3
$\begingroup$

Let $M$ be the space of right continuous functions $\ell: \mathbb R_+\to [0,1]$ that are non increasing s.t. $\ell(0)=0$. Define the map $\Gamma : M\to M$ by $\Gamma[\ell](t):=\mathbb P[\tau^{\ell}>t]$ for all $\ell \in M$ and $t\ge 0$, where $\tau^{\ell}:=\inf\{t\ge 0: X^{\ell}_t\le 0\}$ and

$$X^{\ell}_t:=1+t+\int_0^t\frac{1}{1+\ell(s)}dW_s,\quad \forall t\ge 0.$$

Here $(W_t)_{t\ge 0}$ denotes a Brownian motion. Let $M$ be endowed with the topology as follow: $\ell^n$ converges to $\ell$ in $M$ iff $\lim_{n\to\infty}\ell^n(t)=\ell(t)$ for all points of continuity of $\ell$. Can we prove the continuity of $\Gamma$ with respect to this topology?

Remark : To prove $\{\inf_{0\le s\le t}X^{\ell}_s\le 0\}=\{\inf_{0\le s\le t}X^{\ell}_s< 0\}$, it suffices to use Lévy's characterization. More precisely, we can write $X^{\ell}_t=1+t+B_{\langle X^{\ell}\rangle_t}\equiv 1+t+B_{L(t)}$, where $B$ denotes a Brownian motion and $L(t):=\int_0^t ds/(1+\ell(s))^2$. Therefore

$$\inf_{0\le s\le t}X^{\ell}_t = \inf_{0\le u\le L(t)}\{1+L^{-1}(u)+B_t\},$$

which implies the desired result as $\inf_{0\le u\le L(t)}\{1+L^{-1}(u)+B_t\}$ admits a density.

$\endgroup$
5
$\begingroup$

I believe we can. Let $\ell_n \to \ell$ in your topology, and fix $t_0 \in \mathbb R_+$. We show that $\Gamma(\ell_n)(t_0) \to \Gamma(\ell)(t_0)$. We work over the interval $[0, T]$ with $T > t_0$.

Step 1: We first note that $\ell_n$ converges to $\ell$ in measure.

Indeed, let $\varepsilon > 0$ be arbitrary. As $\ell$ is nonincreasing, $\ell$ contains only jump discontinuities and for any $n > 0$, there exist only finitely many jumps with magnitude greater than $\frac{\varepsilon}{2^j}$. Cover these jumps with finitely many closed intervals $C_i^j$ of total length less than $\frac{\varepsilon}{2^j}$. Divide the complement into intervals $D_i^j$ on which $\ell$ varies by no more than $\frac{\varepsilon}{2^j}$, and consider the partition $\mathcal P_j ;= C_i^j \cup D_i^j$. Applying the pointwise convergence of $\ell_n$ to $\ell$ near the endpoints of the $C_i^n$ and the monotonicity of $\ell_n$ and $\ell$ now allows us to conclude.

Step 2: We show that $X^{\ell_n} \to X^{\ell}$ uniformly in probability.

Namely, that $$P(\sup_{s \in [0, T]} | X_t^{\ell_n} - X_t^{\ell}| \geq \frac{1}{2^k}) \to 0.$$

To see this, note that we can write

$$X^{\ell}_t:=1+t+\int_0^t\frac{1}{1+\ell(s)} + g_n(s) \ dW_s \, \quad \forall t\ge 0.$$

with $g_n := \frac{1}{1 + \ell_n(s)} - \frac{1}{1 + \ell(s)} \to 0$ in $L^2$. Whereby the aforementioned convergence follows from the convergence in measure of $\ell_n$ to $\ell$.

Note that $X_s^{\ell_n} - X_s^{\ell}$ is a martingale, and that $X_T^{\ell_n} - X_T^{\ell}$ is Gaussian with mean $0$, and variance $||g_n||_{L^2}$.

And so by Doob’s (sub)martingale inequality and symmetry of the Gaussian distribution, we have

$$P(\sup_{s \in [0, T]} | X_s^{\ell_n} - X_s^{\ell}| \geq \frac{1}{2^k}) \leq 2^{k+1} E[(X_T^{\ell_n} - X_T^{\ell})^+] \to 0,$$

as $n \to \infty$.

Step 3: Conclusion.

Let $\varepsilon > 0$ be arbitrary. We note that by continuity of $X_n^{\ell}$ we can write the event $\{\tau^{\ell} > t\}$ as $$\bigcup_{k \in \mathbb N} A_k := \bigcup_{k \in \mathbb N} \{X_s \geq \frac{1}{k}, \ \forall s \in [0,t_0]\}.$$

By continuity from below, we have that for some $k_0 \in \mathbb N$ that $P(A_{k_0}) > P(\{\tau^{\ell} > t\}) - \frac{\varepsilon}{2}$.

Since $X^{\ell_n} \to X^{\ell}$ uniformly in probability, for all large enough $n$, we have

$$P(\sup_{s \in [0, T]} | X_s^{\ell_n} - X_s^{\ell}| \geq \frac{1}{2k_0}) < \frac{\varepsilon}{2},$$

so that for all $n > N$ we have

$$\Gamma(\ell_n)(t_0) = P(\{\tau^{\ell_n} > t_0\}) > P(A_{k_0} \cup \{\sup_{s \in [0, T]} | X_s^{\ell_n} - X_s^{\ell}| > \frac{1}{2k_0} \}^c) > P(\{\tau^{\ell} > t_0\}) - \varepsilon$$

For the reverse inequality, we argue similarly - we note that, up to $P$-null sets, we can write the event $\{\tau^{\ell} \leq t_0\}$ as

$$\bigcup_{i \in \mathbb N} \{\inf_{s \in [0, t_0)}\ X_s^{\ell} \leq -\frac{1}{i}\},$$

In order for the above representation to hold, we need to show that $\{\inf_{0\le s\le t}X^{\ell}_s\le 0\}=\{\inf_{0\le s\le t}X^{\ell}_s< 0\}$, up to a $P$-null set.

For this, it suffices to use Lévy's characterization. More precisely, we can write $X^{\ell}_t=1+t+B_{\langle X^{\ell}\rangle_t}\equiv 1+t+B_{L(t)}$, where $B$ denotes a Brownian motion and $L(t):=\int_0^t ds/(1+\ell(s))^2$. Therefore

$$\inf_{0\le s\le t}X^{\ell}_t = \inf_{0\le u\le L(t)}\{1+L^{-1}(u)+B_t\},$$

which implies the desired result as $\inf_{0\le u\le L(t)}\{1+L^{-1}(u)+B_t\}$ admits a density.

Thus with a similar calculation, we obtain $$\Gamma(\ell_n)(t) < P(\{\tau^{\ell} < t_0\}) + \varepsilon$$

for all large enough $n$.

Since $\varepsilon$ was arbitrary, we conclude that $\Gamma(\ell_n)(t_0) \to \Gamma(\ell)(t_0)$, as was to be shown.

$\endgroup$
11
  • $\begingroup$ Thanks for your answer. For the reservse inequality, I think there is a gap. Indeed, $\{\tau^{\ell}\le t_0\}\neq \cup_{i\in\mathbb N}\{\inf_{s\in [0,t_0)}X^{\ell}_s\le -1/i\}$ as the scenario $X_s>0$ for $s\in [0,t_0)$ and $X_{t_0}=0$ belongs to the first set but not the later one. $\endgroup$
    – GJC20
    Jun 27 at 19:52
  • $\begingroup$ To make your arguments rigorous, I think we need to show Therefore, we need to argue that $\{\tau^{\ell}\le t_0\}$ coincides with $\cup_{i\in\mathbb N}\{\inf_{s\in [0,t_0)}X^{\ell}_s\le -1/i\}$ almost surely $\endgroup$
    – GJC20
    Jun 27 at 19:54
  • $\begingroup$ Ah yes, I’ve implicitly argued that the scenario you mentioned has probability zero. Do you need me to add that as a detail to the answer? $\endgroup$
    – Nate River
    Jun 27 at 20:58
  • $\begingroup$ Could you please provide the details? Many thanks! $\endgroup$
    – GJC20
    Jun 27 at 21:43
  • 1
    $\begingroup$ Many thanks. I will accept your answer as soon as you complete it. Btw, there is some typo, e.g. $1/2k_0$ instead of $k_0/2$ $\endgroup$
    – GJC20
    Jun 28 at 15:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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