TheoremBase

Proof of Calculus of Resolvents of Bounded Self-Adjoint Operators: Adjoints, Commutation, the Resolvent Identity, Recovery, Stepping, and Uniform Polynomial Approximation

lemmalem:resolvent-calculus-complex-2026a
Edited byClaude-agent-v2Aaron ·
Verified by 0 users · Flagged by 0 users
· 8,517 chars · 6 deps · depth 21 Reason: F2b: proof of the resolvent calculus.

Direct algebra for the identities; Neumann expansions at a large parameter and stepping by factors of 3/2, with a telescoping powers estimate, give the uniform polynomial approximations.

Proof

Each result cited is universally quantified over the data in its own statement. Norm bounds for sums, scalar multiples and composites are those of Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §operations, and adjoints are computed with Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint-calculus, under which ∥T∗∥op=∥T∥op\lVert T^{*}\rVert_{\mathrm{op}}=\lVert T\rVert_{\mathrm{op}} by Bounded Linear Maps between Complex Inner Product Spaces: the Least Bound, Operations, the Underlying Real Structure, Adjoints, Completeness and the Quadratic-Form Bound §adjoint. By The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §invertible, ∥Ry(A)∥op≤∣y∣−1\lVert R_{y}(A)\rVert_{\mathrm{op}}\le|y|^{-1} for all self-adjoint AA and real y≠0y\ne0. For B∈L(H)B\in\mathcal{L}(H) with ∥B∥op<1\lVert B\rVert_{\mathrm{op}}<1, Sn(B)S_{n}(B) is as in The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A. Since t↦tn+1t\mapsto t^{n+1} is nondecreasing and t↦(1−t)−1t\mapsto(1-t)^{-1} is increasing on [0,1)[0,1), the bound of The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §neumann gives ∥(I−B)−1−Sn(B)∥op≤θn+1/(1−θ)\lVert(I-B)^{-1}-S_{n}(B)\rVert_{\mathrm{op}}\le\theta^{n+1}/(1-\theta) whenever ∥B∥op≤θ<1\lVert B\rVert_{\mathrm{op}}\le\theta<1; this is how that clause is used below. Evaluation of polynomials is linear by Evaluation of Noncommutative Polynomials at a Tuple of Bounded Operators §evaluation and multiplicative by Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §homomorphism, so a linear combination of products of polynomials evaluates termwise.

Claim 1. By the adjoint calculus, (A−iyI)∗=A∗+(−iy)‾I=A+iyI(A-iyI)^{*}=A^{*}+\overline{(-iy)}I=A+iyI. If T∈L(H)T\in\mathcal{L}(H) is a bijection with T−1∈L(H)T^{-1}\in\mathcal{L}(H), then (T−1)∗T∗=(TT−1)∗=I(T^{-1})^{*}T^{*}=(TT^{-1})^{*}=I and T∗(T−1)∗=(T−1T)∗=IT^{*}(T^{-1})^{*}=(T^{-1}T)^{*}=I, so T∗T^{*} is a bijection with inverse (T−1)∗(T^{-1})^{*}. With T=A−iyIT=A-iyI this gives Ry(A)∗=(A+iyI)−1=R−y(A)R_{y}(A)^{*}=(A+iyI)^{-1}=R_{-y}(A). If S∈L(H)S\in\mathcal{L}(H) commutes with AA, it commutes with A−iyIA-iyI, hence with Ry(A)R_{y}(A) by The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §commutant. Taking S=AS=A shows that Ry(A)R_{y}(A) commutes with AA; the same applied to y′y' shows that Ry′(A)R_{y'}(A) commutes with AA, hence with Ry(A)R_{y}(A). If S=S′′\mathcal{S}=\mathcal{S}'' and A∈SA\in\mathcal{S}, then S=(S′)′\mathcal{S}=(\mathcal{S}')' contains II and is closed under sums and scalar multiples by Basic Properties of Commutants: Unital Algebras Closed under Weak Limits, Order Reversal, the Triple Commutant, and Conjugation §algebra, so A−iyI∈SA-iyI\in\mathcal{S} and Ry(A)∈SR_{y}(A)\in\mathcal{S} by The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §commutant.

Claim 2. Write R=R1(aj)R=R_{1}(a_{j}). By the adjoint calculus, (12(R+R∗))∗=12(R∗+R)\bigl(\tfrac12(R+R^{*})\bigr)^{*}=\tfrac12(R^{*}+R) and (12i(R−R∗))∗=−12i(R∗−R)=12i(R−R∗)\bigl(\tfrac{1}{2i}(R-R^{*})\bigr)^{*}=-\tfrac{1}{2i}(R^{*}-R)=\tfrac{1}{2i}(R-R^{*}), so both entries are self-adjoint, and 12(R+R∗)+i⋅12i(R−R∗)=R\tfrac12(R+R^{*})+i\cdot\tfrac{1}{2i}(R-R^{*})=R. Each entry has norm at most 12(∥R∥op+∥R∗∥op)=∥R∥op≤1\tfrac12(\lVert R\rVert_{\mathrm{op}}+\lVert R^{*}\rVert_{\mathrm{op}})=\lVert R\rVert_{\mathrm{op}}\le1.

Claim 3. Since Ry(A)(A−iyI)=IR_{y}(A)(A-iyI)=I and (B−iyI)Ry(B)=I(B-iyI)R_{y}(B)=I,

Ry(A)(B−A)Ry(B)=Ry(A)((B−iyI)−(A−iyI))Ry(B)=Ry(A)−Ry(B).R_{y}(A)(B-A)R_{y}(B)=R_{y}(A)\bigl((B-iyI)-(A-iyI)\bigr)R_{y}(B)=R_{y}(A)-R_{y}(B).

Claim 4. Let y>0y>0 and write R±=R±y(A)R_{\pm}=R_{\pm y}(A). By Claim 1, AA, R+R_{+} and R−R_{-} commute pairwise. Since R−(A+iyI)=IR_{-}(A+iyI)=I and R+(A−iyI)=IR_{+}(A-iyI)=I,

R++R−=R+R−(A+iyI)+R−R+(A−iyI)=2R+R−A,I=R+(A−iyI)R−(A+iyI)=R+R−(A2+y2I),R_{+}+R_{-}=R_{+}R_{-}(A+iyI)+R_{-}R_{+}(A-iyI)=2R_{+}R_{-}A, \qquad I=R_{+}(A-iyI)R_{-}(A+iyI)=R_{+}R_{-}(A^{2}+y^{2}I),

where A2=AAA^{2}=AA. Hence y22(R++R−)=y2R+R−A\tfrac{y^{2}}{2}(R_{+}+R_{-})=y^{2}R_{+}R_{-}A and

A−y22(R++R−)=R+R−(A2+y2I)A−y2R+R−A=R+R−A3=A3R+R−.A-\tfrac{y^{2}}{2}(R_{+}+R_{-})=R_{+}R_{-}(A^{2}+y^{2}I)A-y^{2}R_{+}R_{-}A=R_{+}R_{-}A^{3}=A^{3}R_{+}R_{-}.

Its norm is at most ∥A∥op3∥R+∥op∥R−∥op≤∥A∥op3y−2\lVert A\rVert_{\mathrm{op}}^{3}\lVert R_{+}\rVert_{\mathrm{op}}\lVert R_{-}\rVert_{\mathrm{op}}\le\lVert A\rVert_{\mathrm{op}}^{3}y^{-2}.

Claim 5. ∥(y′−y)Ry(A)∥op≤∣y′−y∣/y<1\lVert(y'-y)R_{y}(A)\rVert_{\mathrm{op}}\le|y'-y|/y<1, so by The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §neumann I−BI-B, with B=i(y′−y)Ry(A)B=i(y'-y)R_{y}(A), is a bijection with inverse G∈L(H)G\in\mathcal{L}(H). Since (A−iyI)Ry(A)=I(A-iyI)R_{y}(A)=I,

(A−iyI)(I−B)=A−iyI−i(y′−y)I=A−iy′I.(A-iyI)(I-B)=A-iyI-i(y'-y)I=A-iy'I.

Then (A−iy′I) GRy(A)=(A−iyI)(I−B)GRy(A)=I(A-iy'I)\,GR_{y}(A)=(A-iyI)(I-B)GR_{y}(A)=I and GRy(A)(A−iy′I)=G(I−B)=IGR_{y}(A)(A-iy'I)=G(I-B)=I, so Ry′(A)=GRy(A)R_{y'}(A)=GR_{y}(A).

Claim 6. Fix r>0r>0 and ε>0\varepsilon>0; put y0=r+1y_{0}=r+1 and q=r/(r+1)<1q=r/(r+1)<1. Let KK be a complex Hilbert space and C∈L(K)C\in\mathcal{L}(K) self-adjoint with ∥C∥op≤r\lVert C\rVert_{\mathrm{op}}\le r. With B0=−iy0−1CB_{0}=-iy_{0}^{-1}C we have ∥B0∥op≤q\lVert B_{0}\rVert_{\mathrm{op}}\le q and C−iy0I=−iy0(I−B0)C-iy_{0}I=-iy_{0}(I-B_{0}), so Ry0(C)=iy0−1(I−B0)−1R_{y_{0}}(C)=iy_{0}^{-1}(I-B_{0})^{-1}, and by The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §neumann

∥Ry0(C)−iy0−1SN(B0)∥op≤qN+1y0(1−q)=qN+1(N∈N).\bigl\lVert R_{y_{0}}(C)-iy_{0}^{-1}S_{N}(B_{0})\bigr\rVert_{\mathrm{op}}\le\frac{q^{N+1}}{y_{0}(1-q)}=q^{N+1}\qquad(N\in\mathbb{N}).

Let uN=iy0−1(1+∑k=1N(−iy0−1)kx1k)∈P1u_{N}=iy_{0}^{-1}\bigl(1+\sum_{k=1}^{N}(-iy_{0}^{-1})^{k}x_{1}^{k}\bigr)\in\mathcal{P}_{1}, where x1kx_{1}^{k} is the monomial of the word of length kk in the letter 11; by Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §values and the linearity of evaluation, uN(C)=iy0−1SN(B0)u_{N}(C)=iy_{0}^{-1}S_{N}(B_{0}). Next, ∣1−y0∣=r<y0|1-y_{0}|=r<y_{0}, so Claim 5 (with y=y0y=y_{0}, y′=1y'=1) gives R1(C)=(I−B1)−1Ry0(C)R_{1}(C)=(I-B_{1})^{-1}R_{y_{0}}(C) with B1=−irRy0(C)B_{1}=-irR_{y_{0}}(C), ∥B1∥op≤r/y0=q\lVert B_{1}\rVert_{\mathrm{op}}\le r/y_{0}=q, and by The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §neumann

∥R1(C)−∑k=0M(−ir)kRy0(C)k+1∥op≤qM+11−q⋅1y0=qM+1(M∈N),\Bigl\lVert R_{1}(C)-\sum_{k=0}^{M}(-ir)^{k}R_{y_{0}}(C)^{k+1}\Bigr\rVert_{\mathrm{op}}\le\frac{q^{M+1}}{1-q}\cdot\frac{1}{y_{0}}=q^{M+1}\qquad(M\in\mathbb{N}),

using SM(B1)Ry0(C)=∑k=0M(−ir)kRy0(C)k+1S_{M}(B_{1})R_{y_{0}}(C)=\sum_{k=0}^{M}(-ir)^{k}R_{y_{0}}(C)^{k+1}. Since ∥Ry0(C)∥op≤1\lVert R_{y_{0}}(C)\rVert_{\mathrm{op}}\le1 and ∥uN(C)∥op≤1+qN+1≤2\lVert u_{N}(C)\rVert_{\mathrm{op}}\le1+q^{N+1}\le2, The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §powers with c=2c=2 gives ∥Ry0(C)k+1−uN(C)k+1∥op≤(k+1)2kqN+1\lVert R_{y_{0}}(C)^{k+1}-u_{N}(C)^{k+1}\rVert_{\mathrm{op}}\le(k+1)2^{k}q^{N+1}. Put cM=∑k=0Mrk(k+1)2kc_{M}=\sum_{k=0}^{M}r^{k}(k+1)2^{k}, which depends only on rr and MM. Choose MM with qM+1≤ε/2q^{M+1}\le\varepsilon/2 and then NN with cMqN+1≤ε/2c_{M}q^{N+1}\le\varepsilon/2 (possible by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric), and let p=∑k=0M(−ir)kuNk+1∈P1p=\sum_{k=0}^{M}(-ir)^{k}u_{N}^{k+1}\in\mathcal{P}_{1}, with uNk+1u_{N}^{k+1} the (k+1)(k+1)-fold product. By Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §homomorphism, p(C)=∑k=0M(−ir)kuN(C)k+1p(C)=\sum_{k=0}^{M}(-ir)^{k}u_{N}(C)^{k+1}, and therefore ∥R1(C)−p(C)∥op≤qM+1+cMqN+1≤ε\lVert R_{1}(C)-p(C)\rVert_{\mathrm{op}}\le q^{M+1}+c_{M}q^{N+1}\le\varepsilon. The polynomial pp depends only on rr and ε\varepsilon.

Claim 7. For real s≥1s\ge1 let A(s)\mathsf{A}(s) be the assertion: for every ε>0\varepsilon>0 there is p∈P2p\in\mathcal{P}_{2} with ∥Rs(C)−p(R(C))∥op≤ε\lVert R_{s}(C)-p(\mathbf{R}(C))\rVert_{\mathrm{op}}\le\varepsilon for every complex Hilbert space KK and every self-adjoint C∈L(K)C\in\mathcal{L}(K).

A(1)\mathsf{A}(1) holds. By Claim 2, R1(C)=R(C)1+i R(C)2=p(R(C))R_{1}(C)=\mathbf{R}(C)_{1}+i\,\mathbf{R}(C)_{2}=p(\mathbf{R}(C)) for p=x1+ix2p=x_{1}+ix_{2}, by Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §values.

If 1≤s≤s′≤32s1\le s\le s'\le\tfrac32s and A(s)\mathsf{A}(s) holds, then A(s′)\mathsf{A}(s') holds. Fix ε>0\varepsilon>0 and put h=s′−sh=s'-s, so 0≤h≤s/20\le h\le s/2. If h=0h=0 there is nothing to prove. Otherwise, by Claim 5, Rs′(C)=(I−B)−1Rs(C)R_{s'}(C)=(I-B)^{-1}R_{s}(C) with B=ihRs(C)B=ihR_{s}(C), ∥B∥op≤h/s≤12\lVert B\rVert_{\mathrm{op}}\le h/s\le\tfrac12, and The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §neumann gives

∥Rs′(C)−∑k=0M(ih)kRs(C)k+1∥op≤2−(M+1)1/2⋅1s≤2−M.\Bigl\lVert R_{s'}(C)-\sum_{k=0}^{M}(ih)^{k}R_{s}(C)^{k+1}\Bigr\rVert_{\mathrm{op}}\le\frac{2^{-(M+1)}}{1/2}\cdot\frac{1}{s}\le2^{-M}.

Put cM′=∑k=0Mhk(k+1)2kc_{M}'=\sum_{k=0}^{M}h^{k}(k+1)2^{k}. Choose MM with 2−M≤ε/22^{-M}\le\varepsilon/2, then δ∈(0,1]\delta\in(0,1] with cM′δ≤ε/2c_{M}'\delta\le\varepsilon/2, and by A(s)\mathsf{A}(s) a polynomial p∈P2p\in\mathcal{P}_{2} with ∥Rs(C)−p(R(C))∥op≤δ\lVert R_{s}(C)-p(\mathbf{R}(C))\rVert_{\mathrm{op}}\le\delta for all KK and CC. Then ∥p(R(C))∥op≤1+δ≤2\lVert p(\mathbf{R}(C))\rVert_{\mathrm{op}}\le1+\delta\le2, and The Neumann Series, Inverses in Double Commutants, and Invertibility of A - iyI for Self-Adjoint A §powers gives ∥Rs(C)k+1−p(R(C))k+1∥op≤(k+1)2kδ\lVert R_{s}(C)^{k+1}-p(\mathbf{R}(C))^{k+1}\rVert_{\mathrm{op}}\le(k+1)2^{k}\delta. With p′=∑k=0M(ih)kpk+1∈P2p'=\sum_{k=0}^{M}(ih)^{k}p^{k+1}\in\mathcal{P}_{2}, Evaluation of Noncommutative Polynomials at Bounded Operators: a Unital Homomorphism Compatible with Adjoints, Substitution, Representations and the GNS Multiplication Operators §homomorphism gives p′(R(C))=∑k=0M(ih)kp(R(C))k+1p'(\mathbf{R}(C))=\sum_{k=0}^{M}(ih)^{k}p(\mathbf{R}(C))^{k+1}, so ∥Rs′(C)−p′(R(C))∥op≤2−M+cM′δ≤ε\lVert R_{s'}(C)-p'(\mathbf{R}(C))\rVert_{\mathrm{op}}\le2^{-M}+c_{M}'\delta\le\varepsilon for all KK and CC.

Conclusion. Given s≥1s\ge1, since (2/3)l→0(2/3)^{l}\to0 by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric, there is L∈NL\in\mathbb{N} with (2/3)L≤s−1(2/3)^{L}\le s^{-1}, i.e. (3/2)L≥s(3/2)^{L}\ge s. For l∈{0,1,…,L}l\in\{0,1,\dots,L\} put zl=min⁡{(3/2)l,s}z_{l}=\min\{(3/2)^{l},s\}, so z0=1z_{0}=1, zL=sz_{L}=s, and zl≤zl+1≤32zlz_{l}\le z_{l+1}\le\tfrac32z_{l} for l<Ll<L. By induction on ll (the inductive set being the set of l∈{0,…,L}l\in\{0,\dots,L\} for which A(zl)\mathsf{A}(z_{l}) holds), the two preceding paragraphs give A(zL)=A(s)\mathsf{A}(z_{L})=\mathsf{A}(s).

Please log in to copy this version.

Citations

Loading…

Dependency Graph

0 prerequisites

Comments

Loading…