• Aucun résultat trouvé

Geometric degree of nonconservativity: Set of solutions for the linear case and extension to the differentiable non-linear case

N/A
N/A
Protected

Academic year: 2021

Partager "Geometric degree of nonconservativity: Set of solutions for the linear case and extension to the differentiable non-linear case"

Copied!
13
0
0

Texte intégral

(1)

HAL Id: hal-01321974

https://hal.archives-ouvertes.fr/hal-01321974

Submitted on 13 Nov 2019

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Geometric degree of nonconservativity: Set of solutions for the linear case and extension to the differentiable

non-linear case

Jean Lerbet, Noël Challamel, François Nicot, Félix Darve

To cite this version:

Jean Lerbet, Noël Challamel, François Nicot, Félix Darve. Geometric degree of nonconservativity: Set

of solutions for the linear case and extension to the differentiable non-linear case. Applied Mathemat-

ical Modelling, Elsevier, 2016, 40 (11-12), pp.5930–5941. �10.1016/j.apm.2016.01.030�. �hal-01321974�

(2)

Geometric degree of nonconservativity: Set of solutions for the linear case and extension to the differentiable non-linear case

J. Lerbet

a,

, N. Challamel

b

, F. Nicot

c

, F. Darve

d

aIBISC, UFRST-UEVE, 40, rue du Pelvoux CE 1455 91020 Evry Courcouronnes cedex, France

bUniversité Européenne de Bretagne, Université de Bretagne Sud LIMATB -UBS -Lorient Centre de Recherche Rue de Saint Maudé - BP 92116 56321 Lorient cedex, France

cIRSTEA, ETNA – Geomechanics Group, Domaine universitaire, 38042 Saint Martin d’Heres, France

dLaboratoire Sols Solides Structures, UJF-INPG-CNRS, BP 53 38041 Grenoble cedex 9, France

Thispaperdealswithnonconservativemechanicalsystemsasthosesubjectedtononcon- servativepositionalforcesandleadingtonon-symmetrictangentialstiffnessmatrices.Ina previouswork,thegeometricdegreeofnonconservativityofsuchsystems,definedasthe minimalnumberofkinematicconstraintsnecessarytoconverttheinitialsystemintoa conservativeoneisfoundtobe,inthelinearframework,thehalfoftherankoftheskew- symmetricpartofthestiffnessmatrix.Inthepresentpaper,newsresultsarereached.First, amoreefficientsolution oftheinitiallinearproblemisproposed.Second,alwaysinthe linearframework,theissueofdescribingthesetofallcorrespondingkinematicconstraints isgivenandreducedtotheoneoffindingalltheLagrangianplanesofasymplecticspace.

Third,theextensiontothelocalnon-linearcaseissolved.Afourdegreeoffreedomsystem exhibitingamaximalgeometricdegreeofnonconservativity(s=2)isusedtoillustrateour results.Theissueoftheglobalnon-linearproblemisnot tackled.Throughoutthepaper, theissueoftheeffectivinessofthesolutionissystematicallyaddressed.

Introduction

Nonconservativeelasticmechanicalsystemsexhibitseveralparadoxicalmechanicalbehaviors.Destabilizingeffectbyad- ditionalfriction iscertainlythe mostfamousparadox ofthesemechanicalsystemsandhasbeen deeplyinvestigated(see [1–3]forexample).Onelessreportedparadoxicaleffectisthedestabilizing effectbyadditionalkinematicalconstraints.J.J.

Thompsonmentionedthiseffectin[4]but,tothebestofourknowledge,thisparadoxicaleffecthadneverbeensystemat- icallyinvestigatedbeforerecently.Thisparadoxicaleffectledtotheso-calledkinematicalstructuralstability(ki.s.s.)issue:

whenandhowisitpossibletodestabilizebyaddingkinematicalconstraint(s)agivenstablesystem?

Duringthelastfiveyears,inasequenceofpapers([5–9]),weelucidatedthiskinematicalstructuralstability(ki.s.s.)issue forthelineardivergencestabilityofbothconservativeandnonconservativeelasticsystemsaswell.Abigpartoftheseworks arealsorelatedtotheso-calledsecondorderworkcriterionintroducedbyHillintheframeworkofplasticityin1958([10])

Corresponding author. Tel.: +33 169477503.

E-mail address: jlerbet@gmail.com (J. Lerbet).

(3)

andindependentlyintroducedandusedintheframeworkofelasticnonconservativesystemsin2004([11]).Themainresult involvesthe symmetric partKs ofthe stiffnessmatrix andthemagnitude ofthe loadparameter aswell but itdoes not dependonthenumberoftheadditionalkinematicconstraints.

Bydualitytotheki.s.s.issue,weinvestigatedin[12]theissuetoconvertby(judicious)additionalkinematicconstraintsa nonconservativesystemintoaconservativeone.Thisissueleadstotheconceptofgeometricdegreedofnonconservativity of.Calculationsshowthatd=sisthehalfsoftherankroftheskew-symmetricpartKa(p)(thatisalwaysevenr=2s).

Inasecondstage, abuildingofthejudicious additionalkinematicconstraintsC1,...,Cs(Rn)s hasbeenproposed thanks totheeigenspacesEλ2

i,i=1,...,softhesymmetricmatrixKa2(p)whosetheeigenvalues−

λ

21,...,

λ

2s arealldouble:each Cimaybe chosenineachdistinct Eλ2

i.Itisworthnotingthat,forbothissues,themechanicalsystem isapproximated byitslinearfirstorderapproximationatagivenequilibriumconfigurationqe.Thatmeansthatisdescribedbythemass matrixMandthestiffnessmatrixK.Ifpisaloadparameter,thenK=K(p).Thenon-symmetryofK(p)(namelyK=Ksor Ka=0)isthenthesignatureofthenon-conservativenatureofthemechanicalsystem.Inourpreviousworks,thesource of the nonconservativity lies in external forces like follower forces actingon elastic system. Hypoelasticity mayalso be anothermechanicalframework leadingtoasimilar mathematicalproblem.Thereexists abroadliteraturecoveringhypoe- lasticity(seeforexample[13–16]).

Inthispaperwe are concernedby finding thecomplete solutionofthe linearcaseandby thegeneralization andthe extensiontothenon-lineardifferentiablecaseabouttothelatterissue.Wethenusethelanguageofanalyticmechanics.Ina firsttime,wereinvestigatethelinearcasebyusingthelanguageofexteriorp-formsandespeciallyexterior1-and2-forms.

Thatallow usto moredeeply highlightthe issueof effectivenessof thecalculation of thesuitable kinematic constraints converting the systeminto a conservative one. Thatalso allow us to investigatethe issueof building the set of all the solutions andto illustrate the geometrical meaning of thesesolutions. To do it, the language of symplectic geometry is systematicallyused.Thatalsosuggeststhegoodwayfortacklingthenon-linearcase.

Thus,inasecond time,we tacklethenon-linearproblemwithappropriate notationsandespeciallythanksto thelan- guageofdifferentialp-forms.Weaccuratelyfocusonthelinkwiththelinearcase.Inathirdstep,thesolutionisproposed byextendingtothenonlinearcasetheconceptofgeometricdegreeofnonconservativityandyieldingageometricmeaning tothecorrespondingnon-linearconstraints.Inthelastpart,theissueofthecalculationoftheappropriatenon-linearcon- straintsisinvestigated.Theproblemofaglobalsolutioninrelationship withthetopologyoftheconfigurationmanifoldis onlyevokedbyjustsettingtheconvenientgeometricframeworkofvectorbundles.Afourdegreeoffreedomsystemcalled theBigonisystem(see[12,17])iscontinuouslyusedthroughoutthepapertoillustratethegeneralresults.

1. Thelinearcase

Inwhatfollowswe referto[12].Weonlyrecall thatforthelinearframework,dynamicequation oftheunconstrained systemread:

MX¨+KX=0, (1)

withKany(namelynon-symmetric)matrixandMsymmetricpositivedefinite.Kisthestiffnessmatrixofthesystemand Mitsmassmatrix.Becauseofthenonconservativityofthe positionalforcesactingon

σ

,K isany.Theminimumnumber

ofkinematicconstraintsallowingtoconvertthesystemintoaconservative one(withacorrespondingsymmetricstiffness matrix)isthegeometricdegreeofnonconservativityof.(1)isdeducedfromtheLagrange equationbytheusualprocess oflinearizationaboutanequilibriumconfiguration.

1.1. Effectivinessofthesolutionproposedin[12]

Inintroduction,we alreadyrecalledthealgebraicmeaningofthe geometricindexordegree ofnonconservativity:this thehalf s ofthe rankr=2s ofKa andthe distinct constraints,viewed asvectorsof Rn, can be chosen in thes distinct eigenspacesEλ2

i,i=1,...,sof Ka2. We now questionthe effectiveness of the buildingof the constraintsas proposed in [12].Todoit,we usethespectral theoremforKa2.Whatdoesmeantheeffectivenessforthespectral theorem?Theusual proof isdone by inductiononthe dimensionof thespace.Forinitializing theinductionreasoning, theD’AlembertGauss theoremis usedforfinding an eigenvalueof thecharacteristicpolynomial of Ka2 andthis theoremisnot effectivein the sensewhereonlyanumericalmethodmayleadto(anapproximationof)theeigenvalues.So,withthesetools,thesolution ofthe linear caseitself is not effective.Remark howeverthat theconstraints are also the criticalpointsof the Rayleigh quotientRassociatedwithKa2 andthat onlythe eigenspacesareinteresting andnot theeigenvalues−

λ

2i,i=1,...,s.The useofRayleigh quotientisthen especiallyrelevant andtheconstraints maybe evaluated bysuccessive minimizationsof R(X)=−XXTKT2aXX.ByMinimaxtheorem,theconstraintsarealsothesolutionsof

dimminF=k max

XF\{0}R

(

X

)

,

fork=1,...,navoiding by thiswaythe useof D’alembert–Gausstheorem.However, thisminimization process gives no analyticexplicitresult.

(4)

1.2. Thelanguageofexteriorp-forms

Now,before addressing ina followingstep thenon-linear case, we still focuson thelinear casewiththehelp ofthe exterior p-forms:welookthematrixKa(skewsymmetricpartofthestiffnessmatrix) nolongerasthe matrixofa linear map ofRn butasthematrixof anexterior 2-formon Rn.We indifferentlynote E=Rn andE its dualspace, thevector spaceofthelinearformsonE.Thus,let

φ

theexterior2-formdefinedonE=Rn by:

φ (

u,

v )

=uTKa

v

, (2)

afteridentifyingavectoru=(u1,...,un)ofRnwiththecolumnvectoru=

u1

.. . un

⎠ofMn1(R).Thankstoabasictheoremof

linearalgebra(see[18]forexample),thereisabasisB=(e1,...,en)ofRnandanumberr=2snsuchthat

φ

(e2i1,e2i)=

φ

(e2i,e2i1)=1forisand

φ

(ei,ej)=0fortheothervaluesofiandj.Inthedualbasis(e1,...,en)of(e1,...,en),the form

φ

reads:

φ

=e1e2+...+e2s1e2s. (3) According to [12], we have to find out a subspace H of Rn such that

φ

(u,

v

)=0

u,

v

H and the constraints, viewed now aslinear formsC1,...,CsE, then belong to H, theorthogonality being then understoodin the sense of duality.

ChoosingeachconstraintCiinthesubspace<e2i1,e2i>spannedbye2i1 ande2iinEleadstothewantedresult.Indeed, suppose tosimplifythat Ci=e2i1 forall i=1,...,s andthat Gis thevector subspacespannedby (e2i1)1is. Letu,

v

H=Gwherethebidualisidentifiedwiththespaceitself.Thus,byuseof(3),ifu,

v

H,e2i1e2i(u,

v

)=e2i1(u)e2i(

v

)e2i(u)e2i1(

v

)=0−0=0andthus

φ

(u,

v

)=0.

IfCiisanyin<e2i−1,e2i>asimilarproofashereafterfordifferentialformsmaybeusedandisnotreproduced.

The effectiveness ofthe building ofthe constraints isnow brought back to theone ofthe basis B=(e1,...,en).The proof isagaindone byinductiononthedimensionnofE (seeforexample[18]pp 30–31).Thisproofiseffectiveandthe following paragraph will highlighthow it is performing on an example.Before dealingwiththe example,an interesting issueistocharacterizeallthesolutions.

1.3. Setofsolutions

For describing the set of solutions, usual concepts of symplectic geometry are used. To do it, we first brought back theissueinthe usualframework ofsymplectic geometry.Theexterior 2-form

φ

doesnotnecessarily definea symplectic structureonRnbecauseithasnotnecessarilyamaximalranknamely

φ

maybedegenerate.Forinstance,thatnecessarily occurswhen n isodd. Letthen Fbe the kernelof

φ

.Then (Rn/F,

φ

˜) isa 2s-dimensionalsymplectic vector space where

φ

˜ is canonically defined by

φ

˜(u¯,

v

¯)=

φ

(x,y) withx (resp, y) anyvector ofthe class u¯ (resp.

v

¯). Remark that thanks to the canonical scalar product (.

|

.) on Rn, one could choose the orthogonal F of F for the scalar product as “canonical”

supplementary spaceof F inRn and

φ

F the restriction of

φ

to F. Then (F,

φ

F) is also a 2s-dimensional symplectic vector spacesandthere arethree possiblemeanings fororthogonalityinF:duality, scalarproduct,and

φ

-orthogonality.

Howeverthescalarproducthasnomeaningon(Rn/F,

φ

˜)andonly

φ

-orthogonalityandorthogonalityfordualitykeepuseful on(Rn/F,

φ

˜).Moreover,whengeneralizingthereasoningfromvectorspacestomanifolds,thenaturalEuclideanstructureof Rn does not exist in the tangent and cotangent spaces. It is then more judicious to avoid the use of this structure. The orthogonalforthedualityofanysubspaceGwillthedenotedbyG.

If Wis any subspace of Rn/F the

φ

-orthogonal or symplectic orthogonal of W isthe vector subspace ofRn/F noted Wφ definedby:u¯∈Wφ⇐⇒

φ

˜(u¯,

v

¯)=0

v

¯W.Because

φ

˜ isnon-degenerate,themap

φ

:Rn/F(Rn/F)definedby

φ

(u¯)(

v

¯)=

φ

˜(u¯,

v

¯)

u¯,

v

¯∈Rn/F isanisomorphism(canonical)andthen

φ

(W)=Wφ foranyWsubspaceofRn/F.

A subspaceLof Rn/F iscalledLagrangian ifL=Lφ.It isalso oftencalleda Lagrangian plane eventhough dimL=s. Astraightforwardcalculationshowsthatthedualbasis ofanybasisofanyLagrangiansubspaceisasolutionoftheinitial problem. Moreaccurately,ifL1 isa Lagrangianplane then thereisa Lagrangiansupplementary spaceL2 ofL1 inRn/F.If (e2i1)1is(resp.(e2i)1is)isanybasisofL1(resp.L2),thenthedualbasis(e2i1)1isof(e2i1)1isinL1isafamilyof constraintssolutionofthemechanicalissue.Infactthisprocessrealizesanysolution.Thus,by thisprocess ,thesetofall thesolutionsoftheproblemisinabijectiverelationshipwiththesetofthebasesofallLagrangianplanesofRn/Fbutthe geometricalmeaning ofthesolutionsliesin thesetofLagrangianspaces noted(

φ

) of(Rn/F,

φ

˜).(

φ

) hasbeendeeply

investigatedespeciallyinrelationshipwiththetheoryofMaslovindex(see[19]forexampleforahighlightingpresentation oftheconstruction ofthisindex).(

φ

)isa s(s+21) dimensionalsubmanifoldoftheGrassmanniannmanifoldofalls-planes ofRn/FandcalledtheLagrangianGrassmanniannof(Rn/F,

φ

˜).

Let (s) the Lagrangian Grassmanniann manifold of the usual R-symplectic vector space (Cs,

ω

) with its canonical symplectic structure. This manifold maybe explicitly describedby andidentified with the set Us(s) of unitary symmet- riccomplexmatricesofMs(C).ALagrangianplaneL(s)isidentifiedwiththematrixULUs(s)bythefollowingway:

xL←→x=ULc(x)where c(x) is theconjugatecolumn vectorof x∈Cs.Ifu isa symplectomorphismfrom (Cs,

ω

) onto (Rn/F,

φ

˜),then(

φ

)=u((s))whichachievesthecompleteandexplicitdescriptionofthesetofsolutionsnamely(

φ

).

Notealsothat thereisanexplicitrepresentationofmatricesofUs(s).IfUbelongstoUs(s),thenU=X+iY withX,Ytwo

(5)

Fig. 1. n d.o.f. Bigoni system.

realsymmetric matrices ofsize s. Because of U is unitary, then X2+Y2=Is andXY=YX.Thus there isa commonba- sisofdiagonalizationof XandYandaR-orthogonal matrixO sothat X=OTDiag(xi)O andY=OTDiag(yi)O.Wededuce thatx2i +y2i =1foralli=1,...,sandwe mayparametrizethe problemby xi=ricos

α

i andyi=risin

α

i.Wededucethat U=OTRVOwithR=diag(ri)andV=diag(eiαj).Remarkalsothat similartoolsliketheLagrangian planesmaybe usedto investigateother objects like thespectrum of Hamiltoniansystems (see forexample [20]). Although some developments seemtobe closetoeachother,theoriginalandunexpectedfactofourowninvestigations liesintheuseofthesetoolsof symplecticgeometryandalgebrainanon-hamiltonianframework.

1.4.Theexample

Aspreviouslyannouncedintheintroduction,weillustratetheseresultswithafourdegreeoffreedomsystem.Wetackle thefourdegree offreedom non-linearZieglersystemwithcompletefollowerforcesateach jointlikeinFig.1withn=4. ThissystemiscalledBigonisystemin[12]becauseoftheexperimentaldevice proposedby thisauthorin[17].Thissame casehasbeenhandledin[12]inthe linearframeworkandisalso usedhereinordertocompare bothlinearapproaches (andalso to illustrate hereafter the non-linear case).In this section, namelyin the linearcase, the linearization isdone abouttheuniqueequilibriumposition

θ

=0R4.

Theusualcaseofauniquefollowerforceattheextremity(namelytheusualZieglersystem)isnotveryinterestingsince thegeometricdegreeofnonconservativityisthenreducedto1(see[12])withanobvioussolution

θ

4=0.Inthiscase,the directionoftheexternalforceremainsconstantandthisforcebecomesconservative!.Accordingtothepreviousnotations,it meansthatdimF=2,s=1andthattheLagrangianspacesareonedimensionalsubspaceofthetwodimensionalsymplectic space(R4/F,

φ

˜).FindingtheonedimensionalLagrangiansubspacesofthissymplecticspaceisequivalenttofindthesetof linearkinematicconstraintssuchthatwhenthesystemundergoesoneoftheseconstraints,itbecomesconservative.

Theforcesystemisthennowsetupby p=(p1,...,p4)(see Fig.1).TheskewsymmetricmatrixKa(p)reads(see[12]) (obviouslyalltheelastictermshavingasymmetricinputinthesystemarenotinvolvedinthismatrix):

Ka

(

p

)

=1 2

⎜ ⎜

0 p2 p3 p4

−p2 0 p3 p4

−p3 −p3 0 p4

−p4 −p4 −p4 0

⎟ ⎟

,

(6)

meaningthatif(

1,...,

4)isthecanonicalbasisofR4 then

φ

=1

2

(

p2

1

2+p3

1

3+p3

2

3+p4

1

4+p4

2

4+p4

3

4

)

.

Here,det(Ka(p))=p22p24showingthat

φ

isnotdegeneratewhenp2p4 =0whichisnowsupposed.Thus,withtheprevious notations,F=

{

0

}

andR4/F=R4and(R4,

φ

)becomesafourdimensionalsymplecticspace.

Wewanttofindabasis(e1,...,e4)sothat

φ

=e1e2+e3e4. (4)

Letuschoose

e1=

1,e2= p2

2

2,e3=

1+

2+

3,e4= p3

2

3+p4

2

4. (5)

Then(4)holdsandforexampletheconstraintsx1=0,x1+x2+x3=0convertthesystemintoaconservativeone.

WefocusnowonthesetofallsolutionsnamelyhereonthesetofLagrangianplanes.

Let

J=

⎜ ⎜

0 1 0 0

−1 0 0 0

0 0 0 1

0 0 −1 0

⎟ ⎟

,

the matrix of the R-symplectic four dimensional vector space (C2,

ω

) in its canonical basis

v

1=(1,0),

v

2=(i,0),

v

3= (0,1),

v

4=(0,i).Let A be thematrix ofa symplectomorphismu from(C2,

ω

) onto (R4,

φ

)in their respectivecanonical basis. Thenthe relation

ω

(x,y)=

φ

(u(x),u(y))forall x,y∈C2 leadstothe usualrelationJ=ATKaA.ButifQ denotesthe change-of-basismatrixtopassfrom(

i)to(ei)thenP=(QT)−1isthe correspondingchange-of-basismatrixtopassfrom (

i)to(ei)andtheformula(4)thenreadsJ=PTKaP.(5)meansthat

Q=

⎜ ⎜

⎜ ⎜

⎜ ⎜

1 0 1 0

0 p2

2 1 0

0 0 1 p3

2

0 0 0 p4

2

⎟ ⎟

⎟ ⎟

⎟ ⎟

.

ItfollowsthatA=(QT)1 andcalculationsgive:

A=2

⎜ ⎜

⎜ ⎜

⎜ ⎜

⎜ ⎝

1 0 1 0

0 1

p2

0 0

−1 −1 p2

1 0

p3

p4 p3

p2p4p3

p4 1 p4

⎟ ⎟

⎟ ⎟

⎟ ⎟

⎟ ⎠

. (6)

Itnowremainstoparametrizethe3(= 2×23)dimensionalGrassmaniann(2)ofLagrangianplanesofC2whichisdoneas aboveinthegeneralcasethroughaparametrizationofUs(2).

Let U=

u1 u2

u3 u4

Us

(

2

)

.

BecauseUissymmetric,u2=u3 andbecauseUisunitary,thefollowingthreeindependentrelationshold:

|

u1

|

2+

|

u2

|

2= 1,

|

u4

|

2+

|

u2

|

2=1, u1c(u2)+u2c(u4)=0. Then

|

u1

|

=

|

u4

|

and we parametrize the problem by u1=cos

α

eiα1,u4= cos

α

eiα4, u2=u3=sin

α

eiα2. The third relation u1c(u2)+u2c(u4)=0 then reads cos

α

sin

α

(ei1α2)+ei2α4))=0. Genericallythat leadsto

α

2=α1+2α4+(2k+1)π2:theparametrizationisgivenby

α

,

α

1,

α

4 andthecorrespondingmatrix U(

α

,

α

1,

α

4)reads:

U

( α

,

α

1,

α

4

)

=

cos

α

eiα1 sin

α

ieiα1+2α4

sin

α

ieiα1+2α4 cos

α

eiα4

.

(7)

Avector

v

=(x,y)(2)ifandonlyif

x

y =

cos

α

eiα1

(

−1

)

ksin

α

ieiα1+2α4

(

−1

)

ksin

α

ieiα1+2α4 cos

α

eiα4

c

(

x

)

c

(

y

)

, (7)

whichgives

x=cos

α

eiα1c

(

x

)

+

(

−1

)

ksin

α

ieiα1+2α4c

(

y

)

y=

(

−1

)

ksin

α

ieiα1+2α4c

(

x

)

+cos

α

eiα4c

(

y

)

. (8) Writing

v

=(x,y)=4

i=1xi

v

iinthecanonicalbasisoftheR-symplecticfourdimensionalvectorspace(C2,

ω

),thenc(

v

)= (c(x),c(y))=4

i=1(−1)i+1xi

v

iand(8)thenreads:

X=

(

x1x2x3x4

)

T∈kerB

( α

,

α

1,

α

4

)

, with

B=

⎜ ⎜

⎜ ⎜

⎜ ⎜

⎜ ⎜

⎜ ⎝

cos

α

cos

α

11 cos

α

sin

α

1

(

−1

)

k+1sin

α

sin

α

1+

α

4

2

(

−1

)

ksin

α

cos

α

1+

α

4

2 cos

α

sin

α

1 cos

α

cos

α

11

(

−1

)

ksin

α

cos

α

1+

α

4

2

(

−1

)

ksin

α

sin

α

1+

α

4

2

(

−1

)

k+1sin

α

sin

α

1+

α

4

2

(

−1

)

ksin

α

cos

α

1+

α

4

2 cos

α

cos

α

41 cos

α

sin

α

4

(

−1

)

ksin

α

cos

α

1+

α

4

2

(

−1

)

ksin

α

sin

α

1+

α

4

2 cos

α

sin

α

4 cos

α

cos

α

41

⎟ ⎟

⎟ ⎟

⎟ ⎟

⎟ ⎟

⎟ ⎠

.

(9) Theseequations define a plane P=P(

α

,

α

1,

α

4) ofR4 and L(

α

,

α

1,

α

4)=A(P(

α

,

α

1,

α

4)) withA givenby (6) isthen the Lagrangianplaneof(R4,Ka)definedby

α

,

α

1,

α

4.Itistheparametrizationofthesetofallsolutionsofourproblem.

This“symplectic” solution issimpleincomparisonwiththeone proposedin[12]thatinvolvedtediouscalculationsfor thecalculationofonlyonesolution.Thismethodallows usmoreovertogetthesetofallsolutionsaswell.Thusitshows thatthiswayisstronglymoreefficientthanthewayusingthespectraltheoremforKa2byavoidingthecalculationsofthe eigenvaluesandtheeigenvectorsofKa2.Suchamoreefficientmethodofbuildingthesetofconvenientconstraintsconverting thenonconservativesystemintoaconservativesystemiscertainlyagood(atleastabetter!)wayfortheextension tothe nonlinearcasewhichisnowtackledinthefollowingsection.

2. Setupofthenon-linearissue.Notations.Linkwiththelinearframework 2.1. Non-linearissueandnotations

Wethen consider nowa mechanical discrete system so that its configuration manifoldisa C realn dimensional manifoldand (q=(q1,...,qn),U) denotes generically a local coordinatesystem. That means that U is an open set ofM andthereisa function

φ

:UM→Rn (C) withforall mM,

φ

(m)=q.We supposethat thesystemissubjected toa positionalforcesystemsothatisdescribedbyadifferential1-formQonMwhoselocalexpressionin(q,U)is:

Q= n

k=1

Q,k

(

q

)

dqk. (10)

Accordingto[18],theexact vocablefordescribing isasemi-basic1-form.Thatmeansthat isdescribedbya1-form (oraPfaff form)

ω

onthetotalspaceTMofthetangentbundle

τ

Msuchthat

ω

isontheimageofthecanonicalvertical

operator.Thelocalexpressionofasuchsemi-basicformisthen

ω

= n

k=1

ω

,k

(

q,q˙

)

dqk. (11)

Thankstothepositionalpropertyoftheforces,

ω

,k(q,q˙)onlydependsonqmeaningonlyoftheprojectionTMMofa pointofTMandmaybeviewedasafunctiononthebasisMofthetangentbundle.Then,(11)takestheform(10).

Ifthereisnoambiguity,weomittheforcesystemandwriteQinsteadQ.Anypdimensional(Creal)submanifold NofMmaylocally in(q,U) bedescribedby afamily (f1,...,fn−p) ofnpindependent(C real)functionsdefinedon

φ

(U)sothatforallq

φ

(U),m=

φ

1(q)N

φ

1(U)f1(q)=...=fnp(q)=0.ThemechanicalsystemwhoseNisthe configurationspaceiscalledasubsystemCof andfunctions(f1,...,fnp)arecalledthelocal(non-linear)expressions oftheconstraintsC.WeindifferentlynoteC orN.

(8)

Thepositionalforce systemactingon issaid conservativeifthereisafunctionhonMsothat inanycoordinate system(q,U):

Q,k

(

q

)

=

h

( φ

−1

(

q

))

qk ,

which isequivalentto Q=dhon Morthat Q islocallyexact onM. Itimpliesthat Q isa closeddifferential 1-formand thendQ=0onM.Locally,byuseofPoincaré’stheoremandafterhavingchosenanappropriatecoordinatesystem,wemay supposethereciprocalpropertytrueoneachUoftheatlascoveringM.Asmentionedabove,theglobalissueinvolvingthe topologyofMisoutofthescopeofthispaper.In[18],thevocable“conservative” meansonlythatthesemibasicPfaff form Qisclosed.Theword“Lagrangian system” isreservedforthecasewherethisformisexactintheframeworkofpositional forcesystem.Inthispaper,onlythelocalextensiontothenon-linearcaseisinvestigated.

Theissueisthenthefollowing:isthereasubmanifoldNofMsothatthephysicalactiononN isconservative?Ifso, findallthepossibleNwiththehighestpossibledimensionpsothatthenumberofconstraintsisthesmallestpossible(be- causeeverysubsystembuiltbyaddingkinematicalconstraintstoaconservativesystemisagainaconservativesystem!!!).

2.2. Linkwiththelinearframework

Wefocusnowonthelinkwiththelinearframeworkof[12].Supposethatthereisaconfigurationqe(equilibriumconfig- urationof)suchthatthelinearapproximationisused.Letx=qqe=(x1,...,xn).ThatmeansthatQisapproximated byitsTaylorexpansiontothefirstorderQ˜leadingto:

Q˜=Q˜

(

x

)

= n

k=1

Q,k

(

qe

)

+ n

=1

Q,k

(

qe

)

q x

dxk. (12)

Thatmeansthatifu=(u1,...,un)∈RnTqeM Q˜

(

x

)(

u

)

=

n

k=1

Q,k

(

qe

)

+ n

=1

Q,k

(

qe

)

q x

uk,

andtheexteriorderivativeofQ˜(x)reads:

dQ˜

(

x

)

=dQ˜

(

0

)

= n

k=1

<k

Q,k

(

qe

)

q

Q,

(

qe

)

qk dxdxk.

Becausethematrix(Q,kq(qe))k,isthestiffnessmatrixK=K(qe),thendQ˜(x)=dQ˜(0)isthelinearform

φ

oftheprevi-

oussection.

Toconcludethisparagraph,itmaybenotedthattheissueoftheextension tothenon-linearframeworkofthedecom- positionofthestiffnessmatrixintoitssymmetricandskewsymmetricpartisinvestigatedin[21].Theproposedsolution doesnotusethelanguageofdifferentialformseventhoughthetrickofPoincaré’sLemmaisstronglyusedinthepaper.The proposeddecompositionaimstoextendtothenon-linearcasetheremarkable(butobvious)followingpropertyoftheskew symmetricmatrixKa:xTKax=n

k,=1Ka,kxkx=0.Inourcontext,thisinterestingextensionisnotusable.

3. Solutionofthenon-linearproblem 3.1. Thesolution

IfthereisasuchpdimensionalsubmanifoldNofM,thentherestrictionQ,NofQtoNmustbeclosed.Locallyboth conditionsareequivalentandthelocalconditioninthelocalcoordinatesystem(q,U)ofMreads:

q

φ (

UN

)

X,YTN dQ

(

q

)(

X

(

q

)

,Y

(

q

))

=0, (13) wherethedifferential2-formdQ2(M).Straightforwardcalculationsgive:

dQ

(

q

)

=

<k

Q,k

(

q

)

q

Q,

(

q

)

qk dqdqk,

andthevectorfieldsX,YbelongtoTNifandonlyif(locallyandwithabovenotations) dfi

(

X

)

=dfi

(

Y

)

=0,

foralli=1,...,np(dependencyofqisomitted).

WesupposenowthattheformdQisregularonMmeaningthatitsclassrisconstantonM.Thenhere,sincetheform dQisitselfa closedform(d2=0),its classisalsoequaltoits rankandiseven: r=2s.sistheuniquenumbersuchthat (dQ)s=0and(dQ)s+1=0.Wethendeducethat2sn.

(9)

Darboux’stheoremgivesthelocalmodelingofdQonanopensetUofMandreads:

dQ= s

k=1

dykdyk+s, (14)

wherey1,...,y2sare2sindependentfunctionsonU.Choosenowthesfunctions fk=ykonUforallk=1,...,s.ThenifX andYTNthenX(fi)=dfi(X)=0=dyi(X)andY(fi)=dfi(Y)=0=dyi(Y)foralli=1,...,s.Thus,

dQ

(

X,Y

)

= s

k=1

dykdyk+s

(

X,Y

)

= s

k=1

dyk

(

X

)

dyk+s

(

Y

)

dyk

(

Y

)

dyk+s

(

X

)

=0.

Itistheproofofthefollowing

Proposition1. Suppose thattheclassof dQ is constant(namely maximal). The(non-linear) degree ofnon-conservativity of isthenthehalfsoftheclass2sofdQandthus p=ns.ThelocaldefinitionofthesubmanifoldNis givenbythefamily f1=0,...,fs=0ofequationsonMwherefiisanylinearcombination(inthevectorspaceonRandnotinthemodulusonthe ringonthefunctionsonR)ofyiandyi+sforalli=1,...,s.

Proof.Theproofhasbeengivenfor fk=ykonUforallk=1,...,s.Supposenowthatfiisanylinearcombinationofyiand yi+sforall i=1,...,smeaningthat fi=

α

iyi+

β

iyi+sfori=1,...,swith(

α

i,

β

i)=(0,0).Then,choosinggi=−α2βi

i+βi2yi+ αi

α2i+βi2yi+swededucedykdyk+s=dfkdgkwhichallowtoconcludebyasameargumentasabove.

Build a family of non-linear constraints converting the system into a conservative one is then brought back to find (thankstoDarboux’stheorem) thedecomposition(14).Thisdecompositioncomes itself(byexterior derivative)fromDar- boux’sdecomposition forthe differential 1-form Q. There are several proofsof the existence of such a decomposition.

Beforeexamininganexample,wefirstasktheeffectivenessofthesolution.Inthefollowingparagraph,theissueoftheset ofallnon-linearsolutionsisonlyformalizedinthelanguageoffiberbundles.

3.2.Effectivenessofthesolution

The issueof the effectivenessof the solutionis a significant problem forthe physicist. Here, theeffectiveness of the calculation ofthe degree ofnonconservativity isclear atleast underthe assumption ofmaximal rank (or class)of dQ. CalculatingthesuccessivepowersofdQleadstothevalueofs.TheexampleinSection3.4illustratesthispoint.

Onthe contrary,the effectivenessof thecalculation ofthe constraints defining an appropriate submanifoldN of Mis broughtbacktotheoneofthecanonicalexpressionofdQthankstoDarboux’stheorem.InSection1,weinvestigatedthe effectivenessofbothsolutionsofthesameissuewithinthelinearframework:theoneproposedin[12]andtheoneofthis paper.Theformerisconstructivebutnotreallyeffectivewhereasthelatteriseffective.

We tackle now the corresponding non-linear issueof the effective calculations of family of constraints fi,i=1,...,s whichisequivalenttoDarboux’s theoremfordQ.There areseveralproofsofDarboux’s theorem.Roughlyspeaking,one mayfindtwo differentkinds ofproof of Darboux’stheorem. Withournotations, thefirst one isdone by inductionon s andwemayfindsuchatypeofproofsinseveralbooksofAnalyticMechanicslikein[18]orin[22].Thenon-effectivestep thenliesinthecalculationoftheflowofano time-dependingvector fieldateachstep oftheinductionnamelyherethe calculationsofthe flows of sno time-depending vector fields(these vectorsfields haveto be calculated ateach step of theinductionreasoning).The secondkindofproofsofDarboux’s theoremusesaMoser’slemma whichisareasoningby homotopywithoutinductiononsbutinvolvingthecalculationoftheflowofatime-dependingvectorfield(seeforexample [23]–[25]).Bothkindsofproofshaveitsownadvantagesanddisadvantagesbutbothneedtocalculatetheflowofnon-linear vectorfields oratleasttointegrate thesevector fieldswhichcannot be done generallyanalyticallybutonly numerically.

Thatisthemainobstacletoan analyticsolutionofthisissue.ThefollowingexampleofthefourdegreeoffreedomBigoni systemleadstosuchasituation.

3.3.Setofsolutions

Theyare asmanysolutionsasfunctionsyk,k=1,...,sinvolvedin(14).Butthispointofviewisnot geometric.Itisa similarissueastheonemetinthelinearcase:asolutionisafamilyoflinearformsbutthesetofsolutionsisdescribedby (themanifoldof)Lagrangianspaces.Heretogetan analogousobjectforthedifferentiablenon-linearcase, onehastouse thelanguageofvectorbundle.The2-differentialformdQπ maybeviewedasa sectionofthevector bundle2(M)ofthe 2-differentialformsonMwhichisitselfa vectorbundleassociatedtothe tangentbundle

τ

(M)=(T(M),pM,M)thefiber beingRn. Supposingthat the class2s ofdQπ is constant onMthen thefield q→kerdQπ(q) definesa vectorsubbundle

τ

0(M) over M whose the fiber is Rk withk=n−2s. The total space of this vector bundle is

qM

{

q

}

×kerdQπ(q). On thequotientvector bundle

τ

˜(M) overM of

τ

(M) by

τ

0(M), the 2-differentialformdQ˜π induced bydQπ(q) on eachfiber TqM/kerdQπ(q)definesastructureofsymplectic vectorbundleover M.Thesetofsolutionsisthen builtbythesetofall LagrangianmanifoldsLofthissymplecticvectorbundle.Theverydifficultissuetobuiltitislettofurtherinvestigations.

Références

Documents relatifs

Moreover, when we consider a linear system in max-plus algebra, we obtain the same results as the residuation theory (ii) the constraint propagation techniques are more general than

Lerbet, J., Challamel, N., Nicot, F., Darve, F.: Geometric degree of nonconservativ- ity: set of solutions for the linear case and extension to the differentiable non linear

So, in this paper, we first prove the theorem algebraically in section 2 and then, we expose the equivalent proof using combinatorial Quantum Field Theory (QFT) in subsection 3.3..

Since the edges of the cube and detected edges in the images are equally affected by the same constant errors due to both chemical shifts and magnetic susceptibility, we assume that

Moreover, when we consider a linear system in max-plus algebra, we obtain the same results as the residuation theory (ii) the constraint propagation techniques are more general than

This wear resistance, which is equivalent in concept to the inverse of wear volume, is directly proportional to hardness, but the constant of proportionality

Huang Multiplicity of positive solutions for some quasi linear elliptic equations in IR N with critical Sobolev exponent J. Differential Equations, 140

[6] I. Giga, Solutions for semilinear parabolic equations in L p and regularity of weak solutions of the Navier- Stokes system, J. Masuda, Nonlinear Evolution Equations and