Honeycombs and Sums of Hermitian Matrices, Revisited

Alexander Postnikov; Ankur Moitra; Dora Woodruff

Honeycombs and Sums of Hermitian Matrices, Revisited

T0 review · 0 major / 5 minor · reviewed 2026-07-10 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Spectra of Hermitian sums equal honeycomb boundary positions because both obey the same four axioms.

desk verdict Clean axiomatic re-proof of Knutson–Tao that treats honeycombs as model organisms; frameworks streamline saturation and the combinatorial half is elementary, with only standard symplectic convexity left external. read the letter →

arxiv 2607.06710 v1 pith:TNL4WNVK submitted 2026-07-07 math.CO math.RT

Ankur Moitra , Alexander Postnikov , Dora Woodruff This is my paper · ORCID

classification math.COmath.RT MSC 05E1015A4214M15

keywords honeycombsHermitianmatricesHornproblemeigenvaluesumsBerenstein-Zelevinskypatternssaturationconjectureframeworkssequences

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper gives a new proof that the possible eigenvalues of a sum of two Hermitian matrices are exactly the possible positions of the south boundary rays of a honeycomb with the other two boundaries fixed. It isolates four shared properties—trivial 1-by-1 case, closure under direct sum, convexity of the set of possibilities, and a splitting rule that decomposes extreme points into smaller direct sums—and proves that any two families of objects satisfying these axioms must be identical. Honeycombs (and a closely related combinatorial object called frameworks) satisfy the axioms by elementary geometric and graph-theoretic arguments, while the matrix spectra satisfy them by more analytic means. The identification therefore shows why honeycombs capture the matrix problem so cleanly: they are model organisms that make the same defining rules obvious and accessible. The approach also yields a short combinatorial proof of the saturation conjecture as a byproduct.

What carries the argument

Horn sequences: set-valued maps from pairs of partitions to subsets of partitions that obey four axioms (base case for n=1, direct-sum closure, convexity of each image as a polytope, and splitting of every extreme point into a direct sum of strictly smaller instances). The axioms force uniqueness, so verifying them for both honeycombs/frameworks and Hermitian spectra identifies the two collections.

What would settle it

Produce an explicit pair of spectra λ, μ for which the set of attainable spectra of A+B is non-convex, or exhibit an extreme point of that set that refuses to split into smaller direct-sum Horn triples while the corresponding honeycomb vertex does (or vice versa).

Watch

Extended reading notes

Core claim

For every n and every pair of nonincreasing real n-tuples λ, μ, the set of possible spectra of A+B equals the set of possible south-boundary positions of n-honeycombs (and of frameworks) with northwest and northeast boundaries given by λ and μ. Both families form Horn sequences—they satisfy the four axioms of base case, direct sum, convexity and extreme-point splitting—and any two Horn sequences coincide.

Load-bearing premise

The set of possible spectra of A+B for fixed spectra of A and B is a convex polytope; this fact is taken as a black box from symplectic geometry rather than proved elementarily inside the paper.

Editorial extensions

If this is right

Honeycombs and frameworks become combinatorial sandboxes for studying finer structure of Horn triples and Littlewood-Richardson coefficients.
The same four axioms give a short proof of the saturation conjecture by reducing integer honeycombs to frameworks whose graphs are forests.
Interior vertices of the Horn polytope are realized by permutation honeycombs (simultaneously diagonalizable matrices).
Any new combinatorial model that satisfies the four axioms is automatically equivalent to the classical Horn sets.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The axiomatic template could certify new models for related spectral problems (products of unitaries, singular-value inequalities) simply by checking the same four properties.
Because frameworks are non-planar 3-valent bipartite graphs, they may connect the Horn problem to scattering diagrams already used in other areas of mathematical physics.
An elementary proof of convexity for the matrix side would make the entire identification purely combinatorial and remove the last external black box.
The paper’s boundary-point splitting (stronger than vertex splitting) supplies a recursive description of all faces of the Horn polytope.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The punchline is a four-axiom characterization (base case, direct sum, convexity, splitting) that forces any two “Horn sequences” to coincide, so Horn_n = Honey_n = Frame_n by induction. That is the actual new contribution: not another proof of the classical equality, but a clean reason why honeycombs capture Hermitian spectra at all.

What works well is the combinatorial side. Frameworks (strictly positive-length 3-valent bipartite embeddings) are a genuine intermediate object; they make the breathing argument for saturation shorter and the vertex-splitting axiom for honeycombs almost immediate. Overlay for direct sum and the BZ-polytope projection for convexity are transparent. The inductive equivalence (Prop. 5.2) is elementary and airtight. The Hermitian splitting argument extends Horn’s classical calculation carefully to multiplicities via a closedness argument and the Perron–Frobenius lemma on stochastic matrices; that part is solid matrix calculus.

The only soft spot is the one the paper itself flags: convexity of Horn_n is imported from Kirwan / Atiyah–Guillemin–Sternberg rather than proved elementarily. That is a standard, long-established theorem for the precise coadjoint-orbit images that arise here, so it is not a fragile black box, but it does mean the “model organism” story is not fully self-contained on the Hermitian side. Everything else checks line-by-line with no free parameters or circularity.

This is for people who already know the Horn problem and want a conceptual reorganization, or for combinatorialists looking for a sandbox for finer LR structure. It does not open a major new technology, but the frameworks notion and the axiomatic packaging are worth having. I would send it to a serious referee without hesitation; the central claim holds under the paper’s own terms.

Referee Report

0 major / 5 minor

Summary. The paper gives an axiomatic proof that the sets Horn_n(\lambda,\mu) of possible spectra of A+B for Hermitian A,B with fixed spectra \lambda,\mu coincide with the sets Honey_n(\lambda,\mu) of South-boundary positions of n-honeycombs (and equivalently Frame_n). It isolates four properties (base case, direct sum, convexity, and extreme-point splitting) that define a Horn sequence, proves that any two such sequences are identical by induction on n (Proposition 5.2), verifies the axioms combinatorially for honeycombs/frameworks (Section 6), and verifies them for Horn triples by elementary matrix calculus plus a classical closedness argument, taking convexity of Horn_n as a black-box input from symplectic geometry (Section 7).

Significance. The result re-proves the Knutson–Tao correspondence by a short, largely elementary route that makes the shared structure of honeycombs and Hermitian spectra transparent. The introduction of frameworks (strictly positive-length 3-valent bipartite embeddings) yields a clean combinatorial proof of the splitting axiom and a concise re-proof of the Saturation Conjecture (Theorem 4.14). The axiomatic characterization is a genuine conceptual contribution: once the four properties are accepted, the identification is formal. Convexity for Horn triples remains an external input, but it is a standard theorem for the coadjoint-orbit images that arise, so the argument is not circular. The paper therefore supplies both a new proof and a useful “model-organism” perspective for further structural questions about Horn triples and LR coefficients.

minor comments (5)

The date line “July 9, 2026” and the arXiv identifier 2607.06710 appear to be placeholders; they should be corrected before publication.
Section 7.1 cites Kirwan and Atiyah–Guillemin–Sternberg for convexity of Horn_n but does not give a precise reference for the fact that the image under the moment map of the product of coadjoint orbits is exactly the set defined by unitary conjugacy in equation (3). A one-sentence pointer would help readers who are not specialists in symplectic geometry.
In the proof of Theorem 6.3 the choice of \epsilon is described informally (“sufficiently small”); an explicit bound in terms of minimal edge lengths and gaps of \nu (as later given for the multiple-path case) would make the argument fully self-contained.
Figure 11 is referenced for path breathing but the caption does not indicate which edges change length; a short annotation would improve readability.
Typographical slips: “Knuton-Tau” (p. 3), “tero-tension” (p. 7), “coinsiding” (p. 15), and the repeated “m-n” for n-m in the inductive step of Proposition 5.2.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: independent axiom checks for Horn triples and honeycombs/frameworks, followed by formal induction; convexity is an external black box, not a self-referential reduction.

full rationale

The central claim (Horn_n(λ,μ)=Honey_n(λ,μ)=Frame_n(λ,μ)) is obtained by verifying that both families are Horn sequences (Definition 5.1) and invoking the purely formal Proposition 5.2 that any two Horn sequences coincide. Base case and direct-sum axioms are elementary for both sides (matrix block-diagonalization; overlay of zero-tension diagrams). Convexity for honeycombs follows at once from the linear inequalities defining Berenstein–Zelevinsky patterns (Proposition 6.1); for Horn triples it is imported as a black-box theorem of symplectic geometry (Kirwan, Atiyah–Guillemin–Sternberg) that is independent of honeycombs and of the present authors. The splitting axiom is proved combinatorially for frameworks by path-breathing (Theorem 6.3) and for Hermitian spectra by explicit differentiation of the unitary orbit map plus the classical Perron–Frobenius lemma on stochastic matrices (Lemmas 7.1–7.19 and the multiplicity-handling argument of 7.5). None of these steps defines one side in terms of the other, fits a parameter to the target, or relies on a load-bearing self-citation of an unverified uniqueness claim. Background citations (KT99, BZ92, GP00) merely supply the standard definitions of honeycombs and BZ patterns; they are not used to smuggle the equality itself. The derivation is therefore self-contained against its own axioms and free of circular reduction.

Assumptions & free parameters 0 free parameters · 5 assumptions · 2 invented entities

The central claim rests on four explicitly stated axioms plus standard linear algebra and one external convexity theorem. No free parameters appear. Frameworks are an invented intermediate entity with clear combinatorial definition and independent combinatorial evidence. Convexity for Hermitian spectra is the sole non-elementary domain assumption.

assumptions (5)

standard math Base case: for n=1, f_1(λ,μ)={λ+μ}
Immediate for both 1 imes1 matrices and single Y-honeycombs (Def. 5.1(i)).
standard math Direct-sum axiom: ν∈f_n(λ,μ) and ν'∈f_m(λ',μ') imply ν⊕ν'∈f_{n+m}(λ⊕λ',μ⊕μ')
Block-diagonal matrices for Horn; overlay of zero-tension diagrams for honeycombs (Def. 5.1(ii)).
domain assumption Convexity: f_n(λ,μ) is a convex polytope
Proved combinatorially for honeycombs via BZ polytopes (Prop. 6.1); for Horn triples imported from Kirwan/AGS symplectic geometry (Section 7.1).
ad hoc to paper Splitting: every extreme point of f_n(λ,μ) decomposes as a direct sum of smaller instances
Stated as Def. 5.1(iv); verified for frameworks by path-breathing (Thm. 6.3) and for Horn by tangent-space rank + Perron–Frobenius (Thm. 7.5).
standard math Any two sequences satisfying the four axioms are identical
Elementary induction on n using Krein–Milman (Prop. 5.2).

invented entities (2)

Frameworks (strictly positive-length 3-valent bipartite embeddings) independent evidence
purpose: Provide a flexible combinatorial model that makes the splitting axiom transparent and shortens the saturation proof.
Defined in Def. 4.8; shown equivalent in measure to honeycombs (Prop. 4.9) but not one-to-one; used crucially in Thm. 6.3 and the saturation argument of Section 4.4.
Horn sequence (sequence of set-valued maps satisfying the four axioms) independent evidence
purpose: Abstract common structure so that equality of any two such sequences follows by induction.
Definition 5.1; the paper verifies that both Horn_n and Honey_n are Horn sequences.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Honeycombs and Sums of Hermitian Matrices, Revisited." pith.science (2026). https://pith.science/paper/TNL4WNVK

@misc{pith2026260706710,
  author       = {Pith},
  title        = {Pith review of: Honeycombs and Sums of Hermitian Matrices, Revisited},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TNL4WNVK}},
  note         = {Machine review of arXiv:2607.06710}
}

abstract

We give a new proof of the celebrated theorem of Knutson and Tao that the spectra of triples $A, B, A+B$ of Hermitian matrices exactly correspond to positions of boundary rays of honeycombs. Most importantly, our proof gives new insights into why honeycombs are related to Hermitian matrices in the first place. Our proof is axiomatic: We distill four essential properties shared by honeycombs and spectra of Hermitian triples, and show that any two objects sharing these four properties must be equivalent. In this way, we argue that honeycombs are `model organisms' for Hermitian triples: they are families of objects satisfying the same defining properties, but in more obvious ways.

Figures

Figures reproduced from arXiv: 2607.06710 by the authors.

**Figure 1.** Compass directions in R 3 Σ=0 As in the introduction, we will call a triple (λ, µ, ν) of spectra a Horn triple if there exist Hermitian matrices A, B, A + B with spectra λ, µ, ν. Definition 3.1. For λ, µ ∈ Cn, Hornn(λ, µ) is the set of all ν ∈ Cn such that (λ, µ, ν) is a Horn triple. Consider the plane R 3 Σ=0 := {(x, y, z) ∈ R 3 | x + y + z = 0}. Inside the plane R 3 Σ=0 we identify six distinguished vectors, which… view at source ↗

**Figure 2.** From left to right are the graphs H1, H2, H3, H4, and H5. segments [ϕ(u), ϕ(v)] on the plane, and the map from boundary rays {v} ∈ E to half-infinite rays emanating from points ϕ(v). Note that we allow different nodes of G to map to the same point on the plane; and we allow some edges of G to map to line segments of length zero. For a positive integer n ∈ N, the honeycomb graph Hn is a certain 3-valent bipartite gra… view at source ↗

**Figure 3.** An embedding of H5 in which all edge lengths le are nonzero, with associated partitions (λ, µ, ν). Boundary rays are labeled by their positions. There are two (linearly equivalent) ways to parametrize honeycombs. We can parametrize an n-honeycomb by the array of positions (pe) assigned to all edges e of Hn (including the boundary rays). Alternatively, for given λ, µ, ν, a honeycomb is uniquely determined by the arra… view at source ↗

Figures from the paper (9 more)

**Figure 4.** Figure 4: Image from [KT99]. Up to rotations, all the possible small neighborhoods around [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: On the left is a honeycomb satisfying Definition 4.6 (a zero-tension diagram). A [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

**Figure 6.** Figure 6: A framework with colored nodes; we give the nodes two colors to emphasize the fact [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

**Figure 7.** Figure 7: Breaking up a neighborhood of a zero-tension diagram into edges and trivalent nodes [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

**Figure 8.** Figure 8: An example of a cycle within a framework which can ‘breathe,’ transforming into the [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

**Figure 9.** Figure 9: A permutation honeycomb. Here ν1 = λ2 + µ1, ν2 = λ1 + µ3, ν3 = λ3 + µ2 a Y , λ2 is matched to µ1, and λ3 is matched to µ2, so this honeycomb is associated to the permutation (σ(1), σ(2), σ(3)) = (3, 1, 2). Why are permutation honeycombs important? First, in Section 6, …

**Figure 10.** Figure 10: The 5-honeycomb shown on the right is the [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]

**Figure 11.** Figure 11: (Here ei is the ith standard basis vector of R n .) Let us make this more precise. Pick a way to direct the path P. If an edge of P points South, Northwest, or Northeast, increase its position (i.e., the constant coordinate of the corresponding line segment in R 3 P=0…

**Figure 11.** Figure 11: A framework for n = 4 with a highlighted path connecting −ν1 to −ν4 that can breathe back and forth. on the boundary of the fundamental Weyl chamber Cn. Now there is a complication in our strategy: namely, as we ‘breathe path’ by ϵ and deform νi and νj as above, then …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 28 canonical work pages

[1]

and Tao, T

Knutson, A. and Tao, T. The honeycomb model of GL_n( C ) tensor products I: Proof of the saturation conjecture. J. Am. Math. Soc. 1999. doi:10.1090/s0894-0347-99-00299-4

work page doi:10.1090/s0894-0347-99-00299-4 1999
[2]

and Tao, T

Knutson, A. and Tao, T. and Woodward, C. The honeycomb model of GL_n( C ) tensor products II: Puzzles determine facets of the Littlewood-Richardson Cone. J. Am. Math. Soc. 2003

work page 2003
[3]

Stable bundles, representation theory and Hermitian operators

Klyachko, A. Stable bundles, representation theory and Hermitian operators. Sel. math. 1998. doi:10.1007/s000290050037

work page doi:10.1007/s000290050037 1998
[4]

Eigenvalues of sums of Hermitian matrices

Horn, A. Eigenvalues of sums of Hermitian matrices. Pacific Journal of Mathematics. 1962

work page 1962
[5]

Eigenvalues, invariant factors, highest weights, and Schubert calculus

Fulton, W. Eigenvalues, invariant factors, highest weights, and Schubert calculus. Bull. Amer. Math. Soc. 2000. doi:10.1090/S0273-0979-00-00865-X

work page doi:10.1090/s0273-0979-00-00865-x 2000
[6]

Horn's problem, Vinnikov curves, and the hive cone

Speyer, D. Horn's problem, Vinnikov curves, and the hive cone. Duke Math. J. 2005. doi:10.1215/S0012-7094-04-12731-0

work page doi:10.1215/s0012-7094-04-12731-0 2005
[7]

Linear Algebra and its Applications , volume =

The symplectic and algebraic geometry of. Linear Algebra and its Applications , volume =. 2000 , issn =. doi:10.1016/S0024-3795(00)00220-2 , author =

work page doi:10.1016/s0024-3795(00)00220-2 2000
[8]

and Zelevinsky, A

Berenstein, A. and Zelevinsky, A. Triple Multiplicities for sl(r+1) and the spectrum of the exterior algebra of the adjoint representation. Journal of Algebraic Combinatorics. 1992. doi:doi.org/10.1023/A:1022429213282

work page doi:10.1023/a:1022429213282 1992

Show all 28 references

[9]

Geometric proofs of Horn and saturation conjectures

Belkale, P. Geometric proofs of Horn and saturation conjectures. Journal of Algebraic Geometry. 2006. doi:10.17615/fs6e-0c86

2006 doi
[10]

and Weyman, J

Derksen, H. and Weyman, J. Semi-invariants of quivers and saturation for Littlewood-Richardson coefficients. Journal of the American Mathematical Society. 2000

2000
[11]

and Postnikov, A

Gleizer, O. and Postnikov, A. Littlewood-Richardson coefficients via Yang-Baxter equation. International Mathematics Research Notices. 2000

2000
[12]

Postnikov , title =

A. Postnikov , title =. arXiv:math/0609764. 2006 , eprint =

2006 arXiv
[13]

and Bourjaily, J

Arkani-Hamed, N. and Bourjaily, J. and Cachazo, F. and Goncharov, A. and Postnikov, A. and Trnka, J. Grassmannian Geometry of Scattering Amplitudes. 2016

2016
[14]

Speyer, D. , year=. L-log-concavity and a proof of the conjecture of. 2601.05007 , archivePrefix=

arXiv
[15]

and Postnikov, A

Lam, T. and Postnikov, A. and Pylyavskyy, P. Schur Positivity and Schur Log-Concavity. American Journal of Mathematics. 2007

2007
[16]

Buch, A. S. and Kresch, A. Purbhoo, K. and Tamvakis, H. The puzzle conjecture for the cohomology of two-step flag manifolds. Journal of Algebraic Combinatorics. 2016

2016
[17]

and Tao, T

Knutson, A. and Tao, T. Honeycombs and sums of Hermitian matrices. Notices of the AMS. 2001

2001
[18]

Kirwan, F. C. Cohomology of Quotients in Complex and Algebraic Geometry. 1984

1984
[19]

On a theorem of Weyl concerning eigenvalues of linear transformations

Fan, K. On a theorem of Weyl concerning eigenvalues of linear transformations. Proc. Nat. Acad. Sci. USA. 1949

1949
[20]

Lidskii, V. B. The proper values of the sum and product of symmetric matrices. Dokl. Akad. Nauk SSSR. 1950

1950
[21]

Thompson, R. C. An inequality for invariant factors. Proc. Amer. Math. Soc. 1982

1982
[22]

An extremum property of sums of eigenvalues

Weilandt, H. An extremum property of sums of eigenvalues. Proc. Amer. Math. Soc. 1955

1955
[23]

Tensor products of semistables are semistable

Totaro, B. Tensor products of semistables are semistable. Geometry and Analysis on Complex Manifolds. 1994

1994
[24]

Ueber Matrizen aus nicht negativen Elementen

Frobenius, G. Ueber Matrizen aus nicht negativen Elementen. Sitzungsberichte der Koniglich Preussichen Akademie der Wissenschaften. 1912

1912
[25]

Lectures on symplectic geometry

Cannas da Silva, A. Lectures on symplectic geometry. 2001

2001
[26]

Deciding Positivity of

B\". Deciding Positivity of. SIAM Journal on Discrete Mathematics , volume =. 2013 , doi =

2013
[27]

Mulmuley, K. D. Narayanan, H. and Sohoni, M. , title =. J. Algebr Comb , volume =. 2012 , doi =

2012
[28]

, title =

Franks, C. , title =. 2018 , isbn =. doi:10.1145/3188745.3188932 , booktitle =

2018 doi

Pith tools

Reviewed July 10, 2026 · model on record in the stance chip above.

[1] [1]

and Tao, T

Knutson, A. and Tao, T. The honeycomb model of GL_n( C ) tensor products I: Proof of the saturation conjecture. J. Am. Math. Soc. 1999. doi:10.1090/s0894-0347-99-00299-4

work page doi:10.1090/s0894-0347-99-00299-4 1999

[2] [2]

and Tao, T

Knutson, A. and Tao, T. and Woodward, C. The honeycomb model of GL_n( C ) tensor products II: Puzzles determine facets of the Littlewood-Richardson Cone. J. Am. Math. Soc. 2003

work page 2003

[3] [3]

Stable bundles, representation theory and Hermitian operators

Klyachko, A. Stable bundles, representation theory and Hermitian operators. Sel. math. 1998. doi:10.1007/s000290050037

work page doi:10.1007/s000290050037 1998

[4] [4]

Eigenvalues of sums of Hermitian matrices

Horn, A. Eigenvalues of sums of Hermitian matrices. Pacific Journal of Mathematics. 1962

work page 1962

[5] [5]

Eigenvalues, invariant factors, highest weights, and Schubert calculus

Fulton, W. Eigenvalues, invariant factors, highest weights, and Schubert calculus. Bull. Amer. Math. Soc. 2000. doi:10.1090/S0273-0979-00-00865-X

work page doi:10.1090/s0273-0979-00-00865-x 2000

[6] [6]

Horn's problem, Vinnikov curves, and the hive cone

Speyer, D. Horn's problem, Vinnikov curves, and the hive cone. Duke Math. J. 2005. doi:10.1215/S0012-7094-04-12731-0

work page doi:10.1215/s0012-7094-04-12731-0 2005

[7] [7]

Linear Algebra and its Applications , volume =

The symplectic and algebraic geometry of. Linear Algebra and its Applications , volume =. 2000 , issn =. doi:10.1016/S0024-3795(00)00220-2 , author =

work page doi:10.1016/s0024-3795(00)00220-2 2000

[8] [8]

and Zelevinsky, A

Berenstein, A. and Zelevinsky, A. Triple Multiplicities for sl(r+1) and the spectrum of the exterior algebra of the adjoint representation. Journal of Algebraic Combinatorics. 1992. doi:doi.org/10.1023/A:1022429213282

work page doi:10.1023/a:1022429213282 1992