Laplace coefficients are Fourier coefficients used in Celestial mechanics calculations

$$ b^n_s (\alpha) \equiv {1 \over \pi} \int_0^{2\pi} {\cos n \phi \over (1 - 2 \alpha \cos \phi + \alpha^2)^s} d \phi $$

with $s = i + 1/2$ (a half integer) and $0<\alpha <1$.

The function $(1 - 2\alpha \cos \phi + \alpha^2)^{-s}$ is analytic (locally) so the Fourier coefficients decay rapidly. For large $n$ and $\alpha$ not small, I think that $b^n_{1/2} \sim c_1 e^{-c_2(1-\alpha)n}$ (with constants $c_1,c_2$). How do I show that this is true and how do I derive constants $c_1,c_2$? If this was a bad guess is there an exponential function that does approximate the coefficients at large $n$?


1 Answer 1


Let $$ f(x,\alpha)=\frac1{c(x)(1+\alpha^2 x^2)(1-2\alpha \cos x+\alpha^2)^s}, $$ where $$c(x)=\sum_{n=-\infty}^\infty\frac1{1+\alpha^2 (x+2\pi n)^2}.$$ This function belongs to $L_2(\mathbb R)$, has an analytic continuation in the strip $|\Im z|<a=\log \alpha^{-1}$ and its Fourier transform at the integer points $\xi=n$ is equal to $b_s^n(\alpha)$. It is known that the wider is the strip the faster is the decay of the Fourier transform, namely as $e^{-b|\xi|}$ for $b\in(0,a)$, see the reference in the answer here. So a rough result implied from there seems to be this: for any $c_2<\log\alpha^{-1}\ $ there exists $c_1>0$ such that $|b_s^n(\alpha)|\le c_1 e^{-c_2n}$.


But the same idea applied to the circle and not the line gives a simpler solution. Consider the function $$f(z)=\frac1{(1+\alpha^2-\alpha(z+1/z))^s}.$$ Then $$f(e^{i \phi})=\frac1{(1-2\alpha \cos \phi+\alpha^2)^s}.$$ The denominator of $f$ has roots $\alpha$, $1/\alpha$. So $f$ is analytic in the ring $R=\{\alpha<|z|<\alpha^{-1}\}\ $. Hence $$f(z)=\frac12\sum_{n=-\infty}^\infty b_s^n(\alpha)z^n,\quad z\in R.$$ From the Cauchy root test for convergence it follows that $\limsup_{n\to\infty} |b_s^n(\alpha)|^{1/n}= \alpha\ $. This can be written in exponential form as desired above.

  • $\begingroup$ @Alice The function is good because its Fourier transform at the integer values gives $b^n_s$. $\endgroup$
    – Andrew
    Jul 13, 2011 at 21:36

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.