2
$\begingroup$

I noticed that

$$t^2 \frac{\partial^3}{\partial t^3}\Delta_t(s)+s^2 \frac{\partial}{\partial s} \Delta_t(s)=0 $$

is satisfied by

$$\Delta_t(s)= - \sqrt{\frac{t}{s}}Y_1{(4\pi\sqrt{ts})}+ \sqrt{\frac{t}{s}}K_1(4\pi \sqrt{ts})$$

for $Y_1$ and $K_1$ Bessel functions.

I noticed this looked similar to the "error term" in the Dirichlet divisor problem

$$\Delta(t)= -\sum_{s \in \Bbb N} d(s) \sqrt{\frac{t}{s}}Y_1{(4\pi\sqrt{ts})}-\sum_{s \in \Bbb N} d(s)\sqrt{\frac{t}{s}}K_1(4\pi \sqrt{ts})$$

I know from number theory that the order of magnitude of $\Delta(t)$ as $t\to \infty$ is close to $O(t^{1/4+\epsilon}).$ Making this exact is the notorious Dirichlet divsor problem.

I wonder if there are techniques from the theory of pde's that I can use to show this asymptotic holds or at least get close.

I observed that

$$t^2\frac{\partial^3}{\partial t^3}\sum_{n=1}^\infty \Psi_n(t,s)=s^2\frac{\partial}{\partial s}\sum_{n=1}^\infty \Psi_n(t,s)+\sum_{n=2}^\infty b(n)\frac{\partial}{\partial s} \Psi_n(t,s)$$

has a solution

$$\Psi_n(t,s)=2 \sqrt{\frac{tn}{s}}K_1(2\sqrt{t ns})$$

where the coefficients are

$$b(n)=\big\lbrace 2,12,24,40,60,84, \cdot\cdot\cdot\big\rbrace.$$

which I noticed correspond to twice the pronic/oblong numbers (if you throw out the first term $2$).

I thought about writing out the coefficients of the sequence $r(n)$ (to be determined) for the PDE corresponding to $Y_1$ as we did for $b(n)$ corresponding to $K_1$.

Then through linearity of solutions we can obtain a PDE involving both $r(n)$ and $b(n)$:

$$t^2\frac{\partial^3}{\partial t^3}\sum_{n=1}^\infty \chi_n(t,s)+s^2\frac{\partial}{\partial s}\sum_{n=1}^\infty \chi_n(t,s)=\sum_{n=2}^\infty r(n)\frac{\partial}{\partial s} \chi_n(t,s)+\sum_{n=2}^\infty b(n)\frac{\partial}{\partial s} \chi_n(t,s)$$

where $$\chi_n(t,s)=-2 \sqrt{\frac{tn}{s}}Y_1{(2\sqrt{tns})}+2 \sqrt{\frac{tn}{s}}K_1(2 \sqrt{tns})$$

Is $r(n)$ a known arithmetic sequence?

$\endgroup$
19
  • $\begingroup$ I'm confused. Is the index $s$ in $\Delta(t)$ related to the variable $s$ in the PDE? $\endgroup$
    – Gonçalo
    Commented May 6 at 2:25
  • 1
    $\begingroup$ @Gonçalo oh yes - I just clarified the matter, let me know if you understand. $\endgroup$
    – zeta space
    Commented May 6 at 4:04
  • $\begingroup$ @Gonçalo essentially I am summing over the $s$ variable in the solution to the original diff eq. $\endgroup$
    – zeta space
    Commented May 6 at 4:06
  • 3
    $\begingroup$ How is $d(s)$ defined for non integer $s$? $\endgroup$
    – Gonçalo
    Commented May 6 at 4:09
  • $\begingroup$ I don't see how you can differentiate w.r.t. $s$ while also summing over $s$ as if it is a discrete index $\endgroup$
    – whpowell96
    Commented May 7 at 18:41

1 Answer 1

1
$\begingroup$

For

$$\Psi_n(t,s)=2 \sqrt{\frac{n t}{s}} K_1\left(2 \sqrt{t n s}\right)\tag{1}$$

one has

$$t^2 \frac{\partial^3\, \Psi_n(t,s)}{\partial t^3}=n\, s^2 \frac{\partial\, \Psi_n(t,s)}{\partial s}\tag{2}$$

s0

$$t^2 \frac{\partial^3}{\partial t^3}\left(\sum\limits_{n=1}^{\infty} \Psi_n(t,s)\right)=s^2 \sum\limits_{n=1}^{\infty} n\, \frac{\partial\, \Psi_n(t,s)}{\partial s}\\\qquad\qquad=s^2 \frac{\partial}{\partial s} \left(\sum_{n=1}^\infty \Psi_n(t,s)\right)+s^2 \sum_{n=1}^\infty (n-1)\frac{\partial\, \Psi_n(t,s)}{\partial s}\tag{3}$$

so $b(n)=n-1$.


For $$\varphi_n(t,s)=2 \sqrt{\frac{n t}{s}} Y_1\left(2 \sqrt{t n s}\right)\tag{4}$$

one has

$$t^2 \frac{\partial^3 \varphi_n(t,s)}{\partial t^3}=-s^2 n \frac{\partial\, \varphi_n(t,s)}{\partial s}\tag{5}$$

so $$t^2 \frac{\partial^3}{\partial t^3}\left(\sum\limits_{n=1}^{\infty} \varphi_n(t,s)\right)=-s^2 \sum\limits_{n=1}^{\infty} n\, \frac{\partial\, \varphi_n(t,s)}{\partial s}\\\qquad=-s^2 \frac{\partial}{\partial s}\left(\sum\limits_{n=1}^{\infty} \varphi_n(t,s)\right)-s^2 \sum\limits_{n=1}^{\infty} (n-1)\, \frac{\partial\, \varphi_n(t,s)}{\partial s}\tag{6}$$

so $r(n)=b(n)=n-1$.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .