Keyboard shortcuts

Press ← or → to navigate between chapters

Press ? to show this help

Press Esc to hide this help

The lattice-walk cross-integral estimate fails

Abstract

Early coherent amplitudes contradict the uniform lattice-walk cross-integral estimate.

All finite indices are natural numbers, i is the imaginary unit, and log is the natural logarithm. Integrals use Lebesgue measure on the real time axis, and the norm of a complex number is its modulus. In the exponentials, the indices are embedded in the real or complex numbers before subtraction, so j − k is the signed difference.

Definition 1.1 (The oscillatory term of the cycle walk).

Formalization. D5/S3/Quantum/Dynamics/LatticeWalkCrossIntegralRefutation.walkTerm (✓ std3).

Citation. Shyam Dhamapurkar; Xiu-Hao Deng (2023). Quantum walk mixing is faster than classical on periodic lattices. DOI: 10.1016/j.physa.2023.129252. URL: https://arxiv.org/abs/2309.16352v1.

Commentary.

For every natural n, integer displacement l and real time t, walkTerm(n,l,t) is the complex double sum in Eq. (B8). It omits the diagonal j = k and the antidiagonal j + k = n. The last exponential equals w raised to l(j − k), where w = exp(2πi/n), so l is the displacement q − p. When n = 0 both index ranges are empty and the sum is zero.

Definition 1.2 (The proposed uniform cross-integral bound).

Formalization. D5/S3/Quantum/Dynamics/LatticeWalkCrossIntegralRefutation.claim (✓ std3).

Citation. Shyam Dhamapurkar; Xiu-Hao Deng (2023). Quantum walk mixing is faster than classical on periodic lattices. DOI: 10.1016/j.physa.2023.129252. URL: https://arxiv.org/abs/2309.16352v1.

Commentary.

Equation (15), the two-dimensional estimate used after Conjecture 1, asserts this bound for every pair of odd, coprime natural orders n₁ > n₂, every pair of natural displacements l₁ < n₁ and l₂ < n₂, and every real T ≥ 0. The two factors are evaluated at the same time t. The right side is 32n₁(n₂ log n₂)² + 32n₂(n₁ log n₁)² and is independent of T.

Theorem 1.3 (A counterexample at unit time).

Proof. Machine-checked in Lean as D5/S3/Quantum/Dynamics/LatticeWalkCrossIntegralRefutation.result (✓ std3). ∎

Resolves. Problems/dhamapurkar-deng-2023-lattice-walk-cross-integral-refutation (refuted) by D5/S3/Quantum/Dynamics/LatticeWalkCrossIntegralRefutation.result.

Source. Repository-derived.

Acknowledgement. Shyam Dhamapurkar; Xiu-Hao Deng (2023). Quantum walk mixing is faster than classical on periodic lattices. DOI: 10.1016/j.physa.2023.129252. URL: https://arxiv.org/abs/2309.16352v1.

Commentary.

Take n₂ = 10¹⁰ + 1, n₁ = 10¹⁰ + 3, l₁ = l₂ = 1 and T = 1. The two orders are odd and coprime: any common divisor divides their difference 2, and an odd number has no even divisor. Both displacements lie in the required ranges.

For an odd natural n ≥ 3 and real t, write A_n(t) = Σ_{0≤j<n} exp(i(t/2)cos(2πj/n)) exp(2πij/n). The cosine difference identity expands |A_n(t)|² into the full double sum. Its diagonal contributes n. Its antidiagonal, disjoint from the diagonal for odd n, contributes −1 by the sum of the nontrivial powers of exp(4πi/n). Consequently walkTerm(n,1,t) = |A_n(t)|² − (n − 1), a real number.

For every natural n ≥ 3, pairing j with n − j cancels the sine-cosine cross terms in the imaginary part of A_n(t). The identity Σ_{0≤j<n} cos²(2πj/n) = n/2 and the inequality u sin u ≥ u²/2 for |u| ≤ 1/2 give Im A_n(t) ≥ nt/8 for 0 < t ≤ 1. Hence |A_n(t)|² ≥ n²/256 on [1/2,1]. The triangle inequality also gives 0 ≤ |A_n(t)|² ≤ n² for every real t.

For odd n₁,n₂ ≥ 3, expand the product using these two real identities. Bound each subtracted squared modulus by nᵢ² and discard the nonnegative constant (n₁ − 1)(n₂ − 1). The product of the squared moduli is nonnegative on [0,1] and at least n₁²n₂²/65536 on [1/2,1]. Thus its real integral satisfies ∫₀¹ walkTerm(n₁,1,t)walkTerm(n₂,1,t) dt ≥ n₁²n₂²/131072 − (n₁ − 1)n₂² − (n₂ − 1)n₁².

At the stated orders, 0 ≤ log n₂ ≤ log n₁ ≤ 34 log 2 < 24, since n₁ < 2³⁴. The proposed upper bound is therefore at most 32·24²(n₁n₂² + n₂n₁²), which is strictly smaller than the integral lower bound above. The norm of the integral is at least its real value, contradicting Eq. (15).

References