LECTURE 9, 18.155, 6 OCTOBER, 2011 Last time I got to the middle of the proof that the Fourier transform of a distribution of compact support is a smooth function. Lemma 1. If u ∈ S 0 (Rn ) has compact support then û ∈ C ∞ (Rn ) is a function of slow growth and more precisely there exists N such that for all α |∂ξα û| ≤ Cα hξiN . (1) Continued. For a distribution of compact support there exists χ ∈ Cc∞ (Rn ) such that χu = u. This allows us to define the function F (ξ) = u(χ(x)e−ix·ξ ) since χ exp(ix · ξ) ∈ Cc∞ (Rn ) ⊂ S(Rn ). This should be the Fourier transform but we still need to check that this is so. Recall that the fact that difference quotient converges uniformly with all x derivatives on compact sets in x (namely the support of χ) (e−ix·(ξ+hej − e−ix·ξ )/h → −ixj e−ix·ξ is what shows that is differentiable and then we can use induction to see that F satisfies the estimates in (1). So we just need to show that û = F, i.e. u(φ̂) = F (φ) ∀ φ ∈ S(Rn ). Since both sides are tempered distributions it is enough to check this for φ of compact support. Think about computing u(χφ̂) in terms of φ. The integral for φ̂ is a Riemann integral (or you can do it using Lebesgue integration too of course) and so is the limit of a sum Z X φ̂(x) = exp(−iξ · x)φ(ξ) = lim 2−nj exp(−iξk,j · x)φ(ξk,j ) j→∞ k Here ξk,j is an ordering of the finite number of points in 2−j Zn in a big ball containing the support of φ and the constant is the volume of the corresponding cube(s) (I hope). As a function of x in a compact set this approximation to the Riemann integral actually converges uniformly with all its derivatives – just differentiate to see this. So X χ(x)2−nj exp(−iξk,j · x)φ(ξk,j ) → χ(x)φ̂(x) in S(Rn ). k 1 2 LECTURE 9, 18.155, 6 OCTOBER, 2011 By the continuity of u this implies X u(χφ̂) = lim u(χ(x)2−nj exp(−iξk,j · x)φ(ξk,j ) j→∞ k → χ(x)φ̂(x) in S(Rn ) Z X =⇒ lim F (ξkj )φ = F (ξ)φ(ξ) j→∞ k again by the properties of Riemann integrals. So indeed û = F. I said we would check that the extension of the Fourier transform (etc) to S 0 (Rn ) was forced by continuity here is the first version. We can say that a sequence uj ∈ S 0 (Rn ) converges weakly to u ∈ S 0 (Rn ) if uj (φ) → u(φ) in C for all φ ∈ S(Rn ). The limit is certainly unique and we can use the Fourier transform on S(Rn ) to prove that S(Rn ) ⊂ S 0 (Rn ) is weakly dense. Lemma 2. If u ∈ S 0 (Rn ) then there exists a sequence uj ∈ S(Rn ) such that φj → u weakly in S 0 (Rn ). Proof. We know that if χ ∈ Cc∞ (Rn ) has χ(x) = 1 in |x| < 1 then χk φ → φ in S(Rn ), where χk (x) = χ( xk ). Now, if u ∈ S 0 (Rn ) then u(χk φ) → u(φ) for all φ ∈ S(Rn ), which means that χk u(φ) → u(φ) which is the statement χk u → u weakly. Here, χk u has compact support, since, for instance taking χ so that χ(x) = 0 in |x| > 2, χ3k χk = χk so χ3k (χk u) = χk u. Thus we have shown that the subset of distributions of compact support, which I will denote Cc−∞ (Rn ) ⊂ S 0 (Rn ), is a weakly dense subspace. So, now (by a standard diagonalization argument) we just need to show that if u ∈ Cc−∞ (Rn ) then we can find a sequence un → u weakly in S 0 (Rn ) with un ∈ S(Rn ). Fourier transform. By the preceeding result, û ∈ C ∞ (Rn ) so χk û ∈ Cc∞ (Rn ) ⊂ S(Rn ) where we know that χk û(φ) → û(φ) ∀ φ ∈ S(Rn ) since for instance χk û → û in L1 or C00 . Anyway, this implies that uk , defined by ubk = χk û is a sequence in S(Rn ) which converges weakly to u. Another consequence of this is something you already know, namely the Riesz Representation for L2 . Here we can say it in the form Lemma 3. If u ∈ S 0 (Rn ) and |u(φ)| ≤ CkφkL2 ∀ φ ∈ S(Rn ) LECTURE 9, 18.155, 6 OCTOBER, 2011 3 then u ∈ L2 (Rn ). Proof. If u ∈ S 0 (Rn ) satisfies the estimate (3) then so does χu if χ ∈ Cc∞ (Rn ) since |χu(φ)| = |u(χφ)| ≤ CkχφkL2 ≤ C 0 kφkL2 . Then F = χ cu also satisfies the same estimate since F is bounded with respect to the L2 norm:|c χu(φ)| = |u(χφ̂)| ≤ Ckχφ̂kL2 ≤ C 0 kφ̂kL2 ≤ C 00 kφkL2 . However, from the result above, χ cu = F is a smooth function and we can easily check that this is in L2 . Namely χ( kξ )F must be Cauchy in L2 . So, using the inverse Fourier transform, which we know maps L2 into itself, we conclude that χu ∈ L2 . Then again we can take a limit of χ( xk )u to see that u ∈ L2 (Rn ). The abstract proof of the Riesz Representation Theorem in a Hilbert space is rather neater, but this concrete one is more a guide to other such constructions – for instance the Schwartz kernel theorem. So, next to the structure theorem for tempered distributions. This can be stated in various ways, let’s try a concise one. Theorem 1 (Schwartz Structure Theorem). Any u ∈ S(Rn ) is of the the form (2) u = hDi2N (hxi2N vN ), vN ∈ C00 (Rn ) ∩ L2 (Rn ) for some N ∈ N. The notation on the right here is hDh2 = 1 + |D|2 = 1 + X j Dj2 = 1 − X ∂j2 j is a differential operator of order 2 so in (2) there is a differential operator of order 2N. A somewhat weaker (but essentially equivalent) statement is that any tempered distribution is a finite sum of products xα ∂ β vα,β with the vα,β bounded and continuous (or L2 ) and this follows just by expanding out the right side of (2). Proof. The main step is a judicious application of the Sobolev embedding theorem and then the Riesz representation theorem. All we know initially is that u ∈ S 0 (Rn ) and that means that for some M, X |u(φ)| ≤ C sup |xα Dβ φ| ∀ φ ∈ S(Rn ). |α|+|β|≤M 4 LECTURE 9, 18.155, 6 OCTOBER, 2011 We start by dividing u by the functon hxi−2N – which we know has slow growth. This means of course that if v = hxi−2N u satisfies the corresponding estimate X |v(φ)| ≤ C sup |xα Dβ (hxi−M φ)| ∀ φ ∈ S(Rn ). |α|+|β|≤M Applying Leibniz formula to commute all the derivatives through, we get a big sum on the right but of the form X |v(φ)| ≤ C sup |gβ (x)Dβ φ)| ≤ kφ|C M ∀ φ ∈ S(Rn ) |β|≤M where if 2N > M then all the gβ are bounded smooth functions the polynomials being swamped by the decay in hxi−2N . Now, we just have a C N norm on the right, no polynomials in it at all. Here is where Sobolev comes in, since that says |v(φ)| ≤ Ckφ|C M ≤ C 0 kφkH M 0 , M 0 > M + n/2. Writing out the Sobolev norm explicitly we conclude that 0 |v̂(φ)| = |v(φ̂)| ≤ C 0 kφ̂|H M 0 = C 00 khξiM φkL2 . Now consider w = hξi−2N v̂ ∈ S 0 (Rn ) where 2N > M 0 , it satisfies kw(φ)k = |v̂(hξi−2N φ)| ≤ kφkL2 so, by the Riesz representation theorem above, w ∈ L2 (Rn ). If we now increase N by say n we get v̂ = hξi2N w0 where hξik w0 ∈ L2 (Rn ) for some k > n/2. This means w0 = v̂N where vN ∈ H k (Rn ) for k > n/2 and hence vN ∈ L2 (Rn ) ∩ C00 (Rn ). So actually we are done since u = hxi2N hDi2N vN . Note that we can make vN more regular, and decay more, by increasing N. Let us denote these doubly-indexes spaces by hxitH s (Rn ) = {u ∈ S 0 (Rn ); u = hxiv, v ∈ H s (Rn )}. These spaces decrease as s increases by increase as t increases. The two results this week, Sobolev embedding and Schwartz representation theorems show that \ S(Rn ) = hxi−s H s (Rn ), s 0 n S (R ) = [ s hxi−s H s (Rn ). LECTURE 9, 18.155, 6 OCTOBER, 2011 5 These show the profound difference between these two spaces. The Schwartz test functions are a ‘projective limit’ of Banach (even Hilbert) spaces – the intersection of a countable collection where the norm increases. The tempered distributions are an ‘inductive limit’ of Hilbert spaces – the union of a countable collection where the norm decreases. Exercise 1. A topology on S 0 (RSn ) is defined by taking as open sets those O ⊂ S 0 (Rn ) such that O ∩ s hxi−s H s (Rn ) is open for all s. Show that this is indeed a topology and that a linear map S 0 (Rn ) −→ C is continuous if and only if it is given by pairing with an element of S(Rn ).