0
$\begingroup$

How can I compute the spectral fractional Laplacian of $(a-x)_+^\alpha$ on $\Omega = (0,1)$? Here the operator is defined as $$(-\Delta)^s u = c_{N,s} \int_0^\infty (e^{t\Delta_N}u(x) - u(x)) t^{-1 - s} dt ,$$ where $e^{t\Delta_N}u$ is the solution of the heat equation with homogeneous Neumann boundary conditions on $\Omega$.

$\endgroup$
2
  • $\begingroup$ I would be surprised if a nice closed-form expression existed. A natural approach would be to expand $(a-x)_+^\alpha$ in the Fourier cosine series $\sum a_n \cos(n \pi x)$ and use the fact that $(-\Delta_N)^s [\cos(n \pi x)] = (n \pi)^{2s} \cos(n \pi x)$. I do not know if there is a simple expression for $a_n$, though. $\endgroup$ Commented Feb 16, 2021 at 1:23
  • $\begingroup$ Right hand integral looks like an operator with respect to $t$: $\int^\infty_0(u(x,t)-u(x,0))t^{-1-s}dt$ with parameter $x$. $\endgroup$
    – user171871
    Commented Feb 16, 2021 at 8:53

2 Answers 2

1
$\begingroup$

The operator $(-∆)^s$ is the Fourier multiplier $\vert\xi\vert^{2s}$ and thus deserves to be denoted by $\vert D\vert^{2s}$. Now the distribution $u_\alpha(x)=x_+^\alpha$ is homogeneous with degree $\alpha$ and thus has a Fourier transform homogeneous with degree $-1-\alpha$. We have in fact $$ \hat{u}_\alpha(\xi)=c_\alpha (\xi-i0)^{-1-\alpha}. $$ You are interested in $v_\alpha(x)=u_\alpha(x-t)$ so that $$ \vert\xi\vert^{2s}\hat{v}_\alpha(\xi)=e^{-it \xi}c_\alpha (\xi-i0)^{-1-\alpha}\vert\xi\vert^{2s}, $$ and $ (\vert D\vert^{2s} v_\alpha)(x)=\int e^{i(x-t) \xi}c'_\alpha (\xi-i0)^{-1-\alpha}\vert\xi\vert^{2s}d\xi =(\vert D\vert^{2s} u_\alpha)(x-t). $ Translations commute with derivations, including fractional versions.

$\endgroup$
1
  • 1
    $\begingroup$ I think Jay is asking about a different operator: the fractional power of the Neumann Laplacian on $(0,1)$. $\endgroup$ Commented Feb 16, 2021 at 1:21
0
$\begingroup$

Let us denote the corresponding eigensystem $\{\lambda_n,\phi_n\}$. If $u(x)=\sum\limits_nu_n\phi_n(x)$, then we have $u(x,t)=\sum\limits_n\exp(-\lambda_n t)u_n\phi_n(x)$. Since $\int\limits^\infty_0(\exp(-\lambda_n t)-1)t^{-1-s}=-\lambda^s_n\frac{\Gamma(1-s)}{s}$ we obtain $$ \text{R.H.I.}=-C_{s,N}\frac{\Gamma(1-s)}{s}\sum\limits_n\lambda^s_nu_n\phi_n(x)$$

$\endgroup$

You must log in to answer this question.

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