Low-dimensional framework conjecture for the partition graph

From papers

Let Fwn\mathrm{Fw}_n denote the boundary framework, let Lr(n)L_r(n) be the vertices of local simplex dimension rr, and let (n)(n) and (1n)(1^n) be the two antenna vertices. Low-dimensional framework conjecture. For every n4n\ge 4,

FwnL1(n)L2(n),FwnL1(n)={(n),(1n)}.\mathrm{Fw}_n\subseteq L_1(n)\cup L_2(n),\qquad \mathrm{Fw}_n\cap L_1(n)=\{(n),(1^n)\}.

Equivalently, every non-antennal framework vertex has local simplex dimension exactly 22. Computation establishes this confinement for 4n304\le n\le 30 and motivates the expectation that higher simplex layers migrate into the interior rather than the boundary framework; the all-nn assertion remains open.

Progress summary

Open

Computations up to size 30 support the conjecture, but no proof for all sizes has been reported.

The conjecture asserts that, for every n4n\ge 4, framework vertices lie only in the first two local simplex layers, with exactly the two antenna vertices in L1(n)L_1(n). It is formulated as Conjecture 5.5 in a 2026 arXiv paper on simplex stratification of the partition graph.

August 2026 computational evidence

The paper reports exhaustive computation for 4n304\le n\le 30, verifying FwnL1(n)L2(n)\mathrm{Fw}_n\subseteq L_1(n)\cup L_2(n) and FwnL1(n)={(n),(1n)}\mathrm{Fw}_n\cap L_1(n)=\{(n),(1^n)\}. It explicitly states that computation does not establish the all-nn claim, which remains an open conjecture; no proof, counterexample, or verification was found in the retrieved sources.

Current status (as of August 2026): The conjecture is verified computationally through n=30n=30, while the assertion for all n4n\ge 4 remains open.

Sources
Sources & referencesView supporting material

Primary source

Fedor B. Lyudogovskiy, “Simplex Stratification and Phase Boundaries in the Partition Graph”, arXiv:2603.23228 (2026).

Solutions 1

Proof

Proof of the complete low-dimensional framework conjecture

The previously established local corner formulas in arXiv:2603.18696, Corollary 3.4, Proposition 3.7, and Theorem 7.1, are used as known input.

Write

λ=(d1m1,,dkmk),d1>>dk>0,gi=didi+1,dk+1=0.\lambda=(d_1^{m_1},\ldots,d_k^{m_k}), \qquad d_1>\cdots>d_k>0, \qquad g_i=d_i-d_{i+1}, \qquad d_{k+1}=0.

For removable corner ii and addable corner jj, the known corner-incidence matrix is

Mij=11{j=i, mi=1}1{j=i+1, gi=1}.M_{ij} = 1-\mathbf1_{\{j=i,\ m_i=1\}} -\mathbf1_{\{j=i+1,\ g_i=1\}}.

Hence the star capacities are

si=k+11{mi=1}1{gi=1},s_i=k+1-\mathbf1_{\{m_i=1\}}-\mathbf1_{\{g_i=1\}},

and every top capacity is at most kk. Consequently

dimloc(λ)={0,λ=(1),k1,λ=(k,k1,,1), k2,k+1,mi2 and gi2 for some i,k,otherwise.(1)\dim_{\mathrm{loc}}(\lambda)= \begin{cases} 0,&\lambda=(1),\\ k-1,&\lambda=(k,k-1,\ldots,1),\ k\ge2,\\ k+1,&m_i\ge2\text{ and }g_i\ge2\text{ for some }i,\\ k,&\text{otherwise}. \end{cases} \tag{1}

The source's earlier definitions in arXiv:2603.21221, Definitions 2.4, 2.6, and 2.8, identify the framework as

Fwn=MnEnEn,\mathrm{Fw}_n=M_n\cup E_n\cup E_n',

where

Mn={(nj,1j):0jn1},M_n=\{(n-j,1^j):0\le j\le n-1\}, En={(nj,j):1jn/2},En={λ:λEn}.E_n=\{(n-j,j):1\le j\le\lfloor n/2\rfloor\}, \qquad E_n'=\{\lambda':\lambda\in E_n\}.

Assume n4n\ge4. The endpoints (n)(n) and (1n)(1^n) have local dimension one.

Every other hook in MnM_n has two distinct block sizes; its upper block has multiplicity one and its lower block has gap one. Therefore no block simultaneously has multiplicity and gap at least two. Since the only two-block staircase (2,1)(2,1) has size three, (1) shows that every non-antennal hook has dimension exactly two.

For a two-row partition (a,b)(a,b) with a>ba>b, both block multiplicities equal one. Again (1) gives local dimension two. If a=b2a=b\ge2, there is one block with multiplicity two and gap a2a\ge2, so (1) also gives dimension two.

Finally, conjugation is an automorphism of the partition graph and therefore preserves local simplex dimension. Every vertex of EnE_n' consequently has dimension two as well.

Thus

FwnL1(n)L2(n),FwnL1(n)={(n),(1n)}(n4).\boxed{ \mathrm{Fw}_n\subseteq L_1(n)\cup L_2(n), \qquad \mathrm{Fw}_n\cap L_1(n)=\{(n),(1^n)\} \quad(n\ge4). }

Equivalently, every non-antennal framework vertex has local simplex dimension exactly two, proving the full conjecture.

0 endorsements
Shivam Patel ·