PROBLEM SET 3, 18.155 BRIEF SOLUTIONS (1) Suppose Ω ⊂ Rn is open, consider (1) (2) C ∞ (Ω) = {φ : Ω −→ C; φ infinitely differentiable}. Take an exhaustion of Ω by an increasing sequence Kj of compact subsets (meaning every compact subset of Ω is contained in one of the Kj ) and show that C ∞ (Ω) is complete as a metric space where X kφ − ψk(k) d(φ, ψ) = 2−k , 1 + kφ − ψk(k) k kφk(k) = sup |Dα φ(x)|. |α|≤k, x∈Kk [Note that these are only seminorms, but I think we did enough before to prove the triangle inequality for d]. Solution: The seminorms in (2) increase with k. Since the Kj exhaust Ω, if kφk(k) = 0 for all k then φ = 0 in Ω. Thus d is a norm – since d(φ, ψ) = 0 implies φ = ψ and we showed the triangle inequality earlier. Completeness follows from the fact that a Cauchy sequence φn is uniformly Cauchy on each Kj so converges, uniformly on compact subsets, to a continuous function φ. The same is true of all derivatives Dα φn → φα . In a small ball around each point in Ω the fundamental theorem of calculus shows that φn (x+sej ) = Rs φn (x)+ 0 ∂j φn (x+tej )dt and passing to the limit, the Riemann integral converges and it follows that ∂j φ = φej for each j. Thus φ is continuously differentiable and by induction infinitely differentiable with derivatives the φα . It follows that φn → φ with respect to each of the seminorms and hence with respect to the metric, since to make d(φn , φ) < 2−k+1 it suffices to make kφn − φk(k) < 2−k−2 . Thus C ∞ (Ω) is complete. It is okay to quote some theorem about uniform convergence of derivatives. (2) Conclude that an element of the dual space denote Cc−∞ (Ω) (this is just apparently weird notation for the continuous linear maps u : C ∞ (Ω) −→ C) is precisely a linear map such that for some 1 2 PROBLEM SET 3, 18.155 BRIEF SOLUTIONS C and k (3) |u(φ)| ≤ Ckφk(k) ∀ φ ∈ C ∞ (Ω). Solution: Each of the seminorms is continuous with respect to the metric, so (3) implies that u is continous on C ∞ (Ω) and conversely, if u is continous then u−1 ({|z| < 1}) contains an open ball around 0 of radius > 0 and hence a seminorm ball kφk(k) < 2−k−2 for k so large that 2−k < /4. Probably enough to quite a general result about Fréchet spaces. (3) Observe that there is a restriction map S(Rn ) −→ C ∞ (Ω) and use this to show that (4) u0 (f ) = u(f ) ∀ f ∈ S(Rn ) defines u0 ∈ S 0 (Rn ). Solution: The restriction map S(Rn ) −→ C ∞ (Ω) is continuous, since the seminorm kφk(k) is bounded by some global norm (5) (1 + |x|2 )k/2 |Dα φ| sup x∈Rn ,|α|≤k which is continuous on S(Rn ). Thus u0 as defined is an element of S 0 (Rn ). (4) Show that the image of the the resulting map Cc−∞ (Ω) −→ S 0 (Rn ) consists of the tempered distributions with compact support contained in Ω. [You might want to use approproate cut-off functions to prove this.] Solution: Each element u0 ∈ S 0 (Rn ) obtained by restriction of u ∈ Cc−∞ (Ω) vanishes on the complement of the compact set Kj for which the estimate (3) holds, since the seminorm, and hence u(φ), vanishes on any φ ∈ Cc∞ (Rn ) with support disjoint from Kj . Conversely, suppose v ∈ S 0 (Rn ) has support K b Ω. Choose χ ∈ Cc∞ (Ω) with χ = 1 in a neighbourhood of K. Then χv = v and we may define u(φ) = v(χφ) for all φ ∈ C ∞ (Ω). This is clearly a linear map and the continuity of v means that (6) |u(φ)| = |v(χφ)| ≤ C sup (1 + |x|2 )k/2 |Dα (χφ)| ≤ C 0 kφ|(j) x,|α|≤k for j ≥ k large enough that supp χ ⊂ Kj . Thus u is continous and u0 (φ) = v(χφ) = v(φ) for φ ∈ S(Rn ). Thus each such distribution arises this way. (5) Let ∆ = D12 + · · · + Dn2 be the (positive) Laplacian on Rn . Show that if u ∈ S 0 (Rn ) and ∆u = 0 then u is a polynomial. Solution: If ∆u = 0 and u ∈ S 0 (Rn ) then |ξ|2 û = 0. This implies that supp û ⊂ {0} and we know that this means that PROBLEM SET 3, 18.155 BRIEF SOLUTIONS 3 û is a finite linear combination of the derivatives of the delta function at 0. Thus u itself is a polynomial – called an harmonic polynomial. (6) For n ≥ 3 compute ∆|x|2−n in |x| > 0 and show that it vanishes; show that |x|2−n is integrable near 0 and hence defines an element U ∈ S 0 (Rn ) (by integration of course). Conclude that ∆U has support at 0 and cannot vanish identically. Solution: In x 6= 0, |x|2−n is a smooth function and we can differentiate directly to see ∂j |x|2−n = (2 − n)|x|1−n (7) xj . |x| Differentiating again, (8) −∂j2 |x|2−n −n = −(2−n)(1−n)|x| x2 x2j 1−n 1 1−n j −(2−n)|x| +(2−n)|x| . |x|2 |x| |x|2 Combining the first and last term and then summing over j gives 0. Now, |x|s is locally integrable across 0 if s > −n so |x|2−n is a well-defined tempered distribution – outside a compact set it is equal to a symbol. So ∆U is well-defined as a tempered distribution and we have shown it has support contained in the origin. It cannot vanish since U is not a polynomial. (7) Complete the discussion above by finding the constant c so that ∆|x|2−n = cδ0 . Solution: Well, it has to be a multiple of δ0 by considering homogeneity. (8) Consider Lojasieciwz Theorem (where the j is wrong) that for every non-trivial (i.e. not identically zero) polynomail P in n variables, has a tempered fundamental solution, U ∈ S 0 (Rn ) such that P (D)U = δ0 . Show that to do this it suffices to show there is F ∈ S 0 (Rn ) such that |P (ξ)|2 f = 1 on Rn . As far as I am aware there are three proofs of this and none are easy. There are estimates by Hörmander, simplifying earlier estimates by Lojasieciwz. There is a clever proof by Atiyah using the Hironaka resolution theorem from algebraic geoemtry and there is a proof showing the meromorphy of |P (ξ)|z . Fortunately it is not important for us but you might like to look at one of them. (9) Prove this theorem for n = 1 – it is still non-trivial but the results from Thursday should allow you to do it. 4 PROBLEM SET 3, 18.155 BRIEF SOLUTIONS (10) In Problem Set 2 you showed that [ (9) S 0 (Rn ) = (1 + |x|2 )N H −2N (Rn ) N where the space on the right is a Hilbert space – the elemenst are tempered distributions of the form u = (1 + |x|2 )N U where U ∈ H −2N (Rn ) and the norm on u is the H −2N (Rn ) norm on U. Give S 0 (Rn ) the corresponding inductive limit topology – sets are open if and only if they meet these Hilbert spaces in open sets (you might want to observe that as N increases the spaces get bigger and the smaller ones are continuously included in the larger ones). Show that the dual of S 0 (Rn ) with respect to this topology is S(Rn ).