A nonglobal local minimum of the unitary scaling landscape
Abstract
A unitary matrix of order three has a nonglobal local scaling minimum.
Let Complex.I be the imaginary unit. A phase point x=(λ,ρ) consists of two real vectors indexed by Fin n. Diagonal unitary scaling gives B with entries Bⱼₖ=exp(((λⱼ+ρₖ : ℝ) : ℂ) Complex.I)Aⱼₖ. Write sum(B) for the sum of all entries, and Ψ_A(x)=landscape A x. The phase space has its usual product topology.
Definition 1.1 (The sum after diagonal scaling).
Formalization. D5/S3/Quantum/Matrix/UnitaryScalingLandscapeRefutation.scalingSum (✓ std3).
Citation. Alexis De Vos; Stijn De Baerdemacker (2014). Scaling a unitary matrix. DOI: 10.1142/S1230161214500139. URL: https://arxiv.org/abs/1401.7883v2.
Commentary.
The double sum is sum(B). It is defined for every complex square matrix A and every pair of real phase vectors, without a unitarity hypothesis. De Vos and De Baerdemacker, Scaling a unitary matrix, Appendix B, define Bⱼₖ=exp(((λⱼ+ρₖ : ℝ) : ℂ) Complex.I)Aⱼₖ and use this sum in their landscape.
Definition 1.2 (The scaling landscape).
Formalization. D5/S3/Quantum/Matrix/UnitaryScalingLandscapeRefutation.landscape (✓ std3).
Citation. Alexis De Vos; Stijn De Baerdemacker (2014). Scaling a unitary matrix. DOI: 10.1142/S1230161214500139. URL: https://arxiv.org/abs/1401.7883v2.
Commentary.
Appendix B defines Ψ_A(λ,ρ)=(n : ℝ)²−‖sum(B)‖². Here the same expression is defined for every A; the conjecture restricts A to the unitary group. The norm is the complex modulus.
Definition 1.3 (Every local minimum is global).
Formalization. D5/S3/Quantum/Matrix/UnitaryScalingLandscapeRefutation.claim (✓ std3).
Citation. Alexis De Vos; Stijn De Baerdemacker (2014). Scaling a unitary matrix. DOI: 10.1142/S1230161214500139. URL: https://arxiv.org/abs/1401.7883v2.
Commentary.
The assertion quantifies over every natural n, including zero, every complex n×n matrix A, and every phase point x. If A is unitary and x is a local minimum of Ψ_A in the sense of IsLocalMin, then Ψ_A(x)≤Ψ_A(y) for every phase point y. This expresses the absence of nonglobal local minima in Appendix B: “We conjecture that all these stationary points are either maxima or saddle points or global minima. In other words: we conjecture that no local minima exist.” Section 2 likewise conjectures that stationary points with 0<Ψ<(n : ℝ)² are saddle points. The separate assertion that global minima have Ψ=0 is not part of claim.
Theorem 1.4 (A unitary counterexample of order three).
Proof. Machine-checked in Lean as D5/S3/Quantum/Matrix/UnitaryScalingLandscapeRefutation.result (✓ std3). ∎
Resolves. Problems/de-vos-de-baerdemacker-2014-unitary-scaling-landscape-refutation (refuted) by D5/S3/Quantum/Matrix/UnitaryScalingLandscapeRefutation.result.
Source. Repository-derived.
Acknowledgement. Alexis De Vos; Stijn De Baerdemacker (2014). Scaling a unitary matrix. DOI: 10.1142/S1230161214500139. URL: https://arxiv.org/abs/1401.7883v2.
Commentary.
Take A=(1/34)[[−2,24,24],[24,1+17 Complex.I,1−17 Complex.I],[24,1−17 Complex.I,1+17 Complex.I]]. Direct multiplication gives A^* A=I, where A^* is the conjugate transpose, so A is unitary. At the zero phase point, sum(A)=49/17 and Ψ_A(0)=9−(49/17)²=200/289. Use one-based indices for the three coordinates. The set U={ (λ,ρ) : |ρ₂−ρ₃|<1/50 } is an open neighbourhood of zero. For zₖ=exp(((ρₖ : ℝ) : ℂ) Complex.I), the triangle inequality and the row-modulus estimate give ‖sum(B)‖≤Σⱼ‖(Az)ⱼ‖≤49/17 on U, hence Ψ_A≥200/289 there. Thus zero is a local minimum. For the comparison, choose ρ=(0,2π/3,π/3) and align the row phases by λⱼ=−arg((Az)ⱼ). The squared row moduli are 433/289, 217/289−√3/34, and 217/289+√3/34. Their moduli are at least 6/5, 4/5, and 8/9, respectively, so the aligned sum satisfies ‖sum(B)‖=Σⱼ‖(Az)ⱼ‖≥26/9>49/17. Consequently this phase point has Ψ_A<200/289, and the local minimum at zero is not global. This counterexample refutes the landscape conjecture of De Vos and De Baerdemacker. It does not refute the existence of a scaling with Ψ_A=0.
References
- Truth anchor:
D5/S3/Quantum/Matrix/UnitaryScalingLandscapeRefutation.claim - Truth anchor:
D5/S3/Quantum/Matrix/UnitaryScalingLandscapeRefutation.landscape - Truth anchor:
D5/S3/Quantum/Matrix/UnitaryScalingLandscapeRefutation.result - Truth anchor:
D5/S3/Quantum/Matrix/UnitaryScalingLandscapeRefutation.scalingSum