Source-linked AI summary

Classical, quantum, and general probabilistic state-discrimination profiles

Mihály Weiner

arXiv:2609.17920v1quant-phcs.ITmath.IT

TL;DR

The paper asks which constraints on state-discrimination profiles are universal and which require classical or quantum structure. It characterizes GPT profiles through measurement optimization and establishes strict classical–quantum–GPT separation for three-state discrimination, alongside results for larger families and prior-weighted profiles.

  • Problem

    The paper asks which restrictions on discrimination profiles follow from universal operational rules and which require specifically classical or quantum structure.

    Source: §1 · p. 2
  • Method

    The paper characterizes the GPT region by showing that optimizing over measurements yields finite upper envelopes of classical success profiles, all realizable by a Cartesian-product construction.

    Source: §1 · p. 2
  • Results

    For three-state profiles, the trine attains the quantum lower bound and is optimal among qubit and arbitrary pure-state triples, while the quantum upper bound 2/3 is not optimal.

    Source: §1 · p. 3
  • Takeaways & Limitations

    A full prior-weighted profile can contain discrimination information not present in the profile’s subfamily samples.

    Source: §9 · p. 19

    The facet theorem establishes existence of a quantum violation for each GPT-violated classical facet, but does not identify quantum maxima or establish simultaneous violations.

    Source: §7 · p. 17

Abstract

from arXiv · show

For a collection of states indexed by $[n]={1,\ldots,n}$, let $F(H)$ denote the optimal unnormalized minimum discrimination error for each nonempty subfamily $H\subseteq[n]$. We call $F$ the discrimination profile and characterize exactly which profiles can arise in a general probabilistic theory (GPT). For fixed $n$, the GPT region $\mathfrak{G}_n$ is a bounded rational polytope, giving a hierarchy $\mathfrak{C}_n\subseteq\mathfrak{Q}_n\subseteq\mathfrak{G}_n$ of classical, quantum, and GPT profiles, analogous to the local--quantum--no-signalling hierarchy in Bell theory. For $n=3$ we determine the classical and GPT polytopes explicitly. Within $\mathfrak{G}_3$, the classical region is characterized by $\beta:=F(123)-F(12)-F(13)-F(23)\le0$, while the GPT maximum is $1$. For qubit triples, the maximal value is attained by the trine, $\beta_{\mathrm{tr}}=3\sqrt3/2-2$, and we prove the dimension-independent quantum bound $\beta_{\mathrm Q}\le2/3$, with a slight further improvement. Hence $0=\beta_{\mathrm C}<3\sqrt3/2-2\le\beta_{\mathrm Q}\le2/3<1=\beta_{\mathrm{GPT}}$. We conjecture $\beta_{\mathrm Q}=\beta_{\mathrm{tr}}$ and prove this for arbitrary pure-state triples, together with further supporting evidence. For arbitrary $n$, every classical facet admitting a GPT violation already admits a quantum violation in dimension at most $3$. A four-state qubit example shows that quantum success profiles need not be submodular, unlike for $n=3$. Finally, we exhibit a four-dimensional quantum triple whose discrete profile is classical for all subfamilies but becomes nonclassical when unequal priors are allowed.

1 Introduction

The paper asks which restrictions on discrimination profiles follow from general operational rules and which require classical or quantum structure. It characterizes the GPT profile region, compares the classical, quantum, and GPT regions, and establishes further results for three and more states.

1 Introduction

  • The motivating question is which profile constraints are universal and which depend on specifically classical or quantum structure.The paper uses the classical–quantum–GPT hierarchy as an organizing framework for this distinction.Source: §1 · p. 2
  • Every finite upper envelope of classical success profiles is realizable in a GPT, yielding an exact description of the GPT region.A Cartesian product of classical state spaces realizes the envelope; optimizing over measurements gives the upper envelope in success coordinates.Source: §1 · pp. 2–3
  • For three states, one additional inequality cuts the classical region out of the GPT region, and the paper determines both regions’ vertices and facets.The quantum supremum of the associated functional is described as a discrimination analogue of a Tsirelson bound.Source: §1 · p. 3
  • A qubit trine attains the lower bound and is optimal among qubit triples and arbitrary pure-state triples; whether it is optimal for all mixed-state triples remains open.The 2/3 upper bound is not optimal; determining the optimal quantum bound remains open.Source: §1 · p. 3
  • For arbitrary n, each classical facet that a GPT can violate can also be violated by quantum states in dimension at most three.A four-state qubit family also shows quantum success profiles need not be submodular, unlike profiles with three hypotheses.Source: §1 · p. 3
  • A four-dimensional quantum triple has a classical discrete profile under uniform priors but a nonclassical prior-weighted profile.The weighted formulation recovers the discrete profile at indicator weights.Source: §1 · p. 3

2 Profiles and universal measurement constraints

This section defines profile regions and establishes universal measurement constraints. It also shows how measurements can be reused across overlapping subproblems to obtain pairwise and spanning-tree error bounds.

2.1 Framework and normalization

  • The framework uses finite-dimensional compact convex GPT state spaces, allows every measurement in the no-restriction convention, and includes classical and quantum models.The regions use finite alphabet sizes and Hilbert-space dimensions, with no closure taken.Source: §2.1 · pp. 3–4
  • The unnormalized error and success values are F(H) and D(H), with equal-prior probabilities obtained by dividing by |H|.The error profile satisfies F({i}) = 0 and 0 ≤ F(H) ≤ |H| − 1; the state list need not contain distinct states.Source: §2.1 · p. 4
  • Mixing preparations with a readable, hypothesis-independent flag makes each of the classical, quantum, and GPT profile regions convex.For quantum states, the construction uses block-diagonal states; the classical and GPT versions use flagged direct sums.Source: §2.1 · p. 5

2.2 One-point gluing and the tree bound

  • Reusing an optimal measurement for A ∪ B yields an error bound for overlapping subproblems by assigning outcomes outside each subproblem to a common hypothesis.The errors of the two reassigned measurements add up to F(A ∪ B).Source: §2.2 · p. 5
  • Iterating the one-point bound over spanning trees gives a tree-based error bound that strengthens the recalled quantum pairwise-average bound in the stated cases.For m identical states, the tree bound is exact at m − 1, while the pairwise-average bound gives m/2; the tree inequality is identified with Hunter’s bound.Source: §2.2 · p. 5

3 A fixed measurement and classical profiles

Fixed measurements yield classical success profiles with a normalized coverage-function structure. Signed coverage coefficients then provide coordinates for describing the classical region and its facets.

3 A fixed measurement and classical profiles

  • Fixed-measurement success profiles are exactly normalized coverage functions, each realizable as the full optimal profile of classical distributions.This characterizes the profile, not just the performance of a particular measurement.Source: §3 · p. 6
  • The classical region is a rational polytope in error coordinates, equivalently represented in success coordinates by normalized coverage profiles.Source: §3 · p. 7
  • An infinite classical alphabet does not enlarge the discrete profile region.Source: §3 · p. 7

3.1 Signed coverage coefficients and classical facets

  • The unique signed coverage coefficients give coordinates beyond the classical region, and their coordinate inequalities are exactly its distinct facets.The classical region has dimension 2^n − n − 1 for n ≥ 3.Source: §3.1 · p. 7

4 Exact characterization of GPT profiles

The paper characterizes GPT discrimination profiles as finite envelopes of classical profiles and gives a realization on Cartesian products of classical simplexes. It also provides a finite linear-programming characterization of the resulting bounded rational polytope.

4 Exact characterization of GPT profiles

  • A GPT profile is characterized by a finite envelope of classical profiles and is realizable on a Cartesian product of classical simplexes.Source: §4 · p. 8

4.1 Necessity

  • For each subfamily, an optimal measurement supplies a classical supporting profile that touches the GPT success profile there and stays below it elsewhere.Source: §4.1 · p. 8
  • Even restricted measurement sets yield the same outer profile region when closed under classical decoding.The argument also covers optima that are only suprema, by taking approximating sequences.Source: §4.1 · p. 8

4.2 Measurements on a Cartesian product

  • On a Cartesian product, every finite-outcome affine measurement is a convex mixture of measurements depending on a single coordinate.Source: §4.2 · pp. 8–9

4.3 Converse realization and dimension

  • For a downward-closed family containing all singletons, the construction realizes that family as the zero set of the full error profile.The realization uses a Cartesian product, not a classical joint probability simplex, whose affine effects would permit simultaneous coordinate readout.Source: §4.3 · p. 9
  • The universal realization can have dimension potentially exponentially larger than the n − 1 dimensions of the cited construction, leaving dimension-efficient realization as an open geometric issue.Source: §4.3 · p. 9

4.4 A finite linear program and the outer polytope

  • A finite linear program characterizes the GPT region as a bounded rational polytope and provides a realization certificate, though its variables grow exponentially.Source: §4.4 · p. 10
  • GPT error profiles need not be supermodular; a controlled loss in the overlap inequality is needed instead.Source: §4.4 · p. 10

5 The complete three-state geometry

The three-state GPT and classical regions have explicit polytope descriptions. Their key distinction is captured by β: classical profiles satisfy β ≤ 0, while GPT profiles can reach 1.

The GPT polytope

  • For three states, G3 is a region in R4 with an irredundant facet description that includes t ≥ x + y, t ≥ x + z, t ≥ y + z, and t ≤ 1 + x, 1 + y, 1 + z.Source: §5 · p. 10

Polytope geometry

  • The GPT feasibility constraints admit a convex-combination construction, and the classical region is obtained by cutting the GPT polytope with the β ≤ 0 hyperplane.The listed GPT vertices have classical realizations except P, whose profile is realized by coordinate readout on the square.Source: §5 · p. 11

The classical polytope

  • The classical three-state polytope has seven facets and vertices O, X, Y, Z, W, and P.Its additional constraint beyond the GPT inequalities is t ≤ x + y + z.Source: §5 · pp. 11–12

The three-state profile witness

  • For three states, β ≤ 0 characterizes classical profiles, while its maximum over the GPT region is 1, attained only at P.The classical region adds the constraint β ≤ 0 to the GPT constraints.Source: §5 · p. 12

6 Quantum violation and dimension-independent bounds

The paper characterizes quantum discrimination bounds for triples, identifies the qubit extremizer, and proves a dimension-independent separation from GPT profiles. It improves the quantum bound but leaves the sharp arbitrary-dimensional value open.

6.1 Operator formulation

  • The proof framework formulates optimal success and its dual as semidefinite programs, with strong duality and the Helstrom formula for pairs.For triples, it abbreviates the total success and the sum of pairwise trace distances to bound β.Source: §6.1 · p. 13

6.2 Qubits and the trine

  • Among qubit triples, equality in the maximum occurs precisely for a pure trine, up to unitary conjugation and permutation.The trine profile is nonclassical.Source: §6.2 · pp. 13–14

6.3 An elementary strict quantum–GPT separation

  • The quantum supremum for triples is at most 2/3 in every finite dimension, strictly below the GPT maximum.Together with the trine witness, this yields strict classical–quantum–GPT separation using one linear functional.Source: §6.3 · p. 15

6.4 Why the endpoint bound is not sharp

  • The arbitrary-dimensional quantum bound is explicitly improved beyond 2/3, but the sharp quantum value remains unresolved.The paper conjectures that the trine value is the sharp bound and proves this for all pure-state triples.Source: §6.4 · p. 15, §6.5 · p. 16

6.5 The remaining conjecture

  • For arbitrary pure-state triples, the conjectured trine bound holds, with equality only for an embedded trine.Removing a common positive semidefinite component cannot decrease β; mixed triples attaining the trine value are also described.Source: §6.5 · p. 16

7 Quantum violations of classical facets beyond three states

Every classical facet that GPT profiles can violate also admits a quantum violation in dimension at most three. Four-state quantum examples additionally show that success profiles need not be submodular.

Quantum violations of classical facets

  • Every classical facet admitting a GPT violation also admits a quantum violation on a Hilbert space of dimension at most three.For n ≥ 4, singleton facets are universal while every facet indexed by a set of size at least two is quantum-violatable; for n = 3, only the triple facet is.Source: §7 · p. 16

A four-state counterexample

  • A four-state qubit profile violates submodularity, although submodularity holds universally for three hypotheses.The strict violations persist under small perturbations, so repeated hypotheses are not essential.Source: §7 · p. 17

Scope of the facet theorem

  • The facet result establishes existence of a quantum violation for each eligible facet, not its quantum maximum or simultaneous violations.It does not imply Q_n = G_n.Source: §7 · pp. 16–17

8 Unequal priors can reveal hidden nonclassicality

A quantum triple can have a classical discrete profile yet a nonclassical profile when unequal priors are considered. The example shows that uniform-prior subfamily data do not determine classicality across all priors.

Prior-weighted profiles

  • A weighted classical profile requires one family of classical distributions to reproduce success for every prior; the family cannot change with the prior.The discrete profile samples only the uniform prior on each subfamily.Source: §8 · p. 17

A classical discrete profile with a nonclassical weighted profile

  • Three density matrices on C4 have a complete discrete profile in the classical region, yet a positive weighted witness rules out every classical model of their weighted profile.The same example is classical under uniform priors on all subfamilies but nonclassical for a suitable unequal weighting.Source: §8 · pp. 18–19

A weighted nonclassicality witness

  • For the entire prior-dependent profile, the single unweighted witness β is no longer a complete test of classicality, because unequal weights rebalance a trine component suppressed in uniform-prior data.For w = (10, 1, 1), the trine-block weights are equal; the common classical block contributes −1/10 to Bw and the trine block contributes 9βtr/10.Source: §8 · p. 19

9 Discussion and open questions

The discussion leaves the exact unrestricted quantum bound on β open and identifies further questions about larger hypothesis sets, realization dimension, and prior-dependent profiles.

Quantum bound

  • The unrestricted quantum optimum remains unknown: the trine is optimal for qubits and pure triples, but whether it is optimal for all mixed triples is open.A proof of the conjectured bound would determine the exact quantum value for the sole nonuniversal classical facet at n = 3, but not the entire quantum region.Source: §9 · p. 19

Larger hypothesis sets

  • For larger n, quantum success profiles can fail submodularity, while every classical facet violated by a GPT profile also has a quantum violation in dimension at most three.Open questions include quantum bounds for these facets, relations among their violations, and new GPT constraints at higher n.Source: §9 · p. 19

Dimension

  • The universal GPT construction can be large, leaving minimal-dimensional realization and comparisons between GPT and Cartesian-product realizations open.The discussion also raises comparisons between fixed-dimensional and unrestricted quantum profile regions.Source: §9 · p. 19

Prior-dependent profiles

  • Profiles over all priors contain more discrimination information than samples over subfamilies, motivating study of finite-prior collections and full prior-dependent profiles.This extension complements rather than replaces the finite polyhedral problem.Source: §9 · p. 19

Author contributions and use of AI tools

The author directed the project and takes responsibility for its mathematical claims and final manuscript, while AI tools contributed across mathematical development and writing.

Author contributions and use of AI tools

  • The author formulated and directed the project, developed its mathematical direction, and takes responsibility for the claims and final manuscript.Conversations with Sol 5.6 and Astra contributed to mathematical material, examples, calculations, code, literature searches, drafting, and revision.Source: Author contributions and use of AI tools · p. 20

A.1 Pure states in arbitrary dimension

The appendix proves trine optimality for pure triples in arbitrary finite dimension, shows how removing common positive mass affects β, and constructs mixed triples attaining the trine value.

A.1 Pure states in arbitrary dimension

  • For pure triples in any finite dimension, β is maximized by the trine, with equality only for an embedded planar trine.The proof uses a pretty-good measurement on the states' span and bounds its success through Gram-matrix eigenvalues and a scalar inequality.Source: §A.1 · pp. 20–21

A.2 Removing a common positive component

  • Removing a common positive component cannot decrease β and strictly increases it when its trace is positive and the original β exceeds −1.Thus a search for values above the trine bound can be restricted to triples with zero common support intersection; this is a support reduction, not a reduction to pure states.Source: §A.2 · p. 21

A.3 Mixed triples attaining the trine value

  • Mixed ancilla tensorings of a trine and flagged mixtures yield mixed triples attaining the trine value, so pure-state optimality does not classify all maximizers.The common support intersection can still be zero.Source: §A.3 · p. 22
Loading 2609.17920v1…