Metrics for Markov Decision Processes with Infinite State Spaces

Size: px
Start display at page:

Download "Metrics for Markov Decision Processes with Infinite State Spaces"

Transcription

1 Metrics for Mrkov Decision Processes with Infinite Stte Spces Norm Ferns School of Computer Science McGill University Montrél, Cnd, H3A 2A7 Prksh Pnngden School of Computer Science McGill University Montrél, Cnd, H3A 2A7 Doin Precup School of Computer Science McGill University Montrél, Cnd, H3A 2A7 Abstrct We present metrics for mesuring stte similrity in Mrkov decision processes (MDPs) with infinitely mny sttes, including MDPs with continuous stte spces. Such metrics provide stble quntittive nlogue of the notion of bisimultion for MDPs, nd re suitble for use in MDP pproximtion. We show tht the optiml vlue function ssocited with discounted infinite horizon plnning tsk vries continuously with respect to our metric distnces. 1 Introduction Mrkov decision processes (MDPs) offer populr mthemticl tool for plnning nd lerning in the presence of uncertinty (Boutilier et l., 1999). MDPs re stndrd formlism for describing multi-stge decision mking in probbilistic environments. The objective of the decision mking is to mximize cumultive mesure of long-term performnce, clled the return. Dynmic progrmming lgorithms, e.g., vlue itertion or policy itertion (Putermn, 1994), llow us to compute the optiml expected return for ny stte, s well s the wy of behving (policy) tht genertes this return. However, in mny prcticl pplictions, the stte spce of n MDP is simply too lrge, possibly infinite or even continuous, for such stndrd lgorithms to be pplied. A typicl mens of overcoming such circumstnces is to prtition the stte spce in the hope of obtining n essentilly equivlent reduced system. One defines new MDP over the prtition blocks, nd if it is smll enough, it cn be solved by clssicl methods. The hope is tht optiml vlues nd policies for the reduced MDP cn be extended to optiml vlues nd policies for the originl MDP. Recent MDP reserch on defining equivlence reltions on MDPs (Givn et l., 2003) hs built on the notion of strong probbilistic bisimultion from concurrency theory. Probbilistic bisimultion ws introduced by Lrsen nd Skou (1991) bsed on bisimultion for nondeterministic (nonprobbilistic) systems due to Prk (1981) nd Milner (1980). Henceforth when we sy bisimultion we will men strong probbilistic bisimultion. In probbilistic setting, bisimultion cn be described s n equivlence reltion tht reltes two sttes precisely when they hve the sme probbility of trnsitioning to clsses of equivlent sttes. The extension of bisimultion to trnsition systems with rewrds ws crried out in the context of MDPs by Givn, Den nd Greig (2003) nd in the context of performnce evlution by Bernrdo nd Brvetti (2003). In both cses, the motivtion is to use the equivlence reltion to ggregte the sttes nd get smller stte spces. The bsic notion of bisimultion is modified only slightly by the introduction of rewrds. However, it hs been well estblished for while now tht use of exct equivlences in quntittive systems is problemtic. A notion of equivlence is two-vlued: two sttes re either equivlent or not equivlent. A smll perturbtion of the trnsition probbilities cn mke two equivlent sttes no longer equivlent. In short, ny kind of equivlence is too unstble to perturbtions of the numericl vlues of the trnsition probbilities. A nturl remedy is to use metrics. Metrics re nturl quntittive nlogues of the notion of equivlence reltion: for exmple the tringle inequlity is nturl quntittive nlogue of trnsitivity. The metrics on which we focus here specify the degree to which objects of interest behve similrly. Much of this work hs been done in very generl setting, using the lbelled Mrkov process (LMP) model (Blute et l., 1997; Deshrnis et l., 2002). Previous metrics (Deshrnis et l., 1999; vn Breugel & Worrell, 2001; Deshrnis et l., 2002b) (more precisely pseudo-metrics or semi-metrics) hve quntittively generlized bisimultion by ssigning distnce

2 zero to sttes tht re bisimilr, distnce one to sttes tht re esily distinguishble, nd n intermedite distnce to those in between. In (vn Breugel & Worrell, 2001) it ws shown how, in simplified setting of finite stte spce LMPs, metric distnces could be clculted in polynomil time. This work, long with tht of (Deshrnis et l., 2002b), ws dpted to finite MDPs in (Ferns et l., 2004). There, we used fixed point theory to construct metrics, ech of which hd bisimultion s its kernel, ws sensitive to perturbtions in MDP prmeters, nd provided bounds on the optiml vlues of sttes. We showed how to compute the metrics up to ny prescribed degree of ccurcy nd then used them to directly ggregte smple finite MDPs. In this pper we present significnt generliztion of these previous results to MDPs with continuous stte spces. The liner progrmming rguments we used in our previous work no longer pply, nd we hve to use mesure theory nd dulity theory on continuous stte spces. The mthemticl theory is interesting in its own right. Although continuous MDPs re of gret interest for prcticl pplictions, e.g. in the res of utomted control nd robotics, the existing methods for mesuring distnces between sttes, for the purpose of stte ggregtion s well s other pproximtion methods re still lrgely heuristic. As result, it is hrd to provide gurnteed error bounds between the correct nd the pproximte vlue function. It is lso difficult to determine the impct tht structurl chnges in the pproximtor would hve on the qulity on the pproximtion. The metrics we define in this pper llow the definition of error bounds for vlue functions. These bounds cn be used s tool in the nlysis of existing pproximtion schemes. The pper is orgnized s follows. In sections 2 nd 3 we provide the theoreticl tools necessry for the construction of our metrics. The ctul construction is crried out in section 4, where we lso rgue tht our metrics re the best for the job. Section 5 provides proof of vlue function continuity with respect to our metrics. In section 6 we provide simple illustrtion of metric use in pproximtion. Finlly, section 7 contins our conclusions nd directions for future work. 2 Bckground 2.1 Mrkov Decision Processes Let (S, A, P, r) be Mrkov decision process (MDP), where S is complete seprble metric spce equipped with its Borel sigm lgebr Σ, A is finite set of ctions, r : S A R is mesurble rewrd function, nd P : S A Σ [0, 1] is lbeled stochstic trnsition kernel, i.e. A, s S, P (s,, ) : Σ [0, 1] is probbility mesure, nd A, X Σ, P (,, X) : S [0, 1] is mesurble function. We will use the following nottion: for A nd s S, P s denotes P (s,, ) nd r s denotes r(s, ). Given mesure P nd integrble function f, we denote the integrl of f with respect to P by P (f). We lso mke the following ssumptions: 1. B := sup s,s, rs rs <. 2. For ech A, r(, ) is continuous on S. 3. For ech A, P s is (wekly) continuous s function of s, i.e. if s n tends to s in S then for every bounded continuous function f : S R, P s n (f) tends to P s (f). The first ssumption is direct consequence of the stndrd ssumption tht rewrds re bounded. The second ssumption is non-stndrd, but very mild. In generl, rewrds in n MDP re not ssumed to vry continuously (e.g., in gol-directed tsks). However, it is generlly ssumed tht there would be finite or countble number of discontinuities. In this cse, it is esy to trnsform the rewrd structure into one tht is continuous nd rbitrrily close to the originl one, e.g. by pplying smoothing sigmoid functions t the points of discontinuity. The third ssumption is continuity ssumption on the trnsition probbilities, nd stisfied by most resonble systems (including physicl systems of interest in control nd robotics). The discounted, infinite horizon plnning tsk in n MDP is to determine policy π : S A tht mximizes the vlue of every stte, V π (s) = E[ t=0 r t s 0 = s, π], where s 0 is the stte t time 0, r t is the rewrd chieved t time t, γ is discount fctor in (0, 1), nd the expecttion is tken by following the stte dynmics induced by π. The function V π is clled the vlue function of policy π. The optiml vlue function V, ssocited with n optiml policy, is the unique solution of the fixed point eqution V (s) = mx A (r s + γp s (V )) nd cn be used to directly determine n optiml policy, provided it is computble. Note tht in generl the optiml vlue function need not be mesurble, in which cse the fixed point eqution would be invlid. However, under ssumptions 1-3, this cnnot

3 be the cse (see theorem of (Putermn, 1994)). In fct, in this cse, the optiml vlue function cn be computed s the limit of sequence of itertes. Define V 0 = 0 nd V n+1 (s) = mx A (r s + γp s (V n )). Then the V n s converge to V in the uniform (mx-norm) metric. Of course, for this computtion to work in prctice it would be desirble to work with smll discretized version of the given MDP. This brings bout the problem of pproximtion, nd finding forml definition which chrcterizes when sttes re equivlent (nd hence cn be lumped together). The correct equivlence reltion is bisimultion. 2.2 Bisimultion Bisimultion is notion of behviourl equivlence, the strongest of whole zoo of equivlence reltions considered in concurrency theory. Bisimultion cn be defined solely in terms of reltions or using fixed point theory (so clled co-induction). The ltter will be useful for our purposes, but first requires some bsic definitions nd tools from fixed point theory on lttices tht cn be found, for exmple, in (Winskel, 1993). Let (L, ) be prtil order. If it hs lest upper bounds nd gretest lower bounds of rbitrry subsets of elements, then it is sid to be complete lttice. A function f : L L is sid to be monotone if x x implies f(x) f(x ). A point x in L is sid to be prefixed point if f(x) x, postfixed point if x f(x) nd fixed point if x = f(x). The importnce of these definitions rises in the following theorem. Theorem Let L be complete lttice, nd suppose f : L L is monotone. Then f hs lest fixed point, which is lso its lest prefixed point, nd f hs gretest fixed point, which is lso its gretest postfixed point. Let REL be the complete lttice of binry reltions on S with the usul subset ordering. We sy set in X is R-closed if the collection of ll those elements of S tht re rechble by R from X is itself contined in X. When R is n equivlence reltion this is equivlent to sying tht X is union of R-equivlence clsses. We write R rst for the reflexive, symmetric, trnsitive closure of R, nd Σ(R) for those Σ-mesurble sets tht re R-closed. Definition 2.2. Define F : REL REL by sf(r)s A, rs = rs nd X 1 This is n elementry theorem sometimes clled the Knester-Trski theorem in the literture. In fct the Knester-Trski theorem is much stronger sttement to the effect tht the collection of fixed points is itself complete lttice. Σ(R rst ), Ps (X) = Ps (X). The gretest fixed point of F is bisimultion. The existence of bisimultion is gurnteed by the fixed-point theorem. Unfortuntely, s n exct equivlence, bisimultion suffers from issues of instbility; tht is, slight numericl differences in the MDP prmeters, r nd P, cn led to vstly different bisimultion prtitions. To get round this, one generlizes the notion of equivlence through metrics. 2.3 Metrics Definition 2.3. A semimetric 2 on S is mp d : S S [0, ) such tht for ll s, s, s : 1. s = s d(s, s ) = 0 2. d(s, s ) = d(s, s) 3. d(s, s ) d(s, s ) + d(s, s ) If the converse of the first xiom holds s well, we sy d is metric. 3 Recll tht function h : S S R is lower semicontinuous (lsc) if whenever (s n, s n) tends to (s, s ), lim inf h(s n, s n) h(s, s ). Here we re considering S S to be endowed with the product topology. Note tht lsc functions re product mesurble. Let M be the set of semimetrics on S tht re lsc on S S nd uniformly bounded, e.g. those ssigning distnce t most 1, nd give it the usul pointwise ordering. Then M is complete lttice. This follows becuse tking the pointwise supremum of n rbitrry collection of lsc functions yields lsc function, nd tking the pointwise supremum of n rbitrry collection of semimetrics yields semimetric. Additionlly, if we tke M with the metric induced by the uniform norm, h = sup s,s h(s, s ), then it is complete metric spce. The rich structure of M llows us to pply both the lttice theoretic fixed-point theorem nd the more fmilir Bnch fixed-point theorem, provided we construct n pproprite mp on M. Since bisimultion involves n exct mtching of rewrds nd probbilistic trnsitions, the pproprite metric generliztion should involve metric on rewrds nd metric on probbility mesures. The choice of rewrd metric is obvious: the usul Eucliden distnce. The choice of probbility metric, however, is not so obvious. 2 They re often clled pseudo-metrics in the literture. 3 For convenience we will use the terms metric nd semimetric interchngebly; however, we relly men the ltter.

4 3 Probbility Metrics There re numerous wys of defining notion of distnce between probbility mesures on given spce (Gibbs & Su, 2002). The prticulr probbility semimetric of which we mke use is known s the Kntorovich metric. Given semimetric h M nd probbility mesures P nd Q on S, the induced Kntorovich distnce, T K (h), is defined by T K (h)(p, Q) = sup f (P (f) Q(f)), where the supremum is tken over ll bounded mesurble f : S R stisfying the Lipschitz condition: f(x) f(y) h(x, y) for ll x, y S. We write Lip(h) for the set of ll such functions. In light of the definition of bisimultion, the importnce of using the Kntorovich distnce is mde evident in the following lemm. Lemm 3.1. Let h M. Then T K (h)(p, Q) = 0 P (X) = Q(X), X Σ(Rel(h)). Proof. Fix ɛ > 0 nd let f Lip(h) such tht T K (h)(p, Q) < P (f) Q(f) + ɛ. WLOG f 0. Choose ψ simple pproximtion (the usul one) to f so tht T K (h)(p, Q) < P (ψ) Q(ψ) + 2ɛ. Let ψ(s) = {c 1,..., c k } where the c i re distinct, E i = ψ 1 ({c i }), nd R = Rel(h). Then ech E i is R- closed, for if y R(E i ) then there is some x E i such tht h(x, y) = 0. So f(x) = f(y) nd therefore, ψ(x) = ψ(y). So y E i. So by ssumption P (ψ) Q(ψ) = c i P (E i ) c i Q(E i ) = 0. Thus, T K (h)(p, Q) = 0. Let X Σ(R). Let K X be compct. Define f(x) = inf k K h(x, k). Since lsc function hs minimum on compct set, we my write f(x) = min k K h(x, k). In fct, f is itself lsc (see Theorem B.5 of (Putermn, 1994)). Since f is mesurble, R(K) = f 1 ({0}) Σ(R). Now, since P is tight (s S is complete seprble metric spce), P (X) = sup P (K) where the supremum is tken over ll compct K X. However, K X implies K R(K) R(X) = X. Since R(K) is mesurble, we hve P (X) = sup P (R(K)). Similrly, Q(X) = sup Q(R(K)). Define g n = mx(0, 1 nf). Then g n decreses to the indictor function on R(K). Also, g n /n Lip(h), so by ssumption P (g n /n) = Q(g n /n). Multiplying by n nd tking limits gives P (R(K)) = Q(R(K)) nd we re done. The Kntorovich metric rose in the study of optiml mss trnsporttion (see (Villni, 2002)): Assume we re given pile of snd nd hole, occupying mesurble spces (X, Σ X ) nd (Y, Σ Y ), ech representing copy of (S, Σ). The pile of snd nd the hole obviously hve the sme volume, nd the mss of the pile is ssumed to be normlized to 1. Let P nd Q be mesures on X nd Y respectively, such tht whenever A Σ X nd B Σ Y, P [A] mesures how much snd occupies A nd Q[B] mesures how much snd cn be piled into B. Suppose further tht we hve some mesurble cost function h : X Y R, where h(x, y) tells us how much it costs to trnsfer one unit of mss from point x X to point y Y. Here we consider h M. The gol is to determine pln for trnsferring ll the mss from X to Y while keeping the cost t minimum. Such trnsfer pln is modelled by probbility mesure λ on (X Y, Σ X Σ Y ), where dλ(x, y) mesures how much mss is trnsferred from loction x to y. Of course, for the pln to be vlid we require tht λ[a Y ] = P [A] nd λ[x B] = Q[B] for ll mesurble A nd B. A pln stisfying this condition is sid to hve mrginls P nd Q, nd we denote the collection of ll such plns by Λ(P, Q). We cn now restte the gol formlly s: minimize h(λ) over λ Λ(P, Q) This is ctully n instnce of n infinite liner progrm. Fortuntely, under very generl circumstnces, it hs solution nd dmits dul formultion. Let us first note tht mesures in Λ(P, Q) cn equivlently be chrcterized s those λ stisfying: P (φ) + Q(ψ) = λ(φ + ψ) for ll (φ, ψ) L 1 (P ) L 1 (Q). As consequence of this chrcteriztion we hve the following inequlity: sup (P (f) Q(f)) T K (h)(p, Q) inf h(λ) f λ Λ(P,Q) (1) where f is restricted to the continuous functions in Lip(h). The leftmost nd rightmost terms in inequlity (1) re exmples of infinite liner progrms in dulity. It is highly nontrivil result tht there is no dulity gp in this cse, s result of the Kntorovich- Rubinstein Dulity Theorem with metric cost function (see (Rchev & Rüschendorf, 1998), specificlly theorems 4.15 & 4.28 nd exmple 4.24, or see (Villni, 2002) for more redble ccount of this topic). In the cse of finite stte spce, this dulity leds to strongly polynomil time lgorithm (in terms of the size of the stte spce) for clculting the Kntorovich metric (Orlin, 1988). Thus, one pproch for clculting the Kntorovich metric is to discretize the liner progrm in some mnner nd solve finite liner progrm (see section 5.3 of (Rchev & Rüschendorf, 1998) for compct S). In further restricted settings, e.g. if S is Eucliden nd h is continuous, more direct

5 pproximtion schemes exist (see section 5.4 of (Anderson & Nsh, 1987)). Issues of efficiency side, the Kntorovich distnce is computble. We conclude this section by noting tht if the stte spce metric is chosen to be the discrete metric, which ssigns distnce 1 to ll pirs of unequl points, then the Kntorovich metric grees with the totl vrition metric, defined s d T V (P, Q) = sup X Σ P (X) Q(X). While simple to define, the totl vrition metric gives n overly strong mesure of the numericl differences cross probbilistic trnsitions to ll mesurble sets. Note, for exmple, tht the distnce between two point msses, δ x nd δ y, is lwys 1, unless x = y exctly. Nevertheless, the totl vrition distnce is commonly used in prctice nd cn led to interesting bounds. 4 Bisimultion Metrics Our development of fixed point metrics mirrors closely the definition of bisimultion. In the following c (0, 1) is discount fctor, in the sme vein s the discount fctor γ used in the definition nd estimtion of vlue functions. It determines the extent to which future trnsitions re tken into ccount when trying to distinguish sttes quntittively. In section 2 we mentioned tht M is uniformly bounded set of lsc semimetrics. Here we fix tht upper bound to be the constnt α defined s B 1 c. Theorem 4.1. Let c (0, 1). Define F c : M M by F c (h)(s, s ) = mx A ( r s r s + ct K(h)(P s, P s )) Then F c hs lest fixed point, d c fix, nd Rel(dc fix ) is bisimultion. Proof. It is esy to see tht F c is monotone on M nd so existence of d c fix follows from the Knester- Trski Theorem. It is importnt to note here tht we re implicitly invoking the leftmost equlity in (1) in order to correctly clim tht the mp tking (s, s ) to T K (h)(ps, Ps ) is lsc. By mens of lemm 3.1 we find tht for ny h in M, Rel(F c (h)) = F(Rel(h)). Thus, Rel(d c fix ) = F(Rel(d c fix )) is fixed point nd so is contined in bisimultion. For the other direction, we consider the discrete bisimultion semimetric; note tht it is not immeditely cler tht it is lsc. Cll it I. Let l be the gretest lower bound in M of {αi }. Then Rel(l). Thus, = F( ) F(Rel(l)) = Rel(F c (l)), which implies F c (l) αi. Since F c (l) M, we must hve F c (l) l. Since d c fix is the lest prefixed point of F c, d c fix l αi, so tht Rel(d c fix ). Thus, we hve estblished existence of metric tht ssigns distnce zero to points exctly in the cse when those points re bisimilr. Of course, d c fix is not the only such metric; the discrete bisimultion metric, for exmple, is nother. However, d c fix is the most suitble cndidte bisimultion metric for MDP nlysis. Before we rgue tht this is the cse, let us first note tht d c fix is in fct unique. Proposition 4.2. For ny h 0 M, d c fix (F c ) n (h 0 ) cn 1 c F c (h 0 ) h 0. In prticulr, lim n (F c ) n (h 0 ) = d c fix, nd dc fix is the unique fixed point of F c. Proof. This is simply n ppliction of the Bnch Fixed Point Theorem. Here we use the dul minimiztion form of T K ( ), s given in (1). Note tht for ll h, h M, nd for ll s, s S, F c (h)(s, s ) F c (h )(s, s ) c mx A (T K(h)(P s, P s ) T K(h )(P s, P s )) c mx A (T K(h h + h )(P s, P s ) T K(h )(P s, P s )) c mx A (T K( h h + h )(P s, P s ) T K(h )(P s, P s )) c mx A ( h h + T K (h )(P s, P s ) T K(h )(P s, P s )) c h h Thus, F c (h) F c (h ) c h h, so tht F c is contrction mpping nd hs n unique fixed point d c fix. As n immedite corollry of theorem 4.1 we find tht bisimultion is closed subset of S S, under the given restrictions on r nd P. So the discrete bisimultion metric, αi, is lsc, nd in prticulr, {(F c ) n (αi )} is fmily of lsc semimetrics decresing to d c fix, ech of which hs bisimultion s its kernel. The first iterte cn be expressed in more fmilir form by noting tht T K (I )(P, Q) = sup X Σ( ) P (X) Q(X), which is the totl vrition distnce of P nd Q s defined over the fully compressed stte spce (see ppendix for proof). The dvntge of using d c fix over ny of these itertes is tht d c fix is sensitive to perturbtions in the MDP prmeters. Formlly, d c fix is continuous in r nd P. Proposition 4.3. Suppose (r i, P i ), i = 1, 2, re MDP prmeters, ech stisfying the ssumptions of section 2, nd set B = mx(b 1, B 2 ). Let d 1 nd d 2 be the corresponding bisimultion metrics given by theo-

6 rem 4.1 with discount fctor c. Then d 1 d c mx r1 r2 + 2Bc (1 c) 2 sup d T V (P1,s, P2,s) This result follows from the unwinding of the fixed point definitions of d 1 nd d 2. Proof. Since Lip( d2 d 2 ) Lip(I ), we first obtin the following inequlity:,s T K (d 2 )(P 1,x, P 1,y) T K (d 2 )(P 2,x, P 2,y) sup (P1,x(f) P1,y(f)) (P2,x(f) P2,y(f)) Lip(d 2) d 2 sup (P1,x( f d 2 ) P 2,x( f d 2 )) Lip(I ) (P1,y( f d 2 ) P 2,y( f d 2 )) d 2 ( sup P1,x(g) P2,x(g) Lip(I ) + sup P1,y(g) P2,y(g) ) Lip(I ) d 2 (d T V (P 1,x, P 2,x) + d T V (P 1,y, P 2,y)) Here we re once more using the minimiztion form of T K. d 1 (x, y) d 2 (x, y) mx A ( r 1,x r 1,y + ct K (d 1 )(P 1,x, P 1,y)) mx A ( r 2,x r 2,y + ct K (d 2 )(P 2,x, P 2,y)) mx A ( r 1,x r 1,y r 2,x r 2,y + c(t K (d 1 )(P 1,x, P 1,y) T K (d 2 )(P 2,x, P 2,y))) mx A ( (r 1,x r 1,y) (r 2,x r 2,y) + c(t K (d 1 )(P 1,x, P 1,y) T K (d 2 )(P 1,x, P 1,y)) + c(t K (d 2 )(P 1,x, P 1,y) T K (d 2 )(P 2,x, P 2,y)))) mx A ( r 1,x r 2,x + r 1,y r 2,y + c d 1 d 2 + 2c d 2 sup d T V (P1,s, P2,s))) s mx A (2 r 1 r2 + c d 1 d 2 + 2c d 2 sup d T V (P1,s, P2,s))) s 2 mx A r 1 r2 + c d 1 d 2 B + 2c( 1 c ) sup d T V (P1,s, P2,s))),s Finlly, note tht proposition 4.2 llows us to clculte distnces up to ny prescribe degree of ccurcy using itertion, provided the Kntorovich metrics cn be efficiently nd suitbly clculted themselves. It remins to be seen if such method will be fesible in prctice. 5 Vlue Function Bounds Theorem 5.1. Suppose γ c. Then V is 1-Lipschitz continuous with respect to d c fix, i.e. V (s) V (s ) d c fix(s, s ). Proof. Ech iterte V n is continuous, nd so ech V n (s) V n (s ) belongs to M. The result now follows by induction nd tking limits. 6 Illustrtion In this section we present toy exmple of metric computtion nd metric pproximtion gurntees. Let S = [0, 1] with the usul Borel sigm-lgebr, A = {, b}, r s = 1 s, r b s = s, P s be uniform on S, nd P b s the point mss t s. Clerly, these MDP prmeters stisfy the required ssumptions. Given ny c (0, 1), we clim d c fix(x, y) = x y 1 c. Denote the RHS by h. Note tht T K (h)(px, Py ) = 0 nd T K (h)(px, b Py b ) = sup f Lip(h) f(x) f(y). Tking f 1 (x) = x 1 c nd f 2(x) = 1 f 1 (x) in Lip(h) we find T K (h)(px, b Py b ) = h(x, y). Thus, F c (h)(x, y) = mx( x y + c 0, x y + c h(x, y)) = x y + c h(x, y) = h(x, y). By uniqueness, d c fix = h. Now consider the following pproximtion. Given ɛ > 0, choose n lrge enough so tht 1 n < (1 c)ɛ. Prtition S s B k = [ k n, k+1 n ), B n 1 = [ n 1 n, 1], for k = 0, 1, 2,..., n 2. Note tht the dimeter of ech B k with respect to d c fix is 1 n(1 c) < ɛ. The n prtitions will be the sttes of finite MDP pproximnt. We obtin the rest of the prmeters by verging over the sttes in prtition. Thus, rb k = 1 2k+1 2n, rb b k = 2k+1 2n, P B k,b l = 1 n, nd P B b k,b l = δ Bk,B l. Assume γ is given nd choose c = γ. Note tht for ll x, y B k, V (x) V (y) dim d c fix B k ɛ. Thus, we would expect tht by verging, nd solving the finite MDP, V (B k ) should differ by t most ɛ from V (x), for ny x B k. In fct, in this cse the vlue functions of the originl MDP nd of the finite pproximnt cn be computed directly nd we cn verify this.

7 For x S, { B k, 1 x + γ V 2(1 γ) if 0 x < 1 2 (x) = x 1 γ if 1 2 { x 1 1 2k+1 V 2n (B k ) = + γ 2(1 γ) if 0 k < n 1 2 2k+1 2n 1 γ if n 1 2 k n 1 Thus, for x B k, V (x) V (B k ) 1 2k+1 1 γ x 2n dim d c fix B k ɛ. 7 Conclusion In this pper we hve constructed metrics for MDPs with continuous stte spces. Ech metric hs bisimultion s its kernel nd is continuous in the MDP prmeters. Most importntly, ech metric bounds the optiml vlue of sttes continuously. Hence, if one ws to ggregte sttes, this metric llows gurntee on the error introduced by this pproximtion. In contrst to previous situtions, in this theoreticl development the most importnt fctor tht we hve to tke into considertion ws the wy in which the rewrds vry cross the stte spce. Wht cn be sid in the cse of generl bounded mesurble, yet not necessrily continuous, rewrd function? In order to generlize our results, we need to estblish the mesurbility of the mp tking pir of sttes to the Kntorovich distnce, nd to generlize lemm 3.1. We re currently working on this development. In the mentime, if the rewrd structure does not stisfy our ssumption, we cn still consider the best lsc pproximtions to rs rs in M. Tht is, we cn replce rs rs by R 1(s, s ) = inf M { rs rs }, nd R2(s, s ) = sup M { rs rs } nd obtin two fixed point semimetrics d c 1 nd d c 2, respectively. Then it is not hrd to modify theorem 4.1 to show tht Rel(d c 2) Rel(d c 1). The ide is tht we re sndwiching bisimultion by nerby closed equivlence reltions. The theoreticl foundtion we estblished cn be used, potentilly in two different wys. The first ide is to use the distnce metric in the process of stte ggregtion, in order to provide finite pproximnt for continuous stte MDP. However, even though our distnce metrics re computble, the computtion methods tht we hve investigted so fr re not stisfctory. Discretizing the Kntorovich liner progrm my result in dded complexity when one considers tht the direct solution might be simple. On the other hnd, more direct methods of clculting the distnce re not currently known in generl. The second ide, which holds lot of promise, is to use our metric s tool for the theoreticl nlysis of existing pproximtion schemes. There re mny heuristic methods for providing vrible resolution or multi-resolution pproximtions to MDPs with continuous stte spces. Using our metrics, the error bounds of these heuristics cn be nlyzed. A second importnt ppliction is in the nlysis of pproximtion schemes which strt with corse pproximnt nd grdully increse the resolution. The distnce metrics cn provide tools for proving tht such schemes converge to correct vlue estimtes in the limit. We re currently pursuing reserch in this direction. Acknowledgments This work hs been supported in prt by funding from NSERC nd CFI. Appendix Lemm 7.1. Suppose C is closed equivlence reltion on S. Then T K (I C )(P, Q) = sup P (X) Q(X). X Σ(C) Proof. For every X Σ(C), the indictor function on X belongs to Lip(I C ). Thus, the RHS is t most the LHS. For the other inequlity, fix positive ɛ nd tke f : S [0, 1] nd ψ = n i=1 c i I Ei s in the proof of lemm 3.1. Let J = {i P (E i ) Q(E i )}. Then T K (I C )(P, Q) 2ɛ P (ψ) (Qψ) = c i (P (E i ) Q(E i )) J c i (P (E i ) Q(E i )) (mx c i ) (P (E i ) Q(E i )) J J 1 (P ( J E i ) Q( J E i )) since J E i belongs to Σ(C). References sup P (X) Q(X) X Σ(C) Anderson, E.J., & Nsh, P. (1987). Liner Progrmming in Infinite-Dimensionl Spces John Wiley nd Sons, Ltd. Bernrdo, M., & Brvetti, M. (2003). Performnce mesure sensitive congruences for Mrkovin process lgebrs. Theoreticl Computer Science, 290, R. Blute, J. Deshrnis, A. Edlt, nd P. Pnngden. Bisimultion for lbelled Mrkov processes. In Proceedings of the Twelfth IEEE Symposium On Logic In Computer Science, Wrsw, Polnd., , 1997.

8 Boutilier, C., Den, T., & Hnks, S. (1999). Decisiontheoretic plnning: Structurl ssumptions nd computtionl leverge. Journl of Artificil Intelligence Reserch, 11, J. Deshrnis, A. Edlt, nd P. Pnngden. Bisimultion for lbeled Mrkov processes. Informtion nd Computtion, vol 179, pges , Deshrnis, J., Gupt, V., Jgdeesn, R., & Pnngden, P. (1999). Metrics for lbeled Mrkov systems. Interntionl Conference on Concurrency Theory (pp ). Deshrnis, J., Gupt, V., Jgdeesn, R., & Pnngden, P. (2002). The metric nlogue of wek bisimultion for probbilistic processes. Logic in Computer Science (pp ). IEEE Computer Society. Ferns, N., Pnngden, P., & Precup, D. (2004) Metrics for finite Mrkov decision processes Proceedings of the 20th conference on Uncertinty in rtificil intelligence (pp ). Gibbs, A. L., & Su, F. E. (2002). On choosing nd bounding probbility metrics. Interntionl Sttisticl Review, 70, (pp ). Givn, R., Den, T., & Greig, M. (2003). Equivlence notions nd model minimiztion in mrkov decision processes. Artificil Intelligence, 147, Lrsen, K., & Skou, A. (1991). Bisimultion through probbilistic testing. Informtion nd Computtion, 94, Milner, R. (1980). A clculus of communicting systems. Lecture Notes in Computer Science Vol. 92. Springer-Verlg. Orlin, J. (1988). A fster strongly polynomil minimum cost flow lgorithm. Proceedings of the Twentieth nnul ACM symposium on Theory of Computing (pp ). ACM Press. Prk, D. (1981). Concurrency nd utomt on infinite sequences. Proceedings of the 5th GI-Conference on Theoreticl Computer Science (pp ). Springer-Verlg. Putermn, M. L. (1994). Mrkov decision processes: Discrete stochstic dynmic progrmming. John Wiley & Sons, Inc. Rchev, S. T., & Rüschendorf L. (1998). Mss Trnsporttion Problems, Vol. I: Theory. Springer, Berlin Heidelberg New York. vn Breugel, F., & Worrell, J. (2001). Towrds Quntittive Verifiction of Probbilistic Trnsition Systems. Proceedings of the 28th Interntionl Colloquium on Automt, Lnguges, nd Progrmming (ICALP), (pp ) Springer-Verlg. vn Breugel, F., & Worrell, J. (2001). An lgorithm for quntittive verifiction of probbilistic trnsition systems. Proceedings of the 12th Interntionl Conference on Concurrency Theory (pp ). Springer-Verlg. Villni, C. (2002). Topics in Mss Trnsporttion. [ seminr/rticles/vilnotes.ps](28/07/03) Winskel, G. (1993). The forml semntics of progrmming lnguges. Foundtions of Computing. The MIT Press.

Metrics for Finite Markov Decision Processes

Metrics for Finite Markov Decision Processes Metrics for Finite Mrkov Decision Processes Norm Ferns chool of Computer cience McGill University Montrél, Cnd, H3 27 nferns@cs.mcgill.c Prksh Pnngden chool of Computer cience McGill University Montrél,

More information

The Regulated and Riemann Integrals

The Regulated and Riemann Integrals Chpter 1 The Regulted nd Riemnn Integrls 1.1 Introduction We will consider severl different pproches to defining the definite integrl f(x) dx of function f(x). These definitions will ll ssign the sme vlue

More information

Advanced Calculus: MATH 410 Notes on Integrals and Integrability Professor David Levermore 17 October 2004

Advanced Calculus: MATH 410 Notes on Integrals and Integrability Professor David Levermore 17 October 2004 Advnced Clculus: MATH 410 Notes on Integrls nd Integrbility Professor Dvid Levermore 17 October 2004 1. Definite Integrls In this section we revisit the definite integrl tht you were introduced to when

More information

Lecture 1. Functional series. Pointwise and uniform convergence.

Lecture 1. Functional series. Pointwise and uniform convergence. 1 Introduction. Lecture 1. Functionl series. Pointwise nd uniform convergence. In this course we study mongst other things Fourier series. The Fourier series for periodic function f(x) with period 2π is

More information

Chapter 5 : Continuous Random Variables

Chapter 5 : Continuous Random Variables STAT/MATH 395 A - PROBABILITY II UW Winter Qurter 216 Néhémy Lim Chpter 5 : Continuous Rndom Vribles Nottions. N {, 1, 2,...}, set of nturl numbers (i.e. ll nonnegtive integers); N {1, 2,...}, set of ll

More information

Review of Calculus, cont d

Review of Calculus, cont d Jim Lmbers MAT 460 Fll Semester 2009-10 Lecture 3 Notes These notes correspond to Section 1.1 in the text. Review of Clculus, cont d Riemnn Sums nd the Definite Integrl There re mny cses in which some

More information

W. We shall do so one by one, starting with I 1, and we shall do it greedily, trying

W. We shall do so one by one, starting with I 1, and we shall do it greedily, trying Vitli covers 1 Definition. A Vitli cover of set E R is set V of closed intervls with positive length so tht, for every δ > 0 nd every x E, there is some I V with λ(i ) < δ nd x I. 2 Lemm (Vitli covering)

More information

Math 1B, lecture 4: Error bounds for numerical methods

Math 1B, lecture 4: Error bounds for numerical methods Mth B, lecture 4: Error bounds for numericl methods Nthn Pflueger 4 September 0 Introduction The five numericl methods descried in the previous lecture ll operte by the sme principle: they pproximte the

More information

Riemann is the Mann! (But Lebesgue may besgue to differ.)

Riemann is the Mann! (But Lebesgue may besgue to differ.) Riemnn is the Mnn! (But Lebesgue my besgue to differ.) Leo Livshits My 2, 2008 1 For finite intervls in R We hve seen in clss tht every continuous function f : [, b] R hs the property tht for every ɛ >

More information

Continuous Random Variables

Continuous Random Variables STAT/MATH 395 A - PROBABILITY II UW Winter Qurter 217 Néhémy Lim Continuous Rndom Vribles Nottion. The indictor function of set S is rel-vlued function defined by : { 1 if x S 1 S (x) if x S Suppose tht

More information

1 Online Learning and Regret Minimization

1 Online Learning and Regret Minimization 2.997 Decision-Mking in Lrge-Scle Systems My 10 MIT, Spring 2004 Hndout #29 Lecture Note 24 1 Online Lerning nd Regret Minimiztion In this lecture, we consider the problem of sequentil decision mking in

More information

MATH34032: Green s Functions, Integral Equations and the Calculus of Variations 1

MATH34032: Green s Functions, Integral Equations and the Calculus of Variations 1 MATH34032: Green s Functions, Integrl Equtions nd the Clculus of Vritions 1 Section 1 Function spces nd opertors Here we gives some brief detils nd definitions, prticulrly relting to opertors. For further

More information

Advanced Calculus: MATH 410 Uniform Convergence of Functions Professor David Levermore 11 December 2015

Advanced Calculus: MATH 410 Uniform Convergence of Functions Professor David Levermore 11 December 2015 Advnced Clculus: MATH 410 Uniform Convergence of Functions Professor Dvid Levermore 11 December 2015 12. Sequences of Functions We now explore two notions of wht it mens for sequence of functions {f n

More information

MAA 4212 Improper Integrals

MAA 4212 Improper Integrals Notes by Dvid Groisser, Copyright c 1995; revised 2002, 2009, 2014 MAA 4212 Improper Integrls The Riemnn integrl, while perfectly well-defined, is too restrictive for mny purposes; there re functions which

More information

THE EXISTENCE-UNIQUENESS THEOREM FOR FIRST-ORDER DIFFERENTIAL EQUATIONS.

THE EXISTENCE-UNIQUENESS THEOREM FOR FIRST-ORDER DIFFERENTIAL EQUATIONS. THE EXISTENCE-UNIQUENESS THEOREM FOR FIRST-ORDER DIFFERENTIAL EQUATIONS RADON ROSBOROUGH https://intuitiveexplntionscom/picrd-lindelof-theorem/ This document is proof of the existence-uniqueness theorem

More information

Math 61CM - Solutions to homework 9

Math 61CM - Solutions to homework 9 Mth 61CM - Solutions to homework 9 Cédric De Groote November 30 th, 2018 Problem 1: Recll tht the left limit of function f t point c is defined s follows: lim f(x) = l x c if for ny > 0 there exists δ

More information

Lecture 1: Introduction to integration theory and bounded variation

Lecture 1: Introduction to integration theory and bounded variation Lecture 1: Introduction to integrtion theory nd bounded vrition Wht is this course bout? Integrtion theory. The first question you might hve is why there is nything you need to lern bout integrtion. You

More information

NOTES AND PROBLEMS: INTEGRATION THEORY

NOTES AND PROBLEMS: INTEGRATION THEORY NOTES AND PROBLEMS: INTEGRATION THEORY SAMEER CHAVAN Abstrct. These re the lecture notes prepred for prticipnts of AFS-I to be conducted t Kumun University, Almor from 1st to 27th December, 2014. Contents

More information

Reinforcement learning II

Reinforcement learning II CS 1675 Introduction to Mchine Lerning Lecture 26 Reinforcement lerning II Milos Huskrecht milos@cs.pitt.edu 5329 Sennott Squre Reinforcement lerning Bsics: Input x Lerner Output Reinforcement r Critic

More information

The final exam will take place on Friday May 11th from 8am 11am in Evans room 60.

The final exam will take place on Friday May 11th from 8am 11am in Evans room 60. Mth 104: finl informtion The finl exm will tke plce on Fridy My 11th from 8m 11m in Evns room 60. The exm will cover ll prts of the course with equl weighting. It will cover Chpters 1 5, 7 15, 17 21, 23

More information

Properties of the Riemann Integral

Properties of the Riemann Integral Properties of the Riemnn Integrl Jmes K. Peterson Deprtment of Biologicl Sciences nd Deprtment of Mthemticl Sciences Clemson University Februry 15, 2018 Outline 1 Some Infimum nd Supremum Properties 2

More information

1 i n x i x i 1. Note that kqk kp k. In addition, if P and Q are partition of [a, b], P Q is finer than both P and Q.

1 i n x i x i 1. Note that kqk kp k. In addition, if P and Q are partition of [a, b], P Q is finer than both P and Q. Chpter 6 Integrtion In this chpter we define the integrl. Intuitively, it should be the re under curve. Not surprisingly, fter mny exmples, counter exmples, exceptions, generliztions, the concept of the

More information

CMDA 4604: Intermediate Topics in Mathematical Modeling Lecture 19: Interpolation and Quadrature

CMDA 4604: Intermediate Topics in Mathematical Modeling Lecture 19: Interpolation and Quadrature CMDA 4604: Intermedite Topics in Mthemticl Modeling Lecture 19: Interpoltion nd Qudrture In this lecture we mke brief diversion into the res of interpoltion nd qudrture. Given function f C[, b], we sy

More information

The Banach algebra of functions of bounded variation and the pointwise Helly selection theorem

The Banach algebra of functions of bounded variation and the pointwise Helly selection theorem The Bnch lgebr of functions of bounded vrition nd the pointwise Helly selection theorem Jordn Bell jordn.bell@gmil.com Deprtment of Mthemtics, University of Toronto Jnury, 015 1 BV [, b] Let < b. For f

More information

Review of basic calculus

Review of basic calculus Review of bsic clculus This brief review reclls some of the most importnt concepts, definitions, nd theorems from bsic clculus. It is not intended to tech bsic clculus from scrtch. If ny of the items below

More information

The First Fundamental Theorem of Calculus. If f(x) is continuous on [a, b] and F (x) is any antiderivative. f(x) dx = F (b) F (a).

The First Fundamental Theorem of Calculus. If f(x) is continuous on [a, b] and F (x) is any antiderivative. f(x) dx = F (b) F (a). The Fundmentl Theorems of Clculus Mth 4, Section 0, Spring 009 We now know enough bout definite integrls to give precise formultions of the Fundmentl Theorems of Clculus. We will lso look t some bsic emples

More information

Strong Bisimulation. Overview. References. Actions Labeled transition system Transition semantics Simulation Bisimulation

Strong Bisimulation. Overview. References. Actions Labeled transition system Transition semantics Simulation Bisimulation Strong Bisimultion Overview Actions Lbeled trnsition system Trnsition semntics Simultion Bisimultion References Robin Milner, Communiction nd Concurrency Robin Milner, Communicting nd Mobil Systems 32

More information

Lecture notes. Fundamental inequalities: techniques and applications

Lecture notes. Fundamental inequalities: techniques and applications Lecture notes Fundmentl inequlities: techniques nd pplictions Mnh Hong Duong Mthemtics Institute, University of Wrwick Emil: m.h.duong@wrwick.c.uk Februry 8, 207 2 Abstrct Inequlities re ubiquitous in

More information

Analytical Methods Exam: Preparatory Exercises

Analytical Methods Exam: Preparatory Exercises Anlyticl Methods Exm: Preprtory Exercises Question. Wht does it men tht (X, F, µ) is mesure spce? Show tht µ is monotone, tht is: if E F re mesurble sets then µ(e) µ(f). Question. Discuss if ech of the

More information

7.2 The Definite Integral

7.2 The Definite Integral 7.2 The Definite Integrl the definite integrl In the previous section, it ws found tht if function f is continuous nd nonnegtive, then the re under the grph of f on [, b] is given by F (b) F (), where

More information

Reinforcement Learning

Reinforcement Learning Reinforcement Lerning Tom Mitchell, Mchine Lerning, chpter 13 Outline Introduction Comprison with inductive lerning Mrkov Decision Processes: the model Optiml policy: The tsk Q Lerning: Q function Algorithm

More information

Lecture 19: Continuous Least Squares Approximation

Lecture 19: Continuous Least Squares Approximation Lecture 19: Continuous Lest Squres Approximtion 33 Continuous lest squres pproximtion We begn 31 with the problem of pproximting some f C[, b] with polynomil p P n t the discrete points x, x 1,, x m for

More information

UNIFORM CONVERGENCE. Contents 1. Uniform Convergence 1 2. Properties of uniform convergence 3

UNIFORM CONVERGENCE. Contents 1. Uniform Convergence 1 2. Properties of uniform convergence 3 UNIFORM CONVERGENCE Contents 1. Uniform Convergence 1 2. Properties of uniform convergence 3 Suppose f n : Ω R or f n : Ω C is sequence of rel or complex functions, nd f n f s n in some sense. Furthermore,

More information

A HELLY THEOREM FOR FUNCTIONS WITH VALUES IN METRIC SPACES. 1. Introduction

A HELLY THEOREM FOR FUNCTIONS WITH VALUES IN METRIC SPACES. 1. Introduction Ttr Mt. Mth. Publ. 44 (29), 159 168 DOI: 1.2478/v1127-9-56-z t m Mthemticl Publictions A HELLY THEOREM FOR FUNCTIONS WITH VALUES IN METRIC SPACES Miloslv Duchoň Peter Mličký ABSTRACT. We present Helly

More information

g i fφdx dx = x i i=1 is a Hilbert space. We shall, henceforth, abuse notation and write g i f(x) = f

g i fφdx dx = x i i=1 is a Hilbert space. We shall, henceforth, abuse notation and write g i f(x) = f 1. Appliction of functionl nlysis to PEs 1.1. Introduction. In this section we give little introduction to prtil differentil equtions. In prticulr we consider the problem u(x) = f(x) x, u(x) = x (1) where

More information

Review of Gaussian Quadrature method

Review of Gaussian Quadrature method Review of Gussin Qudrture method Nsser M. Asi Spring 006 compiled on Sundy Decemer 1, 017 t 09:1 PM 1 The prolem To find numericl vlue for the integrl of rel vlued function of rel vrile over specific rnge

More information

p-adic Egyptian Fractions

p-adic Egyptian Fractions p-adic Egyptin Frctions Contents 1 Introduction 1 2 Trditionl Egyptin Frctions nd Greedy Algorithm 2 3 Set-up 3 4 p-greedy Algorithm 5 5 p-egyptin Trditionl 10 6 Conclusion 1 Introduction An Egyptin frction

More information

2D1431 Machine Learning Lab 3: Reinforcement Learning

2D1431 Machine Learning Lab 3: Reinforcement Learning 2D1431 Mchine Lerning Lb 3: Reinforcement Lerning Frnk Hoffmnn modified by Örjn Ekeberg December 7, 2004 1 Introduction In this lb you will lern bout dynmic progrmming nd reinforcement lerning. It is ssumed

More information

1 The Lagrange interpolation formula

1 The Lagrange interpolation formula Notes on Qudrture 1 The Lgrnge interpoltion formul We briefly recll the Lgrnge interpoltion formul. The strting point is collection of N + 1 rel points (x 0, y 0 ), (x 1, y 1 ),..., (x N, y N ), with x

More information

Notes on length and conformal metrics

Notes on length and conformal metrics Notes on length nd conforml metrics We recll how to mesure the Eucliden distnce of n rc in the plne. Let α : [, b] R 2 be smooth (C ) rc. Tht is α(t) (x(t), y(t)) where x(t) nd y(t) re smooth rel vlued

More information

Math 8 Winter 2015 Applications of Integration

Math 8 Winter 2015 Applications of Integration Mth 8 Winter 205 Applictions of Integrtion Here re few importnt pplictions of integrtion. The pplictions you my see on n exm in this course include only the Net Chnge Theorem (which is relly just the Fundmentl

More information

Math 554 Integration

Math 554 Integration Mth 554 Integrtion Hndout #9 4/12/96 Defn. A collection of n + 1 distinct points of the intervl [, b] P := {x 0 = < x 1 < < x i 1 < x i < < b =: x n } is clled prtition of the intervl. In this cse, we

More information

Recitation 3: More Applications of the Derivative

Recitation 3: More Applications of the Derivative Mth 1c TA: Pdric Brtlett Recittion 3: More Applictions of the Derivtive Week 3 Cltech 2012 1 Rndom Question Question 1 A grph consists of the following: A set V of vertices. A set E of edges where ech

More information

Reversals of Signal-Posterior Monotonicity for Any Bounded Prior

Reversals of Signal-Posterior Monotonicity for Any Bounded Prior Reversls of Signl-Posterior Monotonicity for Any Bounded Prior Christopher P. Chmbers Pul J. Hely Abstrct Pul Milgrom (The Bell Journl of Economics, 12(2): 380 391) showed tht if the strict monotone likelihood

More information

Entropy and Ergodic Theory Notes 10: Large Deviations I

Entropy and Ergodic Theory Notes 10: Large Deviations I Entropy nd Ergodic Theory Notes 10: Lrge Devitions I 1 A chnge of convention This is our first lecture on pplictions of entropy in probbility theory. In probbility theory, the convention is tht ll logrithms

More information

FUNDAMENTALS OF REAL ANALYSIS by. III.1. Measurable functions. f 1 (

FUNDAMENTALS OF REAL ANALYSIS by. III.1. Measurable functions. f 1 ( FUNDAMNTALS OF RAL ANALYSIS by Doğn Çömez III. MASURABL FUNCTIONS AND LBSGU INTGRAL III.. Mesurble functions Hving the Lebesgue mesure define, in this chpter, we will identify the collection of functions

More information

STUDY GUIDE FOR BASIC EXAM

STUDY GUIDE FOR BASIC EXAM STUDY GUIDE FOR BASIC EXAM BRYON ARAGAM This is prtil list of theorems tht frequently show up on the bsic exm. In mny cses, you my be sked to directly prove one of these theorems or these vrints. There

More information

II. Integration and Cauchy s Theorem

II. Integration and Cauchy s Theorem MTH6111 Complex Anlysis 2009-10 Lecture Notes c Shun Bullett QMUL 2009 II. Integrtion nd Cuchy s Theorem 1. Pths nd integrtion Wrning Different uthors hve different definitions for terms like pth nd curve.

More information

CHAPTER 4 MULTIPLE INTEGRALS

CHAPTER 4 MULTIPLE INTEGRALS CHAPTE 4 MULTIPLE INTEGAL The objects of this chpter re five-fold. They re: (1 Discuss when sclr-vlued functions f cn be integrted over closed rectngulr boxes in n ; simply put, f is integrble over iff

More information

A PROOF OF THE FUNDAMENTAL THEOREM OF CALCULUS USING HAUSDORFF MEASURES

A PROOF OF THE FUNDAMENTAL THEOREM OF CALCULUS USING HAUSDORFF MEASURES INROADS Rel Anlysis Exchnge Vol. 26(1), 2000/2001, pp. 381 390 Constntin Volintiru, Deprtment of Mthemtics, University of Buchrest, Buchrest, Romni. e-mil: cosv@mt.cs.unibuc.ro A PROOF OF THE FUNDAMENTAL

More information

Line and Surface Integrals: An Intuitive Understanding

Line and Surface Integrals: An Intuitive Understanding Line nd Surfce Integrls: An Intuitive Understnding Joseph Breen Introduction Multivrible clculus is ll bout bstrcting the ides of differentition nd integrtion from the fmilir single vrible cse to tht of

More information

Numerical integration

Numerical integration 2 Numericl integrtion This is pge i Printer: Opque this 2. Introduction Numericl integrtion is problem tht is prt of mny problems in the economics nd econometrics literture. The orgniztion of this chpter

More information

Theoretical foundations of Gaussian quadrature

Theoretical foundations of Gaussian quadrature Theoreticl foundtions of Gussin qudrture 1 Inner product vector spce Definition 1. A vector spce (or liner spce) is set V = {u, v, w,...} in which the following two opertions re defined: (A) Addition of

More information

CS667 Lecture 6: Monte Carlo Integration 02/10/05

CS667 Lecture 6: Monte Carlo Integration 02/10/05 CS667 Lecture 6: Monte Crlo Integrtion 02/10/05 Venkt Krishnrj Lecturer: Steve Mrschner 1 Ide The min ide of Monte Crlo Integrtion is tht we cn estimte the vlue of n integrl by looking t lrge number of

More information

Math 360: A primitive integral and elementary functions

Math 360: A primitive integral and elementary functions Mth 360: A primitive integrl nd elementry functions D. DeTurck University of Pennsylvni October 16, 2017 D. DeTurck Mth 360 001 2017C: Integrl/functions 1 / 32 Setup for the integrl prtitions Definition:

More information

State space systems analysis (continued) Stability. A. Definitions A system is said to be Asymptotically Stable (AS) when it satisfies

State space systems analysis (continued) Stability. A. Definitions A system is said to be Asymptotically Stable (AS) when it satisfies Stte spce systems nlysis (continued) Stbility A. Definitions A system is sid to be Asymptoticlly Stble (AS) when it stisfies ut () = 0, t > 0 lim xt () 0. t A system is AS if nd only if the impulse response

More information

SYDE 112, LECTURES 3 & 4: The Fundamental Theorem of Calculus

SYDE 112, LECTURES 3 & 4: The Fundamental Theorem of Calculus SYDE 112, LECTURES & 4: The Fundmentl Theorem of Clculus So fr we hve introduced two new concepts in this course: ntidifferentition nd Riemnn sums. It turns out tht these quntities re relted, but it is

More information

Lecture 14: Quadrature

Lecture 14: Quadrature Lecture 14: Qudrture This lecture is concerned with the evlution of integrls fx)dx 1) over finite intervl [, b] The integrnd fx) is ssumed to be rel-vlues nd smooth The pproximtion of n integrl by numericl

More information

Numerical Analysis: Trapezoidal and Simpson s Rule

Numerical Analysis: Trapezoidal and Simpson s Rule nd Simpson s Mthemticl question we re interested in numericlly nswering How to we evlute I = f (x) dx? Clculus tells us tht if F(x) is the ntiderivtive of function f (x) on the intervl [, b], then I =

More information

Numerical Integration

Numerical Integration Chpter 1 Numericl Integrtion Numericl differentition methods compute pproximtions to the derivtive of function from known vlues of the function. Numericl integrtion uses the sme informtion to compute numericl

More information

Numerical Integration

Numerical Integration Chpter 5 Numericl Integrtion Numericl integrtion is the study of how the numericl vlue of n integrl cn be found. Methods of function pproximtion discussed in Chpter??, i.e., function pproximtion vi the

More information

Duality # Second iteration for HW problem. Recall our LP example problem we have been working on, in equality form, is given below.

Duality # Second iteration for HW problem. Recall our LP example problem we have been working on, in equality form, is given below. Dulity #. Second itertion for HW problem Recll our LP emple problem we hve been working on, in equlity form, is given below.,,,, 8 m F which, when written in slightly different form, is 8 F Recll tht we

More information

Variational Techniques for Sturm-Liouville Eigenvalue Problems

Variational Techniques for Sturm-Liouville Eigenvalue Problems Vritionl Techniques for Sturm-Liouville Eigenvlue Problems Vlerie Cormni Deprtment of Mthemtics nd Sttistics University of Nebrsk, Lincoln Lincoln, NE 68588 Emil: vcormni@mth.unl.edu Rolf Ryhm Deprtment

More information

Presentation Problems 5

Presentation Problems 5 Presenttion Problems 5 21-355 A For these problems, ssume ll sets re subsets of R unless otherwise specified. 1. Let P nd Q be prtitions of [, b] such tht P Q. Then U(f, P ) U(f, Q) nd L(f, P ) L(f, Q).

More information

We partition C into n small arcs by forming a partition of [a, b] by picking s i as follows: a = s 0 < s 1 < < s n = b.

We partition C into n small arcs by forming a partition of [a, b] by picking s i as follows: a = s 0 < s 1 < < s n = b. Mth 255 - Vector lculus II Notes 4.2 Pth nd Line Integrls We begin with discussion of pth integrls (the book clls them sclr line integrls). We will do this for function of two vribles, but these ides cn

More information

Czechoslovak Mathematical Journal, 55 (130) (2005), , Abbotsford. 1. Introduction

Czechoslovak Mathematical Journal, 55 (130) (2005), , Abbotsford. 1. Introduction Czechoslovk Mthemticl Journl, 55 (130) (2005), 933 940 ESTIMATES OF THE REMAINDER IN TAYLOR S THEOREM USING THE HENSTOCK-KURZWEIL INTEGRAL, Abbotsford (Received Jnury 22, 2003) Abstrct. When rel-vlued

More information

Bernoulli Numbers Jeff Morton

Bernoulli Numbers Jeff Morton Bernoulli Numbers Jeff Morton. We re interested in the opertor e t k d k t k, which is to sy k tk. Applying this to some function f E to get e t f d k k tk d k f f + d k k tk dk f, we note tht since f

More information

ODE: Existence and Uniqueness of a Solution

ODE: Existence and Uniqueness of a Solution Mth 22 Fll 213 Jerry Kzdn ODE: Existence nd Uniqueness of Solution The Fundmentl Theorem of Clculus tells us how to solve the ordinry differentil eqution (ODE) du = f(t) dt with initil condition u() =

More information

1.9 C 2 inner variations

1.9 C 2 inner variations 46 CHAPTER 1. INDIRECT METHODS 1.9 C 2 inner vritions So fr, we hve restricted ttention to liner vritions. These re vritions of the form vx; ǫ = ux + ǫφx where φ is in some liner perturbtion clss P, for

More information

MAT612-REAL ANALYSIS RIEMANN STIELTJES INTEGRAL

MAT612-REAL ANALYSIS RIEMANN STIELTJES INTEGRAL MAT612-REAL ANALYSIS RIEMANN STIELTJES INTEGRAL DR. RITU AGARWAL MALVIYA NATIONAL INSTITUTE OF TECHNOLOGY, JAIPUR, INDIA-302017 Tble of Contents Contents Tble of Contents 1 1. Introduction 1 2. Prtition

More information

Euler, Ioachimescu and the trapezium rule. G.J.O. Jameson (Math. Gazette 96 (2012), )

Euler, Ioachimescu and the trapezium rule. G.J.O. Jameson (Math. Gazette 96 (2012), ) Euler, Iochimescu nd the trpezium rule G.J.O. Jmeson (Mth. Gzette 96 (0), 36 4) The following results were estblished in recent Gzette rticle [, Theorems, 3, 4]. Given > 0 nd 0 < s

More information

ACM 105: Applied Real and Functional Analysis. Solutions to Homework # 2.

ACM 105: Applied Real and Functional Analysis. Solutions to Homework # 2. ACM 05: Applied Rel nd Functionl Anlysis. Solutions to Homework # 2. Andy Greenberg, Alexei Novikov Problem. Riemnn-Lebesgue Theorem. Theorem (G.F.B. Riemnn, H.L. Lebesgue). If f is n integrble function

More information

arxiv:math/ v2 [math.ho] 16 Dec 2003

arxiv:math/ v2 [math.ho] 16 Dec 2003 rxiv:mth/0312293v2 [mth.ho] 16 Dec 2003 Clssicl Lebesgue Integrtion Theorems for the Riemnn Integrl Josh Isrlowitz 244 Ridge Rd. Rutherford, NJ 07070 jbi2@njit.edu Februry 1, 2008 Abstrct In this pper,

More information

Quadratic Forms. Quadratic Forms

Quadratic Forms. Quadratic Forms Qudrtic Forms Recll the Simon & Blume excerpt from n erlier lecture which sid tht the min tsk of clculus is to pproximte nonliner functions with liner functions. It s ctully more ccurte to sy tht we pproximte

More information

LECTURE. INTEGRATION AND ANTIDERIVATIVE.

LECTURE. INTEGRATION AND ANTIDERIVATIVE. ANALYSIS FOR HIGH SCHOOL TEACHERS LECTURE. INTEGRATION AND ANTIDERIVATIVE. ROTHSCHILD CAESARIA COURSE, 2015/6 1. Integrtion Historiclly, it ws the problem of computing res nd volumes, tht triggered development

More information

Jim Lambers MAT 169 Fall Semester Lecture 4 Notes

Jim Lambers MAT 169 Fall Semester Lecture 4 Notes Jim Lmbers MAT 169 Fll Semester 2009-10 Lecture 4 Notes These notes correspond to Section 8.2 in the text. Series Wht is Series? An infinte series, usully referred to simply s series, is n sum of ll of

More information

A Convergence Theorem for the Improper Riemann Integral of Banach Space-valued Functions

A Convergence Theorem for the Improper Riemann Integral of Banach Space-valued Functions Interntionl Journl of Mthemticl Anlysis Vol. 8, 2014, no. 50, 2451-2460 HIKARI Ltd, www.m-hikri.com http://dx.doi.org/10.12988/ijm.2014.49294 A Convergence Theorem for the Improper Riemnn Integrl of Bnch

More information

Introduction to Some Convergence theorems

Introduction to Some Convergence theorems Lecture Introduction to Some Convergence theorems Fridy 4, 005 Lecturer: Nti Linil Notes: Mukund Nrsimhn nd Chris Ré. Recp Recll tht for f : T C, we hd defined ˆf(r) = π T f(t)e irt dt nd we were trying

More information

Discrete Mathematics and Probability Theory Spring 2013 Anant Sahai Lecture 17

Discrete Mathematics and Probability Theory Spring 2013 Anant Sahai Lecture 17 EECS 70 Discrete Mthemtics nd Proility Theory Spring 2013 Annt Shi Lecture 17 I.I.D. Rndom Vriles Estimting the is of coin Question: We wnt to estimte the proportion p of Democrts in the US popultion,

More information

Main topics for the First Midterm

Main topics for the First Midterm Min topics for the First Midterm The Midterm will cover Section 1.8, Chpters 2-3, Sections 4.1-4.8, nd Sections 5.1-5.3 (essentilly ll of the mteril covered in clss). Be sure to know the results of the

More information

Math 270A: Numerical Linear Algebra

Math 270A: Numerical Linear Algebra Mth 70A: Numericl Liner Algebr Instructor: Michel Holst Fll Qurter 014 Homework Assignment #3 Due Give to TA t lest few dys before finl if you wnt feedbck. Exercise 3.1. (The Bsic Liner Method for Liner

More information

Goals: Determine how to calculate the area described by a function. Define the definite integral. Explore the relationship between the definite

Goals: Determine how to calculate the area described by a function. Define the definite integral. Explore the relationship between the definite Unit #8 : The Integrl Gols: Determine how to clculte the re described by function. Define the definite integrl. Eplore the reltionship between the definite integrl nd re. Eplore wys to estimte the definite

More information

Lecture 3 ( ) (translated and slightly adapted from lecture notes by Martin Klazar)

Lecture 3 ( ) (translated and slightly adapted from lecture notes by Martin Klazar) Lecture 3 (5.3.2018) (trnslted nd slightly dpted from lecture notes by Mrtin Klzr) Riemnn integrl Now we define precisely the concept of the re, in prticulr, the re of figure U(, b, f) under the grph of

More information

1 Sets Functions and Relations Mathematical Induction Equivalence of Sets and Countability The Real Numbers...

1 Sets Functions and Relations Mathematical Induction Equivalence of Sets and Countability The Real Numbers... Contents 1 Sets 1 1.1 Functions nd Reltions....................... 3 1.2 Mthemticl Induction....................... 5 1.3 Equivlence of Sets nd Countbility................ 6 1.4 The Rel Numbers..........................

More information

SUMMER KNOWHOW STUDY AND LEARNING CENTRE

SUMMER KNOWHOW STUDY AND LEARNING CENTRE SUMMER KNOWHOW STUDY AND LEARNING CENTRE Indices & Logrithms 2 Contents Indices.2 Frctionl Indices.4 Logrithms 6 Exponentil equtions. Simplifying Surds 13 Opertions on Surds..16 Scientific Nottion..18

More information

f(x) dx, If one of these two conditions is not met, we call the integral improper. Our usual definition for the value for the definite integral

f(x) dx, If one of these two conditions is not met, we call the integral improper. Our usual definition for the value for the definite integral Improper Integrls Every time tht we hve evluted definite integrl such s f(x) dx, we hve mde two implicit ssumptions bout the integrl:. The intervl [, b] is finite, nd. f(x) is continuous on [, b]. If one

More information

NOTES ON HILBERT SPACE

NOTES ON HILBERT SPACE NOTES ON HILBERT SPACE 1 DEFINITION: by Prof C-I Tn Deprtment of Physics Brown University A Hilbert spce is n inner product spce which, s metric spce, is complete We will not present n exhustive mthemticl

More information

Appendix to Notes 8 (a)

Appendix to Notes 8 (a) Appendix to Notes 8 () 13 Comprison of the Riemnn nd Lebesgue integrls. Recll Let f : [, b] R be bounded. Let D be prtition of [, b] such tht Let D = { = x 0 < x 1

More information

Solution to Fredholm Fuzzy Integral Equations with Degenerate Kernel

Solution to Fredholm Fuzzy Integral Equations with Degenerate Kernel Int. J. Contemp. Mth. Sciences, Vol. 6, 2011, no. 11, 535-543 Solution to Fredholm Fuzzy Integrl Equtions with Degenerte Kernel M. M. Shmivnd, A. Shhsvrn nd S. M. Tri Fculty of Science, Islmic Azd University

More information

Coalgebra, Lecture 15: Equations for Deterministic Automata

Coalgebra, Lecture 15: Equations for Deterministic Automata Colger, Lecture 15: Equtions for Deterministic Automt Julin Slmnc (nd Jurrin Rot) Decemer 19, 2016 In this lecture, we will study the concept of equtions for deterministic utomt. The notes re self contined

More information

Improper Integrals. Type I Improper Integrals How do we evaluate an integral such as

Improper Integrals. Type I Improper Integrals How do we evaluate an integral such as Improper Integrls Two different types of integrls cn qulify s improper. The first type of improper integrl (which we will refer to s Type I) involves evluting n integrl over n infinite region. In the grph

More information

Mapping the delta function and other Radon measures

Mapping the delta function and other Radon measures Mpping the delt function nd other Rdon mesures Notes for Mth583A, Fll 2008 November 25, 2008 Rdon mesures Consider continuous function f on the rel line with sclr vlues. It is sid to hve bounded support

More information

Natural examples of rings are the ring of integers, a ring of polynomials in one variable, the ring

Natural examples of rings are the ring of integers, a ring of polynomials in one variable, the ring More generlly, we define ring to be non-empty set R hving two binry opertions (we ll think of these s ddition nd multipliction) which is n Abelin group under + (we ll denote the dditive identity by 0),

More information

A product convergence theorem for Henstock Kurzweil integrals

A product convergence theorem for Henstock Kurzweil integrals A product convergence theorem for Henstock Kurzweil integrls Prsr Mohnty Erik Tlvil 1 Deprtment of Mthemticl nd Sttisticl Sciences University of Albert Edmonton AB Cnd T6G 2G1 pmohnty@mth.ulbert.c etlvil@mth.ulbert.c

More information

Intuitionistic Fuzzy Lattices and Intuitionistic Fuzzy Boolean Algebras

Intuitionistic Fuzzy Lattices and Intuitionistic Fuzzy Boolean Algebras Intuitionistic Fuzzy Lttices nd Intuitionistic Fuzzy oolen Algebrs.K. Tripthy #1, M.K. Stpthy *2 nd P.K.Choudhury ##3 # School of Computing Science nd Engineering VIT University Vellore-632014, TN, Indi

More information

SOLUTIONS FOR ANALYSIS QUALIFYING EXAM, FALL (1 + µ(f n )) f(x) =. But we don t need the exact bound.) Set

SOLUTIONS FOR ANALYSIS QUALIFYING EXAM, FALL (1 + µ(f n )) f(x) =. But we don t need the exact bound.) Set SOLUTIONS FOR ANALYSIS QUALIFYING EXAM, FALL 28 Nottion: N {, 2, 3,...}. (Tht is, N.. Let (X, M be mesurble spce with σ-finite positive mesure µ. Prove tht there is finite positive mesure ν on (X, M such

More information

2 Fundamentals of Functional Analysis

2 Fundamentals of Functional Analysis Fchgruppe Angewndte Anlysis und Numerik Dr. Mrtin Gutting 22. October 2015 2 Fundmentls of Functionl Anlysis This short introduction to the bsics of functionl nlysis shll give n overview of the results

More information

19 Optimal behavior: Game theory

19 Optimal behavior: Game theory Intro. to Artificil Intelligence: Dle Schuurmns, Relu Ptrscu 1 19 Optiml behvior: Gme theory Adversril stte dynmics hve to ccount for worst cse Compute policy π : S A tht mximizes minimum rewrd Let S (,

More information

QUADRATURE is an old-fashioned word that refers to

QUADRATURE is an old-fashioned word that refers to World Acdemy of Science Engineering nd Technology Interntionl Journl of Mthemticl nd Computtionl Sciences Vol:5 No:7 011 A New Qudrture Rule Derived from Spline Interpoltion with Error Anlysis Hdi Tghvfrd

More information

Introduction to the Calculus of Variations

Introduction to the Calculus of Variations Introduction to the Clculus of Vritions Jim Fischer Mrch 20, 1999 Abstrct This is self-contined pper which introduces fundmentl problem in the clculus of vritions, the problem of finding extreme vlues

More information