Idempotent Method for Continuous-Time Stochastic Control and Complexity Attenuation
|
|
- Shon Pope
- 5 years ago
- Views:
Transcription
1 Idempotent Method for Continuous-ime Stochastic Control and Complexity Attenuation illiam M. McEneaney Hidehiro Kaise Seung Hak Han Dept. of Mech. and Aero. Eng., University of California San Diego, San Diego, CA , USA ( Graduate School of Information Sci., Nagoya University, Furo-cho, Chikusa-ku, Nagoya , Japan Dept. of Mech. and Aero. Eng., University of California San Diego, San Diego, CA , USA ( Abstract: e consider a -plus based numerical method for solution of finite timehorizon control of nonlinear diffusion processes. he approach belongs to the class of curseof-dimensionality-free methods. he -plus distributive property is required. he price to pay is a very heavy curse-of-complexity. hese methods perform well due to the complexityattenuation step. his projects the solution down onto a near-optimal -plus subspace. Keywords: stochastic, nonlinear control, numerical methods. 1. INRODUCION It is now well-known that many classes of deteristic control problems may be solved by max-plus or plus numerical methods. hese methods include maxplus basis-expansion approaches [1], [5], [7], as well as the more recently developed curse-of-dimensionality-free methods [7], [10]. It has recently been discovered that idempotent methods are applicable to stochastic control and games. he methods are related to the above curseof-dimensionality-free methods for deteristic control. In particular, a -plus based method was developed for stochastic control problems [9]. he first such methods for stochastic control were developed only for discrete-time problems. Here, we will remove the severe restriction to discrete-time problems. his extension requires overcog significant technical hurdles. e will first define a parameterized set of operators, approximating the dynamic programg operator. e obtain the solutions to the problem of backward propagation by repeated application of the approximating operators. Using techniques from the theory of viscosity solutions, we show that the solutions converge to the viscosity solution of the Hamilton-Jacobi-Bellman partial differential equation (HJB PDE) associated with the original problem. he problem is now reduced to backward propagation by these approximating operators. he -plus distributive property is employed. A generalization of this distributive property, applicable to continuum versions will be obtained. his will allow interchange of expectation over normal random variables (and other random variables with range in IR m ) with infimum operators. At each time-step, Research partially supported by NSF Grant DMS , AFOSR Grant FA and Grant-in-Aid for Young Scientists (B), No , MEX. the solution will be represented as an infimum over a set of quadratic forms. Use of the -plus distributive property will allow us to maintain that solution form as one propagates backward in time. Backward propagation is reduced to simple standard-sense linear algebraic operations for the coefficients in the representation. e also demonstrate that the assumptions on the representation which allow one to propagate backward one step are inherited by the representation at the next step. he difficulty with the approach is an extreme curse-of-complexity, wherein the number of terms in the -plus expansion grows very rapidly as one propagates. he complexity growth will be attenuated via projection onto a lower dimensional plus subspace at each time step. hat is, at each step, one desires to project onto the optimal subspace relative to the solution approximation. Importantly, the subspace is not set a priori. Using some tools from convex analysis and imax theory, we show the optimal projection is achieved by pruning the current set of quadratic forms.. DEFINIION AND DYNAMIC PROGRAM e begin by defining the specific class of problems which will be addressed here. Let the dynamics take the form dξ s = f µs (ξ s, u s ) ds + σ µs (ξ s, u s ) db s, ξ t = x IR n, (1) where f m (x, u) is measurable, with more assumptions on it to follow. he u s and µ s will be control inputs taking values in U IR p and M =]1, M[ = 1,,, M, respectively. In practice, we often find it useful to allow both a continuum-valued control component and a finite set-valued component, where the latter is used to allow approximation of more general nonlinear Hamiltonians, c.f. [7] for motivation. Also, B, F is an l-dimensional Brownian motion on the probability space (Ω, F, P ), where F 0 contains all the P -negligible elements of F and σ m (x, u) is an n l matrix-valued diffusion coefficient. e will be Copyright by the International Federation of Automatic Control (IFAC) 316
2 exaing a finite time-horizon formulation, with teral time,, and will take initial time t [0, ]. he payoff (to be imized) will be J(t, x, u, µ ) =E. l µs (ξ s, u s ) ds+ Ψ(ξ ) t where Ψ(x) =. inf z Z g (x, z ), (3) where l m (x, u) and g (x, z) are measurable, and (Z, d Z ) is a separable metric space. he value function is V (t, x) = inf J(t, x, u, µ ), u U t,µ M t (4) where U t (resp. M t ) is the set of F t -progressively measurable controls, taking values in U (resp. M) such that there exists a strong solution to (1). () e will assume that the given data in the dynamics and payoff satisfy the following conditions: (A1) U is a closed subset of IR p. (A) here exist L f, L σ, K f, K σ > 0 such that for any x, x IR n, u, u U, m, m M, f m (x, u) f m (x, u ) L f ( x x + u u + m m ), σ m (x, u) σ m (x, u ) L σ ( x x + u u + m m ), f m (x, u) K f (1 + x + u ), σ m (x, u) K σ. (A3) here exist L l, K l > 0 and ν > 0 such that for any x, x IR n, u, u U, m, m M l m (x, u) l m (x, u ) L l (1 + x + x + u + u ) ( x x + u u + m m ), K l + ν u l m (x, u) K l (1 + x + u ). (A4) here exist L Ψ, K Ψ > 0 such that for any x, x IR n, Ψ(x) Ψ(x ) L Ψ (1 + x + x ) x x, K Ψ Ψ(x) K Ψ (1 + x ). At a formal level, one expects that V (t, x) will satisfy the following Hamilton-Jacobi-Bellman partial differential equation (HJB PDE): V t (t, x) + H(x, D xv (t, x), D xv (t, x)) = 0, (5) V (, x) = Ψ(x), x IR n, (6) H(x, D x V (t, x), DxV (t, x)) 1 = inf u U m M tr(σm (x, u)σ m (x, u) DxV (t, x)) +f m (x, u) D x V (t, x) + l m (x, u). (7) Under restrictive conditions, it is known that V (t, x) can be characterized as the unique viscosity solution of (5)/(6). However, for more general conditions including assumptions (A1) (A4), if V (t, x) is[ associated with the ] set of control processes satisfying E u t s ds <, it is shown in [4] that V (t, x) is the unique viscosity solution of (5)/(6) satisfying the quadratic growth condition V (t, x) K(1 + x ), (t, x) [0, ] IR n. (8) See [4, heorem 3.1 and Remark 3.1]. e need to additionally show convergence of discrete-time approximations to the viscosity solution. e approximate the viscosity solution of (5)/(6) by discrete-time stochastic control problems. o define a discrete-time stochastic control value, we introduce a family of parameterized operators F t,s t<s acting on φ : IR n IR: F t,s φ(x) = inf u U,m M lm (x, u)(s t) (9) +E[φ(x + f m (x, u)(s t) + σ m (x, u)(b s B t ))]. Let π N = t 0 = 0 < t 1 < < t N = be a partition of [0, ] with the step size N = t i+1 t i = /N (i = 0, 1,, N 1). e define a discrete-time value function V N (t, x) ((t, x) [0, ] IR n ) associated with π N recursively backward in time: Ψ(x), t =, V N (t, x) = F t,ti+1 V N (10) (t i+1, )(x), t i t < t i+1, where F t,ti+1 V N (t i+1, )(x) is F t,ti+1 φ(x) with φ( ) = V N (t i+1, ). e obtain the following theorem regarding approximation of the viscosity solution. heorem.1. Under (A1) (A4), V N (t, x) converges to a continuous viscosity solution V (t, x) of (5) with (6) as N uniformly on each compact set of [0, ] IR n. Indeed, V (t, x) is the unique viscosity solution satisfying the quadratic growth condition: for some K > 0, V (t, x) K(1 + x ), (t, x) [0, ] IR n. (11) 3. MIN-PLUS DISRIBUIVE PROPERY e will use an infinite version of the -plus distributive property to move a certain infimum from inside an expectation operator to outside. It will be familiar to control and game theorists who often work with notions of nonanticipative mappings and strategies. One version of such appeared in [9]. However, the assumptions in that result are too restrictive for the class of problems we are considering. Instead, we generalize that result to the following, where the proof appears in [6]. heorem 3.1. Let (Z, d Z ) be a separable metric space and (, d ) be a separable Banach space with Borel sets B. Let p be a finite measure on (, B ), and let D =. p( ). Let h : Z IR be Borel measurable. Suppose there exists z Z such that h(w, z) dp (w) < (1) and suppose for given ε > 0, there exists R < such that h(w, z) dp (w) ε. (13) (B R (0)) c inf z Z Also, suppose that given ε > 0 and R <, there exists δ > 0 such that h(w, z) h( w, z) < ε for all z Z and all w, w B R (0) such that d (w, w) < δ. Lastly, we suppose that either Z is countable or h(w, z) is continuous on z for each w (where of course, the 317
3 former supposition can be embedded within the latter, but that is less illuating). hen, inf h(w, z) dp (w) = inf h(w, z(w)) dp (w), z Z z Z where Z. = z : Z Borel measurable. 4. DISRIBUED DYNAMIC PROGRAMMING e will use the above infinite-version of the -plus distributive property in conjunction with the dynamic programg principle of Section. his will yield what we refer to as an idempotent distributed dynamic programg principle (IDDPP). Recall our discrete-time value function, V N (t k, x) given by (10) for t k π N and x IR n. Suppose that at time, t k+1, one has representation V N (t k+1, x) = inf g N k+1(x, z), (14) where (Z k+1, d Zk+1 ) is a separable metric space. Recalling form (3), and letting gn N(x, z) = g (x, z) and Z N = Z, we see that V N (t N, x) = V N (, x) = Ψ(x) has this form. hen the dynamic program of (9), (10) with = /N becomes V N (t k, x) = inf l m (x, u) u U m M [ ( +E inf x + f m (x, u) + σ m (x, u)w, z )] g N k+1 = inf u U m M [ inf l m (x, u) +g N k+1( x + f (x, u, m, w), z )] dp (w), (15) where f (x, u, m, w) = f m (x, u) + σ m (x, u)w, P is the measure corresponding to a Gaussian random variable over IR l with mean zero and covariance I, and = IR l. e outline the approach to be followed. Specifically, we will use the -plus distributive property of heorem 3.1 to move the infimum over Z k+1 outside the integral. Letting Z k+1. = zk+1 : Z k+1 Borel measurable, we will have V N (t k, x) = inf u U m M inf z k+1 Z k+1 l m (x, u) (16) +g N k+1(x + f (x, u, m, w), z k+1 (w))dp (w). Since we will suppose that Z k is bounded and U is possibly unbounded, it is convenient to handle the infimum over u separately in (16). From (16), we can take V N (t k, x) = inf z Z k g N k (x, z), (17) where Z k = M Z k+1 and for x IR n and z = (m, z k+1 ) Z k, gk N (x, z) = inf l m (x, u) (18) u U +g N k+1(x + f (x, u, m, w), z k+1 (w))dp (w). Consequently, the general form of (14) will be inherited from V N (t k+1, ) to V N (t k, ), and one can propagate backward in this manner indefinitely. his is what we referred to above as the IDDPP. e note that (16) itself can be used as an IDDPP if U is bounded. ith significant effort, one obtains our main IDDPP result. heorem 4.1. In addition to (A1) (A3), suppose that (Z, d Z ) and g : IR n Z IR satisfy the following: (i-) (Z, d Z ) is a bounded and closed subset of a separable Banach space X where metric d Z is induced by the norm of X. (ii-) here exist K, L > 0 such that for any x, x IR n, z Z, K g (x, z) K(1 + x ), g (x, z) g (x, z) L(1 + x + x ) x x. (iii-) here exists ˆL > 0 such that for any z, z Z, x IR n, g (x, z) g (x, z ) ˆL(1 + x )d Z (z, z ). Letting Z N = Z and gn N (x, z) = g (x, z), (17) with (18) holds for all k ]0, N[. 5. FULLY QUADRAIC FORMS he above theory was somewhat general in form. ith a small modification, one could include u in the index set, letting Z k = U M Z k+1 rather than Z k = M Z k+1. However, we believe the most computationally useful form will occur when the problem is quadratic in u with U = IR p. Consequently, we now consider such a special problem class, where in particular, we use the fact that the quadratic nature of the control-dependence allows one to analytically obtain the g N k for each z Z k. o be precise, we now suppose f m (x, u) = A m x + B m u + λ m d, σ m (x, u) = σ m, (19) l m (x, u) = 1 x H m 1 x + x H m u + 1 u H m 3 u + λ m 1 x + λ m u + c m, U = IR p, (0) g (x, z) = 1 x Q (z)x + b (z)x + c (z). (1) e do not use the earlier assumptions. Instead, we replace them with assumptions on the problem data in the above form which will yield a special case of the earlier assumptions. e assume: (AQ.1) (Z, d Z ) is a bounded and closed subset of a separable Banach space X where metric d Z is induced by norm X of X. (AQ.) here exists D 1 < such that Q (z), b (z), c (z) D 1 for all z Z. (AQ.3) here exists L < such that max Q (z) Q ( z), b (z) b ( z), c (z) c ( z) Ld Z (z, z) z, z Z. (AQ.4) Q (z) 0 for all z Z. 318
4 (AQ.5) H m 3 > 0 and ( H m 1 H m ) 0 m M. H m H m 3 Note that (AQ.1) (AQ.5), along with the specific form of (19) (1) immediately imply that (A1) (A4) and (i-) (iii-) hold. Recall (18), and suppose that for some k ]0, N 1[, Vk+1(x) N =. V N (t k+1, x) = inf gk+1(x, N z), () where g N k+1(x, z) = 1 x Q k+1 (z)x + b k+1(z)x + c k+1 (z). (3) Note that g N N (x, z) = g (x, z) is of this form. After some computations, we arrive at the following, where we again do not include the details of the algebra. g N k (x, z)= 1 x Q k ( z, m)x + b k ( z, m)x + c k ( z, m) = 1 x Q k (z)x + b k (z)x + c k (z) (4) for appropriate Q k, b k, c k, where z = (m, z) Z k. Using this form, one can verify that conditions (AQ.1) (AQ.4) are inherited. e do not include the details. his yields the following or, but computationally useful, variant of main heorem 4.1. heorem 5.1. Consider the special dynamics and cost of (19) (1). Suppose (AQ.1) (AQ.5) hold. Let gn N (x, z) = g (x, z) and Z N = Z. hen, for all k ]0, N 1[ Vk N (x) = V N (t k, x) = inf gk N (x, z), z Z k where g N k (x, z) is given by (4) with Z k = Z k+1 M, for appropriate Q k, b k, c c. Due to space limitations, we do not include the computations which yield Q k, b k, c k from Q k+1, b k+1, c k COMPLEXIY REDUCION he key to this class of methods lies in the repeated projection of the solution down onto a low-dimensional (-plus) subspace. Importantly, the subspace is chosen at each step so as to imize the error induced by this projection. One sees from heorem 5.1 that after one step of the IDDPP, the set Z k will have the cardinality of the continuum even in the case where Z k+1 is finite. Consequently, the projection down to a finite-dimensional subspace is a critical step. For the class of problems where the solution may be well-represented by a reasonable set of quadratic functions, this method can be expected to perform quite well. It is worth noting that this complexity condition is independent from the dimension of the space; one may have problems with low complexity (by this vaguely defined metric) solutions in high dimensional spaces and vice-versa. e now begin the analysis of the projection of the V N k down to a low-dimensional -plus subspace. his will be a two-step procedure. First, we indicate a simple reduction from an infinite-dimensional index set, Z k, to a, possibly very large but finite, subset. Second, we discuss the theory by which, given a desired dimension of the subspace (i.e., index set size), one finds the optimal subspace of that dimension. e also indicate a computational approach for approximately achieving this projection. e will work entirely in the special case of Section 5. For the reduction from infinite-dimensional to finitedimensional, we simply note that by (AQ.), there exists D < such that (Q k (z), b k (z), c k (z)) z Z k B 3D (0), which is compact. herefore, given ε > 0, there exists a finite ε-net for the set, which yields our finitedimensional -plus projection. As indicated above, the key to computational feasibility will be projection of V k N down to a low-dimensional plus subspace with small error, if such is possible. e will use an approach similar to that in [8]. hat is, we will devise an optimization problem to select the small set of quadratics whose pointwise imum (-plus sum) will constitute the -plus projection. he constraint in the optimization will be slightly relaxed from the ideal constraint. ith this optimization problem formulation, we will see that our problem reduces to selection of a small set of quadratics out of the set whose -plus sum is V k N. Let be the set of all triples of coefficients for quadratic functions on IR n. hat is, we take = (Q, b, c) Q IR n n, symmetric; b n IR n ; c IR. For convenience, we will designate a triple by a single symbol, say τ, where τ will correspond to some triple, say τ = (Q, b, c). e let C Q (IR n ) denote the subset of C(IR n ) consisting of the quadratic functions mapping into IR. e let G : C Q be given as follows. For τ = (Q, b, c), let G[τ](x) = G[(Q, b, c)](x). = 1 x Qx + b x + c. Consider a finite set of quadratic coefficients, M given by. M = τ M, (5). where M =]1, M[ = 1,,... M with M IN. (Here, of course, each τ corresponds to a triple of coefficients.) Note that M may be very large. In particular, we will be interested in the specific case where M corresponds to the set of quadratics which generate V k N, in which case, M = N ε = #Ẑk. Further, in that case, each triple in M which we now designate as τ = ( Q, ˆb, ĉ ) corresponds (via bijection) to some triple (Q k (Ẑ), bk(ẑ), ck(ẑ)). e look for an optimal smaller set of triples, A N = α n n N, N =]1, N[ where N < M. Let Π be the set of probability measures over (IR n, B n ) where B n is the collection of Borel sets over IR n. Integration with respect to a π Π will be designated by π(dx). Let λ be a measure over (IR n, B n ) such that G[τ](x) λ(dx) < for all τ. he optimization IR problem n will take the form Minimize J(A N) =. IR n n N G[α n ](x) λ(dx) (6) subject to G[α n ](x) π(dx) IR n G[τ ](x) π(dx) IR n π Π, n N. (7) e note that (7) is a relaxation of constraint G[α n ](x) G[τ ](x) x IR n. (8) 319
5 Further, constraint (8) is typical in -plus expressions. If there exists n N, x IR n and ε > 0 such that G[α n ]( x) G[τ ]( x) ε, then addition of more α n to A N can never correct this error as one will still have n N G[α n ]( x) G[τ ]( x) ε. e will analyze problem (6),(7) to show that this formulation corresponds to imization of a monotonically increasing, concave function over a cornice set [8], where this class of problems has the property that the imizing set, A N, will be a subset of the original set, M. hat is, the optimal -plus subspace of dimension N will be a subset of of the original set M. For τ, α, we say α τ if G[α](x) π(dx) G[τ](x) π(dx) π Π. IR n IR n Note that is a partial order on. It is also worth noting the following. Lemma 6.1. α τ if and only if G[α](x) G[τ](x) for all x IR n. Given Lemma 6.1, one might wonder why we do not use the condition G[α](x) G[τ](x) for all x IR n to define our partial order. hen combined with the imization operation as on the right-hand side of (7), these are different; the integral definition will be more useful. For π Π, let G π [τ]. = IR n G[τ](x) π(dx). Lemma 6.. For any π Π, G π is a linear functional on. Let S M. = p IR M p [0, 1] M, p = 1. For any p S M, define p M = p τ. e say M = τ τ = p M for some p S M, (9) which is obviously the convex hull of M. Lemma 6.3. For any π Π, G π [τ ] = G π [p M ], p S M or equivalently, IR n G[τ ](x) π(dx) = p S M IR n G[p M ](x) π(dx). One then immediately has: Lemma 6.4. Fix any π Π. Let α M. hen, G π [α] G π [τ ] if and only if G π [α] p S M G π [p M ]. Using Lemma 6.4, our problem (6),(7) becomes Minimize J(A N) =. IR n n N G[α n ](x) λ(dx) (30) subject to G π [α n ] G π [p M ] p S M π Π, n N. (31) Given, we define the upper cone of as U[ ]. = α α τ for some τ. e also define the cornice of as C[ ] =. U[ ] (see [8]), and let C N [ ] =. C[ ] N denote the outer product of C[ ], N times. heorem 6.5. Condition (31) holds if and only if A N C N [ M ]. Combining heorem 6.5 with (30)/(31), we see that problem (6)/(7) is equivalent to: Minimize J(A N ) =. IR n n N G[α n ](x) λ(dx) (3) subject to A N C N [ M ]. (33) Define the inherited partial order on outer product space [ ] N by A 1 N A N where A j N = α j n α j n n N for j 1,, if α 1 n α n for all n N. [ ] N Lemma 6.6. J : IR is monotonically increasing (relative to ) and concave. Now, given the monotonicity and concavity of J, we can apply [8] heorem 3.1 to assert: heorem 6.7. Let J be the optimal value of problem (3),(33), or equivalently, (6),(7). hen, there exists A N = τ n n N such that τ n M for all n N, and such that A N satisfies (33) (equivalently, (7)) and J(A N) = J. he value of heorem 6.7 is that we know that the optimal set of quadratics is a subset of the original M set. It is worth noting that condition A N C N[ M ] is not identical to the condition G[α n ](x) G[τ ](x) x IR n n N. Instead, it is somewhat more conservative. 6.1 ractable Pruning Now that we understand that pruning is optimal (relative to the chosen criterion and constraints), we consider means for approximate optimal pruning. In particular, solution of even the reduced-complexity pruning problem with the above criterion remains computationally demanding. Consequently, we search instead for a tractable suboptimal pruning method. Firstly, one would like to remove those quadratics that do not contribute at all to the value function. Let 0, and suppose ˆm M is such that + G[τ ˆm ](x)π(dx) G[τ ](x)π(dx), IR n \ ˆm IR n (34) for all π Π. One obtains: Proposition 6.8. Suppose G[τ ˆm ](x)π(dx) G[τ ](x)π(dx), IR n \ ˆm IR n for all π Π. hen, 30
6 and IR n = G[τ ](x) π(dx) IR n \ ˆm G[τ ](x) G[τ ](x) π(dx), π Π = \ ˆm G[τ ](x), x IR n. In other words, τ ˆm does not contribute at all to the pointwise imum, and so may be removed without affecting the value function approximation. Further, by the monotonicity of the semigroup operator, no progeny of τ ˆm contribute to the value function approximation in subsequent steps, and so removal of τ ˆm does not affect the solution at any time-step. Now let. ˆm = inf (34) holds. It is apparent that ˆm provides some measure (loosely used) of the merit of τ ˆm relative to the imum. his leads to the following rough heuristic: o achieve a good subset of size N from the original set of M quadratics, those with the lowest ˆm values should be chosen. his leads to the optimization of a submodular functional, which is an issue we will discuss elsewhere. Here, however, we consider the approximate evaluation of ˆm. 6. Gaussians and Convex Program Given a set of quadratics, (Q, b, c ) M indexed as above by its coefficients, corresponding to some set τ M for some set M M, and a specific quadratic, (Q ˆm, b ˆm, c ˆm ) corresponding to τ ˆm, we wish to compute an approximate ˆm. In order to make the problem tractable, we consider only those π Π corresponding to normal distributions. Let D be the set of positive-definite, symmetric n n matrices. he appropriate set of normal distributions may be indexed by (D, x) D IR n. Let (Q ˆm, b ˆm, c ˆm ) correspond to τ ˆm, and π Π correspond to (D, x) D IR n. hen, G[τ ˆm 1 [ 1 ](x)π(dx)= IR n (π) n/ det(d) IR x Q ˆm x n +b ˆm x + c ˆm] exp 1 (x x) D 1 (x x) dx, and after manipulation, one finds this is = 1 x Q ˆm x + b ˆm x + c ˆm + 1 tr [ Q ˆm D ]. Note that this is linear in D and quadratic in x. Let Π denote the set of normal probability measures over IR n. ith a little work, one finds that ˆm may be computed by solution of: [ 1 max z (D, x,z) D IR n+1 x Q ˆm x + b ˆm x + c ˆm + 1 tr [ Q ˆm D ] ] [ 1 subject to: z x Q x + b x + c (35) + 1 tr [ Q D ]] 0 M. his last problem may be converted to an LMI. Following a similar approach to that used in [10], one obtains a relaxed dual problem: ζ ( ) λ c c ˆm 1 ζ λ b b ˆm ( ) ( ) 1 λ b b ˆm 1 λ Q Q ˆm 0, λ Q Q ˆm 0, (λ,ζ) S M IR where S M =. λ IR M λ m [0, 1] m, m λ m = 1, which takes the form of an LMI. hese LMIs are solved repeatedly to detere the contribution of each quadratic to the overall value function approximation. As noted above, the actual pruning corresponds to optimization of a submodular function on the space of subsets of M, and the details are not included due to length restrictions. REFERENCES [1] M. Akian, S. Gaubert and A. Lakhoua, he maxplus finite element method for solving deteristic optimal control problems: basic properties and convergence analysis, SIAM J. Control and Optim., 47 (008), [] G. Barles and P.E. Souganidis, Convergence of approximation schemes for fully nonlinear second order equations, Asymptotic Anal., 4 (1991), [3] G. Cohen, S. Gaubert and J.-P. Quadrat, Duality and Separation heorems in Idempotent Semimodules, Linear Algebra and Applications, 379 (004), [4] F. Da Lio and O. Ley, Uniqueness results for secondorder Bellman-Isaacs equations under quadratic growth assumptions and applications, SIAM J. Control Optim., 40 (006), [5].H. Fleg and.m. McEneaney, A max-plus based algorithm for an HJB equation of nonlinear filtering, SIAM J. Control and Optim., 38 (000), [6] H. Kaise and.m. McEneaney, Idempotent Expansions for Continuous-ime Stochastic Control, Proc. 49th IEEE Conf. Decision and Control, (010). [7].M. McEneaney, Max-Plus Methods for Nonlinear Control and Estimation, Birkhauser, Boston, 006. [8].M. McEneaney, Complexity Reduction, Cornices and Pruning, Proc. of the International Conference on ropical and Idempotent Mathematics, G.L. Litvinov and S.N. Sergeev (Eds.), Contemporary Math. 495, Amer. Math. Soc. (009), [9].M. McEneaney, Idempotent Algorithms for Discrete-ime Stochastic Control, Automatica (to appear). [10].M. McEneaney, A. Deshpande, S. Gaubert, Curse-of-Complexity Attenuation in the Curse-of- Dimensionality-Free Method for HJB PDEs, Proc. ACC 008, Seattle (008). 31
Idempotent Method for Deception Games and Complexity Attenuation
Idempotent Method for Deception Games and Complexity Attenuation William M. McEneaney Dept. of Mech. and Aero. Eng., University of California San Diego, San Diego, CA 92093-0411, USA (email: wmceneaney@ucsd.edu).
More informationIdempotent Method for Dynamic Games and Complexity Reduction in Min-Max Expansions
Idempotent Method for Dynamic Games and Complexity Reduction in Min-Max Expansions William M McEneaney Abstract In recent years, idempotent methods (specifically, max-plus methods) have been developed
More informationIDEMPOTENT METHODS FOR CONTROL AND GAMES
file://\\52zhtv-fs-725v\cstemp\adlib\input\wr_export_130917101852_1868311565... Page 1 of 1 9/17/2013 AFRL-OSR-VA-TR-2013-0492 IDEMPOTENT METHODS FOR CONTROL AND GAMES WILLIAM MCENEANEY UNIVERSITY OF CALIFORNIA,
More informationThe principle of least action and two-point boundary value problems in orbital mechanics
The principle of least action and two-point boundary value problems in orbital mechanics Seung Hak Han and William M McEneaney Abstract We consider a two-point boundary value problem (TPBVP) in orbital
More informationA max-plus based fundamental solution for a class of infinite dimensional Riccati equations
A max-plus based fundamental solution for a class of infinite dimensional Riccati equations Peter M Dower and William M McEneaney Abstract A new fundamental solution for a specific class of infinite dimensional
More informationControlled Diffusions and Hamilton-Jacobi Bellman Equations
Controlled Diffusions and Hamilton-Jacobi Bellman Equations Emo Todorov Applied Mathematics and Computer Science & Engineering University of Washington Winter 2014 Emo Todorov (UW) AMATH/CSE 579, Winter
More informationA CURSE-OF-DIMENSIONALITY-FREE NUMERICAL METHOD FOR SOLUTION OF CERTAIN HJB PDES
A CURSE-OF-DIMENSIONALITY-FREE NUMERICAL METHOD FOR SOLUTION OF CERTAIN HJB PDES WILLIAM M. MCENEANEY Abstract. In previous work of the author and others, max-plus methods have been explored for solution
More informationSébastien Chaumont a a Institut Élie Cartan, Université Henri Poincaré Nancy I, B. P. 239, Vandoeuvre-lès-Nancy Cedex, France. 1.
A strong comparison result for viscosity solutions to Hamilton-Jacobi-Bellman equations with Dirichlet condition on a non-smooth boundary and application to parabolic problems Sébastien Chaumont a a Institut
More informationFenchel-Moreau Conjugates of Inf-Transforms and Application to Stochastic Bellman Equation
Fenchel-Moreau Conjugates of Inf-Transforms and Application to Stochastic Bellman Equation Jean-Philippe Chancelier and Michel De Lara CERMICS, École des Ponts ParisTech First Conference on Discrete Optimization
More informationIntroduction to Bases in Banach Spaces
Introduction to Bases in Banach Spaces Matt Daws June 5, 2005 Abstract We introduce the notion of Schauder bases in Banach spaces, aiming to be able to give a statement of, and make sense of, the Gowers
More informationDuality (Continued) min f ( x), X R R. Recall, the general primal problem is. The Lagrangian is a function. defined by
Duality (Continued) Recall, the general primal problem is min f ( x), xx g( x) 0 n m where X R, f : X R, g : XR ( X). he Lagrangian is a function L: XR R m defined by L( xλ, ) f ( x) λ g( x) Duality (Continued)
More informationStability of Stochastic Programming Problems
Stability of Stochastic Programming Problems W. Römisch Humboldt-University Berlin Institute of Mathematics 10099 Berlin, Germany http://www.math.hu-berlin.de/~romisch Page 1 of 35 Spring School Stochastic
More informationStability of optimization problems with stochastic dominance constraints
Stability of optimization problems with stochastic dominance constraints D. Dentcheva and W. Römisch Stevens Institute of Technology, Hoboken Humboldt-University Berlin www.math.hu-berlin.de/~romisch SIAM
More informationNonlinear L 2 -gain analysis via a cascade
9th IEEE Conference on Decision and Control December -7, Hilton Atlanta Hotel, Atlanta, GA, USA Nonlinear L -gain analysis via a cascade Peter M Dower, Huan Zhang and Christopher M Kellett Abstract A nonlinear
More informationSome SDEs with distributional drift Part I : General calculus. Flandoli, Franco; Russo, Francesco; Wolf, Jochen
Title Author(s) Some SDEs with distributional drift Part I : General calculus Flandoli, Franco; Russo, Francesco; Wolf, Jochen Citation Osaka Journal of Mathematics. 4() P.493-P.54 Issue Date 3-6 Text
More informationProofs for Large Sample Properties of Generalized Method of Moments Estimators
Proofs for Large Sample Properties of Generalized Method of Moments Estimators Lars Peter Hansen University of Chicago March 8, 2012 1 Introduction Econometrica did not publish many of the proofs in my
More informationA Concise Course on Stochastic Partial Differential Equations
A Concise Course on Stochastic Partial Differential Equations Michael Röckner Reference: C. Prevot, M. Röckner: Springer LN in Math. 1905, Berlin (2007) And see the references therein for the original
More informationMean-field dual of cooperative reproduction
The mean-field dual of systems with cooperative reproduction joint with Tibor Mach (Prague) A. Sturm (Göttingen) Friday, July 6th, 2018 Poisson construction of Markov processes Let (X t ) t 0 be a continuous-time
More informationStochastic optimal control with rough paths
Stochastic optimal control with rough paths Paul Gassiat TU Berlin Stochastic processes and their statistics in Finance, Okinawa, October 28, 2013 Joint work with Joscha Diehl and Peter Friz Introduction
More informationNotes on Ordered Sets
Notes on Ordered Sets Mariusz Wodzicki September 10, 2013 1 Vocabulary 1.1 Definitions Definition 1.1 A binary relation on a set S is said to be a partial order if it is reflexive, x x, weakly antisymmetric,
More informationON CALCULATING THE VALUE OF A DIFFERENTIAL GAME IN THE CLASS OF COUNTER STRATEGIES 1,2
URAL MATHEMATICAL JOURNAL, Vol. 2, No. 1, 2016 ON CALCULATING THE VALUE OF A DIFFERENTIAL GAME IN THE CLASS OF COUNTER STRATEGIES 1,2 Mikhail I. Gomoyunov Krasovskii Institute of Mathematics and Mechanics,
More informationAn iterative procedure for constructing subsolutions of discrete-time optimal control problems
An iterative procedure for constructing subsolutions of discrete-time optimal control problems Markus Fischer version of November, 2011 Abstract An iterative procedure for constructing subsolutions of
More informationLebesgue-Stieltjes measures and the play operator
Lebesgue-Stieltjes measures and the play operator Vincenzo Recupero Politecnico di Torino, Dipartimento di Matematica Corso Duca degli Abruzzi, 24, 10129 Torino - Italy E-mail: vincenzo.recupero@polito.it
More informationMATH MEASURE THEORY AND FOURIER ANALYSIS. Contents
MATH 3969 - MEASURE THEORY AND FOURIER ANALYSIS ANDREW TULLOCH Contents 1. Measure Theory 2 1.1. Properties of Measures 3 1.2. Constructing σ-algebras and measures 3 1.3. Properties of the Lebesgue measure
More informationComputational Issues in Nonlinear Dynamics and Control
Computational Issues in Nonlinear Dynamics and Control Arthur J. Krener ajkrener@ucdavis.edu Supported by AFOSR and NSF Typical Problems Numerical Computation of Invariant Manifolds Typical Problems Numerical
More informationThe Lebesgue Integral
The Lebesgue Integral Brent Nelson In these notes we give an introduction to the Lebesgue integral, assuming only a knowledge of metric spaces and the iemann integral. For more details see [1, Chapters
More information2. Function spaces and approximation
2.1 2. Function spaces and approximation 2.1. The space of test functions. Notation and prerequisites are collected in Appendix A. Let Ω be an open subset of R n. The space C0 (Ω), consisting of the C
More informationRecall that if X is a compact metric space, C(X), the space of continuous (real-valued) functions on X, is a Banach space with the norm
Chapter 13 Radon Measures Recall that if X is a compact metric space, C(X), the space of continuous (real-valued) functions on X, is a Banach space with the norm (13.1) f = sup x X f(x). We want to identify
More informationDUALITY IN COUNTABLY INFINITE MONOTROPIC PROGRAMS
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 DUALITY IN COUNTABLY INFINITE MONOTROPIC PROGRAMS ARCHIS GHATE Abstract. Finite-dimensional
More informationRisk-Sensitive Control with HARA Utility
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 46, NO. 4, APRIL 2001 563 Risk-Sensitive Control with HARA Utility Andrew E. B. Lim Xun Yu Zhou, Senior Member, IEEE Abstract In this paper, a control methodology
More informationApproximate dynamic programming for stochastic reachability
Approximate dynamic programming for stochastic reachability Nikolaos Kariotoglou, Sean Summers, Tyler Summers, Maryam Kamgarpour and John Lygeros Abstract In this work we illustrate how approximate dynamic
More informationLinear Quadratic Zero-Sum Two-Person Differential Games Pierre Bernhard June 15, 2013
Linear Quadratic Zero-Sum Two-Person Differential Games Pierre Bernhard June 15, 2013 Abstract As in optimal control theory, linear quadratic (LQ) differential games (DG) can be solved, even in high dimension,
More informationMax Plus Eigenvector Methods for Nonlinear H Problems: Error Analysis
Max Plus Eigenvector Methods for Nonlinear H Problems: Error Analysis William M. McEneaney August 28, 2003 Abstract The H problem for a nonlinear system is considered. The corresponding dynamic programming
More informationNumerical schemes of resolution of stochastic optimal control HJB equation
Numerical schemes of resolution of stochastic optimal control HJB equation Elisabeth Ottenwaelter Journée des doctorants 7 mars 2007 Équipe COMMANDS Elisabeth Ottenwaelter (INRIA et CMAP) Numerical schemes
More informationSMALL SUBSETS OF THE REALS AND TREE FORCING NOTIONS
SMALL SUBSETS OF THE REALS AND TREE FORCING NOTIONS MARCIN KYSIAK AND TOMASZ WEISS Abstract. We discuss the question which properties of smallness in the sense of measure and category (e.g. being a universally
More informationProbability and Measure
Part II Year 2018 2017 2016 2015 2014 2013 2012 2011 2010 2009 2008 2007 2006 2005 2018 84 Paper 4, Section II 26J Let (X, A) be a measurable space. Let T : X X be a measurable map, and µ a probability
More informationState Estimation with Finite Signal-to-Noise Models
State Estimation with Finite Signal-to-Noise Models Weiwei Li and Robert E. Skelton Department of Mechanical and Aerospace Engineering University of California San Diego, La Jolla, CA 9293-411 wwli@mechanics.ucsd.edu
More informationViscosity Solutions of the Bellman Equation for Perturbed Optimal Control Problems with Exit Times 0
Viscosity Solutions of the Bellman Equation for Perturbed Optimal Control Problems with Exit Times Michael Malisoff Department of Mathematics Louisiana State University Baton Rouge, LA 783-4918 USA malisoff@mathlsuedu
More informationBehaviour of Lipschitz functions on negligible sets. Non-differentiability in R. Outline
Behaviour of Lipschitz functions on negligible sets G. Alberti 1 M. Csörnyei 2 D. Preiss 3 1 Università di Pisa 2 University College London 3 University of Warwick Lars Ahlfors Centennial Celebration Helsinki,
More informationAliprantis, Border: Infinite-dimensional Analysis A Hitchhiker s Guide
aliprantis.tex May 10, 2011 Aliprantis, Border: Infinite-dimensional Analysis A Hitchhiker s Guide Notes from [AB2]. 1 Odds and Ends 2 Topology 2.1 Topological spaces Example. (2.2) A semimetric = triangle
More informationAnalysis Qualifying Exam
Analysis Qualifying Exam Spring 2017 Problem 1: Let f be differentiable on R. Suppose that there exists M > 0 such that f(k) M for each integer k, and f (x) M for all x R. Show that f is bounded, i.e.,
More informationIntroduction to Stochastic Optimization Part 4: Multi-stage decision
Introduction to Stochastic Optimization Part 4: Multi-stage decision problems April 23, 29 The problem ξ = (ξ,..., ξ T ) a multivariate time series process (e.g. future interest rates, future asset prices,
More informationINFINITE TIME HORIZON OPTIMAL CONTROL OF THE SEMILINEAR HEAT EQUATION
Nonlinear Funct. Anal. & Appl., Vol. 7, No. (22), pp. 69 83 INFINITE TIME HORIZON OPTIMAL CONTROL OF THE SEMILINEAR HEAT EQUATION Mihai Sîrbu Abstract. We consider here the infinite horizon control problem
More informationLINEAR-CONVEX CONTROL AND DUALITY
1 LINEAR-CONVEX CONTROL AND DUALITY R.T. Rockafellar Department of Mathematics, University of Washington Seattle, WA 98195-4350, USA Email: rtr@math.washington.edu R. Goebel 3518 NE 42 St., Seattle, WA
More informationNew Identities for Weak KAM Theory
New Identities for Weak KAM Theory Lawrence C. Evans Department of Mathematics University of California, Berkeley Abstract This paper records for the Hamiltonian H = p + W (x) some old and new identities
More informationOn duality theory of conic linear problems
On duality theory of conic linear problems Alexander Shapiro School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, Georgia 3332-25, USA e-mail: ashapiro@isye.gatech.edu
More informationApproximate Bisimulations for Constrained Linear Systems
Approximate Bisimulations for Constrained Linear Systems Antoine Girard and George J Pappas Abstract In this paper, inspired by exact notions of bisimulation equivalence for discrete-event and continuous-time
More informationOn a class of stochastic differential equations in a financial network model
1 On a class of stochastic differential equations in a financial network model Tomoyuki Ichiba Department of Statistics & Applied Probability, Center for Financial Mathematics and Actuarial Research, University
More informationA Narrow-Stencil Finite Difference Method for Hamilton-Jacobi-Bellman Equations
A Narrow-Stencil Finite Difference Method for Hamilton-Jacobi-Bellman Equations Xiaobing Feng Department of Mathematics The University of Tennessee, Knoxville, U.S.A. Linz, November 23, 2016 Collaborators
More informationObstacle problems and isotonicity
Obstacle problems and isotonicity Thomas I. Seidman Revised version for NA-TMA: NA-D-06-00007R1+ [June 6, 2006] Abstract For variational inequalities of an abstract obstacle type, a comparison principle
More informationConstructive Proof of the Fan-Glicksberg Fixed Point Theorem for Sequentially Locally Non-constant Multi-functions in a Locally Convex Space
Constructive Proof of the Fan-Glicksberg Fixed Point Theorem for Sequentially Locally Non-constant Multi-functions in a Locally Convex Space Yasuhito Tanaka, Member, IAENG, Abstract In this paper we constructively
More informationOptimal Control and Viscosity Solutions of Hamilton-Jacobi-Bellman Equations
Martino Bardi Italo Capuzzo-Dolcetta Optimal Control and Viscosity Solutions of Hamilton-Jacobi-Bellman Equations Birkhauser Boston Basel Berlin Contents Preface Basic notations xi xv Chapter I. Outline
More informationn [ F (b j ) F (a j ) ], n j=1(a j, b j ] E (4.1)
1.4. CONSTRUCTION OF LEBESGUE-STIELTJES MEASURES In this section we shall put to use the Carathéodory-Hahn theory, in order to construct measures with certain desirable properties first on the real line
More informationPARTIAL REGULARITY OF BRENIER SOLUTIONS OF THE MONGE-AMPÈRE EQUATION
PARTIAL REGULARITY OF BRENIER SOLUTIONS OF THE MONGE-AMPÈRE EQUATION ALESSIO FIGALLI AND YOUNG-HEON KIM Abstract. Given Ω, Λ R n two bounded open sets, and f and g two probability densities concentrated
More informationStability of Feedback Solutions for Infinite Horizon Noncooperative Differential Games
Stability of Feedback Solutions for Infinite Horizon Noncooperative Differential Games Alberto Bressan ) and Khai T. Nguyen ) *) Department of Mathematics, Penn State University **) Department of Mathematics,
More informationMulti-dimensional Stochastic Singular Control Via Dynkin Game and Dirichlet Form
Multi-dimensional Stochastic Singular Control Via Dynkin Game and Dirichlet Form Yipeng Yang * Under the supervision of Dr. Michael Taksar Department of Mathematics University of Missouri-Columbia Oct
More informationWhere is matrix multiplication locally open?
Linear Algebra and its Applications 517 (2017) 167 176 Contents lists available at ScienceDirect Linear Algebra and its Applications www.elsevier.com/locate/laa Where is matrix multiplication locally open?
More informationAPPENDIX C: Measure Theoretic Issues
APPENDIX C: Measure Theoretic Issues A general theory of stochastic dynamic programming must deal with the formidable mathematical questions that arise from the presence of uncountable probability spaces.
More informationLocal semiconvexity of Kantorovich potentials on non-compact manifolds
Local semiconvexity of Kantorovich potentials on non-compact manifolds Alessio Figalli, Nicola Gigli Abstract We prove that any Kantorovich potential for the cost function c = d / on a Riemannian manifold
More informationPart V. 17 Introduction: What are measures and why measurable sets. Lebesgue Integration Theory
Part V 7 Introduction: What are measures and why measurable sets Lebesgue Integration Theory Definition 7. (Preliminary). A measure on a set is a function :2 [ ] such that. () = 2. If { } = is a finite
More informationPROBLEMS. (b) (Polarization Identity) Show that in any inner product space
1 Professor Carl Cowen Math 54600 Fall 09 PROBLEMS 1. (Geometry in Inner Product Spaces) (a) (Parallelogram Law) Show that in any inner product space x + y 2 + x y 2 = 2( x 2 + y 2 ). (b) (Polarization
More informationUniversity of Warwick, EC9A0 Maths for Economists Lecture Notes 10: Dynamic Programming
University of Warwick, EC9A0 Maths for Economists 1 of 63 University of Warwick, EC9A0 Maths for Economists Lecture Notes 10: Dynamic Programming Peter J. Hammond Autumn 2013, revised 2014 University of
More informationTHEOREMS, ETC., FOR MATH 515
THEOREMS, ETC., FOR MATH 515 Proposition 1 (=comment on page 17). If A is an algebra, then any finite union or finite intersection of sets in A is also in A. Proposition 2 (=Proposition 1.1). For every
More informationWiener Measure and Brownian Motion
Chapter 16 Wiener Measure and Brownian Motion Diffusion of particles is a product of their apparently random motion. The density u(t, x) of diffusing particles satisfies the diffusion equation (16.1) u
More informationTHEOREMS, ETC., FOR MATH 516
THEOREMS, ETC., FOR MATH 516 Results labeled Theorem Ea.b.c (or Proposition Ea.b.c, etc.) refer to Theorem c from section a.b of Evans book (Partial Differential Equations). Proposition 1 (=Proposition
More informationDuality and dynamics in Hamilton-Jacobi theory for fully convex problems of control
Duality and dynamics in Hamilton-Jacobi theory for fully convex problems of control RTyrrell Rockafellar and Peter R Wolenski Abstract This paper describes some recent results in Hamilton- Jacobi theory
More informationMATH41011/MATH61011: FOURIER SERIES AND LEBESGUE INTEGRATION. Extra Reading Material for Level 4 and Level 6
MATH41011/MATH61011: FOURIER SERIES AND LEBESGUE INTEGRATION Extra Reading Material for Level 4 and Level 6 Part A: Construction of Lebesgue Measure The first part the extra material consists of the construction
More informationA note on scenario reduction for two-stage stochastic programs
A note on scenario reduction for two-stage stochastic programs Holger Heitsch a and Werner Römisch a a Humboldt-University Berlin, Institute of Mathematics, 199 Berlin, Germany We extend earlier work on
More informationINSTITUTE of MATHEMATICS. ACADEMY of SCIENCES of the CZECH REPUBLIC. A universal operator on the Gurariĭ space
INSTITUTE of MATHEMATICS Academy of Sciences Czech Republic INSTITUTE of MATHEMATICS ACADEMY of SCIENCES of the CZECH REPUBLIC A universal operator on the Gurariĭ space Joanna Garbulińska-Wȩgrzyn Wiesław
More informationNear-Potential Games: Geometry and Dynamics
Near-Potential Games: Geometry and Dynamics Ozan Candogan, Asuman Ozdaglar and Pablo A. Parrilo September 6, 2011 Abstract Potential games are a special class of games for which many adaptive user dynamics
More informationNew Results for Second Order Discrete Hamiltonian Systems. Huiwen Chen*, Zhimin He, Jianli Li and Zigen Ouyang
TAIWANESE JOURNAL OF MATHEMATICS Vol. xx, No. x, pp. 1 26, xx 20xx DOI: 10.11650/tjm/7762 This paper is available online at http://journal.tms.org.tw New Results for Second Order Discrete Hamiltonian Systems
More informationOn a Class of Multidimensional Optimal Transportation Problems
Journal of Convex Analysis Volume 10 (2003), No. 2, 517 529 On a Class of Multidimensional Optimal Transportation Problems G. Carlier Université Bordeaux 1, MAB, UMR CNRS 5466, France and Université Bordeaux
More information6 Classical dualities and reflexivity
6 Classical dualities and reflexivity 1. Classical dualities. Let (Ω, A, µ) be a measure space. We will describe the duals for the Banach spaces L p (Ω). First, notice that any f L p, 1 p, generates the
More information1. Stochastic Processes and filtrations
1. Stochastic Processes and 1. Stoch. pr., A stochastic process (X t ) t T is a collection of random variables on (Ω, F) with values in a measurable space (S, S), i.e., for all t, In our case X t : Ω S
More informationCitation Osaka Journal of Mathematics. 41(4)
TitleA non quasi-invariance of the Brown Authors Sadasue, Gaku Citation Osaka Journal of Mathematics. 414 Issue 4-1 Date Text Version publisher URL http://hdl.handle.net/1194/1174 DOI Rights Osaka University
More informationChapter 4. Measure Theory. 1. Measure Spaces
Chapter 4. Measure Theory 1. Measure Spaces Let X be a nonempty set. A collection S of subsets of X is said to be an algebra on X if S has the following properties: 1. X S; 2. if A S, then A c S; 3. if
More informationPresenter: Noriyoshi Fukaya
Y. Martel, F. Merle, and T.-P. Tsai, Stability and Asymptotic Stability in the Energy Space of the Sum of N Solitons for Subcritical gkdv Equations, Comm. Math. Phys. 31 (00), 347-373. Presenter: Noriyoshi
More information5 Measure theory II. (or. lim. Prove the proposition. 5. For fixed F A and φ M define the restriction of φ on F by writing.
5 Measure theory II 1. Charges (signed measures). Let (Ω, A) be a σ -algebra. A map φ: A R is called a charge, (or signed measure or σ -additive set function) if φ = φ(a j ) (5.1) A j for any disjoint
More informationOn the Bellman equation for control problems with exit times and unbounded cost functionals 1
On the Bellman equation for control problems with exit times and unbounded cost functionals 1 Michael Malisoff Department of Mathematics, Hill Center-Busch Campus Rutgers University, 11 Frelinghuysen Road
More informationWHY SATURATED PROBABILITY SPACES ARE NECESSARY
WHY SATURATED PROBABILITY SPACES ARE NECESSARY H. JEROME KEISLER AND YENENG SUN Abstract. An atomless probability space (Ω, A, P ) is said to have the saturation property for a probability measure µ on
More informationON THE POLICY IMPROVEMENT ALGORITHM IN CONTINUOUS TIME
ON THE POLICY IMPROVEMENT ALGORITHM IN CONTINUOUS TIME SAUL D. JACKA AND ALEKSANDAR MIJATOVIĆ Abstract. We develop a general approach to the Policy Improvement Algorithm (PIA) for stochastic control problems
More informationFast Marching Methods for Front Propagation Lecture 1/3
Outline for Front Propagation Lecture 1/3 M. Falcone Dipartimento di Matematica SAPIENZA, Università di Roma Autumn School Introduction to Numerical Methods for Moving Boundaries ENSTA, Paris, November,
More information1 Stochastic Dynamic Programming
1 Stochastic Dynamic Programming Formally, a stochastic dynamic program has the same components as a deterministic one; the only modification is to the state transition equation. When events in the future
More informationDiscreteness of Transmission Eigenvalues via Upper Triangular Compact Operators
Discreteness of Transmission Eigenvalues via Upper Triangular Compact Operators John Sylvester Department of Mathematics University of Washington Seattle, Washington 98195 U.S.A. June 3, 2011 This research
More informationCharacterizing Robust Solution Sets of Convex Programs under Data Uncertainty
Characterizing Robust Solution Sets of Convex Programs under Data Uncertainty V. Jeyakumar, G. M. Lee and G. Li Communicated by Sándor Zoltán Németh Abstract This paper deals with convex optimization problems
More informationITERATIVE SCHEMES FOR APPROXIMATING SOLUTIONS OF ACCRETIVE OPERATORS IN BANACH SPACES SHOJI KAMIMURA AND WATARU TAKAHASHI. Received December 14, 1999
Scientiae Mathematicae Vol. 3, No. 1(2000), 107 115 107 ITERATIVE SCHEMES FOR APPROXIMATING SOLUTIONS OF ACCRETIVE OPERATORS IN BANACH SPACES SHOJI KAMIMURA AND WATARU TAKAHASHI Received December 14, 1999
More informationEn busca de la linealidad en Matemáticas
En busca de la linealidad en Matemáticas M. Cueto, E. Gómez, J. Llorente, E. Martínez, D. Rodríguez, E. Sáez VIII Escuela taller de Análisis Funcional Bilbao, 9 Marzo 2018 1 Motivation 2 Everywhere differentiable
More informationSUB-OPTIMAL RISK-SENSITIVE FILTERING FOR THIRD DEGREE POLYNOMIAL STOCHASTIC SYSTEMS
ICIC Express Letters ICIC International c 2008 ISSN 1881-803X Volume 2, Number 4, December 2008 pp. 371-377 SUB-OPTIMAL RISK-SENSITIVE FILTERING FOR THIRD DEGREE POLYNOMIAL STOCHASTIC SYSTEMS Ma. Aracelia
More informationEE/ACM Applications of Convex Optimization in Signal Processing and Communications Lecture 17
EE/ACM 150 - Applications of Convex Optimization in Signal Processing and Communications Lecture 17 Andre Tkacenko Signal Processing Research Group Jet Propulsion Laboratory May 29, 2012 Andre Tkacenko
More informationCHAPTER V DUAL SPACES
CHAPTER V DUAL SPACES DEFINITION Let (X, T ) be a (real) locally convex topological vector space. By the dual space X, or (X, T ), of X we mean the set of all continuous linear functionals on X. By the
More informationSolutions to Tutorial 11 (Week 12)
THE UIVERSITY OF SYDEY SCHOOL OF MATHEMATICS AD STATISTICS Solutions to Tutorial 11 (Week 12) MATH3969: Measure Theory and Fourier Analysis (Advanced) Semester 2, 2017 Web Page: http://sydney.edu.au/science/maths/u/ug/sm/math3969/
More informationThe Nearest Doubly Stochastic Matrix to a Real Matrix with the same First Moment
he Nearest Doubly Stochastic Matrix to a Real Matrix with the same First Moment William Glunt 1, homas L. Hayden 2 and Robert Reams 2 1 Department of Mathematics and Computer Science, Austin Peay State
More informationFiltered scheme and error estimate for first order Hamilton-Jacobi equations
and error estimate for first order Hamilton-Jacobi equations Olivier Bokanowski 1 Maurizio Falcone 2 2 1 Laboratoire Jacques-Louis Lions, Université Paris-Diderot (Paris 7) 2 SAPIENZA - Università di Roma
More informationThe small ball property in Banach spaces (quantitative results)
The small ball property in Banach spaces (quantitative results) Ehrhard Behrends Abstract A metric space (M, d) is said to have the small ball property (sbp) if for every ε 0 > 0 there exists a sequence
More informationSUMMARY OF RESULTS ON PATH SPACES AND CONVERGENCE IN DISTRIBUTION FOR STOCHASTIC PROCESSES
SUMMARY OF RESULTS ON PATH SPACES AND CONVERGENCE IN DISTRIBUTION FOR STOCHASTIC PROCESSES RUTH J. WILLIAMS October 2, 2017 Department of Mathematics, University of California, San Diego, 9500 Gilman Drive,
More informationNon commutative Khintchine inequalities and Grothendieck s theo
Non commutative Khintchine inequalities and Grothendieck s theorem Nankai, 2007 Plan Non-commutative Khintchine inequalities 1 Non-commutative Khintchine inequalities 2 µ = Uniform probability on the set
More informationRegularity of the density for the stochastic heat equation
Regularity of the density for the stochastic heat equation Carl Mueller 1 Department of Mathematics University of Rochester Rochester, NY 15627 USA email: cmlr@math.rochester.edu David Nualart 2 Department
More informationLARGE DEVIATIONS OF TYPICAL LINEAR FUNCTIONALS ON A CONVEX BODY WITH UNCONDITIONAL BASIS. S. G. Bobkov and F. L. Nazarov. September 25, 2011
LARGE DEVIATIONS OF TYPICAL LINEAR FUNCTIONALS ON A CONVEX BODY WITH UNCONDITIONAL BASIS S. G. Bobkov and F. L. Nazarov September 25, 20 Abstract We study large deviations of linear functionals on an isotropic
More informationRESTRICTED UNIFORM BOUNDEDNESS IN BANACH SPACES
RESTRICTED UNIFORM BOUNDEDNESS IN BANACH SPACES OLAV NYGAARD AND MÄRT PÕLDVERE Abstract. Precise conditions for a subset A of a Banach space X are known in order that pointwise bounded on A sequences of
More informationAn Empirical Algorithm for Relative Value Iteration for Average-cost MDPs
2015 IEEE 54th Annual Conference on Decision and Control CDC December 15-18, 2015. Osaka, Japan An Empirical Algorithm for Relative Value Iteration for Average-cost MDPs Abhishek Gupta Rahul Jain Peter
More information