A Rank-Count Theory for the Combinatorial Discretizable Distance Geometry Problem

Carlile Lavor; Michael Souza; Wagner da Rocha

REVIEW 3 major objections 3 minor 16 references

A Rank-Count Theory for the Combinatorial Discretizable Distance Geometry Problem

T0 review · 3 major / 3 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read The number of realizations of a Combinatorial DDGP instance is determined by the ranks of two graph-derived binary matrices, provided mirror-separated parameters and a reference solution exist.

desk verdict A genuinely new algebraic framework for counting Combinatorial DDGP realizations, but the main theorem's exactness is packed into the mirror-separation assumption, so the result is conditional until that assumption is shown to hold generically or verified concretely. read the letter →

arxiv 2607.27213 v1 pith:ZGDARY4J submitted 2026-06-01 math.MG math.CO

Michael Souza , Wagner da Rocha , Carlile Lavor This is my paper · ORCID

classification math.MGmath.CO MSC 51K05

keywords CombinatorialDDGPDistanceGeometryProblemsolutioncountingbinarybranchcodesrankformulaaffinespaceoverF2pruningedgespartialreflections

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

This paper claims that for the Combinatorial Discretizable Distance Geometry Problem — a subclass where each new vertex has K predecessors forming a clique, so placements branch two ways — the number of geometric realizations can be computed exactly from the ranks of two binary matrices derived from the vertex-order template and the pruning edges. Previously, counting solutions in the presence of additional distance constraints (pruning edges) was open, because branch reflections no longer act on contiguous suffix intervals. The authors prove that, whenever a reference solution exists and the edge lengths satisfy a 'mirror-separated' genericity condition, the feasible binary branch codes form an affine space over the two-element field F2, so the count is 2^{f + rank[M;V] − rank V}, where f counts 'free' branch decisions. If correct, this turns solution counting from tree search into linear algebra over F2 and unifies the known power-of-two counts for unpruned and DMDGP instances.

What carries the argument

The labeled violation matrix V together with the generator mask matrix M. M sends F2-coefficients of graph-derived generators (cone generators g_q for constrained branch vertices, base generators g_C for predecessor cliques) to net branch masks; V sends the same coefficients to labeled violation patterns indexed by (active edge, mirror clique). The feasible branch-shift space is K_F = M(ker V), and Lemma 1 shows dim K_F = rank[M;V] − rank V. The affine-coset identity Ξ_F = s*_c ⊕ K_F is the load-bearing mechanism: it converts 'which branch flips preserve all pruning distances' into a kernel computation over F2, and the free bits simply multiply the count by 2^f.

What would settle it

Run Branch-and-Prune on a random nondegenerate Combinatorial DDGP instance, collect all feasible branch codes, and compare the set with s*_c ⊕ K_F predicted by the rank formula; any feasible code outside the predicted coset — equivalently, any total count other than 2^{f + rank[M;V] − rank V} — is a concrete counterexample to exactness, showing mirror separation failed for those parameters. A single such instance with exact rational coordinates would settle the matter.

Watch

Extended reading notes

Core claim

The paper's central claim is the Rank-Count Theorem (Theorem 1): for K≥2, if at least one realization exists and the parameters are mirror-separated, then for any reference solution s*, the feasible constrained branch codes satisfy Ξ_F = s*_c ⊕ M(ker V). Equivalently, the feasible branch codes form an affine space over F2, and the total number of realizations is 2^{f + rank[M;V] − rank V}. The argument decomposes branch decisions into constrained and free bits, encodes each partial reflection as a cone or base generator labeled by its mirror clique, and builds a labeled violation matrix V whose rows are pairs (active edge, mirror label). Zero-violation generator combinations provably preserv

Load-bearing premise

The formula stands or falls on the mirror-separation hypothesis: any branch flip outside the algebraically predicted space must break at least one active edge length, because the paper only argues this genericity claim conditionally, not proves it.

Editorial extensions

If this is right

For any ordered Combinatorial DDGP template with pruning edges, the realization count is determined solely by ranks of two binary matrices — no coordinate search over the branch tree is needed to obtain the count.
The count is always a power of two when mirror separation holds and a reference solution exists: each independent feasible shift and each free branch decision doubles the number of realizations.
The framework recovers the classical cases: with no pruning edges, B_c is empty, f = n−K, and the formula gives 2^{n−K}; DMDGP counts appear as the contiguous-order special case.
The labeled violation matrix can be computed purely from the graph template and the set of active edges, so the rank formula is weight-independent and topological in that sense.
The algebraic filter K_F identifies exactly which reflection combinations preserve pruning distances, offering a certificate that could be integrated into Branch-and-Prune solvers to prune without coordinate checks.

Reading between the lines

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

Our inference: if mirror separation holds generically as the paper's Remark 1 suggests, then computing the count is polynomial-time in the size of M and V (rank over F2), meaning exact solution counting for Combinatorial DDGP could become feasible for large instances without enumerating 2^{n−K} branches — a consequence the paper mentions as future work but does not itself prove.
Our inference: the mirror-separation condition is the real empirical question; one could test it by running Branch-and-Prune on random instances and checking whether the feasible set is exactly s*_c ⊕ K_F. If failures occur on a non-negligible set, the 'topological' character of the formula would need qualification.
Our inference: the labeled-violation idea might transfer to other discrete search problems where generators carry group labels, e.g., counting solutions of systems of polynomial equations with reflection symmetries; the rank formula would then be a template for counting without search.
Our inference: because the count is a power of two whenever a reference solution exists, any instance with a number of realizations not of the form 2^m would immediately certify either a mirror-separation failure or a degenerate parameter choice — a cheap empirical falsification check.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 paper builds something real: a labeled-violation matrix and an affine-space theorem for feasible branch codes in Combinatorial DDGP with pruning edges. That's new. Prior counting work covered DMDGP's contiguous orders and the unpruned 2^{n-K} case; this extends to arbitrary predecessor cliques and pruning edges. Lemma 4 is the constructive half, and it's solid: any generator combination in ker V produces a shift that preserves all active distances. I checked the 7-vertex example by hand—the masks, violation rows, ranks (rank[M;V]=4, rank V=2), and the final count of 8 are all internally consistent. The paper is also honest: it states Definition 9 as a hypothesis and flags that it exists to exclude extra solutions.

The soft spot is exactly what the reader's take says: the converse half of Theorem 1 is Definition 9 restated. Mirror separation asserts that every feasible shift lies in K_F = M(ker V), which is precisely the equality the theorem claims. Lemma 4 proves the inclusion K_F ⊆ feasible shifts; the reverse is assumed, not derived. Remark 1 offers a genericity route, but it's conditional on a non-identity criterion for discrepancy functions involving radicals from sphere intersections—neither proved nor checked. And the worked example doesn't enumerate the 128 branch codes to confirm |Ξ|=8, so it doesn't verify mirror separation for that instance either. This is a genuine gap, not a manufactured one.

Still, the gap is clearly disclosed. The paper doesn't pretend mirror separation is proven; it presents a conditional rank-count theorem. If a future version can prove the non-identity criterion for a template family, verify mirror separation computationally for small instances, or anchor the result in the DMDGP special case, this becomes a strong, citable result. As it stands, the framework is valuable and the algebra is coherent, but the central advertised count is conditional on an unverified assumption.

Who gets value? Distance geometry researchers working on DMDGP/DDGP counting, and anyone interested in algebraic methods for discrete geometry. It deserves a serious referee—send it to peer review, not desk reject. The referee should ask for one of those three things: a genericity proof, a verification protocol, or a DMDGP reduction.

Referee Report

3 major / 3 minor

Summary. The paper studies the Combinatorial Discretizable Distance Geometry Problem (Combinatorial DDGP) and proposes an algebraic method for counting realizations when pruning edges are present. It introduces branch codes, cone and base generators, a labeled violation matrix V, and the branch-shift space K_F = M(ker V). Lemma 4 shows constructively that every shift in K_F preserves all active edge lengths. The main Rank-Count Theorem (Theorem 1) then claims that, under a hypothesis called mirror-separated parameters (Definition 9) and assuming a reference solution exists, the feasible constrained branch codes form the affine coset s*_c ⊕ K_F, yielding the count 2^{f + rank[M;V] - rank V}. A seven-vertex example is worked out in detail, giving the claimed count 8.

Significance. If the mirror-separation hypothesis can be proved generic or verified for concrete instances, the rank formula would give a polynomial-time algebraic solution to a previously open counting problem, and the paper's algebraic decomposition of branch shifts is a valuable conceptual step. The internal algebra is coherent: the worked example's matrices, ranks, and kernel satisfy the stated identities, and Lemma 4's forward inclusion is plausible. However, the exact-count claim rests entirely on Definition 9, which is neither derived from more basic assumptions nor verified in any example. The paper therefore delivers a conditional framework rather than a complete resolution; its significance is correspondingly qualified.

major comments (3)

[§6, Definition 9 and Theorem 1] The (⊆) direction of Theorem 1 is not a derivation but a restatement of the mirror-separation hypothesis. Definition 9 asserts exactly that every constrained shift outside K_F changes some active edge length, so the conclusion 'if h∉K_F then a contradiction' is the hypothesis itself. Thus the exactness of the rank-count formula is contingent on an assumption whose content is the absence of the extra solutions the theorem aims to exclude. The paper should state this dependence explicitly and separate the conditional algebraic theorem from the claimed solution of the counting problem.
[§6, Remark 1] The genericity argument for mirror separation is unsupported. The discrepancy functions Δ_{h,e}(θ;s) involve coordinates obtained from sphere intersections, hence contain radicals; it is not automatic that a non-identity criterion implies the failure set has measure zero. Moreover, the criterion must be checked for every h∉K_F and every branch code s, and no such proof or verification is supplied. Consequently, the claim that mirror separation holds for almost all nondegenerate parameters is not established.
[§7] The worked example does not verify mirror separation. It computes the algebraic objects and traces a few selected generator moves, but it does not enumerate the 32 constrained branch codes (or the 128 full codes) to confirm that the true count is 8, nor does it check the Definition 9 condition for every shift outside K_F. Thus the paper provides no evidence that the central hypothesis is satisfiable in any concrete non-vacuous instance. This is a load-bearing gap because the rank formula's exactness depends entirely on that hypothesis.

minor comments (3)

[Abstract] The phrase 'topological solution count' is potentially misleading: the count depends on the mirror-separation hypothesis, which is not shown to be a topological or weight-independent property. The wording should reflect the conditional nature.
[§7.2] The geometric tracing is helpful, but a small table listing the 8 full branch codes (or at least the 4 constrained codes) and their coordinate realizations would make the example self-contained and would allow the reader to independently check the count.
[Lemma 4] The sequential application of mirror blocks is plausible but compressed; a sentence justifying that the current embedding remains a branch embedding after each block application would improve readability.

Circularity Check

1 steps flagged · score 6.0 of 10

Rank-count exactness is built into the mirror-separation hypothesis; the converse half of Theorem 1 is Definition 9 restated.

self definitional [Theorem 1, proof (⊆), Section 6; Definition 9]
"Let sc ∈ ΞF and set h = sc ⊕ s∗c. Extend h by zero on Bf and put s̃ = s∗ ⊕ h. Then s̃ has constrained part sc and is active-feasible. If h /∈ KF, mirror separation gives an active-edge length difference between s∗ and s̃, a contradiction. Hence h ∈ KF and ΞF ⊆ s∗c ⊕ KF."

The reverse inclusion Ξ_F ⊆ s*_c ⊕ K_F is not derived from geometry; it is exactly the defining property of mirror-separated parameters. Definition 9 defines mirror-separation as: for every constrained shift h ∉ K_F and every full branch code s, some active edge changes length. Taking s = s* and h = sc ⊕ s*_c makes the theorem's conclusion the contrapositive of the hypothesis. Thus the exact equality Ξ_F = s*_c ⊕ K_F, and hence the advertised count 2^{f+rank[M;V]−rank V}, reduces by construction to an assumed non-existence of extra feasible shifts. The 7-vertex example does not verify mirror separation for all shifts outside K_F (only a few moves are traced), and Remark 1's genericity claim is conditional on an unproved non-identity criterion for the radical-valued discrepancies Δ_{h,e}.

full rationale

The algebraic machinery—Lemma 1 (rank-nullity), Lemma 2 (block admissibility), Lemma 3 (kernel-induced mirror-compatibility), and Lemma 4 (zero-violation preservation)—is internally consistent and non-circular: it constructively shows K_F ⊆ (feasible shifts from a reference). The circularity is in the converse. Theorem 1's (⊆) step, needed for the exact count, is a one-line application of Definition 9, which was formulated precisely to assert that no feasible shift lies outside K_F. The paper even states 'The following hypothesis excludes exactly those extra solutions,' making the dependence transparent but definitional: the topological count is exact only if the hypothesis holds, and the paper does not prove or verify that hypothesis. The 7-vertex example computes ranks and illustrates some kernel-induced reflections, but it does not enumerate the 16 constrained shifts to certify mirror separation for that instance. Remark 1's genericity argument is conditional on a non-identity criterion for discrepancy functions involving radical coordinates; the criterion is neither proved nor checked. No self-citation is load-bearing, and the forward inclusion is genuine independent content, so this is partial rather than total circularity. Score 6 reflects that the core prediction (the rank-count formula's exactness) reduces by construction to an assumed equality, while a substantial algebraic framework remains non-circular.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The theory's accountability rests on two ad-hoc hypotheses (mirror separation and the non-identity criterion) that together assert the absence of accidental non-algebraic feasible shifts — i.e., the exactness of the count is assumed rather than derived. The remaining machinery (branch masks, generator families, labeled violation matrices, ranks) is cleanly defined from the graph template with no fitted parameters, no invented physical entities, and no constants tuned to data.

assumptions (6)

domain assumption Nondegenerate DDGP parameters (Definition 8): every lateration step is well-defined and each K-sphere intersection gives two distinct reflected choices.
Standard generic-position condition in the DMDGP counting literature; the paper states it is open and dense but does not prove it (Definition 8).
ad hoc to paper Mirror-separated parameters (Definition 9): for every h ∉ K_F and every branch code s, some active edge length changes.
Load-bearing and essentially the converse of Theorem 1. Instance-dependent (defined via θ and the embedding); not verified in the worked example; Remark 1's genericity justification is conditional (see next axiom).
ad hoc to paper Non-identity criterion (Remark 1): for every h ∉ K_F and branch code s, at least one discrepancy Δ_{h,e}(·;s) is not identically zero on the parameter component.
The premise that would make mirror separation hold for 'almost all' parameters; unproved. The Δ functions contain radicals from sphere intersections, so identity-zero is possible; the paper asserts the implication, not the premise.
domain assumption Existence of a viable reference solution (Ξ ≠ ∅).
Theorem 1 requires nonemptiness; in practice one finds a solution (e.g., by Branch-and-Prune) before counting. Stated in the theorem and acknowledged.
domain assumption Generic non-incidence: no active vertex outside a predecessor clique C lies in aff(C).
Invoked in Section 5 to equate 'fixed endpoint lies in C' with length preservation. A standard generic-position assumption, folded into nondegeneracy/mirror separation.
standard math Standard F_2 linear algebra and Euclidean reflection facts (rank-nullity, isometries preserve lengths, Fix/Cone closure of the predecessor DAG).
Background used throughout: rank-nullity over F_2, Euclidean isometries preserve distances, generic sphere intersection gives two points, and the ancestor/descendant closure properties of the dependency graph.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A Rank-Count Theory for the Combinatorial Discretizable Distance Geometry Problem." pith.science (2026). https://pith.science/paper/ZGDARY4J

@misc{pith2026260727213,
  author       = {Pith},
  title        = {Pith review of: A Rank-Count Theory for the Combinatorial Discretizable Distance Geometry Problem},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZGDARY4J}},
  note         = {Machine review of arXiv:2607.27213}
}

abstract

The Distance Geometry Problem (DGP) asks for a geometric realization of a weighted graph with \(n\) vertices in \(\mathbb{R}^K\) such that Euclidean distances between vertices match the given edge weights. When a vertex order where every non-seed vertex has \(K\) predecessors inducing a clique is part of the input, the search space can be discretized via \(K\)-lateration, branching into a binary tree. In this subclass, known as the Combinatorial Discretizable Distance Geometry Problem (Combinatorial DDGP), the goal is to determine the number of realizations satisfying all distance constraints. While the number of realizations is almost always \(2^{n-K}\) when no additional distance constraints are present, a topological solution count in the presence of additional constraints has remained elusive. We develop an algebraic rank-count theory for the feasible binary branch codes, proving that under mirror-separated parameters they form an affine space over \(\mathbb{F}_2\) whenever a viable reference solution exists.

Figures

Figures reproduced from arXiv: 2607.27213 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references

[1]

The K- discretization and K-incident graphs for discretizable distance geometry.Optimization Letters, 14(2):1–14, 2018

Germano Abud, Jorge Alencar, Carlile Lavor, Leo Liberti, and Antonio Mucherino. The K- discretization and K-incident graphs for discretizable distance geometry.Optimization Letters, 14(2):1–14, 2018

2018
[2]

An impossible combinatorial counting method in distance geometry.Discrete Applied Mathematics, 354:83–93, 2024

Germano Abud, Jorge Alencar, Carlile Lavor, Leo Liberti, and Antonio Mucherino. An impossible combinatorial counting method in distance geometry.Discrete Applied Mathematics, 354:83–93, 2024

2024
[3]

Discretization vertex orders in distance geometry.Discrete Applied Mathematics, 197:27–41, 2015

Andrea Cassioli, Oktay Günlük, Carlile Lavor, and Leo Liberti. Discretization vertex orders in distance geometry.Discrete Applied Mathematics, 197:27–41, 2015

2015
[4]

A new algorithm for the K-DMDGP subclass of distance geometry problems with exact distances.Algorithmica, 83(8):2400–2426, 2021

Douglas S Gonçalves, Carlile Lavor, Leo Liberti, and Michael Souza. A new algorithm for the K-DMDGP subclass of distance geometry problems with exact distances.Algorithmica, 83(8):2400–2426, 2021

2021
[5]

The discretizable molecular distance geometry problem.Computational Optimization and Applications, 52(1):115–146, 2012

Carlile Lavor, Leo Liberti, Nelson Maculan, and Antonio Mucherino. The discretizable molecular distance geometry problem.Computational Optimization and Applications, 52(1):115–146, 2012

2012
[6]

On the optimality of finding DMDGP symmetries.Computational and Applied Mathematics, 40(3):98, 2021

Carlile Lavor, Andrés Oliveira, Wagner Rocha, and Michael Souza. On the optimality of finding DMDGP symmetries.Computational and Applied Mathematics, 40(3):98, 2021

2021
[7]

Distance geometry and data science.Top, 28(2):271–339, 2020

Leo Liberti. Distance geometry and data science.Top, 28(2):271–339, 2020. 17

2020
[8]

Counting the number of solutions of kdmdgp instances

Leo Liberti, Carlile Lavor, Jorge Alencar, and Germano Abud. Counting the number of solutions of kdmdgp instances. InInternational Conference on Geometric Science of Information, pages 224–230. Springer, 2013

2013

Show all 16 references

[9]

A branch-and-prune algorithm for the molecular distance geometry problem.International Transactions in Operational Research, 15(1):1–17, 2008

Leo Liberti, Carlile Lavor, and Nelson Maculan. A branch-and-prune algorithm for the molecular distance geometry problem.International Transactions in Operational Research, 15(1):1–17, 2008

2008
[10]

The discretizable molecular distance geometry problem seems easier on proteins

Leo Liberti, Carlile Lavor, and Antonio Mucherino. The discretizable molecular distance geometry problem seems easier on proteins. InDistance Geometry: Theory, Methods, and Applications, pages 47–60. Springer, 2012

2012
[11]

On the number of solutions of the discretizable molecular distance geometry problem

Leo Liberti, Benoît Masson, Jon Lee, Carlile Lavor, and Antonio Mucherino. On the number of solutions of the discretizable molecular distance geometry problem. InInternational Conference on Combinatorial Optimization and Applications, pages 322–342. Springer, 2011

2011
[12]

On the number of realizations of certain henneberg graphs arising in protein conformation.Discrete Applied Mathematics, 165:213–232, 2014

Leo Liberti, Benoît Masson, Jon Lee, Carlile Lavor, and Antonio Mucherino. On the number of realizations of certain henneberg graphs arising in protein conformation.Discrete Applied Mathematics, 165:213–232, 2014

2014
[13]

An analysis on the degrees of freedom of binary representations for solutions to discretizable distance geometry problems

Antonio Mucherino. An analysis on the degrees of freedom of binary representations for solutions to discretizable distance geometry problems. InRecent Advances in Computational Optimization, pages 251–255. Springer, 2021

2021
[14]

The discretizable distance geometry problem.Optimization Letters, 6(8):1671–1686, 2012

Antonio Mucherino, Carlile Lavor, and Leo Liberti. The discretizable distance geometry problem.Optimization Letters, 6(8):1671–1686, 2012

2012
[15]

Exploiting symmetry properties of the dis- cretizable molecular distance geometry problem.Journal of Bioinformatics and Computational Biology, 10(03):1242009, 2012

Antonio Mucherino, Carlile Lavor, and Leo Liberti. Exploiting symmetry properties of the dis- cretizable molecular distance geometry problem.Journal of Bioinformatics and Computational Biology, 10(03):1242009, 2012

2012
[16]

The positioning problem—a draft of an intermediate summary

Yechiam Yemini. The positioning problem—a draft of an intermediate summary. InProceedings of the Conference on Distributed Sensor Networks, pages 137–145. sn, 1978. 18

1978

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.

[1] [1]

The K- discretization and K-incident graphs for discretizable distance geometry.Optimization Letters, 14(2):1–14, 2018

Germano Abud, Jorge Alencar, Carlile Lavor, Leo Liberti, and Antonio Mucherino. The K- discretization and K-incident graphs for discretizable distance geometry.Optimization Letters, 14(2):1–14, 2018

2018

[2] [2]

An impossible combinatorial counting method in distance geometry.Discrete Applied Mathematics, 354:83–93, 2024

Germano Abud, Jorge Alencar, Carlile Lavor, Leo Liberti, and Antonio Mucherino. An impossible combinatorial counting method in distance geometry.Discrete Applied Mathematics, 354:83–93, 2024

2024

[3] [3]

Discretization vertex orders in distance geometry.Discrete Applied Mathematics, 197:27–41, 2015

Andrea Cassioli, Oktay Günlük, Carlile Lavor, and Leo Liberti. Discretization vertex orders in distance geometry.Discrete Applied Mathematics, 197:27–41, 2015

2015

[4] [4]

A new algorithm for the K-DMDGP subclass of distance geometry problems with exact distances.Algorithmica, 83(8):2400–2426, 2021

Douglas S Gonçalves, Carlile Lavor, Leo Liberti, and Michael Souza. A new algorithm for the K-DMDGP subclass of distance geometry problems with exact distances.Algorithmica, 83(8):2400–2426, 2021

2021

[5] [5]

The discretizable molecular distance geometry problem.Computational Optimization and Applications, 52(1):115–146, 2012

Carlile Lavor, Leo Liberti, Nelson Maculan, and Antonio Mucherino. The discretizable molecular distance geometry problem.Computational Optimization and Applications, 52(1):115–146, 2012

2012

[6] [6]

On the optimality of finding DMDGP symmetries.Computational and Applied Mathematics, 40(3):98, 2021

Carlile Lavor, Andrés Oliveira, Wagner Rocha, and Michael Souza. On the optimality of finding DMDGP symmetries.Computational and Applied Mathematics, 40(3):98, 2021

2021

[7] [7]

Distance geometry and data science.Top, 28(2):271–339, 2020

Leo Liberti. Distance geometry and data science.Top, 28(2):271–339, 2020. 17

2020

[8] [8]

Counting the number of solutions of kdmdgp instances

Leo Liberti, Carlile Lavor, Jorge Alencar, and Germano Abud. Counting the number of solutions of kdmdgp instances. InInternational Conference on Geometric Science of Information, pages 224–230. Springer, 2013

2013