Convergence of the JRF iteration for mixed-state discrimination

An analytic proof candidate for convergence of the uniformly initialized iteration, including rank-deficient states and overlapping supports.

Yuxuan Zhang

Full version (PDF, 8 pages) · LaTeX source and provenance (ZIP)

Archival source: Manuscript PDF on Zenodo, preserved in Agentic Proofs for QIQC: Collected Manuscripts, version 1.0 (26 September 2026).

Revision: 19 September 2026. The formal manuscript, independent audit summary, primary-source evidence and reproducible checks have been updated.

Status: a full proof candidate, internally reviewed by AI agents. External expert review and literature novelty remain unconfirmed. The argument is analytic; it has not been checked by a formal proof assistant.

Background

Minimum-error discrimination asks which measurement best identifies a state drawn from a known quantum ensemble. The Ježek–Řeháček–Fiurášek (JRF) iteration [1] gives an explicit update for the measurement. The open question [2] is whether its uniformly initialized iterates converge to an optimal measurement for arbitrary mixed-state ensembles.

Convergence of the success probability alone would not settle the whole question: the measurement itself must converge. Rank-deficient states also matter, since a proof that assumes each state is invertible leaves out part of the original problem.

The statement

Write the weighted states as \(A_i=p_i\rho_i\), with positive priors, and work on the joint signal support \(\mathcal H_0=\operatorname{supp}\sum_i A_i\). For a finite ensemble of \(m\) states, set

\[\Pi_{i,0}=I_{\mathcal H_0}/m,\qquad D_k=\left(\sum_i A_i\Pi_{i,k}A_i\right)^{1/2},\] \[\Pi_{i,k+1}=D_k^{-1}A_i\Pi_{i,k}A_iD_k^{-1}.\]

The manuscript proves that every finite-step inverse exists on this support and that there is a globally optimal minimum-error POVM \(\{\Pi_{i,\infty}\}\) with

\[\sum_i\|\Pi_{i,k}-\Pi_{i,\infty}\|_{\mathrm{HS}}\longrightarrow0.\]

Consequently the success probabilities converge to the optimum. The claim covers arbitrary finite dimension, finitely many outcomes, mixed or pure states, overlapping supports, and individual rank deficiency. It concerns the literal update above, without regularization or averaging.

How the proof works

Compress to the fixed supports. Factor each weighted state on its own support and introduce coherent factors of the POVM. Their square support compressions follow an exact multiplicative recurrence. They stay invertible at every finite step, even when an individual state is singular on the full signal space.

Prove convergence before using inverse bounds. The compressed update maximizes a linearized strongly convex quadratic over a fixed compact convex semialgebraic set. Sufficient decrease, a subgradient estimate at the next iterate, and subsequential continuity give the hypotheses of the Kurdyka–Łojasiewicz convergence theorem of Attouch, Bolte and Svaiter [3, Theorem 2.9]. This gives finite length and convergence of the compressed sequence. It does not by itself give global optimality.

Exclude a singular limiting normalizer. If the limit of \(D_k\) had a kernel, a determinant multiplier in one support block would eventually exceed two at every step. The corresponding bounded compression could not sustain that growth. Thus the limiting normalizer is positive definite, and convergence transfers back to the actual POVM.

Recover global optimality. A cone argument for the ordered, noncommuting products forces every dual inequality \(D_\infty\geq A_i\). The limit also satisfies complementarity, \((D_\infty-A_i)\Pi_{i,\infty}=0\). These two conditions certify the global discrimination optimum.

Verification

The complete revised manuscript received a fresh independent adversarial AI review on 19 September 2026: PASS, exact claim–proof alignment, with no mathematical gap identified. The executable supplement reproduced twice in clean verifier processes. The main mathematical argument is unchanged from the earlier working draft; this is a publication revalidation, not a new discovery. External expert review, formal proof verification and broad novelty confirmation remain outstanding.

The audit checked the support compression, the fixed convex feasible set, the KL hypotheses at the correct iterate, the singular-limit contradiction, the order of the noncommuting products, and reconstruction of the optimal POVM. The supplementary script checks the literal and compressed recurrences on 16 seeded complex ensembles in dimensions two through six, for 1,280 updates. The largest residual was below \(4.22\times10^{-13}\). These finite diagnostics test identities; the universal theorem rests on the written proof.

The eight-page manuscript includes the complete proof and a reproducibility appendix. The source archive contains LaTeX, the checker, detailed results, pinned package versions, the AI audit summary, and a source-evidence manifest. The cited KL theorem and pure-state result were checked against primary full texts. Scientific Agent Skills informed the evidence-tracking and writing process and is cited in the paper.

What remains open

The candidate is intended to cover the full uniformly initialized, finite-dimensional statement in the linked QIQCOP record. Independent expert scrutiny is still needed. Priority is not claimed; the literature search does not establish novelty.

The proof supplies no quantitative convergence rate and makes no assertion for arbitrary nonuniform initialization. Those are further questions beyond the theorem stated here.

References

  1. M. Ježek, J. Řeháček and J. Fiurášek, Finding optimal strategies for minimum-error quantum-state discrimination, Phys. Rev. A 65, 060301(R) (2002).
  2. QIQCOP, Convergence of the JRF iteration for mixed-state discrimination, problem op_76e284219621a785.
  3. H. Attouch, J. Bolte and B. F. Svaiter, Convergence of descent methods for semi-algebraic and tame problems: proximal algorithms, forward–backward splitting, and regularized Gauss–Seidel methods, Mathematical Programming 137, 91–129 (2013), Theorem 2.9; author manuscript.
  4. X. Lü and S.-H. Dong, Iterative algorithm for minimum-error quantum state discrimination: Convergence for pure-state ensembles, Phys. Rev. A 113, 022451 (2026). This is the pure-state result cited in the QIQCOP record; the present claim includes mixed ensembles.