I am interested in the following integral $$\int_{-a}^adx\int_{-a}^a\mathop{\mathrm{d}y}xy\frac{1}{\sqrt{b}}\exp\left[-\frac{(x-y)^2}{2b}\right]\sqrt{\frac{2}{c}}\exp\left[-\frac{(x+y)^2}{4c}\right].$$ Does any one know how to evaluate the above integral? Mathematica was not able to give result for this.

  • $\begingroup$ Well I think it's $0$ for $c=b/2$, because it becomes an odd function in both $x$ and $y$. $\endgroup$
    – snulty
    Jun 23, 2016 at 21:42
  • 1
    $\begingroup$ Try $u=x+y\, ,v=x-y$ the integrals decouple into pieces which can be expresed in terns of error functions $\endgroup$
    – tired
    Jun 23, 2016 at 21:54
  • 1
    $\begingroup$ A bit a context? What is the reason for computing the probability that a normal random variable takes values over a rhombus centered at the origin? $\endgroup$ Jun 23, 2016 at 22:25
  • $\begingroup$ It is to compute some correlations. $\endgroup$
    – titanium
    Jun 24, 2016 at 11:22

1 Answer 1


Let's make this integral a little nicer. First, we deal with the parameters:


$$x=au, \qquad y=av, \qquad b=a^2 B, \qquad c=\frac{a^2 C}{2}$$

$$=\frac{2a^2}{\sqrt{BC}} \int_{-1}^1 \int_{-1}^1 u~v~\exp \left[-\frac{(u-v)^2}{2B} \right]~\exp \left[-\frac{(u+v)^2}{2C} \right]\mathrm{d}u~\mathrm{d}v$$

Let's forget about the coefficients and name the integral:

$$I_0(B,C)=\int_{-1}^1 \int_{-1}^1 u~v~\exp \left[-\frac{(u-v)^2}{2B} \right]~\exp \left[-\frac{(u+v)^2}{2C} \right]\mathrm{d}u~\mathrm{d}v$$

Separating the region into four obvious parts and introducing another integral:

$$I(B,C)=\int_0^1 \int_0^1 u~v~\exp \left[-\frac{(u-v)^2}{2B} \right]~\exp \left[-\frac{(u+v)^2}{2C} \right]\mathrm{d}u~\mathrm{d}v$$

We can easily see that:


Thus, we only need to find $I(B,C)$. Let's expand the squares in the exponentials:

$$I(B,C)=\int_0^1 \int_0^1 u~v~\exp (-\alpha ~u~v )~\exp \left[-\beta (u^2+v^2) \right]\mathrm{d}u~\mathrm{d}v$$

$$\alpha=\frac{1}{C}-\frac{1}{B}, \qquad \beta=\frac{1}{2C}+\frac{1}{2B}$$

Let's introduce a new integral:

$$J(\alpha,\beta)=\int_0^1 \int_0^1 \exp (-\alpha ~u~v )~\exp \left[-\beta (u^2+v^2) \right]\mathrm{d}u~\mathrm{d}v$$

$$I(B,C)=-\frac{\partial J(\alpha,\beta)}{\partial \alpha}$$

Now we notice that the function under the integral is symmetric with respect to the change $u \to v$, thus we can write:

$$J(\alpha,\beta)=2\int_0^1 \int_0^u \exp (-\alpha ~u~v )~\exp \left[-\beta (u^2+v^2) \right]\mathrm{d}u~\mathrm{d}v$$

Now we make a change of variable:

$$v=u t, \qquad dv=u~dt$$

$$J(\alpha,\beta)=2\int_0^1 \int_0^1 \exp (-\alpha ~u^2~t )~\exp \left[-\beta u^2(1+t^2) \right]u~\mathrm{d}u~\mathrm{d}t$$

Another change of variable will be:


$$J(\alpha,\beta)=\int_0^1 \int_0^1 \exp [-(\beta t^2+\alpha ~t+\beta)p ]\mathrm{d}p~\mathrm{d}t$$

The inner integral is elementary and we obtain a single integral:

$$J(\alpha,\beta)=\int_0^1 \frac{1-\exp [-(\beta t^2+\alpha ~t+\beta) ]}{\beta t^2+\alpha ~t+\beta}~\mathrm{d}t$$

$$=\frac{1}{\beta} \int_0^1 \frac{\mathrm{d}t}{(t+\gamma)^2 +1-\gamma^2}-e^{\beta (1-\gamma^2)} \frac{1}{\beta} \int_0^1 \frac{\exp [-\beta (t+\gamma)^2 ]}{(t+\gamma)^2 +1-\gamma^2}~\mathrm{d}t$$

$$\gamma=\frac{\alpha}{2 \beta}=\frac{B-C}{B+C}$$

The first part is an elementary integral (arctangent if $bc>0$).

The second part is related to the error function.

Let's get back to the initial integral and express it in the new form. Note that the exchange $B \to C$ is equivalent to $\alpha \to - \alpha$ and $\beta \to \beta$:


$$=2a^4\sqrt{\frac{2}{bc}} \left(-\frac{\partial J(\alpha,\beta)}{\partial \alpha}+\frac{\partial J(-\alpha,\beta)}{\partial \alpha} \right) =2a^4\sqrt{\frac{2}{bc}} \times$$

$$ \times \int_0^1 \left[ \frac{ 1-\left(\beta t^2 +\alpha t+\beta+1\right) e^{-(\beta t^2 +\alpha t+\beta)}}{(\beta t^2 +\alpha t+\beta)^2}-\frac{ 1-\left(\beta t^2 -\alpha t+\beta+1\right) e^{-(\beta t^2 -\alpha t+\beta)}}{(\beta t^2 -\alpha t+\beta)^2}~\right] t \mathrm{d}t $$

Numerically the first and last expressions are equal, which was checked with Mathematica for random values of $a,b,c$.


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.