of weakly Noetherian constants of motion of nonholonomic systems Francesco Fassò
∗
, Andrea Giacobbe†
and NicolaSansonetto
‡
(December23,2009)
Abstract
Wedevelopamethodtogiveanestimateonthenumberoffunctionallyindependentconstants
ofmotionofanonholonomicsystemwithsymmetrywhichhavethesocalled`weakly
Noethe-rian'property [22]. We show thatthis numberis boundedfromabove by thecorank ofthe
involutiveclosureofacertaindistributionontheconstraintmanifold. Theeectivenessofthe
methodisillustratedonseveralexamples.
Keywords: Nonholonomicsystems,Constantsofmotion,Firstintegrals,Noethertheorem.
MSC:37J15,37J05,70F25.
1 Introduction
Howmany(functionallyindependent)constantsofmotiondoesanonholonomicsystemwith
sym-metryhave? In these terms,this questionis certainlytoovague to be answered. Forholonomic
systemswithsymmetry,anaturalanswerisprovidedbytheconservationoftheenergymomentum
map,whichimpliesalowerboundonthenumberofconstantsofmotion. Inthenonholonomiccase,
instead,thesituation is notascleareven inthe simplestcase ofsystemswithlinearconstraints,
naturalLagrangians(=kineticminuspotentialenergies)andlifted actions,whichisthecasethat
weconsiderin thispaper. Therearetwooppositereasonsforthis:
•
On theonehand, onlycertain componentsof themomentum map areconstantsof motion, andtheir numbermaybedicult toassess. Infact, acomponentof themomentum map isconserved if and only ifits innitesimal generatoris a section of acertain distribution, the
socalledreactionannihilatordistribution
R
◦
,see[23].
•
Ontheotherhand,theclassofconstantsofmotionofnonholonomicsystemswithsymmetry may be larger than just the components of the momentum map. In particular, this classmayinclude functionslinearin themomentaknownasgaugemomenta. Theseare,roughly
speaking, constants of motion generated by certain vector elds which are tangent to the
∗
Università diPadova, Dipartimento diMatematica Pura e Applicata, Via Trieste 63, 35121 Padova, Italy.
(E-mail: fasso@math.unipd.it)
†
Università diPadova, Dipartimento diMatematica Pura e Applicata, Via Trieste 63, 35121 Padova, Italy.
(E-mail: giacobbe@math.unipd.it)
‡
Università di Verona, Dipartimento diInformatica,Cà Vignal2, Strada LeGrazie15, 37134 Verona, Italy.
grouporbits but are notinnitesimal generatorsof the groupaction, see [5, 33, 22]. There
aredierentwaysofcharacterizingthevectoreldswhichproduceconservedgaugemomenta
[28,16,39,22],but heretoo,nogeneralwayofassessingtheirnumberisknown.
Inthissituation,itisverydiculttoanswerinfullgeneralitythequestionraisedabove. Therefore,
inthispaperwetrytoprovidearst,verypartialanswertoit,byproducingabound fromabove
on thenumberof all functionally independent constantsof motionwith acertain propertythe
socalledweak Noetherianpropertyidentied in[22].
Inordertoappreciatetheoriginofthisclassofconstantsofmotionrecallthat,inHamiltonian
mechanics,oneofthefundamentalpropertiesofthelinksymmetriesconservationlawsisthefact
thatthemomentummapofaHamiltonianactiondependsonlyonthegroupaction. Therefore,the
conservation of themomentum map provides conserved quantities forall systemswith invariant
Hamiltonian. This propertyis sometimes called theNoetherian property (or condition) of the
momentummap [35].
Innonholonomicmechanics,however,whichcomponentsofthemomentummapareconserved
quantities depends on the Hamiltonian (and on the constraint manifold), and the Noetherian
propertyisbroken. Infact,itisnotevencompletelyclearwhata`Noetherianconstantofmotion'
shouldbeinthenonholonomiccontext(seetheRemarkattheendofSection3.A).Amongfunctions
linearinthemomenta,someconservedcomponentsofthemomentummap(particularlythosewith
`horizontal'innitesimalgenerators,iftheyexist)dohavetheNoetherianproperty,butingeneral
gauge momenta do not. However, certain gauge momentaparticularly those with horizontal
generatorshaveaweakerproperty: they are conserved quantities of all nonholonomicsystems
with xed constraint manifold and xed kinetic energy, but anyinvariantpotentialenergy [22].
ThisistheweakNoetherian property.
In[22],thispropertywasdenedonlyforconstantsofmotionwhicharelinearinthemomenta,
whichisthecaseofthemomentummapsofliftedactionsandismostcommonlymetinexamples.
Inthisarticle,weconsiderinsteadallweaklyNoetherianconstantsofmotion(eventhenonlinear
ones,if theyexist) withthe aimof establishingatechniquewhich cangivesome informationon
the numberof them which are functionally independent. The reason whyour method doesnot
applytothelinearweaklyNoetherianconstantsofmotionalonewillbecomeclearlater.
Specically,wewillshowthat,underacertainsmoothnesshypothesis,thenumberof
function-allyindependent smoothlocal weaklyNoetherian constantsof motionof anonholonomicsystem
whoseHamiltonianisinvariantunderaliftedactionisgivenbythecorankoftheinvolutiveclosure
∆
∞
ofacertaindistribution
∆
ontheconstraintmanifold. Wewillgiveacompletedescriptionof∆
in thecaseoffreeand properactions,and wewill recallastandardtechniqueto computethe corankof∆
∞
if
∆
isrealanalytic. Inordertoshowthefeasibilityoftheprocedure,wewillapply ittoafewsamplecasesinwhichthiscorankcanbedeterminedanalytically(averticaldisk withvarious symmetrygroups) andto amorecomplexcasewhere theanalysis hastobecarried over
usingsymbolicmanipulationsoftware(aheavyballwhichrollsonasurfaceofrevolution).
ThenumberoflocalweaklyNoetherianconstantsofmotiongivesofcourseonlyanupperbound
onthenumberofgloballydenedconstantsofmotion,whicharethosesignicantforthedynamics.
Nevertheless,knowingthisboundcangivesomeinformationsonthesystem,butalso,canconrm
whether all constantsof motion with this propertyhave been determined. We will analyze this
situationontheexamples.
Ourstudy usesin anessentialwaythe Hamiltonianformulation ofnonholonomic mechanics,
whichisshortlyreviewedinSection2. InSection3weintroducethenotionofweaklyNoetherian
constants of motion and we shortly review the case of gaugemomenta. After this background
material,inSection4werelatetheexistenceoflocalweaklyNoetherianconstantsofmotiontothe
corankof the distribution
∆
∞
, and in Section 5wegiveexplicit expressionsfor thedistribution
∆
in the caseof a freeand properlifted action. Sections 6and 7 are devoted to theexamples. A section of conclusions and perspectives follows; in particular, weoutline there the use of our2 Nonholomic systems
A.Lagrangian formulation. Asalreadymentioned,weshallconsideronlythecaseofsystems
subjecttolinearnoholonomicconstraints.Forgeneralreferences,see[34,12,14,7,9]andreferences
therein.
Asastartingpoint,consideraholonomicmechanicalsystemwith
n
dimensionalconguration manifoldQ
and kinetic energyT (v
q
) =
1
2
a
q
(v
q
, v
q
)
, wherea
is a Riemannian metric onT Q
, subjecttoforcesdescribedbyapotentialenergywhichistheliftofafunctionV
onQ
. Then,the LagrangianoftheholonomicsystemisL = T − V ◦ π
Q
,whereπ
Q
: T Q → Q
isthetangentbundle projection.Forgreaterclarity,weresorttoacoordinatedescriptionwheneverappropriate. Indoingso,we
denote by
(q, ˙q) ∈ R
n
× R
n
(local) bundle coordinates on
T Q
and byadotthe inner productinR
n
andweusethesymbolloc
=
to stressthat therighthandsideofan equalityisanexpression foralocal representativeofthelefthandside. Thus, thekineticenergyislocallywrittenT (q, ˙q)
loc
=
1
2
˙q · A(q) ˙q
forasymmetricandpositivedenite
n × n
matrixA(q)
.A linear nonholonomic constraintis given by anonintegrabledistribution
D
onQ
, that we assumetohaveconstantrankr
,2 ≤ r < n
. ThedistributionD
iscalledtheconstraintdistribution and can belocally dened asthe kernelofk := n − r
linearly independent dierential 1formsτ
1
, . . . , τ
k
onQ
,sothat itsbersareD
q
= ker
τ
1
(q), . . . , τ
k
(q)
.
(1)Inbundlecoordinates
(q, ˙q)
inT Q
,thebersofD
canbewritten asD
q
loc
= { ˙q ∈ R
n
: S(q) ˙q = 0}
for a
k × n
matrixS(q)
with rankk
, which depends smoothlyonq
. Theconstraintdistribution canalsobethoughtofasasubmanifoldD
ofT Q
ofdimension2n − k
,whichinbundlecoordinates isgivenbyD
loc
= {(q, ˙q) ∈ R
2n
: q ∈ Q , ˙q ∈ D
q
}
andwillbecalledthe(Lagrangian)constraintmanifold.
D'Alembert'sprincipleassumesthatthereactionforceexertedbythenonholonomicconstraint
annihilates the bers of (an appropriatejet extension of) thedistribution
D
and leads to a dy-namicalsystemonthesubmanifoldD
,whichisdescribedbyLagrangeequationswithmultipliers. Eliminatingthemultipliers givesthereactionforceasafunction ofthekinematical statev
q
∈ D
andleadstoavectoreld ontheconstraintmanifoldD
,whichgivestheequationsofmotion. In bundlecoordinates,theequationsofmotionhavetheformd
dt
∂L
∂ ˙q
−
∂L
∂q
= R
(2)where
R = R(q, ˙q)
isthereactionforceexertedbythenonholonomicconstraint.The expression of
R
in bundle coordinates can be found, e.g., in [1, 23, 6]. We recall this expressionbecausewewilluseitintheexamplesofSection6. LetΠ(q)
betheA(q)
−1
orthogonal
projectorontotheorthogonalsubspaceof
D
q
= ker S(q)
,that isanddenethevectors
β(q, ˙q) ∈ R
n
,V
0
(q) ∈ R
n
andγ(q, ˙q) ∈ R
k
ashavingcomponentsβ
i
=
X
j,h
∂A
ij
∂q
h
−
1
2
∂A
jh
∂q
i
˙q
j
˙q
h
,
V
i
0
=
∂V
∂q
i
,
γ
a
=
X
j,h
∂S
aj
∂q
h
˙q
j
˙q
h
(i, j, h = 1, . . . , n
,a = 1, . . . , k
). Then,R = Πβ − S
T
SA
−1
S
T
−1
γ + Π V
0
.
(3)B. Local Hamiltonian formulation. We passnow to the Hamiltonian formulation of
non-holonomicallyconstraintsystems,see[40,4,32]forgeneralreferences. Webeginbygivingalocal
description of this formulation, in Darboux coordinates
(q, p)
onT
∗
Q
. We will use such local
formulationforthecomputationofthedistribution
∆
inSection5andintheexamplesofSections 6and 7.TheLegendre transformation
Λ : T Q → T
∗
Q
relativeto theLagrangian
L = T − V ◦ π
Q
isa dieomorphism andthe imageM := Λ(D)
of thesubmanifoldD
ofT Q
is asubmanifold (anda subbundle)ofT
∗
Q
ofdimension
2n−k
,whichwillbecalledthe(Hamiltonian)constraintmanifold. InDarbouxcoordinatesΛ(q, ˙q)
loc
=
(q, A(q) ˙q)
and
M
loc
= {(q, p) ∈ R
2n
: q ∈ Q , A(q)
−1
p ∈ D
q
} .
The equations of motion of the nonholonomicsystem are Hamilton equations with the reaction
force. TheHamiltonian is
H = T ◦ Λ
−1
+ V ◦ π
∗
Q
,whereπ
∗
Q
: T
∗
Q → Q
is thecotangentbundle projection. InDarbouxcoordinates,H(q, p)
loc
=
1
2
p · A(q)
−1
p + V (q)
andtheequationsofmotion
are
˙q =
∂H
∂p
(q, p) ,
˙p = −
∂H
∂q
(q, p) + R q, A(q)
−1
p
.
Theseequationsarethelocalrepresentativeofadynamicalsystemontheconstraintmanifold
M
, thatwecallthe(Hamiltonian)nonholonomicsystem(H, M )
.Notethat it ispossibleto express theconstraintmanifold
M
alsoin terms ofthe 1formsτ
. In fact, the submanifoldD
ofT Q
can be (locally) described as the zerolevel set of the map˜
τ = (˜
τ
1
, . . . , ˜
τ
k
) : T Q → R
k
,whereτ
˜
i
: T Q → R
is givenby˜
τ
i
(v
q
) := hτ
i
(q), v
q
i
for allv
q
∈ D
q
,q ∈ Q
. ThereforeM = ˆ
τ
−1
(0)
forˆ
τ (q, p) := (˜
τ ◦ Λ
−1
)(q, p)
loc
=
S(q)A(q)
−1
p ,
see[39].C. Global Hamiltonian formulation. A local,coordinate descriptionwill not be sucient
forouranalysisand weneedageometric formulationof theequations ofmotion. Inviewofthis,
wepreliminarilyrecall somefacts from symplecticgeometry;seee.g. [31, 35]for detailsonthese
topics.
Let
ω
bethecanonicalsymplecticformofT
∗
Q
. Asusual,
X
F
denotestheHamiltonianvector eld of asmoothfunctionF
onT
∗
Q
, that is,
i
X
F
ω = −dF
. Similarly, the Hamiltonian vector eldX
τ
ofaclosed 1formτ
isdened byi
X
τ
ω = −τ
. IfE
m
isasubspaceofthetangentspaceT
m
(T
∗
Q)
,m ∈ T
∗
Q
,thenitssymplecticorthogonalis denedas
Recall that
(E
ω
m
)
ω
= E
m
and thatE
m
andE
ω
m
havecomplementary dimensions. Thepolar ofa distributionT
onT
∗
Q
isthedistributionT
ω
onT
∗
Q
whosebersarethesymplecticorthogonals
tothebersof
T
.TheHamiltonian formulationof nonholonomicmechanics requires the introduction of a
cer-tain distribution
D
onT
∗
Q
, whose polar is related to the reaction force, see [4, 32, 39, 13, 14].
Specically, at each point
m ∈ T
∗
Q
, theber
D
m
⊆ T
m
(T
∗
Q)
of
D
isthe preimageof theberD
π
Q
(m)
oftheconstraintdistributionD
underthederivativeT π
∗
Q
ofthecotangentbundle projec-tionπ
∗
Q
: T
∗
Q → Q
. With referencetotherepresentation(1)ofD
,thebersofD
arethus given byD
m
= ker
(π
Q
∗
)
∗
τ
1
(m), . . . , (π
Q
∗
)
∗
τ
k
(m)
loc
=
(u
q
, u
p
) ∈ R
n
× R
n
: u
q
∈ D
π
Q
(m)
(4) forall
m ∈ T
∗
Q
. If,asweassumein thiswork,
D
hasconstantrankn − k
, thenD
hasconstant rank2n − k
.Thedistribution
D
isdenedonallofT
∗
Q
,but wewillneedonlyitsrestrictionto
M
. Onthe pointsm ∈ M
, theberD
m
is distinctfromthetangentspaceT
m
M
,eventhoughtheyhavethe samedimension. Infact,T
m
M
isthe annihilatorof thedierentialsof thek
functionsˆ
τ
a
which, inDarbouxcoordinates,arethefunctions[S(q)A(q)
−1
p]
a
. Notealsothat, whileM
dependsboth onD
andΛ
,andthus onthekineticenergyT
,D
anditspolarD
ω
dependonlyon
D
.If, as we assume in this work, the constraints are linear and ideal, then the Hamiltonian
counterpartofthereactionforceisasection
R
of(therestrictiontoM
of)thepolardistributionD
ω
,
seee.g. [4,32,39,13,14]. Thedistribution
D
ω
hasconstantrank
k
andislocallygeneratedbythe vectoreldsX
(π
∗
Q
)
∗
τ
1
, . . . , X
(π
∗
Q
)
∗
τ
k
. Correspondingly,thepolardistribution[T
m
M ]
ω
hasconstant
rank
k
,too,andislocally generatedbyX
τ
ˆ
1
, . . . , X
τ
ˆ
k
.Asprovenin[4,32]ateachpoint
m ∈ M
thesubspacesT
m
M
andD
ω
m
ofT
m
T
∗
Q
aretransversal
andcomplementary,sothat
T
m
(T
∗
Q) = T
m
M ⊕ D
ω
m
∀ m ∈ M .
(5)This implies that, at each point
m ∈ M
, there is a unique splittingu = u
T M
+ u
D
ω
of vectors
u ∈ T
m
(T
∗
Q)
,whereforeachsubspaceE
ofT
m
(T
∗
Q)
,
u
E
∈ E
. Correspondingly,anyvectoreld
X
onT
∗
Q
canbeuniquelydecomposed onthepointsof
M
asX = X
T M
+ X
D
ω
(6) withX
T M
asectionofT M
andX
D
ω
asectionofD
ω
.Splitting (6)isusedtowritegloballytheequationsofmotionofnonholonomicsystems[4,32].
Infact,aconstrainedmotion
t 7→ m
t
∈ M
satisesm = X
˙
H
(m) + R(m) = X
T M
H
(m) + X
D
ω
H
(m) +
R(m)
. Hence,m ∈ T
˙
m
M
andR ∈ D
ω
ifandonlyifthereactionforce satises
R = −X
D
ω
H |M
andt 7→ m
t
satisestheequationofmotion˙
m = X
T M
H
(m) ,
m ∈ M .
Thedynamicalsystemon
M
denedbythevectoreldX
T M
H
isthe(Hamiltonian)nonholonomic system(H, M )
wementionedbefore.Splitting(5)playsacentralrolealsointhestudyofconstantsofmotion [18,39];wewillcome
backonthisinSection4.
Remark: For completeness, wemention heretwofacts that we will notuse. (1)In thecase
of linear constraints, the vector eld
X
T M
H
belongs toD
as well and is therefore a section of the distributionH
= T M ∩ D
alongM
, see [4, 39]. As proven in [4, 16], the symplectic form restrictedtoH
isnondegenerateandT
m
(T
∗
Q) = H
m
⊕ H
ω
m
form ∈ M
. Thesefactsplayacentral rolein thereductionprocedure developedin [4]and inthedistributionalHamiltonianaproach tononholonomicsystemsdevelopedin[17,15]. (2)Thenonholonomicdynamicsconservestheenergy,
thatis,
X
T M
3 Weakly Noetherian constants of motion
A. Denition. We introduce now the notion of weakly Noetherian constants of motion and
review the case of gauge momenta. From now on, it is tacitly understood that
(H, M )
stands foraHamiltoniannonholonomicsystemofthetypespeciedinSection2,whichcorrespondstoacongurationmanifold
Q
,aconstraintdistributionD
,akineticenergyT
andapotentialenergyV
. By a constant of motion of the nonholonomic system(H, M )
we mean a smooth functionF : M → R
suchthat theLie derivativeX
T M
H
(F ) = 0
inall ofM
. Givenan actionΨ
Q
of aLie
group
G
onQ
,wedenotebyΨ
T
∗
Q
itslifttothecotangentbundle
T
∗
Q
.
DenitionFixthecongurationmanifold
Q
,theconstraintdistributionD
onQ
,anactionΨ
Q
of
aLiegroup
G
onQ
,andaΨ
T
∗
Q
invariantkineticenergy
T : T
∗
Q → R
. Then,asmoothfunction
F : M → R
is said to be aweakly Noetherian constant ofmotion relativeto(T, M, Ψ
Q
)
if it is
a constant of motion of all nonholonomic systems
(T + V ◦ π
∗
Q
, M )
withΨ
Q
invariant potential
energy
V : Q → R
.Therestrictionto lifted actionsinthis Denitionis notnecessary,but thisisthecasewewill
considerin thesequel.
Remark: NotionsofNoetherianandofweaklyNoetherianconstantsofmotionofnonholonomic
systemswere givenin [22] in thespecial caseoffunctions linearin themomenta. Thetreatment
therewasin theLagrangiancontextbut,sinceunder ourhypothesestheLegendretransformation
isadieomorphism,everythingtransferstotheHamiltoniancontext. However,thereisasubtlety
involvedinextendingthenotionofNoetherianitytononlinearfunctions,andthisextensioncannot
beachievedbyjustdroppingtheconstancyofthekineticenergy
T
inthepreviousDenition. The factisthat,inamechanicalcontext,onestartswithagivenLagrangianconstraintmanifoldD
and (evenassumingthatthisoperationcouldbedonewithoutchangingD
,whichmightbeunrealistic) changing the kineticenergy changes the Legendre transformationΛ
and hence the Hamiltonian constraintmanifoldM = Λ(D)
. SinceM
is thenaturaldomain of denition ofthe constantsof motion,itisnotcompletelyclearwhatshouldtheappropriatedenitionofNoetherianconstantofmotionbe,ifany. Ifoneconsidersonlyfunctionswhicharedenedinallof
T
∗
Q
,then`horizontal
momenta' (in the sense of the next subsection) are obviously Noetherian [22]. Note that this
dicultycannot beovercomebysimplyworkingin theLagrangiancontext,wherethe constraint
manifold
D
isxed,becausechangingthekineticenergychangesthepullbackofJ
toT Q
.B.WeaklyNoetheriangaugemomenta. Inordertogivesomeperspectiveforthesubsequent
treatmentwereviewveryquicklythepropertiesofweakNoetherianityoftheclassofconstantsof
motionwhichhas been moreextensivelystudied, that formed bymomenta andgaugemomenta.
Formotivations,detailsandexamplessee[22,24]andreferencestherein.
Bya`linear'functionon
M
wemeantherestrictiontoM
ofafunction onT
∗
Q
whichislinear
inthemomenta,that is,which canbewrittenas
F = p · ξ
Q
|
M
withsomevectoreld
ξ
Q
on
Q
. Thevectoreldξ
Q
iscalledageneratorof
F
and,sinceM ⊂ T
∗
Q
,
itisnotunique(see[28,22, 23]forfurtherdetails). Fixnowanaction
Ψ
Q
ofaLiegroup
G
onQ
andletG
denotethedistributiononQ
whosebersG
q
arethetangentspacesT
q
O
q
totheorbitsO
q
ofΨ
Q
. Alinearfunction on
M
iscalledagaugemomentumrelativetotheactionΨ
Q
ifithas
agenerator
ξ
Q
withthefollowingtwoproperties: (1)itisasectionof
G
,and(2)itscotangentliftξ
T
∗
Q
innitesimallypreservestheHamiltonianin
M
,namelyξ
T
∗
Q
(H)|
M
= 0
. Anysuchgenerator is called a gauge simmetry. (Note that agauge momentum may have as well generators whichdo notsatisfy either condition and arenot gaugesymmetries). If the gaugesymmetry
ξ
Q
is an
innitesimalgeneratoroftheaction
Ψ
Q
of themomentum map of thelifted action
Ψ
T
∗
Q
and will be called a momentum. Therefore, in
the sequelwe referonly to gaugemomenta. A gaugemomentum is said to be horizontal if it is
generated by ahorizontal gaugesymmetry, that is, a gaugesymmetry which is asection of the
constraintdistribution
D
.Agaugemomentumisaconstantofmotionofthenonholonomicsystem
(H, M )
ifandonlyifit isgeneratedbyagaugesymmetrywhichisasectionofacertaindistributionR
◦
D,T,V
onQ
[23,22]. This distribution is called the reactionannihilator distribution, depends onD
,T
andV
and is an overdistribution of the constraint distibutionD
, that is, its bers contain those ofD
. This implies,in particular, theverywellknownfact that allhorizontal momentaand gaugemomentaareconstantsofmotion[30,3,16,8,13,5,33,7,19,15]. Thisalsoimpliesthatagaugemomentum
isweaklyNoetherianifandonlyifitisgeneratedbyagaugesymmetry
ξ
Q
whichisasectionofthe
distribution
∩
V
R
◦
D,T,V
which isobtainedby taking theintersection ofallthe distributionsR
◦
D,T,V
corresponding to
Ψ
Q
invariantpotentialenergies
V
. Since∩
V
R
◦
D,T,V
isanoverdistributionofD
,thisimpliesinparticularthathorizontal momenta and horizontal gauge momenta are weakly Noetherian [22]. Moreover,if the action is transitiveon
Q
, or moregenerally ifG
is an overdistribution ofD
, that is,G
q
⊇ D
q
for allq ∈ Q
, then any conservedgaugemomentum is horizontal ([22], Proposition 4) and henceweaklyNotherian.However,theclass ofweakly Noetherianmomenta and gaugemomenta mightbelargerthanthe
horizontalones,apossibilitywhichmightbeparticularlyimportantincasessuchasthegeneralized
Chaplyginsystems,forwhich
D
q
∩ G
q
istrivial.Remark: Thenumberoffunctionallyindependenthorizontalgaugemomenta(or,equivalently,
offunctionallyindependentconstantsofmotionidentiedbythemomentum equation[22])isnot
boundedfromabovebytherankofthedistributionwithbers
G
q
∩ D
q
,unlessthisrankiseither zeroorone. Infact, giventwoormorevectorelds onQ
, liftingtheir linearcombinations with coecient functions produce vectorelds onT Q
which need notbelong to the span of the lifts of thegivenvectorelds. Fora(holonomic)exampleof atwodimensionalLie groupwith threeindipendenthorizontalgaugemomentaseeSection6of [22].
4 On the number of weakly Noetherian constants of motion.
A.First integrals ofdistributions. Asexplainedin theIntroduction,ouraimistodevelopa
methodtocomputeanupperboundonthenumberoffunctionallyindependentweaklyNoetherian
constants of motion. The key will be regarding these functions as `rst integrals' of a certain
distribution. Wethusbeginthisanalysisbyintroducingthenotionofrstintegralsofadistribution
andstudyingsomeoftheirproperties.
For all notions and properties relative to distributions, and in particular for the denition
of smooth and real analytic distributions, we refer to [31, 35] (which call however generalized
distributionwhatwecallheredistribution)andto[29,12].
Giventwodistributions
T
andS
onamanifoldM
,wesaythatT
isanoverdistributionofS
, andwriteT
⊇ S
,iftheirberssatisfyT
m
⊇ S
m
forallm ∈ M
. Weusethetermunder-distribution inasimilarway.Wesaythatasmoothfunction
F : M → R
isanrstintegralofadistributionT
onamanifoldM
ifthedistribution withbersker dF (m)
isanoverdistributionofT
,that is,ker dF (m) ⊇ T
m
∀ m ∈ M .
(7)Alocal rstintegral of
T
isanrstintegraloftherestrictionofT
tosomeopensubsetofM
. Thenotionofrstintegralsofadistributiongeneralizesinanobviouswaythatofrstintegralswe have not been able to nd any reference to it in the literature; therefore, we collect herea
few properties of these objects. Roughlyspeaking, therelevant fact is that the rst integralsof
a distributionare constant on the (StefanSussmann) orbitsof the distribution, which, for non
integrable distributions, have dimensions larger than the rank of the distribution. Thus, rst
integralsofadistributionarerstintegralsofthefoliationthatitgenerates,andthedimensionof
theorbitsputsaboundonthenumberoffunctionallyindependentrstintegralswhichisstricter
thanthatgivenbytherankofthedistribution. Wegivenowaformal,andcomputationallymore
convenient,`innitesimal'statementofthis fact.
Preliminarily we recall that, given a smooth distribution
T
there exists a unique smooth overdistributionT
∞
of
T
which is involutiveand isminimal among all smoothinvolutiveover distribution ofT
. This distribution is called the involutive closure ofT
. IfT
is real analytic, thenT
∞
is real analytic and there is astandardway to compute
T
∞
, which isbased on taking
commutators,seeSection 4.C.Forsomedetails,see[29,12] andreferencestherein.
Proposition 1. Let
T
beasmoothdistributiononM
andU ⊂ M
anopenset. AsmoothfunctionF : U → R
isarst integral ofT
|
U
if andonlyif itisarstintegral ofT
∞
|
U
.Proof Weusethefollowingnotation: if
D
isa(notnecessarilysmooth)distribution,thensmt(D)
denotesthelargestsmoothunderdistributionofD
[29].Preliminarily, we prove that
(T|
U
)
∞
= T
∞
|
U
. One inclusion is obvious:T
∞
|
U
is a smooth involutiveoverdistributionofT
|
U
and hence is an overdistribution of(T|
U
)
∞
. Viceversa, let
B
bethedistribution onM
which coincides with(T|
U
)
∞
on
U
and hasbersT
m
M
form /
∈ U
. Thereforesmt(B)|
U
= (T|
U
)
∞
because the smoothing operationdoesnot change the bersof a
smoothdistribution. Ontheotherhand,
smt(B)
is asmoothoverdistributionofT
which isalso involutive. Hence,smt(B) ⊇ T
∞
and(T|
U
)
∞
= smt(B)|
U
⊇ T
∞
|
U
. Wethus writeT
|
∞
U
for(T
U
)
∞
= T
∞
|
U
. IfF
isarstintegralofT
|
∞
U
thenitis arstintegral ofT
|
U
⊆ T|
∞
U
. Conversely, assume thatF
is arst integralofT
|
U
, that is,ker dF ⊇ T|
U
. The distributionker dF
neednotbesmooth. Nevertheless,sinceT
|
U
issmooth,ker dF ⊇ smt(ker dF ) ⊇
T
|
U
. Since thecommutatorof anytwosmoothsections ofker dF
is asmoothsection ofker dF
, thedistributionsmt(ker dF )
isinvolutive. Thatker dF ⊇ T|
∞
U
followsnowfrom theminimalityofT
|
∞
U
amongthesmoothinvolutiveoverdistributionsofT
|
U
,whichimpliessmt(ker dF ) ⊇ T|
∞
U
. Foreachm ∈ M
,denec
m
(T
∞
) := corank T
m
∞
.
IfT
isasmoothdistribution,soisT
∞
andtheset
M
∞
reg
ofitsregularpointsisopenanddenseinM
[12] (apointissaidtoberegularifthethedistributionhasconstantrankinaneighbourhood ofit). Thus,Proposition1impliesProposition 2. If
T
isa smooth distribution on amanifoldM
,then each pointm ∈ M
∞
reg
has aneighbourhood inwhich thereexistc
m
(T
∞
)
,but not
c
m
(T
∞
) + 1
,functionally independent rst
integralsof
T
.Proof Inaneighbourhood
U
ofaregularpointm
ofT
∞
thesmoothdistribution
T
|
∞
U
hasconstant rank and, being involutive, denes a smooth regular foliation ofU
with leaves of codimensionc
m
(T
∞
)
; in any set of local coordinates adapted to this foliation, thecoordinates transversal to theleavesarerstintegralsoftherestrictionofT
∞
tothecoordinatedomain.
B. Onthe number of weakly Noetherian constants of motion. We consider now a
non-holonomicHamiltoniansystem
(H, M )
of thetypeconsidered so far and, asarst step, wegive acharacterizationofitsconstantsofmotionasrstintegralsofacertaindistribution onthecon-straintmanifold
M
. (FromnowonweshallconsideronlyobjectswhicharedenedonM
andwe shallthereforeworkexclusivelyinM
,notanymoreinT
∗
Q
Wewilluse thefollowing notation. Given anitenumberof smoothvectorelds
X
1
, . . . , X
d
onM
,hX
1
, . . . , X
d
i
m
orhX
1
(m), . . . , X
d
(m)i
denotes the subspace ofT
m
M
spanned byX
1
(m), . . . , X
d
(m)
. Moreover,hX
1
, . . . , X
d
i
denotesthedistributionwithbershX
1
, . . . , X
d
i
m
. Proposition 3. AsmoothfunctionF : M → R
isaconstantofmotionofanonholonomicsystem(H, M )
ifandonly ifker dF (m) ⊇ ker dH(m) ∩ D
m
ω
∩ T
m
M
∀ m ∈ M .
(8)Proof Afunction
F
onM
isaconstantofmotionof(H, M )
ifandonlyifhdF, X
T M
H
i = 0
atall pointsofM
,that is,ker dF (m) ⊇
X
H
T M
m
∀ m ∈ M .
Notethatthe
T M
componentu
T M
ofavectoru ∈ T
m
(T
∗
Q)
isgivenby(u + D
ω
m
) ∩ T
m
M
. Hence,hX
T M
H
i
m
= (hX
H
i
m
+ D
ω
m
) ∩ T
m
M
. RecallingthatX
H
issymplecticallyorthogonaltoker dH
we thushavehX
T M
H
i
m
= [(ker dH(m))
ω
+ D
ω
m
] ∩ T
m
M = (ker dH(m) ∩ D
m
)
ω
∩ T
m
M
.Remarks: (i) Characterization (8) of constants of motion of nonholonomic systems is in a
sensedual tocharacterizationsgivenin [16,18,39]. Since(5)implies
T
m
(T
∗
Q) = (T
m
M )
ω
⊕ D
m
forall
m ∈ M
,onthe pointsofM
thereisasplittingX = X
T M
ω
+ X
D
ofvectoreldson
T
∗
Q
whichisdual to (6). As provenin[39], afunction
F : M → R
is aconstantofmotionof(H, M )
ifandonlyifforone,andthereforeany,extensionF
˜
ofF
oM
,X
D
˜
F
(H)
|M
= 0 .
(9)Notethatthis conditionisequivalentto
X
D
˜
F
(m) ∈
ker dH(m) ∩ D
m
+ T
m
M
ω
∀ m ∈ M ,
whichcanbecomparedto (8). Areformulationofcondition(9) which, likecondition(8),avoids
theuseofextensionsoffunctionso
M
canbegivenintermsofdistributionalHamiltonianvector elds[15].(ii) A function
F
is a constant of motion of a Hamiltonian vector eldX
H
if and only if eitherX
H
(F ) = 0
or,equivalently,X
F
(H) = 0
. Condition (9) is manifestly an analogueof the latter condition. Instead, (8) is an analogueof conditionX
H
(F ) = 0
, that is,ker dF ⊇ hX
H
i
. In fact, if there are no nonholonomic constraints, thenM = T
∗
Q
,D
= T (T
∗
Q)
and thereforeker dH ∩ D
ω
∩ T M = (ker dH)
ω
= hX
H
i
.Based onProposition 3,wecan nowgivethefollowingcharacterizationof weakly Noetherian
constantsofmotion:
Proposition 4. Fixthe congurationmanifold
Q
,the constraintdistributionD
onQ
,anactionΨ
Q
of aLie group
G
onQ
,andaΨ
T
∗
Q
invariant kineticenergy
T : T
∗
Q → R
. Let
I
G
bethe set of allΨ
Q
invariantsmooth real functionson
Q
. For eachm ∈ M
deneI
m
:=
\
V ∈I
G
ker dT (m) + d(V ◦ π
Q
∗
)(m)
(10)∆
m
:= I
m
∩ D
m
ω
∩ T
m
M .
(11)Then, a smooth function
F : M → R
is a weakly Noetherian constant of motion relative to(T, M, Ψ
Q
)
if andonlyif
Proof A function
F
is a weakly Noetherian constant of motion if and only if, at each pointm ∈ M
,inclusion(8)issatisedforallH = T + V ◦ π
∗
Q
withV ∈ I
G
. Omittingforshortnessthe indicationofthepointm
,thisisequivalentto(ker dF )
ω
⊆ ker d(T + V ◦ π
∗
Q
) ∩ D
+ T M
ω
forallV ∈ I
G
,namely(ker dF )
ω
⊆
\
V ∈I
G
ker d(T + V ) ∩ D
+ T M
ω
.
Sincethetransversalityof
D
ω
and
T M
impliesthatofD
andT M
ω
,thesubspaceT
V ∈I
G
ker d(T +
V ) ∩ D
+ T M
ω
equals(I ∩ D) + T M
ω
.Let
∆
bethedistributiononM
withbers∆
m
⊆ T
m
M
asin(12). Inviewof(7),Proposition4 meansthat,if∆
issmooth,thenafunctionF : M → R
isaweaklyNoetherianconstantofmotionif andonlyifitisarstintegralofthedistribution∆
. Therefore,ifc
m
(∆
∞
)
denotesthecodimension
oftheber
∆
∞
m
inT
m
M
,that is,c
m
(∆
∞
) = dim M − rank ∆
∞
m
= (2n − k) − rank ∆
∞
m
,
then by Proposition 2 in a neighbourhood of each regular point
m
of∆
∞
there existc
m
(∆
∞
)
, but notc
m
(∆
∞
) + 1
, functionally independent weakly Noetherian constants of motion relative to
(T, M, Ψ
Q
)
. From a dynamical point of view, the existence of local constants of motion of a
dynamical systemhasno particular signicancetherectication theorem ensures theexistence
of the maximal numberof them. Thus, this fact cannot be considered as anexistence result of
weaklyNoetherian constantsofmotion,butanupperbound ontheirnumber:
Theorem1. Under the hypotheses of Proposition 2, every setof functionally independent global
weakly Noetherian smoothconstants ofmotion relative to
(T, M, Ψ
Q
)
containsatmost
c(∆
∞
) :=
min
m∈M
c
m
(∆
∞
)
elements.In practice, however, the estimate provided by Theorem 1 may be used as a guide for the
search ofasmanyweakly Noetherianconstantsofmotionaspossible,includinghorizontalgauge
momenta.
Of course, the eectiveness of this approach depends on the possibility of constructing the
distribution
∆
∞
,whichinturnrequirestheknowledgeof
∆
. Aswereviewinthenextsubsection, thereisastandardprocedurefortheconstructionof∆
∞
ifthedistribution
∆
isreal analytic(see e.g. Chapter3of [12] fordetails and references). The constructionofthe distribution∆
willbe doneinSection5undersuitableassumptionsontheactionsΨ
Q
.
C.Constructionof
∆
∞
. Weassumenowthat,asithappensintypicalcases,thedistribution
∆
isrealanalytic. Aswehavealreadymentioned,thereastandardwayofconstructingtheinvolutiveclosure of areal analytic distribution[12]. Since anyreal analytic distribution is locally nitely
generated,in anysuciently small open set
U ⊆ M
∞
reg
there arer = rank ∆
independent vector eldsX
1
, . . . , X
d
such that∆
1
:= ∆
|U
hasbers∆
1
m
=
X
1
, . . . , X
d
m
,that is,∆
1
= hX
1
, . . . , X
d
i .
Dene∆
s+1
:= ∆
s
∪ [∆
s
, ∆
s
] ,
s = 1, . . . , n − 1 ,
where[∆
s
, ∆
s
]
isthedistributionwhosesectionsareallthecommutators
[η, ξ]
forη, ξ
sectionsof∆
s
,thatis,
∆
2
= hX
1
, . . . , [X
1
, X
2
], . . .i
etc. Thus,
∆
∞
= ∆
s
where
s
isthesmallestpositiveintegersuchthat∆
s
isintegrableor,inother
words,thesmallestpositiveintegersuch that
∆
s
= ∆
s+1
.
Thus, in order to apply the criterion of Theorem 1 to a specic nonholonomic system with
analyticdistribution
∆
,itsucestondasystemofgeneratorsof∆
in aneighbourhoodofeach ofitsregularpointsandthencomputesucientlymanyLiebracketsofthesegenerators,checkingforlinearindependence.
This procedure canbeimplementedin twoways. Oneis ofcoursethat ofparameterizingthe
constraintmanifold
M
andwriting thegeneratorsof∆
usingthe chosenlocal coordinates. The other is that of making all computations using extensions oM
of the chosen local generatorsX
1
, . . . , X
d
of∆
. Infact,ifX
˜
1
, . . . , ˜
X
d
areextensionsofX
1
, . . . , X
d
oM
,then[ ˜
X
a
, ˜
X
b
]
|M
= [X
a
, X
b
]
∀ a, b
for a known property of related vector elds, see e.g. [35]. We will use both methods in the
examplesbelow.
5 The distribution
∆
for free and proper actionsA. Determinationof
∆
. Thedetermination ofthedistribution∆
requiresthedetermination ofthedistributionI
asin(10). Thisis aneasy taskiftheactionis properandfree,andwelimit ourselvesto thiscase.As above,wedenoteby
G
thedistributiononQ
whosebersG
q
are thetangentspacesT
q
O
q
to the orbitsO
q
of the actionΨ
Q
. Forour purposes it is sucient to determine
∆
in Darboux coordinates(q, p)
. Thus, weidentify the tangentspacesT
(q,p)
T
∗
Q
withR
2n
(or withR
n
× R
n
,when wewantto stressthe individualroles ofthe
q
andp
componentsof thetangentvectors), weidentifythespacesG
q
= T
q
O
q
andD
q
withsubspacesofR
n
andweidentifythespaces
T
(q,p)
M
,∆
(q,p)
andI
(q,p)
with subspacesofR
2n
. Inthesequel, thedot denotesthe scalarproduct in
R
n
and, if
E
is a subspace ofR
n
, then
E
⊥
denotes its orthogonal complement in
R
n
; ifv ∈ R
n
, thenv
⊥
stands for
hvi
⊥
. We will distinguishbetweenrow and column vectorsonly when some
ambiguitymightarise. Asbefore,wedenoteby
A(q)
thekineticmatrix,sothatthekineticenergy isT (q, p) =
1
2
p · A(q)
−1
p
.
Lemma 1. In addition to the assumptions of Proposition 4, assume that the action
Ψ
Q
is free
andproper. Then, ateachpoint
(q, p) ∈ M
,I
(q,p)
=
(v
q
, v
p
) ∈ R
n
× R
n
: v
q
∈ G
q
, v
p
∈ T
p
0
⊥
−
v
q
· T
0
q
kT
0
p
k
2
T
p
0
(13) whereT
p
0
:=
∂T
∂p
(q, p) = A(q)
−1
p
and
T
0
q
:=
∂T
∂q
(q, p) .
(14)Proof Avector
v = (v
q
, v
p
) ∈ I
(q,p)
ifand onlyifd(T + V )v = 0
forallV ∈ I
G
, namely ifand onlyif∂V
∂q
· v
q
= 0
forallV ∈ I
G
andT
0
q
· v
q
+ T
p
0
· v
p
= 0
. Iftheactionis freeandproper,then thecondition∂V
∂q
· v
q
= 0
∀ V ∈ I
G
is equivalentto
v
q
∈ G
q
(see e.g. Theorem 2.5.10of [35]). Ifp = 0
thenT
0
p
= T
q
0
= 0
and there arenoconditionsonv
p
; thisisconsistentwith (13)becauseinthiscaseT
0
p
⊥
= R
n
andT
0
q
= 0
. Ifp 6= 0
thenv
p
∈ T
0
p
⊥
− kT
0
p
k
−2
(v
q
· T
q
0
)T
p
0
.Lemma 2. Under the hypotheses and with the notation of Lemma 1, at any point
(q, p) ∈ M
a vector(u
q
, u
p
) ∈ R
n
× R
n
belongstoI
(q,p)
∩ D
(q,p)
ω
if andonlyifu
q
∈ hT
p
0
i
and
u
p
∈ [D
q
∩ G
q
]
⊥
−
u
q
· T
p
0
kT
0
p
k
2
T
q
0
(15) withT
0
p
andT
0
q
asin (14).Proof Formulas(4)and(13)imply
I
(q,p)
∩ D
(q,p)
=
(v
q
, v
p
) ∈ R
n
× R
n
: v
q
∈ D
q
∩ G
q
, v
p
∈ T
p
0
⊥
−
T
0
q
· v
q
kT
0
p
k
2
T
p
0
.
(16) Thus, a vector(u
q
, u
p
) ∈ R
n
× R
n
is symplectically orthogonal toI
(q,p)
∩ D
(q,p)
if and only ifu
q
· v
p
= u
p
· v
q
forallv
q
∈ D
q
∩ G
q
andallv
p
∈ T
0
p
⊥
− (T
0
q
· v
q
)kT
p
0
k
−2
T
p
0
,that is,ifandonlyifu
q
·
T
p
0
⊥
−
v
q
· T
0
q
kT
0
p
k
2
T
p
0
= u
p
· v
q
∀v
q
∈ D
q
∩ G
q
.
(17)When
v
q
= 0
thisequalityreduces tou
q
· T
0
p
⊥
= 0
,whichgivesu
q
= λT
0
p
forsomeλ ∈ R
. Forthis valueofu
q
,(17)becomes(λT
0
q
+ u
p
) · v
q
= 0
forallv
q
∈ D
q
∩ G
q
. Thisshowsthat[I
(q,p)
∩ D
(q,p)
]
ω
= {(u
q
, u
p
) : u
q
= λT
p
0
, u
p
∈ [D
q
∩ G
q
]
⊥
− λT
q
0
, λ ∈ R} .
Expressing
λ
in termsofu
q
andT
0
p
leadsto(15).Wecannowcompletethecharacterizationof
∆
. Notethat,sinceA
isaninvertiblen×n
matrix andS
isak × n
matrixwithrankk
,SA
−1
S
T
isaninvertible
k × k
matrix. Notealsothat, ifthe vectorsubspaceD
q
∩ G
q
⊥
ofR
n
is
k
dimensionaland hasabasisformedbyvectorsw
1
, . . . , w
k
ofR
n
, then the vector subspace
I
(q,p)
∩ D
(q,p)
ω
ofR
2n
is
(k + 1)
dimensionaland has abasis formedbythevectors(0, w
1
), . . . , (0, w
k
), (T
0
p
, −T
q
0
)
ofR
2n
.
Proposition 5. UnderthehypothesesandwiththenotationofLemma1,atanypoint
(q, p) ∈ M
avector(u
q
, u
p
) ∈ R
n
× R
n
belongsto
∆
(q,p)
if andonlyif itsatises(15) andu
p
∈ A(q)D
q
− S(q)
T
S(q)A(q)
−1
S(q)
T
−1
K(q, p)u
q
(18)where
K(q, p)
isthe matrixwith componentsK(q, p)
ai
=
∂
∂q
i
S(q)A(q)
−1
p
a
,
a = 1, . . . , k ,
i = 1, . . . , n .
Proof Recallthat
∆
m
= [I
m
∩ D
m
]
ω
∩ T
m
M
. ByLemma2,(u
q
, u
p
) ∈ [I
m
∩ D
m
]
ω
ifandonlyif
u
q
andu
p
satisfy(15). SinceM
is(locally)thezerolevelsetofthek
functionsˆ
τ
a
= (S(q)A(q)
−1
p)
a
, avector(u
q
, u
p
) ∈ R
n
isinT
(q,p)
M
ifandonlyifu
q
·
∂ ˆ
τ
a
∂q
(q, p) + u
p
·
∂ ˆ
τ
a
∂p
(q, p) = 0
∀a = 1, . . . , k ,
thatis,ifandonlyif
S(q)A(q)
−1
u
p
= −K(q, p)u
q
.
(19)Given that
D
q
= ker S(q)
it is immediate to verify that this condition is equivalent tou
p
∈
AD − S
T
(SA
−1
S
T
)
−1
Ku
q
.Corollary 1. If
G
actsfreely andproperly onQ
andG
isan overdistributionofD
,then∆
has rankoneatallpointsofM
,except onthe sumbmanifoldp = 0
ofM
whereithas rankzero.Proof We shall provea little bit more, that is, that ifthe action is freeand proper, then the
distribution
∆
satisesdim ∆
(q,p)
= 1 + n − k − dim(G
q
∩ D
q
if p 6= 0
dim ∆
(q,0)
= n − k − dim(G
q
∩ D
q
.
This implies the statement because, if
G
q
∩ D
q
= D
q
, thendim(G
q
∩ D
q
) = n − k
. Letm =
(q, p)
. Notethatdim ∆
m
= 2n − dim ∆
ω
m
. From (11)andthe transversalityofT
m
M
andD
ω
m
it followsthat∆
ω
m
= I
m
∩ D
m
⊕ (T
m
M )
ω
. Hencedim ∆
ω
m
= dim(I
m
∩ D
m
) + dim[T
m
M ]
ω
, wheredim[T
m
M ]
ω
= k
. From(16)itfollowsthat,ifp 6= 0
,thendim(I
m
∩ D
m
) = dim (G
q
∩ D
q
) + (n − 1)
becauseT
0
p
⊥
is a subspace ofR
n
of dimension
n − 1
. If insteadp = 0
, thenT
0
p
⊥
= R
n
anddim(I
m
∩ D
m
) = dim (G
q
∩ D
q
) + n
. Hence,iftheactionΨ
Q
on
Q
is freeandproper,and ifthe grouporbitsaresucientlylarge, sothatG
⊇ D
,thenthedistribution∆
hasrankoneatallpointsexceptatp = 0
,whereitsrank iszero. Thisimpliesthat∆
isintegrableand∆
∞
= ∆
. Therefore,itfollowsfromthestatement
at the beginning of the proofof Theorem 1 that, in this case,givenany
Ψ
T
∗
Q
invariantkinetic
energy
T
onT
∗
Q
,inaneighbourhoodofeachpointof
M
thereisthemaximumnumber2n − k − 1
offunctionallyindependentlocalweaklyNoetherianconstantsofmotion. Thus,inthiscase,whichincludesthecaseoftransitiveactions,ourmethoddoesnotgiveanynewinformationwithrespect
to therectication theorem, but thefact that the local constantsof motioncanbechosen tobe
weaklyNoetherian. Thisis asortofweaklyNoetherian versionoftherecticationtheorem.
If
G
isnotanoverdistributionofD
,instead,thedistribution∆
hasrankgreaterthanoneand maybenotintegrable.Thisputssomelimitationsonthenumberof(evenlocal)weaklyNoetherianconstantsofmotion.
Next,wehave:
Corollary 2. Given
Q
,M
andT
,consider twofree andproper actionsΨ
Q
1
andΨ
Q
2
of twoLie groupsG
1
andG
2
onQ
, whose tangent lifts preserveT
. If the orbits of theΨ
Q
1
action contain those of theΨ
Q
2
action, then any weakly Noetherian constant of motion relative to(T, M, Ψ
Q
2
)
is also aweakly Noetherianconstant ofmotion relativeto(T, M, Ψ
Q
1
)
.Proof Accordingto (13)and (18),thefact that avectortangentto
M
isin∆
depends onthe groupactiononlythroughthetangentspaceto thegrouporbits.Inparticular, iftwodierent groupactionshavethe sameorbitsin
Q
(but possibly dierent orbitsinT Q
),thentheyhavethesameweaklyNoetherianconstantsofmotion.6 Example: the rolling vertical disk.
A.Thesystem. Asarsttestcaseforthemethod,andasanillustrationofit,weconsiderhere
thesystemformedbyadiskwhichisconstrainedtorollwithoutslippingonahorizontalplane,and
tostandvertically. Thisisawellknownsystem,whichhasbeenconsiderede.g. in [16,8,13,23].
Weshallconsidervariousclassesofforcesactingonthedisk,withvarioussymmetrygroups,soas
toshowhowchangingthesymmetrygroupchanges thedimensionof
∆
∞
andthusthebound on
constantsof motion,whichin theconsideredcasescanbedeterminedbyanalyzingtheequations
ofmotion.
The holonomic system has conguration manifold
Q = R
2
× S
1
× S
1
3 (x, y, θ, ϕ)
, where
(x, y) ∈ R
2
are Cartesiancoordinatesof the point of contact,
ϕ
is theangle between thex
axis andtheprojectionofthedisk ontheplane,andθ
isthe anglebetweenaxedradiusof thedisk and the vertical. In orderto simplify thenotation, weassume that massand radius of the diskare both unitary. The nonholonomic noslipping constraint imposes that the point of the disk
in contactwith the plane has zero velocity, that is,
˙x = ˙θ cos ϕ
and˙y = ˙θ sin ϕ
. The ranktwo constraintdistributionhasbersD
(x,y,θ,ϕ)
=
cos ϕ ∂
x
+ sin ϕ ∂
y
+ ∂
θ
, ∂
ϕ
thatis,
ker S
(x,y,θ,ϕ)
forS
(x,y,θ,ϕ)
=
cos ϕ
sin ϕ
−1 0
− sin ϕ cos ϕ
0
0
.
The(Lagrangian) kinetic energy is
T =
1
2
˙x
2
+ ˙y
2
+
1
2
I ˙θ
2
+
1
2
J ˙
ϕ
2
, where
I
andJ
are the appropriatemomentsofinertia. Ifwewriteh = 1/I
,k = 1/J
andifV (x, y, ϕ, θ)
is thepotential energyoftheforcesactingonthedisk,thentheHamiltonianisH(x, y, ϕ, θ, p
x
, p
y
, p
θ
, p
ϕ
) =
p
2
x
+ p
2
y
2
+
h
2
p
2
θ
+
k
2
p
2
ϕ
+ V (x, y, ϕ, θ) .
Theconstraintmanifold
M =
(x, y, θ, ϕ, p
x
, p
y
, p
θ
, p
ϕ
) : p
x
= kp
θ
cos ϕ , p
y
= kp
θ
sin ϕ
is sixdimensional. It is dieomorphic to
R
4
× T
2
and can be globally parameterized with
(x, y, θ, ϕ, p
θ
, p
ϕ
)
. Usingthesecoordinates,theequationsofmotion,whichcanbecomputedfrom (2)and(3),are˙x = hp
θ
cos ϕ ,
˙y = hp
θ
sin ϕ ,
˙θ = hp
θ
,
ϕ = kp
˙
ϕ
,
(20)˙p
θ
= −
1
1 + h
V
0
θ
+ V
x
0
cos ϕ + V
y
0
sin ϕ
,
˙p
ϕ
= V
ϕ
0
.
Theseequationsaresimpleenoughtoallowustointerprettheresultsoftheforthcominganalysis,
whereincreasinglysmallersymmetrygroupsareconsidered. Forsimplicityofexposition,weclassify
thesegroupsviathecorrespondingclassesofinvariantpotentials.
B. No forces. Ifweallowonlyconstantexternalpotentials,thenthesystemisinvariantunder
thegroup
G = R
2
× T
2
,whichactson
Q
bytranslationsalongthecoordinates(x, y, θ, ϕ)
. Sincethisactionistransitive,ourmethodgivestheexistenceofsetsofveweaklyNoetherianlocal constants of motion in a neighbourhood of any point of
M
, see Corollary 1. As we now show,onlyfouroftheseconstantsofmotionaregloballydened;twoofthemcanbechosentobehorizontalmomentaandtheothershorizontalgaugemomenta.
Inordertosee thisobservethat,forconstantpotentials,theequationsofmotion(20)become
˙x = hp
θ
cos ϕ ,
˙y = hp
θ
sin ϕ ,
˙θ = hp
θ
,
ϕ = kp
˙
ϕ
,
˙p
θ
= 0 ,
˙p
ϕ
= 0 .
(21) Thusp
θ
andp
ϕ
are constants of motion, and they are obviously horizontal momenta. Observe now thateachsubsystem(θ, p
θ
)
and(ϕ, p
ϕ
)
performsuniform rotationsonS
1
× R
,with angular
frequencies
hp
θ
andkp
ϕ
respectively. Sinceϕ
grows linearly in time andp
θ
is constant, inte-gratingthersttwoequations(21)showsthattheprojectionof themotionin the(x, y)
plane isgenericallya uniformcircular motion, with frequency
kp
ϕ
. (The motion is, exceptionally, linear ifp
ϕ
= 0
). Thus,motionsofthesystemare genericallyquasiperiodicontori ofdimensionthree, withfrequencieshp
θ
, kp
ϕ
, kp
ϕ
. Sincetwoofthefrequenciesareequaltheclosureofalltheseorbits iscontainedintoriofdimensiontwo. Butsincetheratiohp
θ
/(kp
ϕ
)
variescontinuously,thesetof orbitswhoseclosureis atwodimensionaltorusforms adenseset. Therefore,thereare notmorethan
6 − 2 = 4
independentconstantsofmotionwhich aredenedinopeninvariantsets.Asimplecomputation showsthat twootherglobally dened constantsof motionare
kxp
ϕ
−
hp
θ
sin ϕ
andkyp
ϕ
+ hp
θ
cos ϕ
. They are horizontal gauge momenta because, if the action is transitive,thenanyglobalconstantofmotionwhichislinearinthevelocitiesisahorizontalgaugemomentum,seeProposition4of [22].
Remark: The fthlocal integralgivesthe slope ofthe orbitsonthe twodimensionaltori.
It can be taken e.g. as
hp
θ
ϕ − kp
ϕ
θ
and it is not globally dened on the tori with irrationalkp
θ
/(hp
ϕ
)
. Hence,itis notagaugemomentum eventhoughit islocally alinearcombination of twohorizontalmomenta.C. Potentials
V = V (ϕ)
. Weallow nowpotentialenergieswhich depend ontheangleϕ
, thus restrictingthesymmetrygrouptoG = R
2
× S
1
,whichactsbytranslationsalong
x
,y
andθ
. We beginbydeterminingthedistribution∆
:Fact 1 The bers of
∆
are spannedby thetwovectoreldsX
1
(q, p) = ∂
p
ϕ
(22)X
2
(q, p) = p
x
∂
x
+ p
y
∂
y
+ hp
θ
∂
θ
+ kp
ϕ
∂
ϕ
− kp
ϕ
p
y
∂
p
x
+ kp
ϕ
p
x
∂
p
y
.
(23)Proof The bersof
∆
are thesubpacesof vectors(u
q
, u
p
) ∈ R
n
× R
n
which satisfyconditions
(15)and(18). Firstnotethat,atapoint
q = (x, y, θ, ϕ) ∈ Q
,thedistributionoftangentspacesto theorbitsofthegroupactionhasberG
q
=
∂
x
, ∂
y
, ∂
θ
and hence
D
q
∩ G
q
=
cos ϕ∂
x
+ sin ϕ∂
y
+ ∂
θ
.
Considernowapoint
p = (p
x
, p
y
, p
θ
, p
ϕ
)
suchthat(q, p) ∈ M
andx avectoru
q
∈ R
4
suchthatu
q
∈
T
p
0
= hA
−1
pi
, thatis,u
q
= λ p
x
, p
y
, hp
θ
, kp
ϕ
(24)forsome
λ ∈ R
. Since thekineticmatrixA
isconstant,thevectorT
0
q
:=
∂T
∂q
= 0
andthe second condition(15)reducesto theorthogonalityofu
p
toD
q
∩ G
q
, thatis,u
p
=
α, β, −α cos ϕ − β sin ϕ, γ
(25)forsome
α, β, γ ∈ R
. Simplecomputationsshowthat,onthepointsofM
,K(q, p) =
0
0 0
p
y
cos ϕ − p
x
sin ϕ
0
0 0
−p
x
cos ϕ − p
y
sin ϕ
=
0 0
0
0
0 0
0 −hp
θ
S
T
(SA
−1
S
T
)Ku
q
= λ hkp
θ
p
ϕ
sin ϕ , −hkp
θ
p
ϕ
cos ϕ , 0 , 0
T
= λ kp
y
p
ϕ
, −kp
x
p
ϕ
, 0 , 0
T
,
wherethelastexpressionsfollowfrom theconstraintequations