$$X_t=X_0 + \int_0^t b(s)ds+ \int_0^t \sigma(s)dW_s,\quad \forall t\ge 0,$$

where $X_0\ge 0$ is a random variable of density $\rho$, $(W_t)_{t\ge 0}$ is an independent Brownian motion and $b,\sigma$ are measurable function "as nice as possible". Set $\tau:=\inf\{t\ge 0: X_t\le 0\}$ and $Y_t:=X_{t\wedge \tau}$ for $t\ge 0$. Denote by $p(t,\cdot)$ be the density of $Y_t$ restricted on $(0,\infty)$, i.e.

$$\mathbb E[f(Y_t){\bf 1}_{\{Y_t>0\}}] = \int_0^{\infty}f(x)p(t,x)dx,\quad \mbox{for all continuous and bounded function } f: \mathbb R\to\mathbb R.$$

It is known that $p$ satisfies the Fokker–Planck equation as below :

\begin{eqnarray} \partial_t p &=& \frac{\sigma(t)^2}{2}\partial^2_{xx} p - b(t)\partial_x p,\quad \forall t>0, x>0 \\ p(0,x) &=& \rho(x),\quad \forall x\ge 0 \\ p(t,0) &=& 0,\quad \forall t>0. \end{eqnarray}

From the probabilistic point of view, it is straightforward to see $p\ge 0$ and the function

$$m(t):=\int_0^{\infty}p(t,x)dx \in [0,1]$$

is non-increasing. From the PDE point of view, why the above Fokker–Planck equation together with the initial and boundary conditions implies $p\ge 0$ and $m$ is non-increasing?

Any answer, comments or references are appreciated.

PS : When $b\equiv 0$ and $\sigma\equiv 1$, one has in view of the reflection principle,

$$p(t,x)=\int_0^{\infty}\frac{1}{\sqrt{2\pi t}}\left(\exp\left(-\frac{(x-y)^2}{2t}\right)-\exp\left(-\frac{(x+y)^2}{2t}\right)\right)\rho(y)dy$$

which clearly satisfies the desired properties. For general $b,\sigma$, do we still have the closed-form expression of $p$?


1 Answer 1


Differentiating $m(t)$ and using the equation leads to $$ m'(t):=\int_0^{\infty}\left( \frac{\sigma^2(t)}{2}\partial^2_{xx} p(x,t) - b(t)\partial_x p(x,t)\right)\,dx= -\frac{\sigma(t)^2}{2}\partial_{x} p(0,t)<0. $$ The last inequality follows from the Zaremba-Giraud theorem for the sign of the solution's normal derivative at the boundary. See, for example, Nazarov A.I., A Centennial of the Zaremba–Hopf–Oleinik Lemma.

For $b\equiv0$ $$ p(t,x)=\int_0^{\infty}\frac{1}{\sqrt{2\pi \int_0^t{\sigma^2(\tau)}\,d\tau}}\left(\exp\left(-\frac{(x-y)^2}{2\int_0^t{\sigma^2(\tau)}\,d\tau}\right)-\exp\left(-\frac{(x+y)^2}{2\int_0^t{\sigma^2(\tau)}\,d\tau}\right)\right)\rho(y)dy. $$ For the general case the Green's function of the first BVP cannot be expressed via elementary functions.

  • $\begingroup$ Thank you so much Andrew for the answer. I will accept your solution as soon as I read the paper that you suggested. For the case where $b>0$ is constant, can we expect some "nice" expression for $p$? Here "good" does not mean as explicit as your formula above, but as explicit as possible $\endgroup$
    – GJC20
    Dec 31, 2021 at 18:19
  • $\begingroup$ Happy New Year! $\endgroup$
    – GJC20
    Dec 31, 2021 at 18:19
  • $\begingroup$ Further, why the positivity of $p$ is ensured by the PDE? $\endgroup$
    – GJC20
    Jan 1 at 13:29
  • $\begingroup$ @GJC20 Positivity follows from $\rho\ge0$, $\rho>0$ somewhere and the maximum principle. Or, in this case, from the formula for $p$ in your post. $\endgroup$
    – Andrew
    Jan 1 at 15:17
  • $\begingroup$ Thanks for the quick reply. I mean for the general case where $b,\sigma$ that are not constant. Could you please provide the reference of the maximum principle that can be applied here? Also, for the case where $b$ is constant, can we have some "nice" expression for $p$? $\endgroup$
    – GJC20
    Jan 1 at 16:42

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.