STEIN S METHOD AND THE LAPLACE DISTRIBUTION. John Pike Department of Mathematics Cornell University

Size: px
Start display at page:

Download "STEIN S METHOD AND THE LAPLACE DISTRIBUTION. John Pike Department of Mathematics Cornell University"

Transcription

1 STEIN S METHOD AND THE LAPLACE DISTRIBUTION John Pike Department of Mathematics Cornell University jpike@cornell.edu Haining Ren Department of Mathematics University of Southern California hren@usc.edu Astract. Using Stein s method techniques, we develop a framework which allows one to ound the error terms arising from approximation y the Laplace distriution and apply it to the study of random sums of mean zero random variales. As a corollary, we deduce a Berry-Esseen type theorem for the convergence of certain geometric sums. Our results make use of a second order characterizing equation and a distriutional transformation which is related to zero-iasing. 1. Background and Introduction Beginning with the pulication of Charles Stein s seminal paper [16], proailists and statisticians have developed a wide range of techniques ased on characterizing equations for ounding the distance etween the distriution of a random variale X and that of a random variale Z having some specified target distriution. The metrics for which these techniques are applicale are of the form d H L X, L Z = sup h H E[hX] E[hZ] for some suitale class of functions H, and include as special cases the Wasserstein, Kolmogorov, and total variation distances. The Kolmogorov distance gives the L distance etween the associated distriution functions, so H = {1,a] x : a R}. The total variation and Wasserstein distances correspond to letting H consist of indicators of Borel sets and 1-Lipschitz functions, respectively. The asic idea is to find an operator A such that E[AfX] = for all f elonging to some sufficiently large class of functions F if and only if L X = L Z. For example, Stein showed that Z N, σ if and only if E[A N fz] = E[ZfZ σ f Z] = for all asolutely continuous functions f with f < [16], and shortly thereafter Louis Chen proved that Z Poissonλ if and only if E[A P fz] = E[ZfZ λfz +1] = for all functions f for which the expectations exist []. Similar characterizing operators have since een worked out for several other common distriutions e.g. [6, 11, 1, 13]. 1 Mathematics Suject Classification. 6F5. Key words and phrases. Stein s Method, Laplace distriution, geometric summation. 1

2 STEIN S METHOD AND THE LAPLACE DISTRIBUTION Given a Stein operator A for L Z, one then shows that for every h H, the equation Afx = hx E[hZ] has solution f h F. Taking expectations, asolute values, and suprema yields d H L X, L Z = sup E[hX] E[hZ] = sup E[Af h X]. h H h H In practice, this is usually how one proves that E[AfX] = for f F is sufficient for L X = L Z. The intuition is that since E[AfZ] = for f F, the distriution of X should e close to that of Z when E[AfX] is close to zero. Remarkaly, it is often easier to work with the right-hand side of the aove equation, and the tools for analyzing distances etween distriutions in this manner are collectively known as Stein s method. For more on this rich and fascinating suject, the authors recommend [17, 15, 3, 4]. In this paper, we apply the aove ideas to the Laplace distriution. For a R, >, a random variale W Laplacea, has distriution function { 1 F W w; a, = e w a, w a 1 1 w a e, w a and density f W w; a, = 1 w a e, w R. If W Laplace,, then its moments are given y { E[W k, k is odd ] = k k!, k is even, and its characteristic function is 1 ϕ W t = 1 + t. This distriution was introduced y P.S. Laplace in 1774, four years prior to his proposal of the second law of errors, now known as the normal distriution. Though nowhere near as uiquitous as its younger siling, the Laplace distriution appears in numerous applications, including image and speech compression, options pricing, and modeling sizes of sand particles, diamonds, and eans. For more properties and applications of the Laplace distriution, the reader is referred to the text [1]. Our interest in the Laplace distriution was sparked y the fact that if X 1, X,... is a sequence of random variales satisfying certain technical assumptions and Np X i converges N p Geometricp is independent of the X i s, then the sum p 1 weakly to the Laplace distriution as p [1]. Such geometric sums arise in a variety of settings [9], and the general setup distriutional convergence of sums of random variales is exactly the type of prolem for which one expects Stein s method computations to yield useful results. Indeed, Erol Peköz and Adrian Röllin have applied Stein s method arguments to generalize a theorem due to Rényi concerning the convergence of sums of a random numer of positive random variales to the exponential distriution [1]. By an analogous line of reasoning, we are ale to carry out a similar program for convergence of random sums of certain mean zero random variales to the Laplace distriution.

3 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 3 We egin in Section y introducing a Stein operator which we show completely characterizes the mean zero Laplace distriution. Specifically, we prove Theorem 1.1. Let W Laplace, and define the operator A y Afx = fx f f x. Then E[AgW ] = for every function g such that g and g are locally asolutely continuous and E g W, E g W <. Conversely, if X is any random variale such that E[AgX] = for every twicedifferentiale function g with g, g, g <, then X Laplace,. In Section 3, we use this characterization to ound the distance to the Laplace distriution. For technical reasons, we work in the ounded Lipschitz metric, d BL, which is defined in terms of 1-Lipschitz test functions with sup norm 1. We egin y defining the centered equilirium transformation X X L in terms of the functional equation E[fX] f = 1 E[X ]E[f X L ] for all twice-differentiale functions f such that f, f, and f are ounded. After estalishing that X L exists whenever X has mean zero and finite nonzero variance, we derive the following coupling ound. Theorem 1.. Suppose that X is a random variale with E[X] =, E[X ] =,, and let X L have the centered equilirium distriution for X. Then d BL L X, Laplace, + E X X L. Finally, in Section 4 we apply these tools to the study of random sums of mean zero random variales. As a special case, we show Theorem 1.3. Let X 1, X,... e a [ sequence of independent random variales with E[X i ] =, E[Xi ] =, sup i 1 E X i 3] = ρ <, and let N Geometricp with strictly positive support e independent of the X i s. Then N d BL L p 1 X i, Laplace, p 1 for all p, ρ 6. Characterizing the Laplace Distriution Our first order of usiness is to estalish a characterizing operator for the Laplace distriution. As is typical in Stein s method constructions, we split the proof of Theorem 1.1 into two parts. We egin with Lemma.1. Suppose that W Laplace,. If g and g are locally asolutely continuous with E g W, E g W <, then E[gW ] g = E[g W ].

4 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 4 Proof. Applying Fuini s theorem twice shows that g xe x dx = g 1 y x dy dx = 1 = 1 = 1 ˆ y = 1 = 1 = 1 x e g xe y dxdy y g ye y g dy e y dy g 1 y e z dz dy g ˆ z g ye z dydz g gze z dz g gze z dz g g. e z dz g Setting hy = g y, it follows from the previous calculation that ˆ g xe x dx = g ye y dy = h ye y dy = 1 = 1 = 1 ˆ Summing the aove expressions gives hze z h dz h g ze z g dz + g gze z dz g + g. ˆ E[g W ] = 1 g xe x 1 dx + g xe x dx = 1 [ ˆ 1 gze z g dz + g ˆ 1 + gze z g dz = 1 ˆ 1 gze z g dz ] g = 1 E[gW ] g. Note that since the density of a Laplace, random variale is given y f W w = w 1 e, the density method [3] suggests the following characterizing equation for the Laplace distriution: g w 1 sgnwgw = g w + f W w gw =, f W w

5 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 5 and indeed one can verify that if W Laplace,, then E[g W ] = 1 E[sgnW gw ] for all asolutely continuous g for which these expectations exist. Thus if g is such a function as well, setting Gw = sgnw gw g gives E[g W ] = 1 E[sgnW g W ] = 1 E[G W ] = 1 E[sgnW GW ] = 1 E[gW ] g, so the general form of the equation in Theorem 1.1 can e ascertained y iterating the density method. Alternatively, it is known [15] that if Z Exponential1, then E[g Z] = E[gZ] g for all asolutely continuous g with E g Z <. Thus if g is also asolutely continuous and E g Z <, then E[g Z] = E[g Z] g = E[gZ] g g. Using this oservation, one can derive the equation in Lemma.1 y noting that if J is independent of Z with PJ = 1 = PJ = 1 = 1, then W = JZ has the Laplace, distriution. We include each of these approaches ecause their analogues may e variously applicale in different situations involving the construction of characterizing equations. Oserve that there is an iterative step involved in each case. By manipulating the integral defining E[g W ] or using the usual Stein equation for the exponential along with the representation W = JZ, one arrives at the first order equation E[g W ] = 1 E[sgnW gw ] suggested y one application of the density method. However, we were not ale to get much mileage out of the operator Ãgx = g x 1 sgnxgx, while the second-order operator Agx = gx g g x turned out to e quite effective. This idea of iterating more traditional procedures to otain higher order characterizing equations which are simpler to work with is one of the key insights of this paper. Now, in order to estalish the second part of Theorem 1.1, we will show that any X satisfying the hypotheses has d BL L X, Laplace, = where d BL denotes the ounded Lipschitz distance given y d BL L X, L Y = sup E[hX] E[hY ], h H BL H BL = {h : h 1 and hx hy x y for all x, y R}. The claim will follow since d BL is a metric on the space of Borel proaility measures on R [19]. In keeping with the general strategy laid out in the introduction, we consider the initial value prolem gx g x = hx W h, g = where h H BL and W h := E[hW ], W Laplace,.

6 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 6 Lemma.. For h H BL, W Laplace,, hx = hx W h, a ounded, twice-differentiale solution to the initial value prolem is given y g h x = 1 gx g x = hx, g = e x x e y hydy + e x ˆ x e y hydy. This solution satisfies g h, g h, g h 4 and g h + 3. Proof. The general solution to the homogeneous equation g x gx = is given y g x = C 1 e x + C e x, so, since the associated Wronskian is nonzero, the variation of parameters method suggests that a solution to the inhomogeneous equation g x gx = hx is where u h x = 1 Differentiation gives x g h x = u h xe x + vh xe x e y hydy, vh x = 1 ˆ x g hx = hx u hxe x 1 + hx 1 v hxe x 1 = and thus g hx = 1 hx + u h xe x + vh xe x e y hydy. uh xe x vh xe x, = 1 g h x hx, so g h is indeed a solution. To see that the initial condition is satisfied, we oserve that g h = u h + v h = 1 = hyf W ydy W h e y hydy = f W y hy W h dy f W ydy = W h W h =. Moreover, since h 1, ˆ W h = 1 hxe x dx 1 e x dx = 1, and thus hx hx + W h. Consequently, and u h xe x 1 e x vh xe x 1 e x x ˆ x e y dy = 1 e y dy = 1, for all x R, and the ounds on g h and g h follow. Noting that g h x = 1 g h x hx and thus g h x = 1 g 3 h x 1 h x completes the proof since h and h 1.

7 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 7 With the preceding result in hand, we can finish of the proof of Theorem 1.1 via Lemma.3. If X is a random variale such E[gX] g = E[g X] for every twice-differentiale function g with g, g, g <, then X Laplace,. Proof. Let W Laplace, and, for h H BL, let g h e as in Lemma.. Because g h = and g h, g h, g h are ounded, it follows from the aove assumptions that E[hX] E[hW ] = E[g h X g hx] =. Taking the supremum over h H BL shows that d BL L X, L W =. Before moving on, we oserve that the reason we are working with the ounded Lipschitz distance is that the ounds on g h and its derivatives depended on oth h and h having finite sup norm. As d BL is not especially common at least explicitly in the Stein s method literature, we conclude this section with a proposition relating it to the more familiar Kolmogorov distance d K L X, L Y = sup P{X x} P{Y x}. x R Proposition.4. If Z is an asolutely continuous random variale whose density, f Z, is uniformly ounded y a constant C <, then for any random variale X, d K L X, L Z C + dbl L X, L Z. Proof. We first note that the inequality holds trivially if d BL L X, L Z = as d BL and d K are metrics. Also, since d K P, Q 1 for all proaility measures P and Q, C+ 1, and d BL L X, L Z 1 implies d BL L X, L Z 1, we have d K L X, L Z 1 C + dbl L X, L Z whenever d BL L X, L Z 1. d BL L X, L Z, 1. Now, for x R, ε >, write h x z = 1,x] z = Then for all x R, Thus it suffices to consider the case where { 1, z x, z > x and h x,εz = 1, z x 1 z x ε, z x, x + ε]., z > x + ε E[h x X h x Z] = E[h x X] E[h x,ε Z] + E[h x,ε Z] E[h x Z] ˆ x+ε E[h x,ε X] E[h x,ε Z] + 1 z x f Z zdz x ε E[h x,ε X] E[h x,ε Z] + Cε.

8 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 8 Since d BL L X, L Z, 1, if we take ε = d BL L X, L Z, 1, then εh x,ε H BL and thus E[h x X h x Z] 1 ε E[εh x,εx] E[εh x,ε Z] + Cε 1 ε d BL L X, L Z + Cε = C + dbl L X, L Z. A similar argument using E[h x Z h x X] = E[h x Z] E[h x ε,ε Z] + E[h x ε,ε Z] E[h x X] Cε + E[h x ε,εz] E[h x ε,ε X] shows that E[h x X] E[h x Z] C + dbl L X, L Z for all x R, and the proposition follows y taking suprema. 1 Remark. When C 1, we can take ε = C d BL L X, L Z in the aove argument to otain an improved ound of d K L X, L Z 3 CdBL L X, L Z. To the est of the authors knowledge, the aove proposition is original, though the proof follows the same asic line of reasoning as the well-known ound on the Kolmogorov distance y the Wasserstein distance see Proposition 1. in [15]. It seems that the primary reason for using the Wasserstein metric, d W, is that it enales one to work with smoother test functions while still implying convergence in the more natural Kolmogorov distance. Proposition.4 shows that d BL also upper-ounds d K while enjoying all of the resulting smoothness of Wasserstein test functions and with additional oundedness properties to oot. Moreover, the Wasserstein distance is not always well-defined e.g. if one of the distriutions does not have a first moment, whereas d BL always exists. Finally, d BL is a fairly natural measure of distance since it metrizes weak convergence [19]. However, we always have d W L X, L Z d BL L X, L Z, and it is possile for a sequence to converge in d BL ut not in d W or d K. Furthermore, the ounded Lipschitz metric does not scale as nicely as the Kolmogorov or Wasserstein distances when the associated random variales are multiplied y a positive constant. For the remainder of this paper, we will state our results in terms of d BL with the corresponding Kolmogorov ound eing implicit therein, though one should note that, as with the Wasserstein ound on d K, Kolmogorov ounds otained in this fashion are not necessarily optimal, often giving the root of the true rate. 3. The Centered Equilirium Transformation Our next task is to use the characterization in Theorem 1.1 to otain ounds on the error terms resulting from approximation y the Laplace distriution. To this end, we introduce the following definition.

9 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 9 Definition 3.1. For any nondegenerate random variale X with mean zero and finite variance, we say that the random variale X L has the centered equilirium distriution with respect to X if E[fX] f = 1 E[X ]E[f X L ] for all twice-differentiale functions f such that f, f, and f are ounded. We call the map X X L the centered equilirium transformation. Note that the centered equilirium distriution is uniquely defined ecause E[f X] = E[f Y ] for all twice continuously differentiale functions with compact support implies X = d Y. The nomenclature is in reference to the equilirium distriution from renewal theory, which was used in a similar manner in [1] for a related prolem involving the exponential distriution. Since the characterizing equation for the Laplace distriution involves second derivatives and a variance term, one expects some kind of relation etween X L and the zero ias distriution for X defined y E[XfX] = E[X ]E[f X z ] for all asolutely continuous f for which E XfX < [7] in much the same way as the equilirium distriution is related to the size ias distriution [1]. The following theorem shows that this is indeed the case. Moreover, since the zero ias distriution is defined for any X with E[X] = and VarX,, it will follow that every such random variale has a centered equilirium distriution. Theorem 3.. Suppose that X has mean zero and variance σ,. Let X z have the zero ias distriution with respect to X and let B Beta, 1 e independent of X z. Then X L := BX z satisfies E[fX] f = σ E[f X L ] for all twice-differentiale f with f, f, f <. Proof. Applying the fundamental theorem of calculus, Fuini s theorem, the definition of X z, and the fact that B has density px = x1 [,1] x gives [ˆ 1 ] ˆ 1 E[fX] f = E Xf uxdu = E [Xf ux] du ˆ 1 [ˆ 1 = σ ue[f ux z ]du = σ E = σ E [f BX z ] The assumptions ensure that all of the functions are integrale. ] uf ux z du Corollary 3.3. If X has variance E[X ] = and X L has the centered equilirium distriution with respect to X, then X L is asolutely continuous with density f X Lx = 1 ˆ 1 [ E X; X > x ] dv. v Proof. X L = d BX z with B and X z as in Theorem 3., so the claim follows from the fact [7] that X z is asolutely continuous with density f X zx = 1 E[X; X > x] y the usual method of computing the density of a product.

10 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 1 Remark. In an earlier version of this paper, we estalished the existence of the centered equilirium distriution y showing that for certain random variales X, X L can e otained y iterating the X P ias transformation from [8] with P x = sgnx. Though there may e some merit to such a strategy and it provides another example of how results for higher order Stein operators may e otained y iterating more traditional techniques, in our case it required the rather artificial assumption that the variates in the domain of the transformation have median zero. Those interested in the iterated X P ias approach are referred to the article [5] y Christian Döler, which contains the essential technical details of our original argument and suggests certain improvements. Lemma.3 shows that, up to scaling, the mean zero Laplace distriution is the unique fixed point of the centered equilirium transformation. Thus one expects that if a random variale is close to its centered equilirium transform, then its distriution is close to the Laplace law. Theorem 1. formalizes this intuition. Proof of Theorem 1.. If X is a random variale with E[X ] = < and X L has the centered equilirium distriution for X, then for all h H BL, taking g h as in Lemma., we see that W h E[hX] = E[g h X g hx] = E[ g hx L g hx] E g hx L g hx g h E X L X = + E X X L. For the example in Section 4, we will also need the following complementary result. Proposition 3.4. If Y L has the centered equilirium distriution for Y and E[Y ] =, then E Y Y L E Y E[ Y 3 ]. Proof. We may assume that E[ Y 3 ] < as the inequality is trivial otherwise. The result will follow immediately from the triangle inequality if we can show that E Y L = 1 6 E Y 3, which is what one would otain y formally plugging fy = y 3 into the definition of the transformation Y Y L. Of course, neither f, f, nor f is ounded, so we must proceed y approximation. To this end, define x 3, x n n 3 + 3n x n 3 x n, n < x n + n f n x = n 3 + 3n n + n x 3 n + n x, n + n < x n + n n + n x 3, n + n < x n + n, x > n + n By construction, f n is smooth with compact support and satisfies f n i f i for all n N, i =, 1,, thus fulfilling the conditions in the definition of the centered

11 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 11 equilirium transformation when E[ Y 3 ] <. Moreover, f n i f i pointwise for i =, 1,, so it follows from the dominated convergence theorem that E[fY ] = lim n E[f ny ] = lim n E[f ny L ]. Fatou s lemma shows that f Y L is integrale since E [ f Y L ] [ ] = E lim inf f ny L lim inf E [ f ny L ] = 1 E[fY ] <, n n so another application of dominated convergence gives E[fY ] = lim n E[f ny L ] = E[f Y L ]. The same general argument shows that if Y L has the centered equilirium distriution for Y and E [ Y n ] <, then E[qY ] q = 1 E[Y ]E[q Y L ] for any polynomial q of degree at most n. As is often the case with distriutional transformations defined in terms of functional equations, we find it more convenient to define X X L in terms of a relatively small class of test functions and then argue y approximation when we want to apply the relation more generally. 4. Random Sums The p-geometric summation of a sequence of random variales X 1, X,... is defined as S p = X 1 + X X Np where N p is geometric with success proaility p - that is, P{N p = n} = p1 p n 1, n N - and is independent of all else. A result due to Rényi [14] states that if X 1, X,... are i.i.d., positive, nondegenerate random variales with E[X i ] = 1, then L ps p Exponential1 as p. In fact, just as the normal law is the only nondegenerate distriution with finite variance that is stale under ordinary summation in the sense that if X, X 1, X,... are i.i.d. nondegenerate random variales with finite variance, then for every n N, there exist a n >, n R such that X = d a n X X n + n, it can e shown that if X, X 1, X,... are i.i.d., positive, and nondegenerate with finite variance, then there exists a p > such that a p X X Np =d X for all p, 1 if and only if X has an exponential distriution. Similarly, if we assume that Y, Y 1, Y,... are i.i.d., symmetric, and nondegenerate with finite variance, then there exists a p > such that a p Y Y Np =d Y for all p, 1 if and only if Y has a Laplace distriution. Moreover, it must e the case that a p = p 1. In addition, we have an analog of Rényi s theorem [1]: Theorem 4.1. Suppose that X 1, X,... are i.i.d., symmetric, and nondegenerate random variales with finite variance σ, and let N p Geometricp e independent of the X i s. If X i d X as p, a p N p then there exists γ > such that a p = p 1 γ + op 1 and X has the Laplace distriution with mean and variance σ γ.

12 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 1 A recent theorem due to Alexis Toda [18] gives the following Lindeerg-type conditions for the existence of the distriutional limit in Theorem 4.1. Theorem 4. Toda. Let X 1, X,... e a sequence of independent ut not necessarily identically distriuted random variales such that E[X i ] = and VarX i = σi, and let N p Geometricp independent of the X i s. Suppose that lim n n α σn = for some < α < 1, and for all ɛ >, lim p σ 1 := lim n n n σi > exists, 1 p i 1 pe[xi ; X i ɛp 1 ] =. Then as p, the sum p 1 Np X i converges weakly to the Laplace distriution with mean and variance σ. Remark. The original statement of Toda s theorem is slightly more general, allowing for convergence to a possily asymmetric Laplace distriution. In 11, Peköz and Röllin were ale to generalize Rényi s theorem y using a distriutional transformation inspired y Stein s method considerations [1]. Specifically, for a nonnegative random variale X with E[X] <, they say that X e has the equilirium distriution with respect to X if E[fX] f = E[X]E[f X e ] for all Lipschitz f and use this to ound the Wasserstein and Kolmogorov distances to the Exponential1 distriution. The equilirium distriution arises in renewal theory, ut its utility in analyzing convergence to the exponential distriution comes from the fact that a Stein operator for the exponential distriution with mean one is given y A E fx = f x fx+f, so X Exponential1 is the unique fixed point of the equilirium transformation [15]. This transformation and the similarity etween our characterization of the Laplace and the aove characterization of the exponential inspired our construction of the centered equilirium transformation, and the fact that oth distriutions are stale under geometric summation led us to parallel their argument for ounding the distance etween p-geometric sums of positive random variales and the exponential distriution in order to otain corresponding results for the Laplace case. Our results are summarized in the following theorem. Theorem 4.3. Let N e any N-valued random variale with µ = E[N] < and let X 1, X,... e a sequence of independent random variales, [ independent of N, with N ] [ E[X i ] =, and E[Xi ] = N ] σ i,. Set σ = E = E X i σ i and let M e any N-valued random variale, independent of the X i s and defined on the same space as N, satisfying P{M = m} = σ m P{N m}. σ

13 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 13 Then d BL L µ 1 N X i m=1, Laplace, µ σ σ µ E XM X L Proof. We first note that m σ = P{N = m} σi = M + sup i 1 P{N m}σm, m=1 [ ] σ i E N M 1. so M is well-defined. Now, taking V = µ 1 N X i, we claim that V L = µ 1 M 1 X i + XM L has the centered equilirium distriution with respect to V. Throughout, Xm L is taken to e independent of M, N, and X k for k m. To see that this is so, let f e any function satisfying the assumptions in Definition 3.1. Then, using the notation m X = {X i } i 1, g m X = f µ 1 X i, f s x = f µ 1 s + µ 1 x, letting ν m denote the distriution of S m 1 = and oserving that, y independence, m 1 E[h Xm S L m 1 = s] = E[h Xm] L = σm E[hX m h] = σm E[hX m h S m 1 = s] for all s R and all twice differentiale h with h, h, h ounded, we see that E [f µ 1 m 1 ] ˆ X i + µ 1 X L m = X i, [ ] E f µ 1 s + µ 1 X L m S m 1 = s dν m s ˆ = E [ µf s Xm L S m 1 = s ] dν m s = µ ˆ σm E [f s X m f s S m 1 = s] dν m s = µ ˆ [ σm E fµ 1 s + µ 1 Xm [ = µ σm E f µ 1 fµ 1 s Sm 1 = s m X i f = µ σm E [g m X g m 1 X] µ 1 ] dν m s m 1 X i ]

14 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 14 for all m N, hence E[f V L ] = P{M = m}e [f m=1 = µ σ σ σ m=1 m = E[V ] E µ 1 m 1 X i + µ 1 X L m ] P{M = m}e [g m X g m 1 X] [ ] P{N m} g m X g m 1 X m=1 = E[V ] E[g N X] g X = E[V E[fV ] f. ] Having shown that V L does in fact have the centered equilirium distriution for V, we can apply Theorem 1. with B = E[V ] = σ µ to otain d BL L V, Laplace, B 1 + E V V L B = µ σ N M E XM XM L + sgnn M µ E X M XM L + E σ i=n M+1 N M X i i=n M+1 X i. Finally, since the X i s are independent with mean zero, the Cauchy-Schwarz inequality gives E N M i=n M+1 X i = P{N M = j, N M = k}e j=1 k=1 P{N M = j, N M = k}e j=1 k=1 j=1 k=1 σ i = sup i 1 and the proof is complete. j+k i=j+1 j+k i=j+1 P{N M = j, N M = k}k 1 sup σ i i 1 k=1 X i X i [ ] k 1 P{ N M = k} = sup σ i E i 1 N M 1, Remark. [ When the summands in Theorem 4.3 have common variance σ1, we have N ] σ = E = µσ1, so M is defined y σ 1 P{M = m} = σ m σ P{N m} = 1 P{N m}. µ In other words, M is the discrete equilirium transformation of N defined in [13] for use in geometric approximation. Thus the E[ N M 1 ] term in the ound 1

15 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 15 from Theorem 4.3 may e regarded as measuring how close L N is to a geometric distriution. If the summands are i.i.d., then X M = d X 1 as M is assumed to e independent of all else, so Theorem 1. shows that the E XM XM L term measures how close the distriution of the summands is to the Laplace. To put this into context, recall that if the X i s are i.i.d. Laplace, and N Geometricp, then p 1 N X i Laplace,. We conclude our discussion with a proof of Theorem 1.3, which gives sufficient conditions for weak convergence in the setting of Theorem 4.1. Though it requires that the X i s have uniformly ounded third asolute moments, the condition of symmetry is dropped and the identical distriution assumption is reduced to the requirement that the X i s have common variance. This result is not quite as general as Theorem 4., ut it does provide ounds on the error terms. Proof of Theorem 1.3. We are trying to ound the distance etween the Laplace, distriution and that of p 1 N X i where X 1, X,... are independent mean zero random variales with common [ variance E[Xi ] = and uniformly ounded third asolute moments sup i 1 E X i 3] = ρ <, and N Geometricp is independent of the X i s. In the language of Theorem 4.3, we have µ = 1 p, σ = µ, and σm P{N m} = pp{n m} = p σ 1 p m 1 = p1 p m 1 = P{N = m}, i=m so we can take M = N to otain N d BL L p 1 X i, Laplace, p 1 p 1 + E XN XN L. Applying Proposition 3.4 gives E XN XN L = P{N = n}e Xn Xn L and the result follows n=1 n=1 n=1 P{N = n} E X n + 1 [ 6 E X n 3] P{N = n} E [ Xn ] 1 + ρ 6 = + ρ 6, Acknowledgements The authors would like to thank Larry Goldstein for suggesting the use of Stein s method to get convergence rates in the general setting of Theorem 4. and for many helpful comments throughout the preparation of this paper. Thanks also to Alex Rozinov for several illuminating conversations concerning the material in Section 4 and to the anonymous referees whose careful notes were of great help to us.

16 STEIN S METHOD AND THE LAPLACE DISTRIBUTION 16 References [1] Timothy C. Brown and M. J. Phillips. Negative inomial approximation with Stein s method. Methodol. Comput. Appl. Proa., 14:47 41, MR [] Louis H. Y. Chen. Poisson approximation for dependent trials. Ann. Proaility, 33: , MR [3] Louis H. Y. Chen, Larry Goldstein, and Qi-Man Shao. Normal approximation y Stein s method. Proaility and its Applications New York. Springer, Heidelerg, 11. MR7364. [4] Persi Diaconis and Susan Holmes, editors. Stein s method: expository lectures and applications. Papers from the Workshop on Stein s Method held at Stanford University, Stanford, CA, Institute of Mathematical Statistics Lecture Notes Monograph Series, 46. Institute of Mathematical Statistics, Beachwood, OH, 4. MR [5] Christian Döler. Distriutional transformations without orthogonality relations. ArXiv e- prints, 13. arxiv: [6] Werner Ehm. Binomial approximation to the Poisson inomial distriution. Statist. Proa. Lett., 111:7 16, MR [7] Larry Goldstein and Gesine Reinert. Stein s method and the zero ias transformation with application to simple random sampling. Ann. Appl. Proa., 74:935 95, MR [8] Larry Goldstein and Gesine Reinert. Distriutional transformations, orthogonal polynomials, and Stein characterizations. J. Theoret. Proa., 181:37 6, 5. MR1378. [9] Vladimir Kalashnikov. Geometric sums: ounds for rare events with applications. Risk analysis, reliaility, queueing, volume 413 of Mathematics and its Applications. Kluwer Academic Pulishers Group, Dordrecht, MR [1] Samuel Kotz, Tomasz J. Kozuowski, and Krzysztof Podgórski. The Laplace distriution and generalizations. A revisit with applications to communications, economics, engineering, and finance. Birkhäuser Boston, Inc., Boston, MA, 1. MR [11] Ho Ming Luk. Stein s method for the Gamma distriution and related statistical applications. Thesis Ph.D. University of Southern California. ProQuest LLC, Ann Aror, MI, MR6934. [1] Erol A. Peköz and Adrian Röllin. New rates for exponential approximation and the theorems of Rényi and Yaglom. Ann. Proa., 39:587 68, 11. MR [13] Erol A. Peköz, Adrian Röllin, and Nathan Ross. Total variation error ounds for geometric approximation. Bernoulli, 19:61 63, 13. MR [14] Alfréd Rényi. A characterization of Poisson processes. Magyar Tud. Akad. Mat. Kutató Int. Közl., 1: , MR [15] Nathan Ross. Fundamentals of Stein s method. Proa. Surv., 8:1 93, 11. MR [16] Charles Stein. A ound for the error in the normal approximation to the distriution of a sum of dependent random variales. In Proceedings of the Sixth Berkeley Symposium on Mathematical Statistics and Proaility Univ. California, Berkeley, Calif., 197/1971, Vol. II: Proaility theory, pages Univ. California Press, Berkeley, Calif., 197. MR4873. [17] Charles Stein. Approximate computation of expectations. Institute of Mathematical Statistics Lecture Notes Monograph Series, 7. Institute of Mathematical Statistics, Hayward, CA, MR887. [18] Alexis A. Toda. Weak limit of the geometric sum of independent ut not identically distriuted random variales. ArXiv e-prints, 11. arxiv: [19] Aad W. van der Vaart and Jon A. Wellner. Weak convergence and empirical processes. With applications to statistics. Springer Series in Statistics. Springer-Verlag, New York, MR

EVALUATIONS OF EXPECTED GENERALIZED ORDER STATISTICS IN VARIOUS SCALE UNITS

EVALUATIONS OF EXPECTED GENERALIZED ORDER STATISTICS IN VARIOUS SCALE UNITS APPLICATIONES MATHEMATICAE 9,3 (), pp. 85 95 Erhard Cramer (Oldenurg) Udo Kamps (Oldenurg) Tomasz Rychlik (Toruń) EVALUATIONS OF EXPECTED GENERALIZED ORDER STATISTICS IN VARIOUS SCALE UNITS Astract. We

More information

A Gentle Introduction to Stein s Method for Normal Approximation I

A Gentle Introduction to Stein s Method for Normal Approximation I A Gentle Introduction to Stein s Method for Normal Approximation I Larry Goldstein University of Southern California Introduction to Stein s Method for Normal Approximation 1. Much activity since Stein

More information

On the rate of convergence in limit theorems for geometric sums of i.i.d. random variables

On the rate of convergence in limit theorems for geometric sums of i.i.d. random variables On the rate of convergence in limit theorems for geometric sums of i.i.d. random variables Self-normalized Asymptotic Theory in Probability, Statistics and Econometrics IMS (NUS, Singapore), May 19-23,

More information

THE LINDEBERG-FELLER CENTRAL LIMIT THEOREM VIA ZERO BIAS TRANSFORMATION

THE LINDEBERG-FELLER CENTRAL LIMIT THEOREM VIA ZERO BIAS TRANSFORMATION THE LINDEBERG-FELLER CENTRAL LIMIT THEOREM VIA ZERO BIAS TRANSFORMATION JAINUL VAGHASIA Contents. Introduction. Notations 3. Background in Probability Theory 3.. Expectation and Variance 3.. Convergence

More information

Normal Approximation for Hierarchical Structures

Normal Approximation for Hierarchical Structures Normal Approximation for Hierarchical Structures Larry Goldstein University of Southern California July 1, 2004 Abstract Given F : [a, b] k [a, b] and a non-constant X 0 with P (X 0 [a, b]) = 1, define

More information

A Short Introduction to Stein s Method

A Short Introduction to Stein s Method A Short Introduction to Stein s Method Gesine Reinert Department of Statistics University of Oxford 1 Overview Lecture 1: focusses mainly on normal approximation Lecture 2: other approximations 2 1. The

More information

A converse Gaussian Poincare-type inequality for convex functions

A converse Gaussian Poincare-type inequality for convex functions Statistics & Proaility Letters 44 999 28 290 www.elsevier.nl/locate/stapro A converse Gaussian Poincare-type inequality for convex functions S.G. Bokov a;, C. Houdre ; ;2 a Department of Mathematics, Syktyvkar

More information

Stein s Method and the Zero Bias Transformation with Application to Simple Random Sampling

Stein s Method and the Zero Bias Transformation with Application to Simple Random Sampling Stein s Method and the Zero Bias Transformation with Application to Simple Random Sampling Larry Goldstein and Gesine Reinert November 8, 001 Abstract Let W be a random variable with mean zero and variance

More information

Stein s method and zero bias transformation: Application to CDO pricing

Stein s method and zero bias transformation: Application to CDO pricing Stein s method and zero bias transformation: Application to CDO pricing ESILV and Ecole Polytechnique Joint work with N. El Karoui Introduction CDO a portfolio credit derivative containing 100 underlying

More information

Convergence rates in weighted L 1 spaces of kernel density estimators for linear processes

Convergence rates in weighted L 1 spaces of kernel density estimators for linear processes Alea 4, 117 129 (2008) Convergence rates in weighted L 1 spaces of kernel density estimators for linear processes Anton Schick and Wolfgang Wefelmeyer Anton Schick, Department of Mathematical Sciences,

More information

Normal approximation of Poisson functionals in Kolmogorov distance

Normal approximation of Poisson functionals in Kolmogorov distance Normal approximation of Poisson functionals in Kolmogorov distance Matthias Schulte Abstract Peccati, Solè, Taqqu, and Utzet recently combined Stein s method and Malliavin calculus to obtain a bound for

More information

1 Solution to Problem 2.1

1 Solution to Problem 2.1 Solution to Problem 2. I incorrectly worked this exercise instead of 2.2, so I decided to include the solution anyway. a) We have X Y /3, which is a - function. It maps the interval, ) where X lives) onto

More information

Empirical Processes: General Weak Convergence Theory

Empirical Processes: General Weak Convergence Theory Empirical Processes: General Weak Convergence Theory Moulinath Banerjee May 18, 2010 1 Extended Weak Convergence The lack of measurability of the empirical process with respect to the sigma-field generated

More information

Lecture 3: Expected Value. These integrals are taken over all of Ω. If we wish to integrate over a measurable subset A Ω, we will write

Lecture 3: Expected Value. These integrals are taken over all of Ω. If we wish to integrate over a measurable subset A Ω, we will write Lecture 3: Expected Value 1.) Definitions. If X 0 is a random variable on (Ω, F, P), then we define its expected value to be EX = XdP. Notice that this quantity may be. For general X, we say that EX exists

More information

Problem List MATH 5143 Fall, 2013

Problem List MATH 5143 Fall, 2013 Problem List MATH 5143 Fall, 2013 On any problem you may use the result of any previous problem (even if you were not able to do it) and any information given in class up to the moment the problem was

More information

Stein s Method: Distributional Approximation and Concentration of Measure

Stein s Method: Distributional Approximation and Concentration of Measure Stein s Method: Distributional Approximation and Concentration of Measure Larry Goldstein University of Southern California 36 th Midwest Probability Colloquium, 2014 Stein s method for Distributional

More information

Limiting Distributions

Limiting Distributions We introduce the mode of convergence for a sequence of random variables, and discuss the convergence in probability and in distribution. The concept of convergence leads us to the two fundamental results

More information

Limiting Distributions

Limiting Distributions Limiting Distributions We introduce the mode of convergence for a sequence of random variables, and discuss the convergence in probability and in distribution. The concept of convergence leads us to the

More information

Bounds on the Constant in the Mean Central Limit Theorem

Bounds on the Constant in the Mean Central Limit Theorem Bounds on the Constant in the Mean Central Limit Theorem Larry Goldstein University of Southern California, Ann. Prob. 2010 Classical Berry Esseen Theorem Let X, X 1, X 2,... be i.i.d. with distribution

More information

Metric Spaces and Topology

Metric Spaces and Topology Chapter 2 Metric Spaces and Topology From an engineering perspective, the most important way to construct a topology on a set is to define the topology in terms of a metric on the set. This approach underlies

More information

Notes on Poisson Approximation

Notes on Poisson Approximation Notes on Poisson Approximation A. D. Barbour* Universität Zürich Progress in Stein s Method, Singapore, January 2009 These notes are a supplement to the article Topics in Poisson Approximation, which appeared

More information

MODERATE DEVIATIONS IN POISSON APPROXIMATION: A FIRST ATTEMPT

MODERATE DEVIATIONS IN POISSON APPROXIMATION: A FIRST ATTEMPT Statistica Sinica 23 (2013), 1523-1540 doi:http://dx.doi.org/10.5705/ss.2012.203s MODERATE DEVIATIONS IN POISSON APPROXIMATION: A FIRST ATTEMPT Louis H. Y. Chen 1, Xiao Fang 1,2 and Qi-Man Shao 3 1 National

More information

Perhaps the simplest way of modeling two (discrete) random variables is by means of a joint PMF, defined as follows.

Perhaps the simplest way of modeling two (discrete) random variables is by means of a joint PMF, defined as follows. Chapter 5 Two Random Variables In a practical engineering problem, there is almost always causal relationship between different events. Some relationships are determined by physical laws, e.g., voltage

More information

arxiv: v3 [math.mg] 3 Nov 2017

arxiv: v3 [math.mg] 3 Nov 2017 arxiv:702.0069v3 [math.mg] 3 ov 207 Random polytopes: central limit theorems for intrinsic volumes Christoph Thäle, icola Turchi and Florian Wespi Abstract Short and transparent proofs of central limit

More information

STRONG NORMALITY AND GENERALIZED COPELAND ERDŐS NUMBERS

STRONG NORMALITY AND GENERALIZED COPELAND ERDŐS NUMBERS #A INTEGERS 6 (206) STRONG NORMALITY AND GENERALIZED COPELAND ERDŐS NUMBERS Elliot Catt School of Mathematical and Physical Sciences, The University of Newcastle, Callaghan, New South Wales, Australia

More information

If g is also continuous and strictly increasing on J, we may apply the strictly increasing inverse function g 1 to this inequality to get

If g is also continuous and strictly increasing on J, we may apply the strictly increasing inverse function g 1 to this inequality to get 18:2 1/24/2 TOPIC. Inequalities; measures of spread. This lecture explores the implications of Jensen s inequality for g-means in general, and for harmonic, geometric, arithmetic, and related means in

More information

MAT 570 REAL ANALYSIS LECTURE NOTES. Contents. 1. Sets Functions Countability Axiom of choice Equivalence relations 9

MAT 570 REAL ANALYSIS LECTURE NOTES. Contents. 1. Sets Functions Countability Axiom of choice Equivalence relations 9 MAT 570 REAL ANALYSIS LECTURE NOTES PROFESSOR: JOHN QUIGG SEMESTER: FALL 204 Contents. Sets 2 2. Functions 5 3. Countability 7 4. Axiom of choice 8 5. Equivalence relations 9 6. Real numbers 9 7. Extended

More information

Minimizing a convex separable exponential function subject to linear equality constraint and bounded variables

Minimizing a convex separable exponential function subject to linear equality constraint and bounded variables Minimizing a convex separale exponential function suect to linear equality constraint and ounded variales Stefan M. Stefanov Department of Mathematics Neofit Rilski South-Western University 2700 Blagoevgrad

More information

Lecture 1 Measure concentration

Lecture 1 Measure concentration CSE 29: Learning Theory Fall 2006 Lecture Measure concentration Lecturer: Sanjoy Dasgupta Scribe: Nakul Verma, Aaron Arvey, and Paul Ruvolo. Concentration of measure: examples We start with some examples

More information

A COMPOUND POISSON APPROXIMATION INEQUALITY

A COMPOUND POISSON APPROXIMATION INEQUALITY J. Appl. Prob. 43, 282 288 (2006) Printed in Israel Applied Probability Trust 2006 A COMPOUND POISSON APPROXIMATION INEQUALITY EROL A. PEKÖZ, Boston University Abstract We give conditions under which the

More information

More on Distribution Function

More on Distribution Function More on Distribution Function The distribution of a random variable X can be determined directly from its cumulative distribution function F X. Theorem: Let X be any random variable, with cumulative distribution

More information

arxiv: v2 [math.pr] 15 Nov 2016

arxiv: v2 [math.pr] 15 Nov 2016 STEIN S METHOD, MANY INTERACTING WORLDS AND QUANTUM MECHANICS Ian W. McKeague, Erol Peköz, and Yvik Swan Columbia University, Boston University, and Université de Liège arxiv:606.0668v2 [math.pr] 5 Nov

More information

Generalized Geometric Series, The Ratio Comparison Test and Raabe s Test

Generalized Geometric Series, The Ratio Comparison Test and Raabe s Test Generalized Geometric Series The Ratio Comparison Test and Raae s Test William M. Faucette Decemer 2003 The goal of this paper is to examine the convergence of a type of infinite series in which the summands

More information

#A50 INTEGERS 14 (2014) ON RATS SEQUENCES IN GENERAL BASES

#A50 INTEGERS 14 (2014) ON RATS SEQUENCES IN GENERAL BASES #A50 INTEGERS 14 (014) ON RATS SEQUENCES IN GENERAL BASES Johann Thiel Dept. of Mathematics, New York City College of Technology, Brooklyn, New York jthiel@citytech.cuny.edu Received: 6/11/13, Revised:

More information

X n D X lim n F n (x) = F (x) for all x C F. lim n F n(u) = F (u) for all u C F. (2)

X n D X lim n F n (x) = F (x) for all x C F. lim n F n(u) = F (u) for all u C F. (2) 14:17 11/16/2 TOPIC. Convergence in distribution and related notions. This section studies the notion of the so-called convergence in distribution of real random variables. This is the kind of convergence

More information

Conditional densities, mass functions, and expectations

Conditional densities, mass functions, and expectations Conditional densities, mass functions, and expectations Jason Swanson April 22, 27 1 Discrete random variables Suppose that X is a discrete random variable with range {x 1, x 2, x 3,...}, and that Y is

More information

ADVANCED PROBABILITY: SOLUTIONS TO SHEET 1

ADVANCED PROBABILITY: SOLUTIONS TO SHEET 1 ADVANCED PROBABILITY: SOLUTIONS TO SHEET 1 Last compiled: November 6, 213 1. Conditional expectation Exercise 1.1. To start with, note that P(X Y = P( c R : X > c, Y c or X c, Y > c = P( c Q : X > c, Y

More information

The Stein and Chen-Stein methods for functionals of non-symmetric Bernoulli processes

The Stein and Chen-Stein methods for functionals of non-symmetric Bernoulli processes The Stein and Chen-Stein methods for functionals of non-symmetric Bernoulli processes Nicolas Privault Giovanni Luca Torrisi Abstract Based on a new multiplication formula for discrete multiple stochastic

More information

THE INVERSE FUNCTION THEOREM

THE INVERSE FUNCTION THEOREM THE INVERSE FUNCTION THEOREM W. PATRICK HOOPER The implicit function theorem is the following result: Theorem 1. Let f be a C 1 function from a neighborhood of a point a R n into R n. Suppose A = Df(a)

More information

An introduction to some aspects of functional analysis

An introduction to some aspects of functional analysis An introduction to some aspects of functional analysis Stephen Semmes Rice University Abstract These informal notes deal with some very basic objects in functional analysis, including norms and seminorms

More information

Weak bidders prefer first-price (sealed-bid) auctions. (This holds both ex-ante, and once the bidders have learned their types)

Weak bidders prefer first-price (sealed-bid) auctions. (This holds both ex-ante, and once the bidders have learned their types) Econ 805 Advanced Micro Theory I Dan Quint Fall 2007 Lecture 9 Oct 4 2007 Last week, we egan relaxing the assumptions of the symmetric independent private values model. We examined private-value auctions

More information

Math Stochastic Processes & Simulation. Davar Khoshnevisan University of Utah

Math Stochastic Processes & Simulation. Davar Khoshnevisan University of Utah Math 5040 1 Stochastic Processes & Simulation Davar Khoshnevisan University of Utah Module 1 Generation of Discrete Random Variables Just about every programming language and environment has a randomnumber

More information

0.1 Uniform integrability

0.1 Uniform integrability Copyright c 2009 by Karl Sigman 0.1 Uniform integrability Given a sequence of rvs {X n } for which it is known apriori that X n X, n, wp1. for some r.v. X, it is of great importance in many applications

More information

SOME CONVERSE LIMIT THEOREMS FOR EXCHANGEABLE BOOTSTRAPS

SOME CONVERSE LIMIT THEOREMS FOR EXCHANGEABLE BOOTSTRAPS SOME CONVERSE LIMIT THEOREMS OR EXCHANGEABLE BOOTSTRAPS Jon A. Wellner University of Washington The bootstrap Glivenko-Cantelli and bootstrap Donsker theorems of Giné and Zinn (990) contain both necessary

More information

2 Two-Point Boundary Value Problems

2 Two-Point Boundary Value Problems 2 Two-Point Boundary Value Problems Another fundamental equation, in addition to the heat eq. and the wave eq., is Poisson s equation: n j=1 2 u x 2 j The unknown is the function u = u(x 1, x 2,..., x

More information

Math 328 Course Notes

Math 328 Course Notes Math 328 Course Notes Ian Robertson March 3, 2006 3 Properties of C[0, 1]: Sup-norm and Completeness In this chapter we are going to examine the vector space of all continuous functions defined on the

More information

Lecture Notes on Metric Spaces

Lecture Notes on Metric Spaces Lecture Notes on Metric Spaces Math 117: Summer 2007 John Douglas Moore Our goal of these notes is to explain a few facts regarding metric spaces not included in the first few chapters of the text [1],

More information

MOMENT CONVERGENCE RATES OF LIL FOR NEGATIVELY ASSOCIATED SEQUENCES

MOMENT CONVERGENCE RATES OF LIL FOR NEGATIVELY ASSOCIATED SEQUENCES J. Korean Math. Soc. 47 1, No., pp. 63 75 DOI 1.4134/JKMS.1.47..63 MOMENT CONVERGENCE RATES OF LIL FOR NEGATIVELY ASSOCIATED SEQUENCES Ke-Ang Fu Li-Hua Hu Abstract. Let X n ; n 1 be a strictly stationary

More information

Functional Analysis. Franck Sueur Metric spaces Definitions Completeness Compactness Separability...

Functional Analysis. Franck Sueur Metric spaces Definitions Completeness Compactness Separability... Functional Analysis Franck Sueur 2018-2019 Contents 1 Metric spaces 1 1.1 Definitions........................................ 1 1.2 Completeness...................................... 3 1.3 Compactness......................................

More information

4 Sums of Independent Random Variables

4 Sums of Independent Random Variables 4 Sums of Independent Random Variables Standing Assumptions: Assume throughout this section that (,F,P) is a fixed probability space and that X 1, X 2, X 3,... are independent real-valued random variables

More information

arxiv: v2 [math.pr] 19 Oct 2010

arxiv: v2 [math.pr] 19 Oct 2010 The Annals of Probability 2010, Vol. 38, No. 4, 1672 1689 DOI: 10.1214/10-AOP527 c Institute of Mathematical tatistics, 2010 arxiv:0906.5145v2 [math.pr] 19 Oct 2010 BOUND ON THE CONTANT IN THE MEAN CENTRAL

More information

3 Integration and Expectation

3 Integration and Expectation 3 Integration and Expectation 3.1 Construction of the Lebesgue Integral Let (, F, µ) be a measure space (not necessarily a probability space). Our objective will be to define the Lebesgue integral R fdµ

More information

1 Hoeffding s Inequality

1 Hoeffding s Inequality Proailistic Method: Hoeffding s Inequality and Differential Privacy Lecturer: Huert Chan Date: 27 May 22 Hoeffding s Inequality. Approximate Counting y Random Sampling Suppose there is a ag containing

More information

WEYL-HEISENBERG FRAMES FOR SUBSPACES OF L 2 (R)

WEYL-HEISENBERG FRAMES FOR SUBSPACES OF L 2 (R) PROCEEDINGS OF THE AMERICAN MATHEMATICAL SOCIETY Volume 129, Numer 1, Pages 145 154 S 0002-9939(00)05731-2 Article electronically pulished on July 27, 2000 WEYL-HEISENBERG FRAMES FOR SUBSPACES OF L 2 (R)

More information

Kolmogorov Berry-Esseen bounds for binomial functionals

Kolmogorov Berry-Esseen bounds for binomial functionals Kolmogorov Berry-Esseen bounds for binomial functionals Raphaël Lachièze-Rey, Univ. South California, Univ. Paris 5 René Descartes, Joint work with Giovanni Peccati, University of Luxembourg, Singapore

More information

Expansion formula using properties of dot product (analogous to FOIL in algebra): u v 2 u v u v u u 2u v v v u 2 2u v v 2

Expansion formula using properties of dot product (analogous to FOIL in algebra): u v 2 u v u v u u 2u v v v u 2 2u v v 2 Least squares: Mathematical theory Below we provide the "vector space" formulation, and solution, of the least squares prolem. While not strictly necessary until we ring in the machinery of matrix algera,

More information

PCMI Introduction to Random Matrix Theory Handout # REVIEW OF PROBABILITY THEORY. Chapter 1 - Events and Their Probabilities

PCMI Introduction to Random Matrix Theory Handout # REVIEW OF PROBABILITY THEORY. Chapter 1 - Events and Their Probabilities PCMI 207 - Introduction to Random Matrix Theory Handout #2 06.27.207 REVIEW OF PROBABILITY THEORY Chapter - Events and Their Probabilities.. Events as Sets Definition (σ-field). A collection F of subsets

More information

Part IA Probability. Theorems. Based on lectures by R. Weber Notes taken by Dexter Chua. Lent 2015

Part IA Probability. Theorems. Based on lectures by R. Weber Notes taken by Dexter Chua. Lent 2015 Part IA Probability Theorems Based on lectures by R. Weber Notes taken by Dexter Chua Lent 2015 These notes are not endorsed by the lecturers, and I have modified them (often significantly) after lectures.

More information

Problem set 2 The central limit theorem.

Problem set 2 The central limit theorem. Problem set 2 The central limit theorem. Math 22a September 6, 204 Due Sept. 23 The purpose of this problem set is to walk through the proof of the central limit theorem of probability theory. Roughly

More information

Functional Analysis I

Functional Analysis I Functional Analysis I Course Notes by Stefan Richter Transcribed and Annotated by Gregory Zitelli Polar Decomposition Definition. An operator W B(H) is called a partial isometry if W x = X for all x (ker

More information

STEIN S METHOD, SEMICIRCLE DISTRIBUTION, AND REDUCED DECOMPOSITIONS OF THE LONGEST ELEMENT IN THE SYMMETRIC GROUP

STEIN S METHOD, SEMICIRCLE DISTRIBUTION, AND REDUCED DECOMPOSITIONS OF THE LONGEST ELEMENT IN THE SYMMETRIC GROUP STIN S MTHOD, SMICIRCL DISTRIBUTION, AND RDUCD DCOMPOSITIONS OF TH LONGST LMNT IN TH SYMMTRIC GROUP JASON FULMAN AND LARRY GOLDSTIN Abstract. Consider a uniformly chosen random reduced decomposition of

More information

Math The Laplacian. 1 Green s Identities, Fundamental Solution

Math The Laplacian. 1 Green s Identities, Fundamental Solution Math. 209 The Laplacian Green s Identities, Fundamental Solution Let be a bounded open set in R n, n 2, with smooth boundary. The fact that the boundary is smooth means that at each point x the external

More information

Review of Probability Theory

Review of Probability Theory Review of Probability Theory Arian Maleki and Tom Do Stanford University Probability theory is the study of uncertainty Through this class, we will be relying on concepts from probability theory for deriving

More information

Asymptotic Statistics-III. Changliang Zou

Asymptotic Statistics-III. Changliang Zou Asymptotic Statistics-III Changliang Zou The multivariate central limit theorem Theorem (Multivariate CLT for iid case) Let X i be iid random p-vectors with mean µ and and covariance matrix Σ. Then n (

More information

Measure and Integration: Solutions of CW2

Measure and Integration: Solutions of CW2 Measure and Integration: s of CW2 Fall 206 [G. Holzegel] December 9, 206 Problem of Sheet 5 a) Left (f n ) and (g n ) be sequences of integrable functions with f n (x) f (x) and g n (x) g (x) for almost

More information

MATH 51H Section 4. October 16, Recall what it means for a function between metric spaces to be continuous:

MATH 51H Section 4. October 16, Recall what it means for a function between metric spaces to be continuous: MATH 51H Section 4 October 16, 2015 1 Continuity Recall what it means for a function between metric spaces to be continuous: Definition. Let (X, d X ), (Y, d Y ) be metric spaces. A function f : X Y is

More information

Notions such as convergent sequence and Cauchy sequence make sense for any metric space. Convergent Sequences are Cauchy

Notions such as convergent sequence and Cauchy sequence make sense for any metric space. Convergent Sequences are Cauchy Banach Spaces These notes provide an introduction to Banach spaces, which are complete normed vector spaces. For the purposes of these notes, all vector spaces are assumed to be over the real numbers.

More information

MAS223 Statistical Inference and Modelling Exercises

MAS223 Statistical Inference and Modelling Exercises MAS223 Statistical Inference and Modelling Exercises The exercises are grouped into sections, corresponding to chapters of the lecture notes Within each section exercises are divided into warm-up questions,

More information

Essential Maths 1. Macquarie University MAFC_Essential_Maths Page 1 of These notes were prepared by Anne Cooper and Catriona March.

Essential Maths 1. Macquarie University MAFC_Essential_Maths Page 1 of These notes were prepared by Anne Cooper and Catriona March. Essential Maths 1 The information in this document is the minimum assumed knowledge for students undertaking the Macquarie University Masters of Applied Finance, Graduate Diploma of Applied Finance, and

More information

1 Review of di erential calculus

1 Review of di erential calculus Review of di erential calculus This chapter presents the main elements of di erential calculus needed in probability theory. Often, students taking a course on probability theory have problems with concepts

More information

Notes 6 : First and second moment methods

Notes 6 : First and second moment methods Notes 6 : First and second moment methods Math 733-734: Theory of Probability Lecturer: Sebastien Roch References: [Roc, Sections 2.1-2.3]. Recall: THM 6.1 (Markov s inequality) Let X be a non-negative

More information

Topology. Xiaolong Han. Department of Mathematics, California State University, Northridge, CA 91330, USA address:

Topology. Xiaolong Han. Department of Mathematics, California State University, Northridge, CA 91330, USA  address: Topology Xiaolong Han Department of Mathematics, California State University, Northridge, CA 91330, USA E-mail address: Xiaolong.Han@csun.edu Remark. You are entitled to a reward of 1 point toward a homework

More information

6.1 Moment Generating and Characteristic Functions

6.1 Moment Generating and Characteristic Functions Chapter 6 Limit Theorems The power statistics can mostly be seen when there is a large collection of data points and we are interested in understanding the macro state of the system, e.g., the average,

More information

L p Functions. Given a measure space (X, µ) and a real number p [1, ), recall that the L p -norm of a measurable function f : X R is defined by

L p Functions. Given a measure space (X, µ) and a real number p [1, ), recall that the L p -norm of a measurable function f : X R is defined by L p Functions Given a measure space (, µ) and a real number p [, ), recall that the L p -norm of a measurable function f : R is defined by f p = ( ) /p f p dµ Note that the L p -norm of a function f may

More information

1 Presessional Probability

1 Presessional Probability 1 Presessional Probability Probability theory is essential for the development of mathematical models in finance, because of the randomness nature of price fluctuations in the markets. This presessional

More information

Stat 5101 Notes: Algorithms

Stat 5101 Notes: Algorithms Stat 5101 Notes: Algorithms Charles J. Geyer January 22, 2016 Contents 1 Calculating an Expectation or a Probability 3 1.1 From a PMF........................... 3 1.2 From a PDF...........................

More information

Stat 8112 Lecture Notes Weak Convergence in Metric Spaces Charles J. Geyer January 23, Metric Spaces

Stat 8112 Lecture Notes Weak Convergence in Metric Spaces Charles J. Geyer January 23, Metric Spaces Stat 8112 Lecture Notes Weak Convergence in Metric Spaces Charles J. Geyer January 23, 2013 1 Metric Spaces Let X be an arbitrary set. A function d : X X R is called a metric if it satisfies the folloing

More information

Lecture 3 - Expectation, inequalities and laws of large numbers

Lecture 3 - Expectation, inequalities and laws of large numbers Lecture 3 - Expectation, inequalities and laws of large numbers Jan Bouda FI MU April 19, 2009 Jan Bouda (FI MU) Lecture 3 - Expectation, inequalities and laws of large numbersapril 19, 2009 1 / 67 Part

More information

Real option valuation for reserve capacity

Real option valuation for reserve capacity Real option valuation for reserve capacity MORIARTY, JM; Palczewski, J doi:10.1016/j.ejor.2016.07.003 For additional information aout this pulication click this link. http://qmro.qmul.ac.uk/xmlui/handle/123456789/13838

More information

d(x n, x) d(x n, x nk ) + d(x nk, x) where we chose any fixed k > N

d(x n, x) d(x n, x nk ) + d(x nk, x) where we chose any fixed k > N Problem 1. Let f : A R R have the property that for every x A, there exists ɛ > 0 such that f(t) > ɛ if t (x ɛ, x + ɛ) A. If the set A is compact, prove there exists c > 0 such that f(x) > c for all x

More information

GLUING LEMMAS AND SKOROHOD REPRESENTATIONS

GLUING LEMMAS AND SKOROHOD REPRESENTATIONS GLUING LEMMAS AND SKOROHOD REPRESENTATIONS PATRIZIA BERTI, LUCA PRATELLI, AND PIETRO RIGO Abstract. Let X, E), Y, F) and Z, G) be measurable spaces. Suppose we are given two probability measures γ and

More information

RELATING TIME AND CUSTOMER AVERAGES FOR QUEUES USING FORWARD COUPLING FROM THE PAST

RELATING TIME AND CUSTOMER AVERAGES FOR QUEUES USING FORWARD COUPLING FROM THE PAST J. Appl. Prob. 45, 568 574 (28) Printed in England Applied Probability Trust 28 RELATING TIME AND CUSTOMER AVERAGES FOR QUEUES USING FORWARD COUPLING FROM THE PAST EROL A. PEKÖZ, Boston University SHELDON

More information

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 8 10/1/2008 CONTINUOUS RANDOM VARIABLES

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 8 10/1/2008 CONTINUOUS RANDOM VARIABLES MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 8 10/1/2008 CONTINUOUS RANDOM VARIABLES Contents 1. Continuous random variables 2. Examples 3. Expected values 4. Joint distributions

More information

Statistics 100A Homework 5 Solutions

Statistics 100A Homework 5 Solutions Chapter 5 Statistics 1A Homework 5 Solutions Ryan Rosario 1. Let X be a random variable with probability density function a What is the value of c? fx { c1 x 1 < x < 1 otherwise We know that for fx to

More information

Estimating a Finite Population Mean under Random Non-Response in Two Stage Cluster Sampling with Replacement

Estimating a Finite Population Mean under Random Non-Response in Two Stage Cluster Sampling with Replacement Open Journal of Statistics, 07, 7, 834-848 http://www.scirp.org/journal/ojs ISS Online: 6-798 ISS Print: 6-78X Estimating a Finite Population ean under Random on-response in Two Stage Cluster Sampling

More information

1 Topology Definition of a topology Basis (Base) of a topology The subspace topology & the product topology on X Y 3

1 Topology Definition of a topology Basis (Base) of a topology The subspace topology & the product topology on X Y 3 Index Page 1 Topology 2 1.1 Definition of a topology 2 1.2 Basis (Base) of a topology 2 1.3 The subspace topology & the product topology on X Y 3 1.4 Basic topology concepts: limit points, closed sets,

More information

Laplace s Equation. Chapter Mean Value Formulas

Laplace s Equation. Chapter Mean Value Formulas Chapter 1 Laplace s Equation Let be an open set in R n. A function u C 2 () is called harmonic in if it satisfies Laplace s equation n (1.1) u := D ii u = 0 in. i=1 A function u C 2 () is called subharmonic

More information

On the Third Order Rational Difference Equation

On the Third Order Rational Difference Equation Int. J. Contemp. Math. Sciences, Vol. 4, 2009, no. 27, 32-334 On the Third Order Rational Difference Equation x n x n 2 x (a+x n x n 2 ) T. F. Irahim Department of Mathematics, Faculty of Science Mansoura

More information

Some Background Material

Some Background Material Chapter 1 Some Background Material In the first chapter, we present a quick review of elementary - but important - material as a way of dipping our toes in the water. This chapter also introduces important

More information

NEW FUNCTIONAL INEQUALITIES

NEW FUNCTIONAL INEQUALITIES 1 / 29 NEW FUNCTIONAL INEQUALITIES VIA STEIN S METHOD Giovanni Peccati (Luxembourg University) IMA, Minneapolis: April 28, 2015 2 / 29 INTRODUCTION Based on two joint works: (1) Nourdin, Peccati and Swan

More information

Lecture Notes in Advanced Calculus 1 (80315) Raz Kupferman Institute of Mathematics The Hebrew University

Lecture Notes in Advanced Calculus 1 (80315) Raz Kupferman Institute of Mathematics The Hebrew University Lecture Notes in Advanced Calculus 1 (80315) Raz Kupferman Institute of Mathematics The Hebrew University February 7, 2007 2 Contents 1 Metric Spaces 1 1.1 Basic definitions...........................

More information

Differential Stein operators for multivariate continuous distributions and applications

Differential Stein operators for multivariate continuous distributions and applications Differential Stein operators for multivariate continuous distributions and applications Gesine Reinert A French/American Collaborative Colloquium on Concentration Inequalities, High Dimensional Statistics

More information

ADJOINTS, ABSOLUTE VALUES AND POLAR DECOMPOSITIONS

ADJOINTS, ABSOLUTE VALUES AND POLAR DECOMPOSITIONS J. OPERATOR THEORY 44(2000), 243 254 c Copyright by Theta, 2000 ADJOINTS, ABSOLUTE VALUES AND POLAR DECOMPOSITIONS DOUGLAS BRIDGES, FRED RICHMAN and PETER SCHUSTER Communicated by William B. Arveson Abstract.

More information

or E ( U(X) ) e zx = e ux e ivx = e ux( cos(vx) + i sin(vx) ), B X := { u R : M X (u) < } (4)

or E ( U(X) ) e zx = e ux e ivx = e ux( cos(vx) + i sin(vx) ), B X := { u R : M X (u) < } (4) :23 /4/2000 TOPIC Characteristic functions This lecture begins our study of the characteristic function φ X (t) := Ee itx = E cos(tx)+ie sin(tx) (t R) of a real random variable X Characteristic functions

More information

Metric spaces and metrizability

Metric spaces and metrizability 1 Motivation Metric spaces and metrizability By this point in the course, this section should not need much in the way of motivation. From the very beginning, we have talked about R n usual and how relatively

More information

Part II Probability and Measure

Part II Probability and Measure Part II Probability and Measure Theorems Based on lectures by J. Miller Notes taken by Dexter Chua Michaelmas 2016 These notes are not endorsed by the lecturers, and I have modified them (often significantly)

More information

1.4 The Jacobian of a map

1.4 The Jacobian of a map 1.4 The Jacobian of a map Derivative of a differentiable map Let F : M n N m be a differentiable map between two C 1 manifolds. Given a point p M we define the derivative of F at p by df p df (p) : T p

More information

Exercises Measure Theoretic Probability

Exercises Measure Theoretic Probability Exercises Measure Theoretic Probability Chapter 1 1. Prove the folloing statements. (a) The intersection of an arbitrary family of d-systems is again a d- system. (b) The intersection of an arbitrary family

More information

Chapter 5. Random Variables (Continuous Case) 5.1 Basic definitions

Chapter 5. Random Variables (Continuous Case) 5.1 Basic definitions Chapter 5 andom Variables (Continuous Case) So far, we have purposely limited our consideration to random variables whose ranges are countable, or discrete. The reason for that is that distributions on

More information

17. Convergence of Random Variables

17. Convergence of Random Variables 7. Convergence of Random Variables In elementary mathematics courses (such as Calculus) one speaks of the convergence of functions: f n : R R, then lim f n = f if lim f n (x) = f(x) for all x in R. This

More information