Contents lists available atScienceDirect
Journal
of
Functional
Analysis
www.elsevier.com/locate/jfa
Martingale
representation
for
degenerate
diffusions
A.S. Üstünel
Bilkent University,Math.Dept.,Ankara,Turkey
a r t i c l e i n f o a bs t r a c t
Article history:
Received 10 May 2018 Accepted 14 December 2018 Available online 17 December 2018 Communicated by L. Gross Keywords: Entropy Degenerate diffusions Martingale representation Relative entropy Innovation process
Causal Monge–Ampère equation
Let(W,H,μ) betheclassicalWienerspaceonIRd.Assume
thatX = (Xt) isadiffusionprocesssatisfyingthestochastic
differential equation dXt = σ(t,X)dBt+ b(t,X)dt, where
σ : [0,1]×C([0,1],IRn)→ IRn⊗IRd,b: [0,1]×C([0,1],IRn)→
IRn,B isan IRd-valuedBrownian motion.Wesupposethat
the weak uniqueness of this equation holds for any initial condition.WeprovethatanysquareintegrablemartingaleM w.r.t.tothefiltration(Ft(X),t∈ [0,1]) canberepresentedas
Mt= E[M0] +
t
0
Ps(X)αs(X).dBs
whereα(X) isanIRd-valuedprocessadaptedto(F t(X),t∈
[0,1]),satisfyingE0t(a(Xs)αs(X),αs(X))ds<∞,a= σσ
and Ps(X) denotes a measurable version of the orthogonal
projection from IRd to σ(X
s)(IRn). In particular, for any
h∈ H,wehave E[ρ(δh)|F1(X)] = exp ⎛ ⎝ 1 0 (Ps(X) ˙hs, dBs) −1 2 1 0 |Ps(X) ˙hs|2ds ⎞ ⎠ , (0.1)
where ρ(δh) = exp(01( ˙hs,dBs)− 12 |H|2H). In the case
the process X is adapted to the Brownian filtration, this E-mailaddress:ustunel@fen.bilkent.edu.tr.
https://doi.org/10.1016/j.jfa.2018.12.004
result gives a new development as an infinite series of the L2-functionals of the degenerate diffusions. We also
give an adequate notion of “innovation process” associated to a degenerate diffusion which corresponds to the strong solutionwhentheBrownianmotionisreplacedbyanadapted perturbationofidentity.Thislatterresultgivesthesolution ofthecausalMonge–Ampèreequation.
©2018ElsevierInc.Allrightsreserved.
1. Introduction
Therepresentationof randomvariables with thestochasticintegralswith respect to somebasicprocesseshasalonghistoryandalsoithasveryimportantapplications,e.g., insignal theory, filtering, optimal control, finance, stochastic differential equations, in physics,etc.Letusrecallthequestion:assumethatwearegivenacertainsemimartingale
X indexedby[0,1] forexample.LetF(X)= (Ft(X),t∈ [0,1]) denoteitsfiltration.The
question is under which conditions can we represent any martingale adapted to the filtrationofX as astochastic integralw.r.t. afixed martingaleof thefiltrationF(X)? For the case of Wiener process, this questionhas a very longhistory and it is almost impossibletogiveanexactaccountofthecontributingworks.Toourknowledge,ithas been answeredforthefirst timeintheworkof K.Itô,cf. [10,15]. In[2], C.Dellacherie hasgiven adifferent point of viewto provetherepresentation theorem for theWiener andPoissonprocesses,basedontheuniquenessoftheirlaws.Thecaseofnondegenerate diffusionprocesseshasbeenelucidatedin[12],alsostudiedin[3,11] (cf.alsothereferences there) with also some remarks about the degenerate case.The general case, using the notionofmultiplicityisgivenin[1].
Although the nondegenerate caseis completely settled, thedegenerate case hasnot been solved in a definitive way, in the sense that a unique minimal martingale with respect to which the above mentioned property of representation holds, has not been discovered. In this workwe are doing exactlythis: we prove thatfor adegenerate dif-fusion whose law is unique, there exists a minimal martingale with respect to which everysquareintegrable,F1(X)-measurablefunctionalcanberepresentedasastochastic
integral of an F(X)-adapted process. To do this we first prove the density of a class of F1(X)-measurable stochastic integralsin L2(F1(X)) using the methodlaunched by
C. Dellacherie, then we show that a sequence from this class which is approximating any element of L2(F
1(X)) defines an L2-convergingsequence of processes with values
intherange oftheadjointofthediffusioncoefficient,i.e., withvaluesinσ(t,X)(IRn
). Because of the degeneracy of σ, we can not determine the limit of this sequence but itsimage underthe orthogonalprojection Pt(X): IRd → σ(t,X)(IRn) anditslimit is
perfectlywell-definedandinthisway weseethattheminimalmartingaleforthe repre-sentationofthefunctionalsofthediffusion(Itôprocess)isnothingbutdmt= Pt(X)dBs
Note thateventhe adaptabilityof (mt,t ∈ [0,1]) tothe filtrationF(X) is notevident.
This result of representation gives also existence of non-orthogonal chaos representa-tion of the elements of L2(F
1(X)) as the multiple ordered integrals with respect to
the martingale (mt,t ∈ [0,1]) of elements of {L2(Cn,dt1× · · · × dtn),n ≥ 1}, where Cn ={(t1,. . . ,tn)∈ [0,1]n: t1> . . . > tn}.
Inthethirdsectionwedefinethenotionofinnovationprocessassociated toa degen-eratediffusionprocess1.Inthiscasethedefinitionofinnovationprocessisdifferentfrom theclassicalcaseoftheperturbationofidentitysincewehavetotakeintoaccountalso the actionofthe projectionoperator-valuedprocess(Pt(X),t∈ [0,1]).InSection3we
extendtheinnovationrepresentationtheoremofFujisaki–Kallianpur–Kunita,[8],under the hypothesis of strongexistence and uniquenessto the caseof degenerate diffusions. Thisresultconfirmsthevalidityofthechoicethatwehavedonetodefinetheinnovation process. Inparticular,usingthisinnovationprocesswecancalculatetheconditional ex-pectation ofthe Girsanovexponential of an adapteddrift u with respect to the sigma algebraF1(XU),whereXU isthesolutionofthediffusionstochasticdifferentialequation
where therandominputisequalto U = B + u.
TheresultsofthethirdsectionisthenappliedtothesolutionoftheadaptedMonge– Ampère equation inthe caseof thedegenerate diffusions,which extendsthe resultsof [7,13,14], cf. also[4–6].Inparticularwe calculatetherelative entropyofthelawof XU
with respecttothelawofX by theuseofprecedingresults.
Letusnotetofinishthisintroductionthatmostoftheseresultsareeasilyextendibleto moregeneralsituations,forexamplethestrongexistenceanduniquenesshypothesiscan beweakenedintheentropiccalculations,wehavetriedtofollowmaximumhomogeneity inthehypothesisandthesepossible extensionsmaybetreatedinseparate works. 2. Stochasticintegral representationoffunctionalsofdiffusions
Let X = (Xt,t ∈ [0,1]) be a weak solution of the following stochastic differential
equation:
dXt= b(t, X)dt + σ(t, X)dBt, X0= x, (2.2)
where B = (Bt,t ∈ [0,1]) is an IRd-valued Brownian motion and σ : [0,1] × C([0,1],IRn) → L(IRd,IRn) and b : [0,1]× C([0,1],IRn)→ IRn are measurable maps, adapted to the naturalfiltration of C([0,1],IRn) and of linear growth.Recall that the classical theorem of Yamada–Watanabe says that the strong uniqueness implies the uniqueness inlaw of theaboveSDE, cf. [16,9]. Hence weakuniquenessis easier to ob-tain in the applications. In this section we shall assume the weak uniqueness of the equation andprovethemartingalerepresentationpropertyforthecasewhereσ maybe degenerate.
Moreprecisely, let(Ft(X),t ∈ [0,1]) be thefiltration of X andlet us denote by K
thesetofIRn-valued,(Ft(X),t∈ [0,1])-adaptedprocessesα(X),s.t.
E 1 0 (a(s, X)αs(X), αs(X))ds <∞ , wherea(s,w)= σ(s,w)σ(s,w),s∈ [0,1],w∈ C([0,1],IRn). Theorem1.ThesetΓ={N ∈ L2(F
1(X)): N = E[N ]+
1
0(αs(X),σ(s,X)dBs),α∈ K}
isdensein L2(F 1(X)).
Proof. Suppose that there is some M ∈ L2(F
1(X)) which is orthogonal (in L2) to Γ.
Usingthe usualstoppingtechnique,we canassumethatthecorresponding (Ft(X),t∈
[0,1])-martingaleisboundedandpositivewhoseexpectationisequaltoone.The orthog-onalityimpliesthat(Mt(f (Xt)−
t
0Lf (s,X)ds),t∈ [0,1]) isagaina(local)martingale
foranysmoothfunctionf : IRn→ IR,whereL isdefinedas
Lf (t, X) = 1 2 i,j ai,j(t, X)∂i,jf (Xt) + i bi(t, X)∂if (Xt) . (2.3)
Thisimplies,with PropositionIV.2.1of[9],that,underthemeasureM· P ,X isagain aweak solution of 2.2. Bythe uniqueness inlaw, we get X(M · P )= X(P ), since M
is F1(X)-measurable, we should have M = 1. Hence the functionals of the diffusion
whichareorthogonaltotheabovesetofstochasticintegralsarealmost surelyconstant. Consequently,thesetΓ istotalinL2(F
1(X)). 2
Thefollowing isthe extensionofthemartingalerepresentationtheoremtothe func-tionalsofdegeneratediffusions:
Theorem 2.Denote by Ps(X) a measurable version of the orthogonal projection from
IRd ontoσ(s,X)(IRn)⊂ IRd andletF ∈ L2(F
1(X)) be any randomvariable with zero
expectation.Thenthereexistsaprocessξ(X)∈ L2
a(dt× dP ;IRd),adapted to(Ft(X),t∈ [0,1]), suchthat F (X) = 1 0 (Ps(X)ξs(X), dBs)IRd= 1 0 (ξs(X), Ps(X)dBs)IRn a.s.
Conversely,any stochasticintegralof theform
1
0
where ξ(X) is an (Ft(X),t ∈ [0,1])-adapted, measurable process with E01|Ps(X)ξs(X)|2ds<∞,givesrise toanF1(X)-measurablerandomvariable.
Proof. From Theorem 1, there exists a sequence (Fn(X),n ≥ 1) ⊂ Γ, where Γ is
de-fined inthe statementof Theorem 1, convergingto F (X) in L2. Wecan suppose that
E[Fn(X)]= 0,foranyn≥ 1.Hence
Fn(X) = 1 0 (γsn(X), σ(s, X)dBs)IRd= 1 0 (σ(s, X)γsn(X), dBs)IRd.
As explained above γn is an (Ft(X),t ∈ [0,1])-adapted, IRn-valued process satisfying E01(a(Xs)γsn(X),γsn(X))ds<∞. SinceFn(X)→ F (X) inL2, lim n,m→∞E 1 0 |σ(s, X)γn s(X)− σ(s, X)γsm(X)|2ds = 0 . Let αn s(X) = σ(Xs)γsn(X), as Ps(X)αns(S) = Ps(X)σ(Xs)γsn(X) = σ(Xs)γsn(X), (Ps(X)αns(X),n ≥ 1) converges to some ξs(X) in L2a(ds× dP ;IR d ). As Ps(X) is an
orthogonal projection, (Ps(X)αns(X),n≥ 1) converges to Ps(X)ξs(X) alsoin L2a(ds× dP ;IRd).Therefore 1 0 Ps(X)ξs(X).dBs= lim n 1 0 Ps(X)αns(X).dBs = lim n 1 0 (γns(X), σ(s, X)dBs) = lim n Fn(X) = F (X) .
Let now G ∈ L2(P ) be given by G= 1
0(Ps(X)ηs(X),dBs) and assumethat it is not
F1(X)-measurable.Then G− E[G|F1(X)] is orthogonal toL2(F1(X)).It follows from
thefirstpartofthetheoremthatwecanrepresentE[G|F1(X)] as
1
0(Ps(X)ξs(X),dBs).
Letus defineh= η− ξ,then theorthogonalitymentionedaboveimpliesthat
E ⎡ ⎣ 1 0 (Ps(X)hs, dBs). 1 0 (αs(X), σ(s, X)dBs) ⎤ ⎦ = 0 for any(Ft(X),t∈ [0,1])-adapted,measurable α suchthatE
1
0(a(s,X)αs,αs)ds<∞.
Remark.Letη beanadaptedprocesssuchthatηsbelongstotheorthogonalcomplement
of σ(IRn) in IRd ds× dP -a.s. Then η + ξ can also be used to representF (X). Hence
ξ(X) isnotuniquebutP (X)ξ(X) isalwaysunique.
Theorem3.Let ˙u∈ L2(dt× dP,IRd) beadaptedtotheBrownianfiltration,thenwehave
E ⎡ ⎣ 1 0 ( ˙us, dBs)|F1(X) ⎤ ⎦ = 1 0 (E[Ps(X) ˙us|Fs(X)], dBs) (2.4) almostsurely.
Proof. Wehavetoprovefirstthattherighthandsideof(2.4) isF1(X)-measurable.We
knowfromTheorem 2thattheleftsideof(2.4) canbe representedas
E ⎡ ⎣ 1 0 ( ˙us, dBs)|F1(X) ⎤ ⎦ = 1 0 Ps(X)ξs(X).dBs for some ξ ∈ L2 a(dt× dP ;IRd). Let F (X) = 1 0 Ps(X)αs(X).dBs be any element of L2(F1(X)) withα(X)∈ L2a(dt× dP ;IRd).Wehave E ⎡ ⎣ ⎛ ⎝ 1 0 ˙us.dBs ⎞ ⎠ F (X) ⎤ ⎦ = E 1 0 ( ˙us, Ps(X)αs(X))ds = E 1 0 (Ps(X)E[ ˙us|Fs(X)], Ps(X)αs(X))ds = E ⎡ ⎣ ⎛ ⎝ 1 0 Ps(X)ξs(X).dBs ⎞ ⎠ F (X) ⎤ ⎦ = E 1 0 (Ps(X)ξs(X), Ps(X)αs(X))ds ,
hencePs(X)ξs(X)= Ps(X)E[ ˙us|Fs(X)] ds× dP -a.s.,inparticular
1
0
(E[Ps(X) ˙us|Fs(X)], dBs)
is F1(X)-measurable. Theaboveidentification assures then thevalidity ofthe relation
Corollary 1.Leth∈ H1([0,1],IRd) (i.e., the Cameron–Martin space),denote by ρ(δh) theWick exponential exp(01( ˙hs,dBs)−12
1 0 |˙hs| 2ds), thenwe have E[ρ(δh)|F1(X)] = exp ⎛ ⎝ 1 0 (Ps(X) ˙hs, dBs)− 1 2 1 0 |Ps(X) ˙hs|2ds ⎞ ⎠ . Theorem 4.
• Assume that Ft(X)⊂ Ft(B) forany t ∈ [0,1], where (Ft(B),t ∈ [0,1]) represents thefiltration of theBrownian motion. Definethe martingalem= (mt,t∈ [0,1]) as mt=
t
0Ps(X)dBs,then theset
K ={ρ(δm(h)) : h∈ H} is total in L2(F1(X)), where ρ(δm(h)) = exp
1 0( ˙hs, dms)− 1 2 1 0 |Ps(X) ˙hs| 2ds.
In particular, any element F of L2(F
1(X)) canbe written in aunique way as the
sum F = E[F ] + ∞ n=1 Cn (fn(s1, . . . , sn), dms1⊗ . . . ⊗ dmsn) (2.5)
where Cn isthen-dimensionalsimplexin [0,1]n and fn∈ L2(Cn,ds⊗n)⊗ (IRd)⊗n.
• More generally, without the hypothesis Ft(X) ⊂ Ft(B), forany F ∈ L2(F1(X))∩
L2(F
1(B)), theconclusionsof thefirstpartof thetheorem holdtrue.
Proof. Let F ∈ L2(F
1(X)), assume thatF is orthogonal to K,i.e. E[F ρ(δm(h))] = 0
forany h∈ H.FromCorollary1ρ(δm(h))= E[ρ(δh)|F1(X)],hence
E[F ρ(δh)] = E[F ρ(δm(h))] = 0 ,
hence F isalso orthogonalto E = {ρ(δh): h ∈ H},which istotalinL2(F
1(B),where
F1(B) is theσ-algebra generated bythegoverning Brownianmotion, therefore F = 0.
ConsequentlythespanofK isdenseinL2(F
1(X)).Theorem3andCorollary1allowusto
calculatetheconditionalexpectationsofthemultipleWienerintegralsw.r.t.F1(X) and
theresultwillbe multipleiteratedstochasticintegralsw.r.t.m in theformgivenbythe formula(2.5);herewehavetobecarefulastheoperatorvaluedprocess(Ps(X),s∈ [0,1])
is notdeterministic, thesymmetricinterpretationof theIto–Wienerintegralsw.r.t.the Brownian motionisnolongervalid inourcaseandtheyhaveto bewrittenas iterated Itointegrals. 2
Remark1.Inthistheoremwe needtomake thehypothesis Ft(X)⊂ Ft(B) forany t∈
althoughwe haveTheorem 2 andwhen we iterateit we havea similar representation, but,consistingofafinite numberofterms.Itis notpossibleto pushthisprocedureup toinfinitysincewehavenocontrolatinfinity.
Remark2.Let us note that ifthere is no strongsolution to the equation defining the process X,thechaotic representationproperty mayfail.For example,letU beaweak solutionof
dUt= αt(U )dt + dBt, (2.6)
withU0given.Assumethat(2.6) hasnostrongsolution,asitmayhappeninthefamous
exampleof Tsirelson ([9]), i.e., U isnot measurable w.r.t.the sigmaalgebragenerated byB,thenwehavenochaoticrepresentationpropertyfortheelementsofL2(F1(U )) in
termsoftheiterated stochasticintegralsofdeterministic functionsonCn,n∈ IN w.r.t. B;thecontrarywouldimplytheequalityofF1(U ) andofF1(B),whichwouldcontradict
thenon-existenceofstrongsolutions. 3. Innovationprocess
In this section we assume that the SDE 2.2 has unique strong solution. To fix the ideas,we cansuppose thatσ andb satisfythe followingkind ofLipschitz conditionon thepathspace:forξ,η∈ C([0,1],IRn):
|γ(t, ξ) − γ(t, η)| ≤ K sup
s≤t|ξ(s) − η(s)|
forγ beingequaleithertoσ ortob withcorrespondingEuclideannormsatthelefthand side.
Assumethat ˙u∈ L2(dt× dP,IRd) is aprocess adaptedtothe filtrationofBrownian
motion,letU = (Ut,t∈ [0,1]) bedefined asUt= Bt+
t
0 ˙usds.We denoteby X
U the
strongsolutionoftheequation
dXtU = σ(t, XU)dUt+ b(t, XU)dt
= σ(t, XU)(dBt+ ˙utdt) + b(t, XU)dt
SinceB isthecanonicalBrownianmotion,wehaveXU
t = Xt◦U a.s.Moreover,ifη∈ IRn
isanyvector,we have
(σ(t, XU) ˙ut, η) = ( ˙ut, σ(t, XU)η)
= (Pt(XU) ˙ut, σ(t, XU)η)
where Pt(XU) denotestheorthogonal projectionfrom IRd onto σ(t,XU)(IRn). Hence,
in order to define a reasonably useful concept of innovation, we should estimate the perturbation ˙u simultaneously w.r.t.theboth projections,i.e.,with respectto the con-ditional expectation E[·|Fs(XU)] (which is a projection) and also w.r.t. Ps(XU): Let Z = (Zt,t∈ [0,1]) (to avoidtheambiguity we shall alsouse thenotation ZU if
neces-sary) bedefinedas
Zt= Bt+ t
0
( ˙us− E[Ps(XU) ˙us|Fs(XU)])ds , (3.7)
where (Fs(XU),s∈ [0,1]) denotesthefiltrationofXU.
Proposition 1.The process (0tPs(XU)dZs,t ∈ [0,1]) is an (Ft(XU),t ∈ [0,1])-local martingale.
Proof. Assume to begin that |u|2H = 01| ˙us|2ds ∈ L∞(P ). Let us first prove that the
processunderquestionisadaptedtothefiltration(Ft(XU),t∈ [0,1]):FromtheGirsanov
theorem,theprocess(0tσ(s,XU)dU
s,t∈ [0,1]) isadaptedtothefiltration(Ft(XU),t∈
[0,1]),usingthesamemethodasinTheorem2byreplacingB byU andtheprobability
dP by ρ(−δu)dP , we conclude that the process (0tPs(XU)dUs,t ∈ [0,1]), and hence
theprocess(0tPs(XU)dZs,t∈ [0,1]) isadaptedtothefiltration(Ft(XU),t∈ [0,1]).To
show the(local)martingalepropertyitsufficestowrite that
t 0 Ps(XU)dZs= t 0 Ps(XU)dBs+ t 0 Ps(XU) ˙us− E[ ˙us|Fs(XU)] ds
from which the martingale property follows. The general casefollows from a stopping argument. 2
Remark.NotethatZ = ZU isnotaBrownianmotion.
Theorem 5.Let ˙u∈ L2(dt× dP,IRd) besuchthat E[ρ(−δu)]= 1,thenwehave ζt= E[ρ(−δu)|Ft(XU)] = exp ⎛ ⎝− t 0 Ps(XU)E[ ˙us|Fs(XU)]· dZs− 1 2 t 0 |Ps(XU)E[ ˙us|Fs(XU)]|2ds ⎞ ⎠ . Proof. Assumefirstthat|u|2
H =
1 0 | ˙us|
2ds∈ L∞(P ).Let(XU
t ,t∈ [0,1]) bethe(strong)
solution of dXt = σ(Xt)dUt, where dUt = dBt+ ˙utdt and let f be a C2-function on
IRn. Using the Itô formula, we calculate the Doob–Meyer process associated to the semimartingales (ζt) and(f (XtU)):
f ◦ XU, ζ t=− t 0 Df (XsU), σ(s, XU)Ps(XU)E[ ˙us|Fs(XU)] ds . (3.8)
Letf ∈ C2,usingagaintheItôformulaandtherelation(3.8) weget
f (XtU)ζt= t 0 f (XsU)dζs+ t 0 ζs(Lf )(s, XU)ds + t 0 ζs(Df (XsU), σ(s, XU)dUs) − t 0 (Df (XsU), σ(s, XU)Ps(XU)E[ ˙us|Fs(XU)])ds ,
whereLf isdefinedbytherelation2.3.Thereforetheprocess ⎛ ⎝f(XU t )ζt− t 0 (Lf )(s, XU))ζsds, t∈ [0, 1] ⎞ ⎠
is aP -local martingale, therefore, by the uniquenessin law of thesolution, we should have
E[ζ1F (XU) = E[F (X) = E[F (XU)ρ(−δu)]
for any F ∈ Cb on the path space, and the last inequality follows from the Girsanov
theorem. It suffices then to remark that ζ1 is F1(XU)-measurable by Theorem 3 and
again by the Girsanov theorem which allows us to replace B by U . The general case follows from the usual stopping time argument: let Tn = inf(t :
t
0| ˙us|
2ds ≥ n) and
define ˙uns = 1[0,Tn](s) ˙us and letU
n = B +·
0 ˙u
n
sds.Then XU
n
converges almost surely uniformlytoXU,hencelim
nE[·|Ft(XU n
)]= E[·|Ft(XU)] dt-almostsurelyasbounded
operators on L1(P ) and (P
s(XU n
),n ≥ 1) converges to Ps(X)ds× dP -almost surely.
Moreover(ρ(−δun),n≥ 1) convergesstronglyinL1(P ) toρ(−δu),thereforethegeneral
casefollows. 2
The following result is the generalization of the celebrated innovation’s theorem to thedegeneratecase,cf. [8]:
Theorem6.Let(Mt,t∈ [0,1]) beasquareintegrable(P,(Ft(XU),t∈ [0,1]))-martingale, then it can be represented as a stochastic integral of an (Ft(XU),t ∈ [0,1])-adapted,
IRd-valuedprocess β(XU) inthefollowingway:
Mt= M0+
t
0
P -a.s., where
t
0
|Ps(XU)βs(XU)|2ds <∞ P -a.s., forany t∈ [0,1].
Proof. Assume thatM isa(P,(Ft(XU),t∈ [0,1]))-martingale,thenfor anys< t and A∈ Fs(XU) wehave E Mt ζt 1Aρ(−δu) = E Mt ζt 1Aζt = E[Mt1A] = E[Ms1A] = E Ms ζs 1Aρ(−δu)
where ζ is the optional projection of ρ(−δu) w.r.t. the filtration (Ft(XU),t ∈ [0,1])
as calculated in Theorem 5. Consequently (Mt/ζt,t ∈ [0,1]) is a (Q,(Ft(XU),t ∈
[0,1]))-martingale, where dQ = ρ(−δu)dP . As U is a Q-Brownian motion, from The-orem2,wecanrepresent(Mt/ζt,t∈ [0,1]) as
Mt ζt = c + t 0 (Ps(XU) ˙αs(XU), dUs) ,
then usingtheItôformula
Mt= Mt ζt ζt = c + t 0 ζs(Ps(XU) ˙αs(XU), dUs)− t 0 Ms ζs ζs(Ps(XU)E[ ˙us|Fs(XU)], dZs) − t 0 ζs(Ps(XU) ˙αs(XU), E[ ˙us|Fs(XU)])ds = c + t 0 ζs(Ps(XU) ˙αs(XU), dZs+ Ps(XU)E[ ˙us|Fs(XU)])ds − t 0 Ms(Ps(XU)E[ ˙us|Fs(XU)], dZs)
− t 0 ζs(Ps(XU) ˙αs(XU), E[ ˙us|Fs(XU)])ds = c + t 0 Ps(XU) ζsα˙s(XU)− MsE[ ˙us|Fs(XU)] , dZs
andthiscompletes theproof. 2
4. Entropycalculation andMonge–Ampèreequation
Assumethatl(X) isaprobabilitydensitymeasurable w.r.t.F1(X),i.e., E[l(X)]= 1
andwithfinite entropy:E[l(X)log l(X)]<∞.WewanttofindaprocessU = B + u=
B +0· ˙usds whichisanadaptedperturbationoftheBrownianmotionB suchthat
l(X) = dX U(P ) dX(P ) ◦ X .
ThisproblemiscalledthecausalMonge–Ampèreproblem.Tosimplifythecalculations, we shall assumethatl◦ X isP -a.s. strictly positive.Assumethat sucha U (henceu)
existsandthatu satisfiestheGirsanovtheorem,i.e.,E[ρ(−δBu)]= 1.ThentheGirsanov
theoremimpliesthat
l◦ XUE[ρ(−δu)|F1(XU)] = 1 (4.9)
P -a.s., which is the causal version of the Monge–Ampère equation. From Theorem 2,
l◦ X canberepresentedas l◦ X = exp ⎛ ⎝− 1 0 Ps(X) ˙vs(X)· dBs− 1 2 1 0 |Ps(X) ˙vs(X)|2ds ⎞ ⎠ ,
where (Ps(X) ˙vs(X),s ∈ [0,1]) is adapted to the filtration of X and
1
0 |Ps(X) ˙vs(X)|
2ds<∞ P -a.s. Besides,sinceU isaBrownianmotionunderthe
prob-abilityρ(−δu)dP ,itfollowsfromTheorem 2thatl◦ XU canberepresentedas
l◦ XU = exp ⎛ ⎝− 1 0 Ps(XU) ˙vs(XU)· dUs− 1 2 1 0 |Ps(XU) ˙vs(XU)|2ds ⎞ ⎠ . (4.10) Insertingtherighthandsideof(4.10) andE[ρ(−δu)|F1(XU)] whichisalreadycalculated
inTheorem5intheMonge–Ampèreequation(4.9) andthentakingthelogarithmofthe finalexpression, weobtain
1 0 Ps(XU) E[ ˙us|Fs(XU)] + ˙vs(XU) · dZs +1 2 1 0 |Ps(XU) ˙vs(XU) + Ps(XU)E[ ˙us|Fs(XU)]|2 ds = 0 .
This relationimpliesthat
Ps(XsU)
˙vs(XU) + E[ ˙us|Fs(XU)]
= 0 (4.11)
ds× dP -almostsurely,whichisaquiteelaborate nonlinearequation.From theMonge– Ampèreequation(4.9) wecancalculatetherelativeentropybetweenXU(P ) andX(P ),
denotedbyH(XU(P )|X(P )):
Theorem7.Supposethatl isanX(P )-almostsurelystrictlypositivedensity.Thereexists some ˙u∈ L2(ds× dP ) withE[ρ(−δu)]= 1 with
dXU(P ) dX(P ) = l if andonly if Ps(XsU) ˙vs(XU) + E[ ˙us|Fs(XU)] = 0.
In this casewealso have
H(XU(P )|X(P )) = 1 2E 1 0 |Ps(XU)E[ ˙us|Fs(XU)]|2ds = 1 2E 1 0 |Ps(XU) ˙vs(XU)|2ds . Proof. H(XU(P )|X(P )) = logdX U(P ) dX(P ) dX U(P ) = logdX U(P ) dX(P ) ◦ X UdP = log l◦ XUdP = 1 2E 1 0 |Ps(XU)E[ ˙us|Fs(XU)]|2ds ,
provided that ˙u ∈ L2(ds× dP ) and the first equality follows, the second one is a
consequence of the relation (4.11), it can be also proven directly from the Girsanov theorem. 2
Proposition2.Assumethat l andu are givenasabove.Suppose furthermorethat H(XU(P )|X(P )) = 1 2E 1 0 |Ps(XU) ˙us|2ds . (4.12) Thenthefollowingequation holdstrue:
Ps(XU)dUs+ Ps(XU) ˙vs◦ XUds = Ps(XU)dBs (4.13) almostsurely.In particular,XU satisfies thefollowingstochasticdifferential equation:
dXtU = σ(t, XU)(dBt− ˙vt◦ XUdt) + b(t, XU)dt, (4.14) withthesame initialconditionas X.
Proof. Thehypothesis(4.12) impliesthattheprocess(Pt(XU) ˙ut,t∈ [0,1]) isds-almost
surelyadaptedtothefiltration(Ft(XU),t∈ [0,1]),hencewegetfromtheequality(4.11)
therelation
Pt(XU)( ˙vt◦ XU+ ˙ut) = 0 ,
whichimpliesatoncetherelation(4.13).Toseethenextone,notethat
dXtU = σ(t, XU)(dBt+ ˙utdt) + b(t, XU)dt
= σ(t, XU)(dBt+ Pt(XU) ˙utdt) + b(t, XU)dt
= σ(t, XU)(dBt− Pt(XU) ˙vt◦ XUdt) + b(t, XU)dt
= σ(t, XU)(dBt− ˙vt◦ XUdt) + b(t, XU)dt
where we haveused the factthat σ(t,XU)η = σ(t,XU)P
t(XU)η forany vector inIRd
sincePt(XU) istheorthogonalprojectionofIRd ontoσ(XtU)(IRn). 2
Theorem7canbe extendedas follows Theorem8.Assumethatu∈ L2
a(dt×dP,H) anddenotebyU theprocess(Bt+
t
0 ˙usds,t∈
[0,1]). assume also, asbefore, the Lipschitz hypothesisabout thedrift and diffusion co-efficients,thenthefollowinginequalityholdstrue:
H(XU(P )|X(P )) ≤ 1 2E 1 0 |Ps(XU)E[ ˙us|Fs(XU)]|2ds . (4.15)
Proof. Ifu∈ L∞a (dt×dP,H),thentheclaimwithequality(insteadofinequality)follows
fromTheorem7.Forthecaseu∈ La2(dt×dP,H),defineTn= inf(t> 0:
t 0| ˙us| 2ds> n), then un definedby un(t) = t 0 1[0,Tn](s) ˙usds is inL∞a (dt× dP,H),hencewehave H(XUn(P )|X(P )) = 1 2E 1 0 |Ps(XU n )E[1[0,Tn](s) ˙us|Fs(X Un)]|2ds . (4.16)
As n → ∞, (XUn(P ),n ≥ 1) converges weakly to XU(P ) and the weaklower
semi-continuityoftheentropyimpliesthat
H(XU(P )|X(P )) ≤ lim inf n 1 2E 1 0 |Ps(XU n )E[1[0,Tn](s) ˙us|Fs(X Un )]|2ds = 1 2E 1 0 |Ps(XU)E[ ˙us|Fs(XU)]|2ds ,
where thelimit oftheright handsideof theequation(4.16) followsfrom theLipschitz hypothesis. 2
References
[1]M.H.A.Davis,P.Varaiya,Themultiplicityofanincreasingfamilyofσ-fields,Ann.Probab.2(1974) 958–963.
[2]C.Dellacherie,Intégrales stochastiques par rapport aux processus deWiener ou dePoisson, in: SéminairedeProbabilitésVIII,in:LectureNotesinMathematics,vol. 381,Springer-Verlag,1973, CorrectiondansSPIX,in:LNM,vol. 465.
[3]N. El Karoui,H. Reinhard, Processusde diffusion dansIRn,in: Séminaire deProbabilités, VII (Univ.Strasbourg,annéeuniversitaire1971–1972),in:LectureNotesinMath.,vol. 321,Springer, Berlin,1973,pp. 95–117.
[4]D. Feyel, A.S. Üstünel, Transport of measures on Wiener space and the Girsanov theorem,
C. R. Math.334 (1)(2002)1025–1028.
[5]D.Feyel,A.S.Üstünel,Monge–KantorovitchmeasuretransportationandMonge–Ampèreequation onWienerspace,Probab.TheoryRelatedFields128 (3)(2004)347–385.
[6]D.Feyel,A.S.Üstünel,Monge–Kantorovitchmeasuretransportation,Monge–Ampèreequationand theItôcalculus,in:AdvancedStudiesinPureMathematics,Math.Soc.ofJapan,vol. 41, Mathe-maticalSocietyofJapan,2004,pp. 32–49.
[7]D.Feyel,A.S.Üstünel,M.Zakai,Realizationofpositiverandomvariablesviaabsolutelycontinuous transformationsofmeasureonWienerspace,Probab.Surv.3(2006)170–205(electronic). [8]M.Fujisaki,G.Kallianpur,H.Kunita,Stochasticdifferentialequationsforthenonlinearfiltering
problem,OsakaJ.Math.9(1972)19–40.
[10]K.Itô,MultipleWienerintegrals,J.Math.Soc.Japan3(1951)157–169.
[11]J.Jacod,Calculstochastiqueetproblémesdemartingales,LectureNotesinMathematics,vol. 714, Springer-Verlag,1979.
[12]D.W.Stroock, S.R.S. Varadhan,Diffusion processeswithcontinuouscoefficients 1,Comm. Pure Appl.Math.22(1969)345–400.
[13]A.S.Üstünel,Entropy,invertibilityandvariationalcalculusofadaptedshiftsonWienerspace,J. Funct.Anal.257 (8)(2009)3655–3689.
[14]A.S.Üstünel,VariationalcalculationofLaplacetransformsviaentropyonWienerspaceand appli-cations,J.Funct.Anal.267 (8)(2014)3058–3083.
[15]S.Watanabe,TheJapanesecontributionstomartingales,J.Électron.Hist.Probab.Stat.(J. Elec-tron.Hist.Probab.Stat.)5 (1)(Juin/June2009).
[16]Y. Yamada,S.Watanabe, Onthe uniquenessof solutionsof stochastic differential equations, J. Math.KyotoUniv.11(1971)155–167.