Fourier变换 Schwarz空间

f:ARCf:A\subset \mathbb R\rightarrow \mathbb Cff 是实数到复数的一个映射,则 f(x)=u(x)+icotv(x)f(x) = u(x) + i\cot v(x)u,v:ARu, v:A\rightarrow \mathbb R,定义

Af:=Au+iAv\int_A f := \int_Au+i\int_Av

u,vCk(A)u, v\in C^k(A),则称 fCk(A)f\in C^k(A)

定义1(1-范数)

f:RCf:\mathbb R\rightarrow \mathbb Cff 至多有有限个奇点,若

Rf<+\int_{\mathbb R}|f| < +\infty

则称 fL1(R)f\in L^1(\mathbb R),记 f1=Rf||f||_1=\int_{\mathbb R}|f|ff1-范数

定义2(2-范数)

f:RCf:\mathbb R\rightarrow \mathbb C,若

Rf2<+\int_{\mathbb R}|f|^2 < +\infty

则称 fL2(R)f\in L^2(\mathbb R),记 f2=(Rf2)12||f||_2=\left(\int_{\mathbb R}|f|^2\right )^{\frac{1}{2}}ff2-范数

定义3(Fourier变换)

fL1(R)f\in L^1(\mathbb R)

f^(x)=Rf(y)e2πixydy(xR)\hat{f}(x) = \int_{\mathbb R}f(y)e^{-2\pi ixy}\,dy\quad (x\in \mathbb R)

f^\hat{f}ffFourierFourier 变换,也记为 f^=F(f)\hat{f} = \mathcal{F}(f)


物理应用:

f:RCf:\mathbb R\rightarrow \mathbb C 为波函数,假设 f=ωAωe2πiωxf = \sum_{\omega}A_{\omega}e^{2\pi i\omega x}AωRA_{\omega}\in\mathbb R 为对应 ω\omega 的复振幅,则

A=f^(ω)A = \hat{f}(\omega)

这样就可以将复杂的波函数 ff 拆分为许多简单的波函数之和,相当于对其进行了分解,也就是求出了其傅里叶级数(广义下的)。

定义4(Fourier逆变换)

fL1(R)f\in L^1(\mathbb R)

fˇ(x)=Rf(y)e2πixydy(xR)\check{f}(x) = \int_{\mathbb R}f(y)e^{2\pi i xy}\,dy\quad (x\in\mathbb R)

fˇ\check{f}ffFourierFourier 逆变换,也记为 fˇ=F1(f)\check{f} = \mathcal{F}^{-1}(f)

Schwarz空间 急降函数

S(R):={f:RC, fC(R): k,lN, xkf(l)(x)有界}S(\mathbb R) := \{f:\mathbb R \rightarrow \mathbb C,\ f\in C^{\infty}(\mathbb R):\forall\ k, l\in \mathbb N,\ x^kf^{(l)}(x)\text{有界}\}

则称 S(R)S(\mathbb R)SchwarzSchwarz 空间,若 fS(R)f\in S(\mathbb R),则称 ff急降函数

例子

f(x)=eax2f(x) = e^{-ax^2}a>0a > 0,则 fS(R)f\in S(\mathbb R)


证明: 由于

f= 2axeax2f= (2ax)2eax22aeax2f= (2ax)3eax2+f(l)= Pl(x)eax2Pl(x)l次多项式\begin{aligned} f' =&\ -2axe^{-ax^2}\\ f'' =&\ (-2ax)^2e^{-ax^2}-2ae^{-ax^2}\\ f'''=&\ (-2ax)^3e^{-ax^2}+\cdots\\ &\vdots\\ f^{(l)} =&\ P_l(x)e^{-ax^2}&P_l(x)\text{为}l\text{次多项式} \end{aligned}

xkf(l)(x)=Pl+k(x)eax2C(C为常数)|x^kf^{(l)}(x)| = |P_{l+k}(x)e^{-ax^2}| \leqslant C\quad (C\text{为常数})

(幂指数的阶大于多项式的阶)

命题1(急降函数的性质)

(1). S(R)S(\mathbb R) 为线性空间。

(2). 若 fS(R)f\in S(\mathbb R),则 f(k)S(R)f^{(k)}\in S(\mathbb R)

(3). 若 fS(R)f\in S(\mathbb R)P:RCP:\mathbb R\rightarrow \mathbb C 为多项式,则 P(x)f(x)S(R)P(x)f(x)\in S(\mathbb R)

(4). 若 fS(R)f\in S(\mathbb R),则 fL1(R)f\in L^1(\mathbb R)


证明: (1).(2). 由导数的线性性和连续求导可得。

(3). 由(1)知,只需证明 xkf(x)S(R)x^kf(x)\in S(\mathbb R),由归纳法知,只需证 xf(x)S(R)xf(x)\in S(\mathbb R)

LeibnizLeibniz 公式(乘积函数求导法则)知

xk(xf(x))(l)= xk{xf(l)(x)+(l1)f(l1)(x)}= xk+1f(l)(x)+lxkf(l1)(x)C\begin{aligned} x^k(xf(x))^{(l)} =&\ x^k\{xf^{(l)}(x)+\binom{l}{1}f^{(l-1)}(x)\}\\ =&\ x^{k+1}f^{(l)}(x)+lx^kf^{(l-1)}(x)\leqslant C \end{aligned}

因此 xk(xf(x))(l)x^k(xf(x))^{(l)} 有界,则 xf(x)S(R)xf(x)\in S(\mathbb R)

(4). 由于 f(x)C1, x2f(x)C2(1+x2)f(x)C1+C2f(x)C31+x2|f(x)|\leqslant C_1,\ |x^2f(x)|\leqslant C_2\Rightarrow |(1+x^2)f(x)|\leqslant C_1+C_2\Rightarrow |f(x)|\leqslant\dfrac{C_3}{1+x^2}

RfRC31+x2dx=C3arctanx+=πC3\int_{\mathbb R}|f|\leqslant \int_{\mathbb R}\frac{C_3}{1+x^2}\,dx = C_3\arctan x\biggl|_{-\infty}^{+\infty}=\pi C_3

命题2(急降函数的Fourier变换的性质)

fS(R)f\in S(\mathbb R),则

(1). f(k)^=(2πix)kf^(x)\widehat{f^{(k)}}=(2\pi i x)^k\hat{f}(x)

(2). f^(k)(x)=(2πix)kf(x)^(x)\hat{f}^{(k)}(x) = \widehat{(-2\pi ix)^kf(x)}(x)

P.S: 由此看出,FourierFourier 变换可以将函数求导运算转化为函数与多项式的乘法运算!


证明

(1). 先证明 k=1k=1 的情形,由分部积分得

f^(x)= limN+NNf(y)e2πixydy= limN+{f(y)e2πixyy=Ny=N+2πixNNf(y)e2πixydy}\begin{aligned} \widehat{f'}(x)=&\ \lim_{N\rightarrow +\infty}\int_{-N}^{N}f'(y)e^{-2\pi ixy}\,dy\\ =&\ \lim_{N\rightarrow +\infty}\left\{f(y)e^{-2\pi ixy}\biggl|_{y=-N}^{y=N}+2\pi ix\int_{-N}^{N}f(y)e^{-2\pi ixy}\,dy\right\} \end{aligned}

由于 f(y)C1+y2|f(y)|\leqslant\dfrac{C}{1+y^2},由比较判别法知, limN+f(y)e2πixyy=Ny=N=0\lim\limits_{N\rightarrow +\infty}f(y)e^{-2\pi ixy}\biggl|_{y=-N}^{y=N} = 0

f^(x)=limN+2πixNNf(y)e2πixydy=2πixf^(x)\widehat{f'}(x) = \lim_{N\rightarrow +\infty}2\pi ix\int_{-N}^{N}f(y)e^{-2\pi ixy}\,dy = 2\pi ix\hat{f}(x)

由归纳法得,f(k)^=(2πix)kf^(x)\widehat{f^{(k)}}=(2\pi i x)^k\hat{f}(x)

(2). 由于

f^(x)=Rf(y)e2πixydy 且 f(y)e2πixyf(y)\hat{f}(x)=\int_{\mathbb R}f(y)e^{-2\pi ixy}\,dy\ \text{且}\ |f(y)e^{-2\pi ixy}|\leqslant|f(y)|

又有 fL1(R)f\in L^1(\mathbb R)Rf\int_{\mathbb R}|f| 收敛,则 Rf(y)e2πixydy\int_{\mathbb R}f(y)e^{-2\pi ixy}\,dy 一致收敛,求导和积分可交换顺序,则

f^(x)=R(2πiy)f(y)e2πixydy= 2πixf(x)^(x)\begin{aligned} \hat{f}'(x) =&\int_{\mathbb R}(-2\pi iy)f(y)e^{-2\pi ixy}\,dy\\ =&\ \widehat{-2\pi ixf(x)}(x) \end{aligned}

由归纳法得,f^(k)(x)=(2πix)kf(x)^(x)\hat{f}^{(k)}(x) = \widehat{(-2\pi ix)^kf(x)}(x)

定理3(Fourier变换是Schwarz空间到自身的映射)

fS(R)f\in S(\mathbb R),则 f^S(R)\hat{f}\in S(\mathbb R)


思路: 先进行对任意一个函数进行估计(命题2),转化为另一个急降函数的 FourierFourier 变换,利用急降函数的 FourierFourier 变换有界(命题1),即可。

证明: 设 k,lNk,l\in \mathbb N,记 F(f)=f^(x)\mathcal{F}(f) = \hat{f}(x),由命题2知,

xkf^(l)(x)= xkF((2πix)lf(x))= (2πi)k(2πix)kF((2πix)lf(x))= (2πi)kF(dkdxk((2πix)lf(x)))\begin{aligned} x^k\hat{f}^{(l)}(x) =&\ x^k\mathcal{F}((-2\pi ix)^lf(x))\\ =&\ (2\pi i)^{-k}(2\pi ix)^k\mathcal{F}((-2\pi ix)^lf(x))\\ =&\ (2\pi i)^{-k}\mathcal{F}\left(\frac{d^k}{dx^k}\left((-2\pi ix)^lf(x)\right)\right) \end{aligned}

F(x)=dkdxk((2πix)lf(x))F(x) = \dfrac{d^k}{dx^k}\left((-2\pi ix)^lf(x)\right),由命题1知,F(x)F(x) 为急降函数属于 L1L^1,则

F(F) RF(y)e2πixydy RF(y)dy RF(y)dy<+\begin{aligned} |\mathcal{F}(F)|\leqslant&\ \left|\int_{\mathbb R}F(y)e^{-2\pi ixy}\,dy\right|\\ \leqslant&\ \left|\int_{\mathbb{R}}F(y)\,dy\right|\\ \leqslant&\ \int_{\mathbb R}|F(y)|\,dy < +\infty \end{aligned}

xkf^(l)(x)=(2πi)kF(F)<+f^S(R)\left|x^k\hat{f}^{(l)}(x)\right|=\left|(2\pi i)^{-k}\mathcal{F}(F)\right| < +\infty\\ \Rightarrow\hat{f}\in S(\mathbb R)

Fourier 反演公式

广义 Fubini 定理

f:I×JCf:I\times J\rightarrow \mathbb CI,JRI, J\subset \mathbb R 为区间(有限或者无限),且 f0f\geqslant 0I×Jf<+\int_{I\times J}|f| < +\infty,则

I×Jf(x,y)dxdy=I{Jf(x,y)dy}dx=J{If(x,y)dx}dy\begin{aligned} \int_{I\times J}f(x, y)\,dx\,dy = \int_I\left\{\int_J f(x, y)\,dy\right\}\,dx = \int_J\left\{\int_I f(x, y)\,dx\right\}\,dy \end{aligned}

(《实变函数》中会进行证明)

定理1(Fourier 乘积公式)

f,gS(R)f, g\in S(\mathbb R),则

Rf^g=Rfg^\int_{\mathbb R}\hat{f}g = \int_{\mathbb R}f\hat{g}


证明: 记 F(x,y)=f(x)g(y)e2πixyF(x, y) = f(x)g(y)e^{-2\pi ixy},由于

R2F=R2f(x)g(y)dxdyR2f(x)g(y)dxdy=FubiniRf(x)dxRg(y)dy=f1g1<+\begin{aligned} \int_{\mathbb{R}^2}|F| =&\int_{\mathbb{R}^2}|f(x)g(y)|\,dx\,dy\\ \leqslant&\int_{\mathbb{R}^2}|f(x)||g(y)|\,dx\,dy\\ \xlongequal{\text{Fubini}}&\int_{\mathbb R}|f(x)|\,dx\int_{\mathbb R}|g(y)|\,dy\\ =&||f||_1\cdot||g||_1 < +\infty \end{aligned}

则可对 FF 使用 FubiniFubini 定理:

R2F=R{Rf(x)g(y)e2πixydx}dy=R{Rf(x)g(y)e2πixydy}dx=R{Rf(x)e2πixydx}g(y)dy=Rf(x){Rg(y)e2πixydy}dx=Rf^(y)g(y)dy=Rf(x)g^(x)dx\begin{aligned} \int_{\mathbb{R}^2}F =& \int_{\mathbb{R}}\left\{\int_{\mathbb R}f(x)g(y)e^{-2\pi ixy}\,dx\right\}\,dy = \int_{\mathbb R}\left\{\int_{\mathbb R}f(x)g(y)e^{-2\pi ixy}\,dy\right\}\,dx\\ =&\int_{\mathbb R}\left\{\int_{\mathbb{R}}f(x)e^{-2\pi ixy}\,dx\right\}g(y)\,dy = \int_{\mathbb R}f(x)\left\{\int_{\mathbb R}g(y)e^{-2\pi ixy}\,dy\right\}\,dx\\ =&\int_{\mathbb R}\hat{f}(y)g(y)\,dy=\int_{\mathbb R}f(x)\hat{g}(x)\,dx \end{aligned}

定理2(Fourier 变换反演公式)

fS(R)f\in S(\mathbb R),则 f^S(R)\hat{f}\in S(\mathbb R),定义映射

F:S(R)S(R)f(x)f^(x)\begin{aligned} \mathcal{F}:S(\mathbb R)&\rightarrow S(\mathbb R)\\ f(x)&\mapsto \hat{f}(x) \end{aligned}

F\mathcal{F}S(R)S(\mathbb R) 上的 FourierFourier 变换,设 fˇ(x)=Rf(y)e2πixydy\check{f}(x) = \int_{\mathbb R}f(y)e^{2\pi ixy}\,dy,则 fˇ(x)=f^(x)\check{f}(x) = \hat{f}(x),由于 f^(x)S(R)\hat{f}(x)\in S(\mathbb R),则 fˇ(x)S(R)\check{f}(x)\in S(\mathbb R),定义映射

F:S(R)S(R)f(x)fˇ(x)\begin{aligned} \mathcal{F}^*:S(\mathbb R)&\rightarrow S(\mathbb R)\\ &f(x)&\mapsto \check{f}(x) \end{aligned}

F\mathcal{F}^*F\mathcal{F} 的逆变换,即 F=F1\mathcal{F}^* = \mathcal{F}^{-1}


为证明 FourierFourier 变换为 S(R)S(\mathbb R) 上的变换和 FourierFourier 反演公式,引入 GaussGauss 函数。

Gauss 函数

G:RRG:\mathbb R\rightarrow \mathbb R,定义为

G(x)=eπx2G(x) = e^{-\pi} x^2

GGGaussGauss 函数(钟型函数)。

GaussGauss 函数是重要的核函数,在 FourierFourier 变换下有很多很好的性质,起到重要作用。

性质

(1). RG=1\int_{\mathbb R}G = 1

(2). GS(R)G\in S(\mathbb R)

(3). G^=G\hat{G} = G


证明: (1). 先证明一个重要的等式

I=Rex2dx=πI = \int_{\mathbb R}e^{-x^2}\,dx = \sqrt{\pi}

直接计算得

I2=+ex2dx+ey2dy=FubiniR2e(x2+y2)dxdy=极坐标变换0+{Brer2ds}dr=0+er2{Br1ds}dr= π0+2rer2dr= π0+er2dr2= πI= π\begin{aligned} I^2=&\int_{-\infty}^{+\infty}e^{-x^2}\,dx\cdot\int_{-\infty}^{+\infty}e^{-y^2}\,dy\\ \xlongequal{\text{Fubini}}&\int_{\mathbb{R}^2}e^{-(x^2+y^2)}\,dx\,dy\\ \xlongequal{\text{极坐标变换}}&\int_{0}^{+\infty}\left\{\int_{\partial B_r}e^{-r^2}\,ds\right\}\,dr\\ =&\int_0^{+\infty}e^{-r^2}\left\{\int_{\partial B_r}1\,ds\right\}\,dr\\ =&\ \pi\int_0^{+\infty}2re^{-r^2}\,dr\\ =&\ \pi\int_{0}^{+\infty}e^{-r^2}\,dr^2\\ =&\ \pi\\ \Rightarrow I=&\ \sqrt{\pi} \end{aligned}

RG=Reπx2dx=x=y/x1πRey2dy=1πI=1\begin{aligned} \int_{\mathbb R}G = \int_{\mathbb R}e^{-\pi x^2}\,dx\xlongequal{x = y/\sqrt{x}}\frac{1}{\sqrt{\pi}}\int_{\mathbb R}e^{-y^2}\,dy = \frac{1}{\sqrt{\pi}}\cdot I = 1 \end{aligned}

(2). 通过 SchwarzSchwarz 空间 - 例子a=πa = \pi 即可得证。

(3). (转化为求解微分方程)

G^=Reπy2e2πixydy由比较判别法知,G连续且一致收敛,则求导和积分可交换顺序ddxG^= R(2πiy)eπy2e2πixydy= iRddy(eπy2)e2πixydy= i(eπy2e2πixy++(2πix)Reπy2e2πixydy)= 2πxG^\begin{aligned} \hat{G} =& \int_{\mathbb R}e^{-\pi y^2}e^{-2\pi ixy}\,dy\\ \text{由比较判别法知,}G&\text{连续且一致收敛,则求导和积分可交换顺序}\\ \frac{d}{dx}\hat{G}=&\ \int_{\mathbb R}(-2\pi iy)e^{-\pi y^2}e^{-2\pi ixy}\,dy\\ =&\ i\int_{\mathbb R}\frac{d}{dy}(e^{-\pi y^2})e^{-2\pi ixy}\,dy\\ =&\ i\left(e^{-\pi y^2}e^{-2\pi ixy}\biggl|_{-\infty}^{+\infty}+(2\pi ix)\int_{\mathbb R}e^{-\pi y^2}e^{-2\pi ixy}\,dy\right)\\ =&\ -2\pi x\hat{G} \end{aligned}

G^(0)=Reπy2dy=1\hat{G}(0) = \int_{\mathbb R}e^{-\pi y^2}\,dy = 1,则可转化为求解线性微分方程初值问题,解得

G^=G\hat{G} = G


为证明 定理2 还需对 GaussGauss 函数进行伸缩变换,设 0<δ<10 < \delta < 1,定义

Gδ(x)= eπδx2xRKδ(x)= δ12eπx2/δxR\begin{aligned} G_{\delta}(x) =&\ e^{-\pi\delta x^2} & x\in \mathbb R\\ K_{\delta}(x) =&\ \delta^{-\frac{1}{2}}e^{-\pi x^2/\delta} & x\in \mathbb R \end{aligned}

GδG_{\delta} 是对 GGxx 轴拉伸 1δ\frac{1}{\sqrt{\delta}} 倍,
KδK_{\delta} 是对 GGxx 轴压缩 δ\sqrt{\delta} 倍,yy 轴拉伸 1δ\frac{1}{\sqrt{\delta}} 倍,则

RKδ(x)=Rδ12eπx2/δdx=x=δyReπy2dy=1\begin{aligned} \int_{\mathbb R}K_{\delta}(x) =\int_{\mathbb R}\delta^{-\frac{1}{2}}e^{-\pi x^2/\delta}\,dx\xlongequal{x = \sqrt{\delta}y}\int_{\mathbb R}e^{-\pi y^2}\,dy = 1 \end{aligned}

Gδ

Kδ

引理1(GδG_{\delta}KδK_{\delta} 的关系)

Gδ^=Kδ\widehat{G_{\delta}} = K_{\delta}


证明: 直接计算可得

Gδ^(x)=Reπδy2eπixydy=z=δy δ12Reπz2eπixδ12zdz= δ12G^(x/δ)= δ12G(x/δ)= δ12eπx2/δ= Kδ(x)\begin{aligned} \widehat{G_{\delta}}(x) =& \int_{\mathbb R} e^{-\pi \delta y^2}e^{-\pi ixy}\,dy\\ \xlongequal{z=\sqrt{\delta}y}&\ \delta^{-\frac{1}{2}}\int_{\mathbb R}e^{-\pi z^2}e^{-\pi ix\delta^{-\frac{1}{2}}z}\,dz\\ =&\ \delta^{-\frac{1}{2}}\hat{G}(x/\sqrt{\delta})\\ =&\ \delta^{-\frac{1}{2}}G(x/\sqrt{\delta})\\ =&\ \delta^{-\frac{1}{2}}e^{-\pi x^2/\delta}\\ =&\ K_{\delta}(x) \end{aligned}

引理2(GδG_{\delta}KδK_{\delta}ff 的乘积的积分)

fS(R)f\in S(\mathbb R),当 δ0+\delta\rightarrow 0^+ 时,有

(1). RGδfRf\int_{\mathbb R}G_{\delta}f\rightarrow \int_{\mathbb R}f

(2). RKδff(0)\int_{\mathbb R}K_{\delta}f\rightarrow f(0)


证明: (1). 设 I(δ)Reπδx2f(x)dxI(\delta)\int_{\mathbb R}e^{-\pi\delta x^2}f(x)\,dxδ0\delta\geqslant 0,由比较判别法知,I(δ)I(\delta) 关于 δ\delta[0,+)[0,+\infty) 上一致收敛,则 IC([0,+))I\in C\left([0,+\infty)\right)

limδ0+RGδf=limδ0+I(δ)=I(0)=Rf\lim_{\delta\rightarrow 0^+}\int_{\mathbb R}G_{\delta}f = \lim_{\delta\rightarrow 0^+}I(\delta) = I(0) = \int_{\mathbb R}f

(2).

RKδf=Rδ12eπx2/δf(x)dx=x=δyReπy2f(δy)dy\int_{\mathbb R}K_{\delta}f = \int_{\mathbb R}\delta^{-\frac{1}{2}}e^{-\pi x^2/\delta}f(x)\,dx\xlongequal{x=\sqrt{\delta}y}\int_{\mathbb R}e^{-\pi y^2}f(\sqrt{\delta}y)\,dy

I(δ)=Reπy2f(δy)I(\delta) = \int_{\mathbb R}e^{-\pi y^2}f(\sqrt{\delta}y),由于 eπy2f(δy)Mepiy2\left|e^{-\pi y^2}f(\sqrt{\delta}y)\right|\leqslant M\cdot e^{-pi y^2},由比较判别法知,I(δ)I(\delta)δ\delta 上一致收敛,则 IC([0,+))I\in C([0,+\infty))

limδ0+RKδf=limδ0+I(δ)=I(0)=f(0)Reπy2dy=f(0)RG=f(0)\begin{aligned} \lim_{\delta\rightarrow 0^+}\int_{\mathbb R}K_{\delta}f = \lim_{\delta\rightarrow 0^+}I(\delta)=I(0)=f(0)\int_{\mathbb R}e^{-\pi y^2}\,dy = f(0) \int_{\mathbb R} G = f(0) \end{aligned}

引理3(Fourier 反演在 x=0x=0 处的情况)

fS(R)f\in S(\mathbb R),则

Rf^=f(0)\int_{\mathbb R}\hat{f} = f(0)


证明: 由 定理1 - FourierFourier 乘积公式 得,

RGδf^=RGδ^f=RKδf\int_{\mathbb R}G_{\delta}\hat{f} = \int_{\mathbb R}\widehat{G_{\delta}}f = \int_{\mathbb R}K_{\delta}f

δ)\delta\rightarrow )

Rf^=f(0)\int_{\mathbb R}\hat{f} = f(0)

引理4(Fourier 变换的平移不变性,单射)

fS(R)f\in S(\mathbb R),则

F(f^)(a)=f(a)\mathcal{F}^*(\hat{f})(a) = f(a)

F(F(f))=f\mathcal{F}^*(\mathcal{F}(f)) = f


证明: 记 F(x)=f(x+a)F(x) = f(x+a)xRx\in \mathbb R,则 FS(R)F\in S(\mathbb R)

f(a)=F(0)=RF^=R{Rf(y+a)e2πixydy}dx=z=y+aR{Rf(z)e2πix(za)dz}dx=R{Rf(z)e2πixzdz}e2πixadx=Rf^(x)e2πiaxdx= F(f^(a))\begin{aligned} f(a) = F(0)=\int_{\mathbb R}\hat{F}=&\int_{\mathbb R}\left\{\int_{\mathbb R}f(y+a)e^{-2\pi ixy}\,dy\right\}\,dx\\ \xlongequal{z = y+a}&\int_{\mathbb R}\left\{\int_{\mathbb R}f(z)e^{-2\pi ix(z-a)}\,dz\right\}\,dx\\ =&\int_{\mathbb R}\left\{\int_{\mathbb R}f(z)e^{-2\pi ixz}\,dz\right\}e^{2\pi ixa}\,dx\\ =&\int_{\mathbb R}\hat{f}(x)e^{2\pi iax}\,dx\\ =&\ \mathcal{F}^*(\hat{f}(a)) \end{aligned}

引理5(满射)

fS(R)f\in S(\mathbb R),则

F(F(f))=f\mathcal{F}(\mathcal{F}^*(f)) = f


证明

F(F(f))= F(Rf(y)e2πixydy)=R{Rf(y)e2πizydy}e2πixzdz=zzR{Rf(y)e2πizydy}e2πixydz= F(F(f))(x)=f(x)\begin{aligned} \mathcal{F}(\mathcal{F}^*(f)) =&\ \mathcal{F}\left(\int_{\mathbb R}f(y)e^{2\pi ixy}\,dy\right)\\ =&\int_{\mathbb R}\left\{\int_{\mathbb R}f(y)e^{2\pi izy}\,dy\right\}e^{-2\pi ixz}\, dz\\ \xlongequal{z\rightarrow -z}&\int_{\mathbb R}\left\{\int_{\mathbb R}f(y)e^{-2\pi izy}\,dy\right\}e^{2\pi ixy}\,dz\\ =&\ \mathcal{F}^*(\mathcal{F}(f))(x) = f(x) \end{aligned}

证明定理2(Fourier 反演公式)

根据 引理4引理5 知,fS(R)f\in S(\mathbb R)

F(F(f))=F(F(f))=f\mathcal{F}^*(\mathcal{F}(f)) = \mathcal{F}(\mathcal{F}^*(f)) = f

证明 F\mathcal{F} 是双射:

单射:令 F(f)=g\mathcal{F}(f) = \mathcal{g},则

F(F(f))= F(F(g))f= g\begin{aligned} \mathcal{F}^*(\mathcal{F}(f)) =&\ \mathcal{F}^*(\mathcal{F}(g))\\ \Rightarrow \quad\quad\quad\quad f =&\ g \end{aligned}

满射:设 fS(R)f\in S(\mathbb R),存在 F(f)S(R)\mathcal{F}^*(f)\in S(\mathbb R),使得

F(F(f))=f\mathcal{F}(\mathcal{F}^*(f)) = f

综上,F=F1\mathcal{F}^* = \mathcal{F}^{-1}


Fourier变换 Schwarz空间
https://wty-yy.github.io/posts/1783/
作者
wty
发布于
2021年12月29日
许可协议