I am looking for a time-dependent probability density $f(x,y,t)$ solving the equation $$-\frac{\partial f}{\partial t} = \alpha\cdot \big(y - F(x)\big)\frac{\partial f}{\partial x}+\beta\cdot \big(G(y)-x\big)\frac{\partial f}{\partial y},$$ where the coefficient functions $F$ and $G$ given by\begin{align}F(x) &= a_0 + a_1 x + o(x - x^{0}),\\ G(y) &=b_0 + b_1y + o(y-y^{0}).\end{align}

My question is: What general class of equations does this PDE belong to, and what would be your angle of attack on such problem (if it makes sense at all)?

Remark I: Under the assumption of existence of a solution, my initial angle of attack was to rewrite the equation as $$i \frac{\partial f}{\partial t} = L\,f$$ where $L$ is an operator acting on $f$ given by $L = -i\left[\alpha\big(y - F(x)\big)\frac{\partial}{\partial x}+\beta\big(G(y)-x\big)\frac{\partial}{\partial y}\right]$. Then - I suppose - we can formally express the density $f$ as $$f(x,y,t)=\sum_k c_ke^{-i\lambda_k t}\varphi_k(x,y),$$ where $\lambda_k$ and $\varphi_k$ are the eigenvalues and eigenfunctions corresponding to the operator $L$. I am thus led to considering the equations $$L(\varphi_k) = \lambda_k\,\varphi_k$$ and swap the solutions $\varphi_k$ and $\lambda_k$ back the into the formal solution. But I have not been able to find the eigenfunctions, perhaps my patience was too short.

Remark II: A necessary condition, I suppose, for the existence of a density solving the PDE is the existence of curves $x$ and $y$ solving the "predator-prey" equations $$\frac{dx}{dt} = \alpha\cdot (y - F(x)),\qquad \frac{dy}{dt} = \beta\cdot (G(y)-x).$$ If we neglect the little-o terms, by putting them equal to zero (is that completely illegal?), the curves $x$ and $y$ are given by \begin{align}x(t) &=-\frac{a^0}{a^1} + C_1 e^{-a^1\mu t} + \frac{1}{1-a^1}\left[\frac{b^0}{b^1}+\frac{a^0}{a^1b^1}-C_2 e^{b^1\lambda t}-\frac{C_4}{b^1}e^{-a^1\mu t}\right]\\ y(t) &= \frac{a^1}{1-a^1}\left[\frac{b^0}{b^1}+\frac{a^0}{a^1b^1}-C_5 e^{b^1\lambda t}-\frac{C_6}{b^1}e^{-a^1\mu t}\right] \end{align} where $C_1$, $C_2$, $C_3$, $C_4$, $C_5$, and $C_6$ are arbitrary constants.


1 Answer 1


The temporal part is easy. Since there is no explicit time dependence on the RHS f is seperable as far as f(x,y,t)= R(x,y)T(t). With a separation constant of k, T= $e^{-kt}$.

This leaves the spatial segment. I didn't work it out myself but I would try to use Laplace's transform to turn this into an ODE of one variable and algebraic in the other spatial coordinate.

Hope this helps.


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.