Suppose $f,g$ are real valued on $\mathbb{R}$ (and no further restrictions apart from the obvious requirement that the integrals exist), then when does $\displaystyle\int f(x)g(x)\,dx = \int f(x) \, dx \int g(x) \, dx$? (all integrals indefinite)

This question was posed on another site, and I was wondering whether there are any large classes of functions $f,g$ that work.

Aside from the trivial solution, things like $f(x) = e^{nx}, g(x) = e^{\frac{n}{n-1}x}$ $(n\not= 1)$ work by inspection.

I've tried re-casting the problem as:

$$\int F^2 - \left(\int F\right)^2 = \int G^2 - \left(\int G\right)^2$$

With $F=f+g, G=f-g$, to introduce some symmetry. This gives rise to similarities with the variance formulae but nothing more than that.

Using power series $F(x) = \displaystyle\sum_{n=0}^{\infty} a_nx^n, G(x) = \sum_{n=0}^{\infty} b_nx^n$, and the cauchy product, this equation boils down to:

$$\sum_{m=0}^n \frac{(n-m)(a_ma_{n-m} - b_mb_{n-m})}{(n+1)(n-m+1)}= 0\qquad \forall n\geq 0$$

But again, progress is limited.

  • $\begingroup$ From your recasting, I would rather try to solve $\int F^2 - (\int F)^2 = H$ for $F$ $\endgroup$
    – user14972
    May 21, 2014 at 16:45

2 Answers 2


Differentiate and divide by $f(x) g(x)$, and you get $$ \dfrac{1}{f(x)} \int f(x)\; dx + \dfrac{1}{g(x)} \int g(x)\; dx = 1 $$ If $F(x) = \int f(x)\; dx$ and $G(x) = \int g(x)\; dx$, this says $$ \dfrac{d}{dx} \ln F(x) = \dfrac{F'(x)}{F(x)} = \dfrac{G'(x)}{G'(x) - G(x)}$$ and thus $$F(x) = c \exp \left( \int \dfrac{G'(x)}{G'(x) - G(x)}\; dx \right)$$ or $$ f(x) = \dfrac{c\; g(x)}{g(x) - \int g(x)\; dx} \exp \left( \int \dfrac{g(x)}{g(x) - \int g(x)\; dx}\; dx \right)$$ Since this was obtained by differentiating the original equation, you may have to adjust the constants of integration to make this work.

  • $\begingroup$ Other details to deal with include points where $f(x)=0$ or $g(x)=0$, or where $G'(x) = G(x)$. $\endgroup$
    – user14972
    May 21, 2014 at 16:41
  • $\begingroup$ Assuming $g$ is analytic and is not a constant times the exponential function, such points will be isolated singularities, and the equation will be true away from those singularities. $\endgroup$ May 22, 2014 at 1:37
  • $\begingroup$ Hmmm... poles of $g$ and $g/(g - G)$ may become branch points of their integrals, which do complicate the matter: you'll have to be careful about the branches of $G$ and $F$. $\endgroup$ May 22, 2014 at 1:55

One class of examples satisfy an "orthonormality" condition.

For example, $f_n(x) = e^{1 2 \pi n x}$ on $[0,1]$. Then $\int f_n = 0$ when $n \ne 0$ and $\int f_n f_m = 0$ when $n \ne -m$. Therefore $\int (f_n f_m) = 0 = \int f_n \int f_m$ when $n \ne -m$.


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.