Rigid Two-Step Nilpotent Lie Groups Relative to Multicontact Structures
IRENEVENTURI
ABSTRACT- LetfXi;Ykgdenote a fixed basis of the Lie algebranof aconnected and simply connected nilpotent Lie group N of step two. Under a technical as- sumption onfXi;Ykg, we prove that the Lie algebraMof vector fieldsV onN that satisfy [V;Xi]liXiis finite dimensional, a property that we refer to as rigidity. Our proof allows the explicit count ofdimM.
1. Introduction.
In a broad sense, the study of rigidity of stratified niplotent Lie groups has attractedmuchattentioninrecentyears,startingfromtheseminalworkbyN.
Tanaka ([19] and references therein) and his school (notably Yamaguchi [20], Yatsui[21]and Morimoto[15])andthen moving intomoreexplicitlygeometric settings in the important papers by KoraÂnyi and Reimann [11] and Pansu [17], where various generalised contact-type structures were considered. At a more general and formal level, the notion of multicontact structure was first formulated by Cowling, De Mari, KoraÂnyi and Reimann [4, 5] in the context of the ``boundaries'' G=P of symmetric spaces (G semisimple, P aparabolic subgroup ofG). In particular, the local nature ofG=Pas a nilpotent stratified Lie group stemming from the Iwasawa decomposition ofGKANand from the Bruhat decomposition, together with the root structure of the Lie algebra ofN, leads to a notion of multicontact vector field that can easily be formulated for all stratified nilpotent Lie algebras once an additional sub-stratification of theground layer is singled out. Inthe caseofanIwasawa nilpotentLie algebra nin its standard form as the sum of the (restricted) positive root spaces, the stratification is provided by the height function on the root system, whereas
(*) Indirizzo dell'A.: DIMA, UniversitaÁ di Genova, via Dodecaneso, 35, 16146 Genova.
E-mail: venturi@dima.unige.it
the sub-stratification refers to the simple roots. As explained by KoraÂnyi [10], the problem has a clear differential geometric counterpart: if one selects smooth distributionsDion asmooth manifoldMsuch that the vector fields spanning the distributions satisfy a HoÈrmander type condition, then a mul- ticontact mapping ofMis adiffeomorphism whose differential preserves each distributionDi. Passing to the infinitesimal level yields the notion of multi- contact vector field, that is, a vector field V onM whose flow consists of multicontact mappings. IfVis such avector field, the requirement amounts to the condition that [V;X]2 Difor everyX2 Di. Thus one may formulate a general problem as follows: given a stratified nilpotent Lie algebra nn1 ns (parametrizing the tangent bundle of the connected and simply connected nilpotent Lie groupN with Lie(N)n) together with a multicontactstructure, that is, a decomposition ofn1intothesumofsubspaces n1n(1)1 n(t)1, is the Lie algebraMof vector fields onNwhich pre- serve (under bracket) each direct summandn(i)1 finite dimensional? If yes, we say thatnis rigid and in this case it is natural to ask if we can describe the structure ofM. The remarkable answer in [20] and [4, 5] is that in most cases M 'Lie(G) ifMG=Pand if eachn(1j)is determined bythe structure ofPin terms of simple roots. These questions are formulated by Tanaka and his school in terms of prolongation Lie algebras and rather deep results are presented in [19, 20, 21]. Other very interesting results along these lines have been obtained by Reimann onH-type groups [18], by Warhurst on filiform groups [22], jet spaces [23] and free algebras [24], and by Ottazzi on Hes- senberg manifolds [16].
In this paper we address the following problem: ifnhas step two and we are given certain canonical basesfXigofn1 andfYkgofn2, isnrigid re- lative to the ``finest'' multicontact structure, that is, when eachn(i)1 is just the span of the basis vectorXi? Our main theorem answers this question positively. Its proof is based on a method suggested by Eastwood [14] and combines prolongation notions together with genuine differential geo- metric constructs and provides a method for bounding explicitly the di- mension ofMonce the structure ofnis known.
2. The multicontact system.
2.1 ±Choice of bases.
Let n denote a two-step stratified nilpotent Lie algebra. This means thatnn1n2as a direct sum of vector spaces, wheren1is asubspace of
dimension n which generates the m-dimensional subspace n2 under bracket, that is
[n1;n1]n2:
We select basesfX1;. . .;XngandfY1;. . .;Ymgofn1andn2, respectively, and we assume that (1)
(a)z(n)n2, that is,z(n)\n1 f0g, wherez(n) is the center ofn;
(b) for everyk2mthere existi;j2nsuch thatYk[Xi;Xj]:
(c) for everyi2nthere existj2nandk2msuch that [Xi;Xj]gkijYk, wheregkijis anonzero real constant.
Clearly, assumption (a) does not harm generality, for otherwisenwould split as a direct sum of an abelian Lie algebra and a nilpotent Lie algebra that does satisfy (a). Further, we can always select bases such that (b) is satisfied. Indeed, once we choose a basisfX1;. . .;Xngofn1, the brackets [Xi;Xj] generaten2 by assumption, so we can select a basis ofn2 among them and denote its elements byfY1;. . .;Ymg. As for (c), we donot know if all two-step nilpotent Lie algebras satisfy it or not.
Observe that, by assumption (b), which is not restrictive, for any fixed k2m there exist distinct indices i;j2n such that [Xi;Xj]Yk. Hence there cannot be more directions inn2than distinct pairs (i;j)2nn. This implies thatmn(n 1)=2.
2.1.1 - Examples.
We list below two classes of two-step nilpotent Lie algebras that satisfy our hypotheses. We remark that in all these cases rigidity in our sense is essentially known (see [18] when the center has dimension at least three, and [4, 16]). Our result can thus be seen also as a unified approach.
H-type algebras.For the notation and basic properties of the Heisen- berg type algebras, introduced first in [6], we refer to [3]. Following standard notation, an H-type algebra is a direct sumvz, wherevn1
and zn2, that is orthogonal with respect to a positive definite inner producth;iand for which a mapJ:z!End (v) is defined that satisfies hJYX;X0i hY;[X;X0]i for all Y 2z and all X;X02v. Evidently, JY is skew-symmetric with respect to h;i. The defining condition for such an algebra to be called of Heisenberg type is thatJY2 jYj2I(soJY is non-
(1) Ifnis apositive integer, we writen f1;. . .;ng.
singular for everyY). Among the several properties ofJ, we have [X;JYX] jXj2Y
for everyX2vand everyY2z.
Now, for afixedY60 withjYj 1, the endomorphismJYofvis skew- symmetric and non-singular. Hencevis even dimensional and we can se- lect abasisX1;. . .;X2prelative to which
JY 0 Ip
Ip 0
: We may obviously normalizejXij 1 and obtain
Y [Xi;JYXi] [Xi;Xip] if ip [Xi;Xi p] if i>p;
so that settingYY1, and completing to a basis ofzby choosing vectors Y2;. . .;Ymamong the various brackets [Xi;Xj], yields a basis that satisfies (b). Indeed, for everyithere exists eitherjip(ifip) orji p(if i>p) a ndk1 such that [Xi;Xj] Y1whereg1ij 1.
Hessenberg algebras.This is a class of nilpotent algebras that includes algebras of arbitrary step. They were introduced first in [12], [13] and then studied in [16] from the point of view of rigidity, where it is shown that most of them are rigid, in the multicontact sense that we are interested in.
Given asimple Lie algebragwith Iwasawa decompositiongkan and with root systemSone selects asubsetR Sof positive roots which satisfy the following Hessenberg-type condition: ifb2 Randa2S sa- tisfy b a2S, then b a2 R. The most important property of Hes- senberg type sets is clarified by the next proposition, which is essentially contained in all of the aforementioned references.
PROPOSITION2.1. A subsetR Sis of Hessenberg type if and only if the vectorspacenR0 P
g2R0ggis a (nilpotent) ideal ofnwhereR0Sn R.
PROOF. Suppose R S is of Hessenberg type and let g2 R0 and b2 R. Ifbg2S, then necessarilybg2 R0, for otherwisebg2 R and this implies thatg2 R, contrary to the assumption. ThereforenR0is an ideal inn, obviously nilpotent.
Conversely, ifnR0 is an ideal inn, then it is nilpotent and the following property holds: if g2 R0 and b2 R are such that bg2S, then ne- cessarilybg2 R0. This means that ifbg2 Rfor some positive rootsg andb, then bothbandgmust belong toRfor otherwise at least one of them is inR0and, by assumption, so is their sum. p
We stratifynby height, that is,niis the sum of all root spaces of height equal toi, where the height refers to the number of simple roots that occur in the (unique) decomposition of apositive root as asum of simple roots.
Evidently, ifR is of Hessenberg type, thennR0 is a stratified ideal and therefore the quotient nRn=nR0 is again a stratified nilpotent Lie al- gebra, which we shall refer to as a Hessenberg algebra. Among the more interesting sets Rof Hessenberg type are those that include all positive roots of height smaller than or equal to a positive integerp. If we denote by Rpsuch aset and letn(p)nRp, the sum of the root spaces with roots inRp then it is clear thatn(p)has step exactly equal top.
For example, if gsl(n;R), thenR R2 may be identified with the set of indices that correspond to the first two superdiagonals of annn upper triangular matrix, i.e.f(i;j):1j i2g. ThusnR0consists of the upper triangular matrices whose nonzero entries can only occur in the upper-right corner whose entriesaijsatisfyj i>2, and thereforen(2)is the quotient of the (strictly) upper triangularnnmatricesnmodulonR0. It is natural to identify its first stratumn(2)1 with the first superdiagonal and its second stratumn(2)2 with the second superdiagonal.
Obviously,fXiEi;i1:1in 1g andfYkEk;k2:1kn 2g satisfy both (b) and (c), because
[Xi;Xj] Yi if ji1 0 if j6i1
[Xi;Yk]0 mod nR0 [Y`;Yk]0 mod nR0: (1)
We remark that the dimension ofn(2)1 is exactlyn 1, which can thus be any positive integer.
2.2 ±Multicontact vectorfields.
A vector fieldVon the connected simply connected stratified groupN with Lie algebra denoted by n is said to be a multicontact vector field relative to the basisfX1;. . .;Xng, if it satisfies
[V;Xi]liXi (2)
for alli2nand for some smooth functionsl1;. . .;lnonN. The vector field V is not assumed to be left invariant, so we write
VX
t2n
ftXtX
k2m
gkYk (3)
where theftand thegkare smooth functions onN. Here and in the following we use subscripts to indicate components along the first stratum and su- perscripts to indicate components along the second stratum.
In preparation for what follows we recall some basic notation used for computing Lie brackets. A bracket of elements inn1is of course inn2. For i;j2nwe write
[Xi;Xj]X
k2m
akijYk; (4)
where theakij are real constants. Evidently,akij akji andakij0 for all k2m if and only if [Xi;Xj]0. Moreover, if [Xi;Xj]gkijYk as in as- sumption (c), we haveakijgkijanda`ij0 for all`6k. Finally, dijstands for the Kronecker symbol.
LEMMA2.2. The vectorfield V 2X(N)as in(3)is a multicontact vector field on N if and only if forall i;t2n and all k2m
Xi(gk)X
t2n
aktift
(5)
Xi(ft) ditlt: (6)
PROOF. Fori2nwe have liXiX
t2n
ftXtX
k2m
gkYk;Xi
X
t2n
ft[Xt;Xi] Xi(ft)Xt X
k2m
Xi(gk)Yk
X
t2n
X
k2m
aktiftYk Xi(ft)Xt
X
k2m
Xi(gk)Yk:
Equating the coefficients ofXiandYkwe get (5) and (6). p
2.3 ±Rigidity.
In this section we prove the following
THEOREM2.3. Underthe assumption (c), the Lie algebraMof mul- ticontact vectorfields on a two-step stratified nilpotent connected and simply connected Lie group is finite dimensional.
2.3.1 - Prolongation.
The procedure of prolongation has a long history. It may be traced back to the work of Elie Cartan and was developed by many people in several ways (see e.g. [1, 7, 8, 9, 15, 19, 20] and references therein). A formulation of prolongation in the language of exterior differential systems may be found for example in [2], while from the algebraic point of view the notion was formalized by Tanaka [19]. As explained in [1] the main idea is that ``a generic overdetermined system of partial differential equations may be rewritten as a first order ``closed system'' in which all first partial deri- vatives of the dependent variables are expressed in terms of the variables themselves. To do this, we must introduce extra dependent variables for unknown derivatives until all derivatives of the original and extra vari- ables can be determined as consequences of the original equation''. In this sense, a system may or may not ``close up''. We prove in this section that our system does indeed ``close up''. This can also be expressed by saying that our system admits a finite prolongation.
The linear nature of the equations that express the various ``jet'' variables as functions of the others allows us to prove rigidity as follows.
As usual we first define a vector bundle whose fibers have as many dimensions as the number of variables that have been introduced. Then we introduce a connection on this bundle in such a way that finding a solution to the original system is equivalent to finding a covariant con- stant section of the bundle. Since the space of parallel sections cannot have dimension exceeding the dimension of the bundle, we infer the finite dimensionality of the space of solutions. Furthermore, equality is achieved in the flat case and we show that our connection is indeed flat, thereby obtaining precise dimension counts. Our approach is inspired by that of Eastwood [14].
In this section we focus on the heart of our proof, showing that taking sufficiently many derivatives yields a closed system. Since the computa- tions are slightly involved, it is perhaps helpful to keep in mind the the following picture:
This means that we consider the real valued functions onNlabeledft,gk andlt, witht2nandk2m. We then take derivatives with respect toXi
andYkof the functionsft,gkandltand produce new variablesakt andbkt. All Y-derivatives ofakt and allX-derivatives ofbkt are zero for all choices of the indices if the functionsft,gk are the coefficients of a multicontact vector field as in (3) and if thelt satisfy (2). Hence the system associated to our multicontact problem ``closes up'', that is, the prolongation is finite. The extra variables are defined by
akt Yk(ft); t2n; k2m;
bkt Yk(lt); t2n; k2m:
For computational reasons, we also introduce the variables c`kt Y`(akt); t2n; `;k2m;
although in fact they all vanish in the multicontact case (see Lemma 2.16).
From now on we shall use the symbolgkijto mean nonzero real constants that satisfy assumption (c) without appealing explicitly to it. Evidently, we assume (5) and (6).
The following lemmadescribes Yk(g`) for all possible pairs (k; `)2 2mm.
LEMMA2.4. Forany k; `2m and all i;j2n forwhich[Xi;Xj]gkijYk
we have
Yk(g`) dk`(lilj):
(7)
PROOF. Fixk2mand takei;j2nsuch that [Xi;Xj]gkijYk. Then for any`2m, by (5) and (6)
gkijYk(g`)[Xi;Xj](g`)
Xi(Xj(g`)) Xj(Xi(g`))
XiX
t2n
a`tjft
XjX
r2n
a`rifr
X
t2n
a`tjditltX
r2n
a`ridjrlr
a`ijlia`jilj:
Now, [Xi;Xj]gkijYk implies that a`ij0 if `6k, whereas if `k, then
akij akjigkij60. p
COROLLARY2.5. Letk2mandi; j2nbe such that[Xi; Xj]gkijYk. (i) If there exist p;q2n such thatakpq60, thenliljlplq; (ii) if there exists p2n with p6j such thatakpi60, thenljlp. PROOF. By (4) and (7)
[Xp;Xq](gk)X
`2m
a`pqY`(gk)
akpqYk(gk) X
`2m;`6k
a`pqY`(gk)
akpq(lilj):
But by (5) and (6)
[Xp;Xq](gk)Xp
X
t2n
aktqft
Xq
X
r2n
akrpfr
X
t2n
aktqdptltX
r2n
akrpdqrlr
akpq(lplq):
Hence, ifakpq60, thenlilj lplq, as claimed in (i). Now, ifakpi60 for somep6j, then by part (i)liljlilp, whenceljlp. p Next we analyse the functions aki, showing that the only indices (i;k)2nmfor whichaki is possibly nonzero are those for which we can findj2nsuch that [Xi;Xj]gkijYk.
LEMMA 2.6. Let i2n and k2m. If there is no j2n such that [Xi;Xj]gkijYk, then aki 0.
PROOF. Fix k2m. By assumption (b), there existm;n2n such that [Xm;Xn]Yk. Then for anyi2n
aki Yk(fi)[Xm;Xn](fi)Xm(Xn(fi)) Xn(Xm(fi)):
Thus, if there does not existj2nsuch that [Xi;Xj]gkijYk, theni62 fm;ng and so by (6)Xm(fi)0Xn(fi), whenceaki 0. p
LEMMA2.7. Let i;j;t2n with i6j. Then X
`2m
a`ija`t
0 if t62 fi;jg Xj(li) if ti
Xi(lj) if tj:
8>
<
>: (8)
Thus, if k2m and i;j2n are such that[Xi;Xj]gkijYk, then akt 0for t62 fi;jg, Xi(lj) gkijakj and Xj(li)gkijaki.
PROOF. By definition ofa`t, for anyt2n X
`2m
a`ija`t X
`2m
a`ijY`(ft)[Xi;Xj](ft)Xi(Xj(ft)) Xj(Xi(ft));
hence (8) follows from (6). If [Xi;Xj]gkijYk, then the sum on the left hand side reduces togkijakt and the conclusions follows from (6). p
LEMMA2.8. Let i2n.
(i) Forany k; `2m with k6`,P
t2na`tiakt 0.
(ii) If j2n and k2m are such that[Xi;Xj]gkijYk, then Xi(li)2gkijakj
(9)
PROOF. Sincenis two-step, for anyi2nand anyk; `2mwe have 0[Xi;Yk](g`)Xi(Yk(g`)) Yk(Xi(g`)):
Thus, if `6k, then by (7) Yk(g`)0 and so using (5) and the defi- nition of aki
0Yk(Xi(g`))YkX
t2n
a`tift
X
t2n
a`tiYk(ft)X
t2n
a`tiakt:
Next take`kand suppose thati;j2nare such that [Xi;Xj]gkijYk. Then by (7) we haveYk(gk) (lilj) and so by (5) and (8) we have
0[Xi;Yk](gk)
Xi(Yk(gk)) Yk(Xi(gk))
Xi(lilj) Yk
X
t2n
aktift
Xi(li)gkijakj X
t2n
aktiakt
Xi(li)gkijakj gkjiakj X
t2nt6j
aktiakt
Xi(li)2gkijakj X
t2nt6j
aktiakt;
where we have used thatakji gkij. Finally, by Lemma 2.7 we haveakt 0 for allt62 fi;jgand whentiwe haveakii0, so that all summands in the
last sum vanish. p
Observe that (8) and (9) describe Xj(li) for all possible pairs (i;j)2nn.
LEMMA2.9. If k2m and i;j2n are such that[Xi;Xj]gkijYkand there exists p2n, p6j, such thatakpi 60, then aki 0.
PROOF. By (ii) in Corollary 2.5,ljlp, so thatXj(lj)Xj(lp) and thus, by Lemma 2.8 and Lemma 2.7, respectively, we obtain
2gkijaki X
`2m
a`pja`pX
`2m`6k
a`pja`p
sinceakp0 whenp62 fi;jg. Now, eithera`pj 0 for all`2mwith`6k, in which case aki 0, or else there exists `2m with `6k such that a`pj60. We show now that for all such indices`we havea`p0, so that the sum on the right hand side vanishes, whence the result. Fix any such
`; according to Lemma 2.6 we may suppose that there exists q2nsuch
that [Xp;Xq]g`pqY`, for otherwisea`p0. Two cases arise: either q6j orqj. In the first case, by (ii) in Corollary 2.5 necessarilyljlq and hence ljlp, which gives Xq(lq)Xq(lp). By (9) and (8), respectively, we obtain
2g`qpa`pX
s2m
aspqaspg`pqa`p;
the latter equality expressing [Xp;Xq]g`pqY`. Therefore a`p0 as de- sired. In the second case, [Xp;Xj]g`pjY` and [Xi;Xj]gkijYk; by Lemma 2.8, we obtain that Xj(lj)2g`jpa`p and Xj(lj) 2gkijaki, respectively, so 2g`jpa`p 2gkijaki. Arguing as above we finally get
g`pja`pX
`2m`6k
a`pja`p 2gkijaki 2g`jpa`p:
Thereforea`p0 as desired. p
LEMMA2.10. Let k; `2m with`6k and let i;j;p2n with p6j be such that[Xi;Xj]gkijYkand[Xi;Xp]g`ipY`, then asi0forany s2m.
PROOF. By Lemma2.7 and item (i) of Lemma2.8 0X
s2m
aksja`sgkija`iakpja`p 0X
s2m
a`spaks g`ipaki a`jpakj:
We show now thata`i0 whenever`6k. This is obvious from the first equation ifakpj0. Suppose thenakpj60. By item (ii) of Corollary 2.5, ne- cessarilylilp, which givesXi(li)Xi(lp). By (9) and (8), respectively, we obtain
2g`ipa`pg`pia`p:
Thereforea`p0 and from the first equationa`i0, as desired.
Next we show thataki 0. This is obvious from the second equation if a`jp0. Suppose then a`jp60. By item (ii) of Corollary 2.5, necessarily lilj, which givesXi(li)Xi(lj). By (9) and (8), respectively, we obtain
2gkijakj gkijakj:
Thereforeakj 0 and from the second equationaki 0, as desired.
Finally, takes2mfor every s62 f`;kg. If there exists noq2nsuch that [Xi;Xq]gsiqYs, thenasi0 by Lemma2.6. Conversely, if [Xi;Xq]
gsiqYsfor someq2nthen the previous argument withqin place ofpand
sin place of`givesasi 0. p
Observe that by Lemma 2.6, Lemma 2.7 and Lemma 2.10 the structure of themnmatrixa(aki) is subject to many constraints. Indeed, if we fix both the row indexkand the column indexiand we cannot find another column indexjfor which [Xi;Xj]gkijYk, then Lemma2.6 says thataki 0.
Therefore for any fixed row k, the only column indicesi for whichaki is possibly nonzero are those for which [Xi;Xj]gkijYkfor somej. Take such a pair (i;j). By Lemma2.7 any otherakt vanishes. Hence for any fixed rowk there is at most a pair of column indicesiandjfor which the corresponding entries do not vanish, i.e., every row contains at most two nonzero entries.
But more is true: if for two different rows, say thekthand the`th, we find a coinciding column index, sayi, then by Lemma2.10 we have asi0 for everys. Therefore, every column contains at most one nonzero entry. It follows that, after reordering, we may assume that the matrixa(aki) ha s the following form
The stars refer to nonzero entries; evidently, each block of rows (those with two, those with one and those with no nonzero entries) could be empty. Next we investigate the functionsbki.
LEMMA2.11. Let i;t2n and k2m. Then Xi(akt) ditbkt (10)
and if there is no j2n such that[Xi;Xj]gkijYk, then Xi(aki) bki 0.
PROOF. Sincenis step-two, using the definitions ofaki,bki, and (6) 0[Xi;Yk](ft)
Xi(Yk(ft)) Yk(Xi(ft))
Xi(akt)ditYk(lt)
Xi(akt)ditbkt;
thereby showing (10). It follows from Lemma2.6 that if there is noj2n such that [Xi;Xj]gkijYk, then aki 0, in which case (10) implies
0Xi(aki) bki. p
Observe that (10) describes Xi(akj) for all possible triples (i;j;k)2 2nnm. Also, Lemma 2.6 says that if there is no j2n such that [Xi;Xj]gkijYk, thenaki 0; under these circumstances, because of (10), we havea fortiorithatbki 0.
LEMMA 2.12. If k2m and i;j2n are such that [Xi;Xj]gkijYk then bki bkj.
PROOF. Since [Xi;Xj]gkijYk, (8) gives Xj(li)X
`2m
a`ija`igkijaki: By the definitions ofbki,bkj and by (9) and (10)
gkijbki Yk(li)
[Xi;Xj](li)
Xi(Xj(li))) Xj(Xi(li))
Xi(gkijaki) Xj(2gkijakj)
gkijbki2gkijbkj;
wheregkij60, hence the result. p
LEMMA2.13. Let i;j;t2n with i6j and t62 fi;jg. Then X
`2m
a`ijb`t 0:
Thus, if k2m and i;j2n are such that[Xi;Xj]gkijYk, then bkt 0for t62 fi;jg.
PROOF. ApplyXtto the left hand side of (8), use (10) and Lemma 2.7. p LEMMA2.14. Let k2m and i;j2n be such that[Xi;Xj]gkijYk. If there exists p2n, p6j, such thatakpi60, then bki 0.
PROOF. By Lemma2.9 aki 0. Applying Xi and (10) to aki 0 we
obtain thatbki 0. p
LEMMA2.15. Let k; `2m with`6k and i;j;p2n with p6j be such that[Xi;Xj]gkijYkand[Xi;Xp]g`ipY`then
(i) bsi 0forany s2m;
(ii) bkq b`q 0forany q2n.
PROOF. By Lemma2.10asi0. ApplyingXiand (10) toasi 0 we ob- tain thatbsi 0 for anys2m. This proves (i). As for (ii), first of all we know from Lemma2.13 thatbkq0 ifq62 fi;jgandb`q0 ifq62 fi;pg. As for the remaining indices,by part (i) bsi 0 for any s2m and by Lemma 2.12
bkj bki 0 a ndb`pb`i0. p
Lemma 2.11 tells us that the matrixb(bki) has the same general shape as the matrix a(aki). Furthermore, Lemma2.12 says that whenever a row contains two nonzero entries, they actually coincide. Finally, Lemma 2.15 implies that no row may contain a single nonzero entry. The general shape of thebmatrix is thus
LEMMA2.16. Forall t2n and all`;k2m c`kt 0. Forall i;j2n and all k2m
Xj(bki)0:
(11)
PROOF. Assume firsti6j. By Lemma2.11 we haveXi(aki) bki and Xj(aki)0. Hence,
Xj(bki)Xi(Xj(aki)) Xj(Xi(aki))[Xi;Xj](aki)X
`2m
a`ijY`(aki)X
`2m
a`ijc`ki : In particular, if k2m and i;j2n are such that [Xi;Xj]gkijYk, then Xj(bki)gkijckki and by Lemma 2.12 they are both equal toXj(bkj). Similarly, Xi(bki)Xi(bkj) gkijckkj . Hence (11) will follow (for all possible indicesi;j andk) from the fact thatc`kt 0 for allt2nand all`;k2m.
Fix now k; `2m. By assumption (b) there exist i;j2n such that [Xi;Xj]gkijYk. We show next that for any `2m we have c`kt ck`t 0 whenever t62 fi;jg. Indeed, by Lemma2.7 akt 0 for t62 fi;jg, so that using the definition we obtainc`kt Y`(akt)0 for any`2m. Further, by
definition and by (10)
gkijck`t gkijYk(a`t)Xi(Xj(a`t)) Xj(Xi(a`t))
0 t62 fi;jg Xj(b`i) ti
Xi(b`j) tj;
8>
><
>>
: whose first line proves our claim.
Hereafter we fix k2m and we assume that i;j2n are such that [Xi;Xj]gkijYk. We show next thatckki ckkj 0. By definition ofbki,ckkj , by (9) and Lemma 2.12
0[Xi;Yk](li)
Xi(bki) Yk(2gkijakj)
Xi(bkj) 2Yk(gkijakj)
gkijckkj 2gkijckkj ;
the latter equality following from Xi(bkj) gkijckkj , which was shown above. This proves thatckkj 0 and an analogous argument shows that ckki 0.
Finally we show that whenever`6kwe havec`kt ck`t 0 fort2 fi; jg hence proving the lemma. To this end, we observe that by assumption (b) we may choose p;q2n such that [Xp;Xq]g`pqY` and two cases arise:
either fp;qg \ fi;jg ; or, by symmetry we may assume pi. In the first case, Lemma 2.13 impliesb`ib`j 0 and hence computing as above further implies that
gkijck`t gkijYk(a`t)
0 t62 fi;jg Xj(b`i) ti
Xi(b`j) tj 8>
><
>>
:
we get that the second and third lines vanish, whence ck`t 0 for all t2n. Similarly, by Lemma 2.13, bkpbkq0 and hence computing as above that c`kt 0 for all t2n. In the second case, that is pi, we compute that
gkijck`t gkijYk(a`t)Xi(Xj(a`t)) Xj(Xi(a`t))
0 t62 fi;qg Xj(b`i) ti
0 tq
8>
<
>:
and by Lemma 2.15b`i0, so thatck`t 0 for allt2n. Finally
g`pqc`kt g`pqY`(akt)
0 t62 fi;jg Xq(bki) ti
0 tj
8>
<
>:
and by Lemma 2.15bki 0, so thatc`kt 0 for allt2n. p Intheend wehave that allY-derivatives and allX-derivatives ofaktare zero for all choices of the indices, so the system ``closes up''. Indeed, all the deri- vativesofthe functionsgk;ft;lt;akt;bkt,for anyt2nandk2mareexpressedin terms of the functions themselves, as indicated by equations (5) through (11).
2.3.2 - The connection.
Recall that our strategy to prove rigidity is to introduce a connection on a vector bundle in such a way that finding a solution to the multicontact system is equivalent to finding a covariant constant section of the bundle. The bundle we are referring to is a trivial subbundle of the trivial r-plane bundle p:T!NoverN, whose total space isTRr, withrm2n2mn. The idea is the following: once the multicontact structure has been selected, that is, the basesfX1;. . .;XngandfY1;. . .;Ymghave been fixed in such a way that assumptions (b) and (c) are satisfied, we choose only those directions along which the coordinates of a prolonged multicontact vector field are possibly non vanishing, as described by the lemmas of the previous sections.
The fiber coordinates will thus be indexed by arrays of the form g f a
l b 0
@
1 A
where, first of all,f;l2Rn,g2Rm. As for the vectorsaandb, they will be labeleda(aki) a ndb(b`j) respectively, with indices (i;k) and (j; `) run- ning only in those index sets that correspond to non vanishing coordinates of multicontact vector fields. As already illustrated, the results of the previous section show that for eachk2mthere exists asetAk(consisting at most of two indices) such that akp0 ifp62Ak, whenever akp is the prolonged co- ordinate of a multicontact vector field. Notice that ifi;j2Akit may well be thataki 6akj. We thus introduce the index set
A f(i;k)2nm:i2Akg
in order to parametrize thea-vector, that is we takea(aki) with (i;k)2A.
Similarly, for each`2mthere exists asetB~`(consisting at most of two indices) such thatb`q 0 ifq62B~`, wheneverb`qis the prolonged coordinate of amulticontact vector field. However, ifi6jare inB~`, thenb`ib`j. We thus define anew index setB`consisting at most of a single index in order to achieve a parametrization of the b-vector, by choosing one of i or j.
Finally, we introduce the index set
B f(j; `)2nm:j2B`g
and considerb(b`j) with (j; `)2B. Hencea2Rc(A)andb2Rc(B), where c(A) a ndc(B) are the cardinalities ofAandB, respectively. The bundle we shall be concerned with is thereforep:E!N where the total space is ERs, withsm2nc(A)c(B).
The sections S2G(E) of E are mappings S:N!E such that pSidNthat we arrange in arrays as follows
x7!S(x) f(x) a(x) g(x) l(x) b(x) 0
@
1 A2E (12)
wheregis acolumn vector inRm,f,lare s row vectors inRnanda2Rc(A) and b2Rc(B) are vectors whose entries are arranged in (mostly empty) nmmatrices. For practical purposes, it is convenient to think ofEa s a subbundle of T and hence to regard a and b as nm matrices whose entries are zero in the complement of the index setsAandB. Omitting the dependence onx, one may thus think
and remember that mosta's andb's are zero forS2G(E).
An affine connection on N is an R-bilinear map r:X(N)G(E)!
!G(E), writtenr(X;S) rXS, with the following properties:
(i) rWXSWrXSfor allW2 F(N) a llX2X(N) and allS2G(E);
(ii) rX(WS)X(W)SWrXSfor allW2 F(N), allX2X(N) and all S2G(E).
We will define aconnection onEby specifying its valuesrXiSfor all i2nandrYkSfor allk2mand then using (i). We shall do so by using the right hand sides of the various (affine) equations that have been obtained in the previous section. In order to write explicitly the definition, we para- metrizeSas in (12). Instead of writingrXiSandrYkSas large arrays, we introduce the dual bases (gs),ft,lt, (ast)and (bst). Here is the definition of the various components ofrYkS:
(gs)(rYkS)Yk(gs)dks(lilj) ft(rYkS) Yk(ft) akt if (t;k)2A
Yk(ft) if (t;k)62A
lt(rYkS) Yk(lt) bkt if (t;k)2B Yk(lt) if (t;k)62B
(ast)(rYkS)Yk(ast) (bst)(rYkS)Yk(bst):
(13)
For the sake of precision we remark that in the first line one picks any pair (i;j) with the property that [Xi;Xj]gkijYk (with gkij60), certainly existing by our hypotheses, and the sum lilj is independent of such choice; further, the last two lines are defined for (t;s) in A and B, respectively. Similarly, the definition of the various components of rXiS is:
(gs)(rXiS)Xi(gs) X
t2n
astift
ft(rXiS)Xi(ft)ditlt
lt(rXiS)
Xi(lt) X
(t;`)2A
a`tia`t if t6i
Xi(li) 2gkijakj if tiand (j;k)2A Xi(li) if tiand (j;k)62A 8>
><
>>
:
(ast)(rXiS)Xi(ast)ditbst (bst)(rXiS)Xi(bst):
(14)
Here again the last two lines are defined for (t;s) inAandB, respectively; as for the value oflt(rXiS), the indexjappearing in the right hand side is an index such that [Xi;Xj]gkijYk(withgkij60), certainly existing by our hy- potheses.
PROPOSITION2.17. Formulae(13) and(14)define an affine connec- tion onE.
PROOF. LetXbe any vector field in the basis and letSbe asection. We temporarily change notation and write Sas a vector with componentsSi withiranging in the index setIand observe that formulae (13) and (14) are all of the form
rX(S)
iX(Si)X
j2I
CjSk(j);
where j7!k(j) is some mapping ofI andCj is apossibly zero constant.
Property (i) is automatic from our definition. As for (ii), for any smoothc we have
rX(cS)
iX(cSi)X
j2I
CjcSk(j)
X(c)SicX(Si)X
j2I
CjcSk(j)
X(c)Sic rX(S)i;
as desired. p
2.3.3 - End of proof.
Summarizing, the vector field (3) is a multicontact vector field if and only if the functionsfft;lt;gs;ast;bst :t2n;s2mgsatisfy the equations (5) through (11) and this in turn happens if and only if the sectionSdefined by (12) is a covariant constant section of the bundlep:E!Nwith respect to the connectionr, that is, if it satisfiesrXS0 for anyX2X(N). Recall that the bundle we shall be concerned with is thereforep:E!N where the total space isERs, withsm2nc(A)c(B). Thus, there are at mostslinearly independent such sections, and this proves that the space of multicontact vector fields is at mosts-dimensional.
There are noa priori conditions on thef;l2Rn,g2Rm. Therefore, they may contribute up to 2nm dimensions. Moreover, we consider a(aki) with (i;k)2A and b(b`j) with (j; `)2B. Hencea2Rc(A) and b2Rc(B), wherec(A) a ndc(B) are the cardinalities ofAandB, respectively.
We achieve the total number ofs.
In the next section we improve our result by proving that the connec- tionris actually flat.