18.155 LECTURE 12, 2015 22 OCTOBER RICHARD MELROSE Abstract. Notes before and after lecture – if you have questions, ask! Read: After lecture Continued with operators on Hilbert space, • Bounded operators, B(H). • Invertible operators, GL(H) = {B ∈ B(H); ∃ E ∈ B(H), EB = BE = Id}; we write E = B −1 of course. An inverse is unique if it exists since F B = Id implies F = F BE = E using associativity. • Proposition 1. GL(H) is open, in fact if B ∈ GL(H) and kAk < 1/kB −1 k then B − A ∈ GL(H). Proof by Neumann series: • Lemma 1. If kAk < 1 then Id −A ∈ GL(H). Proof. Consider the Neumann series ∞ X Ai . i=1 This converges since it is Cauchy and B(H) is complete. The Cauchy condition follows from the fact that for N ≥ M k N X Ai k ≤ i=M N X kAki < kAkN (1 − kAk)−1 . i=M The limiting operator E satisfies X AE = EA = Ai = Id −E =⇒ (Id −A)E = E(Id −A) = Id . i>0 • Proof of Proposition. If B ∈ GL(H) and A ∈ B(H) then B − A = B(Id −B −1 A) which has inverse (Id −B −1 A)−1 B −1 since kB −1 Ak ≤ kB −1 kkAk < 1. 1 2 RICHARD MELROSE • The spectrum and resolvent of a bounded operator A ∈ B(H) are complementary subsets of C : Res(A) = {z ∈ C; A − z = A − z Id ∈ GL(H)}, Spec(A) = C \ Res(A). The proposition above shows that if z0 ∈ Res(A), so A − z0 ∈ GL(H) then if |z − z0 | < 1/k(A − z0 )−1 k it follows that A − z ∈ GL(H). Thus Res(A) is open. Moveover, if |z| > kAk then A − z = −z(Id − z −1 A) ∈ GL(H) so Spec(A) ⊂ {|z| ≤ kAk}. • In general it is quite difficult to find the spectrum of a given operator. It cannot be empty. • Let’s consider self-adjoint operators. Proposition 2. If A ∈ B(H) and A = A∗ then Spec(A) ⊂ [−kAk, kAk] and either kAk or − kAk ∈ Spec(A). The first condition is equivalent to Spec(A) ⊂ R, the second says that one of A ± kAk is not invertible. • First observe that if A = A∗ and z ∈ / R then A − z is injective. Indeed, (1) Im((A − z)u, u) = − Im zkuk2 . So if (A − z)u = 0 then u = 0. • Next ‘recall’ a general fact for bounded operators. Lemma 2. For any B ∈ B(H), Ran(B) = (Nul(B ∗ ))⊥ . Proof. This is really just the defining property of the adjoint (Bu, v) = (u, B ∗ v) ∀ u, v ∈ H. Thus if v ⊥ Ran(B) then both sides vanish for all the u and so B ∗ v = 0. Conversely if B ∗ v = 0 then v ⊥ Ran(B). This means Nul(B ∗ ) = (Ran(B))⊥ . The null space is always closed but the range may not be so we can only say ((Ran(B))⊥ )⊥ = Ran(B). Taking perps of both sides gives (2). • From this it follows for A = A∗ and z ∈ / R that A − z̄ is also injective and hence that Ran(A − z) is dense. Other things from today:1. Unbounded self-adjoint operators This is a slightly old-fashioned but still usefull concept. Such an operator A is a linear map (1) A : D −→ H where D ⊂ H is required to be a dense subset. In addition we demand Symmetry: (Au, v) = (u, Av) ∀ u, v ∈ D. This of course looks like self-adjointness but we give it a different name because we require more than this. Namely set D∗ = {u ∈ H; D 3 v 7−→ (Av, u) ∈ C extends to be continous on H} L12 3 then we require D∗ = D. If u ∈ D∗ then the continuity assumptions means that Riesz’ theorem gives us h ∈ H such that (Av, u) = (v, h) for all v ∈ D. Of course if D∗ = D then symmetry implies that h = Au. If L ∈ B(H) is bounded, self-adjoint and has dense range D then L−1 is an unbounded self-adjoint operator and this is almost the general case. If B is a bounded self-adjoint operator and f is a continous function on (−kAk, kAk) then f (B) can be shown to be an unbounded self-adjoint operator. The basic properties of unbounded self-adjoint operators follow from the fact we can show that A + i is surjective as a map from D to H and has bounded inverse which is a bounded, normal operator. Then E = (A + i)−1 (A − i)−1 is 1 a positive self-adjoint operator and B = AE 2 is a bounded self-adjoint operator. The functional calculs for A can be derived from that of B, just use the fact that 1 x(1 + x2 )− 2 is an isomorphism of R to (−1, 1). Department of Mathematics, Massachusetts Institute of Technology E-mail address: rbm@math.mit.edu