clopen.solutions
OP-00087 Lean verified

Polynomial-size vertex test for inverse M-matrix intervals

Submit a result

For $L, U \in \mathbb{R}^{n \times n}$ with $L \le U$ entrywise, let $\mathcal A = \{A \in \mathbb{R}^{n \times n} : L \le A \le U\}$, its entries varying independently and $L_{ij} = U_{ij}$ allowed, and let $C = (L+U)/2$ and $R = (U-L)/2$. Let $D_i$ be the diagonal matrix with $-1$ in position $i$ and $1$ in the other diagonal positions. An inverse M-matrix is the inverse of a nonsingular M-matrix. Johnson and Smith showed that every member of $\mathcal A$ is an inverse M-matrix if and only if every vertex matrix (every member with each entry equal to $L_{ij}$ or $U_{ij}$) is, and Garloff, Al-Saafin and Adm reduced this test to at most $2^{n-1}$ vertex matrices.

Problem. For every $n \ge 1$ and all such $L$ and $U$, is every member of $\mathcal A$ an inverse M-matrix if and only if $C + sD_iRD_j$ is an inverse M-matrix for all $i, j \in \{1, \dots, n\}$ and $s \in \{-1, 1\}$?

Source: Hladík (2020), Conjecture 1; first posted as arXiv:1711.08732 (2017).

Imported from github.com/ajt60gaibb/OpenProblemsInNLA.

Classification

Primary
NA — Numerical Analysis
Secondary
RA
Keywords
interval matrices, inverse M-matrices, OpenProblemsInNLA

References

@incollection{hladik2020,
  author        = {Hlad{\'\i}k, Milan},
  title         = {An Overview of Polynomially Computable Characteristics of Special Interval Matrices},
  booktitle     = {Beyond Traditional Probabilistic Data Processing Techniques: Interval, Fuzzy etc. Methods and Their Applications},
  editor        = {Kosheleva, Olga and Shary, Sergey P. and Xiang, Gang and Zapatrin, Roman},
  series        = {Stud. Comput. Intell.},
  volume        = {835},
  publisher     = {Springer},
  address       = {Cham},
  year          = {2020},
  pages         = {295--310},
  doi           = {10.1007/978-3-030-31041-7_16},
  eprint        = {1711.08732},
  archivePrefix = {arXiv},
}
@article{johnson2002,
  author  = {Johnson, Charles R. and Smith, Ronald L.},
  title   = {Intervals of inverse {M}-matrices},
  journal = {Reliab. Comput.},
  volume  = {8},
  number  = {3},
  year    = {2002},
  pages   = {239--243},
  doi     = {10.1023/A:1015517430475},
}
@article{garloff2021,
  author  = {Garloff, J{\"u}rgen and Al-Saafin, Doaa and Adm, Mohammad},
  title   = {Further matrix classes possessing the interval property},
  journal = {Reliab. Comput.},
  volume  = {28},
  year    = {2021},
  pages   = {56--70},
  url     = {https://reliable-computing.org/reliable-computing-28-pp-056-070.pdf},
}
@misc{openproblemsinnla-iv03,
  author       = {Townsend, Alex},
  title        = {{IV-03}: Polynomial-size vertex test for inverse M-matrix intervals},
  howpublished = {Open Problems in Numerical Linear Algebra, GitHub repository},
  year         = {2026},
  url          = {https://github.com/ajt60gaibb/OpenProblemsInNLA/blob/main/intervals-and-absolute-value-equations/IV-03/README.md},
  note         = {Read at commit e375a6f, 2026-09-30},
}

Canonical Lean statement

A submission resolves this problem by proving the constant Problem_OP_00087.

namespace RegistryProblems.OP_00087
noncomputable section

abbrev Mat (n : ℕ) := Matrix (Fin n) (Fin n) ℝ

/-- An actual nonsingular, entrywise nonnegative matrix whose inverse has
nonpositive off-diagonal entries. No regularity of a whole interval is assumed. -/
def IsInverseM {n : ℕ} (A : Mat n) : Prop :=
  IsUnit A ∧ (∀ i j, 0 ≤ A i j) ∧ ∀ i j, i ≠ j → A⁻¹ i j ≤ 0

def OrderedEndpoints {n : ℕ} (L U : Mat n) : Prop := ∀ i j, L i j ≤ U i j

def intervalFamily {n : ℕ} (L U : Mat n) : Set (Mat n) :=
  {A | ∀ i j, L i j ≤ A i j ∧ A i j ≤ U i j}

def center {n : ℕ} (L U : Mat n) : Mat n := fun i j => (L i j + U i j) / 2
def radius {n : ℕ} (L U : Mat n) : Mat n := fun i j => (U i j - L i j) / 2

def signVector {n : ℕ} (i : Fin n) (k : Fin n) : ℝ := if k = i then -1 else 1

/-- Entrywise expression for C+s D_i R D_j, with the exact original D_i. -/
def vertex {n : ℕ} (L U : Mat n) (s : ℝ) (i j : Fin n) : Mat n :=
  fun k l => center L U k l + s * signVector i k * radius L U k l * signVector j l

/-- OP-00087: Polynomial-size vertex test for inverse M-matrix intervals -/
def _root_.RegistryProblems.Problem_OP_00087 : Prop :=
  ∀ n : ℕ, 1 ≤ n → ∀ L U : Mat n, OrderedEndpoints L U →
    ((∀ A ∈ intervalFamily L U, IsInverseM A) ↔
      ∀ i j : Fin n, ∀ s ∈ ({(-1 : ℝ), 1} : Set ℝ), IsInverseM (vertex L U s i j))

end
end RegistryProblems.OP_00087

Formalised by Sidney Holden (Apache-2.0); reviewed by clemens-admin.

Available to match against in: Lean v4.33.1 with Mathlib v4.33.1.

Resolving this problem in Lean

Add the problems package as a dependency, import this problem's module, and prove the constant. Any published revision works.

In lakefile.toml:

[[require]]
name = "RegistryProblems"
git = "https://github.com/cthalhammer/registry-problems"
rev = "753bebc58e3fee845f4fffec7d24e7c8cc2a689a"

In your proof:

import RegistryProblems.OP_00087

theorem resolution : RegistryProblems.Problem_OP_00087 := by
  sorry

In preprint.toml, mapping the claim to the declaration that establishes it:

[[statements]]
label = "Theorem 1"
lean = "resolution"
latex = "..."
problem = "OP-00087"
claim = "full"

To disprove it instead, prove the negation and claim "disproof", or "counterexample" if the proof exhibits one:

theorem disproof : ¬ RegistryProblems.Problem_OP_00087 := by
  sorry

You can also write the statement out in full; the verifier checks it matches up to definitional equality. Partial results and reductions are not matched against this statement.

Resolution claims

Sidney Holden

Complete all-dimensional inverse-M interval characterization by n-squared negative-sign vertices, implying the full original two-sign criterion without a regularity premise.

Full resolution · verified · canonical statement matched · posted on GitHub 16 Sep 2026
Matthew J. Colbrook

An interval matrix $[C-R,C+R]$ consists entirely of inverse-M matrices if and only if the $n^2$ vertices $C-D_iRD_j$ are inverse-M matrices, where $D_i$ is the diagonal matrix with $-1$ in position $i$ and $1$ elsewhere. This implies the conjectured $2n^2$-vertex criterion, which also tests $C+D_iRD_j$, and shows that half of those tests are unnecessary. The proof combines induction on principal submatrices, …

Full resolution · not verified · fidelity not established · posted on GitHub 11 Sep 2026

Status history

WhenChangeByReason
2026-10-05 Needs verification → Lean verified — Verified here: OpenProblemsInNLA @ e375a6f, IV-03/lean/.

Citing this problem

This identifier is permanent. It will not change if the problem is solved, reclassified or withdrawn.

OP-00087, "Polynomial-size vertex test for inverse M-matrix intervals", clopen.solutions.