Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Support-controlled limits and Rényi bounds for faithful approximations

Definition
CRCD_Quantum_QuantumEntropy_SandwichedRenyiNonNeg_part_3

by JWang226 · Oct 8, 2026 · Mathlib c5ea003 (Lean v4.30.0)

channel-renyi-continuity-jwang226quantum-information

All operator spaces in the entropy statements are finite-dimensional, complex, and nonzero. For α>1\alpha>1α>1, ρ,σ≥0\rho,\sigma\ge0ρ,σ≥0 and ker⁡σ⊆ker⁡ρ\ker\sigma\subseteq\ker\rhokerσ⊆kerρ,

Qα(ρ+εI∥σ+εI)→Qα(ρ∥σ)(ε→0+).Q_\alpha(\rho+\varepsilon I\Vert\sigma+\varepsilon I)\to Q_\alpha(\rho\Vert\sigma) \quad(\varepsilon\to0^+).Qα​(ρ+εI∥σ+εI)→Qα​(ρ∥σ)(ε→0+).

If also ρ≠0\rho\ne0ρ=0, the analogous limit holds for the real trace-normalized divergence DαD_\alphaDα​ in nats. For α>0\alpha>0α>0 and positive semidefinite inputs,

Re⁡Qα(ρ∥σ)=0  ⟺  σ(1−α)/(2α)ρσ(1−α)/(2α)=0.\operatorname{Re}Q_\alpha(\rho\Vert\sigma)=0 \iff\sigma^{(1-\alpha)/(2\alpha)}\rho\sigma^{(1-\alpha)/(2\alpha)}=0.ReQα​(ρ∥σ)=0⟺σ(1−α)/(2α)ρσ(1−α)/(2α)=0.

In particular, its real part is nonzero for a positive-definite pair, and for α>1\alpha>1α>1 it is nonzero under support inclusion and ρ≠0\rho\ne0ρ=0. Additional facts give nonnegativity of the real quasi-entropy, trace continuity under εI\varepsilon IεI perturbations, and Dα(0∥τ)=0D_\alpha(0\Vert\tau)=0Dα​(0∥τ)=0 for every real α\alphaα and arbitrary τ\tauτ under the total-logarithm convention.

For a CPTP map Φ\PhiΦ and its faithful approximation FλF_\lambdaFλ​, 0<λ≤10<\lambda\le10<λ≤1, the part proves

Dα(Fλ(ρ)∥Fλ(σ))≤Dα(ρ∥σ).D_\alpha(F_\lambda(\rho)\Vert F_\lambda(\sigma))\le D_\alpha(\rho\Vert\sigma).Dα​(Fλ​(ρ)∥Fλ​(σ))≤Dα​(ρ∥σ).

For 1/2≤α<11/2\le\alpha<11/2≤α<1, the supplied premises are ρ,σ≥0\rho,\sigma\ge0ρ,σ≥0, ρ≠0\rho\ne0ρ=0, σ≠0\sigma\ne0σ=0 and Re⁡Qα(ρ∥σ)≠0\operatorname{Re}Q_\alpha(\rho\Vert\sigma)\ne0ReQα​(ρ∥σ)=0. For α>1\alpha>1α>1, they are ρ,σ≥0\rho,\sigma\ge0ρ,σ≥0, ρ≠0\rho\ne0ρ=0 and support inclusion. In the below-one regime a CPTP map preserves nonvanishing of Re⁡Qα\operatorname{Re}Q_\alphaReQα​ for positive inputs with ρ≠0\rho\ne0ρ=0 and nonvanishing input quasi-entropy. Finally, positive-definite perturbation paths ρ+εP,σ+εP\rho+\varepsilon P,\sigma+\varepsilon Pρ+εP,σ+εP give continuity of QαQ_\alphaQα​ and DαD_\alphaDα​ for α>0\alpha>0α>0 when ρ,σ,P>0\rho,\sigma,P>0ρ,σ,P>0; no normalization of their traces is required.

Definition code
import Mathlib.Algebra.Central.End
import Mathlib.Algebra.Star.UnitaryStarAlgAut
import Mathlib.Analysis.CStarAlgebra.CompletelyPositiveMap
import Mathlib.Analysis.CStarAlgebra.ContinuousFunctionalCalculus.Continuity
import Mathlib.Analysis.CStarAlgebra.ContinuousFunctionalCalculus.Order
import Mathlib.Analysis.CStarAlgebra.ContinuousFunctionalCalculus.Unique
import Mathlib.Analysis.CStarAlgebra.ContinuousLinearMap
import Mathlib.Analysis.CStarAlgebra.Matrix
import Mathlib.Analysis.CStarAlgebra.Unitary.Span
import Mathlib.Analysis.Complex.Order
import Mathlib.Analysis.Convex.Continuous
import Mathlib.Analysis.Convex.Integral
import Mathlib.Analysis.InnerProductSpace.Adjoint
import Mathlib.Analysis.InnerProductSpace.JointEigenspace
import Mathlib.Analysis.InnerProductSpace.PiL2
import Mathlib.Analysis.InnerProductSpace.Spectrum
import Mathlib.Analysis.InnerProductSpace.StarOrder
import Mathlib.Analysis.InnerProductSpace.TensorProduct
import Mathlib.Analysis.InnerProductSpace.Trace
import Mathlib.Analysis.Matrix.HermitianFunctionalCalculus
import Mathlib.Analysis.Matrix.Order
import Mathlib.Analysis.MeanInequalities
import Mathlib.Analysis.Normed.Lp.PiLp
import Mathlib.Analysis.SpecialFunctions.ContinuousFunctionalCalculus.Rpow.Basic
import Mathlib.Analysis.SpecialFunctions.ContinuousFunctionalCalculus.Rpow.Order
import Mathlib.LinearAlgebra.Dimension.Finite
import Mathlib.LinearAlgebra.Dimension.Free
import Mathlib.LinearAlgebra.Dimension.StrongRankCondition
import Mathlib.LinearAlgebra.Eigenspace.Minpoly
import Mathlib.LinearAlgebra.Lagrange
import Mathlib.LinearAlgebra.Matrix.PosDef
import Mathlib.LinearAlgebra.Matrix.ToLin
import Mathlib.LinearAlgebra.TensorProduct.Basis
import Mathlib.LinearAlgebra.Trace
import Mathlib.MeasureTheory.Group.Integral
import Mathlib.MeasureTheory.Measure.Haar.Basic
import Mathlib.RingTheory.Flat.Basic
import Mathlib.Topology.Algebra.Module.FiniteDimension
import Mathlib.Topology.Algebra.Module.LinearMapPiProd
import Mathlib.Topology.Algebra.Star.Unitary
import Definitions.Def_CRCD_Quantum_QuantumEntropy_HaarUnitary
import Definitions.Def_CRCD_Quantum_QuantumEntropy_SandwichedQuasiJensen_part_3
import Definitions.Def_CRCD_Quantum_QuantumEntropy_SandwichedRenyiNonNeg_part_2
import Definitions.Def_CRCD_Quantum_QuantumEntropy_SandwichedRenyiRelativeEntropy_part_3
import Definitions.Def_CRCD_Quantum_QuantumEntropy_TensorCFC
import Definitions.Def_CRCD_Quantum_QuantumEntropy_YoungInequality_part_2
import Definitions.Def_CRCD_Quantum_QuantumMechanics_QuantumChannel_part_4
import Definitions.Def_CRCD_Quantum_QuantumMechanics_QuantumState
import Definitions.Def_CRCD_Quantum_TraceInequality_BlockDiagonal
import Definitions.Def_CRCD_Quantum_TraceInequality_GeneralizedPerspectiveFunction
import Definitions.Def_CRCD_Quantum_TraceInequality_HilbertSchmidtOperatorSpace
import Definitions.Def_CRCD_Quantum_TraceInequality_JensenOperatorInequality
import Definitions.Def_CRCD_Quantum_TraceInequality_JensenOperatorInequalityIImpIV_part_2
import Definitions.Def_CRCD_Quantum_TraceInequality_LiebAndoTrace_part_2
import Definitions.Def_CRCD_Quantum_TraceInequality_LownerHeinzCore_part_3
import Definitions.Def_CRCD_Quantum_TraceInequality_LownerHeinzTheorem
import Definitions.Def_CRCD_Quantum_TraceInequality_OperatorGeometricMean

/-
Copyright (c) 2025-2026. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
-/





/-!
# Sandwiched Rényi divergence on non-negative operators (Frank–Lieb extension)

This file extends the sandwiched Rényi divergence `D_α(ρ ‖ σ)` from the
positive-definite case to **non-negative** `ρ, σ ≥ 0`, following the convention
of Frank–Lieb (arXiv:1306.5358v3, §I.A).

## Main definitions

* `suppLE ρ σ` — support condition `ker σ ≤ ker ρ` (equivalent to `supp ρ ⊂ supp σ`).
* `sandwichedRenyiDivNN α ρ σ` — Frank–Lieb extension on `EReal`: equals
  `sandwichedRenyiDiv α ρ σ` when `α < 1` or `suppLE ρ σ`, and `⊤ : EReal`
  when `α > 1` and `¬ suppLE ρ σ`.

## Main theorem

* `sandwichedRenyiDivNN_monotone` — **Theorem 1** (Frank–Lieb): for any CPTP map
  `E : CPTP ℋ ℋ`, any `α ∈ [1/2, 1) ∪ (1, ∞)`, and any non-negative `ρ, σ`,
  `D_α^{NN}(E ρ ‖ E σ) ≤ D_α^{NN}(ρ ‖ σ)`.

## Proof structure

The main theorem reduces via case analysis:

* `α > 1`, `¬ suppLE ρ σ`: RHS is `⊤`, trivial.
* otherwise: real-valued inequality, proven via the faithful-approximation
  `F_λ := (1−λ) E + λ · depolarizing` (faithful for `λ > 0`), the perturbed
  Theorem 1 for faithful channels (`sandwichedRenyiDiv_monotone_nonneg_perturbed`),
  and boundary continuity of `sandwichedRenyiDiv` as `ε → 0+`.
-/

namespace SandwichedRenyiRelativeEntropy

open QuantumState QuantumChannel MeasureTheory TensorProduct
open scoped ComplexOrder NNReal Topology

universe u

set_option linter.style.longLine false

/-! ### Support condition `suppLE` -/





/-! ### Frank–Lieb explicit formula -/



/-! ### Auxiliary spectral / order lemmas -/









/-! ### Support preservation under positive maps -/



/-! ### Outer-product helpers and the depolarizing-channel Kraus decomposition -/











/-! ### Faithful approximating channel `F_λ := (1−λ) E + λ · depolarizing` -/

















/-! ### Continuity of `CFC.rpow` and `sandwichedQuasi` on the full non-negative cone

For a **non-negative exponent** `p ≥ 0` the map `x ↦ x^p` is continuous on all of
`ℝ≥0` (no pseudo-inverse discontinuity at `0`), so `A ↦ CFC.rpow A p` is continuous
on the whole non-negative cone `{A | 0 ≤ A}`, not just on the strictly-positive
`pdSetLM`. This is the analytic engine for the `α < 1` boundary continuity, where the
exponents `β = (1−α)/(2α) > 0` and `α > 0` are both non-negative. -/















/-! ### Boundary continuity of `sandwichedRenyiDiv` along the perturbation path -/

/-! #### Eigenvector formulas for the continuous functional calculus -/



















/-- **Boundary-continuity of `sandwichedQuasi` (α > 1) under `suppLE`.** Under the support
    condition, `Q_α(ρ+εI ‖ σ+εI) → Q_α(ρ ‖ σ)` as `ε → 0⁺`. The pseudo-inverse blow-up of
    `(σ+εI)^β` (β < 0) on `ker σ` is killed by `suppLE` (`rpow_conj_tendsto_of_suppLE`); the
    outer `Tr(·^α)` is continuous on the non-negative cone (`α > 0`). -/
 lemma sandwichedQuasi_tendsto_of_suppLE
    {ℋ : Type u} [Qudit ℋ] [Nontrivial ℋ]
    {α : ℝ} (hα_gt : 1 < α)
    {ρ σ : L ℋ} (hρ : 0 ≤ ρ) (hσ : 0 ≤ σ) (hsupp : suppLE ρ σ) :
    Filter.Tendsto
      (fun ε : ℝ => sandwichedQuasi α (ρ + (ε : ℂ) • (1 : L ℋ)) (σ + (ε : ℂ) • (1 : L ℋ)))
      (nhdsWithin 0 (Set.Ioi 0))
      (nhds (sandwichedQuasi α ρ σ)) := by
  have hαpos : 0 < α := by linarith
  set β : ℝ := (1-α)/(2*α) with hβ
  have hM := _root_.SandwichedRenyiRelativeEntropy.rpow_conj_tendsto_of_suppLE hα_gt hρ hσ hsupp
  have hMnn : ∀ A B : L ℋ, 0 ≤ B → (0:L ℋ) ≤ CFC.rpow A β * B * CFC.rpow A β :=
    fun A B hB => conjugate_nonneg_of_nonneg hB CFC.rpow_nonneg
  have hcont : ContinuousWithinAt (fun X : L ℋ => Tr (CFC.rpow X α)) {X : L ℋ | 0 ≤ X}
      (CFC.rpow σ β * ρ * CFC.rpow σ β) := by
    have h1 : ContinuousOn (fun X : L ℋ => CFC.rpow X α) {X : L ℋ | 0 ≤ X} :=
      _root_.QCProve2mePrivate.Quantum_QuantumEntropy_SandwichedRenyiNonNeg_rpow_continuousOn_nonneg (le_of_lt hαpos)
    have h2 : Continuous (fun A : L ℋ => Tr A) := LinearMap.continuous_of_finiteDimensional _
    exact (h2.comp_continuousOn h1).continuousWithinAt (hMnn σ ρ hρ)
  have hMwithin : Filter.Tendsto
      (fun ε : ℝ => CFC.rpow (σ+(ε:ℂ)•(1:L ℋ)) β * (ρ+(ε:ℂ)•(1:L ℋ)) * CFC.rpow (σ+(ε:ℂ)•(1:L ℋ)) β)
      (nhdsWithin 0 (Set.Ioi 0))
      (nhdsWithin (CFC.rpow σ β * ρ * CFC.rpow σ β) {X : L ℋ | 0 ≤ X}) := by
    rw [tendsto_nhdsWithin_iff]
    refine ⟨hM, ?_⟩
    filter_upwards [self_mem_nhdsWithin] with ε hε
    have hε1 : (0:L ℋ) ≤ (ε:ℂ)•(1:L ℋ) :=
      smul_nonneg (Complex.zero_le_real.mpr (le_of_lt hε)) zero_le_one
    exact hMnn _ _ (add_nonneg hρ hε1)
  have hcomp := Filter.Tendsto.comp hcont hMwithin
  have heqQ : ∀ ρ' σ' : L ℋ,
      sandwichedQuasi α ρ' σ' = Tr (CFC.rpow (CFC.rpow σ' β * ρ' * CFC.rpow σ' β) α) := by
    intro ρ' σ'; rfl
  simp_rw [heqQ]
  exact hcomp

/-- **Orthogonality core**: for `0 < α` and `ρ, σ ≥ 0`, the (real part of the)
    quasi-entropy vanishes iff the conjugated operator does:
    `Q_α(ρ‖σ).re = 0 ↔ σ^β ρ σ^β = 0` (where `β = (1−α)/(2α)`). The latter is the
    operator form of orthogonality `ρ ⊥ σ`. -/
 lemma sandwichedQuasi_re_eq_zero_iff
    {ℋ : Type u} [Qudit ℋ] [Nontrivial ℋ]
    {α : ℝ} (hα0 : 0 < α) {ρ σ : L ℋ} (hρ : 0 ≤ ρ) (hσ : 0 ≤ σ) :
    (sandwichedQuasi α ρ σ).re = 0 ↔
      CFC.rpow σ ((1 - α) / (2 * α)) * ρ * CFC.rpow σ ((1 - α) / (2 * α)) = 0 := by
  set β : ℝ := (1 - α) / (2 * α) with hβ
  set M : L ℋ := CFC.rpow σ β * ρ * CFC.rpow σ β with hM
  have hM_nn : (0 : L ℋ) ≤ M := _root_.SandwichedRenyiRelativeEntropy.rpow_conj_nonneg β hρ hσ
  have hMα_nn : (0 : L ℋ) ≤ CFC.rpow M α := CFC.rpow_nonneg
  have hQ_eq : sandwichedQuasi α ρ σ = Tr (CFC.rpow M α) := by
    unfold sandwichedQuasi; rw [← hβ, ← hM]
  rw [hQ_eq]
  constructor
  · intro htr
    have hMα0 : CFC.rpow M α = 0 := by
      by_contra hne
      exact absurd htr (ne_of_gt (trace_re_pos_of_ne_zero hMα_nn hne))
    have hαinv : α * (1 / α) = 1 := by rw [mul_one_div, div_self (ne_of_gt hα0)]
    have hcomp : CFC.rpow (CFC.rpow M α) (1 / α) = M := by
      have key := CFC.rpow_rpow_of_exponent_nonneg M α (1 / α) hα0.le (by positivity) hM_nn
      rw [hαinv, CFC.rpow_one M hM_nn] at key
      exact key
    rw [hMα0, CFC.zero_rpow (one_div_ne_zero (ne_of_gt hα0))] at hcomp
    exact hcomp.symm
  · intro hM0
    rw [hM0, CFC.zero_rpow (ne_of_gt hα0), map_zero, Complex.zero_re]

/-- **Support-overlap ⟹ `Q_α ≠ 0` (α > 1).** Under `suppLE ρ σ` with `ρ ≠ 0`, the
    quasi-entropy is non-zero. Indeed `Q_α.re = 0 ⟺ σ^β ρ σ^β = 0`; expanding in the
    eigenbasis of `σ`, this forces `⟨b_k, ρ b_i⟩ = 0` for all `i, k` (using `suppLE` to
    kill the `ker σ` directions and positivity of `eigᵢ^β` elsewhere), hence `ρ = 0`,
    contradicting `ρ ≠ 0`. -/
 lemma sandwichedQuasi_re_ne_zero_of_suppLE
    {ℋ : Type u} [Qudit ℋ] [Nontrivial ℋ]
    {α : ℝ} (hα_gt : 1 < α) {ρ σ : L ℋ}
    (hρ : 0 ≤ ρ) (hσ : 0 ≤ σ) (hsupp : suppLE ρ σ) (hρ0 : ρ ≠ 0) :
    (sandwichedQuasi α ρ σ).re ≠ 0 := by
  intro hzre
  have hαpos : 0 < α := by linarith
  set β : ℝ := (1-α)/(2*α) with hβ
  rw [_root_.SandwichedRenyiRelativeEntropy.sandwichedQuasi_re_eq_zero_iff hαpos hρ hσ] at hzre
  apply hρ0
  set n := Module.finrank ℂ ℋ with hn_def
  have hn : Module.finrank ℂ ℋ = n := rfl
  have hσ_pos : σ.IsPositive := (LinearMap.nonneg_iff_isPositive σ).mp hσ
  have hσ_sym : σ.IsSymmetric := hσ_pos.isSymmetric
  have hρ_sym : ρ.IsSymmetric := ((LinearMap.nonneg_iff_isPositive ρ).mp hρ).isSymmetric
  set b := hσ_sym.eigenvectorBasis hn with hb
  set eig := hσ_sym.eigenvalues hn with heig
  have h_eig_nn : ∀ i, 0 ≤ eig i := fun i => hσ_pos.nonneg_eigenvalues hn i
  have h_eig_apply : ∀ i, σ (b i) = ((eig i : ℝ):ℂ) • b i := hσ_sym.apply_eigenvectorBasis hn
  have hb_ne : ∀ i, b i ≠ 0 := fun i => b.orthonormal.ne_zero i
  have h_ker : ∀ k, eig k = 0 → ρ (b k) = 0 :=
    fun k hk => hsupp (LinearMap.mem_ker.mpr (by rw [h_eig_apply k, hk]; simp))
  have h_supp_zero : ∀ i j, (eig i = 0 ∨ eig j = 0) → inner ℂ (b j) (ρ (b i)) = (0:ℂ) := by
    intro i j hij
    rcases hij with hi | hj
    · rw [h_ker i hi]; simp
    · rw [show inner ℂ (b j) (ρ (b i)) = inner ℂ (ρ (b j)) (b i) from (hρ_sym (b j) (b i)).symm,
          h_ker j hj]; simp
  have hspec0 : ∀ i, eig i ∈ spectrum ℝ σ :=
    fun i => mem_spectrum_real_of_eigenvector (hb_ne i) (h_eig_apply i)
  have hS0b : ∀ i, CFC.rpow σ β (b i) = (((eig i)^β : ℝ):ℂ) • b i :=
    fun i => _root_.SandwichedRenyiRelativeEntropy.rpow_apply_eigenvector hσ β (h_eig_apply i) (hspec0 i)
  have hSsym : (CFC.rpow σ β).IsSymmetric :=
    ((LinearMap.nonneg_iff_isPositive _).mp CFC.rpow_nonneg).isSymmetric
  have hcoeff : ∀ i k, inner ℂ (b k) ((CFC.rpow σ β * ρ * CFC.rpow σ β) (b i))
      = (((eig i)^β : ℝ):ℂ) * (((eig k)^β : ℝ):ℂ) * inner ℂ (b k) (ρ (b i)) := by
    intro i k
    rw [Module.End.mul_apply, Module.End.mul_apply, hS0b i, map_smul, map_smul, inner_smul_right,
        show inner ℂ (b k) (CFC.rpow σ β (ρ (b i)))
          = inner ℂ (CFC.rpow σ β (b k)) (ρ (b i)) from (hSsym (b k) (ρ (b i))).symm,
        hS0b k, inner_smul_left, Complex.conj_ofReal]
    ring
  have hzero_all : ∀ i k, inner ℂ (b k) (ρ (b i)) = (0:ℂ) := by
    intro i k
    by_cases h : eig i = 0 ∨ eig k = 0
    · exact h_supp_zero i k h
    · push_neg at h
      have hi' : 0 < eig i := lt_of_le_of_ne (h_eig_nn i) (Ne.symm h.1)
      have hk' : 0 < eig k := lt_of_le_of_ne (h_eig_nn k) (Ne.symm h.2)
      have hc := hcoeff i k
      rw [hzre, LinearMap.zero_apply, inner_zero_right] at hc
      have hne : (((eig i)^β : ℝ):ℂ) * (((eig k)^β : ℝ):ℂ) ≠ 0 := by
        rw [← Complex.ofReal_mul]
        exact Complex.ofReal_ne_zero.mpr (ne_of_gt (mul_pos (Real.rpow_pos_of_pos hi' β)
          (Real.rpow_pos_of_pos hk' β)))
      exact (mul_eq_zero.mp hc.symm).resolve_left hne
  refine LinearMap.ext fun x => ?_
  rw [LinearMap.zero_apply]
  conv_lhs => rw [← b.sum_repr' x]
  rw [map_sum]
  refine Finset.sum_eq_zero fun i _ => ?_
  rw [LinearMap.map_smul,
      show ρ (b i) = ∑ k, inner ℂ (b k) (ρ (b i)) • b k from (b.sum_repr' (ρ (b i))).symm]
  rw [show (∑ k, inner ℂ (b k) (ρ (b i)) • b k) = 0 from
      Finset.sum_eq_zero fun k _ => by rw [hzero_all i k, zero_smul]]
  rw [smul_zero]

/-- Continuity of `Tr` on perturbed states: `Tr(ρ + εI) → Tr ρ` as `ε → 0+`. -/
 lemma trace_tendsto_perturbed
    {ℋ : Type u} [Qudit ℋ] [Nontrivial ℋ]
    (ρ : L ℋ) :
    Filter.Tendsto
      (fun ε : ℝ => (Tr (ρ + (ε : ℂ) • (1 : L ℋ)) : ℂ))
      (nhdsWithin 0 (Set.Ioi 0))
      (nhds (Tr ρ)) := by
  have h_eq : ∀ ε : ℝ,
      (Tr (ρ + (ε : ℂ) • (1 : L ℋ)) : ℂ) =
      Tr ρ + (ε : ℂ) * (Module.finrank ℂ ℋ : ℂ) := by
    intro ε
    rw [map_add, LinearMap.map_smul, LinearMap.trace_one, smul_eq_mul]
  simp_rw [h_eq]
  have h_real : Filter.Tendsto
      (fun ε : ℝ => Tr ρ + (ε : ℂ) * (Module.finrank ℂ ℋ : ℂ))
      (nhds 0) (nhds (Tr ρ + (0 : ℂ) * (Module.finrank ℂ ℋ : ℂ))) := by
    refine Filter.Tendsto.add tendsto_const_nhds ?_
    refine Filter.Tendsto.mul ?_ tendsto_const_nhds
    exact (Complex.continuous_ofReal.tendsto 0)
  simpa using h_real.mono_left nhdsWithin_le_nhds

/-- **Boundary-continuity** of `sandwichedRenyiDiv` for non-negative `ρ, σ` with `suppLE ρ σ`,
    `α > 1`, and `ρ ≠ 0`.

    Derived from `sandwichedQuasi_tendsto_of_suppLE` (continuity of the quasi-relative
    entropy under `suppLE`), continuity of `Tr`, and continuity of `log`. Under `ρ ≠ 0` the
    denominator `(Tr ρ).re > 0` (`trace_re_pos_of_ne_zero`) and, under `suppLE`, the numerator
    `Q_α(ρ‖σ).re ≠ 0` (`sandwichedQuasi_re_ne_zero_of_suppLE`), so the limiting quotient is a
    continuity point of `log`. -/
 lemma sandwichedRenyiDiv_tendsto_of_suppLE
    {ℋ : Type u} [Qudit ℋ] [Nontrivial ℋ]
    {α : ℝ} (hα_gt : 1 < α)
    {ρ σ : L ℋ} (hρ : 0 ≤ ρ) (hσ : 0 ≤ σ) (hsupp : suppLE ρ σ) (hρ0 : ρ ≠ 0) :
    Filter.Tendsto
      (fun ε : ℝ => sandwichedRenyiDiv α (ρ + (ε : ℂ) • (1 : L ℋ)) (σ + (ε : ℂ) • (1 : L ℋ)))
      (nhdsWithin 0 (Set.Ioi 0))
      (nhds (sandwichedRenyiDiv α ρ σ)) := by
  unfold sandwichedRenyiDiv
  have h_tr : (Tr ρ).re ≠ 0 := ne_of_gt (trace_re_pos_of_ne_zero hρ hρ0)
  have h_Q : (sandwichedQuasi α ρ σ).re ≠ 0 :=
    _root_.SandwichedRenyiRelativeEntropy.sandwichedQuasi_re_ne_zero_of_suppLE hα_gt hρ hσ hsupp hρ0
  -- (1) `Q_α(ρ+εI ‖ σ+εI) → Q_α(ρ ‖ σ)` as ε → 0+ (deep step under `suppLE`).
  have hQ : Filter.Tendsto
      (fun ε : ℝ =>
        (sandwichedQuasi α (ρ + (ε : ℂ) • (1 : L ℋ))
          (σ + (ε : ℂ) • (1 : L ℋ))).re)
      (nhdsWithin 0 (Set.Ioi 0))
      (nhds (sandwichedQuasi α ρ σ).re) := by
    have hQ_C := _root_.SandwichedRenyiRelativeEntropy.sandwichedQuasi_tendsto_of_suppLE hα_gt hρ hσ hsupp
    exact (Complex.continuous_re.tendsto _).comp hQ_C
  -- (2) `Tr(ρ+εI) → Tr ρ` (immediate, linear).
  have hT : Filter.Tendsto
      (fun ε : ℝ => (Tr (ρ + (ε : ℂ) • (1 : L ℋ))).re)
      (nhdsWithin 0 (Set.Ioi 0))
      (nhds (Tr ρ).re) :=
    (Complex.continuous_re.tendsto _).comp (_root_.SandwichedRenyiRelativeEntropy.trace_tendsto_perturbed (ℋ := ℋ) ρ)
  -- (3)+(4) Quotient and `log`: continuous since both limits are nonzero.
  have hLog : Filter.Tendsto
      (fun ε : ℝ =>
        Real.log
          ((sandwichedQuasi α (ρ + (ε : ℂ) • (1 : L ℋ))
              (σ + (ε : ℂ) • (1 : L ℋ))).re /
            (Tr (ρ + (ε : ℂ) • (1 : L ℋ))).re))
      (nhdsWithin 0 (Set.Ioi 0))
      (nhds (Real.log ((sandwichedQuasi α ρ σ).re / (Tr ρ).re))) := by
    have hRatio : Filter.Tendsto
        (fun ε : ℝ =>
          (sandwichedQuasi α (ρ + (ε : ℂ) • (1 : L ℋ))
              (σ + (ε : ℂ) • (1 : L ℋ))).re /
            (Tr (ρ + (ε : ℂ) • (1 : L ℋ))).re)
        (nhdsWithin 0 (Set.Ioi 0))
        (nhds ((sandwichedQuasi α ρ σ).re / (Tr ρ).re)) :=
      Filter.Tendsto.div hQ hT h_tr
    have h_div_ne : (sandwichedQuasi α ρ σ).re / (Tr ρ).re ≠ 0 :=
      div_ne_zero h_Q h_tr
    exact (Real.continuousAt_log h_div_ne).tendsto.comp hRatio
  exact hLog.const_mul _

/-! ### Helpers for the `α < 1` main-theorem case -/

/-- `sandwichedRenyiDiv α 0 τ = 0`: with `ρ = 0`, `Tr ρ = 0` and `Real.log (· / 0) = 0`. -/
 lemma sandwichedRenyiDiv_zero_left
    {ℋ : Type u} [Qudit ℋ] [Nontrivial ℋ] (α : ℝ) (τ : L ℋ) :
    sandwichedRenyiDiv α (0 : L ℋ) τ = 0 := by
  unfold sandwichedRenyiDiv
  have h0 : (Tr (0 : L ℋ)).re = 0 := by simp
  rw [h0, div_zero, Real.log_zero, mul_zero]

/-- For positive-definite `ρ, σ` (in `pdSetLM`), the quasi-entropy is non-zero: `σ^β ρ σ^β`
    is a product of units (hence a unit, hence `≠ 0`), so by `sandwichedQuasi_re_eq_zero_iff`
    its `Q_α.re ≠ 0`. -/
lemma sandwichedQuasi_re_ne_zero_of_pdSetLM
    {ℋ : Type u} [Qudit ℋ] [Nontrivial ℋ]
    {α : ℝ} (hα0 : 0 < α) {ρ σ : L ℋ}
    (hρ : ρ ∈ pdSetLM (ℋ := ℋ)) (hσ : σ ∈ pdSetLM (ℋ := ℋ)) :
    (sandwichedQuasi α ρ σ).re ≠ 0 := by
  have hrpow_unit : IsUnit (CFC.rpow σ ((1 - α) / (2 * α))) :=
    isUnit_of_pdSetLM (pdSetLM_rpow_ne hσ)
  have hρ_unit : IsUnit ρ := isUnit_of_pdSetLM hρ
  intro hzero
  rw [_root_.SandwichedRenyiRelativeEntropy.sandwichedQuasi_re_eq_zero_iff hα0 (nonneg_of_pdSetLM hρ) (nonneg_of_pdSetLM hσ)] at hzero
  exact (((hrpow_unit.mul hρ_unit).mul hrpow_unit).ne_zero) hzero

/-- The real part of the quasi-entropy is non-negative (it is `Tr` of a non-negative operator). -/
 lemma sandwichedQuasi_re_nonneg
    {ℋ : Type u} [Qudit ℋ] [Nontrivial ℋ]
    (α : ℝ) (ρ σ : L ℋ) :
    0 ≤ (sandwichedQuasi α ρ σ).re := by
  have hpos : (0 : L ℋ) ≤ CFC.rpow
      (CFC.rpow σ ((1 - α) / (2 * α)) * ρ * CFC.rpow σ ((1 - α) / (2 * α))) α := CFC.rpow_nonneg
  have h := ((LinearMap.nonneg_iff_isPositive _).mp hpos).trace_nonneg
  rw [Complex.le_def] at h
  exact h.1

/-- **Faithful-approximation DPI** (`α < 1`): for each `λ ∈ (0,1]`, the faithful channel
    `F_λ` satisfies `D_α(F_λρ ‖ F_λσ) ≤ D_α(ρ‖σ)`. Obtained by taking `ε → 0⁺` in the
    perturbed PD inequality (`F_λρ, F_λσ` are positive-definite). -/
 lemma sandwichedRenyiDiv_faithfulApprox_le
    {ℋ 𝒦 : Type u} [Qudit ℋ] [Nontrivial ℋ] [Qudit 𝒦] [Nontrivial 𝒦]
    (E : CPTP ℋ 𝒦) {α : ℝ} (hα_ge : (1 : ℝ) / 2 ≤ α) (hα_lt : α < 1)
    {ρ σ : L ℋ} (hρ : 0 ≤ ρ) (hσ : 0 ≤ σ) (hρ0 : ρ ≠ 0) (hσ0 : σ ≠ 0)
    (hQρσ : (sandwichedQuasi α ρ σ).re ≠ 0)
    {lam : ℝ} (hlam0 : 0 < lam) (hlam1 : lam ≤ 1) :
    sandwichedRenyiDiv α ((faithfulApprox E lam hlam0.le hlam1).toFun ρ)
        ((faithfulApprox E lam hlam0.le hlam1).toFun σ) ≤ sandwichedRenyiDiv α ρ σ := by
  have hα0 : 0 < α := by linarith
  have hTrρ : (Tr ρ).re ≠ 0 := ne_of_gt (trace_re_pos_of_ne_zero hρ hρ0)
  set F := faithfulApprox E lam hlam0.le hlam1 with hF
  have hF1 : F.toFun 1 ∈ pdSetLM (ℋ := 𝒦) := faithfulApprox_one_pdSetLM E hlam0 hlam1
  have hFρ_pd : F.toFun ρ ∈ pdSetLM (ℋ := 𝒦) := faithfulApprox_pdSetLM E hlam0 hlam1 hρ hρ0
  have hFσ_pd : F.toFun σ ∈ pdSetLM (ℋ := 𝒦) := faithfulApprox_pdSetLM E hlam0 hlam1 hσ hσ0
  have hFρ_nn : (0 : L 𝒦) ≤ F.toFun ρ := nonneg_of_pdSetLM hFρ_pd
  have hFσ_nn : (0 : L 𝒦) ≤ F.toFun σ := nonneg_of_pdSetLM hFσ_pd
  have hF1_nn : (0 : L 𝒦) ≤ F.toFun 1 := nonneg_of_pdSetLM hF1
  have hQF : (sandwichedQuasi α (F.toFun ρ) (F.toFun σ)).re ≠ 0 :=
    sandwichedQuasi_re_ne_zero_of_pdSetLM hα0 hFρ_pd hFσ_pd
  have hTrFρ : (Tr (F.toFun ρ)).re ≠ 0 :=
    ne_of_gt (trace_re_pos_of_ne_zero hFρ_nn (isUnit_of_pdSetLM hFρ_pd).ne_zero)
  have h_pert : ∀ ε : ℝ, 0 < ε →
      sandwichedRenyiDiv α (F.toFun ρ + (ε : ℂ) • F.toFun 1) (F.toFun σ + (ε : ℂ) • F.toFun 1)
        ≤ sandwichedRenyiDiv α (ρ + (ε : ℂ) • (1 : L ℋ)) (σ + (ε : ℂ) • (1 : L ℋ)) := by
    intro ε hε
    have hpd := sandwichedRenyiDiv_monotone_nonneg_perturbed F hα_ge (ne_of_lt hα_lt) hρ hσ hF1 hε
    have hsmul := LinearMap.map_smul F.toCompletelyPositiveMap.toLinearMap (ε : ℂ) (1 : L ℋ)
    have eρ : F.toFun (ρ + (ε : ℂ) • (1 : L ℋ)) = F.toFun ρ + (ε : ℂ) • F.toFun (1 : L ℋ) := by
      have h1 := LinearMap.map_add F.toCompletelyPositiveMap.toLinearMap ρ ((ε : ℂ) • (1 : L ℋ))
      rw [hsmul] at h1; exact h1
    have eσ : F.toFun (σ + (ε : ℂ) • (1 : L ℋ)) = F.toFun σ + (ε : ℂ) • F.toFun (1 : L ℋ) := by
      have h1 := LinearMap.map_add F.toCompletelyPositiveMap.toLinearMap σ ((ε : ℂ) • (1 : L ℋ))
      rw [hsmul] at h1; exact h1
    rw [eρ, eσ] at hpd
    exact hpd
  have h_RHS := _root_.SandwichedRenyiRelativeEntropy.sandwichedRenyiDiv_tendsto_nonneg_lt hα0 hα_lt hρ hσ zero_le_one hQρσ hTrρ
  have h_LHS := _root_.SandwichedRenyiRelativeEntropy.sandwichedRenyiDiv_tendsto_nonneg_lt hα0 hα_lt hFρ_nn hFσ_nn hF1_nn hQF hTrFρ
  haveI : (nhdsWithin (0 : ℝ) (Set.Ioi 0)).NeBot := nhdsWithin_Ioi_neBot (le_refl 0)
  exact le_of_tendsto_of_tendsto h_LHS h_RHS
    (Filter.eventually_of_mem self_mem_nhdsWithin h_pert)

/-- **Orthogonality reflection** (`α < 1`): a CPTP map cannot create orthogonality.
    If `ρ ≠ 0` and `Q_α(ρ‖σ) ≠ 0` (i.e. `ρ`, `σ` are not orthogonal), then
    `Q_α(Eρ‖Eσ) ≠ 0`.

    **Proof** (no Stinespring needed): the faithful DPI gives, for each `λ ∈ (0,1]`,
    `D_α(F_λρ ‖ F_λσ) ≤ D_α(ρ‖σ)`; since `α < 1` (so `1/(α−1) < 0`) and `Tr(F_λρ) = Tr ρ`,
    this is equivalent to `Q_α(ρ‖σ) ≤ Q_α(F_λρ ‖ F_λσ)`. As `λ → 0⁺`, `Q_α(F_λρ‖F_λσ) →
    Q_α(Eρ‖Eσ)` by nonneg-cone continuity, so `0 < Q_α(ρ‖σ) ≤ Q_α(Eρ‖Eσ)`. -/
 lemma sandwichedQuasi_re_ne_zero_of_CPTP
    {ℋ 𝒦 : Type u} [Qudit ℋ] [Nontrivial ℋ] [Qudit 𝒦] [Nontrivial 𝒦]
    (E : CPTP ℋ 𝒦) {α : ℝ} (hα_ge : (1 : ℝ) / 2 ≤ α) (hα_lt : α < 1)
    {ρ σ : L ℋ} (hρ : 0 ≤ ρ) (hσ : 0 ≤ σ) (hρ0 : ρ ≠ 0)
    (hQ : (sandwichedQuasi α ρ σ).re ≠ 0) :
    (sandwichedQuasi α (E.toFun ρ) (E.toFun σ)).re ≠ 0 := by
  have hα0 : 0 < α := by linarith
  have hαm : α - 1 < 0 := by linarith
  have hTrρ_pos : 0 < (Tr ρ).re := trace_re_pos_of_ne_zero hρ hρ0
  have hb_pos : 0 < (sandwichedQuasi α ρ σ).re :=
    lt_of_le_of_ne (_root_.SandwichedRenyiRelativeEntropy.sandwichedQuasi_re_nonneg α ρ σ) (Ne.symm hQ)
  have hβne : (1 - α) / (2 * α) ≠ 0 := div_ne_zero (by linarith) (by positivity)
  have hσ0 : σ ≠ 0 := by
    intro h
    apply hQ
    rw [_root_.SandwichedRenyiRelativeEntropy.sandwichedQuasi_re_eq_zero_iff hα0 hρ hσ, h, CFC.zero_rpow hβne, zero_mul, mul_zero]
  -- `Q_α(ρ‖σ) ≤ Q_α(F_λρ ‖ F_λσ)` for each `λ`.
  have hQ_ge : ∀ (l : {l : ℝ // 0 < l ∧ l ≤ 1}),
      (sandwichedQuasi α ρ σ).re ≤
        (sandwichedQuasi α ((faithfulApprox E l.val l.property.1.le l.property.2).toFun ρ)
          ((faithfulApprox E l.val l.property.1.le l.property.2).toFun σ)).re := by
    rintro ⟨lam, hlam0, hlam1⟩
    simp only
    set F := faithfulApprox E lam hlam0.le hlam1 with hF
    have hFρ_pd : F.toFun ρ ∈ pdSetLM (ℋ := 𝒦) := faithfulApprox_pdSetLM E hlam0 hlam1 hρ hρ0
    have hFσ_pd : F.toFun σ ∈ pdSetLM (ℋ := 𝒦) := faithfulApprox_pdSetLM E hlam0 hlam1 hσ hσ0
    have ha_pos : 0 < (sandwichedQuasi α (F.toFun ρ) (F.toFun σ)).re :=
      lt_of_le_of_ne (_root_.SandwichedRenyiRelativeEntropy.sandwichedQuasi_re_nonneg _ _ _)
        (Ne.symm (sandwichedQuasi_re_ne_zero_of_pdSetLM hα0 hFρ_pd hFσ_pd))
    have hTrF : (Tr (F.toFun ρ)).re = (Tr ρ).re := by
      rw [← F.trace_map ρ]
    have hD := _root_.SandwichedRenyiRelativeEntropy.sandwichedRenyiDiv_faithfulApprox_le E hα_ge hα_lt hρ hσ hρ0 hσ0 hQ hlam0 hlam1
    -- Convert `D_F ≤ D_ρ` to `Q_ρ ≤ Q_F`.
    unfold sandwichedRenyiDiv at hD
    rw [hTrF] at hD
    -- `(1/(α-1)) log (Q_F/t) ≤ (1/(α-1)) log (Q_ρ/t)` with `1/(α-1) < 0` ⟹ flip.
    have hlog : Real.log ((sandwichedQuasi α ρ σ).re / (Tr ρ).re) ≤
        Real.log ((sandwichedQuasi α (F.toFun ρ) (F.toFun σ)).re / (Tr ρ).re) := by
      have hinv_neg : 1 / (α - 1) < 0 := one_div_neg.mpr hαm
      by_contra hcon
      push_neg at hcon
      exact absurd hD (not_le.mpr (by
        apply mul_lt_mul_of_neg_left hcon hinv_neg))
    rw [Real.log_le_log_iff (by positivity) (by positivity),
        div_le_div_iff_of_pos_right hTrρ_pos] at hlog
    exact hlog
  -- `λ → 0⁺`: `Q_α(F_λρ‖F_λσ).re → Q_α(Eρ‖Eσ).re`.
  have hEρ : (0 : L 𝒦) ≤ E.toFun ρ := map_nonneg E.toCompletelyPositiveMap hρ
  have hEσ : (0 : L 𝒦) ≤ E.toFun σ := map_nonneg E.toCompletelyPositiveMap hσ
  have hcont := sandwichedQuasi_re_continuousOn_nonneg (ℋ := 𝒦) hα0 hα_lt.le
  have hcwa : ContinuousWithinAt
      (Function.uncurry (fun (ρ σ : L 𝒦) => (sandwichedQuasi α ρ σ).re))
      ({A : L 𝒦 | 0 ≤ A} ×ˢ {A : L 𝒦 | 0 ≤ A}) (E.toFun ρ, E.toFun σ) :=
    hcont (E.toFun ρ, E.toFun σ) ⟨hEρ, hEσ⟩
  have hpair : Filter.Tendsto
      (fun l : {l : ℝ // 0 < l ∧ l ≤ 1} =>
        (((faithfulApprox E l.val l.property.1.le l.property.2).toFun ρ),
         ((faithfulApprox E l.val l.property.1.le l.property.2).toFun σ)))
      (Filter.comap (fun l : {l : ℝ // 0 < l ∧ l ≤ 1} => l.val) (nhdsWithin 0 (Set.Ioi 0)))
      (nhdsWithin (E.toFun ρ, E.toFun σ) ({A : L 𝒦 | 0 ≤ A} ×ˢ {A : L 𝒦 | 0 ≤ A})) := by
    rw [tendsto_nhdsWithin_iff]
    refine ⟨(faithfulApprox_tendsto E ρ).prodMk_nhds (faithfulApprox_tendsto E σ), ?_⟩
    filter_upwards with l
    exact Set.mk_mem_prod
      (nonneg_of_pdSetLM (faithfulApprox_pdSetLM E l.property.1 l.property.2 hρ hρ0))
      (nonneg_of_pdSetLM (faithfulApprox_pdSetLM E l.property.1 l.property.2 hσ hσ0))
  have hlim := hcwa.tendsto.comp hpair
  haveI : (Filter.comap (fun l : {l : ℝ // 0 < l ∧ l ≤ 1} => l.val)
      (nhdsWithin 0 (Set.Ioi 0))).NeBot := by
    refine Filter.comap_neBot fun t ht => ?_
    obtain ⟨U, hU_open, hU0, hU_sub⟩ := mem_nhdsWithin.mp ht
    obtain ⟨δ, hδ_pos, hball⟩ := Metric.mem_nhds_iff.mp (hU_open.mem_nhds hU0)
    have hx_pos : (0 : ℝ) < min (δ / 2) 1 := lt_min (by positivity) one_pos
    refine ⟨⟨min (δ / 2) 1, hx_pos, min_le_right _ _⟩, ?_⟩
    apply hU_sub
    refine ⟨hball ?_, hx_pos⟩
    simp only [Metric.mem_ball, Real.dist_eq, sub_zero, abs_of_pos hx_pos]
    exact lt_of_le_of_lt (min_le_left _ _) (by linarith)
  have hle : (sandwichedQuasi α ρ σ).re ≤ (sandwichedQuasi α (E.toFun ρ) (E.toFun σ)).re :=
    ge_of_tendsto hlim (Filter.Eventually.of_forall hQ_ge)
  exact ne_of_gt (lt_of_lt_of_le hb_pos hle)

/-! ### Real-valued monotonicity (used in the main theorem) -/

/-- `Q_α` continuity along a pd perturbation path `ε ↦ (ρ+εP, σ+εP)` (all pd). -/
 lemma sandwichedQuasi_tendsto_pd
    {ℋ : Type u} [Qudit ℋ] [Nontrivial ℋ]
    {α : ℝ} (_hα0 : 0 < α) {ρ σ P : L ℋ}
    (hρ : ρ ∈ pdSetLM (ℋ := ℋ)) (hσ : σ ∈ pdSetLM (ℋ := ℋ)) (hP : P ∈ pdSetLM (ℋ := ℋ)) :
    Filter.Tendsto
      (fun ε : ℝ => (sandwichedQuasi α (ρ + (ε : ℂ) • P) (σ + (ε : ℂ) • P)).re)
      (nhdsWithin 0 (Set.Ioi 0))
      (nhds (sandwichedQuasi α ρ σ).re) := by
  have hcont := sandwichedQuasi_re_continuousOn_pdSetLM (ℋ_aux := ℋ) α
  set S : Set (L ℋ × L ℋ) := pdSetLM (ℋ := ℋ) ×ˢ pdSetLM (ℋ := ℋ) with hS
  set g : ℝ → L ℋ × L ℋ := fun ε => (ρ + (ε : ℂ) • P, σ + (ε : ℂ) • P) with hg_def
  have hg_cont : Continuous g := by fun_prop
  have hg0 : g 0 = (ρ, σ) := by simp [hg_def]
  have hg_tendsto : Filter.Tendsto g (nhdsWithin 0 (Set.Ioi 0)) (nhdsWithin (ρ, σ) S) := by
    rw [tendsto_nhdsWithin_iff]
    refine ⟨?_, ?_⟩
    · rw [← hg0]; exact (hg_cont.tendsto 0).mono_left nhdsWithin_le_nhds
    · filter_upwards [self_mem_nhdsWithin] with ε hε
      have hεP : (ε : ℂ) • P ∈ pdSetLM (ℋ := ℋ) := pdSetLM_pos_smul hP hε
      refine Set.mk_mem_prod ?_ ?_
      · rw [add_comm]; exact pdSetLM_add_nonneg (nonneg_of_pdSetLM hεP) hρ
      · rw [add_comm]; exact pdSetLM_add_nonneg (nonneg_of_pdSetLM hεP) hσ
  have hcwa : ContinuousWithinAt
      (Function.uncurry (fun (ρ σ : L ℋ) => (sandwichedQuasi α ρ σ).re)) S (ρ, σ) :=
    hcont (ρ, σ) ⟨hρ, hσ⟩
  have hcomp := (hcwa.tendsto).comp hg_tendsto
  simpa only [Function.comp_def, Function.uncurry, hg_def] using hcomp

/-- `D_α` continuity along a pd perturbation path. -/
 lemma sandwichedRenyiDiv_tendsto_pd
    {ℋ : Type u} [Qudit ℋ] [Nontrivial ℋ]
    {α : ℝ} (hα0 : 0 < α) {ρ σ P : L ℋ}
    (hρ : ρ ∈ pdSetLM (ℋ := ℋ)) (hσ : σ ∈ pdSetLM (ℋ := ℋ)) (hP : P ∈ pdSetLM (ℋ := ℋ)) :
    Filter.Tendsto
      (fun ε : ℝ => sandwichedRenyiDiv α (ρ + (ε : ℂ) • P) (σ + (ε : ℂ) • P))
      (nhdsWithin 0 (Set.Ioi 0))
      (nhds (sandwichedRenyiDiv α ρ σ)) := by
  unfold sandwichedRenyiDiv
  have hQc := _root_.SandwichedRenyiRelativeEntropy.sandwichedQuasi_tendsto_pd hα0 hρ hσ hP
  have hTr0 : (Tr ρ).re ≠ 0 :=
    ne_of_gt (trace_re_pos_of_ne_zero (nonneg_of_pdSetLM hρ) (isUnit_of_pdSetLM hρ).ne_zero)
  have hQ0 : (sandwichedQuasi α ρ σ).re ≠ 0 := sandwichedQuasi_re_ne_zero_of_pdSetLM hα0 hρ hσ
  have hTc : Filter.Tendsto (fun ε : ℝ => (Tr (ρ + (ε : ℂ) • P)).re)
      (nhdsWithin 0 (Set.Ioi 0)) (nhds (Tr ρ).re) := by
    have hform : ∀ ε : ℝ, (Tr (ρ + (ε : ℂ) • P)).re = (Tr ρ).re + ε * (Tr P).re := by
      intro ε
      rw [map_add, map_smul, smul_eq_mul, Complex.add_re, Complex.re_ofReal_mul]
    simp_rw [hform]
    have h2 : Filter.Tendsto (fun ε : ℝ => (Tr ρ).re + ε * (Tr P).re)
        (nhds 0) (nhds ((Tr ρ).re + 0 * (Tr P).re)) :=
      tendsto_const_nhds.add ((continuous_id.mul continuous_const).tendsto 0)
    simpa using h2.mono_left nhdsWithin_le_nhds
  have hRatio : Filter.Tendsto
      (fun ε : ℝ => (sandwichedQuasi α (ρ + (ε : ℂ) • P) (σ + (ε : ℂ) • P)).re /
        (Tr (ρ + (ε : ℂ) • P)).re)
      (nhdsWithin 0 (Set.Ioi 0)) (nhds ((sandwichedQuasi α ρ σ).re / (Tr ρ).re)) :=
    Filter.Tendsto.div hQc hTc hTr0
  have hLog := (Real.continuousAt_log (div_ne_zero hQ0 hTr0)).tendsto.comp hRatio
  exact hLog.const_mul _

/-- **Faithful-approximation DPI** (`α > 1`): for each `λ ∈ (0,1]`, `D_α(F_λρ ‖ F_λσ) ≤
    D_α(ρ‖σ)`. Take `ε → 0⁺` in the perturbed PD inequality: the LHS limit at the PD pair
    `(F_λρ, F_λσ)` uses PD-continuity, the RHS limit uses `suppLE`-boundary continuity. -/
 lemma sandwichedRenyiDiv_faithfulApprox_le_gt
    {ℋ 𝒦 : Type u} [Qudit ℋ] [Nontrivial ℋ] [Qudit 𝒦] [Nontrivial 𝒦]
    (E : CPTP ℋ 𝒦) {α : ℝ} (hα_gt : 1 < α)
    {ρ σ : L ℋ} (hρ : 0 ≤ ρ) (hσ : 0 ≤ σ) (hρ0 : ρ ≠ 0) (hsupp : suppLE ρ σ)
    {lam : ℝ} (hlam0 : 0 < lam) (hlam1 : lam ≤ 1) :
    sandwichedRenyiDiv α ((faithfulApprox E lam hlam0.le hlam1).toFun ρ)
        ((faithfulApprox E lam hlam0.le hlam1).toFun σ) ≤ sandwichedRenyiDiv α ρ σ := by
  have hα_ge : (1 : ℝ) / 2 ≤ α := by linarith
  have hα0 : 0 < α := by linarith
  have hσ0 : σ ≠ 0 := by
    rintro rfl
    apply hρ0
    refine LinearMap.ext fun x => ?_
    rw [LinearMap.zero_apply]
    exact LinearMap.mem_ker.mp (hsupp (LinearMap.mem_ker.mpr (by simp)))
  set F := faithfulApprox E lam hlam0.le hlam1 with hF
  have hF1 : F.toFun 1 ∈ pdSetLM (ℋ := 𝒦) := faithfulApprox_one_pdSetLM E hlam0 hlam1
  have hFρ_pd : F.toFun ρ ∈ pdSetLM (ℋ := 𝒦) := faithfulApprox_pdSetLM E hlam0 hlam1 hρ hρ0
  have hFσ_pd : F.toFun σ ∈ pdSetLM (ℋ := 𝒦) := faithfulApprox_pdSetLM E hlam0 hlam1 hσ hσ0
  have h_pert : ∀ ε : ℝ, 0 < ε →
      sandwichedRenyiDiv α (F.toFun ρ + (ε : ℂ) • F.toFun 1) (F.toFun σ + (ε : ℂ) • F.toFun 1)
        ≤ sandwichedRenyiDiv α (ρ + (ε : ℂ) • (1 : L ℋ)) (σ + (ε : ℂ) • (1 : L ℋ)) := by
    intro ε hε
    have hpd := sandwichedRenyiDiv_monotone_nonneg_perturbed F hα_ge (ne_of_gt hα_gt) hρ hσ hF1 hε
    have hsmul := LinearMap.map_smul F.toCompletelyPositiveMap.toLinearMap (ε : ℂ) (1 : L ℋ)
    have eρ : F.toFun (ρ + (ε : ℂ) • (1 : L ℋ)) = F.toFun ρ + (ε : ℂ) • F.toFun (1 : L ℋ) := by
      have h1 := LinearMap.map_add F.toCompletelyPositiveMap.toLinearMap ρ ((ε : ℂ) • (1 : L ℋ))
      rw [hsmul] at h1; exact h1
    have eσ : F.toFun (σ + (ε : ℂ) • (1 : L ℋ)) = F.toFun σ + (ε : ℂ) • F.toFun (1 : L ℋ) := by
      have h1 := LinearMap.map_add F.toCompletelyPositiveMap.toLinearMap σ ((ε : ℂ) • (1 : L ℋ))
      rw [hsmul] at h1; exact h1
    rw [eρ, eσ] at hpd
    exact hpd
  have h_RHS := _root_.SandwichedRenyiRelativeEntropy.sandwichedRenyiDiv_tendsto_of_suppLE hα_gt hρ hσ hsupp hρ0
  have h_LHS := _root_.SandwichedRenyiRelativeEntropy.sandwichedRenyiDiv_tendsto_pd hα0 hFρ_pd hFσ_pd hF1
  haveI : (nhdsWithin (0 : ℝ) (Set.Ioi 0)).NeBot := nhdsWithin_Ioi_neBot (le_refl 0)
  exact le_of_tendsto_of_tendsto h_LHS h_RHS
    (Filter.eventually_of_mem self_mem_nhdsWithin h_pert)
end SandwichedRenyiRelativeEntropy


/- Mechanically delaborated from the original compiled Lean 4.30 ConstantInfo. -/
public theorem SandwichedRenyiRelativeEntropy.sandwichedQuasi_re_ne_zero_of_CPTP._simp_1_2.{u} :
∀ {α : Type u} [inst : PseudoMetricSpace.{u} α] {x y : α} {ε : Real},
  Eq.{1} (Membership.mem.{u, u} (Metric.ball.{u} x ε) y) (LT.lt.{0} (Dist.dist.{u} y x) ε) :=
fun {α} [PseudoMetricSpace.{u} α] {x y} {ε} => propext Metric.mem_ball.{u}
Source
https://github.com/Hayata-Yamasaki-Group/lean-quantum/blob/bf1c4f6aaec84948f1a1c76c0728432813404a0f/Quantum/QuantumEntropy/SandwichedRenyiNonNeg.lean#L1280-L1775

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me