Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Adiabatic theorem with T ≥ (10⁴/δ²)(‖H′‖³/γ⁴ + ‖H′‖‖H″‖/γ³)

Disproved
QAlgorithms.Nannicini.adiabatic_theorem

by Goku · Oct 13, 2026 · Mathlib 0df444a (Lean v4.33.1)

lecture-notesnanniciniquantum-information

Let H(s)H(s)H(s), s∈[0,1]s \in [0,1]s∈[0,1], be a time-dependent Hamiltonian on a finite-dimensional space Cι\mathbb{C}^\iotaCι: each H(s)H(s)H(s) is Hermitian and the entries of HHH are twice differentiable on [0,1][0,1][0,1], with entrywise derivatives H′(s)H'(s)H′(s), H′′(s)H''(s)H′′(s). Let N1,N2N_1, N_2N1​,N2​ bound the operator norms, ∥H′(s)∥≤N1\|H'(s)\| \le N_1∥H′(s)∥≤N1​ and ∥H′′(s)∥≤N2\|H''(s)\| \le N_2∥H′′(s)∥≤N2​ for all s∈[0,1]s \in [0,1]s∈[0,1].

Let λ:[0,1]→R\lambda : [0,1] \to \mathbb{R}λ:[0,1]→R be a continuous eigenvalue branch and ∣φ(s)⟩|\varphi(s)\rangle∣φ(s)⟩ a unit eigenvector of H(s)H(s)H(s) with eigenvalue λ(s)\lambda(s)λ(s), for each s∈[0,1]s \in [0,1]s∈[0,1]. Let γ>0\gamma > 0γ>0 be a spectral gap around λ\lambdaλ: for every sss, λ(s)\lambda(s)λ(s) is a simple eigenvalue of H(s)H(s)H(s) and every other eigenvalue μ\muμ of H(s)H(s)H(s) satisfies ∣μ−λ(s)∣≥γ|\mu - \lambda(s)| \ge \gamma∣μ−λ(s)∣≥γ.

Let δ>0\delta > 0δ>0 and let TTT satisfy

T≥104δ2(N13γ4+N1N2γ3).T \ge \frac{10^4}{\delta^2}\left(\frac{N_1^3}{\gamma^4} + \frac{N_1 N_2}{\gamma^3}\right).T≥δ2104​(γ4N13​​+γ3N1​N2​​).

Let ∣ψ(s)⟩|\psi(s)\rangle∣ψ(s)⟩ solve the adiabatic evolution

d∣ψ(s)⟩ds=iT H(s)∣ψ(s)⟩,s∈[0,1],∣ψ(0)⟩=∣φ(0)⟩.\frac{d|\psi(s)\rangle}{ds} = iT\,H(s)|\psi(s)\rangle, \quad s \in [0,1], \qquad |\psi(0)\rangle = |\varphi(0)\rangle .dsd∣ψ(s)⟩​=iTH(s)∣ψ(s)⟩,s∈[0,1],∣ψ(0)⟩=∣φ(0)⟩.

Then there is θ∈R\theta \in \mathbb{R}θ∈R with

∥eiθ∣ψ(1)⟩−∣φ(1)⟩∥≤δ.\bigl\| e^{i\theta} |\psi(1)\rangle - |\varphi(1)\rangle \bigr\| \le \delta .​eiθ∣ψ(1)⟩−∣φ(1)⟩​≤δ.

This is the adiabatic theorem in the form of Ambainis and Regev. A slow enough evolution carries an eigenvector of H(0)H(0)H(0) to the corresponding eigenvector of H(1)H(1)H(1), up to a global phase. It is the basis of adiabatic optimization: start in the known ground state of a simple Hamiltonian and interpolate to a Hamiltonian that encodes the problem.

Formalization Note The source's ∥H′∥=max⁡s∥H′(s)∥\|H'\| = \max_s\|H'(s)\|∥H′∥=maxs​∥H′(s)∥, ∥H′′∥\|H''\|∥H′′∥ and γ=min⁡sγ(s)\gamma = \min_s \gamma(s)γ=mins​γ(s) (Defs. 9.6–9.9) are replaced by arbitrary bounds N1,N2N_1, N_2N1​,N2​ and an arbitrary lower bound γ>0\gamma > 0γ>0 on the instantaneous gaps. The required lower bound on TTT increases in N1,N2N_1, N_2N1​,N2​ and decreases in γ\gammaγ, so this is equivalent to the source's statement and avoids junk values of real suprema. Derivatives at s=0,1s = 0, 1s=0,1 are one-sided. Only the precise final-state claim is stated: the source's "approximately remains … for all sss" carries no stated error.

Preamble
import Definitions.Def_QAlgorithms_Defs_Adiabatic
set_option autoImplicit false
Formal statement
namespace QAlgorithms.Nannicini

open QAlgorithms.Stabilizer


/-- Nannicini pp.212–213, Theorem 9.11 (Adiabatic theorem). Let `H(s)` be a time-dependent
Hamiltonian, let `λ(s)` be an eigenvalue of `H(s)`, and let `|φ(s)⟩` be an eigenstate of `H(s)`
with eigenvalue `λ(s)`. Assume that there is a spectral gap `γ > 0` around `λ(s)`. Assume further
that, from the initial state `|φ(0)⟩`, we apply the time evolution `d|ψ(s)⟩/ds = iT H(s)|ψ(s)⟩`,
`s ∈ [0, 1]`, `|ψ(0)⟩ = |φ(0)⟩`, and for some `δ > 0`, `T` satisfies
`T ≥ (10⁴/δ²)(‖H′‖³/γ⁴ + ‖H′‖‖H″‖/γ³)`. Then the system approximately remains in the
instantaneous eigenstate `|φ(s)⟩` with eigenvalue `λ(s)` for all `s`, and in particular, the final
state `|ψ(1)⟩` has Euclidean distance at most `δ` from `|φ(1)⟩`, up to global phase:
`‖e^{iθ}|ψ(1)⟩ − |φ(1)⟩‖ ≤ δ` for some `θ`.

Stated is the precise claim, the "in particular" about `ψ(1)`; the unquantified "approximately
remains … for all s" carries no stated error. Hermiticity of `H(s)` is part of `HasGapAround`.
`‖H′‖, ‖H″‖, γ` are the bounds `N1, N2, γ` of the module docstring. -/
theorem adiabatic_theorem {ι : Type*} [Fintype ι] [DecidableEq ι] (H H' H'' : ℝ → Matrix ι ι ℂ)
    (hH' : ∀ s ∈ Set.Icc (0 : ℝ) 1, ∀ i j,
      HasDerivWithinAt (fun t => H t i j) (H' s i j) (Set.Icc 0 1) s)
    (hH'' : ∀ s ∈ Set.Icc (0 : ℝ) 1, ∀ i j,
      HasDerivWithinAt (fun t => H' t i j) (H'' s i j) (Set.Icc 0 1) s)
    (lam : ℝ → ℝ) (hlam : ContinuousOn lam (Set.Icc 0 1)) (φ : ℝ → EuclideanSpace ℂ ι)
    (hφ : ∀ s ∈ Set.Icc (0 : ℝ) 1, IsState (φ s) ∧ act (H s) (φ s) = (lam s : ℂ) • φ s)
    (γ : ℝ) (hγ : 0 < γ) (hgap : ∀ s ∈ Set.Icc (0 : ℝ) 1, HasGapAround (H s) (lam s) γ)
    (N1 N2 : ℝ) (hN1 : ∀ s ∈ Set.Icc (0 : ℝ) 1, specNorm (H' s) ≤ N1)
    (hN2 : ∀ s ∈ Set.Icc (0 : ℝ) 1, specNorm (H'' s) ≤ N2)
    (δ : ℝ) (hδ : 0 < δ) (T : ℝ)
    (hT : 10 ^ 4 / δ ^ 2 * (N1 ^ 3 / γ ^ 4 + N1 * N2 / γ ^ 3) ≤ T)
    (ψ : ℝ → EuclideanSpace ℂ ι) (hψ : SolvesAdiabaticPlus H T ψ) (hψ0 : ψ 0 = φ 0) :
    ∃ θ : ℝ, ‖Complex.exp ((θ : ℂ) * Complex.I) • ψ 1 - φ 1‖ ≤ δ := by
  sorry




end QAlgorithms.Nannicini
Source
Giacomo Nannicini, Quantum algorithms for optimizers, 2025, v5; notes dated 9 July 2025, PDF p. 212, Theorem 9.11. https://arxiv.org/pdf/2408.07086v5
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

Let ι\iotaι be a finite index set and let H,H′,H′′:R→Cι×ιH, H', H'' : \mathbb{R} \to \mathbb{C}^{\iota\times\iota}H,H′,H′′:R→Cι×ι be matrix-valued functions. Write I=[0,1]I = [0,1]I=[0,1]. All derivatives below are taken within III, so they are one-sided at s=0s = 0s=0 and s=1s = 1s=1.

Smoothness of HHH.

  • For every s∈Is \in Is∈I and every entry (i,j)(i,j)(i,j), the function t↦H(t)ijt \mapsto H(t)_{ij}t↦H(t)ij​ has derivative H′(s)ijH'(s)_{ij}H′(s)ij​ at sss.
  • Likewise t↦H′(t)ijt \mapsto H'(t)_{ij}t↦H′(t)ij​ has derivative H′′(s)ijH''(s)_{ij}H′′(s)ij​ at sss.
  • No continuity of H′′H''H′′ is assumed.

Tracked eigenpair. Let λ:R→R\lambda : \mathbb{R}\to\mathbb{R}λ:R→R be continuous on III, and let φ:R→Cι\varphi : \mathbb{R}\to\mathbb{C}^\iotaφ:R→Cι satisfy, for each s∈Is \in Is∈I:

  • ∥φ(s)∥=1\|\varphi(s)\| = 1∥φ(s)∥=1;
  • H(s)φ(s)=λ(s)φ(s)H(s)\varphi(s) = \lambda(s)\varphi(s)H(s)φ(s)=λ(s)φ(s).

No continuity or differentiability of φ\varphiφ is assumed. If ι\iotaι is empty, the unit-norm condition cannot hold, and the theorem is vacuous.

Gap. Let γ>0\gamma > 0γ>0. For each s∈Is \in Is∈I, assume all three of the following:

  • H(s)H(s)H(s) is Hermitian.
  • In the increasingly sorted list of eigenvalues of H(s)H(s)H(s), counted with multiplicity, exactly one position equals λ(s)\lambda(s)λ(s). So λ(s)\lambda(s)λ(s) is a simple eigenvalue.
  • Every listed eigenvalue μ≠λ(s)\mu \ne \lambda(s)μ=λ(s) satisfies ∣μ−λ(s)∣≥γ|\mu - \lambda(s)| \ge \gamma∣μ−λ(s)∣≥γ.

Norm bounds. Let N1,N2∈RN_1, N_2 \in \mathbb{R}N1​,N2​∈R satisfy ∥H′(s)∥≤N1\|H'(s)\| \le N_1∥H′(s)∥≤N1​ and ∥H′′(s)∥≤N2\|H''(s)\| \le N_2∥H′′(s)∥≤N2​ for all s∈Is \in Is∈I. Here ∥⋅∥\|\cdot\|∥⋅∥ is the spectral (Euclidean operator) norm, so N1,N2≥0N_1, N_2 \ge 0N1​,N2​≥0 necessarily.

Evolution time. Let δ>0\delta > 0δ>0, and let T∈RT \in \mathbb{R}T∈R satisfy

104δ2(N13γ4+N1N2γ3)≤T.\frac{10^4}{\delta^2}\left(\frac{N_1^3}{\gamma^4} + \frac{N_1N_2}{\gamma^3}\right) \le T.δ2104​(γ4N13​​+γ3N1​N2​​)≤T.

Evolved state. Let ψ:R→Cι\psi : \mathbb{R} \to \mathbb{C}^\iotaψ:R→Cι satisfy, for every s∈Is \in Is∈I,

ψ′(s)=i T H(s)ψ(s)\psi'(s) = i\,T\,H(s)\psi(s)ψ′(s)=iTH(s)ψ(s)

(note the sign +i+i+i), with ψ(0)=φ(0)\psi(0) = \varphi(0)ψ(0)=φ(0).

Conclusion. There exists a real θ\thetaθ with

∥eiθψ(1)−φ(1)∥≤δ.\|e^{i\theta}\psi(1) - \varphi(1)\| \le \delta.∥eiθψ(1)−φ(1)∥≤δ.
Human review
  • Endorsed by Shuze Chen · Oct 13, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Goku · Oct 13, 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