LECTURE 11, 18.155, 18 OCTOBER 2011 (1) Lemma 1. For any s ∈ R, m ∈ N, u ∈ H s (Rn ) and Dα u ∈ H s (Rn ) for all α ∈ N with |α| = m is equivalent to u ∈ H s+m (Rn ). In fact it suffices to have u ∈ H s and Djm u ∈ H s for j = 1, . . . , n. Proof. hξis û ∈ L2 and ξjm hξis û ∈ L2 j = 1, . . . , n together imply that hξis+m û ∈ L2 . k Lemma 2. The elements of C∞ (Rn ) – the functions with all derivatives up to order k ∈ N bounded and continuous – are multipliers on H s (Rn ) provided |s| ≤ k. Proof. Certainly true for s = k, since the product exists in S 0 (Rn ) and expanding out a derivative of it, up to order k, by Leibniz’ formula gives a sum where each term is in L2 . Next consider the case 0 < s < k = 1. Then we may use the characterization of H s you did in the homework last week. This shows that Z |u(x) − u(y)|2 2 kukL2 + dxdy n+2s |x−y|≤1 |x − y| is the square of an equivalent norm on H s (Rn ). Here I have dropped the other part of the ‘Hölder’ type integral since it is controlled by the L2 norm anyway: Z Z |u(x) − u(y)|2 1 |u(x)|2 dxdy ≤ 4 dxdy ≤ Ckuk2L2 . n+2s n+2s |x − y| |x−y|≥1 |x − y| |x−y|≥1 1 Now observe that if u ∈ H s and f ∈ C∞ (bbRn ), then using the estimate |f (x) − f (y)| ≤ sup |Df ||x − y| Z |f (x)u(x) − f (y)u(y)|2 dxdy |x − y|n+2s |x−y|<1 Z |f (x)u(x) − f (x)u(y) + f (x)u(y) − f (y)u(y)|2 = dxdy |x − y|n+2s |x−y|<1 Z Z |u(x) − u(y)|2 |u(y)|2 ≤ 2 sup |f | dxdy + 2 sup |Df | dxdy n+2s n+2s−2 |x−y|<1 |x − y| |x−y|≤1 |x − y| 1 2 LECTURE 11, 18.155, 18 OCTOBER 2011 so this is bounded in terms of H s and we can use density of S(Rn ) in H s . Now, the case of general s > 1 follows by considering the largest integer l ≤ s and combining these two arguments to show that Dα (f u) ∈ H s−l for all |α| = l, so f u ∈ H s (Rn ). The case of negative s follows by duality since u ∈ H s with s < 0 iff if |u(φ)| ≤ CkφkH −s and then |f u(φ)| = |u(f φ)| ≤ Ckf φk)H −s ≤ C 0 kf kC∞ k kφkH −s by the discussion for s > 0. (2) Now, back to homogeneity. We are in the middle of discussing the space of distributions, on the line, which are homogeneous of degree z ∈ C – meaning u(φt ) = tz+1 u(φ), ∀ φ ∈ Cc∞ (R), φt (x) = φ(x/t), t > 0 (where the +1 comes from the homogeneity of dx.) The convergence of the difference quotient φt (x) − φ(x) → −xφ0 (x) as t → 1 t−1 is uniform with all derivatives and uniform support – certainly in S(R) but actually in Cc∞ (R). Anyway, we are assuming u ∈ S 0 (R) (although this is unnecessary) so from (2) it follows that d u(φt ) − u(φ) → u(−xφ0 (x) = ( (xu))(φ) = (z + 1)u(φ) t−1 dx so u satisfies the Euler’s differential equation du x = zu on R. dx So this is a little exercise in differential equations. Consider what happens in x > 0 – in fact we can see that if χ ∈ Cc∞ (R) has support in x > 0 then 1 dχu = zχu − χ0 u. dx x We need to check that χu ∈ C 1 and then equation holds in the classical sense and since we can take χ = 1 on any interval we want, du x = zu =⇒ u(x) = cxz+ on x > 0 dx LECTURE 11, 18.155, 18 OCTOBER 2011 3 by integration. So, it is only the regularity we need. Fix χ ∈ Cc∞ (R) with support contained in [a, b] where b > a > 0. We can choose another χ1 ∈ Cc∞ (Rn ) with χ0 = 1 on [a − , b + ] for some a > > 0 and then successive χk ’s, k ≥ 1, with ‘nested supports’ χk (x) = 0 if x < a − /k or x > b + /k and χk = 1 if x ≥ a − /(k + 1) or x ≤ b + /(k + 1). Thus χk χk+1 = χk+1 . Now, χ0 u has compact support so is in some H s where s may be negative as far as we know. The differential equation (2) for χ = χk shows that dχk u 1 1 = zχk u − χ0k u = zχk (χk−1 u) − χ0k (χk−1 u). dx x x Proceeding inductively this shows that χk u ∈ H s+k for all k. Namely, it is true for k = 0 and if true for k − 1 then χk u = χk χk−1 u ∈ H s+k−1 and (2) shows that dχk u/dx ∈ H s+k−1 as well, hence χk u ∈ H s+k . So, by the Sobolev embedding theorem, taking k > |s| + 1, this actually shows that u ∈ C 1 (a, b) on any interval (a, b) with b > a > 0 and the argument leading to (2) is justified. The same thing works in x < 0 of course, so we conclude that if u is homogeneous of degree z then ( c+ xz+ if x > 0 u= c− cz− if x < 0. Now, let’s assume that Re z > −1 since then we showed last time that xz± are well-defined tempered distributions and what (2) shows is that (1) v = u − c+ xz+ − c− xz− ∈ S 0 (R) has support contained in {0}. So, in fact we know that v can only be a finite sum of multiples of derivatives of the delta function. However, we can check immediately that δ(φt ) = φt (0) = φ(0) = δ(φ) which means that δ is homogeneous of degree −1. (Not 0 becuase of the z + 1 in the definition of distributional homogeneity.) The derivative dk δ/dxk clearly has homogeneity −1 − k by the same argument, so no combination of these can be homogeneous of degree z with Re z > −1 (except the zero combination). 4 LECTURE 11, 18.155, 18 OCTOBER 2011 So, finally we see that v = 0 in (1) and (2) The space of homogeneous distributions on R of degree z, Re z > −1 is of dimension two spanned by xz± . (3) Now, to the case Re z ≤ 1. The Fourier transform of test functions gives, by changing variable, Z b φt = e−ixξ φ(x/t)dx = tφ̂(ξt). So, if u is homogeneous of any degree z, (3) û(φt ) = u(φbt ) = tu(φ̂(·t)) = t(1/t)z+1 u(φ̂) = t−z û(φ), shows that û is homogeneous of degree Z = −z − 1 (so that t−z = tZ+1 . Thus, if Re z < 0 then û is homogeneous of degree Z with Re Z = − Re z −1 > −1. There is some overlap! So by applying the Fourier transform we conclude that the spac of tempered distributions homogeneous of degree z is always of dimension two – although the spanning part of (2) is not always true. (4) What really happens? If φ ∈ S(R) then Z ∞ z F (z) = x+ (φ) = xz φ(x)dx for Re z > −1 0 is actually a holomorphic function there and extends to a meromorphic function of z ∈ C with poles at z ∈ −R. Holomorphy in Re z > 0 follows by differntiating the integral – which is easily seen to be once continuously differentiable in u and v if z = u + iv and to satisfy ∂¯z F (z) = 0 which is one definition of holomorphy. Moreover the ‘functional equation’ xz+ (φ0 ) = −zxz−1 (φ), Re z > 0 follows by integration by parts. This however can be written 1 xz+ (φ) = − xz+1 (φ0 ) Re z > 1 z+1 + and then iterated to get (4) xz+ (φ) = (−1) k k Y 1 z+k dk φ x ( ), Re z > k z + j + dxk j=1 for any positive integer k. However, the right side is then holomorphic, as a function of z, in Re z > −k − 1 so, by the uniqueness of holomorphic extension, the left side must extend to a LECTURE 11, 18.155, 18 OCTOBER 2011 5 holomorphic function as stated – with the only poles arising from the vanishing of the z + j. Proposition 1. The distribution xz+ is well-defined by analytic continuation, i.e. (4) for all z ∈ C \ (−N) and the residue at k−1 z = −k is a multiple of ddk−1 xδ . Proof. The continuity of φ 7−→ xz+ (φ) also follows from (4). The residue at z = −k is dk φ 1 dk−1 φ 1 (−1)k−1 dk−1 δ − x0+ ( k ) = δ( k−1 ) = (φ). (k − 1)! dx (k − 1)! dx (k − 1)! dxk−1 Essentially the same thing is true for xz− with sign changes. From this you can check that a certain combination xz− + ak xz− has no pole at z = −k and therefore the value there defines a homogeneous distribution of degree −k. The upshot is that xz± span the homogeneous distributions for z ∈ / −N but for the dk−1 δ negative integers this combination, plus dxk−1 span the homogeneous distributions. Just to make it doubly clear The function x−k defined on x > 0 is not the restriction of homogeneous distribution (which would have to be of degree −k) which vanishes in x < 0. However, it is the restriction to x > 0 of an element of S 0 (R) which vanishes in x < 0 it just cannot be homogeneous. Exercise 1. Can you find a distribution with this property? Hint: Subtract (z+k) times the residue and evaluate at z = −k.