 =2 (15) tvar 1/2 that is useless for n=1. According to what we have said above it is not surprising 6 that this ratio is exact for n = 1. We clearly appreciate that the variational result does not provide the kind of information that chemists would like to have because it only predicts first–order reactions. From the discussion above we conclude that no chemist will resortto the varia- tionalexpressionsinthestudyofchemicalreactions.Thereisnoreasonwhatsoever fortheuseofanunreliableapproximateexpressionwhenonehasasimpleexactan- alyticaloneathand.Besides,wehaveclearlyprovedthatthevariationalexpressions are utterly misleading. 4. Lambert equation He [1] also applied the Ritz method to the Lambert equation k2 y′(x)2 ′′ y (x)+ y(x)=(1 n) (16) n − y(x) and arrived at the variational formulation 1 J(y)= n2y2n−2y′2+k2y2n dt (17) 2 − Z (cid:0) (cid:1) By means of the transformation z =yn He obtained 1 J(z)= z′2+k2z2 dt (18) 2 − Z (cid:0) (cid:1) that leads to the Euler–Lagrangeequation z′′+k2z =0 (19) January 29, 2009 14:50 WSPC/INSTRUCTION FILE HeRitz 6 P. Amore and F. M. Fern´andez Obviously, the solution to this linear equation is straightforward. Ifwesubstitutethetransformationz =ynintoequation(16)weobtainequation (19) in a more direct way. Therefore, there is no necessity for the variational Ritz method. 5. Soliton solution He [1] also studied the KdV equation ∂u(x,t) ∂u(x,t) ∂3u(x,t) 6u(x,t) + =0 (20) ∂t − ∂x ∂x3 and looked for its travelling–wavesolutions in the frame u(x,t)=u(ξ), ξ =x ct (21) − The function u(ξ) satisfies the nonlinear ordinary differential equation ′′′ ′ ′ u (ξ) cu(ξ) 6u(ξ)u(ξ)=0 (22) − − where the prime indicates differentiation with respect to ξ. Then He [1] integrated thisequation(takingtheintegrationconstantarbitrarilyequaltozero)andobtained u′′(ξ) cu(ξ) 3u(ξ)2 =0 (23) − − By means of the so called semi–inverse method He [1] obtained the variational integral ∞ 1 du 2 c J = + u2+u3 dt (24) 2 dξ 2 Z0 " (cid:18) (cid:19) # Choosing the trial function u=pcosh−2(qξ) (25) where p and q are variational parameters, He [1] obtained p=c/2 and q =√c/2. By substitution of equation (25) into equation (22) we obtain the same values of p and q in a more direct way and with less effort. Therefore,there is no need for the variational method for the successful treatment of this problem. January 29, 2009 14:50 WSPC/INSTRUCTION FILE HeRitz Ritzmethod for nonlinear problems 7 6. Bifurcation He [1] also applied the Ritz method to the most popular Bratu equation u′′(x)+λeu(x) =0, u(0)=u(1)=0 (26) that has been studied by severalauthors [5] (and references therein). Here we only cite those papers that are relevant to present discussion. He [1] derived the action integral 1 1 J = u′2 λeu dx (27) 2 − Z0 (cid:18) (cid:19) and proposed the simplest trial function that satisfies the boundary conditions: u(x)=Ax(1 x) (28) − Curiously, He [1] appeared to be unable to obtain an exact analytical solution for the integral; however, it is not difficult to show that A2 π J(A)= λeA/4erf √π/2 (29) 6 − A r (cid:0) (cid:1) We cannot exactly solve dJ(A)/dA=0 for A but we can solve it for λ: 4A5/2 λ= (30) 3 √π(A 2)eA/4erf √A/2 +2√A − h (cid:16) (cid:17) i The analysis of this expression shows that λ(A) exhibits a maximum λ = c 3.569086042 at A = 4.727715383. Therefore there are two variational solutions c for each 0 < λ < λ , only one for λ = λ and none for λ > λ . This conclusion c c c agreeswiththerigorousmathematicalanalysisoftheexactsolution[5]thatwewill discuss below. Besides, the critical value of the adjustable parameter A is also a c root of d2J(A)/dA2 =0. The exact solution to the one–dimensional Bratu equation (26) is well–known. Curiouslyenough,He[1],Deebaetal[6],andKhury[7]showedawrongexpression. A correct one is (notice that one can write it in different ways) cosh[θ(x 1/2)] u(x)= 2ln − (31) − cosh(θ/2) (cid:26) (cid:27) January 29, 2009 14:50 WSPC/INSTRUCTION FILE HeRitz 8 P. Amore and F. M. Fern´andez where θ is a solution to 2θ2 λ= (32) cosh(θ/2)2 The critical λ–value is the maximum of λ(θ), and we easily obtain it from the root of dλ(θ)/dθ =0 that is given by eθc(θ 2) θ 2=0 (33) c c − − − Theexactcriticalparametersareθ =2.399357280andλ =3.513830719thatlead c c ′ to u(0) = 4. We appreciate that the variational approach provides a reasonable c qualitative (or even semi quantitative) description of the problem. We may try a perturbation approach to the Bratu equation in the form of a Taylor series in the parameter λ: ∞ u(x)= u (x)λj (34) j j=0 X where, obviously, u (x)=0. From the exact expression we obtain 0 λ λ2 λ3 u′(0)= + + +... 0.5λ+0.0417λ2+0.00625λ3+... (35) 2 24 160 ≈ while the variational approach also yields a reasonable result λ λ2 43λ3 u′(0)= + + +... 0.5λ+0.05λ2+0.00768λ3+... (36) 2 20 5600 ≈ It seems that the Ritz method alreadyproduces satisfactoryresults for this kind of two–point boundary value problems. Anothersimple variationalfunctionthatsatisfiesthe sameboundaryconditions is u(x)=Asin(πx) (37) It leads to the following variational integral: A2π2 J(A)= λ[I (A)+L (A)] (38) 0 0 4 − where I (z) and L (z) stand for the modified Bessel and Struve functions [8], re- ν ν spectively. From the minimum condition we obtain Aπ3 λ= (39) 2 2+π[I (A)+L (A)] 1 1 { } January 29, 2009 14:50 WSPC/INSTRUCTION FILE HeRitz Ritzmethod for nonlinear problems 9 It is not difficult to show that this trial function yields better critical parameters: ′ λ = 3.509329130 and u(0) = 3.756549365. Besides, one can easily derive the c c approximate perturbation expansion exactly 4λ 4λ2 4 3π2+16 λ3 4 π2+18 λ4 ′ u(0)= + + + +... π2 π4 3π8 π10 (cid:0) (cid:1) (cid:0) (cid:1) 0.405λ+0.0411λ2+0.00641λ3+... (40) ≈ Notice that although the coefficient of λ is not exact the remaining ones are more accurate than those of the preceding trial function. ′ Fig. 1 shows the exactslope at originu(0) in terms of λ and the corresponding estimatesgivenbythetwovariationalfunctions.Weappreciatethatthevariational ′ approach proposed by He [1] yields the solution with smaller u(0) (lower branch) more accurately than the other one (upper branch). This comparison between the exact and approximate solutions for a wide range of values of λ was not carried out before; He [1] simply compared the two slopes at origin for just λ = 1. The other trialfunction (37) yields a better overallapproximationatthe expense of the accuracyforsmallvaluesofλ.Somemoreelaboratedapproaches,liketheAdomian decomposition method, fail to provide the upper branch [6]; therefore, the Ritz method seems to be suitable for the analysis of this kind of nonlinear problems. WemayconcludethattheRitzvariationalmethodprovidesausefulinsightinto the Bratuequation.However,oneshouldnotforgetthatthereis arelativelysimple exact solution to this problem and that the generalization of the approach in the form of a power series proposed by He [1] u(x)=Ax(1 x)(1+c x+c x2+...) (41) 1 2 − may surely lead to rather analytically intractable equations. 7. Conclusions Historically,scientistshavedevelopedperturbational,variationalandnumericalap- proaches to solve nontrivial mathematical problems in applied mathematics and January 29, 2009 14:50 WSPC/INSTRUCTION FILE HeRitz 10 P. Amore and F. M. Fern´andez theoretical physics. In some cases, where the exact solution exists but is given by complicated special functions, an approximate simpler analytical solution may nonetheless be ofpracticalutility. However,He [1]chose examples where either the Ritz method does not provide any useful insight, or the exact analytical solutions areassimpleasthe approximateones,orthe directderivationofthe exactresultis more straightforwardthan the use of the variational method. From the discussions in the preceding sections we may conclude that He’s application of the Ritz varia- tional method [4] does not show that the approach is suitable for the treatment of nonlinear problems. In most of the cases studied here the straightforward analysis of the problem yields either more information or the same result in a more direct way. To be fair we should mention that the Ritz variational method provides a rea- sonable bifurcation diagram by means of relatively simple trial functions as shown in Fig. 1. However,evenin this case the utility of the approachis doubtful because thereexistsaremarkablysimpleanalyticalsolutiontothatequation.Thetreatment of a nontrivial example is necessary to assert the validity of the approach. He’s choice of the rate equation for chemical reactions [1] is by no means a happy one (without mentioning the mistakes in the calculations). In this case the exact solution is quite simple and the variational ansatz is unsuitable for practi- cal applications. We may argue that a trial function with the correct asymptotic behaviour would yield meaningful results. In fact, it may even produce the exact result; but one should not forget that such a success wouldobviously be due to the factthatthereexistaremarkablysimpleexactsolutionavailablebystraightforward integration. Assaidbefore,presentresultsshowthatHe[1]failedtoprovethattheRitzvari- ational method provides a successful way of treating strongly nonlinear problems. Of course,the main ideas behind that variationalmethod are correct,and the case of Bratu equation suggests that it may be possible to find appropriate trial func- tions for the successful treatment of some problems. Unfortunately, the remaining