Return to computing page for the first course APMA0330
Return to computing page for the second course APMA0340
Return to computing page for the fourth course APMA0360
Return to Mathematica tutorial for the first course APMA0330
Return to Mathematica tutorial for the second course APMA0340
Return to Mathematica tutorial for the fourth course APMA0360
Return to the main page for the course APMA0330
Return to the main page for the course APMA0340
Return to the main page for the course APMA0360
Introduction to Linear Algebra with Mathematica

Preface


We denote by 𝔏¹(ℝ) the vector space of Lebesgue integrable functions on the real line. Its elements are equivalent classes of functions that are equal almost everywhere. The space 𝔏¹(ℝ) or 𝔏¹ is the vector space of Lebesgue integrable functions, defined "up to equality almost everywhere," with the norm
\[ \| f \|_1 = \int_{-\infty}^{\infty} |f(x)|\,{\text d} x < \infty . \]

In this section, we define the Fourier transform of integrable functions from 𝔏¹(ℝ) and derive its main properties.

Fourier transform in 𝔏¹


Let f(x) be a function, real- or complex-valued, depending on one real variable x. The Fourier transform (or spectrum) of f, if it exists, is the complex-valued function defined for the real variable ξ by \begin{equation} \label{EqFourier1.1} \hat{f} (\xi ) = \mbox{V.P.} \int_{\mathbb{R}} f(x)\, e^{-\mathbf{j}\xi x}{\text d}x , \end{equation} where "V.P." stands for the Cauchy principal value and j or ⅉ denotes the imaginary unit on complex plane ℂ so ⅉ² = −1. An integral operator of Eq.\eqref{EqFourier1.1} is called after Fourier and symbolically denoted by \[ \hat{f} = ℱ\left[ f \right] \quad \mbox{or} \quad \hat{f}(\xi ) = ℱ\left[ f(x)) \right]_{x\to\xi} . \]
It is also convenient to denote by fF the Fourier transform because there is no universal notation for this integral transformation.
The conjugate Fourier transform of a function f is given (when it exists) by \[ \hat{f}^{\ast} (\xi ) = ℱ^{\ast}\left[ f \right] (\xi ) = \mbox{V.P.} \int_{\mathbb{R}} f(x)\, e^{\mathbf{j}\xi x}{\text d}x . \]
The Fourier transform of a Lebesgue integrable function on ℝ definitely exists because \( \displaystyle \quad \left\vert e^{{\bf j}\xi x} \right\vert = 1 . \ \) Due to a wide range of applications, there is a need to evaluate the Fourier integral for functions not necessarily of 𝔏. For example, if g : x → (sinx)/x, then g is not integrable, but the following limit (in Cauchy principal value sense)
\[ \lim_{N\to +\infty} \int_{-N}^N \frac{\sin x}{x}\,e^{-\mathbf{j}\xi x} {\text d}x \]
exists for all real values of ξ.    
Example 1: Consider the “rectangle” function (also know as a characteristic function) \[ \Pi (x) = \begin{cases} 1, & \quad \mbox{if } -1 \le x \le 1 , \\ 0, & \quad \mbox{otherwise}. \end{cases} \] Then its Fourier transform is a sinc function (or the cardinal sine), as Mathematica claims
Integrate[Exp[-I*x*t], {x, -1, 1}]
(2 Sin[t])/t
\[ \int_{-1}^1 e^{-{\bf j}\xi x} {\text d}x = 2\,\frac{\sin\xi}{\xi} \quad (\xi \ne 0). \] Hence, \[ \hat{\Pi} (\xi ) = 2\cdot \mbox{sinc}(\xi ) = 2 \times\begin{cases} \frac{\sin\xi}{\xi} , & \quad \mbox{if } \xi \ne 0 , \\ 1, & \quad \mbox{if } \xi = 0. \end{cases} \]    ■
End of Example 1
Theorem 1: (Main properties of Fourier transform) Let f,g ∈ 𝔏¹(ℝ). Then
  1. ℱ(f + g) = ℱ(f) + ℱ(g).
  2. For any complex number α, \[ ℱ\left( \alpha\, f \right) = \alpha ℱ\left( f \right) \]
  3. If f (also denoted by \( \displaystyle \ \overline{f} \) ) is complex conjugate of f, then \[ ℱ\left( f^{\ast}\right) = \hat{f}(-\xi )^{\ast} = \overline{\hat{f}(-\xi )} . \]
  4. Let us denote the shift of function f by fy(x) = f(x + y) ror y ∈ ℝ. Then \[ \hat{f}_y (\xi ) = \hat{f}(\xi )\,e^{-{\bf j}\xi y} . \]
  5. \( \displaystyle \quad \left\vert \hat{f} (\xi ) \right\vert \leqslant \int \left\vert f(x) \right\vert {\text d}x = \| f \|_1 . \)
  6. For positive λ, let φ(x) = λ fx), then \[ \hat{\varphi}(\xi ) = f \left( \frac{\xi}{\lambda} \right) . \]
  1. Linearity: for f,g ∈ 𝔏¹(ℝ), we have \[ \hat {f+g}(\xi ) =\int _{\mathbb{R}}(f(x)+g(x))e^{-{\bf j}\xi x}\, {\text d}x = \int_{\mathbb{R}}f(x)\.e^{-{\bf j}\xi x}\, {\text d}x + \int_{\mathbb{R}}g(x)e^{-{\bf j}\xi x}\, {\text d}x = \hat {f}(\xi ) +\hat {g}(\xi ). \]
  2. Homogeneity: for α ∈ ℂ and f ∈ 𝔏¹(ℝ), we have \[ \hat {\alpha f}(\xi )=\int _{\mathbb{R}}\alpha f(x)e^{-i\xi x}\, dx=\alpha \int _{\mathbb{R}}f(x)e^{-i\xi x}\, dx=\alpha \hat {f}(\xi ). \]
  3. Conjugation: Let \( \displaystyle \quad f^*(x)=\overline{f(x)}. \ \) Then \[ \hat {f^*}(\xi )=\int _{\mathbb{R}}\overline{f(x)}\, e^{-i\xi x}\, dx. \] Take complex conjugate of ℱ[f](-ξ): \[ \hat {f}(-\xi )=\int _{\mathbb{R}}f(x)e^{-i(-\xi )x}\, dx=\int _{\mathbb{R}}f(x)e^{i\xi x}\, dx. \] Thus, \[ \overline{\hat {f}(-\xi )}=\overline{\int _{\mathbb{R}}f(x)e^{i\xi x}\, dx}=\int _{\mathbb{R}}\overline{f(x)e^{i\xi x}}\, dx=\int _{\mathbb{R}}\overline{f(x)}\, e^{-i\xi x}\, dx=\hat {f^*}(\xi ). \] So \[ \hat {f^*}(\xi )=\overline{\hat {f}(-\xi )}. \]
  4. Translation: \( \displaystyle \quad \hat{f_y}(\xi ) =\hat{f}(\xi )\,e^{-{\bf j}\xi y} \quad \) for fy(x) = f(x+y). Then \[ \hat {f_y}(\xi )=\int _{\mathbb{R}}f(x+y)e^{-i\xi x}\, dx. \] Use the change of variable u = x+y, so x = u-y, dx = du: \[ \hat {f_y}(\xi )=\int _{\mathbb{R}}f(u)e^{-i\xi (u-y)}\, du=e^{i\xi y}\int _{\mathbb{R}}f(u)e^{-i\xi u}\, du=e^{i\xi y}\hat {f}(\xi ). \] If your convention uses \( \displaystyle \quad e^{-{\bf j}\xi x} \quad \) in the transform, the factor is \( \displaystyle \quad e^{{\bf j}\xi y}. \ \) If the statement you were given has \( \displaystyle \quad e^{-{\bf j}\xi y}, \ \) then the underlying sign convention in the exponential is the opposite. With the above definition, the correct factor is ejξy.
  5. If your notes use a slightly different Fourier transform convention (e.g., \( \displaystyle \quad e^{+{\bf j}\xi x} \quad \) or 2π factors), the proofs are identical, but the signs and scaling factors adjust accordingly.
   
Example 2: Each property becomes much clearer when you see a concrete 𝔏¹(ℝ) function and compute its transform explicitly. The following standard convention is used: \[ \hat {f}(\xi )=\int _{\mathbb{R}}f(x)e^{-{\bf j}\xi x}\, {\text d} x. \]
  1. Linearity: \( \displaystyle \ \hat {f+g} =\hat {f}+\hat {g} . \)

    Take \[ f(x) = e^{-|x|},\qquad g(x) =\mathbf{1}_{[-1,1]}(x). \] Then \begin{align*} \hat {f}(\xi ) &= \int_{\mathbb{R}} e^{-|x|}e^{-{\bf j}\xi x}\, {\text d}x = \frac{2}{1+\xi ^2}. \\ \hat {g}(\xi ) &= \int _{-1}^1 e^{-{\bf j}\xi x}\, {\text d}x = \frac{2\sin \xi }{\xi }. \end{align*}

    Integrate[Exp[-Abs[x]]*Exp[-I*x*t], {x, -Infinity, Infinity}, Assumptions -> {t \[Element] Reals}]
    2/((-I + t) (I + t))
    Integrate[Exp[-I*x*t], {x, -1, 1}, Assumptions -> {t \[Element] Reals}]
    (2 Sin[t])/t
    Now compute ℱ[ f + g ]: \[ \hat {f+g}(\xi ) =\int_{\mathbb{R}}(e^{-|x|} +\mathbf{1}_{[-1,1]}(x))\, e^{-{\bf j}\xi x}\, {\text d}x =\frac{2}{1+\xi^2} + \frac{2\sin \xi }{\xi }. \] This matches ℱ[ f ](ξ) + ℱ[ g ](ξ).
  2. Scalar multiplication: ℱ[ α f ] = α ℱ[ f ](ξ).

    Let \( \displaystyle \quad f(x) =e^{-x^2} , \ \) then the Gaussian transform is known: \[ \hat {f}(\xi ) = \sqrt{\pi }\, e^{-\xi ^2/4}. \]

    Integrate[Exp[-x^2]*Exp[-I*x*t], {x, -Infinity, Infinity}, Assumptions -> {t \[Element] Reals}]
    E^(-(t^2/4)) Sqrt[\[Pi]]
    Then \[ \hat {3{\bf j}\,f}(\xi ) =3{\bf j}\,\sqrt{\pi }\, e^{-\xi ^2/4} = 3{\bf j}\, \hat {f}(\xi ). \]
  3. Conjugation: \( \displaystyle \quad \hat {f^*}(\xi ) =\overline{\hat {f}(-\xi )} . \)

    Let \[ f(x) = e^{{\bf j}\,x}\mathbf{1}_{[0,1]}(x). \] Then \[ f^*(x)=\overline{e^{ix}}\mathbf{1}_{[0,1]}(x)=e^{-ix}\mathbf{1_{\mathnormal{[0,1]}}}(x). \] Compute: \[ \hat {f}(\xi )=\int _0^1e^{{\bf j}x}e^{-{\bf j}\xi x}\, {\text d}x =\int_0^1 e^{-{\bf j}(\xi -1)x}\, {\text d}x = \frac{1-e^{-{\bf j}(\xi -1)}}{{\bf j}(\xi -1)}. \]

    Integrate[Exp[I*x]*Exp[-I*x*t], {x, 0, 1}, Assumptions -> {t \[Element] Reals}]
    (I (-1 + E^(-I (-1 + t))))/(-1 + t)
    Similarly, \[ \hat {f^*}(\xi ) = \int _0^1 e^{-{\bf j}\,x} e^{-{\bf j}\xi x}\, {\text d}x =\frac{1-e^{-{\bf j}(\xi +1)}}{{\bf j}(\xi +1)}. \]
    Integrate[Exp[-I*x]*Exp[-I*x*t], {x, 0, 1}, Assumptions -> {t \[Element] Reals}]
    (I (-1 + E^(-I (1 + t))))/(1 + t)
    Now we check the identity: \[ \overline{\hat {f}(-\xi )}=\overline{\frac{1-e^{-i(-\xi -1)}}{i(-\xi -1)}}=\frac{1-e^{-i(\xi +1)}}{i(\xi +1)}=\hat {f^*}(\xi ). \]
  4. Translation: \( \displaystyle \quad \hat {f_y}(\xi ) = \hat {f}(\xi )e^{{\bf j}\xi y} . \)

    Let f(x) = e-|x| again, and choose y = 2. Then \[ f_2(x)=f(x+2)=e^{-|x+2|}. \] We know ℱ[ f ](ξ) = 2/(1 + ξ²).

    Integrate[Exp[-Abs[x]]*Exp[-I*x*t], {x, -Infinity, Infinity}, Assumptions -> {t \[Element] Reals}]
    2/((-I + t) (I + t))
    The theorem predicts: \[ \hat {f_2}(\xi )=\frac{2}{1+\xi ^2}\, e^{{\bf j}2\xi }. \] You can verify this directly by substituting u = x + 2 in the integral.
    Integrate[Exp[-Abs[x + 2]]*Exp[-I*x*t], {x, -Infinity, Infinity}, Assumptions -> {t \[Element] Reals}]
    (2 E^(2 I t))/((-I + t) (I + t))
  5. 𝔏¹-bound: \( \displaystyle \quad |\hat {f}(\xi )|\leq \| f\|_1 . \)

    Take \[ f(x) =\frac{1}{1+x^2}. \] Then \[ \| f\|_1 =\int _{\mathbb{R}}\frac{{\text d}x}{1+x^2} =\pi . \]

    Integrate[1/(1 + x^2), {x, -Infinity, Infinity}]
    \[Pi]
    The Fourier transform is known: \[ \hat {f}(\xi ) =\pi e^{-|\xi |}. \]
    Integrate[Exp[-I*x*t]/(1 + x^2), {x, -Infinity, Infinity}, Assumptions -> {t \[Element] Reals}]
    E^-Abs[t] \[Pi]
    Check the bound: \[ \pi = \sup \left\{ \pi\,e^{-|\xi |} \right\} = \| \hat{f} \|_{\infty} \leqslant \| f \|_1 = \pi . \]

    This shows the inequality is sharp at ξ = 0.

  6. Scaling: if φ(x) = λ f(λ x), then \( \displaystyle \quad \hat {\varphi }(\xi ) =\hat {f}(\xi /\lambda ) . \)

    Let \( \displaystyle \quad f(x) =\mathbf{1}_{[-1,1]}(x). \quad \) Then \[ \hat {f}(\xi )=\frac{2\sin \xi }{\xi }. \] Choose \[ \varphi (x)=3\, f(3x)=3\, \mathbf{1}_{[-1/3,\, 1/3]}(x). \] The theorem predicts: \[ \hat {\varphi }(\xi )=\hat {f}(\xi /3)=\frac{2\sin (\xi /3)}{\xi /3}. \] You can also compute directly: \[ \hat {\varphi }(\xi )=\int _{-1/3}^{1/3}3e^{-i\xi x}\, dx=3\cdot \frac{2\sin (\xi /3)}{\xi }=\frac{2\sin (\xi /3)}{\xi /3}, \] which matches perfectly.

   ■
End of Example 2
Theorem 2: If f ∈ 𝔏¹, then its Fourier transform fF = ℱ( f ) is bounded and uniformly continuous on \( \displaystyle \ \hat{\mathbb{R}} . \)
We have \[ \hat{f}(\xi + \eta ) - \hat{f}(\xi ) = \int f(x) \left( e^{-{\bf j} (\xi + \eta ) x} - e^{-{\bf j} \xi x} \right) {\text d} x \] Hence, \[ \left\vert \hat{f}(\xi + \eta ) - \hat{f}(\xi ) \right\vert \leqslant \int |f(x)| \left\vert e^{-{\bf j}\eta x} -1 \right\vert {\text d} x \] The integral on the right is independent of ξ, the integrand is bounded by 2 |f(x)| and tends to zero everywhere where η → 0.
The continuity of the Fourier transform is an important fact. It means that if a sequence ( fₙ )n∈ℕ of integrable functions converges to f ∈ 𝔏¹, in the sense of the 𝔏¹ norm, namely,
\[ \int \left\vert f_n (x) - f(x) \right\vert {\text d} x \,\underset{n\to\infty}{\longrightarrow} \,0 , \]
then there will be convergence in the sense of the 𝔏 norm, that is, uniform convergence of the sequence \( \displaystyle \quad ( \hat{f}_n )_{n\in \mathbb{R}} \to \hat{f} . \)    
Example 3: Not every continuous function on ℝ satisfying condition \[ \lim_{|\xi | \to \infty} Q(\xi ) = 0 \] is a Fourier transform of a function from 𝔏¹. We define function Q(ξ) by formula \[ Q(\xi ) = \begin{cases} \frac{1}{\ln \xi} , &\quad \mbox{for }\ \xi > e , \\ \frac{\xi}{e} , &\quad \mbox{for }\ 0 \leqslant \xi \leqslant e , \\ -Q(-\xi ) , &\quad \mbox{for }\ \xi < 0 . \end{cases} \] It can be checked that function Q(ξ) is continuouis on whole line ℝ by verifying that limits from left and right are the same at points ξ = 0 and ξ = e because at other points this function is continuous. This function also vanishes at infinity because lnξ → ∞ as ξ → ∞.

Now we prove that there is no function q(x) ∈ 𝔏¹ such that ℱ[q] = Q.

We consider the following limit: \[ \lim_{N\to\infty} \int_e^N \frac{Q(\xi )}{\xi}\,{\text d}\xi = \lim_{N\to\infty} \int_e^N \frac{{\text d}\xi}{\xi\,\ln\xi} = \lim_{N\to\infty} \ln \ln N = \infty . \] This result shows that integral \[ \int_e^{\infty} \frac{Q(\xi )}{\xi}\,{\text d}\xi \qquad \mbox{diverges} . \]

Suppose there exists function q(x) ∈ 𝔏¹ such that ℱ[q] = Q, \[ Q(\xi ) = ℱ[q] = \int_{-\infty}^{\infty} q(x)\, e^{-{\bf j}\xi x} {\text d} x . \] Since function Q(ξ) is odd, we have \[ Q(\xi ) = - \int_{-\infty}^{\infty} q(x)\, e^{{\bf j}\xi x} {\text d} x . \] Adding these two equations, we get \[ Q(\xi ) = \int_{-\infty}^{\infty} q(x) \left[ e^{-{\bf j}\xi x} - e^{{\bf j}\xi x} \right] {\text d} x = -2{\bf j} \int_{-\infty}^{\infty} q(x) \,\sin (\xi x)\,{\text d} x . \] We break it into \begin{align*} Q(\xi ) &= - 2{\bf j}\int_{-\infty}^{0} q(x) \,\sin (\xi x)\,{\text d} x - 2{\bf j}\int_{0}^{\infty} q(x) \,\sin (\xi x)\,{\text d} x \\ &= - 2{\bf j}\int_{0}^{\infty} \left[ q(x) - q(-x) \right] \sin (\xi x)\,{\text d} x \\ &= \int_{0}^{\infty} q_0 (x)\,\sin (\xi x)\,{\text d} x , \qquad q_0 = -2{\bf j} \left[ q(x) - q(-x) \right] . \end{align*} It is clear that q₀ ∈ 𝔏¹([0,∞)). Then we have \begin{align*} \int_e^N \frac{Q(\xi )}{\xi}\,{\text d}\xi &= \int_e^N \frac{{\text d}\xi}{\xi} \int_{0}^{\infty} q_0 (x)\,\sin (\xi x)\,{\text d} x \\ &= \int_{0}^{\infty} q_0 (x)\,{\text d}x \int_e^N \frac{\sin (\xi x)}{\xi}\,{\text d}\xi . \end{align*} Fubini's theorem guarantees that exchange of order of integration is valid since function q₀sin(ξx)/ξ is absolutely integrable. The inner integral \[ \int_e^N \frac{\sin (\xi x)}{\xi}\,{\text d}\xi \qquad\mbox{converges as } N\to\infty . \] Hence, the integral \[ q_0 (x) \int_e^N \frac{\sin (\xi x)}{\xi}\,{\text d}\xi \] admits an estimate \[ \left\vert q_0 (x) \int_e^N \frac{\sin (\xi x)}{\xi}\,{\text d}\xi \right\vert \le \left\vert q_0 (x) \right\vert . \] Since q₀ ∈ 𝔏&sip1;([0,∞)), Riemann--Lebesgue lemma allows us to take the limit under the sign of integral \[ \lim_{N\to\infty} \int_e^N \frac{\sin (\xi x)}{\xi}\,{\text d}\xi = \int_e^{\infty} \frac{\sin (\xi x)}{\xi}\,{\text d}\xi < \infty . \] We got a contradiction because the integral in the left hand-side diverges. Therefore, our assumption that there exists q(x) which Fourier transform in Q, is wrong!    ■

End of Example 3
The 𝔏 norm measures the largest magnitude a function takes—ignoring sets of measure zero. The 𝔏 space consists of all measurable functions whose magnitude is bounded almost everywhere. Formally, it is the space of essentially bounded functions.
For a measurable function f : Ω → ℝ (or ℂ) on a measure space (Ω, ℱ, μ), 𝔏-norm of f is defined as \[ \| f \|_{\infty} = \min \left\{ M \in [0, \infty ]\ \mid \ |f| \le M \mbox{ almost everywhere} \right\} = \operatorname*{ess\,sup} (f) . \] We say that f is an 𝔏 function if ∥f < ∞.
The 𝔏-norm ∥f is sometimes called the essential supremum of |f|, and 𝔏 functions are sometimes said to be essentially bounded or bounded almost everywhere. Note that a continuous function on ℝ is 𝔏 if and only if it is bounded, in which case ‖∥f is equal to the supremum of |f|. The essential supremum is not the same as the pointwise supremum. This distinction is exactly what makes 𝔏 the correct dual of 𝔏¹ duality requires ignoring measure-zero differences. For this norm, the following famous inequalities hold.

Minkowski’s Inequality: If f and g are 𝔏 functions, then f + g is 𝔏, and

Hölder's Inequality: If f is an 𝔏¹ function and g is an 𝔏 function, then their product f g is Lebesgue integrable and

\[ \| f\, g \|_{1} \leqslant \| f \|_{1} \cdot \| g \|_{\infty} . \]
Corollary 1: \( \displaystyle \quad \| \hat{f} \|_{\infty} \leqslant \| f \|_1 ,\quad \forall \ f \, \in \) 𝔏¹(ℝ). In other words, the Fourier transformation ℱ : 𝔏¹(ℝ) ⇾ 𝔏(ℝ) is a continuous linear operator. Furthemore, the Fourier transform of f belongs to ℭ0(ℝ), the space of bounded continuous functions that vanish at infinity.

   

Example 4: Let us consider an indicator function \[ f(x) = {\bf 1}_{[-1,1]} = \chi_{[-1,1]} = \begin{cases} 1, &\quad -1 \le x \le 1 , \\ 0, &\quad \mbox{otherwise}. \end{cases} \] Its 𝔏¹-norm \[ \| f \|_1 = \int_{\mathbb{R}} |f(x)|\,{\text d}x = \int_{-1}^1 1\cdot {\text d}x = 2 . \] Its Fourier transform is \[ \hat{f}(\xi ) = \int_{-1}^1 e^{-{\bf j}x\xi} \, {\text d}x = 2\,\frac{\sin\xi}{\xi} . \]
Integrate[Exp[-I*x*t], {x, -1, 1}]
(2 Sin[t])/t
With natural continous extension \( \displaystyle \quad \hat{f}(0) = 2 \quad \) (the Fourier integral at ξ = 0,). Now we check inequality formulated in Colloraly. We know the general estimate, based on Euler's formula \[ \left\vert \hat{f} \right\vert = \left\vert \int_{-1}^1 e^{-{\bf j}x\xi} \, {\text d}x \right\vert \le \int_{-1}^1 \left\vert e^{-{\bf j}x\xi} \right\vert {\text d}x = \| f \|_1 . \] So \[ \| \hat{f} \|_{\infty} = 2 = \| f \|_1 . \] Hence, for this characteristic function, the corollary is not only true, it is sharp.

This example also shows that the Fourier operator ℱ : 𝔏¹(ℝ) → 𝔏(ℝ) is continuous. The inequality \[ \| \hat{f}\|_{\infty} \le \| f \|_1 \] says exactly that the operator norm of ℱ (viewed as a map from 𝔏¹(ℝ) to 𝔏(ℝ)) is at most 1. In other words, if fₙ → f in 𝔏¹, then \[ \| \hat{f}_n - \hat{f} \|_{\infty} \le \| f_n - f\|_1 \to 0 , \] so Fourier transforms ℱ[fₙ] converges to ℱ[f] in 𝔏.    ■

End of Example 4
Theorem 3: If F and its derivative f = F′ belong to 𝔏¹(ℝ), then \[ \hat{F}(\xi ) = \frac{1}{\mathbf{j}\xi} \,\hat{f} . \]
Let F ∈ 𝔏¹(ℝ) and suppose its derivative f = F′ also lies in 𝔏¹(ℝ). We want to show that for ξ ≠ 0, \[ \hat{F}(\xi ) = \mathbf{j}\xi\,\hat{f}(\xi ), \] where the Fourier transform is defined by \[ \hat{F}(\xi ) = \int_{\mathbb{R}} F(x)\,e^{−\mathbf{j}x\xi}\,{\text d}x. \]

We start from the Fourier transform of f. Since f = F′, we compute \[ \hat{f}(\xi ) = \int_{\mathbb{R}} F′(x)\,e{−\mathbf{j}x\xi} \,{\text d}x. \]

Use integration by parts with \[ \begin{split} u&= e^{−\mathbf{j}x\xi}, \quad \Longrightarrow \quad {\text d}u = −\mathbf{j}\xi\,e^{−\mathbf{j}x\xi}\, {\text d}x, \\ {\text d}v &= F′(x)\,{\text d}x, \quad \Longrightarrow \quad v=F(x). \end{split} \] Then \[ \hat{f}(\xi ) = \int_{\mathbb{R}} F(x)\,e^{−\mathbf{j}x\xi}\,\mathbf{j}\xi\,{\text d}x . \]

Vanishing of the boundary term. We claim \[ \lim_{x\to\pm\infty} \,F(x)\, e^{−\mathbf{j}x\xi} = 0. \] This follows because:

  • F ∈ 𝔏¹(ℝ) and F′ ∈ 𝔏¹(ℝ),
hence F(x) → 0 as ∣x∣ → ∞ (a standard result: absolutely continuous functions with integrable derivative and integrable value must vanish at infinity). Thus, the boundary term is zero: \[ \left[ F(x)\,e^{−\mathbf{j}x\xi} \right]_{x=-\infty}^{x=\infty} = 0. \]

We make a conclusion: So we obtain \[ \hat{f}(\xi ) = \mathbf{j}\xi\,\hat{F}(\xi ). \] For ξ ≠ 0, rearranging gives \[ \hat{F}(\xi ) = \mathbf{j}\xi\,\hat{f}(\xi ). \] Final remark The identity holds for all ξ ≠ 0. At ξ = 0, the expression jξ1 is undefined, and the relation must be interpreted separately (typically in a limiting or distributional sense). This completes the proof.

   
Example 5: Let \[ F(x) = e^{-|x|} , \qquad x \in \mathbb{R} . \] Then F ∈ 𝔏¹(ℝ), since \[ \int_{\mathbb{R}} | F(x) |\,{\text d}x = \int_{\mathbb{R}} e^{-|x|}\,{\text d}x = 2\,\int_0^{\infty} e^{-x} \,{\text d} x = 2 < \infty . \] The (districutional) derivative of F is \[ F' (x) = f(x) = - \mbox{sgn}(x) \, e^{-|x|} , \] where the signum function is \[ \mbox{sgn}(x) = \begin{cases} \phantom{-}1 , &\quad \mbox{if } x > 0 , \\ \phantom{-}0 , &\quad \mbox{if } x = 0, \\ -1 , &\quad \mbox{if } x < 0 . \end{cases} \] Clearly, f ∈ 𝔏¹(ℝ) as well: \[ \int_{\mathbb{R}} | f(x) |\,{\text d}x = \int_{\mathbb{R}} e^{-|x|}\,{\text d}x = 2\,\int_0^{\infty} e^{-x} \,{\text d} x = 2 < \infty . \] We use the Fourier transform convention \[ \hat{f}(\xi ) = \int_{\mathbb{R}} f(x)\, e^{-\mathbf{j} x\xi} \,{\text d} x = \frac{2\mathbf{j}\xi}{1 + \xi^2} . \]
-Integrate[ Sign[x]*Exp[-Abs[x]]*Exp[-I*x*t], {x, -Infinity, Infinity}, Assumptions -> t > 0]
(2 I t)/((-I + t) (I + t))
The Fourier transform of F is \[ \hat{F}(\xi ) = \frac{2}{1 + \xi^2} , \qquad \xi \in \mathbb{R} , \]
Integrate[Exp[-Abs[x]]*Exp[-I*x*t], {x, -Infinity, Infinity}, Assumptions -> t > 0]
2/((-I + t) (I + t))
because \[ \int_0^{\infty} e^{-x} e^{-\mTHBF{J}X\XI}\,{\text d}x = \int_0^{\infty} e^{-x\left( 1 + \mathbf{j}\xi \right)} \,{\text d}x = \frac{1}{1 + \mathbf{j}\xi} . \] Theorem states thst under hypotheses F, f ∈ 𝔏¹(ℝ), and f = F′, \[ \hat{F}(\xi ) = \frac{1}{\mathbf{j}\xi}\,\hat{f}(\xi ) . \] Thus, the identity holds for all ξ ≠ 0. At ξ = 0, we have \[ \hat{F}(0) = \int_{\mathbb{R}} e^{-|x|} \,{\text d}x = 2, \qquad \hat{f}(0) = \int_{\mathbb{R}} f(x) \,{\text d}x = 0 , \] so the expression ℱ[f](ξ)/(ⅉξ) is not defined at ξ = 0, but \[ \lim_{\xi\to 0} \,\frac{1}{\mathbf{j}\xi}\, \hat{f}(\xi ) = \lim_{\xi\to 0} \,\frac{1}{\mathbf{j}\xi}\, \frac{2\mathbf{j}\xi}{1 + \xi^2} = 2 = \hat{F}(0) , \] showing that the right-hand side admits a continuous extension at ξ = 0 equal to ℱ[F](0).

This example explicitly verifies the theorem’s formula on a nontrivial pair , and also illustrates the subtlety at ξ = 0, where the identity holds in the sense of continuous extension.    ■

End of Example 5
The characteristic function of an interval [𝑎,b], denoted as χ[𝑎,b](x) or 1[𝑎,b](x), is a simple function that equals 1 if x is inside the interval [𝑎,b] and 0 otherwise

Theorem 4 (Riemann--Lebesgue lemma): If f ∈ 𝔏¹(ℝ) is an integrable function, then its Fourier transform tends to 0 at infinity: \[ \lim_{\xi\to\infty} \left\vert \hat{f}(\xi ) \right\vert = 0. \] In other words, ℱ : 𝔏¹(ℝ) → ℭ₀, the space of continuous functions vanishing at infinity.
First, we prove the theorem for characteristic functions, then for step functions, and the general case will follow from the latter because the set of step functions is dense in 𝔏¹.

Let us consider the Fourier transformation of the characteristic function \begin{align*} \Phi (a,b,\xi ) &= \int_{-\infty}^{\infty} \chi_{[a,b]} (x)\, e^{-{\bf j}x\xi} {\text d} x = \int_a^b e^{-{\bf j}x\xi} {\text d} x \\ &= \frac{e^{-{\bf j}b\xi} - e^{-{\bf j}a\xi}}{-{\bf j}\xi} \end{align*} This formula shows that the Fourier transform of a characteristic function is continuous on all real axis and \[ | \Phi (a,b,\xi )| = \frac{|e^{-{\bf j}b\xi} - e^{-{\bf j}a\xi} |}{|-{\bf j}\xi|} \le \frac{2}{|\xi |} . \] This gives \[ \lim_{\xi\to\infty} | \Phi (a,b,\xi )| = 0 . \] Any step function is a linear combination of characteristic functions: \[ h(x) = \sum_{k=1}^n \alpha_k \chi_{[a_k , b_k ]} (x) . \] Then its Fourier transform is \begin{align*} \hat{h} &= ℱ[h] = ℱ\left\{ \sum_{k=1}^n \alpha_k \chi_{[a_k , b_k ]} (x) \right\} \\ &= \sum_{k=1}^n \alpha_k ℱ\left[ \chi_{[a_k , b_k ]} \right] \\ &= \sum_{k=1}^n \alpha_k \frac{e^{-{\bf j}b_k \xi} - e^{-{\bf j}a_k \xi}}{-{\bf j}\xi} , \end{align*} which can be estimated \[ \left\vert \hat{h} \right\vert \le \frac{2}{|\xi |}\sum_{k=1}^n \left\vert \alpha_k \right\vert . \] Therefore, the Riemann--Lebesgue lemma is valid for step functions. It is known that the class of step functions is dense in 𝔏¹, which means that any Lebesgue integrable function f can be approximated with step functions: \( \displaystyle \quad h(x) = \lim_{n\to\infty} h_n (x) , \quad \) for some sequence of step functions (hₙ), namely, \[ \lim_{n\to\infty} \| f - h_n \| = 0 . \] For any ε > 0, there exists positive integer N such that for all nN, \[ \| f - h_n \| < \varepsilon . \] We estimate the Fourier transform of this difference \begin{align*} \| \hat{f} - \hat{h}_n \| &= \left\vert \int_{\mathbb{R}} \left[ f(x) - h_n (x) \right] e^{-{\bf j} x\xi} {\text d} x \right\vert \\ & \le \int_{\mathbb{R}} \left\vert f(x) - h_n (x) \right\vert {\text d} x = \| f - h_n \|_1 < \varepsilon , \quad n \ge N . \end{align*} This inequality shows that ℱ[f] is a uniform limit with respect to ξ ∈ (−∞, ∞) of a sequence of continuous functions that vanish at infinity. Hence, ℱ[f] posses all properties expressed in Riemann--Lebesgue lemma. Indeed, for any ε > 0, there exists N > 0 such that for all nN, we have \[ \left\vert \hat{f} - \hat{h}_n \right\vert < \varepsilon \qquad \forall n \ge N . \] Then \[ 0 \le \left\vert \hat{f} \right\vert < \left\vert \hat{h}_n \right\vert + \varepsilon \] For this ε > 0, find positive number M such that from inequality |ξ| > M follows inequality \( \displaystyle \quad \left\vert \hat{h}_n \right\vert < \varepsilon . \ \) In this case, the relation \( \displaystyle \quad 0 \le \left\vert \hat{f} \right\vert < \left\vert \hat{h}_n \right\vert + \varepsilon \ \) becomes \[ 0 \le \left\vert \hat{f} \right\vert < 2\varepsilon , \] if |ξ| > M. This is equivalent to \[ \lim_{|\xi | \to\infty} \left\vert \hat{f} (\xi ) \right\vert = 0 . \]

This lemma is also valid for conjugate Fourier transform: ℱ[f](ξ) → 0 as ξ → ∞.

Separating real and imaginary parts, we conclude that for any absolutely integrable function f,

\[ \int f(x)\,\cos (\xi x)\,{\text d} x \,\to\,0, \qquad \int f(x)\,\sin (\xi x)\,{\text d} x \,\to\,0, \]
as |ξ| → ∞.    
Example 6: Recall the Riemann–Lebesgue Lemma: if f ∈ 𝔏¹(ℝ), then its Fourier transform \[ ℱ[f] \equiv \hat{f}(\xi ) = \int_{\mathbb{R}} f(x)\,e^{−\mathbf{j}\xi x}\,{\text d}x \] satisfies \[ \hat{f}(\xi ) \to 0 \quad \mbox{as } ∣\xi ∣ \to \infty . \] We illustrate this with an example by choosing a simple 𝔏¹ function. Let \[ f(x) = \mathbf{1}_{[0,1]}(x) = \chi_{[0,1]}(x) , \] be the indicator (characteristic) function of the interval [0,1]. Clearly, f ∈ 𝔏¹(ℝ) and ∥f∥₁ = 1.

Compute its Fourier transform: \[ \hat{f}(\xi ) = \int_0^1 e^{−\mathbf{j}x\xi} \,{\text d}x . \] This integral is elementary if you have an access to a computer algebra system: \[ \hat{f}(\xi ) = \left[ \left( −\mathbf{j}\xi \right)^{-1} e^{−\mathbf{j}x\xi} \right]_{x=0}^{x=1} = \frac{\mathbf{j} \left( \cos\xi -1 \right) + \sin\xi }{\xi} , \quad \xi \ne 0. \]

Integrate[Exp[-I*x*t], {x, 0, 1}]
(I (-1 + Cos[t]) + Sin[t])/t
At ξ=0, we have ℱ[f](0) = 1.

Now we analyze the asymptotic behavior. We estimate the magnitude: \[ \left\vert \hat{f}(\xi ) \right\vert = \left\vert \frac{\mathbf{j} \left( \cos\xi -1 \right) + \sin\xi }{\xi} \right) \le \frac{3}{|\xi |} . \] Hence, \[ ∣\hat{f}(\xi ) \mapsto 0 \quad \mbox{as } ∣\xi ∣ \to \infty . \]

This example clearly demonstrates the Riemann–Lebesgue phenomenon:

The function f is localized in space (compact support). Its Fourier transform is spread out but decays to zero. The decay here is of order O(1/∣ξ∣), reflecting the jump discontinuities of f.

Remark: Smoother functions exhibit faster decay. For instance, if f is continuously differentiable with f′ ∈ 𝔏¹, then \[ \hat{f}(\xi ) = O\left( ∣\xi ∣^{-1} \right) , \] and higher smoothness yields even faster decay. This example provides a concrete and computable instance of the Riemann–Lebesgue Lemma in action.    ■

End of Example 6

 

  1. Carslaw, H.S., Introduction to the theory of Fourier's series and integrals, Third edition, Dover Books on Advanced Mathematics, 1950.
  2. Titchmarsh, E.C., Introduction to the theory of Fourier integrals, Oxford at the Clarendon Press, 1937.

 

Return to Mathematica page
Return to the main page (APMA0340)
Return to the Part 1 Basic Concepts
Return to the Part 2 Fourier Series
Return to the Part 3 Integral Transformations
Return to the Part 4 Parabolic PDEs
Return to the Part 5 Hyperbolic PDEs
Return to the Part 6 Elliptic PDEs
Return to the Part 6P Potential Theory
Return to the Part 7 Numerical Methods