scieee Open visual document viewer

Optimal control, contact dynamics and Herglotz variational problem

de León Rodríguez, Manuel,Lainz Valcázar, Manuel,Muñoz Lecanda, Miguel Carlos

Abstract

In this paper, we combine two main topics in mechanics and optimal control theory: contact Hamiltonian systems and Pontryagin maximum principle. As an important result, among others, we develop a contact Pontryagin maximum principle that permits to deal with optimal control problems with dissipation. We also consider the Herglotz optimal control problem, which is simultaneously a generalization of the Herglotz variational principle and an optimal control problem. An application to the study of a thermodynamic system is provided.

Full text

Jou nal o Nonlinea Science (2023) 33:9 h ps://doi.o g/10.1007/s00332-022-09861-2 Op imal Con ol, Con ac Dynamics and He glo z Va ia ional P oblem Manuel de León1,2 ·Manuel Lainz1·Miguel C. Muñoz-Lecanda3 Recei ed: 25 June 2020 / Accep ed: 7 Oc obe 2022 / Published online: 11 No embe 2022 © The Au ho (s) 2022 Abs ac In his pape , we combine wo main opics in mechanics and op imal con ol heo y: con ac Hamil onian sys ems and Pon yagin maximum p inciple. As an impo an esul , among o he s, we de elop a con ac Pon yagin maximum p inciple ha pe mi s o deal wi h op imal con ol p oblems wi h dissipa ion. We also conside he He glo z op imal con ol p oblem, which is simul aneously a gene aliza ion o he He glo z a ia ional p inciple and an op imal con ol p oblem. An applica ion o he s udy o a he modynamic sys em is p o ided. Keywo ds Con ac Hamiol onian sys ems ·Op imal con ol ·He glo z p inciple · P esymplec ic sys ems ·Pon yagin maximum p inciple Ma hema ics Subjec Classi ica ion 37J55 ·49S05 ·70Q05 ·34H05 ·49K15 ·49K20 · 93C15 Con en s 1 In oduc ion ............................................. 2 2 P econ ac Hamil onian Sys ems ................................... 4 Communica ed by Alain Go iely. BManuel de León [email p o ec ed] BManuel Lainz [email p o ec ed] Miguel C. Muñoz-Lecanda [email p o ec ed] 1Ins i u o de Ciencias Ma emá icas (CSIC-UAM-UC3M-UCM), Mad id, Spain 2Real Academia de Ciencias Exac as, Físicas y Na u ales, Mad id, Spain 3Depa men o Ma hema ics, Uni e si a Poli ècnica de Ca alunya, Ba celona, Spain 123 9Page 2 o 46 Jou nal o Nonlinea Science (2023) 33 :9 2.1 Con ac Mani olds and Hamil onian Sys ems .......................... 5 2.2 P econ ac Mani olds and Hamil onian Sys ems ........................ 6 2.2.1 P econ ac Hamil onian Sys ems and he Cons ain Algo i hm ............. 7 2.2.2 Mo phisms o P econ ac Hamil onian Sys ems ..................... 8 2.3 The Lag angian Fo malism ................................... 9 2.3.1 The He glo z Va ia ional P inciple ............................ 10 3 A Quick Su ey on Op imal Con ol and Pon yagin Maximum P inciple ............. 11 3.1 The Op imal Con ol P oblem .................................. 11 3.1.1 S a emen o he P oblem ................................. 11 3.1.2 The Ex ended Op imal Con ol P oblem ......................... 12 3.2 The Pon yagin Maximum P inciple .............................. 13 3.3 The P esymplec ic App oach o PMP .............................. 14 4 Dynamics o Vec o Fields as Con ac Dynamics .......................... 16 4.1 The Gene al Case ........................................ 16 4.2 The Case M=R×Mo..................................... 16 4.2.1 The Symplec ic Case ................................... 17 4.2.2 The Rela ion wi h Con ac Dynamics .......................... 18 5 The Con ac Dynamics App oach o Pon yagin Maximum P inciple ............... 19 5.1 S a emen o he P oblem .................................... 20 5.2 No mal Solu ions: po=λo=/0................................ 21 5.3 Abno mal Solu ions: po=λo=0............................... 21 6 He glo z Va ia ional P oblem as an Op imal Con ol P oblem ................... 23 6.1 S a emen o he P oblem .................................... 23 6.2 Op imal Con ol App oach o he He glo z Va ia ional P oblem ................ 24 6.3 Applica ion o he P esymplec ic Fo m o he Pon yagin Maximum P inciple ........ 26 6.3.1 The Ex ended P oblem .................................. 26 6.3.2 Solu ion o he Ex ended P oblem wi h he P esymplec ic Fo m o he Pon yagin Max- imum P inciple ...................................... 27 6.4 The Final Resul s ........................................ 29 7 He glo z Op imal Con ol P oblem ................................. 30 7.1 S a emen o he P oblem .................................... 30 7.2 Solu ion o He glo z Op imal Con ol P oblem ......................... 31 7.3 Con ac Fo mula ion o he No mal Solu ions ......................... 33 7.4 Reduc ion o he P oblem .................................... 35 8 Applica ion: Op imal Con ol on The modynamic Sys ems ..................... 37 8.1 Homogeneous Hamil onian Sys ems and Con ac Sys ems ................... 37 8.2 Con ol o Con ac Sys ems ................................... 38 8.3 Applica ion o The modynamic Sys ems ............................ 39 8.4 Example: Gas–Pis on–Dampe Sys em ............................. 40 9 Conclusions and Fu u e Wo k .................................... 43 Re e ences ................................................ 44 1 In oduc ion This pape ies o combine wo impo an opics in mechanics and con ol heo y: Hamil onian con ac sys ems and Pon yagin maximum p inciple in op imal con ol. On he one hand, Hamil onian con ac sys ems a e ge ing a g ea popula i y in ecen imes because hey allow o desc ibe dissipa ion dynamics, and se e al o he ypes o physical sys ems in he modynamics, quan um mechanics, ci cui heo y, con ol heo y, e c. (see o ins ance B a e i 2019;Go o2016; Kholodenko 2013; Rami ez e al. 2017; de León and Sa dón 2017; Gase e al. 2020b; Simoes e al. 2020; Sussmann 1999). Recen ly, a gene aliza ion o con ac geome y has been de eloped o desc ibe ield heo ies wi h dissipa ion (Gase e al. 2020a,c). In ac , he Hamil onian 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 3 o 46 9 o mula ion in he scena io o con ac s uc u es exhibi s e y di e en cha ac e is ics o i s coun e pa in symplec ic mani olds. Indeed, hese di e ences a e based on he ac ha in he con ac case hey a e Jacobi s uc u es, mo e gene al han hose o Poisson ela ed o he symplec ic ones. In a ia ional e ms, one can show ha con ac Hamil onian equa ions can be de i ed om he so-called He glo z p inciple, which includes as a pa icula case he classical Hamil on p inciple. On he o he hand, he Pon yagin maximum p inciple (PMP), see Pon yagin e al. (1962), Ba be o-Liñan and Muñoz-Lecanda (2009) and e e ences he ein, is he mos use ul ins umen o inding solu ions o an op imal con ol p oblem. In ac , he PMP is he pa adigm in he heo y o op imal con ol, and since i s o mula ion has ne e ceased esea ch on i s inc edible p ope ies, om e y di e en poin s o iew, al hough we will ocus he e on i s mo e geome ic aspec s. An immedia e issue a ising om possible applica ions is ha o s udying p oblems o op imal con ol om he poin o iew o Hamil onian con ac sys ems, and he e o e o sys ems wi h dissipa i e p ope ies among many o he s. And, hen, i seems e y na u al o ask whe he a Pon yagin maximum p inciple could be de eloped o deal wi h a con ac con ol p oblem. To ou knowledge he ela ionship be ween con ac Hamil onian sys ems and he Pon yagin maximum p inciple was i s no iced in Ohsawa (2015) and de eloped in Jó´zwikowski and Respondek (2016). T ying o look o bo h opics wi h a common iewpoin , we conside wea he he solu ion cu es o he Pon yagin maximum p inciple admi a o mula ion in e ms o Hamil onian con ac sys ems in an adequa e mani old. Con e sely, we examine i he He glo z a ia ional p oblems can be unde s ood as a pa icula class o op imal con ol p oblems. Wi h all his in mind, he pape is s uc u ed as ollows. Sec ions 2and 3a e dedica ed o e iew he elemen s o Hamil onian con ac sys ems and Pon yagin maximum p inciple, bo h necessa y o unde s and he objec o he manusc ip . So, Sec . 2is de o ed jus o ecall he main no ions and esul s abou con ac Hamil onian sys ems, including he so-called He glo z p inciple, a na u al ex ension o he well-known Hamil on p inciple. As we said abo e, his sec ion will acili a e a be e unde s anding o he es o he pape . Sec ion 3is dedica ed o he Pon yagin maximum p inciple in se e al o mula- ions. We in oduce he classical op imal con ol p oblem, he associa ed ex ended sys em, he classical Pon yagin maximum p inciple and i s ans o ma ion in o he symplec ic and p esymplec ic o mula ions. This las one is used in se e al sec ions o he a icle. In Sec . 4we discuss an in e es ing pa icula case o Hamil onian dynamics; indeed, gi en a ec o ield Xon a mani old M, one can de ine he comple e li o X o he co angen bundle T∗Mwhich is jus he Hamil onian ec o ield co esponding o he Hamil onian unc ion de e mined by X: jus i s e alua ion on he poin s o he co angen bundle. Hence he dynamics o a gene al ec o ield is desc ibed as he co esponding o a Hamil onian ec o ield in a symplec ic mani old. Bu his dynamics on T∗Mis iche han one could expec . In ac , i he mani old Mdecomposes as M=R×Mo, and he ec o ield has a pa icula symme y p ope y, one has a e y na u al se ing o iden i y wo di e en geome ic beha iou s acco ding o he alue o he momen un po 123 9Page 4 o 46 Jou nal o Nonlinea Science (2023) 33 :9 co esponding o he global coo dina e xo. Indeed, one is a (p e) symplec ic geome y, when po=0, and he second one, a con ac geome y, when po=/0. Sec ions 5,6and 7a e he bulk o he pape . Sec ion 5is in a b oade sense a di ec applica ion o Sec . 3. We conside an op imal con ol sys em gi en by (M,U,X,I,xa,xb)whe e M=R×Mo, ha is, we s udy he so-called ex ended sys em associa ed o an op imal con ol p oblem de ined by a ec o ield depending on con ols, X(x,u), and a cos unc ion F. Applying Theo em 5in Sec . 3, we know ha his p oblem is equi alen o sol ing he dynamics o he p esymplec ic sys em (T∗M×U,ω,H), whe e X=F∂ ∂xo+Xi∂ ∂xi,H=Fpo+Xipiis he linea Hamil- onian gi en by X, and ωis he p esymplec ic o m ob ained by li ing he canonical symplec ic o m, ωM∈2(T∗M), oT∗M×U. He e, U ep esen s ob iously he space o con ols. The co esponding p esymplec ic algo i hm p o ides he solu ions, and we can dis inguish wo cases: he egula one, when he con ols can be ob ained as unc ions o he es o a iables, o he singula one, ha p oduces highe o de condi ions. Again, he e olu ion o he momen um pois cons an , and his pe mi s, as abo e, o discuss he cases whe e po=0o po= 0. Wi h his in mind, we a e able o s a e he Con ac Pon yagin maximum p inciple (Theo em 4). Sec ion 6is jus de o ed o in e p e he He glo z p inciple as an Op imal Con ol P oblem, and de i e he He glo z equa ions o mo ion using he co esponding Pon- yagin p inciple. In Sec . 7we s a e he He glo z Op imal Con ol P oblem and ind he solu ion equa ions. In his si ua ion, he ex emal condi ion, gi en as an in eg al o he cos unc ion in he classical op imal con ol p oblems, is changed in o an ex emal condi ion on he solu ions o a di e en ial equa ion on a new a iable o be maximized. This p oblem is a gene aliza ion o he classical op imal con ol sys ems in he sense ha we ob ain he classical equa ions i he cos unc ion and he ex emal condi ion is like in he classical si ua ion. Finally, in Sec . 8we apply he abo e esul s o an example coming om The modynamics. Being awa e ha in p ac ical applica ions o op imal con ol i is necessa y o use mo e gene al classes o unc ions and mappings, as i is usual in his kind o heo e ical app oaches, all he mani olds and mappings a e conside ed as o C∞-class. The usual Eins ein con en ion o summa ion indices will be unde s ood unless indica ed. As gene al e e ences o no a ions and basic esul s on geome y, mechanics and con ol we use (Ab aham and Ma sden 1978; Bullo and Lewis 2005; Bloch 2015). 2 P econ ac Hamil onian Sys ems In his sec ion we e iew he necessa y heo y o con ac mani olds, con ac and p econ ac dynamical sys ems, in bo h Hamil onian and Lag angian o mula ions, and He glo z a ia ional p inciple and i s gene alized Eule –Lag ange equa ions. See A nold (1978), B a e i (2017), B a e i e al. (2017), de León and Lainz-Valcáza (2019),Gase e al.(2020a), Geiges (2008), Guen he e al. (1996), Lainz-Valcáza and de León (2019) and Liu e al. (2018) o de ails. 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 5 o 46 9 2.1 Con ac Mani olds and Hamil onian Sys ems Acon ac mani old (M,η)is a (2n+1)-dimensional mani old equipped wi h a con ac o m η, ha is a 1- o m sa is ying η∧(dη)n= 0. Then, he e exis a unique ec o ield R, called he Reeb ec o ield, such ha iRdη=0,iRη=1.(1) Gi en (M,η), he e is a Da boux heo em o con ac mani olds: a ound each poin in Mone can ind local Da boux coo dina es (qi,pi,z)such ha η=dz−pidqi,R=∂ ∂z.(2) As an example, and a na u al model, we ha e he ex ended co angen bundle T ∗Q× Ro an n-dimensional mani old Q, which ca ies a na u al con ac o m ηQ=dz−θQ,(3) whe e θQis he pullback o he Liou ille 1- o m o T∗Q,θQ=pidqi, being (qi,pi,z) he na u al bundle coo dina es o T∗Q×R. I (M,η)is a con ac mani old, he map: ¯ :TM →T∗M, → ι dη+η( )η. is a ec o bundle isomo phism o e M. Gi en a Hamil onian unc ion H:M→R, we can de ine a dynamical sys em. The iple (M,η,H)is called a con ac Hamil onian sys em. The associa ed Hamil onian ec o ield XHis he solu ion o he ollowing equa ion ¯ (XH)=dH−(R(H)+H)η. (4) In Da boux coo dina es, XHhas he local exp ession XH=∂H ∂pi ∂ ∂qi−∂H ∂qi+pi ∂H ∂z∂ ∂pi +pi ∂H ∂pi −H∂ ∂z.(5) The e o e, an in eg al cu e (qi( ), pi( ), z( )) o XHsa is ies he di e en ial equa- ions dqi d =∂H ∂pi , dpi d =− ∂H ∂qi−pi ∂H ∂z, 123 9Page 6 o 46 Jou nal o Nonlinea Science (2023) 33 :9 dz d =pi ∂H ∂pi −H. 2.2 P econ ac Mani olds and Hamil onian Sys ems Le ηbe a 1- o m on an m-dimensional mani old M. We de ine he cha ac e is ic dis ibu ion o ηas C=ke η∩ke dη⊆TM,(7) which we suppose o be egula , ha is, o cons an ank. We say ha ηis a 1- o m o class c i he ank o he dis ibu ion Cis m−c. The e exis some cha ac e iza ions o his no ion o a 1- o m gi en in he ollowing (Godbillon 1969). P oposi ion 1 Le ηbe a one- o m on an m-dimensional mani old M. Then, he ol- lowing s a emen s a e equi alen : 1. The o m ηis o class 2 +1. 2. A e e y poin o M, η∧(dη) =/0,η∧(dη) +1=0.(8) 3. A ound any poin o M, he e exis local Da boux coo dina es x1,...x ,y 1,...y , z, u1,...us, whe e 2 +s+1=m, such ha η=dz−  i=1 yidxi.(9) In hese Da boux coo dina es, he cha ac e is ic dis ibu ion o ηis gi en by C=∂ ∂uaa=1,...,s.(10) A pai (M,η) o a mani old Mequipped wi h a o m ηas abo e will be called ap econ ac mani old (see Godbillon 1969). The o m ηwill be called a p econ ac o m. Rema k 1 The dis ibu ion Cis in olu i e and i gi es ise o a olia ion o M.I he quo ien π:M→M/Chas a mani old s uc u e, hen he e is a unique 1- o m ˜ηsuch ha π∗˜η=η. F om a di ec compu a ion, ˜ηis a con ac o m on M/C. This jus i ies he name o p econ ac o m. Gi en (M,η), he ollowing map :TM →T∗M → ι dη+η( )η, (11) 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 7 o 46 9 is a mo phism o ec o bundles o e Mand i s ke nel is C. AReeb ec o ield o (M,η)is a ec o ield Ron Msuch ha ιRdη=0,η(R)=1,(12) o , equi alen ly (R)=η. We no e ha he e exis Reeb ec o ields in e e y p econ ac mani old. Indeed we can de ine local ec o ields R=∂ ∂zin Da boux coo dina es and can ex end i using pa i ions o uni y. Howe e , unlike on con ac mani olds, hey a e no unique. In ac , gi en a Reeb ec o ield Rand any sec ion Co C, we ha e ha R=R+Cis ano he Reeb ec o ield. 2.2.1 P econ ac Hamil onian Sys ems and he Cons ain Algo i hm Ap econ ac Hamil onian sys em is a p econ ac mani old (M,η) wi h a smoo h unc ion H:M→Rcalled he Hamil onian. We deno e i by (M,η,H). Fo a p econ ac Hamil onian sys em (M,η,H), gi en a submani old M⊂M, aHamil onian ec o ield along Mis a ec o ield X∈X(M), such ha X|M∈ X(M)and solu ion o he equa ion (X)=dH−(H+R(H))η, (13) a he poin s o M, and being Rany Reeb ec o ield. I can be seen ha , i his equa ion holds o one Reeb ec o ield, i will hold o all o hem. No ice ha , since is no an isomo phism, hen (13) migh no ha e solu ions a e e y poin o he mani old M. Fu he mo e, solu ions, i hey exis s, a eno necessa ily unique. Indeed, adding a sec ion Co C o a solu ion Xgi es ise o a new solu ion X=X+C. In o de o ob ain he maximal submani old along which Hamil onian ec o ields a e de ined, we can de elop a cons ain algo i hm.Todoso,le γH= dH−(H+R(H))η ∈1(M)and de ine induc i ely M0=M, and o any posi i e in ege i, Mi={p∈Mi|(γH)p∈(TpMi−1)},(14) whe e we assume ha all Mia e mani olds. The algo i hm will e en ually s op, ha is, we will ind a posi i e in ege isuch ha Mi=Mi−1. We call his submani old he inal cons ain submani old M .I M has posi i e dimension, he e will exis Hamil onian ec o ields along M .The pai (M ,X)will be called a Hamil onian ec o ield solu ion o he Hamil onian p econ ac sys em (M,η,H). A use ul cha ac e iza ion o such pai s is gi en by he ollowing P oposi ion 2 X is a Hamil onian ec o ield along M o (M,η,H)i and only i , a he poin s o M, η(X)=−H,(15a) 123 9Page 8 o 46 Jou nal o Nonlinea Science (2023) 33 :9 LXη=gη, (15b) whe e g :M→R. Mo eo e , i his holds, hen g =−R(H) o any Reeb ec o ield R. P oo Le Xbe a Hamil onian ec o ield along M. By he de ini ion o , equa ion (13), a he poin s o M, becomes ιXdη+η(X)η =dH−(H+R(H))η, (16) and, by con ac ion wi h R, we ob ain η(X)=−H.(17) Combining (16) and (17), we deduce ιXdη+dιXη=−R(H)η, (18) bu he le -hand side o his equa ion equals LXηby Ca an’s o mula, hence X ul ills (15) a he poin s o M. Now assume ha Xsa is ies (15) on he poin s o M. Once again, by con ac ion o (15b) wi h a Reeb ec o ield R,weha e g=ιRLX(η) =ιR(ιXdη+d(η(X))) =−ιR(dH)=−R(H). (19) Combining his wi h (15), we can easily e ie e (16).  2.2.2 Mo phisms o P econ ac Hamil onian Sys ems Le (M,η,H)and (¯ M,¯η, ¯ H)be p econ ac Hamil onian sys ems. A map F:M→¯ M is said o be a con o mal mo phism o p econ ac sys ems i F∗¯η= ηand F∗¯ H= H o some non- anishing unc ion :M→R.I =1, we say ha Fis a s ic mo phism o p econ ac sys ems. Theo em 1 Le F :M→¯ M be a con o mal mo phism o p econ ac sys ems. Assume ha X,¯ X a e F- ela ed ec o ields de ined along submani olds M⊆M and ¯ M= F(M)⊆¯ M, espec i ely. The e o e, i ¯ X is a Hamil onian ec o ield along ¯ M, hen X is also a Hamil onian ec o ield along M. P oo Since ¯ Xis a Hamil onian ec o ield, i s sa is ies (15) along ¯ M ¯η( ¯ X)=−¯ H,(20a) L¯ X¯η=¯g¯η. (20b) Pulling back by F, we ob ain η(X)=− H,(21a) 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 9 o 46 9 LX( η) =(¯g◦F) η. (21b) F om his exp ession, we ob ain η(X)=−H,(22a) LX(η) =gη, (22b) whe e g=¯g◦F−(LX )/ . Hence Xis a Hamil onian ec o ield.  Obse e ha i Fis a di eomo phism, hen we ha e a bijec i e co espondence be ween pai s o Hamil onian ec o ields along submani olds. 2.3 The Lag angian Fo malism Unlike T∗Q×R, he mani old TQ×Rdoes no ha e a canonical con ac s uc u e. Howe e , gi en a Lag angian unc ion L :TQ ×R→Rone can cons uc he 1- o m ηL=dz−θL,(23) whe e θLis he associa ed Lag angian 1- o m, which in bundle coo dina es (qi, i,z) is w i en as θL=∂L ∂ idqi.(24) The Lag angian Lis said o be egula i i s Hessian ma ix wi h espec o he eloci ies, (Wij)=∂2L ∂ i∂ j,(25) is egula . One can see ha ηLis con ac o m when Lis egula . Fu he mo e, ηLis a p e- con ac o m when (Wij)has cons an ank (see de León and Lainz-Valcáza 2019, Sec ion). The ene gy o he Lag angian is EL=(L)−Lwhe e is he canonical Liou ille ec o ield on TQ,= i∂ ∂ i, ex ended in he usual way o TQ×Rwi h he same local exp ession. Hence, p o ided Lis such ha (Wij)has ull ( esp. cons an ) ank we ha e ha (TQ×R,η L,EL)is a con ac ( esp. p econ ac ) Hamil onian sys em. Le ξLbe a Hamil onian ec o ield o his con ac o p econ ac sys em. F om a di ec compu- a ion one can see ha e e y in eg al cu e (qi( ), i( ), z( )) o ξLis a solu ion o he He glo z equa ions: d d ∂L ∂ i−∂L ∂qi=∂L ∂ i ∂L ∂z,(26) 123 9Page 16 o 46 Jou nal o Nonlinea Science (2023) 33 :9 na u al p ojec ion M×U→M, conside ing he con ols as he elemen s o he ib es. The local equa ions a e he same ha we ha e ob ained in he i ial case o he con ols. 4 Dynamics o Vec o Fields as Con ac Dynamics I is well known ha he in eg al cu es o a ec o ield in a mani old Mcan be ob ained as p ojec ion o in eg al cu es o a Hamil onian ec o ield in he co angen bundle. We can ex end his dynamics o he con ac associa ed mani old TM×R,as in Eq.(3), wha gi es he addi ional equa ion ˙z=0, ha is in a i ial way. We wan o ob ain a non- i ial ex ension. In his sec ion we s udy how o ob ain hese in eg al cu es as solu ions o a con ac dynamical sys em in an adequa e con ac mani old, a leas in he case ha he o iginal ec o ield has some symme y p ope ies. He e we eco e a simila si ua ion we had in he Pon yagin maximum p inciple in i s symplec ic app oach. See Sec . 3. 4.1 The Gene al Case Le Mbe a mani old and X∈X(M)a ec o ield. Le ˆ X:T∗M→R he na u al unc ion de ined by ˆ X(α) =α(X)=<α, X>. In a canonical coo dina e sys em (xi,pi)in T∗M, we ha e ha ˆ X(x,p)=piXi. As i is well known, i ωM=−dθMis he symplec ic canonical 2- o m in T∗M, we can conside he Hamil onian symplec ic sys em (T∗M,ωM,ˆ X). Then he Hamil- onian ec o ield Yˆ X∈X(T∗M), de ined by i(Yˆ X)ωM=dˆ X, has local exp ession X=Xi∂ ∂xi,⇒Yˆ X=Xi∂ ∂xi−pj ∂Xj ∂xi ∂ ∂pi i (xi)and (xi,pi)a e coo dina es o Mand T∗M espec i ely. By his local exp ession we ha e ha Yˆ X=X∗, whe e X∗is he so-called canonical li ing o X∈X(M) o T∗M. The in eg al cu es o Yˆ Xp ojec ed o Ma e he in eg al cu es o Xas we can see by di ec obse a ion o he abo e local exp ession. Wi h his me hod, we ha e ans o med any ec o ield in a Hamil onian one bu doubling he dimension. Fo de ails abou hese cons uc ions we e e o de León and Rod igues (1989), Yano and Ishiha a (1973). Obse e ha he Hamil onian ˆ Xdepends linea ly on he momen a. 4.2 The Case M=R×Mo In his Sec ion we analyze he speci ic case whe e in he mani old M he e is a pa icula “di ec ion”, ha is M=R×M0. This si ua ion allows us o spli up he in eg al cu es o a ec o ield, wi h a symme y p ope y, in o wo di e en classes: one ollowing a symplec ic geome y and he o he class unde a con ac geome y. This s udy is 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 17 o 46 9 connec ed wi h he ideas de eloped in Ohsawa (2015) whe e he con ac s uc u e is associa ed o he p ojec i e mani old associa ed o T∗M. 4.2.1 The Symplec ic Case Suppose now ha we ha e one di ec ion specially iden i ied in he angen bundle o he mani old, ha is M=R×Mo. When necessa y we deno e by (xo,xi)a coo dina e sys em in Mand (xo,xi,po,pi)i s na u al ex ension o T∗M. Le X∈X(M)and suppose ha ∂ ∂xo,X=0. In coo dina es his means ha , i X=Xo∂ ∂xo+Xi∂ ∂xi, hen he coo dina es Xoand Xio he ec o ield Xdo no depend on xo. In pa icula his implies ha Xis p ojec able o Mo. Rema k 4 Wha is he meaning o his si ua ion? Suppose we ha e wo ec o ields Xo,X∈X(M)wi h [Xo,X]=0. Then a ound any egula poin o Xowe can choose a local coo dina e sys em (U,xo,xi), wi h i=1,...,n,i dimM=1+n, and U⊂Man open se , wi h Xo|U=∂/∂xo. Hence we ha e he abo e si ua ion bu locally. In his case he local decomposi ion {xo}×{xi}is no unique. This is wha we called abo e “pa icula symme y p ope y” o he ec o ield X. We can obse e ha i is a common si ua ion a leas locally. This is a si ua ion we a e going o ackle when ying o ela e con ac s uc u es and op imal con ol. The a iable xowill co espond o he cos unc ion Fas we ha e seen in Sec . 3in ou e iew o he Pon yagin maximum p inciple. I we p oceed in his case as abo e in he gene al si ua ion, wi h i(X∗)ωM=dH, whe e he Hamil onian unc ion His de ined by H=ˆ X=poXo+piXi hen he co esponding Hamil onian ec o , using [∂/∂xo,X]=0, is gi en by X∗=Xo∂ ∂xo+Xi∂ ∂xi−0∂ ∂po −po ∂Xo ∂xi+pj ∂Xj ∂xi∂ ∂pi . The associa ed sys em o di e en ial equa ions is: ˙xo=Xo,˙po=0,˙xi=Xi,˙pi=−po ∂Xo ∂xi−pj ∂Xj ∂xi This is he desc ip ion o he Hamil onian sys em (T∗M,ωM,H)wi h H=ˆ X. 123 9Page 18 o 46 Jou nal o Nonlinea Science (2023) 33 :9 4.2.2 The Rela ion wi h Con ac Dynamics Obse e ha he ec o ield X∗is angen o he submani old de ined by po= cons an , hence we can educe he p oblem o hose hype su aces o T∗M.Weha e wo di e en si ua ions and, by compa ison wi h he si ua ion o he op imal con ol and he symplec ic Pon yagin maximum p inciple, we will call no mal and abno mal si ua ions. (a) The no mal si ua ion po=/0 Fo λo∈R,λo=/0, le N⊂T∗Mbe he submani old de ined by po=λoand le j:N→T∗Mbe he na u al inclusion. Ob iously he dimension o Nis odd, hence i canno be a symplec ic mani old. We deno e by (xo,xi,pi) he coo dina es induced in Nby he coo dina es we ha e in T∗M. Conside now he canonical 1- o m θM∈1(T∗M)and le η=−j∗θM, hen we ha e he ollowing esul Lemma 1 (N,η)is a con ac mani old. The Reeb ec o ield is R =−1 λo ∂ ∂xo. The p oo is di ec using i s local exp ession, η=−λodxo−pidxi. The minus sign comes om a con en ion in he de ini ion o he symplec ic o m in T∗Mand he 1- o m and 2- o m in a con ac mani old. Le HN=j∗Hbe he es ic ion o H o N. We ha e ha , locally, HN= λoXo+piXiand we ha e a Hamil onian con ac sys em gi en by (N,η,HN).Le XN∈X(N)be he co esponding con ac Hamil onian ec o ield, ha is: i(XN)η =−HN,i(XN)dη=dHN−(L(R)HN)η whose local exp ession is XN=Xo∂ ∂xo+Xi∂ ∂xi−λo ∂Xo ∂xi+pj ∂Xj ∂xi∂ ∂pi ,(34) wi h he usual no a ion con using he unc ions on T∗Mand hei es ic ions o N. Wi h his in mind, we ha e ha : Theo em 6 The ec o ield X∗∈X(T∗M)is angen o N and, on he poin s o N, i is equal o XN. Hence he no mal in eg al cu es o he ec o ield X∗a e solu ions o a Hamil- onian con ac dynamics on a co esponding con ac mani old. The con ac sys em is (N,η,HN). Commen : A li le calculus He e we gi e he co esponding calculus o ob ain he exp ession in (34). We ha e ha HN=λoXo+piXiand η=−λodxo−pidxi. Deno ing XNby XN=ao∂ ∂xo+ai∂ ∂xi+bi ∂ ∂pi 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 19 o 46 9 he i s con ac dynamical equa ion is: i(XN)η =−HN⇒−λoao−aipi=−λoXo−piXi and he second one i(XN)dη=dHN−(L(R)HN)η ⇒ −bidxi+aidpi=λo ∂Xo ∂xidxi+Xidpi+pj ∂Xj ∂xidxi. Hence ai=Xi,bi=−λo ∂Xo ∂xi−pj ∂Xj ∂xi,ao=Xo as we wan ed. (b) The abno mal si ua ion po=0 This case co esponds o λo=0 and he submani old No⊂T∗Mde ined by po=0. Le jo:No→T∗Mbe he na u al inclusion and ηo=j∗ oθM. Obse e ha ηo=−pidxiis no a con ac o m. In ac , as mo=dim Mo,we ha e ha ηo∧(dηo)mo−1=/0, bu ηo∧(dηo)mo=0. We can conside he 2- o m ωo=dηo, he Hamil onian Ho=j∗ oHand he p esymplec ic mani old (No,ω o,Ho). Obse e ha ke ωo={ ∂ ∂xo}. The Hamil onian p esymplec ic equa ion i(Xo)ωo=dHo gi es he solu ion Xo=Xi∂ ∂xi−pj ∂Xj ∂xi ∂ ∂pi +A∂ ∂xo, whe e Ais a bi a y and co esponds o ke ωo. In ac we ha e ha ˙xo=A. I does no exis any cons ain because he ec o ield Xois de ined on he whole mani old No. This is because he only cons ain is gi en by LTHo=0 wi h T∈ ke ωoand his is ul illed globally on No. Commen : Obse e ha T∗M=λ∈RNλ, hence wi h his decomposi ion we ob ain all he solu ions o he ini ial Hamil onian p oblem on T∗Mgi en by he Hamil onian H. 5 The Con ac Dynamics App oach o Pon yagin Maximum P inciple Following he ideas o he p e ious sec ions, we s udy a con ac app oach o he Pon yagin maximum p inciple, in pa icula o he so-called no mal solu ions o he op imal con ol p oblem. In pa icula we will ob ain he no mal solu ions o an op imal 123 9Page 20 o 46 Jou nal o Nonlinea Science (2023) 33 :9 con ol p oblem as p ojec ion o he in eg al cu es o a Hamil onian con ac sys em in adequa e mani olds. The abno mal solu ion can be ob ained wi h ano he di e en app oach gi en a he end o his sec ion. 5.1 S a emen o he P oblem Le (M,U,X,I,xa,xb)be an op imal con ol p oblem. We know by Theo em 5 ha o sol e his p oblem we need o s udy he associa ed Hamil onian p esymplec ic sys em (T∗M×U,ω,H), ha is o ob ain an in eg al cu e o he ec o ield XH solu ion o he equa ion i(XH)ω =dH, whe e ω=π∗ 1ωo=dxo∧dpo+dxi∧dpi,H=ˆ X=poF+piXi and π1:T∗M×U→T∗M. Recall ha ke ω={∂/∂ua}. The solu ion o he equa ion i(XH)ω =dHis gi en by: XH=F∂ ∂xo+Xi∂ ∂xi−λo ∂F ∂xi+pj ∂Xj ∂xi∂ ∂pi +Aa∂ ∂ua.(35) Obse e ha his solu ion exis s all o e he mani old T∗M×Uand ha pois cons an o e e y cu e solu ion o he p oblem. The las e m co esponds o he elemen s o ke ω. The minimali y, compa ibili y, condi ions a e ∂H ∂ua=0 o e e y a, a e used o de e mine he con ols. As we said in Sec . 3, i he compa ibili y equa ions allows us o de e mine he con ols u1,...,uk, ha is we can ob ain ua=ψ(xo,xi,po,pi), hen we say ha he op imal con ol p oblem is egula , o he wise i is called singula . In he singula case, i is necessa y o apply an algo i hm o cons ain s, ha is o go o highe o de condi ions, o ob ain he con ols pe haps on a submani old o T∗M×U. Suppose ha we a e in he egula si ua ion, hence we ha e de e mined he con ols by he compa ibili y condi ions. Wi h he egula i y assump ion as he con ols uahas been de e mined, we ha e ha XHis p ojec ed o he mani old T∗Mand has componen s only in (xo,xi,po,pi).Then we ha e: XH=F∂ ∂xo+Xi∂ ∂xi−λo ∂F ∂xi+pj ∂Xj ∂xi∂ ∂pi (36) because we a e in he symplec ic case. We know ha , o all he solu ions o he associa ed p esymplec ic o mula ion, we ha e ha he momen po( )is a cons an . Following he p e ious sec ion, we will y o classi y he solu ions acco ding o he eal alue o po. Hence we de ine and s udy (a) No mal solu ions: hose wi h po=λo=/0. (b) Abno mal solu ions: hose wi h po=λo=0. 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 21 o 46 9 5.2 No mal Solu ions: po=o=/0 Le N⊂T∗Mbe he submani old de ined by po=λoand j:N→T∗Mbe he na u al inclusion. We deno e by (xo,xi,pi) he coo dina es induced in Nby he coo dina es we ha e in T∗M. Conside now he canonical 1- o m θM∈1(T∗M)and le η=−(j)∗θM, hen we ha e ha Lemma 2 (N,η)is a con ac mani old. The Reeb ec o ield is R =−1 λo ∂ ∂xo. Le HN=(j)∗H he es ic ion o H o N, hen HN=λo(X)o+pi(X)iand we ha e a Hamil onian con ac sys em gi en by (N,η,HN).Le XN∈X(N)be he co esponding con ac Hamil onian ec o ield, ha is he solu ion o he equa ions i(XN)η =−HN,i(XN)dη=dHn−(L(R)HN)η whose local exp ession is XN=Xo∂ ∂xo+Xi∂ ∂xi−λo ∂Xo ∂xi+pj ∂Xj ∂xi∂ ∂pi .(37) Wi h he usual no a ion deno ing by he same names he unc ions on T∗Mand hei es ic ions o N. Wi h his in mind and ollowing Sec . 4.2.1,weha e ha : P oposi ion 3 The ec o ield XH∈X(T∗M)is angen o N and, on he poin s o N, i is equal o X N. Hence, o e e y u∈U, all he no mal solu ions o he op imal con ol p oblem a e solu ions o a con ac Hamil onian p oblem. 5.3 Abno mal Solu ions: po=o=0 Le No⊂T∗M he submani old de ined by po=0. Le jo:No→T∗Mbe he na u al inclusion and ηo=j∗ oθM. As abo e, ηo=−pidxiis no a con ac o m and we ha e ha ηo∧(dηo)mo=0. We can conside he 2- o m ωo=dηo, he Hamil onian Ho=j∗ oHand he p esymplec ic mani old (No,ω o,Ho). Obse e ha ke ωo={ ∂ ∂xo}. The Hamil onian p esymplec ic equa ion i(Xo)ωo=dHo gi es he solu ion Xo=Xi∂ ∂xi−pj ∂Xj ∂xi ∂ ∂pi +Aa∂ ∂ua, 123 9Page 22 o 46 Jou nal o Nonlinea Science (2023) 33 :9 whe e Aaa e a bi a y and co espond o ke ωo. And i does no exis any cons ain because he ec o ield Xois de ined on he whole mani old No. No e: We can also sol e he p econ ac p oblem gi en by (Nu o,η u o,Hu o). Commen : Obse e ha T∗M=λ∈RNλ, hence wi h his decomposi ion we ob ain all he solu ions o he Hamil onian p oblem on T∗Mgi en by he Hamil onian H. Some o hem, he no mal solu ions, as con ac p oblems, and he abno mal solu ions as symplec ic ones. Wi h all his in mind, we ha e p o ed he ollowing Theo em 7 (Con ac Pon yagin maximum p inciple) Conside he op imal con ol p oblem (M,U,X,I,xa,xb), wi h U ⊂Rkan open se . Le ˆσ:I→T∗M×U= T∗R×T∗Mo×U, ˆσ=(σT∗M,σ U), be a solu ion o he p esymplec ic Pon yagin maximum p inciple o such p oblem and suppose we a e in he egula case, ha is he minimali y condi ions (∂ H/∂u)=0, o e e y u ∈U and o e e y ∈I allows o de e mine he con ols. Then (a) i ˆσis a no mal solu ion wi h po=λo=/0, hen s i is an in eg al cu e o he con ac Hamil onian sys em (N,η,HN), as desc ibed abo e, wi h HN= λoF+piXi. (b) i ˆσis an abno mal solu ion, hen i is an in eg al cu e o he p esymplec ic Hamil onian sys em (No,ω o,Ho), as desc ibed abo e, wi h Ho=piXi. Fo he no mal solu ions, hey sa is y he di e en ial equa ions: ˙xo=∂H ∂po =F,˙po=∂H ∂xo=0,(⇒po=c ) ˙xi=∂H ∂pi =Xi,˙pi=− ∂H ∂xi=−po ∂F ∂xi−pj ∂Xj ∂xi ∂H ∂u1=0,..., ∂H ∂uk=0 whe e H=λoF+piXiwi h λo=/0. Fo he abno mal solu ions, he co esponding di e en ial equa ions a e ˙xi=∂H ∂pi =Xi,˙pi=− ∂H ∂xi=−pj ∂Xj ∂xi ∂H ∂u1=0,..., ∂H ∂uk=0 whe e H=piXi. Commen : We no e ha in Ohsawa (2015), an app oach o Pon yagin maximum p inciple is gi en in e ms o con ac sys ems. Indeed, he au ho wo ks in he p ojec- i iza ion o he co angen bundle, PT∗(M0×R). In ha app oach, he no mal and abno mal solu ions a e uni ied, and he abno mal ones co espond o he hype plane a in ini y. Howe e his mani old is no a con ac mani old in he sense we a e using, so we a e o ced o emo e he hype plane a in ini y and ob ain T∗M0×R. 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 23 o 46 9 This can be seen as a di e en o mula ion o Theo em 7in ou pape , whe e we ea sepa a ely bo h kind o solu ions, which co espond o di e en geome ies (con ac and p esymplec ic). In addi ion, in ha e e ence he au ho does no s udy he ela ionship be ween he He glo z a ia ional p inciple and op imal con ol, which is he ocus o ou pape . 6 He glo z Va ia ional P oblem as an Op imal Con ol P oblem In Sec . 2.3.1 we ha e s udied he He glo z a ia ional p inciple; he e we ob ained he con ac equa ions o a Hamil onian con ac sys em as solu ion o a a ia ional p oblem bu wi h a gene aliza ion o he Hamil on a ia ional p inciple. This mo e gene al p inciple was s a ed and sol ed in 1930 by Gus a He glo z, see He glo z (1930) and Guen he e al. (1996). The idea was o change he in eg al s a emen on he cu es solu ion o he p oblem by a di e en ial equa ion de ined p ecisely by he Lag angian unc ion. In e es in his app oach has been inc easing since he las e e ed publica ion and i s ela ion wi h con ac dynamics and dissipa ion sys ems, see o example Geo gie a and Guen he (2002) and B a e i e al. (2017) and e e ences he ein. In his sec ion we app oach He glo z p inciple as an op imal con ol p oblem and ind he co esponding di e en ial equa ions, he gene alized Eule –Lag ange equa ions, wi h a new p oo h ough he Pon yagin maximum p inciple. 6.1 S a emen o he P oblem We begin ecalling he s a emen o he He glo z a ia ional p oblem as we did in Sec . 2.3.1. Le Qbe a smoo h mani old and F:TQ×R→Ra smoo h unc ion and conside he ollowing p oblem: He glo z a ia ional p oblem: Find cu es =(γ, ζ ) :I=[a,b]→Q×R, such ha (1) end poin s condi ions: γ(a)=qa,γ(b)=qb,ζ(a)=0, (2) is an in eg al cu e o ˙z=F(q, ,z):˙ ζ=F(γ ( ), ˙γ( ), ζ( )), o e e y ∈I, and (3) ex eme condi ion: ζ(b)is maximum o e all cu es sa is ying (1) and (2). Obse e ha we ha e conside ed he di e en ial equa ion ˙z=F(q, ,z)depending on he cu es γ. In he case ha he unc ion Fdoes no depend on he a iable z, ha is F:TQ→R, hen he di e en ial equa ions is ˙z=F(γ, ˙γ), hence by in eg a ion, he p oblem is he classical a ia ional one de ined by: ind he cu es γ( )minimizing S[γ]=b a F(γ ( ), ˙γ( )d wi h ini ial condi ions γ(a)=qa,γ(b)=qb. 123 9Page 24 o 46 Jou nal o Nonlinea Science (2023) 33 :9 As we know, He glo z ob ained ha he cu es γsolu ion o his p oblem sa is y he so-called gene alized Eule –Lag ange equa ions d d ∂F ∂ iγ −∂F ∂qi−∂F ∂z ∂F ∂ i=0. In his sec ion we will ob ain hese di e en ial equa ions as an applica ion o he Pon yagin maximum p inciple o a sui able op imal con ol p oblem associa ed o he He glo z a ia ional p oblem. To do so, we begin by gi ing a geome ic s a emen o he He glo z p oblem. Gi en he unc ion F:TQ ×R:→ R, conside he igh up iangle o he ollowing diag am TR τo TQ×R τQ×IR π2 Z R I ˆ  Q×R whe e Z∈X(R,π 2)is he ec o ield on Ralong he p ojec ion π2de ined by Z=F∂ ∂z. Now aking he ull diag am, we ha e he ollowing p oblem associa ed wi h he ec o ield Z Geome ic He glo z a ia ional p oblem: Find cu es :I=[a,b]→Q×R, =(γ, ζ), such ha (1) end poin s condi ions: (a)=(qa,0), γ (b)=qb, (2) is an in eg al cu e o Z:˙ ζ=F(˜ ( )), o e e y ∈I, whe e ˜ =(γ = (γ, ˙γ),ζ), and (3) ex eme condi ion: ζ(b)is maximum o e all cu es sa is ying (1) and (2). Ob iously he wo abo e p oblems a e equi alen . The di e ence is only in he lan- guage used o s a e hem. 6.2 Op imal Con ol App oach o he He glo z Va ia ional P oblem Associa ed o he unc ion F:TQ×R:→ R, conside he ollowing diag am T(Q×R) τQ×R TQ×RτQ×IR Y Q×R 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 25 o 46 9 whe e Yis he ec o ield on Q×Ralong he p ojec ion τQ×IR:TQ×R→Q×R de ined by Y((q, ),z)=((q,z), , F), which in local coo dina es Yis gi en by Y= i∂ ∂qi+F∂ ∂z. This ec o ield co esponds o he sys em o o dina y di e en ial equa ions: ˙qi= i,˙z=F(qi, i,z). Obse e ha he i s sumand o he ec o ield Yis a canonical ec o ield along he p ojec ion τQ:TQ →Q, i co esponds o he iden i y map ITQ :TQ →TQ. Hence he ec o ield Yis associa ed in a na u al way o he unc ion F. These elemen s de ine a con ol sys em wi h ec o ield Y ∈X(Q×R,τQ×IR), on he s a e space Q ×R, and wi h he ib es o TQas he se o con ols; ha is o e e y s a e (q,z)∈Q×R, he con ols a e he elemen s ∈TqQ. On his con ol sys em we s a e he ollowing op imal con ol p oblem: Conside he diag am T(Q×R) τQ×R TQ×RτQ×IR Y Q×R I  ˜   whe e, i =(γ, ζ), hen ˜ =(γ ,ζ)=((γ, ˙γ),ζ). Fo a cu e :I→Q×R, we ake i s canonical li ing o he angen bundle, :I→T(Q×R), ha is: i =(γ, ζ) hen =((γ, ζ ), ( ˙γ, ˙ ζ)). We say ha a cu e is an in eg al cu e o he ec o ield Yi : =Y◦˜ , ( ˙γ,˙ ζ) =Y(γ, ˙γ,ζ)=( i(γ, ˙γ),F(γ, ˙γ,ζ)), which, in local coo dina es, is a solu ion o he abo e sys em o di e en ial equa ions: ˙qi= i,˙z=F(qi, i,z). Hence we ha e he op imal con ol p oblem gi en by: Op imal con ol p oblem associa ed o He glo z a ia ional p oblem: Find cu es :I=[a,b]→Q×R,=(γ, ζ ), such ha (1) end poin s condi ions: (a)=(qa,0), γ (b)=qb, (2) is an in eg al cu e o Y:( )=Y(˜ ( )), o e e y ∈Iand (3) op imal condi ion: ζ(b)is maximum o e all cu es sa is ying (1) and (2). 123 9Page 32 o 46 Jou nal o Nonlinea Science (2023) 33 :9 whe e Z=X+Y, ha is Z=Xi∂ ∂xi+F∂ ∂z, locally. And he cu es a e =(γM,γ z), γ=(, γU)=(γM,γ z,γ U). Then we ha e ano he equi alen s a emen : He glo z op imal con ol p oblem: Find cu es γ:I=[a,b]→M×R×U, γ=(γM,γ z,γ U),=(γm,γ z)such ha (1) end poin s condi ions: γM(a)=xa,γM(b)=xb,γ z(a)=0, (2) Mis an in eg al cu e o Z:˙ M=Z◦γ, and (3) op imal condi ion: γz(b)is maximum o e all cu es sa is ying (1), and (2). Condi ion (2) is w i en as (˙γ,˙γz)=Z◦γ, ha is ˙xi=Xi(x,z,u), ˙z=F(x,z,u). (45) which a e he same se o di e en ial equa ions as Eq. (44). Hence bo h p oblems a e equi alen . In he sequel we e e o his second o m. Obse e ha wi h his app oach, we ha e a classical op imal con ol p oblem and we can ind i s solu ion ollowing he me hod o Sec . 3, in pa icula by applying he weak p esymplec ic o m o he Pon yagin maximum p inciple, Theo em 5.In his case, he unc ion o op imize is one o he di ec ions o s a e space which is gi en by z. We begin, as usual, by ex ending he ec o ield, hence ob aining he ex ended sys em adding a new a iable xo o he a iable z o maximize. The new ec o ield is X=F∂ ∂xo+Xi∂ ∂xi+F∂ ∂z∈X(R×M×R). Then he associa ed Hamil onian is H(xo,po,xi,pi,z,pz,u)=poF+piXi+pzF, de ined on he mani old T∗(R×M×R)×U. The p esymplec ic o m is ω= dxo∧dpo+dxi∧dpi+dz∧dpz, wi h ke nel gi en by he angen ec o ields o U, and he Hamil onian ec o ield XH, solu ion o he equa ion iXHω=dH, is locally gi en by XH=F∂ ∂xo+0∂ ∂po +Xi∂ ∂xi+F∂ ∂z −po ∂F ∂xi+pj ∂Xj ∂xi+pz ∂F ∂xi∂ ∂pi −po ∂F ∂z+pi ∂Xi ∂z+pz ∂F ∂z∂ ∂pz +Aa∂ ∂ua, whe e he las e m co esponds o he ke nel o ω. Obse e ha his solu ion exis s all o e he mani old T∗(R×M×R)×Uand ha pois cons an o e e y cu e solu ion o he p oblem. 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 33 o 46 9 Being a p esymplec ic sys em, he compa ibili y equa ions a e gi en by i(Z)dH= 0 o e e y Z∈ke ω, ha is equa ions ∂H ∂u1=0,..., ∂H ∂uk=0 (46) which, oge he wi h he equa ions coming om he ec o ield XH,gi eusase o equa ions o sol e he op imal con ol p oblem. Recall ha hese compa ibili y condi ions a e he same ha he op imali y ones. As in o dina y op imal con ol p oblems, suppose ha he compa ibili y equa- ions allow us o de e mine he con ols u1,...,uk, ha is we can ob ain ua= ψa(xo,xi,po,pi), hen we say ha he op imal con ol p oblem is egula ,o h- e wise i is called singula . In he singula case, i is necessa y o apply an algo i hm o cons ain s, ha is o go o highe o de condi ions, o ob ain he con ols pe haps on a submani old o T∗(R×M×R)×U. The di e en ial equa ions associa ed wi h he abo e ec o ield XH, oge he wi h equa ions (46) a e he solu ion equa ions o he He glo z op imal con ol p oblem. Rema k 5 To unde s and he signi icance o hese equa ions, we can compa e he abo e se o equa ions wi h he co esponding ones o a classical op imal con ol sys em. Apa om he compa ibili y condi ions, which a e he same, he ec o ield solu ion, see Theo em 5, was gi en by XN=F∂ ∂xo+Xi∂ ∂xi−λo ∂F ∂xi+pj ∂Xj ∂xi∂ ∂pi .(47) Compa ing his ec o ield XNwi h he abo e XH, in his las we ha e a new a i- able, z, hence wo new e ms, one o ˙zand he o he o ˙pz. Mo eo e , he e m co esponding o pihas changed. Bu i he cos unc ion Fdoes no depend on z, hen we ha e ha ˙pz=0, hence pz=cons an , and bo h equa ions, he classical and he He glo z op imal con ol, a e he same. In ac in his las case, we can change he di e en ial equa ion ˙z=F(x,z,u) and he op imali y condi ion by he in eg al o be op imized b a F(x,u)d and we ob ain exac ly he classical p oblem. Hence, as we p oposed a he beginning o he sec ion, we ac ually ha e a gene al- iza ion o he classical op imal con ol p oblem om he poin o iew o he equa ions sol ing he p oblem. 7.3 Con ac Fo mula ion o he No mal Solu ions We can analyze he se o no mal solu ions, ha is po=/0, in he aim o Sec . 4.2.2 and ob ain hese solu ions as in eg al cu es o con ac dynamical sys ems. 123 9Page 34 o 46 Jou nal o Nonlinea Science (2023) 33 :9 To p oceed suppose we a e in he egula si ua ion, ha is he maximali y condi ions allows us o de e mine he con ols. To s udy his si ua ion we can ix he con ols, hey a e de e mined by he las equa ions solu ion o he p oblem, and analyze he o he equa ions as solu ions o a symplec ic p oblem. Then, once ixed u=uo, ou mani old is T∗(R×M×R). In his mani old we can analyze he p oblem as a con ac dynamical sys em. Fo a gi en λo∈R,λo=/0, conside he submani old Nλo⊂T∗(R×M×R), gi en by po=λoand he na u al injec ion jλo:Nλo→T∗(R×M×R).Le η=−j∗ λoθ∈1(Nλo). Then we ha e Lemma 4 Fo e e y ixed u ∈U, he mani old (Nλo,η)is a con ac mani old. I s Reeb ec o ield is gi en by Rλo=−1 λo ∂ ∂xo The p oo is s aigh o wa d using he local exp ession o η η=−λodxo−pidxi−pzdz. Le HNλo=j∗ λoHand conside he Hamil onian con ac sys em gi en by (Nλo,η,HNλo).Le Z∈X(Nλo) he co esponding Hamil onian ec o ield, ha is he solu ion o he con ac equa ions i(Z)η =−HNλo,i(Z)dη=dHNλo−(L(Rλo)HNλo)η whose local exp ession is XH=F∂ ∂xo+0∂ ∂po +Xi∂ ∂xi+F∂ ∂z −po ∂F ∂xi+pj ∂Xj ∂xi+pz ∂F ∂xi∂ ∂pi −po ∂F ∂z+pi ∂Xi ∂z+pz ∂F ∂z∂ ∂pz . Wi h he abo e exp essions and commen s we ha e p o en he Theo em 11 The no mal solu ions o he p oblem 7.2 co esponding o po=λo=/0 a e he p ojec ions o R×M×R×U o he cu es solu ion o he con ac Hamil onian p oblem gi en by (Nλo,η,HNλo). The co esponding di e en ial equa ions o he cu es solu ion o his Hamil onian con ac p oblem a e : ˙xo=F ˙xi=Xi ˙z=F 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 35 o 46 9 ˙pi=−po ∂F ∂xi−pj ∂Xj ∂xi−pz ∂F ∂xi ˙pz=−po ∂F ∂z−pi ∂Xi ∂z−pz ∂F ∂z Toge he wi h he maximiza ion condi ion, ha is he cons ain s ob ained om he compa ibili y o he p esymplec ic equa ion ∂H ∂u1=0,......, ∂H ∂uk=0 7.4 Reduc ion o he P oblem We ema k ha his p oblem is a gene aliza ion o He glo z a ia ional p inciple. On he p e ious sec ion, we showed ha he equa ions ob ained h ough he Pon yagin maximum p inciple could be educed o ob ain he He glo z equa ion. In his sec ion, we show ha a simila educ ion can be applied in his mo e gene al case. We see om he di e en ial equa ions abo e ha , aking he same ini ial condi ion o bo h a iables, we will ha e xo=z o he solu ions o p oblem 7.2. Then one o hem is i ele an o he p oblem, we can elimina e i . As he momen um co esponding o xois cons an , we can elimina e he pai (xo,po). Obse e ha , in ac , pzis also i ele an o he p oblem. Indeed, we can educe he dimension o he s a e space o he p oblem; his new mani old is wha we will now cons uc . Conside he Hamil onian H0:W0=T∗M×R×U→R, (xi,pi,z,ua)→ piXi(xi,z,ua)−p ∗ 1F(x,z,u), (48) whe e p 1:T∗M×R×U→M×R×Uis he na u al p ojec ion. Also conside he canonical con ac o m on T∗M×R η0=dz−pidxi.(49) Theo em 12 The no mal solu ions o he p oblem 7.2 co esponding o po=λoa e he p ojec ions o R×M o he cu es solu ion o he con ac Hamil onian p oblem gi en by (T∗M×R,η 0,H0). The equa ions o mo ion o he a o emen ioned Hamil onian p oblem a e ˙xi=Xi,(50a) ˙pi=pi ∂F ∂z−pj ∂Xj ∂xi+∂F ∂xi−∂Xj ∂zpipj,(50b) ˙z=F(50c) 123 9Page 36 o 46 Jou nal o Nonlinea Science (2023) 33 :9 subjec ed o he cons ain s ∂H ∂ua=∂F ∂ua−pj ∂Xj ∂ua=0.(50d) Rema k 6 In he case ha he p oblem is singula , one would wo k ins ead wi h he p econ ac sys em (T∗M×R×U,η 0,H0), applying he app op ia e cons ain algo- i hm. P oo Le γbe a solu ion o he He glo z op imal con ol p oblem. By Theo em 11, we know ha he e exis s a solu ion cu e σo he co esponding con ac sys em on Nλ0. In o de o p o e his heo em, we will p ojec σon o a solu ion o he sys em (T∗M×R,η 0,H0). Fi s o all, no ice ha he solu ions sa is y x0=z, hence σwill lie on he subman- i old j:˜ Nλ0→Nλ0de ined by x0=z. The dynamical ec o ield Xo he p econ ac sys em (Nλ0,η λ0,Hλ0)is angen o he submani old ˜ Nλ0. Indeed, he es ic ion o X o ˜ Nλ0a e jus he equa ions o mo ion ˜ Xo he induced p econ ac sys em (˜ Nλ0,˜ηλ0=j∗ηλ0,˜ Hλ0=j∗Hλ0).In coo dina es ˜ηλ0=(−λ0−pz)dz−pidxi,(51a) ˜ Hλ0=(λ0+pz)F+piXi(51b) Conside he ollowing commu a i e diag am, ˜ Nλ0 W0R×M×R M×R τ λ0 τ0π1 (52) whe e λ0xi,z,pi,pz=xi,−(λo+pz)pi,z.(53) No ice ha λ0is a subme sion and a con o mal equi alence o p econ ac sys ems: ∗ λ0η0=−(λo+pz)˜ηλ0,(54a) ∗ λ0H0=−(λo+pz)˜ Hλ0,(54b) By Theo em 1p ojec ions o he solu ion cu es o he p econ ac sys em on ˜ Nλ0a e solu ion cu es o he con ac sys em on TM×R. 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 37 o 46 9 As a consequence o his heo em, we can ob ain again he He glo z equa ions. Conside he He glo z p oblem in Sec . 6.1 o a Lag angian L:TQ×R→R. No ice ha his p oblem is a pa icula case o he He glo z op imal con ol p oblem, whe e •Con ols a e he eloci ies ua= i. •The cos unc ion is he Lag angian F=L. •The con ol equa ion is X= i∂ ∂xi. The solu ions o his p oblem a e gi en by Theo em 12: ˙qi= i,(55) ˙pi=pi ∂L ∂z+∂L ∂qi(56) ˙z=L(57) wi h he cons ain s ∂L ∂ i=pi,(58) which a e p ecisely He glo z equa ions. 8 Applica ion: Op imal Con ol on The modynamic Sys ems One possible applica ion o his heo y is he s udy o he modynamic p ocesses which minimize o maximize some he modynamic po en ial. As an example, we apply ou o malism o he con ol sys ems conside ed in Van de Scha and Maschke (2017). The ela ion be ween symplec ic and con ac mani olds ia he symplec i ica ion p ocedu e has pe mi ed o go deepe in he geome ic desc ip ion o he modynamic sys ems. This way has been explo ed in Balian and Valen in (2001) (see also A nold 1978; Libe mann and Ma le 1987; Ibáñez e al. 1997). 8.1 Homogeneous Hamil onian Sys ems and Con ac Sys ems The e is a close ela ionship be ween homogeneous symplec ic and con ac sys ems, see o example Van de Scha and Maschke (2017) whe e his ela ion is s udied. He e we b ie ly ecall he ideas we need o ollow he example. In he gene al case, i π:M→Bis a ec o bundle, a unc ion F:M→R is homogeneous i , o any ep∈Mp=π−1(p)wi h π(ep)=p∈B,weha e F(λep)=λF(ep). In his si ua ion he unc ion Fcan be p ojec ed o he p ojec i e bundle P(M)o e Bob ained by p ojec i iza ion on e e y ib e. We a e in e es ed in he case ha M=T∗(Q×R)→Q×R, wi h na u al coo dina es (qi,z,Pi,Pz) Le Hbe an homogeneous Hamil onian unc ion on T∗(Q×R). Locally, we ha e ha H(qi,z,λPi,λPz)=λH(qi,z,Pi,Pz), o all λ∈R. Equi alen ly, one can 123 9Page 38 o 46 Jou nal o Nonlinea Science (2023) 33 :9 w i e Hqi,z,Pi,Pz=−Pzhqi,−Pi/Pz,z,(59) o Pz=/0, whe e h:T∗Q×R→R,h(qi,pi,z)=H(qi,z,−pi,−1)is well de ined. Wi h he abo e changes, we ha e iden i ied he mani old T∗Q×Ras he p ojec i e bundle P(T∗(Q×R)) o he co angen bundle T∗(Q×R) aking ou he poin s a in ini y, ha is he subse de ined by {Pz=0}. Following Van de Scha and Maschke (2017, Sec ion 4.1), he map :T∗(Q×R) {pz=0}→T∗Q×R (qi,z,Pi,Pz)→(qi,Pi/Pz,z)=(qi,pi,z), (60) sends he Hamil onian symplec ic sys em (T∗(Q×R) {pz=0},ωQ×R,H)on o he Hamil onian con ac sys em (T∗Q×R,ηQ,h), whe e ωQ×R=dqi∧dPi+dz∧dPz and ηQ=dz−pidqi. Obse e ha he na u al coo dina es o T∗Q×R, deno ed by (qi,pi,z), co espond o he homogeneous coo dina es in he p ojec i e bundle. In ac , he map is he p ojec i iza ion; i.e., he map ha sends each poin in he ibe s o T∗(Q×R) o he line ha passes h ough i and he o igin. I can be shown ha p o ides a bijec ion be ween con o mal con ac omo - phisms and homogeneous symplec omo phisms. Mo eo e , maps homogeneous Lag angian submani olds L⊆T∗(Q×R)on o Legend ian submani olds L= φ(L)⊆T∗Q×R. See Van de Scha and Maschke (2017) and Sec . 8.3 o mo e de ails on his opics. 8.2 Con ol o Con ac Sys ems On he con ac na u al mani old T∗Q×R, wi h coo dina es (qi,pi,z), assume ha we a e gi en a pa ame ized amily o Hamil onians h:T∗Q×R×U→R,U⊂Rk, wi h Hamil onian con ac ec o ields Xhu, whe e hu(qi,z,pi)=h(qi,z,pi,u). Then we can de ine he con ol sys em Z(q,p,z,u)=Xhu(q,p,z), whe e he ollowing diag am is commu a i e: T(T∗Q×R) τT∗Q×R T∗Q×R×U Z πT∗Q×R I γ 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 39 o 46 9 A cu e γ:I→T∗Q×R×Uis an in eg al cu e o Z, ha is =Z◦γ,i in local coo dina es sa is ies he di e en ial equa ions dqi d =∂hu ∂pi , dpi d =− ∂hu ∂qi−pi ∂hu ∂z, dz d =pi ∂hu ∂pi −hu. One can conside he He glo z op imal con ol p oblem gi en by Z, as we s a ed in Sec . 7.2. Then, by Theo em 12, we know ha he no mal solu ions a e he p ojec ions o he solu ions o he con ac sys em (T∗(T∗Q)×R,η T∗Q,H), whe e H=pqi ∂hu ∂pi −ppi ∂hu ∂qi−pi ∂hu ∂z−pi ∂hu ∂pi +hu.(61) 8.3 Applica ion o The modynamic Sys ems We conside he modynamic sys ems in he so-called en opy ep esen a ion. Hence he he modynamic phase space, ep esen ing he ex ensi e a iables, is he mani old T∗Q×R, equipped wi h i s canonical con ac o m ηQ=dS−pidqi.(62) The local coo dina es on he con igu a ion mani old Q×Ra e (qi,S), whe e Sis he o al en opy and qi’s deno e he es o ex ensi e a iables. O he a iables, such as he in e nal ene gy, may be chosen ins ead o he en opy, by means o a Legend e ans o ma ion. The s a e o a he modynamic sys em always lies on he equilib ium submani old L⊆T∗Q×R, which is a Legend ian submani old, ha is, η|TL=0 and dim L= dim Q=n. The pai (T∗Q×R,L)is a he modynamic sys em. The equa ions (locally) de ining La e called he s a e equa ions o he sys em. On a he modynamic sys em (T∗Q×R,L), one can conside he dynamics gene - a ed by a Hamil onian ec o ield XHassocia ed o a Hamil onian h. I his dynamics ep esen s quasis a ic p ocesses, meaning ha a e e y ime he sys em is in equilib- ium, ha is, i s e olu ion s a es emain in he submani old L, i is equi ed o he con ac Hamil onian ec o ield Xh o be angen o L. This happens i and only i h anishes on L. Equi alen ly, by Sec . 8.1, one can conside he ex ended he modynamic phase space T ∗(Q×R)wi h i s canonical symplec ic o m ωQ×R=dqi∧dPi+dS∧dPS.(63) 123 9Page 40 o 46 Jou nal o Nonlinea Science (2023) 33 :9 In his o mula ion, a he modynamic sys em is a uple (T∗(Q×R), L)), whe e L is a homogeneous Lag angian submani old. Dynamics a e gi en by a homogeneous Hamil onian K. See Van de Scha and Maschke (2017) o de ails and ecall we ha e iden i ied, in Sec . 8.1, he bundle T∗Q×Rwi h he p ojec i e bundle P(T∗(Q×R)). Po - he modynamic sys ems we e in oduced in Van de Scha and Maschke (2017), bu in a homogeneous symplec ic o malism. De ini ion 1 (Po - he modynamic sys em)Apo - he modynamic sys em on T∗(Q× R)is de ined as a pai (L,K), whe e he homogeneous Lag angian submani old L⊂ T∗(Q×R)speci ies he s a e p ope ies. The dynamics is gi en by he homogeneous Hamil onian dynamics wi h pa ame ized homogeneous Hamil onian K:= Ka+ Kc αuα:T∗(Q×R)→R,u∈Rk,Kc:T∗(Q×R)→Rk, wi h Ka,Kcbo h equal o ze o on he poin s o L, and Kaas he in e nal Hamil onian. One need he addi ional condi ion ∂K ∂S|L≥0,(64) so ha he second law o he modynamics holds. Using he esul s o Sec . 8.1, we could ins ead conside he ollowing con ac o mula ion. De ini ion 2 (Po - he modynamic sys em, con ac o malism) A po - he modynamic sys em on (T∗Q×R,ηQ)is de ined as a pai (L,h), whe e he Legend ian submani old L⊂T∗Q×Rspeci ies he s a e p ope ies. The dynamics is gi en by he con ac Hamil onian dynamics wi h pa ame ized con ac Hamil onian h=ha+hc αuα: T∗Q×R→R,u∈Rm,hc:T∗Q×R→Rk, wi h ha,hcze o on L, and he in e nal Hamil onian hasa is ying ∂h ∂S|L≥0,(65) so ha he second law o he modynamics holds. Ou heo y p o ides ools o unde s and which o he a ailable he modynamic p ocesses minimize he en opy p oduc ion o he sys em. Obse e ha we can conside p ocesses ha maximize o minimize o he he modynamic a iables, such as he ene gy, ia a Legend e ans o m. 8.4 Example: Gas–Pis on–Dampe Sys em We end his sec ion wi h an explici example which can be ound in Van de Scha and Maschke (2017). Conside an adiaba ically isola ed cylinde closed by a pis on con aining a gas wi h in e nal ene gy U(V,S). The ex ended phase space has he ollowing ex ensi e a iables • he momen um o he pis on π, 123 Jou nal o Nonlinea Science (2023) 33 :9 Page 41 o 46 9 • he olume o he gas V, • he ene gy E, • he en opy S. They co espond o Q×Rwi h local coo dina es (V,π,E,S). The Legend ian sub- mani old is gi en by L=(V,π,E,pV,pπ,pE,S)|E=π2 2m+U(S,V), pV=−pE ∂U ∂V, pπ=−pE π m,pE=1/∂U ∂S(66) The ene gy is hen gi en by h=pV π m+pπ−∂U ∂V−dπ m−d(π m)2 ∂U ∂S +pπ+pE π mu,(67) whe e dis he diame e o he pis on and mis i s mass. The Hamil onian ec o ield is gi en by Xh=π m ∂ ∂V+−πd m+u−∂U ∂V∂ ∂π +πu m ∂ ∂E +⎛ ⎝⎛ ⎝pπ ∂2U ∂V∂S−π2d∂2U ∂S2 m2∂U ∂S2⎞ ⎠pV+pπ ∂2U ∂V2−π2d∂2U ∂V∂S m2∂U ∂S2⎞ ⎠ ∂ ∂pV +⎛ ⎝⎛ ⎝pπ ∂2U ∂V∂S−π2d∂2U ∂S2 m2∂U ∂S2⎞ ⎠pπ+dp π m−pEu m−pV m+2πd m2∂U ∂S⎞ ⎠ ∂ ∂pπ +⎛ ⎝pπ ∂2U ∂V∂S−π2d∂2U ∂S2 m2∂U ∂S2⎞ ⎠pE ∂ ∂pE +π2d m2∂U ∂S∂ ∂S(68) We cons uc he con ac Hamil onian sys em (T∗(T∗Q)×R,η T∗Q,H)as in (61): H=− dp π m−pEu m−pV m+2πd m2∂U ∂SPπ −pπ ∂2 (∂V)2U(V,S)−π2d∂2 ∂V∂SU(V,S) m2∂U ∂S 2PV −πd m−u+∂ ∂VU(V,S)Ppπ+πPpEu m+πPpV m−π2d m2∂U ∂S , (69) 123