Let $b\in \mathbb R$ and $\sigma>0$ be given. For a fixed probability distribution $\mu_0$ on $\mathbb R$ s.t.


consider the mean field SDE :

$$dX_t = \mathbf{1}_{\{X_t>0\}} \left[ bdt + \frac{\sigma}{1 + m_t {\bf 1}_{\{b>0\}}} dW_t \right],\quad \mbox{for all } t\ge 0,~~~~~~~~~~~~~~~~(\ast)$$

where $X_0\sim \mu_0$ is independent of the Brownian motion $(W_t)_{t\ge 0}$ and

$$m_t:=\int_{(0,\infty)}\mu_t(dx),\quad \mbox{for all } t\ge 0.$$

How can we show the existence and uniqueness of the (weak) solution to $(\ast)$?

Any answers, remarks or references are highly appreciated!


The case for $b\le 0$ is trivial. Indeed, $(\ast)$ reduces to $dX_t = \mathbf{1}_{\{X_t>0\}} \big[ bdt + \sigma dW_t \big]$ and the solution is given as $X_t=Y_{t\wedge \tau}$, where $Y_t:=X_0+bt+\sigma W_t$ and $\tau:=\inf\{t\ge 0: Y_t\le 0\}$. For the case $b>0$, $(\ast)$ turns to be

$$dX_t = \mathbf{1}_{\{X_t>0\}} \left[ bdt + \frac{\sigma}{1 + m_t} dW_t \right].$$

I do not find any literature on the existence of its solution.


A heuristic argument is as follows : Let $\ell:\mathbb R_+\to [0,1]$ be some "nice" function. Consider the process

$$Y^{\ell}_t: = X_0+ bt+\int_0^t\frac{\sigma}{1+\ell(s)}dW_s.$$

Then we have $(Y^{\ell}_{t\wedge \tau^{\ell}})_{t\ge 0}$ is a solution to $(\ast)$ if $\mathbb P[\tau^{\ell}>t]=m_t$ for all $t\ge 0$, where $\tau^{\ell}:=\inf\{t\ge 0: Y^{\ell}_t\le 0\}$. Thus it remains to calculate the probability $\mathbb P[\tau^{\ell}>t]$ in terms of $\ell$. Is there any reference for this computation?


1 Answer 1


This is a rough idea. Fix an arbitrary $T>0$. For $0\le t\le T$ and $x\ge 0$, define $u(t,x;T):=\mathbb P[\tau^{\ell}>T|Y^{\ell}_t=x]$. One has by definition $u(T,x;T)={\bf 1}_{\{x>0\}}$ for all $x\ge 0$ and $u(t,x;T)=0$ for all $0\le t\le T$.

For $0\le t<T$ and $x>0$, it follow from the tower property that

$$u(t,x;T)=\mathbb E\big[\mathbb E[{\bf 1}_{\{\tau^{\ell}>T\}}|Y^{\ell}_{s+h}]|Y^{\ell}_t=x\big]=\mathbb E\big[u(t+h,Y^{\ell}_{t+h};T)|Y^{\ell}_t=x\big], \quad \mbox{for all } 0<h<T-t.$$

Admitting the regularity of $u$, one has by Ito's formula

$$\frac{1}{h}\mathbb E\left[\int_{t}^{t+h}\left(\partial_tu(s,Y^{\ell}_s;T) + b\partial_xu(s,Y^{\ell}_s;T) +\frac{\sigma^2}{2(1+\ell(s))^2}\partial_{xx}^2 u(s,Y^{\ell}_s;T)\right)ds~\Big|Y^{\ell}_t=x\right]~~=~~0.$$

Letting $h\to 0$, it yields

$$\partial_tu(t,x;T) + b\partial_xu(t,x;T) +\frac{\sigma^2}{2(1+\ell(t))^2}\partial_{xx}^2 u(t,x;T)=0.$$

Assume $X_0=\bar x>0$. Then one has coupled equations:

\begin{equation} \partial_tu(t,x;T) + b\partial_xu(t,x;T) +\frac{\sigma^2}{2(1+\ell(t))^2}\partial_{xx}^2 u(t,x;T)=0,~ \forall 0\le t<T, x>0 \\ u(T,x;T)={\bf 1}_{\{x>0\}},~ \forall x\ge 0 \\ u(t,0;T) = 0,~ \forall 0\le t\le T \\ u(0,\bar x,;T) = \ell(T),~ \forall T\ge 0. \end{equation}


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.