10

Consider $L^2(\mathbb{R})$ as an Hilbert space with inner product $(\cdot ,\cdot)$. Define $$\psi_n(x)=e^{-\frac{x^2}{2}}H_n(x),$$ where $H_n(x)$ is the Hermite Polynomials. How do you show that $\{\psi_n(x)\}$ is a orthogonal basis for $L^2(\mathbb{R})$? (We can make it orthonormal by normalize it).

I showed $(\psi_i,\psi_j)=0$ for all $i\neq j$. To conclude the $\{\psi_n(x)\}$ forms a basis. We need to show that $\text{span}(\psi_n(x))$ is a dense subset of $L^2$. How do we prove that?

Arctic Char
  • 16,007
nagnag
  • 1,773
  • You prove that you can approximate any element by linear combinations of the elements in the basis. How exactly you do this, depends on the basis, of course. – Mariano Suárez-Álvarez Feb 20 '15 at 23:08
  • I edited the problem. However, this is a Hilbert space basis not vector space basis. Is showing linear independence any easier? – nagnag Feb 20 '15 at 23:09
  • If you read what I wrote, you'll notice that I used the word «approximate». – Mariano Suárez-Álvarez Feb 20 '15 at 23:10
  • @amathnerd : You already showed the independence when you showed that the familly is orthogonal – Tryss Feb 20 '15 at 23:10
  • If you know that the elements are ortjogonal, then they are linearly independent, by exactly the same argument that works over any inner product space, Hilbert or not. – Mariano Suárez-Álvarez Feb 20 '15 at 23:11
  • @MarianoSuárez-Alvarez Thanks, I forgot the orthogonality implies independence part. – nagnag Feb 20 '15 at 23:15
  • To show that ${ e^{-x^{2}/2}H_{n}}{n=0}^{\infty}$ is an orthogonal basis, it is enough to show (a) orthogonality and (b) denseness of the span of this set in $L^{2}(\mathbb{R})$. Denseness is shown by proving that if $g \in L^{2}(\mathbb{R})$ satisfies $\int{-\infty}^{\infty} g(x)e^{-x^{2}/2}H_{n}(x)dx = 0$ for all $n \ge 0$, then $g=0$ a.e.. A generalization of denseness is found in this problem: http://math.stackexchange.com/questions/1032900/xnfx-n-in-mathbbn-subset-l-2a-b-as-a-complete-system . A solution is outlined as a problem in a book of Kolmogorov and Fomin. – Disintegrating By Parts Feb 20 '15 at 23:29
  • @T.A.E.Thanks, I truly appreciate it – nagnag Feb 20 '15 at 23:30

1 Answers1

13

As you say, you need to prove that $$ \overline{\text{span}(\psi_n(x))} = L^2(\mathbb{R}). $$ For that, it suffices to prove that if $f\in L^2(\mathbb{R})$ satisfies $$ (f, \psi_n) = 0, $$ for all $n$, then $f = 0$.

Note that the linear span of Hermite polynomials is equal to the linear span of all polynomials (look at the degrees of Hermite polynomials), so it's enough to prove that $$ \int_{\mathbb{R}} f(x) x^n e^{-x^2/2} dx = 0, \; \forall n\geq 0, \Rightarrow f \equiv 0. $$ Now, use that the set of continuous functions with compact support is dense in $L^2(\mathbb{R})$, and by Weierstrass theorem, you can approximate a continuous function on a compact domain uniformly by polynomials. Then, take a sequence of continuous functions with compact support $\{g_n\}$ such that $$ g_n \rightarrow f\cdot e^{-x^2/2} \ \ \ \in L^2(\mathbb R) $$ and, for each $g_n$, a sequence of polynomials $P_{n,m}$ such that $$ P_{n,m} \cdot I_{[supp(g_n)]} \rightarrow g_n \quad \text{(uniformly in $m$)}, $$ where $I_{[supp(g_n)]}$ is the indicatrix of the support of $g_n$. Now consider $$ A_{n,m} = \int \left(P_{n,m}(x) \cdot I_{[supp(g_n)]}(x) - f(x)e^{-x^2/2}\right) \cdot f(x) e^{-x^2/2} dx.$$ By the triangular inequality and choosing $m(n)$ large enough for each $n$, we have $$A_{n,m(n)} \rightarrow 0, \quad \text{as }n\to \infty,$$ but also $$ A_{n,m(n)} \rightarrow -\int f(x)^2 e^{-x^2} dx, \quad \text{as }n \to \infty. $$ Then $$ 0 = -\int f(x)^2 e^{-x^2} dx \Rightarrow f \equiv 0 $$ which concludes the argument.

Arctic Char
  • 16,007
V.A.S
  • 174
  • 1
    Thank you but could you clearify why it it sufficient to prove that if $f \in L^2(\mathbf{R})$ satisfy $(f,\phi_n)=0$ for all $n$, then $f=0$? – TOMILO87 Aug 22 '22 at 18:49
  • 2
    @TOMILO87 Consider $(f,\phi_n) = 0 \ \forall n$ and $f \neq 0$, then $\frac{f}{\lVert f \rVert}$ is a new unit vector that is orthogonal to all $\phi_n$, from which follows that $\phi_n$ can't be a complete orthonomal system. – Jahi02 May 20 '23 at 20:04
  • Why $A_{n,m(n)} \rightarrow -\int f(x)^2 e^{-x^2} dx$ hold? – lyn Aug 06 '23 at 02:22
  • Is this proof really valid? – lyn Aug 16 '23 at 16:23