Past exam of the mathematics course of the University of Cambridge / 2001 / iii / Paper 21 (Advanced Probability)

For a submartingale $(Z_n)$ and $a<b$, let $U_N(a,b)$ count the completed upcrossings of $[a,b]$ by time $N$. The Doob upcrossing inequality givesFor a martingale $(X_n)$, a useful sharper form isHere $x^{+}=\max (x,0)$ and $x^{-}=\max (-x,0)$. To see the martingale form, use a bounded predictable process $H_k\in \{0,1\}$ that holds one unit after an observation at or below $a$ and releases it on the next observation at or above $b$. Its gain $G_N={\sum }_{k=1}^N{H}_k(X_k-X_{k-1})$ is at least $(b-a)U_N-(X_N-a{)}^{-}$: completed trades earn at least $b-a$, and any unfinished trade loses at most $(X_N-a{)}^{-}$. Since $\mathbb{E}{G}_N=0$, the bound follows. For the submartingale version, apply the same strategy to $(Z_n-a{)}^{+}$, itself a submartingale. Each purchase occurs at value zero, so its gain is at least $(b-a)U_N$. The complementary predictable strategy has nonnegative expected gain, bounding this strategy's expected gain by $\mathbb{E}(Z_N-a{)}^{+}-\mathbb{E}(Z_0-a{)}^{+}$, and in particular by the stated right side.

The almost-sure martingale convergence theorem says that if $C={\sup }_n\mathbb{E}|X_n|<\infty$, then $X_n$ converges almost surely to a finite, integrable random variable $Y$, with $\mathbb{E}|Y|\le C$. Indeed, the Doob upcrossing inequality bounds the expected total number of upcrossings of every rational interval by $(C+|a|)/(b-a)$. By monotone convergence, each such total count is finite almost surely; intersect these probability-one events over all rational $a<b$. If the lower and upper limits of a sample path differed, some rational interval strictly between them would be crossed infinitely often. Thus each remaining path has a limit in the extended real line. The Fatou lemma givesso the limit is finite almost surely and integrable. Almost-sure convergence by itself neither preserves expectations nor supplies this uniform first-moment bound.

For the loss of expectation, let $({\eta }_k)$ be independent variables with the fair Bernoulli distribution and setConditional on the past, survival doubles the value with probability $1/2$ and otherwise sets it to zero, so this is a nonnegative martingale. An infinite string of successes has probability zero. The process therefore eventually vanishes almost surely, giving yes: $X_0=1$ and $Y=0$ almost surely are possible. Its expectations remain one at every finite time; the process is not uniformly integrable.

A nonintegrable limit is also possible for a signed martingale. Take independent ${\xi }_k$ withand let $M_0=1$, $M_n=1+{\sum }_{k=1}^n{\xi }_k$. Every increment has mean zero and a finite absolute expectation, so each $M_n$ is integrable and $(M_n)$ is a martingale. Since ${\sum }_k\mathbb{P}({\xi }_k\ne 0)<\infty$, the Borel-Cantelli lemma makes only finitely many increments nonzero almost surely; hence the finite limit $Y=1+{\sum }_k{\xi }_k$ exists almost surely. To compute its absolute expectation, let $A_k$ be the event that only the $k$th increment is nonzero. Independence givesThe infinite product is positive because its tail logarithms are bounded in absolute value by a constant times the summable series ${\sum }_j{2}^{-j}$. The events $A_k$ are disjoint, and on $A_k$, $|Y|\ge 4^k-1$. ThereforeThus yes: a finite almost-sure martingale limit can have infinite absolute expectation, even with deterministic $M_0=1$. This is almost-sure martingale convergence to a nonintegrable limit; the missing hypothesis is the uniform first-moment bound. For a nonnegative martingale, in contrast, the Fatou lemma would force the limit to be integrable.

Write $s_i=p_i+q_i>0$ and $d_i=p_i-q_i$ for $i=0,1$. For the parity correction $h_{\theta }(k)=k-\theta 1_{\{k\text{ odd}\}}$, the conditional expectations of the increments of $h_{\theta }(X_n)$ areThe holding move contributes zero. Thus the period-two martingale corrector for a birth-death chain exists precisely whenThis is the condition that $h_{\theta }$ be a harmonic function for a Markov chain. The process has integrable values from a fixed initial state because $|X_n|\le |X_0|+n$; the Markov property therefore turns the zero conditional increments into the required martingale property. Conversely, both parities are reached with positive probability because all right and left probabilities are positive, so the two conditional equations are necessary. Expanding their numerator also gives the equivalent balance condition $p_0{p}_1=q_0{q}_1$. Positivity makes $|\theta |<1$.

Put $L=-aN$, $U=bN$. The exit time $T_N$ is almost surely finite. Indeed, set $m=U-L$ and $p_{\ast }=\min (p_0,p_1)>0$. From any interior point, consecutive right steps reach $U$ in at most $m$ moves, with probability at least $p_{\ast }^m$. The Markov property gives $\mathbb{P}(T_N>km)\le (1-p_{\ast }^m{)}^k$. This is the finite-interval exit times have geometric tails argument. The walk cannot skip a boundary because its jumps have size at most one.

Under the balance condition, the stopped martingale $h_{\theta }(X_{n\wedge T_N})$ is bounded, because the stopped position stays in $[L,U]$. Apply the optional stopping theorem at $n\wedge T_N$, then dominated convergence as $n\to \infty$. Starting from zero givesWith ${\epsilon }_a=1_{\{aN\text{ odd}\}}$ and ${\epsilon }_b=1_{\{bN\text{ odd}\}}$,The bounded parity corrections disappear after division by $N$; the holding probability of a Markov chain may be positive or zero and may differ between parities.

For failure of balance, use scale increments for a birth-death chain rather than the nonexistent bounded correction. Let $u(k)$ be the upper exit probability from $k$ for boundaries $L,U$. The Markov property yieldswhere $p_k,q_k$ depend only on parity. Choose positive numbers $w_j$ with $w_{j+1}=(q_j/p_j)w_j$. Then $u(k)={\sum }_{j=L+1}^k{w}_j/{\sum }_{j=L+1}^U{w}_j$ satisfies these equations and boundary values. Stopping this bounded harmonic function proves that it is the exit probability. In particular,If $\rho <1$, the positive-index sum stays bounded as $N\to \infty$, while the negative-index contribution diverges geometrically, so ${\pi }_N\to 0$. If $\rho >1$, the negative-index sum stays bounded while the positive-index sum diverges, so ${\pi }_N\to 1$. Therefore the complete classification isThus a fixed nonzero imbalance drives the macroscopic exit to the corresponding side, for every positive $a,b$.

Let $\varphi (u)=\int e^{iuy}\mu (dy)$. Since $1-\cos (uy)\ge 0$, the Tonelli theorem giveswith the fraction defined to be one at zero. For $|x|\ge 1$, $\sin x/x\le \sin 1$. To verify the constant, the function is even; on $[1,\pi ]$ it decreases because $(x\cos x-\sin x{)}^{\prime }=-x\sin x<0$ on $(0,\pi )$ and $x\cos x-\sin x$ starts from zero. On $[\pi ,\infty )$, $\sin x/x\le 1/\pi <\sin 1$. Consequently the integrand is at least $1-\sin 1$ on $|y|\ge \lambda$, proving the characteristic-function tail boundFor pointwise convergence ${\varphi }_n\to \varphi$, where $\varphi$ is a characteristic function, the dominated convergence theorem applies on each finite integration interval, since the integrands lie between zero and two. HenceContinuity of $\varphi$ at zero and $\varphi (0)=1$ make the right side tend to zero as $\lambda \to \infty$. Given $\epsilon >0$, choose $\lambda$ so it is below $\epsilon /2$. The tail probability is then below $\epsilon$ for all sufficiently large $n$. Enlarge the compact interval to control the finitely many remaining measures, each of which has vanishing tails. This proves uniform tightness of the entire sequence, not just of its tail. It is the characteristic-function tightness bound principle with the sharper displayed constant.

For the integrable mean-zero summands, differentiation at zero is justified byThe dominated convergence theorem gives ${\varphi }^{\prime }(0)=i\mathbb{E}{X}_1=0$. DefineIt is continuous away from zero by continuity of the characteristic function, and continuous at zero by the zero derivative. Since $|\varphi (u)|\le 1$, the finite geometric series yields, for $n\ge 1$,and the case $n=0$ has both sides zero. Thus the requested uniform-in-$n$ estimate isThis is the characteristic-function derivative under an absolute first moment estimate.

Independence makes the characteristic function of $S_n/n=(X_1+\cdots +X_n)/n$ equal to $\varphi (u/n{)}^n$. Therefore $|1-\varphi (u/n{)}^n|\le |u|\psi (u/n)$. Applying the already proved characteristic-function tail bound to its distribution, for every $\epsilon >0$,The convergence follows because ${\sup }_{0\le u\le 1/\epsilon }\psi (u/n)\to 0$, and the remaining integral of $u$ is finite. Hence $\boxed{S_n/n\to 0\text{ in probability}}$, the mean-zero weak law of large numbers. This deduction uses the tail bound directly and does not presuppose the Lévy continuity theorem.

False. In fact, standard one-dimensional Brownian motion satisfies ${\mathrm{liminf}}_{t\to \infty }{B}_t=-\infty$ almost surely. For each integer $m\ge 1$, let ${\tau }_{-m}=\inf \{t:B_t=-m\}$. The Brownian reflection principle, applied to $-B$, giveswhere $\Phi$ is the standard normal distribution function. Thus each negative integer is reached almost surely. Intersecting these countably many probability-one events, all negative integer levels are reached on the same path. Their hitting times must become arbitrarily large, because a continuous path is bounded on every compact time interval. Hence the path cannot tend to positive infinity. Applying the same argument to positive levels also gives ${\mathrm{limsup}}_{t\to \infty }{B}_t=+\infty$ almost surely.

True. The negative-level hitting times from the preceding calculation are almost surely finite for every negative integer, and occur at unbounded times. ThereforeThis is the almost-sure negation in the original PDF, not another assertion of convergence. The stronger oscillation statement $\mathrm{liminf}{B}_t=-\infty$ and $\mathrm{limsup}{B}_t=+\infty$ rules out that convergence path by path.

True. The unit-time increments $Z_k=B_k-B_{k-1}$ are independent and identically distributed random variables with standard normal distribution, hence integrable with mean zero. The strong law of large numbers gives $B_n/n\to 0$ almost surely at integer times.

To control the intervals between integers, put $M_n={\sup }_{0\le u\le 1}|B_{n+u}-B_n|$. Stationary increments and the Doob L2 maximal inequality give $\mathbb{E}{M}_n^2\le 4$. Thus the Markov inequality yieldsThe Borel-Cantelli lemma, followed by a countable intersection over positive rational $\epsilon$, implies $M_n/n\to 0$ almost surely. For $n\le t\le n+1$,Both terms tend to zero on the same probability-one event, proving $\boxed{B_t/t\to 0\text{ almost surely}}$ for continuous time as well.

False. The preceding interpolation argument proves that the event $\{B_t/t\to 0\}$ has probability one. Its complementary event, which is the nonconvergence assertion printed here, therefore has probability zero. The normalization by $t$ dominates both the integer-time increments and the fluctuations within each unit interval.

False. For every deterministic $t>0$, Brownian scaling gives $B_t/\sqrt{t}\sim N(0,1)$. In particular, for any $\epsilon >0$,independently of $t$. Almost-sure convergence would imply convergence in probability, but these fixed positive tail probabilities rule out convergence in probability to zero. The square-root normalization retains fluctuations of order one.

True. The preceding distributional argument alone shows that almost-sure convergence to zero cannot hold with probability one. To establish almost-sure nonconvergence, rather than merely failure of a probability-one convergence claim, use the Kolmogorov zero-one law.

Let $A=\{B_n/\sqrt{n}\to 0\}$. For every fixed $m$, subtracting $B_m/\sqrt{n}\to 0$ shows thatThus $A$ belongs to the tail sigma-algebra of the independent unit-time increments; it is a tail event. The Kolmogorov zero-one law gives $\mathbb{P}(A)\in \{0,1\}$. If its probability were one, then $B_n/\sqrt{n}\to 0$ in probability, contrary to its standard normal distribution at every $n$. Therefore $\mathbb{P}(A)=0$. Continuous-time convergence would imply integer-time convergence, soThe more general Brownian fluctuations exceed the square-root scale result is consistent with this direct tail-event proof. No independence of the normalized values themselves has been assumed.

For the continuous adapted process $X$, the supremum on each compact time interval is attained. Let $D_t=(\mathbb{Q}\cap [0,t])\cup \{t\}$. Continuity gives ${\sup }_{s\le t}{X}_s={\sup }_{r\in D_t}{X}_r$, and henceEach event inside the union is in ${\mathcal{F}}_r\subseteq {\mathcal{F}}_t$, so this proves carefully that $T$ is a stopping time. Merely checking rational times for $X_r\ge \lambda$ would miss a path that touches the level at an isolated irrational time; the approximation from below avoids this issue. This is the closed-level hitting times of continuous adapted processes criterion. For $\lambda =0$, nonnegativity makes $T=0$ and the later inequality is immediate.

For $\lambda >0$, put $A=\{T\le t\}$ and $\tau =T\wedge t$. On $A$, continuity implies $X_{\tau }\ge \lambda$. To justify the optional sampling step without extra filtration assumptions, round $\tau$ upwards to a finite dyadic grid in $[0,t]$, obtaining stopping times ${\tau }_n\downarrow \tau$. The event $A$ belongs to ${\mathcal{F}}_{\tau }$, and hence to ${\mathcal{F}}_{{\tau }_n}$: directly, $A\cap \{\tau \le u\}=\{T\le u\}$ for $u<t$, while for $u\ge t$ it is $A$. Discrete optional sampling for the sampled submartingale givesAs $n\to \infty$, continuity gives $X_{{\tau }_n}\to X_{\tau }$. Nonnegativity and the Fatou lemma therefore yieldConsequently the Doob maximal inequality for a nonnegative submartingale takes the precise form

For the second-moment bound, suppose first that $\mathbb{E}{X}_t^2<\infty$; otherwise the inequality with an infinite right side is automatic. Set $M_R=X_t^{\ast }\wedge R$. The layer cake representation and the preceding tail estimate giveThe integrands are nonnegative, so the interchange is justified by the Tonelli theorem. Since $M_R$ is bounded, division is legitimate when its norm is nonzero, giving $\Vert M_R{\Vert }_2\le 2\Vert X_t{\Vert }_2$; the zero-norm case is immediate. Finally, monotone convergence as $R\to \infty$ givesThis truncated layer-cake proof of the L2 maximal inequality proves square integrability of the maximum rather than assuming it during the Cauchy-Schwarz inequality step.

For $t>0$, put $M_t={\sup }_{0\le s\le t}{B}_s$. For $\lambda >0$, reflect the Brownian path after its first hit of $\lambda$. The Strong Markov property and symmetry of subsequent Brownian increments make this reflection preserve the path distribution. It exchanges the events $\{M_t\ge \lambda ,B_t<\lambda \}$ and $\{B_t>\lambda \}$. Since $B_t$ has no atom at $\lambda$, the Brownian reflection principle givesThe case $\lambda =0$ follows by decreasing positive levels to zero: the probability is one. At $t=0$, $M_0=0$ and the probability is zero for every $\lambda \ge 0$.

The displayed tail probabilities are exactly those of $|B_t|$, so $M_t$ has the half-normal distribution of $\sqrt{t}|Z|$ with $Z\sim N(0,1)$. ThusFor $A_t={\sup }_{s\le t}|B_s|$, the pointwise inequality $A_t\ge M_t$ gives the lower non-strict bound. To prove strictness when $t>0$, fix $\lambda >0$. Reflection at $\lambda$ also givesSubtracting this from $\mathbb{P}(B_t<-\lambda )=\mathbb{P}(B_t>\lambda )$ givesOn this event $A_t\ge |B_t|>\lambda >M_t$, so $A_t^2-M_t^2$ is strictly positive with positive probability. Finally, $|B_t|$ is a continuous nonnegative submartingale by the conditional Jensen inequality, and part (a) gives $\Vert A_t{\Vert }_2\le 2\Vert B_t{\Vert }_2=2\sqrt{t}$. Its square is therefore integrable, and the strict pointwise comparison on a positive-probability event givesThis is the strict comparison of one-sided and absolute Brownian maxima. The strict lower inequality printed with $t\ge 0$ needs the qualification $t>0$: at $t=0$ both maxima vanish and all three quantities are zero.

By the usual definition of a Lévy process, $X_0=0$ almost surely. The two martingale assumptions giveFor a continuous Lévy process, the jump measure in the Lévy–Khintchine formula is zero, and the process has the form of a Brownian Gaussian component plus a deterministic drift. Thus its characteristic function isThe first moment makes $m=0$, and the second moment makes ${\sigma }^2=1$. ConsequentlyEquivalently, continuity and the martingale identity $X_t^2-t$ identify the quadratic variation as $t$, so the Lévy characterization of Brownian motion gives the same conclusion. A continuous nonzero drift is excluded by the mean-zero martingale assumption.

When paths move only by jumps of size $\pm 1$, there is neither a Gaussian component nor a continuous drift. The Lévy measure is supported on $\{-1,1\}$ and has finite mass because these points are bounded away from zero. The process is therefore a Compound Poisson process, expressible as a difference of independent Poisson processeswhere the rates are ${\lambda }_{+},{\lambda }_{-}$ respectively. The martingale assumptions implyHence $\boxed{{\lambda }_{+}={\lambda }_{-}=1/2}$. The embedded jump chain is a simple symmetric random walk, and the total jump rate is one. In particular,This is the symmetric Skellam distribution, with the explicit mass functionFor $k\ge 0$, sum the joint probabilities of $N_{+}=j+k,N_{-}=j$; for $k<0$, interchange the two variables. The characteristic function is $\mathbb{E}{e}^{iu{X}_1}=\exp (\cos u-1)$, consistent with the rate-one symmetric Poisson difference process. The condition that paths move only by jumps is essential to exclude a separate continuous component.

For the exit law, the two identified processes have the same argument. Starting at zero, neither can skip $-a$ or $b$: Brownian motion is continuous, and the symmetric Poisson difference process has unit jumps with integer boundary levels. Therefore $|X_{t\wedge T}|\le K=\max (a,b)$, and if $T$ is finite, $X_T\in \{-a,b\}$.

To justify finiteness rather than presume it, apply the optional stopping theorem to the martingale $X_t^2-t$ at the bounded stopping time $t\wedge T$:The first boundary hitting time is a stopping time in both cases: in the continuous case use the continuous-path hitting criterion, and in the jump case inspect the successive jump times. Monotone convergence now gives $\mathbb{E}T\le K^2$, so $T<\infty$ almost surely. The stopped position tends to $X_T$ as $t\to \infty$ and is bounded, making dominated convergence applicable. Stopping $X$ at $t\wedge T$ and taking this limit yields $\mathbb{E}{X}_T=0$.

If $\pi =\mathbb{P}(X_T=-a)$, then $0=-a\pi +b(1-\pi )$. Thus in both cases the entire distribution isThis is the same balanced gambler's ruin law despite the different one-time distributions. As a further consistency check, passing to the limit in the stopped square-martingale identity givesAll uses of an unbounded exit time were obtained by limits of bounded-time sampling with an explicit bound on the stopped position.