0
$\begingroup$

I have a question on Lawler – Notes on the Bessel process, on page 4. Let $X_t$ be one-dimensional Brownian motion, and we want to use $N_t$ as a measure-changing (local) martingale, defined as $$N_t=\left(\frac{X_t}{X_0}\right)^a\exp\left(-\frac{a(a-1)}{2}\int_0^t\frac{ds}{X_s^2}\right),\quad t<T_0,$$

up to the first visit to zero. It satisfies the SDE $$dN_t=\frac{a}{X_t}N_tdX_t,\quad N_0=1.$$

Then it is supposed in the note that $0<\varepsilon<X_0<R$, and $\tau=T_{\varepsilon}\wedge T_R$, as up to the first visit to $\varepsilon$ or $R$. Then it claims that $N_s$ stopped by $\tau$, $$dN_{t\wedge\tau}=\frac{a}{X_{t\wedge\tau}}1(t<\tau)N_{t\wedge\tau}dX_t,$$

is uniformly bounded. But why is this true?

I have a rough thought that $N_{t\wedge\tau}=N_0+\int_0^t\frac{a}{X_{s\wedge\tau}}N_{s\wedge\tau}dX_{s\wedge\tau}\leq N_0+\frac{a}{\varepsilon}\int_0^tN_{s\wedge\tau}dX_{s\wedge\tau}$, a.s. as $t<\tau$. Then if we have the stochastic version of Grönwall's inequality, i.e. if we can read $N_{t\wedge\tau}\leq N_0\exp(\frac{a}{\varepsilon}X_{t\wedge\tau})$, a.s., then it is uniformly bounded for sure. But I am quite pessimistic about whether this holds or not.

$\endgroup$

2 Answers 2

2
$\begingroup$

On the time interval $[0,\tau]$, the exponential factor in the definition of $N_t$ is bounded below by $0$ and above by $\exp(Kt)$ for some constant $K=K(\epsilon,R)\ge 0$. Therefore $0<N_{t\wedge\tau}\le(R/\epsilon)^a\exp(K\cdot (t\wedge \tau))$. This is "uniform" provided you ignore the dependence on $t$.

$\endgroup$
1
  • $\begingroup$ Can we get the boundedness from the SDE itself? I am wondering whether we can have some results without solving for that explicit form. $\endgroup$
    – MikeG
    Mar 18, 2022 at 23:08
0
$\begingroup$

This isn't quite the "stochastic Gronwall" you're looking for, but starting with the equation: $$ N_{t\wedge{\tau}}=1+a\int_{0}^{t\wedge{\tau}}\frac{N_{s}}{X_{s}}dX_{s} $$ Taking the expectation of the square of both sides (okay, technically you need to truncate first and later take a limit, but I trust you can fill in the gaps) we get: $$ \mathbb{E}N_{t\wedge{\tau}}^{2}\leq{1+a^{2}\int_{0}^{t\wedge{\tau}}\mathbb{E}\big(\frac{N_{s\wedge{\tau}}^{2}}{X_{s\wedge{\tau}}^{2}}\big)ds}\leq{1+\frac{a^{2}}{\varepsilon^{2}}\int_{0}^{t}\mathbb{E}N_{s\wedge{\tau}}^{2}ds} $$ Applying the usual Gronwall inequality to the function $f(s)=\mathbb{E}N_{s\wedge{\tau}}^{2}$, we get that the local martingale $(N_{t\wedge{\tau}})_{t\geq{0}}$ is square integrable. This is far from the almost sure bound you want, but it's something.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.