Let E be the vector space of functions f f f defined and continuous on R \mathbf{R} R , and such that the integral ∫ − ∞ + ∞ f ( x + ξ ) e − ξ 2 / 2 d ξ \int_{-\infty}^{+\infty} f(x + \xi) e^{-\xi^2/2} d\xi ∫ − ∞ + ∞ f ( x + ξ ) e − ξ 2 /2 d ξ converges for all x ≥ 0 x \geq 0 x ≥ 0 . The operator U defined by
U λ ξ ( f ( ξ ) ) = 1 2 π ∫ − ∞ + ∞ e − ξ 2 / 2 f ( x + ξ ) d ξ
U_\lambda^\xi \left( f(\xi) \right) = \frac{1}{\sqrt{2\pi}} \int_{-\infty}^{+\infty} e^{-\xi^2/2} f(x + \xi) d\xi
U λ ξ ( f ( ξ ) ) = 2 π 1 ∫ − ∞ + ∞ e − ξ 2 /2 f ( x + ξ ) d ξ
is then defined on E and is clearly a composition operator. The space E contains all the exponentials e λ x e^{\lambda x} e λ x (λ \lambda λ arbitrary complex), and one has
u ( λ ) = 1 2 π ∫ − ∞ + ∞ e − ( ξ 2 / 2 ) + λ ξ d ξ = 1 2 π e λ 2 / 2 ∫ − ∞ + ∞ e − ( ξ − λ ) 2 / 2 d ξ = e λ 2 / 2
u(\lambda) = \frac{1}{\sqrt{2\pi}} \int_{-\infty}^{+\infty} e^{-(\xi^2/2) + \lambda \xi} d\xi \\
= \frac{1}{\sqrt{2\pi}} e^{\lambda^2/2} \int_{-\infty}^{+\infty} e^{-(\xi - \lambda)^2/2} d\xi = e^{\lambda^2/2}
u ( λ ) = 2 π 1 ∫ − ∞ + ∞ e − ( ξ 2 /2 ) + λ ξ d ξ = 2 π 1 e λ 2 /2 ∫ − ∞ + ∞ e − ( ξ − λ ) 2 /2 d ξ = e λ 2 /2
(cf. III, p. 120, exerc. 24, and VII, p. 313, formula (22)). One has n ! α n = U 0 ξ ( ξ n ) = 1 2 π ∫ − ∞ + ∞ e − ξ 2 / 2 ξ n d ξ n! \alpha_n = U_0^\xi (\xi^n) = \frac{1}{\sqrt{2\pi}} \int_{-\infty}^{+\infty} e^{-\xi^2/2} \xi^n d\xi n ! α n = U 0 ξ ( ξ n ) = 2 π 1 ∫ − ∞ + ∞ e − ξ 2 /2 ξ n d ξ . For every integer n n n one can write
∑ k = 0 n ∫ − ∞ + ∞ ∣ λ ξ ∣ k k ! e − ξ 2 / 2 d ξ ≤ 2 ∫ 0 + ∞ e − ( ξ 2 / 2 ) + ∣ λ ∣ ξ d ξ .
\sum_{k=0}^n \int_{-\infty}^{+\infty} \frac{|\lambda \xi|^k}{k!} e^{-\xi^2/2} d\xi \leq 2 \int_0^{+\infty} e^{-(\xi^2/2) + |\lambda| \xi} d\xi.
k = 0 ∑ n ∫ − ∞ + ∞ k ! ∣ λ ξ ∣ k e − ξ 2 /2 d ξ ≤ 2 ∫ 0 + ∞ e − ( ξ 2 /2 ) + ∣ λ ∣ ξ d ξ .
The series ∑ n = 0 ∞ e − ξ 2 / 2 ( λ ξ ) n n ! \sum_{n=0}^{\infty} e^{-\xi^2/2} \frac{(\lambda \xi)^n}{n!} ∑ n = 0 ∞ e − ξ 2 /2 n ! ( λ ξ ) n can therefore be integrated term-by-term over R \mathbf{R} R (II, p. 72, cor. 1), which proves that the series ∑ n = 0 ∞ α n λ n \sum_{n=0}^{\infty} \alpha_n \lambda^n ∑ n = 0 ∞ α n λ n converges absolutely for every λ ∈ C \lambda \in \mathbf{C} λ ∈ C , and has a sum equal to u ( λ ) = e λ 2 / 2 = ∑ n = 0 ∞ λ 2 n 2 n n ! u(\lambda) = e^{\lambda^2/2} = \sum_{n=0}^{\infty} \frac{\lambda^{2n}}{2^n n!} u ( λ ) = e λ 2 /2 = ∑ n = 0 ∞ 2 n n ! λ 2 n ; thus the operator U U U is regular. Applying the lemma mentioned above shows that α 2 n = 1 / 2 n n ! , α 2 n + 1 = 0 \alpha_{2n} = 1/2^n n!,\ \alpha_{2n+1} = 0 α 2 n = 1/ 2 n n ! , α 2 n + 1 = 0 for every n ≥ 0 n \geq 0 n ≥ 0 ; the operator U U U is thus of order 0. One has
1 u ( λ ) = e − λ 2 / 2 = ∑ n = 0 ∞ ( − 1 ) n λ 2 n 2 n n ! ,
\frac{1}{u(\lambda)} = e^{-\lambda^2/2} = \sum_{n=0}^{\infty} \frac{(-1)^n \lambda^{2n}}{2^n n!},
u ( λ ) 1 = e − λ 2 /2 = n = 0 ∑ ∞ 2 n n ! ( − 1 ) n λ 2 n ,
the series being absolutely convergent for every λ ∈ C \lambda \in \mathbf{C} λ ∈ C ; another application of the lemma shows that β 2 n = ( − 1 ) n ( 2 n ) ! 2 n n ! , β 2 n + 1 = 0 \beta_{2n} = \frac{(-1)^n (2n)!}{2^n n!},\ \beta_{2n+1} = 0 β 2 n = 2 n n ! ( − 1 ) n ( 2 n )! , β 2 n + 1 = 0 ; further, the series ∑ n = 0 ∞ λ n n ! u n ( x ) \sum_{n=0}^{\infty} \frac{\lambda^n}{n!} u_n(x) ∑ n = 0 ∞ n ! λ n u n ( x ) is absolutely convergent for every λ ∈ C \lambda \in \mathbf{C} λ ∈ C and every x ∈ R x \in \mathbf{R} x ∈ R , and one has
∑ n = 0 ∞ λ n n ! u n ( x ) = exp ( − λ 2 2 + λ x ) = exp ( x 2 2 ) exp ( − 1 2 ( λ − x ) 2 ) .
\sum_{n=0}^{\infty} \frac{\lambda^n}{n!} u_n(x) = \exp \left( -\frac{\lambda^2}{2} + \lambda x \right) = \exp \left( \frac{x^2}{2} \right) \exp \left( -\frac{1}{2} (\lambda - x)^2 \right).
n = 0 ∑ ∞ n ! λ n u n ( x ) = exp ( − 2 λ 2 + λ x ) = exp ( 2 x 2 ) exp ( − 2 1 ( λ − x ) 2 ) .
On applying the Taylor formula to the function exp ( − x 2 / 2 ) \exp(-x^2/2) exp ( − x 2 /2 ) one then obtains the following expression for the polynomials u n ( x ) u_n(x) u n ( x ) :
u n ( x ) = ( − 1 ) n e λ 2 / 2 d n d x n ( e − x 2 / 2 ) .
u_n(x) = (-1)^n e^{\lambda^2/2} \frac{d^n}{dx^n} (e^{-x^2/2}).
u n ( x ) = ( − 1 ) n e λ 2 /2 d x n d n ( e − x 2 /2 ) .
This polynomial is called the Hermite polynomial of degree n n n , and is often denoted by H n ( x ) \mathrm{H}_n(x) H n ( x ) . The formulae (7), (8) and (9) of VI, p. 273, here give
d H n d x = n H n − 1 ( x )
\frac{d \mathrm{H}_n}{dx} = n \mathrm{H}_{n-1}(x)
d x d H n = n H n − 1 ( x )
H n ( x + y ) = ∑ k = 0 n ( n k ) H n − k ( x ) y k
\mathrm{H}_n(x+y) = \sum_{k=0}^n \binom{n}{k} \mathrm{H}_{n-k}(x) y^k
H n ( x + y ) = k = 0 ∑ n ( k n ) H n − k ( x ) y k
1 2 π ∫ − ∞ + ∞ e − ξ 2 / 2 H n ( x + ξ ) d ξ = x n
\frac{1}{\sqrt{2\pi}} \int_{-\infty}^{+\infty} e^{-\xi^2/2} \mathrm{H}_n(x+\xi) d\xi = x^n
2 π 1 ∫ − ∞ + ∞ e − ξ 2 /2 H n ( x + ξ ) d ξ = x n
and the formula (27) of VI, p. 278, becomes, for h = 0 h = 0 h = 0
2 π f ( x ) = ∑ m = 0 n ( ∫ − ∞ + ∞ e ξ 2 / 2 f ( m ) ( ξ ) d ξ ) H m ( x ) m !
\sqrt{2\pi} f(x) = \sum_{m=0}^n \left( \int_{-\infty}^{+\infty} e^{\xi^2/2} f^{(m)}(\xi) d\xi \right) \frac{\mathrm{H}_m(x)}{m!}
2 π f ( x ) = m = 0 ∑ n ( ∫ − ∞ + ∞ e ξ 2 /2 f ( m ) ( ξ ) d ξ ) m ! H m ( x )
− ∫ − ∞ + ∞ d ξ ∫ 0 ξ H n ( x + η ) n ! e − ( ξ + η ) 2 / 2 f ( n + 1 ) ( x + ξ − η ) d η .
- \int_{-\infty}^{+\infty} d\xi \int_0^\xi \frac{\mathrm{H}_n(x+\eta)}{n!} e^{-(\xi+\eta)^2/2} f^{(n+1)}(x+\xi-\eta) d\eta.
− ∫ − ∞ + ∞ d ξ ∫ 0 ξ n ! H n ( x + η ) e − ( ξ + η ) 2 /2 f ( n + 1 ) ( x + ξ − η ) d η .
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.
Nothing.