Problem 1 ¶ (1) Prove the last lemma in lecture 12.
(2) Conclude that the Fourier transform F : S → S \mathcal F: \mathcal S\to \mathcal S F : S → S , f ↦ F f f\mapsto \mathcal Ff f ↦ F f is continuous with respect to the topology on S \mathcal S S .
Problem 2 ¶ Let K ( x ) = e − π x 2 K(x) = e^{-\pi x^2} K ( x ) = e − π x 2 be the Gaussian function. Similar to what we did with Λ and Λ ϵ \Lambda_\epsilon Λ ϵ , we define K δ ( x ) = 1 δ e − π x 2 δ K_\delta(x) = \frac{1}{\sqrt \delta}e^{-\frac{\pi x^2}{\delta}} K δ ( x ) = δ 1 e − δ π x 2 .
(1) Show that K δ K_\delta K δ is a good kernel on R \mathbb{R} R when δ → 0 \delta \to 0 δ → 0 , in the sense that ∫ − ∞ + ∞ K δ ( x ) d x = 1 ; ∫ − ∞ + ∞ ∣ K δ ( x ) ∣ d x ≤ M \int_{-\infty}^{+\infty}K_\delta(x)dx = 1; \int_{-\infty}^{+\infty}|K_\delta(x)|dx \leq M ∫ − ∞ + ∞ K δ ( x ) d x = 1 ; ∫ − ∞ + ∞ ∣ K δ ( x ) ∣ d x ≤ M ; and for η > 0 \eta> 0 η > 0 , ∫ ∣ x ∣ > η ∣ K δ ( x ) ∣ d x → 0 \int_{|x|>\eta}|K_\delta(x)|dx \to 0 ∫ ∣ x ∣ > η ∣ K δ ( x ) ∣ d x → 0 when δ → 0 \delta \to 0 δ → 0 .
(2) If f f f is continuous and bounded on R \mathbb{R} R , then f ∗ K δ → f f*K_\delta \to f f ∗ K δ → f uniformly when δ → 0 \delta \to 0 δ → 0 .
(3) Calculate the Fourier transform of K δ K_\delta K δ .
(4) Let T T T be an operator on S S S given by ( T , f ) : = lim δ → 0 ( T K δ , f ) (T,f):=\lim_{\delta\to 0} (T_{K_\delta},f) ( T , f ) := lim δ → 0 ( T K δ , f ) , calculate T T T .
(5) Show that f ( 0 ) = ∫ − ∞ + ∞ F f ( s ) d s f(0) = \int_{-\infty}^{+\infty}\mathcal Ff(s)ds f ( 0 ) = ∫ − ∞ + ∞ F f ( s ) d s for f ∈ S f\in \mathcal S f ∈ S .
(6) Give an alternative proof of inversion theorem by looking at F ( y ) = f ( y + x ) F(y) = f(y+x) F ( y ) = f ( y + x ) .
Problem 3 ¶ (1) For k ≥ 0 k\geq 0 k ≥ 0 , show that for every f ∈ C c k ( R ) f\in \mathscr{C}_c^k(\mathbb{R}) f ∈ C c k ( R ) , there exists a sequence f n f_n f n in C c ∞ ( R ) \mathscr{C}_c^{\infty}(\mathbb{R}) C c ∞ ( R ) such that f n → f f_n\to f f n → f in C c k ( R ) \mathscr{C}_c^k(\mathbb{R}) C c k ( R ) .
(2) Let T T T be a distribution of order k k k . Show that u u u extends uniquely to a continuous linear functional on C c k ( R ) \mathscr{C}_c^k(\mathbb{R}) C c k ( R ) . In other words, define T ( φ ) T(\varphi) T ( φ ) for every φ ∈ C K k ( R ) \varphi\in \mathscr{C}_K^k(\mathbb{R}) φ ∈ C K k ( R ) and show that ∣ ( T , φ ) ∣ ≤ C ∑ 0 ≤ l ≤ k ∥ φ ∥ l , K |(T,\varphi)|\leq C\sum_{0\leq l\leq k}\|\varphi\|_{l,K} ∣ ( T , φ ) ∣ ≤ C ∑ 0 ≤ l ≤ k ∥ φ ∥ l , K , where C C C is a constant depending on K K K .
Use (1), approximate φ with a sequence φ n ∈ C K ∞ ( R ) \varphi_n \in \mathscr{C}_K^{\infty}(\mathbb{R}) φ n ∈ C K ∞ ( R ) . Show that lim n → ∞ T ( φ n ) \lim_{n\to \infty} T(\varphi_n) lim n → ∞ T ( φ n ) exists, and define T ( φ ) T(\varphi) T ( φ ) to be the limit. Next, show that the limit is well-defined, i.e. show that the limit does not depend on choice approximate sequence.
Problem 4 (optional) ¶ (1) Let T : S → S T:\mathcal S\to \mathcal S T : S → S be a linear operator. Show that if T T T commutes with both d d x \frac{d}{dx} d x d and 2 π i x 2\pi i x 2 πi x , i.e. T d d x f = d d x T f T\frac{d}{dx}f = \frac{d}{dx}Tf T d x d f = d x d T f , T ( x f ) = x ( T f ) T(xf) = x(Tf) T ( x f ) = x ( T f ) , then T = c I T = cI T = c I for some constant c c c , i.e. T f = c f Tf = cf T f = c f for every f ∈ S f\in S f ∈ S .
Show that for f ∈ S f\in \mathcal S f ∈ S with f ( x 0 ) = 0 f(x_0) =0 f ( x 0 ) = 0 , then there exists ϕ ( x ) ∈ S \phi(x)\in \mathcal S ϕ ( x ) ∈ S such that f ( x ) = x ϕ ( x ) f(x) = x\phi(x) f ( x ) = x ϕ ( x ) . Then T T T commutes with x x x ⟹ \implies ⟹ ( T f ) ( x 0 ) = 0 (Tf)(x_0) = 0 ( T f ) ( x 0 ) = 0 .
Find ϕ ( x ) ∈ C ∞ ( R ) \phi(x)\in \mathscr{C}^{\infty}(\mathbb{R}) ϕ ( x ) ∈ C ∞ ( R ) such that f ( x ) = ( x − x 0 ) ϕ ( x ) f(x) = (x-x_0)\phi(x) f ( x ) = ( x − x 0 ) ϕ ( x ) . You may need the following fact from calculus: If f ( 0 ) = 0 f(0) = 0 f ( 0 ) = 0 , then f ( x ) = ∫ 0 1 ( d d t f ( t x ) ) d t f(x) = \int_{0}^1 (\frac{d}{dt}f(tx))dt f ( x ) = ∫ 0 1 ( d t d f ( t x )) d t . If ϕ ∈ S \phi \in \mathcal S ϕ ∈ S then we are done. But we only have ϕ ∈ C ∞ \phi \in \mathscr{C}^{\infty} ϕ ∈ C ∞ . We can use the result of problem 5 to modify the ϕ to be a Schwartz function. Note that the function f ( x ) x − x 0 \frac{f(x)}{x-x_0} x − x 0 f ( x ) is rapidly decreasing but has differentiablilty issue at x 0 x_0 x 0 , so you can glue the smooth ϕ with it to produce rapidly decreasing smooth function satisfying our need. This implies T f = g ⋅ f Tf = g\cdot f T f = g ⋅ f for some g ∈ S g\in \mathcal S g ∈ S .
For y ∈ R y\in \mathbb{R} y ∈ R , let e y f ( x ) : = f ( x ) − f ( y ) e − ( x − y ) 2 e_y f(x):= f(x) - f(y)e^{-(x-y)^2} e y f ( x ) := f ( x ) − f ( y ) e − ( x − y ) 2 . Then e y f ∈ S e_yf \in \mathcal S e y f ∈ S and e y f ( y ) = 0 e_yf(y) = 0 e y f ( y ) = 0 ⟹ T e y f ( y ) = 0 \implies Te_y f(y) = 0 ⟹ T e y f ( y ) = 0 . What can you conclude from this? Show that T T T commutes with d d x \frac{d}{dx} d x d implies g g g is a constant.
(2) Let T : f ↦ ( F 2 f ) − T:f\mapsto (\mathcal F^2f)^- T : f ↦ ( F 2 f ) − . Show that T T T satisfies (1) and determine c c c for T T T .
(3) Conclude that the Fourier transform F : S → S \mathcal F: \mathcal S\to S F : S → S is an isomorphism without directly using the Fourier inversion theorem. (This is an alternative proof of the inversion theorem.)
Problem 5 ¶ Prove that f ∈ S f\in S f ∈ S and T ∈ S ′ T\in S' T ∈ S ′ then f ∗ T f*T f ∗ T is a tempered distribution.
When estimating ∥ f ∗ g ∥ ( k , l ) \|f*g\|_{(k,l)} ∥ f ∗ g ∥ ( k , l ) , it will be useful to replace ∣ x ∣ k |x|^k ∣ x ∣ k by ( 1 + ∣ x ∣ ) k (1+|x|)^k ( 1 + ∣ x ∣ ) k (they are equivalent), and use the elementary inequality ( 1 + ∣ x ∣ ) ≤ ( 1 + ∣ x − y ∣ ) ( 1 + ∣ y ∣ ) (1+|x|)\leq (1+|x-y|)(1+|y|) ( 1 + ∣ x ∣ ) ≤ ( 1 + ∣ x − y ∣ ) ( 1 + ∣ y ∣ ) to handle the f ( x − y ) f(x-y) f ( x − y ) term in convolution.