CHAPTER 8 Integrable systems and number theory Peter H. van der Kamp, Jan A. Sanders and Jaap Top 1. Introduction The evolution equation (KdV) where is the derivative of with respect to , was derived to describe water waves in shallow channels [KdV95]. It bears the names of Korteweg and de Vries and has become the prototype integrable nonlin- ear partial differential equation since the work in [Miu68, MGK68, SG69, Gar71, KMGZ70, GGKM74]. The KdV equation is for the field of inte- grability what the harmonic oscillator is for quantum mechanics: the system for which every method works. KdV has infinitely many conservation laws, infinitely many (co)symmetries, it has (co)symplectic and recursion opera- tors, a Lax-pair, a bilinear form, the Painlev´ e property, stable local solutions and it can be solved by the inverse scattering method. Once all these properties are established, the obvious next question is: are there more equations like this? The goal we have in mind is the classi- fication of ’integrable’ partial differential equations. A few choices have to be made here, for instance, what kind of equations are to be classified and what exactly will be our definition of integrability. We will focus on the existence of infinitely many generalized symmetries (a precise definition follows in the next section). We limit ourselves to polynomial equations, mainly because for these the symbolic calculus can be used. For classification results in the general case we refer to [Zak91], in particular [M ˇ SS91]. We remark here that these lists give a different kind of classification, since they allow for much larger classes of transformations. Contrary to our analysis they can only classify one order at the time and cannot exclude the possibility that higher order equations are integrable. The classification results given here, which are based on the thesis of Jing Ping Wang [Wan98], and following publica- tions, do allow, at least in the scalar case, a complete classification up till all orders. This breakthrough was made possible by the use of the symbolic 187
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
CHAPTER 8
Integrable systemsand number theory
PeterH. vanderKamp,JanA. SandersandJaapTop
1. Intr oduction
Theevolutionequation��������������� (KdV) �where �� is the ��� � derivative of ����� ����� with respectto � , wasderived todescribewaterwavesin shallow channels[KdV95]. It bearsthenamesofKorteweg and de Vries and hasbecomethe prototypeintegrablenonlin-earpartialdifferentialequationsincethework in [Miu68, MGK68, SG69,Gar71, KMGZ70, GGKM74 ]. TheKdV equationis for thefield of inte-grability whattheharmonicoscillatoris for quantummechanics:thesystemfor whicheverymethodworks.KdV hasinfinitely many conservationlaws,infinitely many (co)symmetries,it has(co)symplecticandrecursionopera-tors,aLax-pair, abilinearform, thePainleveproperty, stablelocalsolutionsandit canbesolvedby theinversescatteringmethod.
Onceall thesepropertiesareestablished,theobviousnext questionis:aretheremoreequationslike this? Thegoalwe have in mind is theclassi-ficationof ’integrable’partialdifferentialequations.A few choiceshave tobemadehere,for instance,whatkind of equationsareto beclassifiedandwhat exactly will be our definition of integrability. We will focuson theexistenceof infinitely many generalizedsymmetries(a precisedefinitionfollows in thenext section).
We limit ourselvesto polynomialequations,mainly becausefor thesethesymboliccalculuscanbeused.For classificationresultsin thegeneralcasewereferto [Zak91], in particular[MSS91]. Weremarkherethattheselists givea differentkind of classification,sincethey allow for muchlargerclassesof transformations.Contraryto our analysisthey canonly classifyoneorderat the time andcannotexcludethe possibility that higherorderequationsare integrable. The classificationresultsgiven here,which arebasedon the thesisof Jing Ping Wang [Wan98], and following publica-tions,do allow, at leastin the scalarcase,a completeclassificationup tillall orders.Thisbreakthroughwasmadepossibleby theuseof thesymbolic
187
188 PETER H. VAN DER KAMP, JAN A. SANDERS AND JAAP TOP
methodandthe subsequentapplicationof numbertheoreticandalgebraicgeometricmethodsandgivesus an extremelyelegantapplicationof puremathematics.As an illustration of the impact of this kind of theory, letus mentionthe fact that only a decadeago it was perfectly legitimate topublishpaperson the searchof integrableseventhandninth orderscalarpolynomialequations,cf. [Ger96, Ger93, GKZ90]. The presentresultsshow why theseeffortswereall in vain.
2. Scalarequations
2.1. Generalizedsymmetries. Let � ��� � and � ��� � befunctionsof ������ anda finite numberof its derivatives �� !� "$#&%"(' # . The function � ��� � iscalledageneralizedsymmetryof theequation����� � ��� � if ) �*�!�,+ � ��� �satisfies) �-� � � ).� �0/1��+32 � . The word ‘generalized’is includedin thisdefinition to stressthe fact that � in principle dependsnot only on � , butalsoon someof its derivatives �� .
A completelyalgebraicdescriptionof thisnotion,limited to polynomialequations,is obtainedasfollows. Write 465 �87:9;� � �<� � � 2 �>=>=>=;? for thepoly-nomial ring in infinitely many variables�@�A��� � ��� � � 2 �B=>=>= andlet CEDF4 .Write G ' for thederivationon 4 definedby G ' ���� � �H�� &IJ� for all KMLAN .Then G ' � CO� �QP SRT� �� UIJ�WV CV �� =
Denoteby XBY the uniquederivation on 4 satisfying XZY ��� � � C andXBY\[:G ' � G ' [:XZY . Thisoperatorcan(formally) bewrittenas
XZY �QP 3RT� G ' � CO� VV �� =This derivation XBY is the prolongationof the evolutionaryvectorfieldwithcharacteristicC , cf. [Olv93, equation5.6]. Also, XBY ��] � � G_^ � C�� , whereG_^is theFrechetderivativeof ] , cf. [Olv93, proposition5.25].
Extend XZY to a derivation on the dual numbers4 9;+ ? (with + 2 � N )by XBY ��]��`+ba � � XBY ��] � �`+ XBY �ca � for ] � a Dd4 . With thesenotations,�dDA4 is a generalizedsymmetryof � DA4 (or, of the equation���e�� ��� � ��� � � 2 �>=>=>=f� ), if
Hence� is a generalizedsymmetryof theequation����� � ��� � ��� �>=>=>=v� pre-ciselywhentheLie bracket 9 �w�$��? � XBgW�@nhXZtT� vanishes.
If onewrites x�Y for the automorphismof 4 9;+ ? which sends]_�*aq+ to]y�z�cae� XBY ��] ��� + , thenyet anotherway to definea symmetry� of � is bydemandingthat xt � �s� � x�g � �i� .
Notethatthebracketdefinedhereis indeedaLie bracketon 4 , i.e., it isbilinearandit satisfies9 C��$C�? � N and 9{9 C�� ] ?c� a ? �9|9 ] � a ?}�$C�? �9|9~a �$C�?c� ] ? � N .Onewayto verify this,is bynotingthat X3� Y�� ^�� � XBYS[TX$^�n!X$^�[uXBY is thestandardLie bracketof XBY and Xj^ in theLie algebraof all derivationson 4 . As usualfor C�D�4 thelinearoperator]��� 9 C�� ] ? is calledtheadjointof C andit iswritten asad� C�� . Thesetof generalizedsymmetriesof ����� � ��� � ��� �B=>=>=f�is preciselythe kernelof ad� �s� . Note that by the generaltheory of Liealgebras,thesegeneralizedsymmetriesof � alsoform a Lie algebra(withthesameLie bracket).
DEFINITION 2.1. An equation���� � is calledintegrable if thespaceof generalizedsymmetriesof � is infinite dimensional(over 7 ), andalmostintegrableof depth(at least,at most) � if thespaceof generalizedsymme-tries of � is exactly (at least,at most) � -dimensional.Whenan equationis almostintegrablebut not integrablewe saythat it is almostintegrableoffinitedepth.
EXAMPLE 2.2. For any equation����� � ��� � , thefunctions�<� and � ��� �aregeneralizedsymmetriesasis easilyverified. 7 -linear combinationsof��� and � ��� � arecalledtrivial symmetries;all otheronesnontrivial.
EXAMPLE 2.3. A nontrivial symmetryof theKdV equation�����Q��������<� is �����`�� ��������� N� ����� 2 �A�� � 2 �<� �ascanbeverifiedby a tediousbut straightforwardcalculation.In thenextsectionwewill explainhow onemayfind suchasymmetry.
2.2. � -homogenuityandgrading. Oneassociatesweightsto themono-mialsin 4 by fixing some�wDw� andassigningto themonomial�� ��<�U�U���� ��theweight ��� � K �O� =B=>= � K�� . If everymonomialin � hasthesameweight,thentheequation����� � is called � -homogeneous. Notethatthis dependsonthechoiceof � ; for example,in theKdV equationonehas� ��� � �*��������<� with two monomialsof weight
�and � � � , respectively. Henceonly
with thechoice � �H� this equationis � -homogeneous(of weight�). One
seesthatwith this choice,alsothesymmetry���� �� ������ ���� ����� 2 � �� � 2 ���is � -homogeneous,of weight � . In thesecasestheweightequalstheorder,which is by definitionthehighest� suchthat �� occursin theexpression.
190 PETER H. VAN DER KAMP, JAN A. SANDERS AND JAAP TOP
EXAMPLE 2.4. We fix � � � and do the Lie-bracket calculationofthesymmetryof weight � of the � -homogeneousKdV, usinghomogeneityandgrading. Put � �·���:�z����� . The symmetry � we try to find, canbe written as � � � �� � � �� � � 2� with � � D ¢ � . If the linear part � �� ofthesymmetryis nonzero,memayafter rescalingsupposethat � �� �`��� . Itcommuteswith � �� �Q��� , hencethe
¢ �-partof 9 �w�$��? is indeedN . Because
of homogeneitythequadraticanddegreethreepartsof � canbewritten as� �� ��¸¹�º��������E¸ 2 � 2 ��� and � 2� �*¸«�(����� 2� .The
Eventhoughthis illustratestheuseof thesymbolicmethodon a rathertrivial example,alreadyhereonecanseethe simplification it brings andthepossibility to apply for instancetheinvarianttheoryof thepermutationgroupto attacktheclassificationproblem.
if �såæN .With thesenotations,oneof thestatementsin Proposition8 is that forCwD ¢ Ø and �sL®N onehas Ê9��� �$C�? � × Ø ÆC .
EXAMPLE 2.7. We now turn backto our KdV computationandtry tofind thequadraticandcubicpartsof a � -homogeneoussymmetryof weight� �»� . Wewrite this symmetryas
194 PETER H. VAN DER KAMP, JAN A. SANDERS AND JAAP TOP
This is a polynomialwhich indeeddeterminesa unique � � . Now considerthe cubic term � 2 3± 2 . To solve the equation 9;��� �$� 2 S± 2 ? �ç9������ �(� � ? � Ncorrespondingto the
Proof.Bothsidesof theequalityarelinearin both C and ] , henceit sufficesto considerthe case C � ��êj�që and ]z� �� S� Ø . Here a straightforwardcalculationproves the result (in fact, it is hardly more work to stateandproveananalogousresultfor CwD ¢ � and ] D ¢´¶ ). Õ
This lemmaimpliesthatasolution � 2 S± 2 existspreciselywhen
DEFINITION 2.8. An element� D ¢is callednonlinearinjective if9 �w��ôy?<D§í &IJ� implies ô½Dlí &IJ� for all ô½Dlí ��K�åðN .
Note that ad� �s� definesa linear mapfrom í to itself, for all K , andhencean inducedlinear map: í ã í �IJ� � í ã í &IJ� for all K . Nonlinearinjective preciselymeansthat theseinducedmapsareinjective, for KMåAN .
DEFINITION 2.9. One calls � D ¢relative õ -prime with respectto�D ¢ if for all KiLðõ andfor all ô½Dlí onehasthat 9 ���(ô!?�D ad� �s� � í ã �
implies ô½D ad� �s� � í ��ö òT÷ í �IJ� .To explain this terminology, notethat for ôøD�í and �*D ¢ theclass9 ����ôy?�ö ò.÷ í &IJ� only dependson the Lie bracket 9 � � �(ô!? of ô with the
linearpart � � of � . Similarly, ad� �s� � í ã í �IJ� � � ad� � � � � í ã í �IJ� � . Nowin thesymboliclanguage,takingtheLie bracketof alinearterm o � �j�T� andaterm C,D ¢ correspondsto multiplying ÆC by o � � × � . If two polynomials]T� � ] 2 of thiskindarerelativelyprimein theusualsense,then]u� ÆC is divisibleby ] 2 preciselywhen ÆC is divisible by ] 2 . Thus, the polynomialsbeingrelativeprimeimpliesthatthecorresponding� � and � � (andhence� and� ) arerelative õ -prime,for all õ .
Thefollowing implicit functiontheoremfor filteredLie-algebras,whichis to befoundin [SW98] andin [Wan98, Section2.9],canbeusedto provetheexistenceof infinitely many symmetries.
THEOREM 2.10(Sanders,Wang). Let í bea filteredLie algebra whichis completewith respectto thefiltration topology. Suppose�w�$� and ôùDí � satisfyú 9 �w�$��? � Nú � is nonlinearinjectiveú � is relatively õ -primewith respectto �ú 9 �w�(ô!?ÅDlí ëú 9 ���(ô!?�Dlí � .Thena unique ûô½D§í ë existssuch that ô � ûô is a symmetryof both � and� , i.e., 9 �w��ô � ûô!? � N �A9 ���(ô � ûô!?c=
Theproofof thisis actuallyrathersimple:since9 �w�$��? � N and 9 ôü���l?ÅDí ë , it follows that 9 �w� 9 ���(ô!?|? � n 9 ��� 9 ôü���l?|?:DEí ë . Nonlinearinjectivityof � now implies 9 ���(ô!?ýDzí ë . Moreover, the sameequalityshows that9 ��� 9 �w�(ô!?{?\D ad� �s� � í ë ã í ë�IJ� � . Since � is relatively õ -primewith respectto � , it follows that 9 �w�(ô!?�þ 9 �w��ô ¡ ?�ö òT÷ í ëcIJ� for someô ¡ Dlí ë . Hence9 �w��ô`n®ô ¡ ?ÿDæí ëcIJ� and,usingnonlinearinjectivity of � asbefore,also9 ����ô§n_ô ¡ ?ÅD§í ëcIJ� . By induction,thesameargumentyieldsfor every �§åæNan elementô!¡ªD*í ë for which 9 �w�(ô½n®ô!¡v? and 9 ���(ô½n®ôy¡v? arein í ëcI�Ë .Completenessof í finishestheargument.
EXAMPLE 2.11. Wecannow provethattheKdV equationhasinfinitelymany symmetriesusingthis implicit functiontheorem.So,take � �Q�����
196 PETER H. VAN DER KAMP, JAN A. SANDERS AND JAAP TOP���<� and � �*����� �� ������� ���� ����� 2 � �� � 2 ��� . Considerany oddinteger � andput ô �®�� UI 2 � � � � � 2 S± 2 with � � �$� 2 S± 2 asobtainedin Example2.7.ú
Wehavethat 9 �w�$��? �½9;���3�à����� � ���U� �� �����3� ���� ����� 2 � �� � 2 ��� ? � Nq=úThe actionof ad� �s� on í ã í &IJ� equalsthat of ad����� � . This iseasilyverified to be injective for Keå N (e.g.,usingthe symbolicmethod).úTake CÈD ¢ . Usingthesymboliclanguageonefindsthat 9 ���$C�?�Dad� �s� � í ã í &IJ� � implies C�D ad� �s� � í ã í &IJ� � whenever
× � and× � arerelativeprime.It is aneasytaskto verify thatboth× �� and
× ��areirreducible.Because
× ¥  � #�� ² й��Ãr� × ³±�� , all× with � � � � �
and K§å �are irreducibleas well. It follows that � is relative
3-primewith respectto � .úWehaveshown in Example2.7 that 9 �w�(ôy?ÅD§í � .úSince 9;��� � �� ? � N , it follows that 9 ���(ôy?ÅD§í � .
The implicit function theoremthereforeyields a Cauchysequencefor thefiltration topology � ô � ô � 3R�� , with all ô D®í � and 9 �w�(ô � ô ? and9 ����ô � ô ? elementsof í . Thepartof ô � ô which is � -homogeneousof weight � �»� thenhasbebeindependentof � for �°åyå®N , anddefinesanontrivial symmetryof both � and � . Thisshowstheexistenceof infinitelymany independentsymmetriesof theKdV equation.
3. Classificationresults
3.1. Positiveweight. Usingthesymbolicmethodandtheimplicit func-tion theorem,thepapers[SW98] and[SW00] classifyall � -homogeneousequationsof theform ���<���� ´� C ����� �>=>=>=Z� �� S±�� �whichhave infinitely many independentsymmetries,in thecase�lL®N . Webriefly indicatethestrategy for thecase�§åæN .
Usingdiophantineapproximationtheory, F. Beukers[Beu97] provedausefulresultconcerningthemutualdivisibility of the � -polynomials:
PROPOSITION 9. For integers á ������K�L � with � ¿� á , thepolynomials× �½� ¾ �<� =>=>= � ¾ &IJ� � nȾ � nð=>=B=�n�¾ &IJ�havetheproperty ��� ÷ � × � × Ø � � � exceptin thefollowingcases.
cubic terms,thenit cannothave a nontrivial symmetry. Supposenow thatthe equation� of order � hasno quadraticterms. If � hasa nontrivialsymmetry, thenit hasa nontrivial � -homogeneousone.This meansin par-ticular that its linearpart containsexactly one � Ø . Now first of all both �and á have to beodd.Wefind � 2Ø � × 2Ø× 2 � 2 =In particular,
theimplicit functiontheorem,it followsthat � hasa � -homogeneoussym-metryof order
�.
If � doeshavequadraticterms,asimilarargumentshowsthatit is in thehierarchyof an equationof order2,3,5or 7. A ratherextensive computeralgebracomputationwasusedto show thatif agiven � th orderequationhasa nontrivial symmetry, thenthesymbolicexpressionof its quadraticpart isdivisibleby � ¾ 2� � ¾ � ¾ 2 � ¾ 22 � . Thismeanstheequationis in thehierarchyofsomefifth orderequation.Therestrictionthatthe � -homogeneousequationneedsto havea quadraticor acubicpart,reducesthepossiblevaluesof theweight �låðN to afinite set.Eachcasehasto becheckedseparately. A sys-temof order � � � � � needsto have a symmetryof order �¹� � ��� , respectively.This resultsin thelist of tenequationsin table3.1.
For theoddorderequationsin this list which haveaquadraticpart,onemorething hadto beproven. Thesesystemsarerelative3-primewhile thedivisibility resultsonly show that thereexists infinitely many symmetries
modulo¢ 2 . Oneprovesin this casethat Ê9 � � �$� � ? is divisibleby
198 PETER H. VAN DER KAMP, JAN A. SANDERS AND JAAP TOP
� ¾ ��� ¾ 2 � � ¾ �� ¾ � � � ¾ 2 � ¾ � � if either ¾ �� ¾ 2 or ¾ � ¾ 2 divides � � . It follows
Also noncommutativeequationscanbeconsidered,in which case����� 2shouldbereadasa tensorproduct,that is �����æ� 2 , andonehasto assumethereis no relation between�����A� 2 and � 2 �A��� . We omit the resultsobtainedin this case.They canbefoundin [OS98,OW00].
4. Systemsof equations
In this survey we will only treat the caseof systemsof two evolutionequations.It shouldbeclearfrom this whatthecorrespondingnotionsanddefinitionsin thegeneralcaseof
equationsare.
We take two functions �s�`���� ����� and ) � ) ��� ����� . As before,the � thderivativewith respectto � is denoted�� and ) , respectively. By ��� and ) �onedenotesthederivativewith respectto � . Theequationsconsideredhavetheform � ���<� � ��� � ��� �>=B=>=B��)���) � �B=>=>=v�) ����:��� � �<� �>=>=>=Z��)���) � �>=>=>=f�in which �w� � arefunctionsof � and ) andof finitely many of theirderiva-tives �� ��) � . The maximal � suchthat �� or ) appearsin oneof �w� � iscalledtheorderof theequation,or of �w� � . Thedimensionof theequationis thenumberof functionsinvolved,which is � in our case.
Write 4 5 �ç7:9;� � �<� �>=>=>=Z��)���) � �>=>=>=�? for the ring of polynomialsover 7in infinitely many variables�,�0��� � ��� �>=>=B=B��) � ) � ��) � �B=>=>= . Onefixesa 7 -linearderivation X ' on 4 definedby X ' ���� � � �� &IJ� and X ' � ) � � ) &IJ� . For
INTEGRABLE SYSTEMS AND NUMBER THEORY 199
any pair � C�� ] �WD§4�� 4 thereis aunique7 -linearderivationon 4 , denotedby X  Y(� ^ à , satisfying X  Y(� ^ à ��� � � C and X  Y�� ^ à � ).� � ] . Note that X  Y(� ^ à �}a �is in fact the Frechetderivative of a in the direction � C�� ] � , c.f. [Olv93,5.24],alsocalledtheGateauxderivative([Mag78, A1]). Oneextendssuchaderivationto thering of dualnumbers4 9;+ ? by X  Y�� ^ à � � �°+�� � � X  Y(� ^ à � ��� �+ X  Y(� ^ à ��� � .
With thesenotations,� �k� � � �$� 2 � D*4��°4 is a symmetryof � �� � � �(� 2 �´D§4��l4 preciselywhen� XBg ���_�F+ � � � � � �B���ý�h+ � � � �����h+ X ' � � � �$�>={=|��) �h+ � 2 ��) ���h+ X ' � � 2 �$�>=>=>=f�XBg � ) �F+ � 2 � � � 2 ���ý�h+ � � � ���O�h+ X ' � � � �$�>={=|��) �h+ � 2 ��) ���h+ X ' � � 2 �$�>=>=>=v�j=Completelyanalogousto thescalarcase,onecalculatesthatthis is equiva-lent to thevanishingof aLie bracket 9 �w�(��? on 4��l4 , definedby9 �w�$��?<5 �½� XBg � � � ��nhXBt � � � �j��XBg � � 2 ��nFXBt � � 2 ���$=The symmetriesof � form a sub-Lie-algebraof 4 �»4 . The 7 -linearcombinationsof ����� ��) � � and � � � �(� 2 � arecontainedin this; they arecalledtrivial symmetries.Thesystem� is calledintegrableif thissub-Lie-algebrahasinfinite dimensionover 7 .
4.2. Homogenuity and grading. As in thescalarcase,write¢
for thesubspaceof 4 consistingof all polynomialswith constantterm N . Givenintegers ��� � ���ELAN , the linearsubspace
¢ �Ó� ¶ is by definitionthe 7 -spanofall monomials�� � �U�U���� � ) � � �U�U� ) � � with K ��� =B=>= � K�� � ¬ ��� =>=>= � ¬ ¶ � � .Thisdefinesthreegradingson
¢, and
¢ � ¨ �Ó� ¶ � ¢ �Ó� ¶ .Given real numbers� � � �$� 2 � and , a pair � �w� � � D ¢ � ¢ is called� � � �$� 2 � -homogeneousof weight if all
4.4. Example: a degenerateintegrable system. In this sectionwedemonstratetheuseof thesymbolicmethodandthe implicit functionthe-oremby proving the integrability of somedegeneratesystemconsistingoftheKdV equationcoupledto apurelynon-linearequationwith aparameter.
This systemhasinfinitely many symmetriesfor any 3 , asis shown asfol-lows. Write � (with � odd)for the � th ordersymmetryin 7:9;� � ��� �>=>=>=Z� �� ?of the KdV equation. It follows from [Olv93, 5.31] that every � is theG ' -imageof a uniqueelementin 7:9�� � ��� �>=B=>=B� �� ?¹ï ¢ ; this elementis de-noted G ±��' � � � . As is shown in loc. cit., they satisfytherecursive relation� >I 2 � G 2' � � � �æ�S� � ÿ�ð��� G ±��' � � � . A directcalculationnow showsthatfor everyodd �,L � , thepair
In fact,muchmorecanbeproven: by takingthese� ¹� 3�� and � � 3�� asinput in theimplicit functiontheorem,it wasshown in [vdK02] (usingthesymbolicmethod)that � � 3�� alsohassymmetriesof every even order(atleastfor 3 ¿� � ). As a specialcase,this provesa conjectureof Foursov[Fou00]).
5. Classificationresults
As in the scalarcase,to use the symbolic methodin order to clas-sify integrablehomogeneoussystemsoneneedsdivisibility resultsfor
×-
polynomials. In the presentsituation,these(homogeneous)polynomialsdependon integers ��� á åzN anda vector ¸ �ç�c¸«� �B=>=>=B� ¸ Ø �ÿD 7 Ø IJ� ; theyaredenoted
If � �`� and ¸ � � ã � then C �EDÓ� � 2 � n 2� ���à�GFH? �JIFK@ � ��� �LIFH? ��FK@ � inwhich Fe�NM 2$O PDÓ� .
In all othercasesC # � is irr educiblein 7:9 � � ? � @ ? .Theideaof theproof is to show thatthenumberof singularitieson the
curvegivenby C # � -� N is too smallfor thecurve to bereducible.Thiswasactuallycarriedout by Frits Beukers.
Otherprogresscanbemadeundertheassumptionthat theorderof theequationis two. This is carriedout in [SW01]. With helpof Taylor’s ex-pansionit is possibleto show thefollowing.
THEOREM 5.3. Suppose � " DRQ � �c7 � havethepropertythatthereexistsmorethanvalue á DS6 such that
× �2 9~¸ ? divides× �Ø 9 " ? . Thenall such á have
thesameparity andweare in oneof thefollowingcases:
With thesetwo theoremsandtheimplicit functiontheoremwecandrawthefollowing conclusions:ú
Undertheusualassumptions,if the �VUXW -ordersystemwithoutqua-dratictermsis integrable,thenits only eigenvalueis n � andit hasarbitraryordersymmetries.úThisanalysisis alsousefulin dimension
å � . However, only for �H� it givesa completeanswer:we only have two eigenvalues,soeither ¸q� � ¸ 2 or ¸u� is equalto ¸«� , or they areall differentfrom ¸«�andthereforeequalto oneanother.úAlthough we only considerthe integrability problem,the resultsequallyapply to almostintegrablesystems(that is, systemswithonly afinite numberof nontrivial generalizedsymmetries).
Thereis still a lot to bedonehere.Themutualdivisibility of thepolynomi-als × � 9 ) ? �8¸����e�4?y�!@ � n " � n ÙY? n @
INTEGRABLE SYSTEMS AND NUMBER THEORY 203
(with ) ���c¸ � " � Ù � � ) is well understoodfor" �0Ùµ� andwe know when
thequadraticoneappearsasa factorof anotherone,but thatis it. Progressherewould have immediateimplicationsin theclassificationtheoryof sys-temsof evolutionequationswith respectto theexistenceof symmetries.
5.3. Quadratic terms. Weproceedasin theprevioussubsection.Firstwe assumethat our two dimensionalsystemhasnonzeropart � ±�� � � or� � � ±�� . Numbertheoreticalmethodsand‘experimentalmathematics’willleadus to many integrablesystemsat any order. Moreover we will finda hugesetof almostintegrablesystems,that is to say systemswith a fi-nite numberof symmetries.It wasobserved andconjectured,cf. [Fok80,I S80, Fok87], that theexistenceof one(or a few) symmetriesimplies theexistenceof infinitely many symmetries.This turnsout to bewrong.Coun-terexampleswerefound in [Bak91, vdKS99]. A (� -adic)methodto provethat the numberof symmetriesis finite hasbeendeveloped,cf. [BSW98,vdKS01]. We will first concentrateon integrablesystems.After this, weindicatehow the � -adicmethodswork. Finally, we fix theorderandshowhow to classifythegeneralsystemof ordertwo usingnumbertheory.
5.3.1. Integrable Z -systems.Supposethat � ±�� � � is non-zero.Thesys-temcontainsthefollowing subsystemwhich wewill analyzeon its own.Z �9�¸q� � ¸ 2 ? � �s�r5 � ����� ¸q���� :� � � ) � ��) � �>=>=>=f�) �Å� ¸ 2 ) whereq� � ¸ 2 D 7 and � is aquadraticpolynomialin ) � ��) � ��) 2 �>=>=>= . Wecallthis a Z –system.The (only) conditionfor Z Ø 9 " � � " 2 ? � �i� to bea symmetryof Z �9~¸q� � ¸ 2 ? � �s� reads� �9�¸q� � ¸ 2 ? Æ� � � Ø 9 " � � " 2 ? Æ���with the � –functions� �9�¸q� � ¸ 2 ? � ¾ � ��¾ 2 � �8¸q�B� ¾ ��� ¾ 2 � n ¸ 2 � ¾ � � ¾ 2 �j=If � Ø 9 " � � " 2 ? Æ� is divisible by � ¹9~¸q� � ¸ 2 ? we have a symmetricpolynomialexpressionfor Æ� which canbetransformedback.Becausethe ¾ � -degreeofÆ� is assumedto be smallerthan � , the function � ¹9~¸q� � ¸ 2 ? cannotdivideÆ� . Therefore� �9�¸q� � ¸ 2 ? shouldhave a commonfactorwith � Ø 9 " � � " 2 ? incasea nontrivial symmetry(with eigenvalues
THEOREM 5.5(Lech,Mahler). Let ��� � � 2 �>=>=>=Z� ��ê � Ù>� � Ù 2 �>=B=>=B� Ù$ê D 7ba N .Supposethatnoneof theratios �� ã �T� with K ¿� ¬ is a rootof unity. Thentheequality ÙU�Ó� Ø � �EÙ 2 � Ø 2 � =B=>= �EÙ$ê$� Øê � Nholdsfor at mostfinitelymanyintegers á .
In [BSW98] it is shown thatasa consequenceof this, theonly factorsof � –functions(with a nonzero� ) which appearin infinitely many otherones� Ø , havezeroesin a setof theform £ Nq�>n � ����� �� � I�T� � c� ¦ . Thefollowinglist of all integrableZ –systemswith quadraticpart ) 2� givenin [BSW01] isobtainedusingthis. It alsousedanalgorithmof C.J.Smyth(cf [BS01]) thatsolvespolynomialequationsfor rootsof unity. For eachsystemin the list,
INTEGRABLE SYSTEMS AND NUMBER THEORY 205
all � suchthata (nontrivial) symmetryZ exists,aregiven.
We now presenta moredirectmethodthantheonein [BSW01]. Thismakes it possibleto treathigher orders. Expressing ¹� ��� I�S� in termsof��� � c� and ?z� �cI ��cI c� yields an equationthat can be solved for roots ofunity. As an example, this was carriedout for � ß � � , and the values� correspondingto solutions � � ? were plotted. Becausethe set of rootsis invariant under � �� �� and � �� I� , the upperhalf unit disc is takenasa fundamentaldomain. Inspectingthe patternsformedby the values �obtainedin this way, canbedescribedasexperimentalmathematics.
To explain the resultswhich wereat first found experimentallyin thisway, notethat any �,D 7la � canbedescribedby fixing two unit vectorsm �(x in theupperhalf planeandsayingthat � is theintersectionof thelines¸ m and n � � " x .
THEOREM 5.6. Let� (¤�`Dn6 . Let
m �$x be � � th rootsof unity. Leto ��£V@ Dqp�¥ @ ¿��I@ �b¥ @ ¥ ¿� � ¦ . To theintersectionpoint �ÝD o of thelines¸ m and n � � " x , therecorrespondsan integrable Z -system.Anyintegrable Z -systemis a symmetryof such a system.
Theproof of thefirst statementis simple,it follows from substituting�in
`Ø � ��� I�b� . Theratioof eigenvaluesof theintegrableZ -systemis givenby� � � � � ã � � � �b� andtheorderof thesymmetriesis a multiple of � . The
secondstatementfollows from theLech-Mahlertheorem.Thenumberof integrablesystemsof this form canbecalculatedandit
canbeverifiedwhethersuchasystemis in a lowerhierarchy.
THEOREM 5.7. Let r � £V@ D o ¥ @à�sI@ ¿� n � �b¥ @à� �2 ¥ ¿� �2 ¦ . TheZ -systemsdescribedin theorem5.6thatcorrespondto �_DRr havenoothersymmetries.
206 PETER H. VAN DER KAMP, JAN A. SANDERS AND JAAP TOP
FIGURE 2. Thezeroesof � –polynomialsof integrablesys-temswith order � N inside the unit disc as intersectionsofstars.
The ideaof theproof is, to write
`Ø � ��� I�b� in termsof rootsof unity
mand x andto performthetransformation
x 2 � t � m 2 �vuwt =This leadsto thediophantineequation
(5.1)
, � n u� n t 0 Ø �ù� n u Ø� n t Ø
INTEGRABLE SYSTEMS AND NUMBER THEORY 207
Frits Beukersshowed thatwhen á å � , this equationhasno solutionsinrootsof unity u � t with u � t ¿� T � and u ¿��t � It . Applying the inversetransformationto u � tà� n � resultsin x � TyK m �(x � T � .
Onemayobserve that the remainingintegrable Z -systemsdo not haveothersymmetriesaswell. To show this, oneprovesthat the diophantineequation � � ��u � �Q� S±�� � � ��u �hasno solutionswith ��å � and u a root of unity ¿� T � . This followsby comparinga � -adicvaluationof thetwo sides.This observation in factcompletestheclassificationof integrableZ -systems.
EXAMPLE 5.8. Take � � � . Theline 3 M ²é O n � intersectstheimaginaryaxis in thepoint � �yx � K . This is a zeroof � � 9 � � �Bn � � ? . Thepolynomialdividing all � –functionsof thecorrespondingsymmetriesis
5.3.2. Almostintegrable Z -systems.Many Z -systemshaveonly finite-ly many independentsymmetries.An efficient methodfor computingallZ -systemsof a given orderwith a symmetryof someotherfixed order istheuseof resultants.
We fix integers � ¿� á andcalculateall ��� � (with � ¿�{� � �¶ ) suchthat` �� �3� � � � ` Ø � ��� � � � N . In the following we disregardthe trivial factorsof
` whichare � �ªn � � � � � n � � for all � andalso � � � � � �E�p� � � for odd � .
LEMMA 5.3.2. Take �zå � = To obtain all eigenvaluesof � th order Z -systemswith a symmetryof order á ¿� � onecalculatesthe resultantof` �� �3� � � and
`Ø � ��� � � with respectto � andappliesthemap � �� �cI � â �cI � à â to
208 PETER H. VAN DER KAMP, JAN A. SANDERS AND JAAP TOP
orderfactor ô . This implies that the � th order Z –systemwith eigenvalues¸q�i� � � � � ¸ 2 �A� � � �b� andquadraticpartcorrespondingto � �9�¸q� � ¸ 2 ? ã ôhasasymmetryof order á . <
We will now discuss� -adictechniquesto answerthequestion,whethera particularsystemhasmorethanoneindependentsymmetry, andif so,ofwhat order. In fact, the methodwill enableus to show that a systemhasonly finitely many generalizedsymmetries.
LEMMA 5.3.3 (Hensel). A polynomial CÉD¤ä Ëb9;Ú ? has a zero in ä Ë ,providedthe following holds. There existsan 3 � Dðä Ë such that C � 3 � �!þN!ö òT÷ � and
� Y� � � 3 � � ¿þQNyö òT÷ � .The (standard)proof is to constructa Cauchysequencein ä Ë using
Newton iteration,startingfrom 3 � .Themethodof Skolem. Let ¬ beapositiveinteger. Given� -adicintegersÙ$ and� -adicunits �� for �yß K ß ¬ , oneputs
� Ø � �P &ÐJ� Ù$ ~? Ø � where?3 Dwä Ë is definedby � � � ?3 ��*� Ë>±�� . For example,with Ù$ ��½� n � � and ¬ � � and ���µ� � � � , � 2 � � �l� , ���!�L�:� � � and ��¼y� � � � � wehave
` �� ��� � � �®� � .LEMMA 5.3.4(Skolem). If � �ê ¿þzN mod� then �J� � �ê(I �  Ë>±��cà ¿� N .PROOF. Notethat � �ê�I �  Ë>±��cà �8o � &ÐJ� Ù( � ê � � �4?S ��� � þ � �ê mod �<= <LEMMA 5.3.5(Skolem). If �Få � and � �ê � N and � �ê ¿þÉN mod � then�J�_åæN wehave� �ê(I �  ËU±��cà ¿� N .PROOF. Assume� �ê(I �  Ë>±��cà � N , then
210 PETER H. VAN DER KAMP, JAN A. SANDERS AND JAAP TOP
To apply theselemmasin our situation,recall that we arehave a pair� ¿� á anda solution �3� � to thesystem Ø � ��� � � � ` ¹� ��� � � � N . We wantto find conditionson otherintegers � suchthat
` ê�� ��� � � � N . For this, onesearchesa prime number� suchthat the involvedroots � and � arein ä Ë ,andoneconsidersthecorresponding� �ê (with ¬ � � ). Onenow checkstheconditionsin thelemmasfor all ��(�� n � .
EXAMPLE 5.10. Here is how to apply the methodof Skolem to theBakirov system.Recall that the resultantof
Theexpression� �Ø �`Ø � �3� � � vanishesmodulo � when á D£ N.� � �*�¹� � � �e� � � Nq� ��� � � � ¦ andis non-zerofor other á ( �bÞ . This
in factimpliesthatthereis noothersymmetryof orderlessthan �bÞ .With thefirst lemmaof Skolemoneconcludesthatany symmetryhasorder þ á ö ò.÷ � � with á D £ Nq� � �*�¹� � � �e� � � N.� ��� � � � ¦ .
However, � �Ø ¿þ`Nyö òT÷ � 2 when á D £S�e� � � Nq� ��� � � � ¦ . There-forewecannotapplythesecondlemmain thesecases.úIn ä � z � wefind thezeroes� � ��� � � � � � =>=>= and �ÿ� � � ��w�e� � �=B=>= . The correspondingexpression� �Ø � ` Ø � �3� � � for N ß á (�>Þ N satisfies� �Ø þQNyö òT÷ � only when á D £ Nq� � �*�¹� � ¦ . However,for theseá onefindsthat � �Ø is nonzeromodulo � . Both lemmasof Skolemcanbeappliedandit follows thatthereis no nontrivialsymmetryexceptat order
C � �b� � � � z �»��� � j � � � � � � � � Nb� � � ��� � � ¼ � � � Nb� � � � � � 2 �»�b� � � � =With � � � � we find C � �U� �sþ C � � ���,þ N mod � , both correspondingto � -adic zeroes. With N ßká ( ��� one finds
`Ø � �U� � � ��� � N only ifá D £ Nq� � �*�q� ��� ¦ . For thesevaluesof á the associated� �Ø is not zero
modulo � . The Skolem lemmasimply that the only non-trivial symmetryhasorder11.
The fact that the degreeof C is Þ indicatesthat therearetwo different� th ordersystemswith a symmetryof order �b� . Theargumentgivensofar,shows the lack of othersymmetriesfor only oneof them. To prove it fortheothersystem,it sufficesto show that C is irreducibleover | . This is thecase,asfollows,e.g.,from thefactthat C � �b� is irreduciblemodulo
Example: � � � � á � � � . ThepolynomialC � �b� � � � 2 � ��� ��� � � Nb� ��� � � � �V� �»� Þ � z � � ��� j� � �S� � � � �b� � �»� Þ � ¼ � � � � � � � Nb� 2� ��� � �splits into distinct linear factorsin ä ��� � 9 ��? . It is irreducibleover | . Mod-ulo � � � N � , the pair �c� � �¹�*��NH��� is a zeroof
`Ø � ��� � ��ö ò.÷ � when á D£ Nq� � � � � � � � � � � �*� � �q¦ . The pair �c� � �.� � N�N«� is a zero of
`Ø � ��� � ��ö òT÷ �
when á D £ N.� � � � � � � � � � � � � � �.¦ . Usingbothpairswe canapplySkolem’sfirst lemmafor all N ß·á ( � N Þ except £ Nq� � � � � � � ¦ , and for thesere-mainingvalueswe could apply the secondlemma. This methodis quite
INTEGRABLE SYSTEMS AND NUMBER THEORY 213
successfulhere,sincewe couldnot find any prime( ( Þq� � � ) for which thenormalprocedureworks.
With theseimprovementsof the � -adic methodwe have beenable toprove
THEOREM 5.11. Take� (®�7( ��� ����( á (®� � ���.� and á ¿� ��� � �f�
when � � � . Thenall � th order non-integrable Z -systemswith a symmetryof order á arealmostintegrableof depth � .
Counterexamplesto Fokasconjecture. Theexceptionalcasein theorem5.11disprovesa conjecturemadein [Fok87], whereFokassuggestedthatif a scalarequationpossessesat leastonetime-independentnon-Lie pointsymmetry, then it possessesinfinitely many. Similarly for n-componentequationsoneneeds� symmetries.
Counterexample[vdKS99, vdKS01]: take � � �.� á � ��� � �f� . Theresultantof
` j�� ��� � � and
` ���B� �3� � � with respectto � aswell astheresultantof
¥ �� 3 � � � �c¸ n � � ¥ � �implies ¥�3:¥ � ¥�3 � � ¥ , i.e., �>3 � n �2 . Togethertheseimply 3 � n �2b� 2 .Then ¸à� n � T � K . Since ¸ is invariantunder 3 �� �� thesecondpair givesthesamevaluesfor ¸ . Wedefine� � 3\� á � � � ¸!� �� � Ø n � ¸ n �� � Ø n 3 Ø � �� 3 � � � Øwhere3 � n �2 � 2 . Its valueonly dependson ¸ since
� � �� � á � � � � 3\� á � .Noticethat
� � 3�� á � � N if andonly if
N � � �� � Ø n � n �� � Ø n �q� x �� � Ø � ò�� á��� =Solvingthis,weobtain á þ � ö ò.÷ �or á � � =It follows that
" � K Ø � � � nFK�� Ø when¸e� n � T � K�=Following similar reasoningall possibleeigenvaluesandordersof possi-ble symmetriesareobtainedfor all possiblecombinations� � �$� 2 . Finallythe implicit function theoremis usedto prove integrability of the systemsinvolved.
6. Conclusion
Theapplicationof numbertheoryin theanalysisof integrablesystemsis quite successfulandpromising. It is anotherunexpectedapplicationofpuremathematicsandit illustratesthe needof communicationamongthedifferentbranchesof mathematicsandmathematicalphysics.
Bibliography
[Bak91] I. M. Bakirov. On the symmetriesof somesystemof evolution equations.Technicalreport, Akad. Nauk SSSRUral. Otdel. Bashkir. Nauchn.Tsentr,Ufa,1991.
[Beu97] Frits Beukers. On a sequenceof polynomials. J. Pure Appl. Algebra,117/118:97–103,1997.
[BS01] FritsBeukersandChrisJ.Smyth.Cyclotomicpointsoncurves.In MillennialConferenceon NumberTheory, May 21-26,2000,Urbana-Champaign. AKPeters,2001.
[BSW98] Frits Beukers,JanA. Sanders,andJingPingWang.Onesymmetrydoesnotimply integrability. J. DifferentialEquations, 146(1):251–260,1998.
[Fok80] A. S.Fokas.A symmetryapproachto exactlysolvableevolutionequations.J.Math.Phys., 21(6):1318–1325,1980.
[Fok87] A. S. Fokas.Symmetriesandintegrability. Studiesin AppliedMathematics,77:253–299,1987.
[Fou00] Mikhail V. Foursov. On integrablecoupledKdV-type systems.InverseProb-lems, 16(1):259–274,2000.
[Gar71] Clifford S. Gardner. Korteweg-de Vries equationand generalizations.IV.The Korteweg-deVries equationas a Hamiltoniansystem.J. Math. Phys.,12:1548–1551,1971.
[GGKM74] Clifford S. Gardner, John M. Greene,Martin D.Kruskal, and Robert M.Miura. Korteweg-deVriesequationandgeneralization.VI. Methodsfor ex-actsolution.Comm.PureAppl.Math., 27:97–133,1974.
[Ger93] Vladimir P. Gerdt.Computeralgebra,symmetryanalysisandintegrability ofnonlinearevolution equations.Internat. J. Modern Phys.C, 4(2):279–286,1993.
[Ger96] V. P. Gerdt. Homogeneityof integrability conditionsfor multi-parametricfamiliesof polynomial–non-linearevolutionequations.Math.Comput.Simu-lation, 42(4-6):399–408,1996.Symboliccomputation,new trendsanddevel-opments(Lille, 1993).
[GD75] I.M. Gel’fand and L.A. Diki ı. Asymptotic propertiesof the resolvent ofSturm-Liouvilleequations,andthealgebraof Korteweg-deVries equations.RussianMath.Surveys, 30 (1975),no.5, 77–113.
[GKZ90] V. P. Gerdt,N. A. Kostov, andA. Yu. Zharkov. Nonlinearevolutionequationsandsolvingalgebraicsystems:the importanceof computeralgebra.In Soli-tonsandapplications(Dubna,1989), pages120–128.World Sci.Publishing,RiverEdge,NJ,1990.
[I S80] N. H. Ibragimov andA. B. Sabat.Evolution equationswith a nontrivial Lie-Backlundgroup.Funktsional.Anal. i Prilozhen., 14(1):25–36,96,1980.
[KdV95] D. J.Korteweg andG.deVries.Onthechangeof form of longwavesadvanc-ing in a rectangularcanal,anda new type of long stationarywaves.Philos.Mag., 39(5):422–443,1895.
[KMGZ70] Martin D. Kruskal, RobertM. Miura, Clif ford S. Gardner, and NormanJ.Zabusky. Korteweg-de Vries equationand generalizations.V. Uniquenessandnonexistenceof polynomial conservation laws. J. MathematicalPhys.,11:952–960,1970.
[Lec53] C. Lech.A noteon recurringsequences.Arkiv. Mat., 2:417–421,1953.[Mag78] F. Magri. A simplemodelof integrableHamiltonianequation.J. Math.Phys.,
19(5):1156–1162,1978.[MGK68] RobertM. Miura, Clif ford S. Gardner, andMartin D. Kruskal.Korteweg-de
[MSS91] A. V. Mikhaılov, A. B. Sabat,andV. V. Sokolov. Thesymmetryapproachtoclassificationof integrableequations.In Whatis integrability?, SpringerSer.NonlinearDynamics,pages115–184.Springer-Verlag,Berlin, 1991.
[Olv93] PeterJ. Olver. Applicationsof Lie groupsto differential equations, volume107 of GraduateTexts in Mathematics. Springer-Verlag,New York, secondedition,1993.
[OS98] PeterJ. Olver and Vladimir V. Sokolov. Integrableevolution equationsonassociativealgebras.Comm.Math.Phys., 193(2):245–268,1998.
[OW00] Peter J. Olver and Jing Ping Wang. Classification of integrable one-componentsystemson associative algebras.Proc. LondonMath. Soc.(3),81(3):566–586,2000.
[SG69] C. H. SuandC. S.Gardner. Korteweg-deVries equationandgeneralizations.III. Derivation of the Korteweg-deVries equationandBurgersequation.J.Math.Phys., 10:536–539,1969.
[SW98] JanA. Sandersand Jing Ping Wang.On the integrability of homogeneousscalarevolutionequations.J. DifferentialEquations, 147(2):410–434,1998.
[SW00] JanA. SandersandJingPingWang.On the integrability of non-polynomialscalarevolutionequations.J. DifferentialEquations, 166(1):132–150,2000.
[SW01] J.A. SandersandJ.P. Wang.On the integrability of systemsof secondorderevolution equationswith two components.TechnicalReportWS–557,VrijeUniversiteitAmsterdam,Amsterdam,2001.Submittedto Journalof Differ-entialEquations.