From copulas to dependence nuclei: what Sklar's theorem contributes
Sklar's theorem states that any multivariate joint distribution can be decomposed into its univariate marginals and a copula that carries the dependence structure; when marginals are continuous, the copula is unique [1][2]. This separation is the conceptual foundation the new paper imports into quantum information: once local measurement bases are fixed, the Born rule produces an ordinary classical joint probability table, and one can ask what dependence remains after marginals are removed [1]. The paper is explicit that the discrete case is subtler, because the Sklar copula is generally not unique for discrete marginals, so it adopts the dependence-nucleus viewpoint of Geenens, in which probability tables related by positive local reweightings are treated as having the same dependence [1]. For a strictly positive 2×2 table, the odds ratio is a complete coordinate for this nucleus, and each nucleus has a unique uniform-margin representative that is the discrete counterpart of a copula [1]. This methodological choice is what makes the reconstruction theorem possible without selecting an arbitrary copula representative [1].
The all-context reconstruction theorem and its double-spin-flip ambiguity
The central result concerns two non-product two-qubit states ρ and σ. If, for every pair of local binary projective measurements, the margin-free dependence quantity Q agrees, then σ equals ρ or its double spin flip, where the double spin flip is (σy⊗σy)ρ*(σy⊗σy) [1]. The proof writes each state in Fano–Bloch form and shows that non-product states necessarily have full-rank reduced states, so all row and column marginals of every Born table are strictly positive even when individual entries vanish [1]. This positivity guarantees that the denominator in Q is strictly positive in every local context, so Q remains well defined for pure entangled and rank-deficient mixed states [1]. The equality of Q across all contexts forces the connected correlation bilinear forms to be proportional, and a careful inertia argument then forces the proportionality constant to be one, leaving only the two sign branches that correspond to identity or double spin flip [1]. The theorem therefore says that information discarded context by context is not lost when the entire compatible family of contexts is considered together [1].
Product states as the degenerate boundary of relational reconstruction
The theorem is stated for non-product states, and the paper shows why product states form the natural degenerate case: if ρ = ρA⊗ρB, the Born probabilities factorize in every local context, so p++p−− = p+−p−+ for all measurement directions, and the connected correlation vanishes identically [1]. Conversely, if that factorization holds in every local context, the correlation matrix factorizes as T = abT and the state is product [1]. This means the absence of nontrivial binary dependence across all local contexts is equivalent to the state being product, so the reconstruction problem has a clean relational boundary rather than a technical exception [1]. The paper also notes a simple case where both reduced states are maximally mixed: every Born table already has uniform marginals, and the all-context Q data directly determine the full correlation matrix because the map c ↦ 2c/(1+c²) is one-to-one on [−1,1] [1]. This example illustrates how the dependence data can recover the correlation matrix without any marginal information [1].
Compressing dependence to a scalar recovers global quantum discord
The paper then considers a coarser description: instead of retaining the full dependence nucleus in each context, it compresses the dependence in a single Born table to the classical total correlation, also called multi-information [1]. For continuous variables, copula entropy satisfies Tcl = −Hc, so the total correlation is exactly the negative copula entropy; the paper notes that for discrete Born probabilities it uses Tcl directly because the discrete Sklar copula is not unique [1]. Optimizing this scalar over all local measurement contexts gives Cmax(ρ) = max_B Tcl(pρ,B), the largest classical total correlation that can be made manifest in a single local measurement context [1]. The paper states that this quantity, combined with the quantum total correlation, reproduces global quantum discord [1]. This connects the all-context dependence reconstruction to a known measure of quantum correlations, and it shows that the scalar compression of dependence is not merely a technical simplification but lands on an established discord-type quantity [1].
Lineage, competing approaches, and where the conclusion stops
The new result sits in a lineage that includes Mermin's demonstration that a complete set of subsystem correlations determines the density operator, and quantum self-testing, where observed correlations can uniquely determine the underlying state and measurements [1]. The question here is different: even after each Born table is individually stripped of its marginals, does the entire context-indexed dependence family still determine ρ [1]? The paper also distinguishes its construction from a previously proposed quantum copula defined directly at the level of the quantum state, rather than context-wise from Born statistics [1]. On the practical side, precursor work on quantum state tomography with locally purified density operators and local measurements shows that local measurement data can reconstruct mixed states with a gradient-based algorithm scaling as O(N log N D³) for pure 1D states, which is a different reconstruction philosophy: it keeps reduced density matrices rather than discarding marginals [3]. Validation work on two-qubit tomography compares optimal, MUB, standard, Pauli, and JKMW protocols and finds that the optimal protocol, which measures all density-matrix elements one by one, gives the most reliable results among the analyzed linear-inversion protocols [5]. Multicopy neural-network methods reduce measurement requirements by 67% compared to full quantum state tomography and maintain roughly twice the fidelity of standard QST under equivalent noise, showing that alternative routes to quantum correlations can outperform full reconstruction in resource terms [6]. Competing evidence on multipartite correlations shows that the multipartite symmetric quantum discord is nearly zero if and only if the state is approximately locally recoverable after measurements on each system, and that the conditional entanglement of multipartite information is a faithful entanglement measure vanishing exactly for fully separable states [4]. That work gives an operational and recoverability-based interpretation of discord-type quantities, which is a different explanatory route from the all-context dependence reconstruction [4]. The boundaries of the new theorem are explicit: it applies to non-product two-qubit states and local binary projective measurements [1]. The paper states that extensions to multiqubit and continuous-variable systems are discussed, and that for two bosonic modes the continuous-variable reconstruction problem asks whether the all-context copula family still identifies the state after local marginals are discarded context by context [1]. Whether one should enlarge the context family to more general local measurements, and whether doing so changes the state-separating power of the all-context dependence family, remain open questions [1]. The double-spin-flip ambiguity is discrete and does not imply that ρ and its flip are physically identical as quantum states [1]. The paper also notes that the discrete copula is not unique, which is why the dependence-nucleus equivalence class is used instead of an arbitrary Sklar copula [1]. These are the limits within which the conclusion should be read: the theorem is a structural reconstruction result for a specific state class and measurement family, not a claim that all quantum states are determined by margin-free dependence data [1].
About These Sources
This research page is built on 6 studies (5 peer-reviewed, 1 preprint) — published from 2016 to 2026, 5 from 2024 or later — selected as the most relevant from 8 studies that passed quality screening, drawn from 62 papers retrieved from a database of over 500 million.
Sources used in this answer
Sklar's Theorem and Quantum State Reconstruction from All-Context Dependence
Applies Sklar's theorem to Born statistics and proves that non-product two-qubit states are determined up to at most a double-spin-flip ambiguity by dependence nuclei over all local binary projective measurement contexts, with scalar optimization reproducing global quantum discord [1].
Copulas for Geostatistical Data: Foundations, Modeling Principles and Statistical Inference
Provides the foundational copula framework for separating marginal distributions from dependence structure, including the uniqueness of the copula for continuous marginals and the inversion-copula construction [2].
Quantum state tomography with locally purified density operators and local measurements
Presents a precursor quantum state tomography method using locally purified density operators and local measurements, with a gradient-based algorithm scaling as O(N log N D³) for pure 1D states and demonstrated on 1D pure and mixed states and 2D pure states up to 8×8 [3].
Multipartite quantum correlations and local recoverability.
Provides competing evidence that multipartite symmetric quantum discord is nearly zero if and only if the state is approximately locally recoverable after measurements on each system, and that conditional entanglement of multipartite information is a faithful entanglement measure [4].
Priority Choice Experimental Two-Qubit Tomography: Measuring One by One All Elements of Density Matrices.
Validates two-qubit tomography protocols experimentally and finds that optimal tomography, measuring all density-matrix elements one by one, gives the most reliable results among the analyzed linear-inversion protocols [5].
Resource-efficient quantum correlation measurements via multicopy neural network methods.
Validates resource-efficient quantum correlation measurement using multicopy measurements and neural networks, achieving a 67% reduction in measurement requirements compared to quantum state tomography and roughly twice the fidelity of standard QST under equivalent noise [7].