betaFeedback

WorksMathematics

Lorentzian entropy, subgaussian matrix concentration, and sharp functional inequalities for all matroids

Abstract

We prove complete entropy contraction for homogeneous Lorentzian laws and derive subgaussian matrix concentration and sharp functional inequalities for every matroid. Uniform retention of ℓ\ell out of rr particles contracts matrix entropy by the optimal universal factor ℓ/r\ell/r, in every finite matrix dimension and positive field, including repeated labels. The matrix Laplace comparison retains matrix variance and arbitrary Hermitian offsets. It gives subgaussian bounds for arbitrary Hermitian observables with bounded exchange differences, and Chernoff and Bernstein bounds for basis sums. Every unit-edge matroid basis graph satisfies the cube's small-set and Gross inequalities, with equality classified and optimal universal complete entropy-decay rate four. Its spectrum lies in {0,2}∪[3,∞)\{0,2\}\cup[3,\infty), with binary direct-factor contrasts spanning the eigenspace at two. We classify extremal down-up eigenfunctions, resolving Alaniz Pintos's degree-one conjecture and the equality problem for the Feder–Mihail walk. For degree at least two, every matrix entropy extremizer extends uniquely from its two-particle marginal. Equality depends only on the normalized pair law, with noncommutativity confined to mutually exclusive labels. We classify power-entropy extremizers for 1<p<21<p<2 and give a fourth-order criterion for affine equality. Retained-level Gibbs reconstruction has sharp dimension-independent stability. For multiaffine laws whose generating polynomial and complement are Lorentzian, we determine the exact simultaneous all-field quantum marginal and matrix Hölder regions and classify the sharp half-marginal fields. Connected sharp laws are determinantal through rank three; stable nondeterminantal examples exist in every higher rank.

Introduction

Uniform retention of ℓ\ell out of rr particles from a homogeneous Lorentzian law contracts matrix entropy by the sharp universal factor ℓ/r\ell/r, at every positive field and in every finite matrix dimension. Repeated labels are allowed. For r≥2r \ge2, every matrix extremizer extends uniquely from its two-particle marginal; its noncommutative structure is confined to labels that never occur together.

For matroids, complete contraction gives nonlinear subgaussian concentration and matrix Chernoff and Bernstein bounds. The equality theory identifies all extremal down-up eigenfunctions and, for the unit-edge basis graph, the spectral separation Spec⁡(LG)⊆{0,2}∪[3,∞)\operatorname{Spec}(L_G) \subseteq\{0,2\} \cup[3,\infty). The eigenvalue-two space consists exactly of binary direct-factor contrasts, and the optimal universal complete entropy-decay rate is four.

Contraction, profiles and trace duality form the common entrance (§ 2–§ 4). The concentration bounds and all-field marginal region can then be read directly. For spectral applications, continue through scalar translations to § 9 and § 10; the quantum equality route passes through reconstruction and § 7 and § 8.

Complete contraction

Let f(z)=∑∣α∣=raαzαf(z) = \sum_{|\alpha|=r} a_\alpha z^\alpha be a nonzero homogeneous Lorentzian polynomial of degree r≥1r \ge1, and fix a positive field λ\lambda. Its coefficient law is

μ(α)=aαλαf(λ).\mu(\alpha) = \frac{a_\alpha\lambda^\alpha}{f(\lambda)}.

An occupation α\alpha contains αi\alpha_i particles of label ii. Retain a uniformly chosen set of ℓ\ell particles, write μℓ\mu_\ell for the retained law, and let Fℓ(β)=E[F(α)∣β]F_\ell(\beta) = \mathbb{E}[F(\alpha) \mid\beta] under the reference deletion coupling. For a positive semidefinite matrix function put

Hμ(F)=EμTr⁡(Flog⁡F)−Tr⁡[(EμF)log⁡(EμF)].\mathcal{H}_\mu(F) = \mathbb{E}_\mu\operatorname{Tr}(F \log F) - \operatorname{Tr}[(\mathbb{E}_\mu F)\log(\mathbb{E}_\mu F)].

No trace normalization is imposed.

Theorem 1.1 (Complete particle-deletion contraction). For every 0≤ℓ≤r0 \le\ell\le r and every positive semidefinite matrix function of any finite size,

Hμℓ(Fℓ)≤ℓrHμ(F).(1)\mathcal{H}_{\mu_\ell}(F_\ell) \le\frac{\ell}{r}\mathcal{H}_\mu(F). \tag*{(1)}

The coefficient is uniform in the positive field and matrix size, and is optimal over the class of Lorentzian laws.

The word complete records the uniformity in matrix size. For a normalized classical–quantum state, the entropy is relative entropy from μ⊗ρQ\mu\otimes\rho_Q, using the actual quantum marginal ρQ\rho_Q. The input classical marginal may differ from μ\mu. If it equals μ\mu, the entropy is mutual information. Deleting one particle and refilling conditionally gives complete modified logarithmic Sobolev constant at least 1/r1/r for the generator I−KI-K.

Lieb’s three-matrix inequality turns the Lorentzian Hessian signature into a quadratic entropy inequality. Conditioning on two unmarked particles identifies this defect with the second difference of the deletion profile. Convexity then gives contraction, while the same defects control strictness and equality.

Cryan–Guo–Mousa established scalar deletion contraction and scalar concentration for homogeneous strongly log-concave subset laws [17]. The scalar entropy framework also includes [7, 4]; Lee–Seo give the one-particle occupation version with repeated labels [33 Lemma 3.5 and Remark 3.6]. The quadratic converse is exact: scalar contraction, complete contraction and one positive eigenvalue are equivalent. Infinitesimal tests on coordinate derivatives at one positive field therefore recognize Lorentzianity (Corollary 4.4).

Quadratic models for entropy equality

For r≥2r \ge2, absorb the positive field and normalize f(1)=1f(1) = 1. Its translation space is

Nf={h:Dhf=0}=ker⁡∇2f(1).\mathcal{N}_f = \{h : D_h f = 0\} = \ker\nabla^2 f(1).

It determines all scalar extremizers. For matrices, the mean and the one-particle conditional densities also determine every extremizer, by Gibbs reconstruction. The following theorem constructs every equality density from its two-particle marginal.

Theorem 1.2 (Quadratic extension). Restrict to active labels and put W=∇2f(1)/(r(r−1))W = \nabla^2 f(1)/(r(r-1)) and f2(z)=zTWzf_2(z) = z^{\mathsf{T}}Wz. For each finite matrix dimension, two-particle conditional expectation is a bijection between the complete entropy equality densities of ff and those of f2f_2, normalized by Tr⁡EF=1\operatorname{Tr} EF = 1. Given a quadratic extremizer EijE_{ij}, set σ=∑ijwijEij\sigma= \sum_{ij} w_{ij}E_{ij}. On supp σ\sigma, put

Gi=∑jWijEij(W1)i,Bi=σ−1/2Giσ−1/2.G_i = \frac{\sum_j W_{ij}E_{ij}}{(W\mathbf{1})_i}, \qquad B_i = \sigma^{-1/2}G_i\sigma^{-1/2}.

Its unique extension is

Fα=σ1/2(∏iBiαi)σ1/2.F_\alpha= \sigma^{1/2}\left(\prod_i B_i^{\alpha_i}\right)\sigma^{1/2}.

Factors occurring in the same occupation commute. The extension has Hμ(F)/r=HW(E)/2H_\mu(F)/r = H_W(E)/2. More generally, any two Lorentzian laws with the same normalized WW have canonically identified equality families, even when their degrees differ.

The finite marginal test and normal form (Theorem 7.1) combine Petz recovery, Hayden–Jozsa–Petz–Winter’s Markov structure theorem [27], and Lorentzian translations. For matroid basis laws, the quantum register is independent of the simplified basis; only parallel-class choices carry equality information (Corollary 7.7). When every two distinct active labels can occur together, logarithmic extremizers are simultaneously diagonalizable. In this case, fixing the classical marginal leaves only constant equality encodings, although a singular Hessian makes ℓ/r\ell/r the sharp supremum, approached by qubit encodings. In general, informative fixed-marginal equality exists exactly when two distinct active labels never co-occur, and a qubit suffices.

Power equality for 1<p<21 < p < 2 places the exclusive classes in orthogonal reducing subspaces; those inputs have linear profiles for every trace function (Theorem 8.2). For affine matrix translation families, a quartic sum of squares decides exact equality for the whole family (Theorem 8.3). This criterion applies to every homogeneous occupation law with the stated translation identities.

Extremal matroid walks and spectral separation

For the spectral statements, let MM be a loopless rank-rr matroid with at least two bases, r≥1r \ge1, and let μ\mu be uniform on B(M)\mathcal{B}(M). The down-up walk PduP_{\mathrm{du}} removes one uniformly chosen element and chooses a uniform basis completing the retained face, allowing the removed element to return. Write CC for the parallel classes. The basis graph GG joins bases differing by one exchange, and its positive unit-edge Laplacian is

(LGg)(B)=∑B′∼B(g(B)−g(B′)).(L_Gg)(B) = \sum_{B' \sim B}\left(g(B)-g(B')\right).

The constant-edge Feder–Mihail walk is PFM=I−pLGP_{\mathrm{FM}} = I - pL_G, where 0<p≤1/Δ(G)0 < p \le1/\Delta(G) and Δ(G)\Delta(G) is the maximum degree. We use the gap 1−λ2(P)1-\lambda_2(P), where λ2(P)\lambda_2(P) is the second-largest eigenvalue.

Theorem 1.3 (Complete classification of spectral equality). With these normalizations, the following statements hold.

(1) The down-up gap is at least 1/r1/r, and

ker⁡(Pdu−(1−1/r)I)={B↦∑i∈Bhi:∑i∈Chi=0 for every C}.(2)\ker(P_{\mathrm{du}}-(1-1/r)I)=\left\{B\mapsto\sum_{i\in B}h_i:\sum_{i\in C}h_i=0\text{ for every }C\right\}. \tag*{(2)}

This space has dimension ∑C(∣C∣−1)\sum_C(|C|-1). Consequently the gap equals 1/r1/r if and only if MM has a nontrivial parallel class.

(2) If MM has kk direct summands isomorphic to U1,2U_{1,2}, then

Spec⁡(LG)⊆{0,2}∪[3,∞),dim⁡ker⁡(LG−2I)=k.(3)\operatorname{Spec}(L_G)\subseteq\{0,2\}\cup[3,\infty),\qquad\dim\ker(L_G-2I)=k. \tag*{(3)}

The eigenvalue-two space is spanned by the kk binary contrasts 1{a∈B}−1{b∈B}\mathbf{1}_{\{a\in B\}}-\mathbf{1}_{\{b\in B\}} on these summands. The Feder–Mihail gap equals 2p2p exactly when k>0k>0; if k=0k=0, it is at least 3p3p. The threshold three is sharp.

Part (1) proves more than the existence of a degree-one extremizer asked for in [2 Conjecture 8.1]: it identifies every extremizer. It gives an exact criterion on the original matroid for the structural question in [2, Conjecture 8.2]. Part (2) answers [2 Problem 8.4]. Parallel elements alone suffice for down-up equality; the unit-edge walk singles out binary direct components. The proofs appear in Corollaries 5.8 and 10.11 and Theorem 10.7.

The extremal-space theorem holds at every proper retained level for all positive-field Lorentzian laws: the modes are precisely the Hessian-null linear counts. A weighted upper-star inequality and parallel-class decomposition give the unit-edge separation; triangles attain the threshold three.

The same decomposition expresses all retained-level matroid walks through contraction minors and shows that scalar equality tilts preserve their spectra simultaneously (Corollary 9.5). The general splitting theorem uses derivative laws, permits repeated particles, and needs no Lorentzian hypothesis (Theorem 9.3).

A general matrix Laplace theorem

Let LL be the positive generator of a finite irreducible reversible chain with invariant law μ\mu and rates q(x,y)q(x,y). For a Hermitian observable AA, define

ΓL(A)(x)=12∑yq(x,y)(A(x)−A(y))2.\Gamma_L(A)(x)=\frac{1}{2}\sum_y q(x,y)(A(x)-A(y))^2.

Suppose that α>0\alpha>0 satisfies the complete entropy-production inequality

αHμ(F)≤EμTr⁡[(LF)log⁡F](F>0).\alpha H_\mu(F)\leq E_\mu\operatorname{Tr}[(LF)\log F]\qquad(F>0).

Theorem 1.4 (Matrix Laplace comparison). If ΓL(A)(x)⪯V\Gamma_L(A)(x)\preceq V for a fixed matrix V⪰0V\succeq0 and every xx, then for every Hermitian offset KK and every real θ\theta,

EμTr⁡eK+θA≤Tr⁡exp⁡(K+θEμA+θ2αV).(4)E_\mu\operatorname{Tr}e^{K+\theta A}\leq\operatorname{Tr}\exp\left(K+\theta E_\mu A+\frac{\theta^2}{\alpha}V\right). \tag*{(4)}

References

  1. [1]R. Adamczak and I. Kavvadiас, Matrix concentration inequalities for dependent binary random variables, preprint (2025). arXiv:2504.08138.arxiv.org/abs/2504.08138
  2. [2]L. Alaniz Pintos, Exact spectral gap of the down-up walk on matroids with parallel elements, preprint (2026). https://doi.org/10.5281/zenodo.21745073.
  3. [3]V. L. Alev and L. C. Lau, Improved analysis of higher order random walks and applications, Proceedings of STOC 2020, 1198–1211. https://doi.org/10.1145/3357713.3384317. https://arxiv.org/abs/2001.02827.
  4. [4]N. Anari, V. Jain, F. Koehler, H. T. Pham, and T.-D. Vuong, Entropic independence I: modified log-Sobolev inequalities for fractionally log-concave distributions and high-temperature Ising models, arXiv:2106.04105v2, 5 November 2021. https://arxiv.org/abs/2106.04105v2.
  5. [5]N. Anari, K. Liu, S. Oveis Gharan, and C. Vinzant, Log-concave polynomials II: high-dimensional walks and an FPRAS for counting bases of a matroid, arXiv:1811.01816v3 (2019). https://arxiv.org/abs/1811.01816v3.
  6. [6]N. Anari, K. Liu, S. Oveis Gharan, and C. Vinzant, Log-concave polynomials III: Mason’s ultra-log-concavity conjecture for independent sets of matroids, Proceedings of the American Mathematical Society 152 (2024), no. 5, 1969–1981. https://doi.org/10.1090/proc/16724. https://arxiv.org/abs/1811.01600v1. Theorem numbering follows arXiv version 1.
  7. [7]N. Anari, S. Oveis Gharan, and C. Vinzant, Log-concave polynomials I: entropy and a deterministic approximation algorithm for counting bases of matroids, Duke Mathematical Journal 170 (2021), 3459–3504. https://arxiv.org/abs/1807.00929.
  8. [8]I. Bengtsson, Å. Ericsson, M. Kuś, W. Tadej, and K. Życzkowski, Birkhoff’s polytope and unistochastic matrices, N = 3 and N = 4, Communications in Mathematical Physics 259 (2005), 307–324. https://doi.org/10.1007/s00220-005-1392-8. https://arxiv.org/abs/math/0402325v2.
  9. [9]A. J. Bernstein, Maximally connected arrays on the n-cube, SIAM Journal on Applied Mathematics 15 (1967), 1485–1489. https://doi.org/10.1137/0115129.
  10. [10]M. Berta, D. Sutter, and M. Walter, Quantum Brascamp–Lieb dualities, Communications in Mathematical Physics 401 (2023), 1807–1830. doi:10.1007/s00220-023-04678-w. arXiv:1909.02383.DOI
  11. [11]J. Borcea and P. Brändén, The Lee–Yang and Pólya–Schur programs. I. Linear operators preserving stability, Inventiones Mathematicae 177 (2009), 541–569. https://arxiv.org/abs/0809.0401v3. Proposition numbering follows arXiv version 3.
  12. [12]P. Brändén, Obstructions to determinantal representability, Advances in Mathematics 226 (2011), 1202–1212. https://doi.org/10.1016/j.aim.2010.08.003. https://arxiv.org/abs/1004.1382.
  13. [13]P. Brändén and J. Huh, Lorentzian polynomials, Annals of Mathematics 192 (2020), 821–891. https://arxiv.org/abs/1902.03719.
  14. [14]E. A. Carlen and E. H. Lieb, Remainder terms for some quantum entropy inequalities, Journal of Mathematical Physics 55 (2014), 042201. https://doi.org/10.1063/1.4871575. https://arxiv.org/abs/1402.3840v2.
  15. [15]Q. Chen and R. W. Yeung, Partition-symmetrical entropy functions, IEEE Transactions on Information Theory 62 (2016), no. 10, 5385–5402. https://doi.org/10.1109/TIT.2016.2600580. https://arxiv.org/abs/1407.7405v2.
  16. [16]Y.-B. Choe, J. G. Oxley, A. D. Sokal, and D. G. Wagner, Homogeneous multivariate polynomials with the half-plane property, Advances in Applied Mathematics 32 (2004), 88–187. https://doi.org/10.1016/S0196-8858(03)00078-2. https://arxiv.org/abs/math/0202034.
  17. [17]M. Cryan, H. Guo, and G. Mousa, Modified log-Sobolev inequalities for strongly log-concave distributions, Annals of Probability 49 (2021), 506–525. https://arxiv.org/abs/1903.06081.
  18. [18]G. De Palma and C. Rouzé, Quantum concentration inequalities, Annales Henri Poincaré 23 (2022), 3391–3429. https://doi.org/10.1007/s00023-022-01181-1. https://arxiv.org/abs/2106.15819.
  19. [19]P. Diaconis and R. C. Griffiths, An introduction to multivariate Krawtchouk polynomials and their applications, Journal of Statistical Planning and Inference 154 (2014), 39–53. https://arxiv.org/abs/1309.0112.
  20. [20]D. Ellis, N. Keller, and N. Lifshitz, On a biased edge isoperimetric inequality for the discrete cube, arXiv:1702.01675v2 (2018). https://arxiv.org/abs/1702.01675v2.
  21. [21]L. Gao, M. Junge, and N. LaRacuente, Fisher information and logarithmic Sobolev inequality for matrix-valued functions, Annales Henri Poincaré 21 (2020), 3409–3478. https://doi.org/10.1007/s00023-020-00947-9. https://arxiv.org/abs/1807.08838.
  22. [22]L. Gao, M. Junge, N. LaRacuente, and H. Li, Complete positivity order and relative entropy decay, Forum of Mathematics, Sigma 13 (2025), e31, 1–51. https://doi.org/10.1017/fms.2024.117. https://arxiv.org/abs/2209.11684.
  23. [23]A. Garg, T. Kathuria, and N. Srivastava, Scalar Poincaré implies matrix Poincaré, preprint (2020). https://arxiv.org/abs/2006.09567.
  24. [24]L. H. Harper, Optimal assignments of numbers to vertices, Journal of the Society for Industrial and Applied Mathematics 12 (1964), 131–135. https://doi.org/10.1137/0112012.
  25. [25]S. Hart, A note on the edges of the n-cube, Discrete Mathematics 14 (1976), 157–163. https://doi.org/10.1016/0012-365X(76)90058-3.
  26. [26]N. J. A. Harvey and N. Olver, PiPage rounding, pessimistic estimators and matrix concentration, Proceedings of SODA 2024, 926–945. https://doi.org/10.1137/1.9781611973402.69. https://arxiv.org/abs/1307.2274.
  27. [27]P. Hayden, R. Jozsa, D. Petz, and A. Winter, Structure of states which satisfy strong subadditivity of quantum entropy with equality, Communications in Mathematical Physics 246 (2004), 359–374. https://doi.org/10.1007/s00220-004-1049-z. https://arxiv.org/abs/quant-ph/0304007v2.
  28. [28]D. Huang and J. A. Tropp, Nonlinear matrix concentration via semigroup methods, Electronic Journal of Probability 26 (2021), paper no. 8, 1–31. https://doi.org/10.1214/20-EJP578. https://arxiv.org/abs/2006.16562.
  29. [29]T. Kathuria, A matrix Bernstein inequality for strong Rayleigh distributions, preprint (2020). https://arxiv.org/abs/2011.13340.
  30. [30]T. Kaufman, R. Kyng, and F. Soldá, Scalar and matrix Chernoff bounds from ℓ∞-independence, Proceedings of SODA 2022, 3732–3753. arXiv:2111.02177.arxiv.org/abs/2111.02177
  31. [31]J. Kurihara, T. Uyematsu, and R. Matsumoto, Secret sharing schemes based on linear codes can be precisely characterized by the relative generalized Hamming weight, IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences E95-A (2012), no. 11, 2067–2075. https://doi.org/10.1587/transfun.E95.A.2067.
  32. [32]R. Kyng and Z. Song, A matrix Chernoff bound for strongly Rayleigh distributions and spectral sparsifiers from a few random spanning trees, Proceedings of FOCS 2018, 373–384. arXiv:1810.08345.arxiv.org/abs/1810.08345
  33. [33]J. Lee and J. Seo, Antiferromagnetic models are clique-minimizing, arXiv:2608.17920 (2026). https://arxiv.org/abs/2608.17920.
  34. [34]E. H. Lieb, Convex trace functions and the Wigner–Yanase–Dyson conjecture, Advances in Mathematics 11 (1973), 267–288. https://doi.org/10.1016/0001-8708(73)90011-X.
  35. [35]E. H. Lieb and M. B. Ruskai, Proof of the strong subadditivity of quantum-mechanical entropy, J. Math. Phys. 14 (1973), 1938–1941. https://doi.org/10.1063/1.1666274.
  36. [36]L. Mackey, M. I. Jordan, R. Y. Chen, J. Farrell, and J. A. Tropp, Matrix concentration inequalities via the method of exchangeable pairs, Annals of Probability 42 (2014), 906–945. https://doi.org/10.1214/13-AOP892. https://arxiv.org/abs/1201.6002.
  37. [37]R. Montenegro and J.-B. Son, Edge isoperimetry and rapid mixing on matroids and geometric Markov chains, Proceedings of STOC 2001, 704–711. https://doi.org/10.1145/380752.380876.
  38. [38]S. Murai, T. Nagaoka, and A. Yazawa, Strictness of the log-concavity of generating polynomials of matroids, Journal of Combinatorial Theory, Series A 181 (2021), 105351. https://arxiv.org/abs/2003.09568.
  39. [39]J. A. Tropp, User-friendly tail bounds for sums of random matrices, Foundations of Computational Mathematics 12 (2012), 389–434. arXiv:1004.4389.DOI
  40. [40]C. Wang and A. Gohari, Matrix Φ-entropic measures of dependence, arXiv:2610.29576v1 (2026). https://arxiv.org/abs/2610.29576v1.

Paper details

vlnc:W_fqc36dh8
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT

View statement
Looking for related papers…
Contents