Module Code MAT00018M MMath and MSc Examinations 2019-20 Department: Mathematics Title of Exam: Stochastic Processes Time Allowed: 2 hours Allocation of Marks: The marking scheme shown on each question is indicative only. Question: 1 2 3 4 Total Marks: 28 23 16 33 100 Instructions for Candidates: Answer all four questions. Please write your answers in black or blue ink; pencil is acceptable for graphs and diagrams. Do not use red ink. You should simplify your answers as far as possible, but not resort to numerical approximations. No calculators may be used. Materials Supplied: Green booklet. Do not write on this booklet before the exam begins. Do not turn over this page until instructed to do so by an invigilator. Page 1 (of 6) MAT00018M 1 (of 4). (a) Give a definition of a Poisson process N = {N(t) : t ≥ 0} with rate λ. [4] (b) Write down the formula for the transition probabilities P (N(t) = j|N(s) = i) for any t ≥ s ≥ 0 and j ≥ i ≥ 0. [4] (c) Assuming that P (N(1) = j) = 3j j! e−3, j = 0, 1, 2, 3, . . . calculate, providing explanation for each step of your calculations, E[N(1)]. [6] Page 2 (of 6) continued on next page continued from previous page MAT00018M 1 (of 4) cont. (d) The photons emitted by an LED result from the recombination of electron- hole pairs. Assume that the number of photons is modelled using a Poisson process with rate of 10 photons per second. (i) What is the probability that exactly 8 photons have been emitted during the first 1 second? [2] (ii) What is the probability that at most 1 photon has been emitted during the first 1 second? [2] (iii) What is the probability that at least 1 photon has been emitted during the first 1 second? [2] (iv) What is the probability that exactly 2 photons arrive during the first 1 second and exactly 3 photons arrive during the first 3 seconds? [2] (v) It is known that exactly 4 photons have been emitted during the first 3 seconds. What is the probability that exactly 1 photon has been emitted in the first 1 second? [2] (vi) How many seconds would you expect to have to wait until the first photon has been emitted? [2] (vii) Alice was counting the emitted photons. After she has counted the first 1000 photons, she was replaced by her colleague Bob. What is the probability that exactly 5 photons have been counted by Bob during the first 2 seconds on his watch? [2] State explicitly any properties of the Poisson process used in your calcula- tions. [Total: 28] Page 3 (of 6) Turn over MAT00018M 2 (of 4). (a) Give a definition of a Brownian Motion W = {W (t) : t ≥ 0}. [5] (b) Assuming that W (t), for t > 0, has density pt(x) = 1√ 2pit e− x2 2t , x ∈ R, calculate the probability P (W (1) ≤ 0) . [5] (c) Assuming that W (t), for t > 0, has density pt(x) = 1√ 2pit e− x2 2t , x ∈ R, show that for t ≥ 0 the following equalities hold: E [ e4W (1/2) ] = e4, E [ (W (1))2 ] = 1, E [ (W (5))3e−(W (5)) 2] = 0. State without proof any result you are using. [13] [Total: 23] Page 4 (of 6) MAT00018M 3 (of 4). (a) Suppose that N = N(t), t ≥ 0 is a birth-death process with birth rates λ0, λ1, λ2, · · · ≥ 0 and death rates µ1, µ2, · · · ≥ 0. Write down the Master (or the forward Kolmogorov) equations for the tran- sition probabilities pij(t) = P ( N(t) = j|N(0) = i), t ≥ 0, where i, j are natural numbers such that j ≥ i ≥ 0. [4] (b) Assume that b, a > 0 are two fixed numbers. Consider a system consisting of K telephone lines. We assume that K ≥ 3. If n lines are busy, where n ≤ K, then with probability nbh, one of them will be freed within small time interval of length h > 0. If n lines are busy, where n ≤ K − 1, then with probability ah, a new call will arrive. The probability that within that time interval two or more conversations will terminate or two or more calls will arrive, is negligible. Let N(t), for t ≥ 0, be the number of occupied telephone lines. Assume that N = (N(t)), t ≥ 0, is a birth-death process. Find explicit expressions for the birth and death rates of the process N . [4] (c) Explain why the stationary distribution for the birth-death process from the previous part exists. Find the stationary distribution for the birth-death process from the previous part. Hint: You can use a general formula for the stationary distribution of a birth- death process without deriving it. [8] [Total: 16] Page 5 (of 6) Turn over MAT00018M 4 (of 4). Suppose that W = {W (t)}, t ≥ 0 is a Brownian Motion. (a) Define a step process and then define the Itoˆ integral for such a process. Let ξ = (ξ(t))t≥0 be a stochastic processes given by ξ(t) = 1 if 0 ≤ t ≤ 3, (W (2))2 if 3 < t ≤ 4, W (4) if 4 < t ≤ 5, 0 if t > 5. Explain why ξ is a step process. Evaluate the Itoˆ integral I(ξ) of ξ. Note that the Itoˆ integral I(ξ) is also denoted by ∫ 5 0 ξ(s) dW (s). Compute the mean value of this Itoˆ integral. [6] (b) State (without a proof) the Itoˆ isometry. Compute the variance of this Itoˆ integral I(ξ) from part (a). [4] (c) State the Itoˆ Lemma in the simple form. Assume that T > 0 is a given number. Using the Itoˆ Lemma show that∫ T 0 tW (t) dW (t) = 1 2 T (W (T ))2 − 1 2 ∫ T 0 (W (t))2 dt− 1 4 T 2. [6] Using the last result show that E[ ∫ T 0 tW (t) dW (t)] = 0. [3] (d) Using the Itoˆ Lemma or otherwise, find the solution to the following stochas- tic differential equation dX(t) = 3X(t)dt+ 4X(t) dW (t), (1) with initial condition X(0) = 1. [14] [Total: 33] Page 6 (of 6) End of examination. SOLUTIONS: MAT00018M 1. (a) A Poisson process with rate λ > 0 is a stochastic process N = {N(t) : t ≥ 0} taking values in S = {0, 1, . . . } such that: a) N(0) = 0; b) N has independent increments. c) P(N(t+ h) = k + i|N(t) = k) = λh+ o(h) if i = 1 o(h) if i > 1 1− λh+ o(h) if i = 0 0 if i < 0; for any small h > 0. 4 Marks (b) The transition probabilities for the Poisson process are given by P (N(t) = j|N(s) = i) = P (N(t− s) = j − i) = (λ(t− s))j−i (j − i)! e −λ(t−s). (2) 4 Marks (c) In the following λ = 3. E(N(1)) = ∞∑ k=0 xkpk = ∞∑ k=0 k λk k! e−λ = e−λ ∞∑ k=1 λk−1λ (k − 1)! = λe −λ ∞∑ k=1 λk−1 (k − 1)! = λe−λ ∞∑ n=0 λn n! (n = k − 1) = λe−λeλ = λ = 3. 6 Marks (d) We measure time in units equal to 1 second. The rate λ = 10. We put the numerical value of λ at the last moment. We do not use calculator to make calculations. Note that for s = 0 and i = 0 the formula from part (b) takes the following form P (N(t) = j) = (λ t)j j! e−λt. (3) (i) We use formula (3) with t = 1, λ = 10 and j = 8, so that tλ = 10. P(N(t) = j) = (tλ)j j! e−tλ = (tλ)8 8! e−tλ = 108 8! e−10 So P(N(1) = 8) = 108 8! e−10. 7 SOLUTIONS: MAT00018M 2 Marks (ii) We calculate, with t = 1 and λ = 10, P(N(t) ≤ 1) = P(N(t) = 0) + P(N(t) = 1) We begin with find the first term. For this aim we again use formula (3) with t = 1, λ = 10 (so that tλ = 10) but now with j = 0. We have P(N(t) = 0) = (tλ)0 0! e−tλ = (tλ)0 0! e−tλ = 1 1 e−10 = e−10 Next, we calculate the second term. By the same methods P(N(t) = 1) = (tλ)1 1! e−tλ = (tλ)1 1! e−tλ = λe−λ = 10 e−10 Hence, we infer that P(N(1) ≤ 1) = 10 e−10 + e−10 = 11 e−10 2 Marks (iii) We calculate, using properties of the probability measure and the result from the last part to get P(N(t) ≥ 1) = 1− P(N(t) < 1) = 1− P(N(t) = 0) = 1− e−tλ = 1− e−10, where the last but one equality follows from the previous part. 2 Marks (iv) Put s = 1 and i = 2, t = 3 and j = 3. Then we will have. The most important step is write the formula for the desired probability using the increments: P(N(s) = i, N(t) = j) = P(N(s) = i, N(t)−N(s) = j − i) Because the r.v.’s N(t)−N(s) and N(s) are independent = P(N(s) = i)P(N(t)−N(s) = j − i) by applying formula (3) twice = (sλ)i i! e−sλ ( (t− s)λ)j−i (j − i)! e −(t−s)λ = (λ)2 2! e−λ ( 2× λ)1 1! e−2×λ = λ3e−3λ = 103e−30 8 SOLUTIONS: MAT00018M So P(N(1) = 2 and N(3) = 3) = 103e−30 2 Marks (v) By the definition of the conditional expectation we have, with s = 1, t = 3, i = 1 and j = 4, we have P(N(1) = 1|N(3) = 4) = P(N(1) = 1, N(3) = 4) P(N(3) = 4) = P(N(1) = 1, N(3)−N(1) = 4− 1) P(N(3) = 4) by the independe of r.v.’s N(3)−N(1) and N(1) = P(N(1) = 1)P(N(3)−N(1) = 4− 1) P(N(3) = 4) = (1×λ)1 1! e−1×λ × [(3−1)×λ]4−1 (4−1)! e −(3−1)×λ (3×λ)4 4! e−3×λ = λe−λ × [2λ]3 3! e−2λ (3×λ)4 4! e−3×λ since the exponentials cancel = λ × [2λ]3 3! (3×λ)4 4! = 23 × 4!× λ4 34 × 3!× λ4 since the λ4 cancel = 25 34 = 32 81 2 Marks (vi) We define τ1 = the smallest time t : N(t) = 1 This is a stopping time and we know from Lectures, Theorem 3.5, that τ1 has an exponential distribution with parameter λ = 10. Hence E(τ1) = 1 λ = 1 10 . 2 Marks (vii) We put M(t) = N(t+ τ1)−N(τ1), t ≥ 0. By Theorem 3.4, M = (M(t)), t ≥ 0 is also a Poisson process with the same rate λ = 10 as N . Therefore, by formula (3) with t = 2, λ = 10 and j = 5 we get 9 SOLUTIONS: MAT00018M P(M(t) = 5) = (tλ)5 5! e−tλ = (20)5 5! e−20. 2 Marks Total: 28 Marks 2. (a) A Brownian Motion W = {W (t) : t ≥ 0} is a stochastic process such that (i) W (0) = 0 a.s.; (ii) For any s ≥ 0, t > 0,W (t+ s)−W (s) ∼ N(0, t); (iii) W has independent increments; (iv) The sample paths of W are continuous a.s.. 5 Marks (b) We have P (W (1) ≤ 0) = P (W (1) ∈ (−∞, 0]) = ∫ 0 −∞ p1(x)dx = ∫ 0 −∞ 1√ 2pi exp ( −x 2 2 ) dx = 1 2 [ ∫ 0 −∞ 1√ 2pi exp ( −x 2 2 dx ) + ∫ ∞ 0 1√ 2pi exp ( −x 2 2 dx )] = 1 2 ∫ ∞ −∞ 1√ 2pi exp ( −x 2 2 ) dx = 1 2 . We used the fact that the function f : x 7→ 1√ 2pit exp ( −x2 2t ) is even so that ∫ 0 −∞ f(x) dx = ∫ ∞ 0 f(x) dx. The last equality was stated in the Lectures. 5 Marks (c) We start with the most difficult part. Since the r.v. W (t), has density pt(x) = 1√ 2pit e− x2 2t , x ∈ R, we get, by the change of measure Theorem, 10 SOLUTIONS: MAT00018M E [ e4W (t) ] = ∫ ∞ −∞ e4xpt(x) dx = ∫ ∞ −∞ e4x 1√ 2pit e− x2 2t dx = 1√ 2pit ∫ ∞ −∞ e4x− x2 2t dx Now we have, for any λ ∈ R, 1√ 2pit ∫ ∞ −∞ eλx− x2 2t dx = . . . λx− x 2 2t = − 1 2t [ x2 − 2tλx+ (tλ)2 − (tλ)2 ] = − 1 2t [( x− λt)2 − (tλ)2] = −(x− tλ) 2 2t + tλ2 2 . . . = e tλ2 2 1√ 2pit ∫ +∞ −∞ e− (x−tλ)2 2t dx, y = x− tλ√ t , dx = √ tdy = e tλ2 2 ( 1√ 2pi ∫ +∞ −∞ e− x2 2 dx ) = e tλ2 2 . Putting λ = 4 and t = 1/2 we deduce that E [ e4W (1/2) ] = e 1/2×42 2 = e4. A solution to the last case is based on the same principle that an integral∫∞ −∞ f(x) dx is equal to 0 when f is an odd function. As above, by the change of measure Theorem, E [ (W (t))3e−(W (t)) 2] = ∫ ∞ −∞ x3e−x 2 pt(x) dx = ∫ ∞ −∞ x3e−x 2 1√ 2pit e− x2 2t dx = 0 because the function f(x) = x3e−x 2 e− x2 2t is odd. The second case is the most difficult. Again by the change of measure Theorem with g(x) = x2 to conclude that E [ W (1)2 ] = E [ g(W (1)) ] = ∫ +∞ −∞ g(x)p1(x) dx = ∫ +∞ −∞ x2p1(x) dx = ∫ +∞ −∞ x2√ 2pi e− x2 2 dx = 1√ 2pi ∫ +∞ −∞ x2e− x2 2 dx. 11 SOLUTIONS: MAT00018M By the integration by parts formula we have we have∫ +∞ −∞ x2e− x2 2 dx = lim n→∞ ∫ n −n x2e− x2 2 dx = lim n→∞ ∫ n −n (−x)(−xe−x 2 2 )dx = lim n→∞ ∫ n −n (−x)(e−x 2 2 )′dx = lim n→∞ [ − xe−x 2 2 |n−n − ∫ n −n −e−x 2 2 dx ] = lim n→∞ 2n e n2 2 + ∫ +∞ −∞ e− x2 2 dx = ∫ +∞ −∞ e− x2 2 dx, (4) where lim n→∞ 2n e n2 2 = 0 by l’Hospital rule. Therefore we infer that E [ W (1)2 ] = 1√ 2pi ∫ +∞ −∞ e− x2 2 dx = 1. 13 Marks Total: 23 Marks 3. (a) Recall that birth-death process with birth rates λ0, λ1, . . . and death rates µ1, µ2, . . . , where λi ≥ 0 and µi ≥ 0, is a Markov process N = {N(t) : t ≥ 0} with state space N such that for any t ≥ 0 and any small h ≥ 0, P(N(t+ h) = n+m|N(t) = n) = λnh+ o(h) if m = 1 µnh+ o(h) if m = −1 o(h) if m > 1 1− (λn + µn)h+ o(h) if m = 0 0 if |m| ≥ 2. The master equation expresses the rate of change in the probability of being in a state n as the difference between the rate at which the system enters the state and the rate at which it leaves the state. In the case of a birth-death process this gives d dt pij(t) = λj−1pi,j−1(t) + µj+1pi,j+1(t)− (λj + µj)pij(t). (5) where we use the convention from the lectures that λ−1 = µ0 = p−1 = 0. 4 Marks 12 SOLUTIONS: MAT00018M (b) In this problem, the births are represented by a new call arriving and the deaths are represented by a line being freed. Thus the birth rates are λn = { a, for n = 0, 1, 2, · · · , K 0, for n = K,K + 1, · · · . and the death rates are µn = { µb, for n = 0, 1, 2, · · · , K 0, for n = K,K + 1, · · · . 4 Marks (c) The stationary distribution for this birth-death process exists because ∞∑ n=1 λ0λ1 · · ·λn−1 µ1µ2 · · ·µn = K∑ n=1 λ0λ1 · · ·λn−1 µ1µ2 · · ·µn <∞ (6) In our case, λn−1 µn = a nb Hence, K∑ n=1 λ0λ1 · · ·λn−1 µ1µ2 · · ·µn = K∑ n=1 a b · · · a nb = K∑ n=1 ( a b )n 1 n! The stationary distribution exists and can be calculated as follows. pi0 = ( 1 + K∑ n=1 ( a b )n 1 n! )−1 pin = λ0λ1 · · ·λn−1 µ1µ2 · · ·µn pi0 = 1 n! (a b )n( 1 + K∑ n=1 ( a b )n 1 n! )−1 , 1 ≤ n ≤ K. (7) 8 Marks Total: 16 Marks 4. (a) Definition Let (Ω,F ,P) be a probability space and let W = W (t)t≥0 be a Brownian Motion w.r.t a filtration (Ft)t≥0. A step process is a process (ξ(t))t≥0 satisfying the following conditions (i) (ξ(t))t≥0 is adapted to (Ft)t≥0, 13 SOLUTIONS: MAT00018M (ii) E|ξ(t)|2 <∞ for all t ≥ 0, (iii) there exists a finite sequence (t0, . . . , tN) such that 0 = t0 < t1 < · · · < tN = T and ξ(t) = { ξ(ti), if t ∈ [ti, ti+1) for i = 0, . . . , N − 1 0, if t ≥ tN . For such a step process ξ = (ξ(t))t≥0 we define its Itoˆ integral by I(ξ) = N−1∑ i=0 ξ(ti)(W (ti+1)−W (ti)). (8) The stochastic process ξ is a simple process as well, with N = 3 and t0 = 0 < t1 = 3 < t2 = 4 < t3 = 5. Then by formula (8) we have Y = I(ξ) = N−1∑ i=0 ξ(ti)(W (ti+1)−W (ti)) = 3−1∑ i=0 ξ(ti)(W (ti+1)−W (ti)) = 1(W (3)−W (0)) + (W (2))2(W (4)−W (3)) +W (4)(W (5)−W (4)) By the properties of the Itoˆ integral E [ I(ξ) ] = 0. Alternatively, one can calculate E [ I(ξ) ] = E [ 1(W (3)−W (0)) + (W (2))2(W (4)−W (3)) + W (4)(W (5)−W (4)) ] = · · · = 0. Remarks. Bookwork and Unseen 6 Marks (b) Theorem [Itoˆ isometry for step processes)] For a step process ξ = (ξ(t))t≥0 we have E ( [I(ξ)]2 ) = E([ ∫ ∞ 0 ξ(t)dW (t)]2) = E( ∫ ∞ 0 |ξ(s)|2dt). (9) We use the Itoˆ isometry to calculate E(Y 2). E(Y 2) = E(|I(ξ)|2) = ∫ ∞ 0 E|ξ(t)|2 dt = ∫ 5 0 E|ξ(t)|2 dt+ ∫ ∞ 5 E|ξ(t)|2 dt = ∫ 5 0 E|ξ(t)|2 dt. 14 SOLUTIONS: MAT00018M Therefore, E(Y 2) = ∫ 5 0 E|ξ(t)|2 dt = 3−1∑ i=0 E(|ξti+|2)(ti+1 − ti) = E(1)2(3− 0) + E((W (2))2)2(4− 3) + E(W (4))2(5− 4) = 1 · 3 + 3 · 4 · 1 + 4 · 1 = 19. Remarks. Bookwork and Unseen 4 Marks (c) Theorem (The simple version of the) Itoˆ Lemma Assume that f(t, x) is of C1,2 class. Define η(t) = f(t,W (t)), t ≥ 0. Then η(t) is an Itoˆ process and for t ≥ 0, one has η(t) = f(0, 0) + ∫ t 0 [∂f ∂s (s,W (s)) + 1 2 ∂2f ∂x2 (s,W (s)) ] ds + ∫ t 0 ∂f ∂x (s,W (s)) dW (s). We apply the above result for a function f(t, x) = t 1 2 x2 which obviously satisfies the above assumptions. Note that f only depends on the x variable and thus ∂f ∂t = 1 2 x2, ∂f ∂x = tx, 1 2 ∂2f ∂x2 = t, f(0, 0) = 0. Thus, for every T ≥ 0, 1 2 T (W (T ))2 = f(0, 0) + ∫ T 0 [∂f ∂t (t,W (t)) + 1 2 ∂2f ∂x2 (t,W (t)) ] dt + ∫ T 0 ∂f ∂x (t,W (t)) dW (t) = 0 + ∫ T 0 [1 2 (W (t))2 + 1 2 t ] dt+ ∫ T 0 tW (t) dW (t) = 1 2 ∫ T 0 (W (t))2 dt+ 1 4 T 2 + ∫ T 0 tW (t) dW (t) Rearranging, we get∫ T 0 tW (t) dW (t) = 1 2 T (W (T ))2 − 1 2 ∫ T 0 (W (t))2 dt− 1 4 T 2. 15 SOLUTIONS: MAT00018M 6 Marks Hence, since E [ (W (t))2 ] = t so that E [ ∫ T 0 (W (t))2 dt ] = T 2 2 we deduce that E [∫ T 0 tW (t) dW (t) ] = 0. 3 Marks (d) Method based on use of Itoˆ formula. We look for a solution of the form X(t) = f ( t,W (t) ) , t ≥ 0 for an appropriate function f(t, x) (to be found). The simplified version of Itoˆ’s formula then gives, because ∂f ∂t (t, x) = 0 and f(t, x) = f(x), dX(t) = ( ∂f ∂t (t,W (t)) + 1 2 ∂2f ∂x2 (t,W (t)) ) dt+ ∂f ∂x (t,W (t))dW (t) Comparing the above with equation to be solved, i.e. dX(t) = 3X(t)dt+ 4X(t) dW (t) we deduce that f should satisfy, since X(t) = f ( W (t) ) , ∂f ∂x (t,W (t)) = 4f ( t,W (t) ) ∂f ∂t (t,W (t)) + 1 2 ∂2f ∂x2 ( W (t) ) = 3f ( W (t) ) We can replace W (t) above by x to deduce ∂f ∂x (x) = 4f(t, x) ∂f ∂t (t, x) + ∂2f ∂x2 (x) = 3f(t, x) Denoting y(x) = f(t, x), for fixed t, the first of these two equations becomes dy dx = 4y(x) Solving it we get 16 SOLUTIONS: MAT00018M y(x) = ze4x where, obviously, the constant z depends on t. So we get f(t, x) = z(t)e4x We need to find z(t). For this purpose we use the first of the above two equations. We get z′(t)e4x + 1 2 × 42z(t)e4x = 3z(t)e4x Dividing by e4x we get z′(t) = −5z(t) This is a linear homogeneous equation whose solution is z(t) = z0e −5t Therefore, f(t, x) = z0e −5te4x and so X(t) = z0e −5t+4W (t) Since X(0) = 1, we infer that X(t) = e−5t+4W (t) 14 Marks Total: 33 Marks 17
欢迎咨询51作业君