{"id":"46660e06-3cdf-4911-ae12-7acfb3afc404","arxiv_id":"1908.10085","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For any set of m compatible quantum measurements, a parent measurement exists with at most d^2(m(o-1)+1) non-vanishing outcomes, a bound linear in the number of measurements, with small cases where the maximal complexity is required.","lead":"This paper asks how many outcomes a single 'parent' measurement needs in order to simulate a set of compatible quantum measurements. It proves a general upper bound that grows linearly with the number of measurements, not exponentially, while also giving small examples where the parent still needs every available outcome.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified.","rationale":"The reader's weakest assumption correctly identifies the cone representation and the D-versus-D+1 issue as the part of the proof to scrutinize, but on inspection the concern does not land. The proof explicitly adds the identity coordinate, which turns the affine normalization constraint into a linear coordinate of the cone; consequently the standard conic Carathéodory bound is D, not D+1. A linear dependence among the k > D sub-measurement vectors yields both the identity-coordinate relation Σ λ_a C_a = 0 and the needed measurement-coordinate relations, so the new parent C′_a = (1−γλ_a)C_a is normalized, positive, and reproduces every child measurement, including the omitted o-th outcome. The choice γ = 1/max λ_a is legitimate because the identity-coordinate relation forces at least one positive λ_a. The reduction is exactly the standard proof of Carathéodory for conic combinations and is internally consistent. No machine-checked proof exists, but the derivation is short and can be verified by hand; the numerical examples are not independently run, but the central upper bound does not depend on them. The verdict should remain unchanged.","tokens_in":12785,"tokens_out":20998,"duration_ms":227408,"concrete_test":"Implement the Appendix B reduction in exact arithmetic for a small non-trivial case, e.g. m = 5, o = 2, d = 2, with D = 24: start from a random parent with all 32 elements strictly positive and its compatible children, then repeatedly apply Eqs. (29)–(30) whenever the support exceeds D. Verify after every step that all M_{a|x} are reproduced, normalization Σ C′_a = I holds, and each C′_a is positive semidefinite, and confirm the process reaches support ≤ D.","verdict_should_be":"UNCHANGED","load_bearing_attack":"No significant objection identified. The central upper bound of Eq. (9) is proved by a valid Carathéodory-style reduction in Appendix B. The only delicate point is the role of the extra d² identity coordinate: including it makes the representation conic, so the applicable constant is D = d²(m(o−1)+1), not D+1. Linear dependence among k > D vectors gives both Σ_a λ_a C_a = 0 and Σ_{a:a_x=i} λ_a C_a = 0 for i < o; the o-th outcome is then reproduced through normalization. Because C_a ≥ 0 and at least one λ_a > 0, choosing γ = 1/max λ_a keeps every C′_a positive, preserves normalization and all children, and removes at least one outcome. Iterating terminates with at most D non-vanishing elements. I could not find a countervailing gap in this argument.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript investigates the minimal number of non-vanishing POVM elements required for a parent measurement of a set of jointly measurable measurements. The authors recall the canonical parent with o^m outcomes and prove in Appendix B that every set of m measurements with o outcomes on C^d admits a parent with at most d^2(m(o-1)+1) non-vanishing elements (Eq. (9)), a bound linear in m. They exhibit simple cases, including the noisy X/Z qubit pair and a qutrit example with m=3 and o=2, where the maximal number o^m of parent elements is necessary; the latter is certified by dual SDP witnesses. They also construct large compatible sets from partitions of a single parent, report numerical sampling of boundary complexity, translate the parent bound into a bound on hidden states in EPR-steering LHS models, and introduce probabilistic parents that can reduce parent size below the deterministic complexity.","tokens_in":12890,"tokens_out":19488,"duration_ms":197981,"significance":"If the main bound is valid, the result is a significant structural contribution: it replaces the naive exponential upper bound o^m on parent size with a linear-in-m bound, and it gives a concrete upper bound that can guide SDP-based searches for parent measurements. The proof is an elementary, parameter-free Carathéodory reduction, and the paper also provides explicitly checkable dual certificates for the small examples, together with an accompanying notebook for the numerical parts. The steering application is constructive and the resulting LHS bound is a useful byproduct. The paper opens a new set of questions about typical and worst-case parent complexity; the SDP-witness technique for proving lower bounds on parent size is likely to be reusable.","major_comments":[],"minor_comments":[{"comment":"The proof of the bound should explicitly state that the linear dependence is taken in the full space R^D including the d^2 identity coordinates, and that the vector lambda_a is chosen so that max_a lambda_a > 0. As printed, Eq. (27) displays only the child constraints and Eq. (30) silently assumes a positive maximum; the intended argument is correct, but these details should be made explicit.","section":"Appendix B, Eqs. (27)-(30)"},{"comment":"The expression for omega^{a*} contains what appears to be a duplicated additive term; as printed, a reader cannot verify the feasibility conditions (23) or the negative values in (25). Please correct the formula and, if space allows, include the explicit verification of the claimed inequalities.","section":"Appendix A, Eq. (21)"},{"comment":"The set O(N) is used where O(C) is meant in two of the displayed constraints; this typo should be fixed so that the SDP is stated consistently.","section":"Appendix A, Eq. (15)"},{"comment":"The sentence 'Indeed, {K1,K2,K3} is a valid POVM' should refer to {K0,K1,K2}, since K2 is defined as the complement of K0 and K1.","section":"Probabilistic Parents, Eq. (11)"},{"comment":"The histograms of boundary complexity are based on 1000 sampled points with no error bars or convergence checks; a brief statement about sampling uncertainty would make the numerical claim more precise.","section":"Figure 1 and Appendix C"},{"comment":"The assertion that typical compatible measurements 'will live in the interior' and hence require conic combinations of D points 'and not fewer' is heuristic; it should be labelled as an observation or conjecture rather than a proven statement.","section":"Complexity of typical measurements"},{"comment":"There are several typographical errors and inconsistent superscript and equation-number formatting (e.g., 'develep' in the main text); a careful proofread is needed before publication.","section":"Throughout"}],"recommendation":"minor_revision","confidential_remarks":"To the editor: The central theorem is sound and the paper is a good fit for the journal. The required changes are local and expository: the authors should make the identity-coordinate and sign conventions in Appendix B explicit, correct the typos in Appendix A and Eq. (11), and add a brief sampling-uncertainty note for the numerical histograms. I would be comfortable accepting once these revisions are made."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper you'll get from me is a genuine result, not a repackaging. Skrzypczyk, Hoban, Sainz, and Linden prove that any set of m compatible POVMs with o outcomes on C^d has a parent with at most d^2(m(o-1)+1) non-vanishing elements. That bound is linear in m, replacing the naive exponential o^m, and it is proved cleanly in Appendix B by a Carathéodory-style reduction on the cone of sub-measurements. I read the argument carefully: the extra d^2 identity coordinate is what lets you write the normalization as a conic equality, the linear dependence among more than D vectors gives a gamma that zeros out one element while preserving the children, and iteration terminates at D. The proof is sound.\n\nThe second solid piece is the certified example: three qutrit measurements, each two-outcome, that are compatible but require all eight parent elements. The SDP dual witnesses in Appendix A are explicit, and the negative values are small but unambiguous. That's a real counterpoint to the bound, showing exponential complexity can occur for small m. The Bell-number observation on the other side—one six-outcome parent gives 201 compatible children—is a nice structural complement. The transfer to EPR-steering LHS models follows from the known incompatibility-steering correspondence and is straightforward, with the caveat that Appendix D writes the map using rho^{-1}; strictly speaking you need full rank or a support restriction for the reduced state. Minor, fixable.\n\nSoft spots are mostly presentation. The citation at Eq. (6) is an empty bracket, the notebook link is just 'here' without a URL, and the boundary sampling in Appendix C is heuristic—interesting histograms, but no convergence or error analysis. The probabilistic parents section is openly preliminary; it raises a good question but doesn't resolve it. None of this affects the central theorem or the example.\n\nThis is a paper I'd send to a serious referee. The main bound is new, the proof is short and checkable, and the example is certified. I'd recommend accept after a light revision that fills in the missing citation and link and adds a sentence about the rank-deficient steering case. It belongs in the reading group.","headline":"Skrzypczyk et al. prove a linear-in-m bound on parent-measurement size and certify an exponential small-case example, so the central result is solid and worth refereeing.","tokens_in":13436,"tokens_out":4893,"would_cite":true,"duration_ms":48917,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P15","52A20"],"pacs":["03.65.Ta"],"model":"deepseek-v4-flash","headline":"For every family of $m$ compatible quantum measurements, a parent measurement exists with at most $d^2(m(o-1)+1)$ outcomes.","keywords":["joint measurability","POVM","parent measurement","measurement incompatibility","Carathéodory theorem","semidefinite programming","EPR steering","local hidden state model"],"falsifier":"Search for a counterexample where the bound is tighter than the canonical parent, for instance $m=5$ binary-outcome measurements on a qubit, where the claimed bound is 24 parent elements and the canonical parent has 32. Checking all sparsity patterns of size 24 with the compatibility SDP and finding a set whose minimal parent has 25 or more non-vanishing elements would refute equation (9).","tokens_in":12581,"feed_emoji":"⚛️","tokens_out":9245,"duration_ms":85077,"temperature":0.7,"pith_summary":"Quantum measurements that can be performed together are called compatible, and a single parent measurement can simulate them by coarse-graining its outcomes. The naive parent has $o^m$ outcomes, one for every combination of the children's outcomes, which grows exponentially with the number $m$ of measurements. This paper proves that such exponential parents are never necessary once $m$ is large: every set of $m$ compatible measurements on $\\mathbb{C}^d$ with $o$ outcomes has a parent with at most $d^2(m(o-1)+1)$ non-vanishing POVM elements. It also exhibits small cases where the exponential parent genuinely is required, so the bound is not vacuous but instead marks a transition in how complex compatibility can be.","feed_headline":"Parent measurements grow linearly, not exponentially","feed_subtitle":"A new bound caps parent size at a constant times the number of measurements, not an exponential.","key_machinery":"The load-bearing object is the joint-measurability cone: the collection of sub-normalised deterministic response functions embedded in $\\mathbb{R}^{D}$ with $D=d^2(m(o-1)+1)$. Its extremal rays are exactly the deterministic sub-measurements, and Carathéodory's theorem for cones states that any point in a $D$-dimensional cone is a non-negative combination of at most $D$ extremal rays. The proof repeatedly uses any linear dependence among more than $D$ parent elements to zero one element while preserving the children, terminating with a parent of size at most $D$. A complementary mechanism is the SDP dual: for a fixed sparsity pattern, weak duality produces a witness operator whose negativity certifies that no parent with that pattern exists, which is how the paper proves maximal complexity in the small examples.","core_discovery":"The central discovery is a linear-in-$m$ bound on the complexity of joint measurability. Writing a set of compatible measurements as a point in a real vector space of dimension $D=d^2(m(o-1)+1)$---one coordinate block for each sub-normalised measurement element plus extra coordinates for the normalisation---the paper identifies compatible sets with points in a convex cone whose extremal rays are deterministic sub-measurements. Carathéodory's theorem for cones then implies that any such point is a conic combination of at most $D$ extremal rays, and translating back yields a parent POVM with at most $D$ non-vanishing elements. The same geometric argument gives a corresponding bound on the number of hidden states in local-hidden-state models for EPR steering, since steering assemblages and compatible measurements are in one-to-one correspondence. The paper also constructs maximally complex examples---a pair of qubit measurements requiring all four parent elements and a triple of qutrit measurements requiring all eight---and develops an SDP-duality witness method to certify that no smaller parent exists.","pith_inferences":["If the constant $d^2$ can be lowered by using the affine dimension of the joint-measurability set rather than its cone dimension, the linear bound would carry over to those smaller constants; the paper's geometry suggests the bound is not an accident of the particular embedding.","The memory-time trade-off indicates a potential parallel strategy: enumerate sparsity patterns independently, so compatibility of large sets could be attacked by many small SDPs running in parallel rather than one huge SDP.","The probabilistic-parent example hints at a more general trade-off between randomisation and parent size; one could test whether the minimum deterministic parent size and the minimum probabilistic parent size are separated by a constant factor or grow with $m$.","One could use the same SDP-duality witness technique to study complexity beyond the small examples, for instance to map the regions of $m,o,d$ where the bound is tight, which the paper leaves open."],"forward_implications":["The memory needed to decide joint measurability by semidefinite programming can be capped: instead of storing $o^m$ parent elements, one may search over parents with at most $d^2(m(o-1)+1)$ non-zero elements.","Every EPR-steering assemblage that admits a local-hidden-state model has a model with at most $d^2(m(o-1)+1)$ hidden states, because the steering-to-compatibility map preserves the parent size.","A single $o$-outcome parent measurement is automatically a parent for many children: all the distinct partitions of its outcomes, of which there are $B_o-2$, the Bell number minus the two trivial partitions.","In small cases the upper bound is not tight and exponential parents are genuinely needed: the two-outcome qubit pair needs its full four-element parent, and the three-outcome qutrit triple needs all eight elements.","In parameter regimes where the bound is tight, typical interior compatible sets are maximally complex, while on the boundary the sampled complexity is spread over many values, indicating a rich boundary structure."],"supporting_citations":[{"why":"Establishes the joint-measurability framework and the parent-measurement formulation the paper studies.","marker":"[3]"},{"why":"Supplies the semidefinite-programming formulation and duality theory used to certify lower bounds on complexity.","marker":"[6]"},{"why":"Bell numbers count the children generated by partitioning a parent's outcomes.","marker":"[7]"},{"why":"Carathéodory's theorem for cones is the geometric engine behind the linear upper bound.","marker":"[8]"},{"why":"Defines EPR steering and local-hidden-state models, connecting compatibility complexity to steering complexity.","marker":"[10]"},{"why":"Establishes the one-to-one correspondence between incompatibility and steering used to transfer the bound.","marker":"[13]"},{"why":"Provides the converse part of the steering-incompatibility correspondence needed for the LHS model bound.","marker":"[14]"}],"fun_headline_variants":["Linear bound tames measurement incompatibility","Parent measurements: linear not exponential","Compatible measurements have linear-size parents","Joint measurability: complexity capped linearly","Quantum measurements: linear memory, not exponential"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof relies on representing a compatible set as a point in a cone of exactly $d^2(m(o-1)+1)$ dimensions, so that any such point is a combination of at most that many special rays; if the normalisation condition requires one more ray, the bound rises by one.","fun_headline_variants_meta":{"raw":{"variants":["Linear bound tames measurement incompatibility","Parent measurements: linear not exponential","Compatible measurements have linear-size parents","Joint measurability: complexity capped linearly","Quantum measurements: linear memory, not exponential"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00055,"raw_usage":{"total_tokens":2597,"prompt_tokens":891,"completion_tokens":1706,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":507,"completion_tokens_details":{"reasoning_tokens":1644}},"tokens_in":507,"tokens_out":1706,"duration_ms":13636,"temperature":1.0,"reasoning_tokens":1644,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:54:32.040820+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Search for a counterexample where the bound is tighter than the canonical parent, for instance $m=5$ binary-outcome measurements on a qubit, where the claimed bound is 24 parent elements and the canonical parent has 32. Checking all sparsity patterns of size 24 with the compatibility SDP and finding a set whose minimal parent has 25 or more non-vanishing elements would refute equation (9).","supporting_citations":[{"cited_title":"Busch, P","cited_arxiv_id":null,"evidence_quote":"Establishes the joint-measurability framework and the parent-measurement formulation the paper studies."},{"cited_title":"Guerini and M","cited_arxiv_id":null,"evidence_quote":"Supplies the semidefinite-programming formulation and duality theory used to certify lower bounds on complexity."},{"cited_title":"Boyd and L","cited_arxiv_id":null,"evidence_quote":"Bell numbers count the children generated by partitioning a parent's outcomes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Carathéodory's theorem for cones is the geometric engine behind the linear upper bound."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines EPR steering and local-hidden-state models, connecting compatibility complexity to steering complexity."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the converse part of the steering-incompatibility correspondence needed for the LHS model bound."}],"review_version":1}