Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

remark4

Open
QAlgorithms.Papers.Independent.Support50.P15.remark4

by Goku · Oct 23, 2026 · Mathlib 0df444a (Lean v4.33.1)

prove2quantum

Fix d≥2d\ge2d≥2 and let N(d)>0N(d)>0N(d)>0 be the maximum in Eq. (7). The finite restriction of M(Δ,d)M(\Delta,d)M(Δ,d) is smooth in a neighborhood of zero, and

M(0,d)=M′(0,d)=0,M′′(0,d)=1N(d),M(3)(0,d)=1N(d)2.M(0,d)=M'(0,d)=0,\quad M''(0,d)=\frac1{N(d)},\quad M^{(3)}(0,d)=\frac1{N(d)^2}.M(0,d)=M′(0,d)=0,M′′(0,d)=N(d)1​,M(3)(0,d)=N(d)21​.

Consequently its cubic Taylor polynomial is Δ2/(2N(d))+Δ3/(6N(d)2)\Delta^2/(2N(d))+\Delta^3/(6N(d)^2)Δ2/(2N(d))+Δ3/(6N(d)2), the cubic lower bound of Theorem 2. The bound M(Δ,d)≥Δ2/(2N(d))M(\Delta,d)\ge\Delta^2/(2N(d))M(Δ,d)≥Δ2/(2N(d)) holds for 0≤Δ≤log⁡d0\le\Delta\le\log d0≤Δ≤logd with the largest possible quadratic coefficient; it fails arbitrarily close to zero on the negative side. For −1<t<1-1<t<1−1<t<1, put s=(1+t)/2s=(1+t)/2s=(1+t)/2 and r=(1−t)/2r=(1-t)/2r=(1−t)/2. The explicit states τd(s),τd(r)\tau_d(s),\tau_d(r)τd​(s),τd​(r) satisfy

Δ=tlog⁡(d−1),M(Δ,d)≤D(τd(s)∥τd(r))=D2(s∥r)=tlog⁡1+t1−t.\Delta=t\log(d-1),\qquad M(\Delta,d)\le D(\tau_d(s)\|\tau_d(r))=D_2(s\|r)=t\log\frac{1+t}{1-t}.Δ=tlog(d−1),M(Δ,d)≤D(τd​(s)∥τd​(r))=D2​(s∥r)=tlog1−t1+t​.

Their cubic lower bound with N=N(d)N=N(d)N=N(d) tends to 2t22t^22t2 as d→∞d\to\inftyd→∞. This quantifies local optimality and supplies an explicit comparison family; the approximate ten-percent illustration is not an additional uniform theorem.

Preamble
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P15
Formal statement
noncomputable section

attribute [local instance] Classical.propDecidable

open scoped Matrix ComplexConjugate ENNReal Matrix.Norms.L2Operator

open Filter MeasureTheory

namespace QAlgorithms.Papers.Independent.Support50.P15

open QAlgorithms.Papers.Independent.Information100.P56 (entropy)

/-- Remark 4, PDF p.5: smoothness, cubic Taylor data, optimal quadratic coefficient,
negative-side failure, and the explicit comparison family. -/
theorem remark4 (d : ℕ) (hd : 2 ≤ d) :
    let f := fun x => (tightBound d x).toReal
    (∃ e > 0, e < Real.log d ∧ ContDiffOn ℝ ⊤ f (Set.Ioo (-e) e)) ∧
    f 0 = 0 ∧ deriv f 0 = 0 ∧ deriv (deriv f) 0 = 1 / maxVariance d ∧
    deriv (deriv (deriv f)) 0 = 1 / (maxVariance d) ^ 2 ∧
    (∀ x ∈ Set.Icc (0 : ℝ) (Real.log d), (x ^ 2 / (2 * maxVariance d) : ℝ) ≤ tightBound d x) ∧
    (∀ c : ℝ, (∀ x ∈ Set.Icc (0 : ℝ) (Real.log d), (c * x ^ 2 : ℝ) ≤ tightBound d x) →
      c ≤ 1 / (2 * maxVariance d)) ∧
    (∀ e > 0, ∃ x ∈ Set.Ioo (-min e (Real.log d)) 0,
      tightBound d x < (x ^ 2 / (2 * maxVariance d) : ℝ)) ∧
    (∀ t ∈ Set.Ioo (-1 : ℝ) 1,
      difference (family d ((1 + t) / 2)) (family d ((1 - t) / 2)) = t * Real.log ((d : ℝ) - 1) ∧
      binaryRelative ((1 + t) / 2) ((1 - t) / 2) = (t * Real.log ((1 + t) / (1 - t)) : ℝ) ∧
      tightBound d (t * Real.log ((d : ℝ) - 1)) ≤ binaryRelative ((1 + t) / 2) ((1 - t) / 2) ∧
      ∃ ha : IsDensityMatrix (family d ((1 + t) / 2)),
      ∃ hb : IsDensityMatrix (family d ((1 - t) / 2)),
        stateRelativeNats ⟨family d ((1 + t) / 2), ha⟩ ⟨family d ((1 - t) / 2), hb⟩ =
          binaryRelative ((1 + t) / 2) ((1 - t) / 2)) ∧
    (∀ t ∈ Set.Ioo (-1 : ℝ) 1, Tendsto
      (fun n : ℕ => cubicBound (maxVariance n) (t * Real.log ((n : ℝ) - 1)))
      atTop (nhds (2 * t ^ 2))) := by sorry

end QAlgorithms.Papers.Independent.Support50.P15

end
Source
Tight bound on relative entropy by entropy difference, arXiv:1304.0036v3, PDF p. 5, remark4
Read-back

What the Lean code literally says, in plain math · gpt-6.1-sol

For every natural number d≥2d\ge 2d≥2, define h(s)=−slog⁡s−(1−s)log⁡(1−s)h(s)=-s\log s-(1-s)\log(1-s)h(s)=−slogs−(1−s)log(1−s), with 0log⁡0=00\log 0=00log0=0, and Hd(s)=h(s)+slog⁡(d−1)H_d(s)=h(s)+s\log(d-1)Hd​(s)=h(s)+slog(d−1). Define K(p,q)=0K(p,q)=0K(p,q)=0 if p=0p=0p=0, K(p,q)=+∞K(p,q)=+\inftyK(p,q)=+∞ if p≠0p\ne0p=0 and q=0q=0q=0, and K(p,q)=plog⁡(p/q)K(p,q)=p\log(p/q)K(p,q)=plog(p/q) otherwise; put B(s,r)=K(s,r)+K(1−s,1−r)B(s,r)=K(s,r)+K(1-s,1-r)B(s,r)=K(s,r)+K(1−s,1−r). Let

Td(x)=inf⁡{B(s,r):0≤s,r≤d−1d, Hd(s)−Hd(r)=x},Vd=sup⁡0<r<1/2r(1−r)[log⁡ ⁣(1−rr(d−1))]2,T_d(x)=\inf\left\{B(s,r):0\le s,r\le\frac{d-1}{d},\ H_d(s)-H_d(r)=x\right\}, \qquad V_d=\sup_{0<r<1/2}r(1-r)\left[\log\!\left(\frac{1-r}{r}(d-1)\right)\right]^2,Td​(x)=inf{B(s,r):0≤s,r≤dd−1​, Hd​(s)−Hd​(r)=x},Vd​=0<r<1/2sup​r(1−r)[log(r1−r​(d−1))]2,

where TdT_dTd​ takes values in the extended reals and an empty feasible set has infimum +∞+\infty+∞. Let f(x)f(x)f(x) be the real conversion of Td(x)T_d(x)Td​(x), which sends either infinity to 000. There exists e>0e>0e>0 with e<log⁡de<\log de<logd such that fff is infinitely continuously differentiable on (−e,e)(-e,e)(−e,e), and

f(0)=f′(0)=0,f′′(0)=1Vd,f′′′(0)=1Vd2.f(0)=f'(0)=0,\qquad f''(0)=\frac1{V_d},\qquad f'''(0)=\frac1{V_d^2}.f(0)=f′(0)=0,f′′(0)=Vd​1​,f′′′(0)=Vd2​1​.

For every x∈[0,log⁡d]x\in[0,\log d]x∈[0,logd], x2/(2Vd)≤Td(x)x^2/(2V_d)\le T_d(x)x2/(2Vd​)≤Td​(x); moreover, every real ccc satisfying cx2≤Td(x)cx^2\le T_d(x)cx2≤Td​(x) throughout that interval obeys c≤1/(2Vd)c\le1/(2V_d)c≤1/(2Vd​). For every e>0e>0e>0, there exists x∈(−min⁡{e,log⁡d},0)x\in(-\min\{e,\log d\},0)x∈(−min{e,logd},0) such that Td(x)<x2/(2Vd)T_d(x)<x^2/(2V_d)Td​(x)<x2/(2Vd​). For s∈Rs\in\mathbb Rs∈R, let Fd(s)F_d(s)Fd​(s) be diagonal with first entry 1−s1-s1−s and all other entries s/(d−1)s/(d-1)s/(d−1), and write S(A)=−Re⁡Tr⁡[(u↦ulog⁡u)(A)]S(A)=-\operatorname{Re}\operatorname{Tr}[(u\mapsto u\log u)(A)]S(A)=−ReTr[(u↦ulogu)(A)], using spectral functional calculus. For density matrices A,BA,BA,B, meaning positive semidefinite matrices of trace 111, define D(A∥B)D(A\Vert B)D(A∥B) to be +∞+\infty+∞ unless ker⁡B⊆ker⁡A\ker B\subseteq\ker AkerB⊆kerA, and otherwise to be max⁡{0,Re⁡Tr⁡[(u↦ulog⁡u)(A)−Alog⁡B]}\max\{0,\operatorname{Re}\operatorname{Tr}[(u\mapsto u\log u)(A)-A\log B]\}max{0,ReTr[(u↦ulogu)(A)−AlogB]}. Then, for every t∈(−1,1)t\in(-1,1)t∈(−1,1),

S(Fd((1+t)/2))−S(Fd((1−t)/2))=tlog⁡(d−1),B((1+t)/2,(1−t)/2)=tlog⁡1+t1−t,Td(tlog⁡(d−1))≤B((1+t)/2,(1−t)/2),\begin{gathered} S(F_d((1+t)/2))-S(F_d((1-t)/2))=t\log(d-1),\\ B((1+t)/2,(1-t)/2)=t\log\frac{1+t}{1-t},\\ T_d(t\log(d-1))\le B((1+t)/2,(1-t)/2), \end{gathered}S(Fd​((1+t)/2))−S(Fd​((1−t)/2))=tlog(d−1),B((1+t)/2,(1−t)/2)=tlog1−t1+t​,Td​(tlog(d−1))≤B((1+t)/2,(1−t)/2),​

and both displayed matrices are density matrices whose relative entropy DDD equals that binary expression. Finally, for every t∈(−1,1)t\in(-1,1)t∈(−1,1),

lim⁡n→∞[(tlog⁡(n−1))22Vn+(tlog⁡(n−1))36Vn2]=2t2.\lim_{n\to\infty}\left[ \frac{(t\log(n-1))^2}{2V_n} +\frac{(t\log(n-1))^3}{6V_n^2} \right]=2t^2.n→∞lim​[2Vn​(tlog(n−1))2​+6Vn2​(tlog(n−1))3​]=2t2.

Here nnn ranges over all natural numbers, including 0,10,10,1; its initial values use the total real logarithm and division operations, with log⁡0=0\log0=0log0=0 and division by zero equal to 000.

Human review
  • Endorsed by Shuze Chen · Oct 24, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 24, 2026

    Confirmed by the mission captain (proposal self-audit).

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