Skip to content
Shadow Theory

Section 5 9 October 2026

Ordered resolvents for an unbounded closed-form interaction

Reading position 6 of 16

5 Ordered resolvents for an unbounded closed-form interaction

Theorem 5.1 (Closed-form factor transfer and physical weights)

Let A0A_0 and A=A0+W∗JWA=A_0+W^*JW be self-adjoint operators defined by closed semibounded forms on the same V\mathcal V, with equivalent positive shifted form norms. Here W:V→HauxW:\mathcal V\to\mathcal H_{\rm aux} is bounded and J=J∗J=J^* is bounded on Haux\mathcal H_{\rm aux}. The adjoint W∗W^* maps Haux\mathcal H_{\rm aux} to V∗\mathcal V^*. Actual form closure is an assumption; a formal factorization alone does not supply it.

For ζ=E+iη\zeta=E+i\eta, η>0\eta>0, set

R0=(A0−ζ)−1,R=(A−ζ)−1,G0=WR0W∗,G=WRW∗. R_0=(A_0-\zeta)^{-1},\quad R=(A-\zeta)^{-1},\quad G_0=WR_0W^*,\quad G=WRW^* .

All products are bounded between the indicated form spaces, and

T=(I+JG0)−1=I−JG,RW∗=R0W∗T,G=G0T. T=(I+JG_0)^{-1}=I-JG,\qquad RW^*=R_0W^*T,\qquad G=G_0T . (5.1)

In particular

Im⁡G=T∗(Im⁡G0)T=η(RW∗)∗(RW∗),(RW∗)∗Tph(RW∗)=T∗[(R0W∗)∗Tph(R0W∗)]T.\begin{align}\operatorname{Im}G &=T^*(\operatorname{Im}G_0)T =\eta(RW^*)^*(RW^*), \tag{5.2}\\ (RW^*)^*T_{\rm ph}(RW^*) &=T^*\big[(R_0W^*)^*T_{\rm ph}(R_0W^*)\big]T . \tag{5.3}\end{align}

The second formula is an equality of bounded auxiliary-space forms for any positive closed physical form TphT_{\rm ph} with V⊂Dom⁡Tph1/2\mathcal V\subset\operatorname{Dom}T_{\rm ph}^{1/2} continuously. No commutation with A0,A,WA_0,A,W or JJ is required.

Proof

Choose K0=A0+c0≥IK_0=A_0+c_0\ge I. The spectral theorem gives the exact identity

∥K01/2R0K01/2∥=sup⁡λ∈σ(A0)λ+c0∣λ−ζ∣<∞. \|K_0^{1/2}R_0K_0^{1/2}\| =\sup_{\lambda\in\sigma(A_0)} \frac{\lambda+c_0}{|\lambda-\zeta|}<\infty . (5.4)

Thus R0R_0 extends boundedly V∗→V\mathcal V^*\to\mathcal V. The same argument with a positive shift of AA and equivalence of form norms gives this extension for RR. Their form inverse identities therefore give both ordered resolvent identities

R−R0=−R0W∗JWR=−RW∗JWR0. R-R_0=-R_0W^*JWR=-RW^*JWR_0.

Sandwiching by W,W∗W,W^* yields G−G0=−G0JG=−GJG0G-G_0=-G_0JG=-GJG_0. It follows by multiplication in the displayed order that both products of I+JG0I+JG_0 and I−JGI-JG equal II. This proves invertibility and the formula for TT. The first resolvent identity now gives RW∗=R0W∗(I−JG)RW^*=R_0W^*(I-JG), proving (5.1).

For a self-adjoint resolvent, Im⁡R=ηR∗R\operatorname{Im}R=\eta R^*R. The identity extends to dual inputs by continuity: both sides define bounded sesquilinear forms on V∗\mathcal V^*, and Hilbert inputs are dense there. Sandwiching with W∗W^* proves the second equality in (5.2). Substitution of RW∗=R0W∗TRW^*=R_0W^*T proves its first equality. The same substitution into the quadratic form of TphT_{\rm ph} gives (5.3). Continuity of that form on V\mathcal V justifies each product even when TphT_{\rm ph} is unbounded on H\mathcal H.

□

The resolvent inverse in this theorem exists off the real axis because the closed self-adjoint parents exist. A Neumann criterion is sufficient for a quantitative bound, not necessary for the identity:

∥JG0∥≤θ<1⟹∥T∥≤(1−θ)−1. \|JG_0\|\le\theta<1\quad\Longrightarrow\quad \|T\|\le(1-\theta)^{-1}. (5.5)

If LF−1=W∗bLF^{-1}=W^*b with bounded bb, an earned Im⁡G0/π≤s0,ηI\operatorname{Im}G_0/\pi\le s_{0,\eta}I implies

1πIm⁡[(LF−1)∗R(LF−1)]≤∥b∥2s0,η(1−θ)2I \frac1\pi\operatorname{Im} [(LF^{-1})^*R(LF^{-1})] \le \frac{\|b\|^2s_{0,\eta}}{(1-\theta)^2}I

when its full form-dual sandwich is used. Likewise (5.3) transfers an earned physical-energy resolvent bound. Absorptive response alone does not replace that physical-energy bound. For a form-dual coupling this is first a resolvent sandwich: one uses the bounded near factors of Theorem 4.3, or the ordered temporal estimate below, rather than treating LF−1LF^{-1} as a bounded Hilbert-space coupling without proof.

Equality in (5.4) is over the actual spectrum. Replacing it by a whole half-line above a spectral floor gives an upper bound, not in general an equality. Spectral gaps matter for that distinction, whereas the shift that makes K0K_0 positive does not create such gaps. At a real eigenvalue no inverse has been asserted. As η\eta decreases, genuine poles and thresholds may make both the baseline response and the quantitative inverse ceiling large.