3
$\begingroup$

In an effort to solve a delay partial differential equation $$\partial_t f(t,x)= \Phi(x) f(t,x)+\Psi(x) f(t,x-\alpha),$$ with $$f(0,x)=1,\hspace{0.3cm} f(t,0)=1$$ Where $\alpha$ is the delay ( a real number) and in particular I have $\Phi(x)=e^{-x}+e^{x}-2$ and $\Psi(x)=1-e^{-x}$,

I used Laplace transform $$ L(f)=\int_{0}^{t} e^{-st} f(x,t) dt: =U(s,t) $$ and I got a delay equation with no derivatives involved : $$s\hspace{0.1cm}U(s,x)-1 = \Phi(x) U(s,x)+\Psi(x) U(s,x-\alpha)$$

This is where I stopped. Does anyone know how I can find an explicit formula for $U(s,x)$ from the last equation? Or is there any other way to solve the delay PDE expressed on top? Obviously the separation of variables method doesn't work here because it doesn't satisfy initial and boundary conditions.

$\endgroup$
4
  • 2
    $\begingroup$ I would not call this a delay equation, and a PDE is also not a good name. This is a nonlocal equation, your "delay" is not in $t$... $\endgroup$ Nov 6, 2018 at 7:07
  • $\begingroup$ I cannot check it now, but did you try separating the variables? $\endgroup$ Nov 6, 2018 at 7:08
  • $\begingroup$ Closely related: mathoverflow.net/questions/315480/… $\endgroup$ Nov 17, 2018 at 2:00
  • $\begingroup$ The condition $f(t,0)=1$ is redundant. Both $\Phi(0)$ and $\Psi(0)$ are zero, so $f(t,0)=f(0,0)=1$. I can see no reason why there should be a closed form solution. $\endgroup$ Mar 31, 2021 at 18:30

1 Answer 1

1
$\begingroup$

Firstly, we remove the $\Phi$ contribution by writing $$ f(t,x)=e^{t\Phi(x)}h(t,x). $$ This gives $$ \partial h(t,x)=e^{-t\Phi(x)}\Psi(x)e^{-t\Phi(x-\alpha)}h(t,x-\alpha). $$ that grants that $h(t,0)=1$ and $h(0,x)=1$. Then, we notice that $$ h(t,x-a)=\sum_{n=0}^\infty\frac{(-\alpha)^n}{n!}h^{(n)}(t,x) $$ that means that the equation is non-local. This can be rewritten formally as $$ h(t,x-\alpha)=e^{-\alpha\partial_x}h(t,x). $$ If we take $$ h(t,x)=\int_{-\infty}^\infty\frac{dk}{2\pi}e^{ikx}{\tilde h}(t,k) $$ it is easy to see that $$ \int_{-\infty}^\infty\frac{dk}{2\pi}e^{ikx}\partial_t{\tilde h}(t,k)= A(t,x)\int_{-\infty}^\infty\frac{dk}{2\pi}e^{ik(x-a)}{\tilde h}(t,k) $$ where we have set $A(t,x)=e^{-t\Phi(x)}\Psi(x)e^{-t\Phi(x-\alpha)}$. Therefore, $$ \int_{-\infty}^\infty dxe^{-ik'x}\int_{-\infty}^\infty\frac{dk}{2\pi}e^{ikx}\partial_t{\tilde h}(t,k)= \int_{-\infty}^\infty dxe^{-ik'x}A(t,x)\int_{-\infty}^\infty\frac{dk}{2\pi}e^{ik(x-\alpha)}{\tilde h}(t,k) $$ i.e. $$ \partial_t{\tilde h}(t,k')=\int_{-\infty}^\infty\frac{dk}{2\pi}{\tilde A}(t,k'-k)e^{-ik\alpha}{\tilde h}(t,k) $$ being $$ {\tilde A}(t,k'-k)=\int_{-\infty}^\infty dxe^{-i(k'-k)x}A(t,x). $$ This result says that this equation can be amenable to numerical methods but is not solvable exactly.

$\endgroup$

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.