Optimal control of Hamilton-Jacobi
Transcription
Optimal control of Hamilton-Jacobi
Introduction Main results Proofs Further results Optimal control of Hamilton-Jacobi-Bellman equations P. Jameson Graber Commands (ENSTA ParisTech, INRIA Saclay) 17 January, 2014 P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Motivation Background Table of Contents 1 Introduction Motivation Background 2 Main results Assumptions Existence of minimizers Characterization of minimizers 3 Proofs Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions 4 Further results Well-posedness of weak solutions Long time average behavior P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Motivation Background Objective We consider an optimal control problem: Problem (1) For a given cost functional K and finite measure dm0 , find (Z Z ) Z T inf K (t, x, f (t, x))dxdt − u(0, x)dm0 (x) 0 over functions u, f such that f ∈ C ([0, T ] × TN ) and u solves −∂t u(t, x) + H(x, Du(t, x)) = f (t, x), u(T , x) = uT (x). (2) Find a PDE characterization of (weak) minimizers. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Motivation Background Motivation Purpose: to steer an evolving set. Model: given a set A of admissible controls and a given region ΩT ⊂ RN , let Ωt be the backward reachable set at time t defined by Ωt := {y (t) : ẏ = c(y , α), α ∈ A, y (T ) ∈ ΩT }. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Motivation Background Model Strategy: use obstacles. Obstacle construction: Let f ≥ 0 be continuous. Define Kτ , τ ∈ [0, T ] by Kτ := {x : f (τ, x) > 0}. Consider the modified “blocked” evolving set Ω̃t = {y (t) : ẏ = c(y , α), α ∈ A, y (T ) ∈ ΩT , y (τ ) ∈ / Kτ ∀ τ ∈ [t, T ]}. Following [Bokanowski, Forcadel, Zidani 2010], we have a characterization: let uT ≥ 0 (continuous) be such that uT = 0 precisely on ΩT . Z T Ω̃t = {y (t) : uT (y (T )) + f (s, y (s)) ≤ 0, ẏ = c(y , α), α ∈ A}. t P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Motivation Background Model Ω̃t = {u(t, ·) ≤ 0} characterized by an optimal control problem: ( ) Z T u(t, x) := inf uT (y (T )) + f (s, y (s))ds : ẏ = c(y , α), α ∈ A, y (t) = x . t u solves a Hamilton-Jacobi-Bellman equation: −∂t u(t, x) + H(x, Du(t, x)) = f (t, x), u(T , x) = uT (x). Conclusion Steering the front by obstacles is thus related to optimizing solutions to HJB equations. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Motivation Background Related model A related open problem: Optimal control via velocity Given a cost functional C, find Z inf C(c) − u(T , x)dm0 (x) , where the infimum is taken over u, c satisfying ut + c(t, x)|Du| = 0, u(0, x) = u0 (x). Find a characterization of the minimizers. Very little appears to be known. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Motivation Background Table of Contents 1 Introduction Motivation Background 2 Main results Assumptions Existence of minimizers Characterization of minimizers 3 Proofs Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions 4 Further results Well-posedness of weak solutions Long time average behavior P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Motivation Background Mass transportation In 2000, Benamou and Brenier considered a fluid mechanics formulation of the Monge-Kantorovich problem (MKP). For ρ0 (x) ≥ 0 and ρT (x) ≥ 0 probability densities on Rd , we say M : Rd → Rd transfers ρ0 to ρT if ∀A ⊂ Rd bounded, Z Z ρT (x)dx = ρ0 (x)dx. x∈A M(x)∈A The Lp MKP is to find Z p dp (ρ0 , ρT ) = inf M |M(x) − x|p ρ0 (x)dx. dp is called the Lp Kantorovich (or Wasserstein) distance. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Motivation Background PDE characterization of MKP Theorem (Benamou-Brenier, 2000) The square of the L2 Kantorovich distance is equal to the infimum of Z T Z ρ(t, x)|v (t, x)|2 dxdt, T Rd 0 among all (ρ, v ) satisfying ∂t ρ + ∇ · (ρv ) = 0 for 0 < t < T and x ∈ Rd , and the initial-final condition ρ(0, ·) = ρ0 , ρ(T , ·) = ρT . The optimality conditions are given by v (t, x) = ∇φ(t, x), where ∂t φ + 1 |∇φ|2 = 0. 2 P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Motivation Background Dolbeault-Nazaret-Savaré distances In 2009, Dolbeault, Nazaret, and Savaré, building on Benamou-Brenier, introduced a new class of transport distances between measures, given by minimizing Z 1Z |v (t, x)|2 m(ρ(t, x))dxdt 0 Rd subject to ∂t + ∇ · (m(ρ)v ) = 0, ρ(0, ·) = ρ0 , ρ(1, ·) = ρ1 . In 2013, Cardaliaguet, Carlier, and Nazaret studied geodesics for this class of distances when m(ρ) = ρα , getting optimality conditions ∂t ρ + ∇ · ( 21 ρα ∇φ) = 0, ρ > 0 ⇒ ∂t φ + α4 ρα−1 |∇φ|2 = 0, ρ ≥ 0, ∂t φ ≤ 0, ρ(0, ·) = ρ0 , ρ(1, ·) = ρ1 . P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Motivation Background Mean Field Games In 2007, Lasry and Lions give an overview of the following system of equations: ∂u ∂t ∂m ∂t − ν∆u + H(x, ∇u) = V [m] in Q × (0, T ) + ν∆m + ∇ · ∂H ∂p (x, ∇u)m = 0 in Q × (0, T ) u|t=0 = V0 [m(0)] on Q, m|t=T = m0 on Q. There are two characterizations of the system: 1 asymptotic behavior of differential games with large number of players, 2 optimal control of Fokker-Planck equation, or of Hamilton-Jacobi equation (by duality). P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Assumptions Existence of minimizers Characterization of minimizers Table of Contents 1 Introduction Motivation Background 2 Main results Assumptions Existence of minimizers Characterization of minimizers 3 Proofs Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions 4 Further results Well-posedness of weak solutions Long time average behavior P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Assumptions Existence of minimizers Characterization of minimizers Assumptions Hamiltonian structure. H = H(x, p) is Lipschitz in the first variable: |H(x, p) − H(y , p)| ≤ L|x − y ||p|. H(x, ·) is subadditive and positively homogeneous, hence convex, with linear bounds c0 |p| ≤ H(x, p) ≤ c1 |p|. H(x, p) := sup{−c(x, a) · p : a ∈ A} Cost function. k = k(t, x, m) continuous and strictly increasing with 1 |m|q−1 − C ≤ k(t, x, m) ≤ C |m|q−1 + C . C K ∗ (t, x, ·) a primitive of k, K (t, x, ·) its Fenchel conjugate. We will assume that K (t, x, 0) = 0 and that K (t, x, f ) is increasing in f for f ≥ 0. Note that 1 p |f | − C ≤ K (t, x, f ) ≤ C |f |p + C . C Final-initial conditions. uT : TN → R is Lipschitz, m0 ∈ L∞ (TN ) P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Assumptions Existence of minimizers Characterization of minimizers Comparison with the literature 1 This is the first study with linearly bounded Hamiltonian. 2 In [Cardaliaguet, 2013] the Hamiltonian is supposed to satisfy 1 C |p|r − C2 ≤ H(x, p) ≤ |p|r + C rC r r for some r > N(q − 1) ∨ 1 and, for some θ ∈ [0, N+1 ), |H(x, p) − H(y , p)| ≤ C |x − y |(|p| ∨ 1)θ . The Hamiltonian H(x, p) = c(x)|p|r is forbidden by this restriction. By contrast, H(x, p) = c(x)|p| is permitted under our assumptions. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Assumptions Existence of minimizers Characterization of minimizers Existence theorem Relaxed set The set K̃ will be defined as the set of all pairs (u, f ) ∈ BV × Lp such that u ∈ L∞ , u(T , ·) = uT in the sense of traces, and −∂t u + H(x, Du) ≤ f in the sense of measures, by which we mean that f − H(x, Du) + ∂t u is a non-negative Radon measure. Relaxed problem: Find inf (u,f )∈K̃ A(u, f ). Lemma The relaxed problem has the same infimum as the original. Theorem (PJG) There exists a minimizer to the relaxed problem. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Assumptions Existence of minimizers Characterization of minimizers Characterization of minimizers Heuristically, use Lagrange multipliers to get system (MFG) −∂t u + H(x, Du) = k(t, x, m(t, x)) u(T , x) = uT (x) ∂t m − div (m∂p H(x, Du)) = 0 m(0, x) = m0 (x) (state equation) (adjoint equation) which characterizes minimizers (u, f ) of the relaxed problem (the optimal choice is f = k(t, x, m)). Problems: H not differentiable, weak solutions otherwise hard to define. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Assumptions Existence of minimizers Characterization of minimizers Definition of weak solutions A pair (u, m) ∈ BV ((0, T ) × TN ) × Lq ((0, T ) × TN ) is called a weak solution to the MFG system if it satisfies the following conditions. 1 u satisfies Z Z Z tZ u(0)m0 dx = u(t)m(t)dx + k(s, x, m)mdxds, TN TN Z Z u(t)m(t)dx = TN 2 Z 0 T TN Z uT m(T )dx + TN k(s, x, m)mdxds, t TN for almost all t ∈ [0, T ]. We also have −∂t u + H(x, Du) ≤ k(t, x, m) in the sense of measures. Moreover, u(T , ·) = uT in the sense of traces m satisfies the continuity equation ∂t m + div (mv) = 0 in (0, T ) × TN , m(0) = m0 in the sense of distributions, where v ∈ L∞ ((0, T ) × TN ; RN ) is a vector field such that v(t, x) ∈ c(x, A) a.e. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Assumptions Existence of minimizers Characterization of minimizers Well-posedness Theorem (PJG) (i) There exists a weak solution (u, m) of (MFG). Moreover, if u ∈ W 1,p , then the Hamilton-Jacobi equation −∂t u − v · Du = f holds pointwise almost everywhere in {m > 0}, where v is the bounded vector field appearing in the definition. (ii) If (u, m) and (u 0 , m0 ) are both weak solutions (MFG), then m = m0 almost everywhere while u = u 0 almost everywhere in the set {m > 0}. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Table of Contents 1 Introduction Motivation Background 2 Main results Assumptions Existence of minimizers Characterization of minimizers 3 Proofs Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions 4 Further results Well-posedness of weak solutions Long time average behavior P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Smooth problem Denote by K0 the set of maps u ∈ C 1 ([0, T ] × TN ) such that u(T , x) = uT (x). The “smooth problem” is to compute the infimum over K0 of the functional Z TZ Z A(u) = K (t, x, −∂t u + H(x, Du))dxdt − u(0, x)dm0 (x). 0 TN TN Lemma The “smooth problem” is equivalent to the original problem, i.e. inf A(u) = u∈K0 inf f ∈C ([0,T ]×TN ) J (f ). Proof: uses the superposition principle. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Dual problem Define K1 to be the set of all pairs (m, w) ∈ L1 ((0, T ) × TN ) × L1 ((0, T ) × TN ; RN ) such that m ≥ 0 almost everywhere, w(t, x) ∈ m(t, x)c(x, A) almost everywhere, and ∂t m + div (w) = 0 m(0, ·) = m0 (·) in the sense of distributions. The “dual problem” is to compute the infimum over K1 of Z Z TZ B(m, w) = uT (x)m(T , x)dx + K ∗ (t, x, m(t, x))dxdt. TN 0 P. Jameson Graber TN Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Proof of duality Lemma The smooth and dual problems really are in duality, i.e. inf A(u) = − u∈K0 min B(m, w). (m,w)∈K1 Moreover, the minimum on the right hand side is achieved by a pair (m, w) ∈ K1 , of which m is unique, and which must satisfy the following: (t, x) 7→ K ∗ (t, x, m(t, x)) ∈ L1 ((0, T ) × TN ) and m ∈ Lq ((0, T ) × TN ). Proof is an application of the Fenchel-Rockafellar duality Theorem. Let X := C 1 ([0, T ] × TN ; R), Y := C ([0, T ] × TN ; R) × C ([0, T ] × TN ; RN ). Let Λ : X → Y be given by Λ(u) = (∂t u, Du). P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Proof of duality (cont) Lemma inf u∈K0 A(u) = − min(m,w)∈K1 B(m, w). Then define the functionals F : X → R and G : Y → R by R − TN u(0, x)dm0 (x) if u(T , ·) = uT (·) F (u) = ∞ otherwise RT R G (a, b) = 0 TN K (t, x, −a + H(x, b))dxdt One can show the lemma is equivalent to max −F ∗ (Λ∗ (m, w)) − G ∗ (−(m, w)) (m,w)∈Y 0 =− min (m,w)∈Y 0 F ∗ (Λ∗ (m, w)) + G ∗ (−(m, w)). Apply Fenchel-Rockafellar to finish. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Table of Contents 1 Introduction Motivation Background 2 Main results Assumptions Existence of minimizers Characterization of minimizers 3 Proofs Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions 4 Further results Well-posedness of weak solutions Long time average behavior P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Original and relaxed problems equivalent Below we prove the following: Proposition A(u, f ) = − inf min B(m, w). (m,w)∈K1 (u,f )∈K̃ Equivalently, inf A(u, f ) = inf A(u). (u,f )∈K̃ u∈K0 First part of the proof: Since for u ∈ K0 we have that (u, −ut + H(x, Du)) ∈ K̃, it follows that inf (u,f )∈K̃ A(u, f ) ≤ inf u∈K0 A(u). It remains to show other inequality. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions A technical lemma Lemma For u ∈ BV and f an integrable function, the following are equivalent: −∂t u + H(x, Du) ≤ f in the sense of measures, and for every continuous vector field ṽ = ṽ(t, x) ∈ c(x, A) we have Z T Z Z T Z φ∂t u + φṽ · Du ≤ − 0 TN f φdtdx. 0 TN for every continuous function φ : [0, T ] × TN → [0, ∞). Remark: the crucial thing is that we only need to consider ṽ continuous. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Proof of technical lemma Lemma ZZ −∂t u + H(x, Du) ≤ f ⇔ − ZZ φ(∂t u + ṽ · Du) ≤ f φ ∀ṽ cont. One direction is obvious. For the other direction, let > 0. By Lusin’s Theorem, there is a compact set K with |Du|([0, T ] × TN \ K ) ≤ and Du/|Du| continuous on K . By continuity and convexity of c(·, A) we construct ṽ(t, x) ∈ c(x, A) by partition of unity so that −ṽ · Du/|Du| ≥ H(x, Du/|Du|) − . Use this to obtain the estimate Z T Z Z T Z φH(x, Du) ≤ C + 0 TN φṽ · Du. 0 TN Then let → 0. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Key lemma Lemma Suppose u ∈ BV ∩ L∞ satisfies −∂t u + H(x, Du) ≤ f in the sense of measures. Then for any m ∈ Lq satisfying the continuity equation ∂t m + div(mv) = 0 for some vector field v with v(t, x) ∈ c(x, A) a.e., we have Z Z Z tZ u(0)m(0)dx ≤ u(t)m(t)dx + fmdxds, TN TN Z Z u(t)m(t)dx ≤ TN Z 0 T TN Z uT m(T )dx + TN fmdxds, t TN for almost every t ∈ (0, T ). Proof: obtain smooth approximations of m and mv through time scaling and convolution, apply previous lemma and integration by parts. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Proof of equivalence Second half of the proof: Let (m, w) be a minimizer of the dual problem. It suffices to show that for (u, f ) ∈ K̃, we have A(u, f ) ≥ −B(m, w). This essentially follows from the last lemma. Indeed, we have Z T Z Z A(u, f ) = K (t, x, f (t, x))dxdt − Z u(0, x)m0 (x)dx TN 0 T TN Z Z ∗ ≥ fm − K (t, x, m)dxdt − 0 Z TN T Z ≥− 0 u(0, x)m0 (x)dx TN K ∗ (t, x, m)dxdt − TN Z uT (x)m(T , x)dx = −B(m, w). TN P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Table of Contents 1 Introduction Motivation Background 2 Main results Assumptions Existence of minimizers Characterization of minimizers 3 Proofs Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions 4 Further results Well-posedness of weak solutions Long time average behavior P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Upper bound lemma Lemma Suppose u is a continuous viscosity solution to the Hamilton-Jacobi equation −∂t u + H(x, Du) = f ≥ 0 in a region [0, T ] × U for some open set U in Euclidean space. Then for any 0 ≤ t < s ≤ T and for any 0 < β < 1, we have the following estimate: u(t, x) − u(s, y ) ≤ C (1 − β 2 )−N/2p kf kp |t − s|α , ∀|x − y | ≤ βc0 (s − t), where α = 1 − (N + 1)/p and the constant C depends on p, N, and c0−1 . Cf. [Cardaliaguet 2013]. This lemma gives us upper bounds. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Passing to the limit Let (un , fn ) be a minimizing sequence. WLOG, un and fn are Lipschitz, hence differentiable a.e., and fn ≥ 0. We have: fn bounded in Lp by estimates on K , so fn → f weakly. This implies un pointwise bounded. The above implies H(x, Dun ) is bounded in L1 , so Dun is as well. The above implies ∂t un is bounded in L1 , so un is bounded in BV . (∂t un , Dun ) → (∂t u, Du) in the weak measure sense, while un → u in L1 . Conclusion: (u, f ) is the minimizer. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Table of Contents 1 Introduction Motivation Background 2 Main results Assumptions Existence of minimizers Characterization of minimizers 3 Proofs Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions 4 Further results Well-posedness of weak solutions Long time average behavior P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Introduction Main results Proofs Further results Minimizers are weak solutions Suppose (u, f ) ∈ K̃ is a minimizer of relaxed problem and (m, w) ∈ K1 is a minimizer of dual problem. We know that −∂t u + H(x, Du) ≤ f in the sense of measures, so Z tZ Z u(t)m(t) − u(0)m0 ≥ 0, fm + Z 0 T Z Z fm + uT m(T ) − u(t)m(t) ≥ 0. t RT R R The main point is to get 0 fm + uT m(T ) − u(0)m0 = 0, so that the above inequalities become equality. Everything else comes directly from the definition of K̃ and K1 and the properties of minimizers already proved. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Minimizers are weak solutions, part 2 (u, f ) ∈ K̃ is a minimizer of relaxed problem and (m, w) ∈ K1 is a minimizer of dual problem. By the duality theorem, Z T Z 0= K (t, x, f ) + K ∗ (t, x, m) + Z uT m(T ) − u(0)m0 0 Z ≥ T Z Z fm + uT m(T ) − u(0)m0 ≥ 0, 0 which implies the desired result. It also implies K (t, x, f (t, x)) + K ∗ (t, x, m(t, x)) = f (t, x)m(t, x)a.e. so f (t, x) = k(t, x, m). P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Weak solutions are also minimizers Suppose (u, m) is a weak solution. We set w = mv and f = k(·, ·, m). Then (u, f ) is a minimizer of the relaxed problem and (m, w) a minimizer of the dual problem. For example: ZZ Z A(u 0 , f 0 ) = K (t, x, f 0 ) − u 0 (0)m0 ZZ Z ≥ K (t, x, f ) + ∂f K (t, x, f )(f 0 − f ) − u 0 (0)m0 ZZ Z 0 = K (t, x, f ) + m(f − f ) − u 0 (0)m0 ZZ Z = K (t, x, f ) + mf 0 + m(T )uT − m0 u(0) − u 0 (0)m0 ZZ Z ≥ K (t, x, f ) − m0 u(0) = A(u, f ). P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions Weak solutions are unique Suppose (u1 , m1 ) and (u2 , m2 ) are both weak solutions. 1 m1 and m2 are both minimizers of dual problem, so m1 = m2 =: m. 2 f := k(·, ·, m), then both (u1 , f ) and (u2 , f ) are minimizers of relaxed problem. 3 u := max{u1 , u2 }; we show −∂t u + H(x, Du) ≤ f in the sense of measures. 4 Then (u, f ) is a minimizer of relaxed problem, so Z ZZ Z ZZ K (t, x, f ) − ui (0)m0 = K (t, x, f ) − u(0)m0 . 5 Combine with previous inequalities and use u ≥ ui to get u = ui in {m > 0}. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Well-posedness of weak solutions Long time average behavior Table of Contents 1 Introduction Motivation Background 2 Main results Assumptions Existence of minimizers Characterization of minimizers 3 Proofs Duality Equivalence of original and relaxed problems A priori bounds Passing to the limit Existence of weak solutions Uniqueness of weak solutions 4 Further results Well-posedness of weak solutions Long time average behavior P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Well-posedness of weak solutions Long time average behavior Recent work The following is joint (i) (ii) (iii) work with Pierre Cardaliaguet. We study −∂t φ + H(x, Dφ) = f (x, m) ∂t m − div (mDp H(x, Dφ)) = 0 φ(T , x) = φT (x), m(0, x) = m0 (x) and the corresponding ergodic problem (i) λ + H(x, Dφ) = f (x, m(x)) (ii) −div(mD R p H(x, Dφ)) = 0 (iii) m ≥ 0, Td m = 1 P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Well-posedness of weak solutions Long time average behavior Assumptions 1 2 (Initial-final conditions) m0 ∈ C (Td ), m0 > 0, φT : Td → R Lipschitz. (Conditions on the Hamiltonian) H : Td × Rd → R is continuous in both variables, convex and differentiable in the second variable, with Dp H continuous in both variables. Also, 1 C |p|r − C ≤ H(x, p) ≤ |p|r + C . rC r 3 (Conditions on the coupling) Let f be continuous on Td × (0, ∞), strictly increasing in the second variable, satisfying 1 |m|q−1 − C ≤ f (x, m) ≤ C |m|q−1 + C ∀ m ≥ 1. C 4 The relation holds between the growth rates of H and of F : r > max{d(q − 1), 1}. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Well-posedness of weak solutions Long time average behavior Optimal control problems Optimal control of HJ equation: find the infimum of T Z Z F ∗ (x, −∂t φ + H(x, Dφ))dxdt − A(φ) = Td 0 Z φ(0, x)dm0 (x) Td over the set of maps φ ∈ C 1 ([0, T ] × Td ) such that φ(T , x) = φT (x). Optimal control of transport equation: find the infimum of Z B(m, w ) = Td L1 ((0, T ) over (m, w ) ∈ a.e. t ∈ (0, T ), and T Z Z mH ∗ x, − φT m(T )dx + Td 0 × Td ) × L1 ((0, T ) w + F (x, m)dxdt m × Td ; Rd ), m ≥ 0 a.e., R Td m(t, x)dx = 1 ∂t m + div (w ) = 0, m(0, ·) = m0 (·) in the sense of distributions. Theorem (Cardaliaguet, PJG) The optimal control problems above are in duality, i.e. inf A(φ) = − φ∈K0 P. Jameson Graber min B(m, w ). (m,w )∈K1 Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Well-posedness of weak solutions Long time average behavior Relaxed problem K will be defined as the set of all pairs (φ, α) ∈ BV × L1 such that Dφ ∈ Lr ((0, T ) × Td ), φ(T , ·) ≤ φT in the sense of traces, α+ ∈ Lp ((0, T ) × Td ), φ ∈ L∞ ((t, T ) × Td ) for every t ∈ (0, T ), and −∂t φ + H(x, Dφ) ≤ α. in the sense of distribution. Find Z TZ Z inf A(φ, α) = inf F ∗ (x, α(t, x))dxdt− (φ,α)∈K (φ,α)∈K 0 Td φ(0, x)m0 (x)dx. Td Theorem (Cardaliaguet, PJG) The relaxed problem has a minimizer. Moreover, inf (φ,α)∈K A(φ, α) = − B(m, w ) = inf A(φ). min (m,w )∈K1 P. Jameson Graber φ∈K0 Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Well-posedness of weak solutions Long time average behavior Definition of weak solution A pair (φ, m) ∈ BV ((0, T ) × Td ) × Lq ((0, T ) × Td ) is called a weak solution if it satisfies the following conditions. 1 Dφ ∈ Lr and the maps mf (x, m), mH ∗ (x, −Dp H(x, Dφ)) and mDp H(x, Dφ) are integrable, 2 φ satisfies −∂t φ + H(x, Dφ) ≤ f (x, m) in the sense of distributions, the boundary condition φ(T , ·) ≤ φT in the sense of trace and the following equality ZZ m (H(x, Dφ) − hDφ, Dp H(x, Dφ)i − f (x, m)) dxdt Z = (φT m(T ) − φ(0)m0 )dx 3 m satisfies the continuity equation ∂t m − div (mDp H(x, Dφ)) = 0 in (0, T ) × Td , m(0) = m0 in the sense of distributions. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Well-posedness of weak solutions Long time average behavior Well-posedness Theorem (Cardaliaguet, PJG) (i) If (m, w ) ∈ K1 is a minimizer of the dual problem and (φ, α) ∈ K̃ is a minimizer of the relaxed problem, then (φ, m) is a weak solution and α(t, x) = f (x, m(t, x)) almost everywhere. (ii) Conversely, if (φ, m) is a weak solution, then there exist functions w , α such that (φ, α) ∈ K is a minimizer of the relaxed problem and (m, w ) ∈ K1 is a minimizer of the dual problem. (iii) If (φ, m) and (φ0 , m0 ) are both weak solutions, then m = m0 almost everywhere while φ = φ0 almost everywhere in the set {m > 0}. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Well-posedness of weak solutions Long time average behavior Main result Let φf ∈ C 1 on Td . Let (φT , mT ) be a “good” solution of −∂t φ + H(x, Dφ) = f (x, m) ∂t m − div (mDp H(x, Dφ)) = 0 φ(T , x) = φf (x), m(0, x) = m0 (x). and (λ̄, φ̄, m̄) be a solution of the ergodic MFG system. Define ψ T (s, x) = φT (sT , x), µT (s, x) = mT (sT , x) ∀(s, x) ∈ (0, 1)×Td . Theorem (Cardaliaguet, PJG) As T → +∞, (µT ) converges to m̄ in Lθ ((0, 1) × Td ) for any θ ∈ [1, q), ψ T /T converges to the map s → λ̄(1 − s) in Lθ ((δ, 1) × Td ) for any θ ≥ 1 and any δ ∈ (0, 1). P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations Introduction Main results Proofs Further results Well-posedness of weak solutions Long time average behavior Thank you! P. Cardaliaguet and P. Graber, “Mean field games systems of first order,” submitted to ESAIM: Control, Optimization, and Calculus of Variations. Available on the arXiv. P. Graber, “Optimal control of first-order Hamilton-Jacobi equations with linearly bounded Hamiltonians,” to appear in Applied Mathematics and Optimization. Also available on the arXiv. P. Jameson Graber Optimal control of Hamilton-Jacobi-Bellman equations