Let $f = \sum_{n=0}^{\infty} a_n e_n $ where $e_n$ are an ONB of $L^2[0,1].$

Now assume we have that $$\frac{d}{dx}e_n = \lambda_n e_n.$$

Assume $f \in H^1[0,1],$ so i.e. $||f'||_{L^2} < \infty$

I want to say now that $$\frac{d}{dx} f = \sum_{n} a_n \lambda_n e_n$$ but I am not sure how to justify the interchange of summation and differentiation.

So I am trying to make rigorous here what physicists always do.

  • $\begingroup$ I think you mean $L^2$ in your first sentence? $\endgroup$ – Jose27 Feb 26 '15 at 23:25
  • $\begingroup$ @Jose27 ah, yes of course, thank you. $\endgroup$ – Roadrunner34 Feb 27 '15 at 0:37

The following answers your question, but makes some additional assumptions (see below).

Since $f' \in L^2$, and since $(e_n)_n$ is an ONB of $L^2$, we know

$$ f' = \sum_n \langle f', e_n\rangle e_n. $$

So your claim is equivalent to $\langle f', e_n \rangle = \lambda_n a_n$ for all $n$. By using partial integration, we get

\begin{eqnarray*} \langle f', e_n \rangle &=& \int_0^1 f'(x) \overline{e_n (x)} \, dx \\ &=& [f \cdot \overline{e_n}] \bigg|_0^1 - \int f(x) \overline{e_n ' (x)} \, dx \\ &=& [f \cdot \overline{e_n}] \bigg|_0^1 - \overline{\lambda_n} \int f(x) \overline{e_n(x)} \,dx \\ &=& [f \cdot \overline{e_n}] \bigg|_0^1 - \overline{\lambda_n} a_n . \end{eqnarray*}

If we now make the additional assumption that $f$ and $e_n$ are periodic of period $1$ (or if $f(0) = f(1) = 0$ or $e_n (0) = e_n(1) = 0$), the boundary terms vanish.

In this case, your claim is equivalent to $\lambda_n = -\overline{\lambda_n}$, i.e. to the condition that $\lambda_n$ is purely imaginary. That this is the case can be checked (under the same assumptions as above) by choosing $f = e_n$.


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.