Why does it seem that boundary terms vanish when using Fourier transform instead of integrating by parts

254 Views Asked by At

Ok after plowing through MSE posts my confusion has finally reached its climax.

Say I would like to obtain a formula as simple as

$$\int_{\mathbb{R}} f \partial_x \overline{g} dx = -\int_{\mathbb{R}} \partial_x f \overline{g} dx$$

If $f,g,f',g'$ are $L^2(\mathbb{R})$ functions, then these integrals are well-defined. Using the Fourier transform we can even prove this formula, using $\widehat{f'} = -i\xi\widehat{f}$. So far so good.

But what about boundary terms?

By integration by parts, we would obtain

$$\int_{\mathbb{R}} f \partial_x \overline{g} dx = f\overline{g}|_{-\infty}^{\infty}-\int_{\mathbb{R}} \partial_x f \overline{g} dx$$

So in order to reconcile these two equations, we need $\lim_{|x|\rightarrow \infty} f\overline{g} = 0$. But this is certainly not the case for arbitrary $L^2(\mathbb{R})$ functions. So what the heck is going on?

Since I assume $f,g \in H^1(\mathbb{R})$ we know that $f$ and $g$ already go to zero at infinity. But then does that mean boundary terms always vanish in $H^1(\mathbb{R})$?

But then, when would there be boundary terms in the first place? In order to define these integrals, I need the derivatives to lie in $L^2$, otherwise something like $\int f'\overline{g}$ doesn't make sense.

So in which situation would the boundary terms not vanish?

What am I not seeing here?

Am I going insane?

1

There are 1 best solutions below

4
On BEST ANSWER

The Schwartz functions are dense in $L^2$, so it suffices that these operations can be done on them, and we can then extend to all of $L^2$. A function $f ∈ C^∞(ℝ^n)$ is Schwartz if for all multi-indices $α,β$ there exists constants $C_{αβ}$ such that $$ \sup_{x∈ℝ^n}|x^β∂^αf(x)| ≤ C_{αβ}. $$ In the one-dimensional case, $α$ and $β$ are just natural numbers. In particular, this means that $f(x) → 0$ as $|x|→∞$, and there will hence never be any boundary terms.

As you may be aware of, the usual formula for the Fourier transform, $\hat f(ξ) = ∫e^{-ix·ξ}f(x)\,\mathrm dx$, is not well-defined on $L^2$. It is, however, well-defined on Schwartz functions, and this allows us to extend the Fourier transform to $L^2$.