Estimation.NPIV

Nonparametric instrumental variables: the conditional-moment operator, sieve primal analysis, ill-posedness measures, and doubly robust NPIV functionals.

DR 11 core · 3 supporting · 3 submodules Doubly robust estimation of linear functionals of the NPIV structural function. Operator 39 core · 32 supporting · 4 submodules Extends the inverse-problem system with population operators for NPIV estimation. Primal 28 core · 16 supporting · 3 submodules The sieve primal NPIV analysis: projected residuals, variance and bias terms, and the rate theorem for the structural function.
Setup 8 core · 0 supporting This file provides the substrate for the non-iterated Tikhonov regularized adversarial estimator (TRAE) of Bennett, Kallus, Mao, Newey, Syrgkanis, and Uehara (2023).

NPIV / TRAE setup: linear inverse-problem functional system

This file provides the substrate for the non-iterated Tikhonov regularized adversarial estimator (TRAE) of Bennett, Kallus, Mao, Newey, Syrgkanis, and Uehara (2023). The target is a scalar linear functional of a primal solution to a linear inverse problem; the main algebraic object is a doubly robust score built from the primal solution and from the solution of a dual inverse problem.

Mirrors the following definitions from doc/basic_concepts/po/estimation/trae_inverse_problems.tex:

* def:est-trae-system — the linear inverse-problem functional system. * def:est-trae-dual-solution — the dual solution q₀. * def:est-trae-dr-functional — the doubly-robust functional Θ and its pointwise score φ_{h,q}.

The candidate sets Hbar, Qbar are kept as plain sets of ℝ-valued functions; the Hilbert, Lp, and projection structure is added by later operator-system files. This setup layer records the moment-equation contract and the linearity needed for the mixed-bias proof.

structure InverseProblemSystem reviewed
Causalean.Estimation.NPIV

A linear inverse-problem functional system packages an observation space together with measurable treatment-side and instrument-side covariate projections of the observed data, measurable closed candidate sets for the primal and dual nuisances, each closed under subtraction, observation-level moment maps that are linear in their nuisance argument, jointly measurable and integrable against every candidate, whose cross product q(Z)h(X) is integrable as well, and a primal nuisance in the candidate set satisfying the inverse-problem moment equation against every dual candidate (def:est-trae-system).

Definition (Lean source)
Ω :
μ :
Observation type.
𝒲 :
Type*
"Action" / treatment-side covariate type used by `Hbar`-side functions.
𝒳 :
Type*
Instrument-side covariate type used by `Qbar`-side functions.
𝒵 :
Type*
inst𝒲 :
inst𝒳 :
inst𝒵 :
Observed data point.
W :
Ω → 𝒲
meas_W :
@Measurable Ω 𝒲 _ inst𝒲 W
Treatment-side projection from `𝒲` to `𝒳`.
xOf :
𝒲 → 𝒳
Instrument-side projection from `𝒲` to `𝒵`.
zOf :
𝒲 → 𝒵
meas_xOf :
@Measurable 𝒲 𝒳 inst𝒲 inst𝒳 xOf
meas_zOf :
@Measurable 𝒲 𝒵 inst𝒲 inst𝒵 zOf
Closed candidate set for the primal nuisance (`Hbar ⊆ L²(P_X)` in the TeX; here we keep it as a plain set of measurable ℝ-valued functions).
Hbar :
Set (𝒳 → ℝ)
Closed candidate set for the dual nuisance (`Qbar ⊆ L²(P_Z)` in the TeX; here we keep it as a plain set of measurable ℝ-valued functions).
Qbar :
Set (𝒵 → ℝ)
meas_of_Hbar :
∀ h ∈ Hbar, Measurable h
meas_of_Qbar :
∀ q ∈ Qbar, Measurable q
Closure under subtraction (the only algebraic property the mixed-bias proof needs; standing in for the closed-linear-subspace assumption).
Hbar_sub :
∀ h₁ ∈ Hbar, ∀ h₂ ∈ Hbar, (fun x => h₁ x - h₂ x) ∈ Hbar
Qbar_sub :
∀ q₁ ∈ Qbar, ∀ q₂ ∈ Qbar, (fun z => q₁ z - q₂ z) ∈ Qbar
Observation-level moment map `m(W; q)`, linear in the dual nuisance argument `q`.
m :
𝒲 → (𝒵 → ℝ) → ℝ
Observation-level moment map `m_e(W; h)`, linear in the primal nuisance argument `h`.
m_e :
𝒲 → (𝒳 → ℝ) → ℝ
Linearity of `m` over subtraction in the dual argument.
m_sub :
∀ (w : 𝒲) (q₁ q₂ : 𝒵 → ℝ), m w (fun z => q₁ z - q₂ z) = m w q₁ - m w q₂
Linearity of `m_e` over subtraction in the primal argument.
m_e_sub :
∀ (w : 𝒲) (h₁ h₂ : 𝒳 → ℝ), m_e w (fun x => h₁ x - h₂ x) = m_e w h₁ - m_e w h₂
Joint measurability of `ω ↦ m(W ω; q)` for `q ∈ Qbar`.
measurable_m :
∀ q ∈ Qbar, Measurable (fun ω => m (W ω) q)
Joint measurability of `ω ↦ m_e(W ω; h)` for `h ∈ Hbar`.
measurable_m_e :
∀ h ∈ Hbar, Measurable (fun ω => m_e (W ω) h)
Integrability of `ω ↦ m(W ω; q)` for `q ∈ Qbar`.
integrable_m :
∀ q ∈ Qbar, Integrable (fun ω => m (W ω) q) μ
Integrability of `ω ↦ m_e(W ω; h)` for `h ∈ Hbar`.
integrable_m_e :
∀ h ∈ Hbar, Integrable (fun ω => m_e (W ω) h) μ
Integrability of the cross product `q(Z) · h(X)` for `h ∈ Hbar, q ∈ Qbar`, where `X = xOf ∘ W` and `Z = zOf ∘ W`.
integrable_qh :
∀ h ∈ Hbar, ∀ q ∈ Qbar, Integrable (fun ω => q (zOf (W ω)) * h (xOf (W ω))) μ
Primal nuisance candidate `h₀ ∈ Hbar` satisfying the moment equation.
h₀ :
𝒳 → ℝ
h₀_mem :
h₀ ∈ Hbar
Primal moment identity (Definition 4.5, item 5): `E[m(W; q)] = E[h₀(X) q(Z)]` for every `q ∈ Qbar`.
primal_moment :
∀ q ∈ Qbar, ∫ ω, m (W ω) q ∂μ = ∫ ω, h₀ (xOf (W ω)) * q (zOf (W ω)) ∂μ
Causalean.Estimation.NPIV.InverseProblemSystem · Causalean/Estimation/NPIV/Setup.lean:43
def X reviewed
Causalean.Estimation.NPIV.InverseProblemSystem

Treatment-side covariate as a random variable X := xOf ∘ W.

Definition (Lean source)
def X (S : InverseProblemSystem Ω μ) (ω : Ω) : S.𝒳 := S.xOf (S.W ω)
Causalean.Estimation.NPIV.InverseProblemSystem.X · Causalean/Estimation/NPIV/Setup.lean:141 · uses InverseProblemSystem
def Z reviewed
Causalean.Estimation.NPIV.InverseProblemSystem

Instrument-side covariate as a random variable Z := zOf ∘ W.

Definition (Lean source)
def Z (S : InverseProblemSystem Ω μ) (ω : Ω) : S.𝒵 := S.zOf (S.W ω)
Causalean.Estimation.NPIV.InverseProblemSystem.Z · Causalean/Estimation/NPIV/Setup.lean:144 · uses InverseProblemSystem
def θ₀ reviewed
Causalean.Estimation.NPIV.InverseProblemSystem

Scalar target θ₀ := E[m_e(W; h₀)] (Definition 4.5, item 6).

Definition (Lean source)
noncomputable def θ₀ (S : InverseProblemSystem Ω μ) : ℝ := ∫ ω, S.m_e (S.W ω) S.h₀ ∂μ
Causalean.Estimation.NPIV.InverseProblemSystem.θ₀ · Causalean/Estimation/NPIV/Setup.lean:147 · uses InverseProblemSystem
def phiVal reviewed
Causalean.Estimation.NPIV.InverseProblemSystem

The pointwise pseudo-outcome at a sample value w : 𝒲: φ_{h,q}(w) := m_e(w; h) + m(w; q) − q(zOf w) h(xOf w).

Definition (Lean source)
noncomputable def phiVal (S : InverseProblemSystem Ω μ) (h : S.𝒳 → ℝ) (q : S.𝒵 → ℝ) (w : S.𝒲) : ℝ := S.m_e w h + S.m w q - q (S.zOf w) * h (S.xOf w)
Causalean.Estimation.NPIV.InverseProblemSystem.phiVal · Causalean/Estimation/NPIV/Setup.lean:151 · uses InverseProblemSystem
def phi reviewed
Causalean.Estimation.NPIV.InverseProblemSystem

The pseudo-outcome along the random observation W: φ_{h,q}(ω) := φ_{h,q}(W ω).

Definition (Lean source)
noncomputable def phi (S : InverseProblemSystem Ω μ) (h : S.𝒳 → ℝ) (q : S.𝒵 → ℝ) (ω : Ω) : ℝ := S.phiVal h q (S.W ω)
Causalean.Estimation.NPIV.InverseProblemSystem.phi · Causalean/Estimation/NPIV/Setup.lean:160 · uses InverseProblemSystem
structure DualSolution reviewed
Causalean.Estimation.NPIV.InverseProblemSystem

DualSolution S q₀ says q₀ ∈ Qbar and q₀ solves the dual moment identity E[m_e(W; h)] = E[q₀(Z) h(X)] for every h ∈ Hbar (def:est-trae-dual-solution).

Definition (Lean source)
q₀ :
S.𝒵 → ℝ
Membership of `q₀` in the dual candidate set.
mem :
q₀ ∈ S.Qbar
The dual moment identity.
identity :
∀ h ∈ S.Hbar, ∫ ω, S.m_e (S.W ω) h ∂μ = ∫ ω, q₀ (S.zOf (S.W ω)) * h (S.xOf (S.W ω)) ∂μ
Causalean.Estimation.NPIV.InverseProblemSystem.DualSolution · Causalean/Estimation/NPIV/Setup.lean:166 · uses InverseProblemSystem
def Θ reviewed
Causalean.Estimation.NPIV.InverseProblemSystem

The doubly-robust functional from def:est-trae-dr-functional: Θ(h, q) := E[m_e(W; h) + m(W; q) − q(Z) h(X)] = E[φ_{h,q}(W)].

Definition (Lean source)
noncomputable def Θ (S : InverseProblemSystem Ω μ) (h : S.𝒳 → ℝ) (q : S.𝒵 → ℝ) : ℝ := ∫ ω, S.phi h q ω ∂μ
Causalean.Estimation.NPIV.InverseProblemSystem.Θ · Causalean/Estimation/NPIV/Setup.lean:179 · uses InverseProblemSystem
Mixed­Bias 1 core · 2 supporting This file proves the doubly robust mixed-bias identity for linear inverse-problem functionals. ★ mixed_bias_identity

Mixed-Bias Identity for NPIV

This file proves the doubly robust mixed-bias identity for linear inverse-problem functionals. It shows that, given a dual solution, the functional error factors as the expectation of the product of the dual nuisance error and the primal nuisance error, with immediate cancellation at either truth.

theorem mixed_bias_identity reviewed
Causalean.Estimation.NPIV

Mixed-bias / DR identityprop:est-trae-mixed-bias. Fix a linear inverse-problem functional system, and suppose q₀ solves the associated dual moment equation. Then for any primal candidate function h in the closed candidate set Hbar and any dual candidate function q in the closed candidate set Qbar, the bias of the doubly-robust functional Θ(h, q) relative to the true target θ₀ equals the expectation of the product (q₀(Z) − q(Z))·(h(X) − h₀(X)) of the dual and primal nuisance errors.

Formal statement
q₀ :
S.𝒵 → ℝ
hq₀ :
S.DualSolution q₀
h :
S.𝒳 → ℝ
hh :
h ∈ S.Hbar
q :
S.𝒵 → ℝ
hq :
q ∈ S.Qbar
S.Θ h q - S.θ₀ = ∫ ω, (q₀ (S.Z ω) - q (S.Z ω)) * (h (S.X ω) - S.h₀ (S.X ω)) ∂μ
Proof (Lean source)
theorem mixed_bias_identity (S : InverseProblemSystem Ω μ) {q₀ : S.𝒵 → ℝ} (hq₀ : S.DualSolution q₀) {h : S.𝒳 → ℝ} (hh : h ∈ S.Hbar) {q : S.𝒵 → ℝ} (hq : q ∈ S.Qbar) : S.Θ h q - S.θ₀ = ∫ ω, (q₀ (S.Z ω) - q (S.Z ω)) * (h (S.X ω) - S.h₀ (S.X ω)) ∂μ := by -- Integrability witnesses for each piece appearing in the proof. have h_int_meh := S.integrable_m_e h hh have h_int_meh₀ := S.integrable_m_e S.h₀ S.h₀_mem have h_int_mq := S.integrable_m q hq have h_int_qh := S.integrable_qh h hh q hq have h_int_q₀h := S.integrable_qh h hh q₀ hq₀.mem have h_int_q₀h₀ := S.integrable_qh S.h₀ S.h₀_mem q₀ hq₀.mem have h_int_qh₀ := S.integrable_qh S.h₀ S.h₀_mem q hq -- Reordered version (the primal moment uses `h₀(X) · q(Z)` not `q(Z) · h₀(X)`). have h_int_h₀q : Integrable (fun ω => S.h₀ (S.xOf (S.W ω)) * q (S.zOf (S.W ω))) μ := by have h := h_int_qh₀ simpa [mul_comm] using h -- Step 1: distribute the LHS over its three summands. have hΘ_eq : S.Θ h q = (∫ ω, S.m_e (S.W ω) h ∂μ) + (∫ ω, S.m (S.W ω) q ∂μ) - (∫ ω, q (S.zOf (S.W ω)) * h (S.xOf (S.W ω)) ∂μ) := by unfold InverseProblemSystem.Θ InverseProblemSystem.phi InverseProblemSystem.phiVal calc ∫ ω, S.m_e (S.W ω) h + S.m (S.W ω) q - q (S.zOf (S.W ω)) * h (S.xOf (S.W ω)) ∂μ = (∫ ω, S.m_e (S.W ω) h + S.m (S.W ω) q ∂μ) - ∫ ω, q (S.zOf (S.W ω)) * h (S.xOf (S.W ω)) ∂μ := integral_sub (h_int_meh.add h_int_mq) h_int_qh _ = (∫ ω, S.m_e (S.W ω) h ∂μ) + (∫ ω, S.m (S.W ω) q ∂μ) - ∫ ω, q (S.zOf (S.W ω)) * h (S.xOf (S.W ω)) ∂μ := by rw [integral_add h_int_meh h_int_mq] -- Step 2: dual identity applied at `h - h₀ ∈ Hbar` gives -- ∫ m_e(W; h) − ∫ m_e(W; h₀) -- = ∫ q₀(Z) · (h(X) − h₀(X)) ∂μ. have h_sub_mem : (fun x => h x - S.h₀ x) ∈ S.Hbar := S.Hbar_sub h hh S.h₀ S.h₀_mem have h_dual := hq₀.identity (fun x => h x - S.h₀ x) h_sub_mem have h_me_sub_pw : ∀ ω, S.m_e (S.W ω) (fun x => h x - S.h₀ x) = S.m_e (S.W ω) h - S.m_e (S.W ω) S.h₀ := fun ω => S.m_e_sub (S.W ω) h S.h₀ have h_me_diff_eq : (∫ ω, S.m_e (S.W ω) h ∂μ) - (∫ ω, S.m_e (S.W ω) S.h₀ ∂μ) = ∫ ω, q₀ (S.zOf (S.W ω)) * (h (S.xOf (S.W ω)) - S.h₀ (S.xOf (S.W ω))) ∂μ := by rw [← integral_sub h_int_meh h_int_meh₀] have hcongr : (fun ω => S.m_e (S.W ω) h - S.m_e (S.W ω) S.h₀) = (fun ω => S.m_e (S.W ω) (fun x => h x - S.h₀ x)) := by funext ω exact (h_me_sub_pw ω).symm rw [hcongr, h_dual] -- Step 3: primal moment identity at `q`: -- ∫ m(W; q) ∂μ = ∫ h₀(X) · q(Z) ∂μ. have h_prim := S.primal_moment q hq -- Step 4: rewrite the RHS as a sum of three atomic integrals. have hRHS_eq : (∫ ω, (q₀ (S.Z ω) - q (S.Z ω)) * (h (S.X ω) - S.h₀ (S.X ω)) ∂μ) = (∫ ω, q₀ (S.zOf (S.W ω)) * (h (S.xOf (S.W ω)) - S.h₀ (S.xOf (S.W ω))) ∂μ) + (∫ ω, S.h₀ (S.xOf (S.W ω)) * q (S.zOf (S.W ω)) ∂μ) - (∫ ω, q (S.zOf (S.W ω)) * h (S.xOf (S.W ω)) ∂μ) := by -- Pointwise expansion: (q₀ − q)·(h − h₀) = q₀·(h − h₀) + h₀·q − q·h. have h_int_diff_q₀ : Integrable (fun ω => q₀ (S.zOf (S.W ω)) * (h (S.xOf (S.W ω)) - S.h₀ (S.xOf (S.W ω)))) μ := by have : (fun ω => q₀ (S.zOf (S.W ω)) * (h (S.xOf (S.W ω)) - S.h₀ (S.xOf (S.W ω)))) = (fun ω => q₀ (S.zOf (S.W ω)) * h (S.xOf (S.W ω)) - q₀ (S.zOf (S.W ω)) * S.h₀ (S.xOf (S.W ω))) := by funext ω; ring rw [this]; exact h_int_q₀h.sub h_int_q₀h₀ -- ∫ (q₀ − q)·(h − h₀) = ∫ q₀·(h − h₀) + ∫ h₀·q − ∫ q·h have h_pw : ∀ ω, (q₀ (S.Z ω) - q (S.Z ω)) * (h (S.X ω) - S.h₀ (S.X ω)) = q₀ (S.zOf (S.W ω)) * (h (S.xOf (S.W ω)) - S.h₀ (S.xOf (S.W ω))) + S.h₀ (S.xOf (S.W ω)) * q (S.zOf (S.W ω)) - q (S.zOf (S.W ω)) * h (S.xOf (S.W ω)) := fun ω => by simp only [InverseProblemSystem.X, InverseProblemSystem.Z] ring have hcongr : (fun ω => (q₀ (S.Z ω) - q (S.Z ω)) * (h (S.X ω) - S.h₀ (S.X ω))) = (fun ω => q₀ (S.zOf (S.W ω)) * (h (S.xOf (S.W ω)) - S.h₀ (S.xOf (S.W ω))) + S.h₀ (S.xOf (S.W ω)) * q (S.zOf (S.W ω)) - q (S.zOf (S.W ω)) * h (S.xOf (S.W ω))) := by funext ω; exact h_pw ω rw [hcongr] calc ∫ ω, q₀ (S.zOf (S.W ω)) * (h (S.xOf (S.W ω)) - S.h₀ (S.xOf (S.W ω))) + S.h₀ (S.xOf (S.W ω)) * q (S.zOf (S.W ω)) - q (S.zOf (S.W ω)) * h (S.xOf (S.W ω)) ∂μ = (∫ ω, q₀ (S.zOf (S.W ω)) * (h (S.xOf (S.W ω)) - S.h₀ (S.xOf (S.W ω))) + S.h₀ (S.xOf (S.W ω)) * q (S.zOf (S.W ω)) ∂μ) - ∫ ω, q (S.zOf (S.W ω)) * h (S.xOf (S.W ω)) ∂μ := integral_sub (h_int_diff_q₀.add h_int_h₀q) h_int_qh _ = (∫ ω, q₀ (S.zOf (S.W ω)) * (h (S.xOf (S.W ω)) - S.h₀ (S.xOf (S.W ω))) ∂μ) + (∫ ω, S.h₀ (S.xOf (S.W ω)) * q (S.zOf (S.W ω)) ∂μ) - ∫ ω, q (S.zOf (S.W ω)) * h (S.xOf (S.W ω)) ∂μ := by rw [integral_add h_int_diff_q₀ h_int_h₀q] -- Step 5: combine. rw [hΘ_eq, hRHS_eq, ← h_me_diff_eq, ← h_prim] unfold InverseProblemSystem.θ₀ ring
2 supporting declarations (lemmas, instances)
  • Θ_h₀_eq_θ₀ theorem — DR cancellation at the truth on the primal side: Θ(h₀, q) = θ₀ for every q ∈ Qbar.
    q₀ :
    S.𝒵 → ℝ
    hq₀ :
    S.DualSolution q₀
    q :
    S.𝒵 → ℝ
    hq :
    q ∈ S.Qbar
    S.Θ S.h₀ q = S.θ₀
    Proof (Lean source)
    theorem Θ_h₀_eq_θ₀ (S : InverseProblemSystem Ω μ) {q₀ : S.𝒵 → ℝ} (hq₀ : S.DualSolution q₀) {q : S.𝒵 → ℝ} (hq : q ∈ S.Qbar) : S.Θ S.h₀ q = S.θ₀ := by have hmb := mixed_bias_identity S hq₀ S.h₀_mem hq apply sub_eq_zero.mp rw [hmb] simp
    Causalean.Estimation.NPIV.Θ_h₀_eq_θ₀ · Causalean/Estimation/NPIV/MixedBias.lean:171
  • Θ_q₀_eq_θ₀ theorem — DR cancellation at the truth on the dual side: Θ(h, q₀) = θ₀ for every h ∈ Hbar.
    q₀ :
    S.𝒵 → ℝ
    hq₀ :
    S.DualSolution q₀
    h :
    S.𝒳 → ℝ
    hh :
    h ∈ S.Hbar
    S.Θ h q₀ = S.θ₀
    Proof (Lean source)
    theorem Θ_q₀_eq_θ₀ (S : InverseProblemSystem Ω μ) {q₀ : S.𝒵 → ℝ} (hq₀ : S.DualSolution q₀) {h : S.𝒳 → ℝ} (hh : h ∈ S.Hbar) : S.Θ h q₀ = S.θ₀ := by have hmb := mixed_bias_identity S hq₀ hh hq₀.mem apply sub_eq_zero.mp rw [hmb] simp
    Causalean.Estimation.NPIV.Θ_q₀_eq_θ₀ · Causalean/Estimation/NPIV/MixedBias.lean:183
Source­Condition 2 core · 0 supporting This file records the β-source condition for the primal NPIV nuisance and the Tikhonov bias bounds assumed by the primal rate theorem.

Source Condition and Tikhonov Bias

This file records the β-source condition for the primal NPIV nuisance and the Tikhonov bias bounds assumed by the primal rate theorem. The source condition expresses the target nuisance as a spectral power of the normal operator applied to an admissible witness, while the bias bundle stores the strong and weak approximation inequalities later discharged by spectral calculus.

structure SourceCondition reviewed
Causalean.Estimation.NPIV

The β-source condition represents the true primal nuisance as the image, under a nonnegative spectral power of the normal NPIV operator, of an admissible witness function in the primal candidate class, via the spectral identity expressing the nuisance as that power of the operator applied to the witness, inside the space.

Definition (Lean source)
S :
β :
The source smoothness exponent is non-negative.
beta_nonneg :
0 ≤ β
The pre-image of `h₀` under the spectral lift `(T*T)^{β/2}`.
w₀_fun :
S.𝒳 → ℝ
`w₀` lies in the primal candidate set `Hbar`.
w₀_mem :
w₀_fun ∈ S.Hbar
Spectral identity: `(h₀)_{L²} = (T†T)^{β/2} (w₀)_{L²}` inside `Lp ℝ 2 μ`.
spectral_identity :
S.hL2 S.h₀_mem
= Complexification.realCFC S.Tstar_T (fun x : ℝ => rpow (max x 0) (β/2)) (S.hL2 w₀_mem)
Causalean.Estimation.NPIV.SourceCondition · Causalean/Estimation/NPIV/SourceCondition.lean:51 · uses OperatorSystem
structure TikhonovBiasBound reviewed
Causalean.Estimation.NPIV

The Tikhonov bias-bound bundle records, at a positive regularization level, the properties of the corresponding population Tikhonov solution: a nonnegative constant for which its squared distance to the true nuisance in the strong candidate-space norm and its squared distance to the true nuisance in the weak operator-image norm are each bounded by that constant times a power of the regularization level, together with the population strong-convexity inequality it satisfies relative to every other candidate.

Definition (Lean source)
S :
β lambda :
sc :
The Tikhonov regularization level is positive.
lambda_pos :
0 < lambda
The population Tikhonov solution at level `λ`.
h_lambda_star_fun :
S.𝒳 → ℝ
`h*_λ` lies in the primal candidate set `Hbar`.
h_lambda_star_mem :
h_lambda_star_fun ∈ S.Hbar
Constant absorbing the proof's `≲`.
C :
The constant is non-negative.
C_nonneg :
0 ≤ C
Strong-metric Tikhonov bias bound: `‖h*_λ − h₀‖²_{L²(P_X)} ≤ C · ‖w₀‖_{L²(P_X)} · λ^{min(β, 2)}`.
strong_bias :
S.strongNorm (S.hL2 h_lambda_star_mem - S.hL2 S.h₀_mem) ^ 2
≤ C * S.strongNorm (S.hL2 sc.w₀_mem) * lambda ^ (min β 2)
Weak-metric Tikhonov bias bound: `‖T(h*_λ − h₀)‖²_{L²(P_Z)} ≤ C · ‖w₀‖_{L²(P_X)} · λ^{min(β+1, 2)}`.
weak_bias :
S.weakNorm (S.hL2 h_lambda_star_mem - S.hL2 S.h₀_mem) ^ 2
≤ C * S.strongNorm (S.hL2 sc.w₀_mem) * lambda ^ (min (β + 1) 2)
**Population strong convexity** at `h*_λ` (proof sketch lines 287–304 of `doc/basic_concepts/po/estimation/trae_inverse_problems.tex`). For any candidate `ĥ ∈ Hbar`, λ‖ĥ − h*_λ‖²_{L²(P_X)} + ‖T(ĥ − h*_λ)‖²_{L²(P_Z)} ≤ ‖T(ĥ − h₀)‖²_{L²(P_Z)} − ‖T(h*_λ − h₀)‖²_{L²(P_Z)} + λ(‖ĥ‖²_{L²(P_X)} − ‖h*_λ‖²_{L²(P_X)}). Holds because `h*_λ` is the population minimizer of `F(h) := ‖T(h − h₀)‖² + λ‖h‖²` over the closed linear subspace `Hbar`, so the first-order term in the quadratic expansion vanishes. Spectral-calculus files discharge this inequality from the source condition and Tikhonov construction.
strong_convexity :
∀ h,
∀ hh : h ∈ S.Hbar,
lambda * (S.strongNorm (S.hL2 hh - S.hL2 h_lambda_star_mem)) ^ 2
+ (S.weakNorm (S.hL2 hh - S.hL2 h_lambda_star_mem)) ^ 2
≤ (S.weakNorm (S.hL2 hh - S.hL2 S.h₀_mem)) ^ 2
- (S.weakNorm (S.hL2 h_lambda_star_mem - S.hL2 S.h₀_mem)) ^ 2
+ lambda * ((S.strongNorm (S.hL2 hh)) ^ 2 - (S.strongNorm (S.hL2 h_lambda_star_mem)) ^ 2)