I have a simple gaussian integral: $\int^{\infty}_{-\infty}dx\:e^{i\alpha x^2}$.

If $\alpha \in \mathbb{R}$, then:

$$ \int^\infty_{-\infty} dx\; e^{i \, \alpha x^2} = \sqrt{\frac \pi {-i \alpha}} \qquad \qquad \alpha <0 $$

Now if $\alpha \in \mathbb{C}$ then we obtain the same answer but with different conditions:

$$ \int^\infty_{-\infty} dx\; e^{i \, \alpha x^2} = \sqrt{\frac \pi {-i \alpha}} \qquad \qquad Im(\alpha)>0 $$

These can be combined into a simple answer with an OR statement:

$$ \int^\infty_{-\infty} dx\; e^{i \, \alpha x^2} = \sqrt{\frac \pi {-i \alpha}} \qquad \qquad Im(\alpha)=0 \, \& \, Re(\alpha) <0 \quad|| \quad Im(\alpha)>0 $$

When I I ask Mathematica to solve this for me

Integrate[E^(I x^2 a), {x, -∞, ∞}]

Mathematica returns only one of these cases:

ConditionalExpression[Sqrt[π]/Sqrt[-I a], Im[a] > 0]

I have tried specifying $Im(\alpha)\geq 0$ in the Assumptions but Mathematica disregards this and gives the same result. It can of course find the answer for $\alpha \in \mathbb{R}$ but it cannot seem to give a fully general answer where both conditions are present.

  • $\begingroup$ Where did this integral come up? $\endgroup$ – mjw Mar 7 '19 at 18:45

If $\alpha$ is complex, then yes, The imaginary part of alpha should be strictly greater than zero. If $\alpha$ is real, does it converge? If you complete the contour in the complex plane, with a semicircle, and replace $x$ by $z=x+ i y$, I do not see the integral converging along the semicircle.


Okay, I looked the integral up in Gradshteyn and Ryzhik, Table of Integrals, Series, and Products, 6$^\textrm{th}$ edition, p. 333. Looks like the integral for $\alpha$ real converges for $\alpha<0$ but with limits zero to infinity.

$\displaystyle \int_0^\infty e^{-i\lambda x^2} dx = \frac{1}{2} \sqrt{\frac{\pi}{\lambda}} e^{-i\pi/4}, \quad (\lambda>0)$.

We can infer from this (let $w=-x$ $\Rightarrow$ $dw = - dx$)

$\displaystyle \int_{-\infty}^\infty e^{-i\lambda x^2} dx = \sqrt{\frac{\pi}{\lambda}} e^{-i\pi/4}, \quad (\lambda>0)$.

Replacing $\lambda$ by $-\lambda$ we get the complex conjugate:

$\displaystyle \int_{-\infty}^\infty e^{i\lambda x^2} dx = \sqrt{\frac{\pi}{-\lambda}} e^{i\pi/4}, \quad (\lambda<0)$.

Combining these:

$\displaystyle \int_{-\infty}^\infty e^{-i\lambda x^2} dx = \sqrt{\frac{\pi}{\lambda}} e^{ - i\pi \,\textrm{sign }{(\lambda)} /4}, \quad (\lambda \ne 0, \lambda \in \Re)$.

This is consistent with what Mathematica (Version, Mac OS X) gives:

Assuming[Element[a,Reals],Integrate[Exp[I a x^2],{x,-Infinity,Infinity}]


Sqrt[Pi]/2 (1+ I Sign[a])/Sqrt[Abs[a]]

If anybody has an idea how to compute this integral with contour integration (or otherwise) from first principals, that would be interesting!

Also, we still haven't answered why Mathematica assumes that $\alpha$ is not real if $\alpha \in \mathbb{C}$!

$\mathbf{ADDITIONAL \,\, UPDATE:}$

This will take into account each case, and output the result or indicate if the integral does not converge:

q[a_] := Integrate[Exp[I a x^2], {x, -Infinity, Infinity}]; 
integral[a_] := If[Element[a, Reals], Assuming[Element[a, Reals], q[a]], q[a]];

Here are a few examples:

enter image description here

  • 1
    $\begingroup$ I don't think Mathematica gave the proper response $\alpha<0$. If it converges for $\alpha<0$ why not $\alpha>0$. and does it converge? $\endgroup$ – mjw Mar 7 '19 at 15:38
  • $\begingroup$ I have assumed $x \in \mathbb{R}$ so convergence in that sense has not been accounted for. $\endgroup$ – OldTomMorris Mar 7 '19 at 16:28
  • $\begingroup$ Well, I believe Gradshteyn and Ryzhik's listing. We can then make some substitutions to compute what happens for $\alpha>0$. Mathematica (ver. 11.2) gives an answer that is consistent. Obviously, the integral diverges for $\alpha=0$. Mathematica's result $\rightarrow \infty$ there. $\endgroup$ – mjw Mar 7 '19 at 18:44
  • $\begingroup$ "Also, we still haven't answered why Mathematica assumes that 𝛼 is not real if 𝛼 ∈ ℂ!". This is exactly my problem! $\endgroup$ – OldTomMorris Mar 8 '19 at 10:26
  • 1
    $\begingroup$ Agreed! I've updated my answer to output what we would have liked to have seen for cases when $\alpha$ is real and when $\alpha$ has a nonzero imaginary part. $\endgroup$ – mjw Mar 8 '19 at 18:36

By default, Mathematica assumes all symbols are complex valued, so this is what you get if you don't specify. You can see all the variations by making assumptions:

Integrate[E^(I x^2 a), {x, -∞, ∞}, Assumptions -> #] 
          & /@ {a ∈ Complexes, a ∈ Reals, Im[a] == 0, 
                Im[a] < 0, Im[a] > 0, Re[a] == 0, Im[a] >= 0, a == 0, a < 0}

One of these doesn't converge and another is equal to infinity, but they do give the full range of possibilities.


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.