I am interested in the numerical calculation of $$ F(\eta)=\frac{2}{π}\int_0^{+\infty}\sin(t\eta-t^3)\frac{dt}{t}\quad\text{for $\eta\ge 0$}. $$ I believe that the function $F$ is bounded, but I do not know its upper bound. It is easy to get $F(0)=-1/3$ and various other values for $\eta \in [0,40]$ but getting above the value $40$ for $\eta$ seems to raise computational difficulties for Mathematica. It is also interesting to see that $F$ takes values strictly above 1 with the seemingly largest value $1.5$ for $\eta =3$.

As a motivation, one may note that with $g(t)= e^{it^3}\text{pv}\frac{1}{πt}$, the operator of convolution with $g$ is bounded on $L^2(\mathbb R)$ whenever its Fourier transform is bounded. We have here $F=\hat g$, up to some normalization, and a precise bound in $L^\infty$ for $F$ would provide the $\mathcal B(L^2)$ operator-norm of that convolution.


Nemo's representation of $F(\eta)$ in terms of a hypergeometric function can be evaluated without difficulty for large $\eta$: $$F(\eta)=\frac{\sqrt{3} {\eta}^2 \Gamma \left(\frac{2}{3}\right) \; _1F_2\left(\frac{2}{3};\frac{4}{3},\frac{5}{3};-\frac{{\eta}^3}{27}\right)}{6\pi }-\frac{12 {\eta} \; _1F_2\left(\frac{1}{3};\frac{2}{3},\frac{4}{3};-\frac{{\eta}^3}{27}\right)}{6\Gamma \left(-\frac{1}{3}\right)}-\frac{1}{3}$$

$$F(\eta)\rightarrow 1+\sqrt[4]{3} \sqrt{2/\pi}\frac{1}{\eta^{3/4}} \left[\sin \left(\frac{2 \eta^{3/2}}{3 \sqrt{3}}\right)-\cos \left(\frac{2 \eta^{3/2}}{3 \sqrt{3}}\right)\right],\;\;\eta\gg 1.$$

see plot, blue is $F(\eta)$, gold is the large-$\eta$ approximation (nearly indistinguishable for $\eta>5$). So $F(\eta)$ oscillates around 1, with an amplitude that decays as $\eta^{-3/4}$. The maximum 1.5487 is reached at $\eta= 3.37213$.

| cite | improve this answer | |
  • $\begingroup$ Great! Thank you very much. $\endgroup$ – Bazin Feb 11 at 9:14
  • $\begingroup$ In your asymptotic expression, it seems that the term $\sqrt{2π}$ should be replaced by $\sqrt{2/π}$. $\endgroup$ – Bazin Feb 11 at 12:40
  • $\begingroup$ Sorry to bother you again with that matter. I would be very grateful if you could double-check your constant $\sqrt{2\pi}$, which I believe should be replaced by $\sqrt{2/\pi}$. Thanks. $\endgroup$ – Bazin Feb 13 at 10:31
  • $\begingroup$ you are right, it was my typing error, apologies for the confusion. $\endgroup$ – Carlo Beenakker Feb 13 at 10:42
  • $\begingroup$ Thanks a lot for caring to check this; of course I renew my thanks for your very helpful and thorough answer to my question. $\endgroup$ – Bazin Feb 13 at 16:05

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.