Pith. sign in

REVIEW 3 major objections 8 minor 30 references

Spectral Outer-Approximation Algorithms for Binary Semidefinite Problems

T0 review · 3 major / 8 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read This paper claims that binary quadratically constrained quadratic programs, rewritten as binary semidefinite programs, can be solved faster with a spectral outer-approximation algorithm whose cuts come from a common eigenbasis of the…

desk verdict Genuinely useful empirical results on Boolean least squares, but Algorithm 2's spectral initialization is under-specified and the marketed spectral mechanism isn't what drives the wins. read the letter →

arxiv 2506.18265 v1 pith:BC4PYX6Z submitted 2025-06-23 math.OC

Daniel de Roux , Zedong Peng , David E. Bernal Neira This is my paper · ORCID
classification math.OC MSC 90C2290C1090C2090C26
keywords binarysemidefiniteprogrammingouterapproximationeigenvectorcutssimultaneousdiagonalizationquadraticallyconstrainedquadraticprogramsintegerBooleanleastsquaresknapsack
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 binary quadratically constrained quadratic programs, once rewritten exactly as binary semidefinite programs, can be solved efficiently by a specialized outer-approximation scheme that uses eigenvector-based cuts. The cuts come from a matrix that simultaneously diagonalizes the objective and a weighted sum of the constraints, yielding a polyhedral or second-order outer approximation of the semidefinite cone. The authors report that the resulting algorithm solves cardinality-constrained Boolean least squares instances with up to 100 variables in seconds and is competitive with or faster than established integer-semidefinite solvers on the tested problems. If correct, this makes integer semidefinite programming a practical route for a broad class of binary quadratic problems.

What carries the argument

The load-bearing object is the simultaneous diagonalizer: a matrix $U$ whose columns are eigenvectors of both the objective matrix $C$ and an aggregation $A(q_2) = \sum_i q_{2,i} A_i$ that commutes with $C$, found by solving the linear program $\min f(y)$ subject to $C A(y) = A(y) C$. Its columns $v_1,\dots,v_n$ define the second-order cuts $v^\top (X - xx^\top) v \geq 0$ that outer-approximate $X - xx^\top \succeq 0$; the identity aggregation $q_1$ with $\sum_i q_{1,i} A_i = I$ enters through the augmented matrix $\begin{pmatrix} X & x \\ x^\top & 1 \end{pmatrix}$ used in the second-order formulation. Lemma 2 shows that disaggregating a dual positive semidefinite matrix into eigenvector terms turns one linear cut into several second-order cuts that together imply it, which is what allows the outer approximation to tighten until the lower and upper bounds meet.

What would settle it

Run the algorithm on a binary semidefinite reformulation of a BQCQP whose quadratic constraint matrices are all diagonal and have no positive linear combination equal to the identity; if the solver still succeeds quickly, the identity aggregation is not load-bearing, and if it fails, the reported speed-ups depend on an aggregation the paper never explicitly constructs.

Watch

Extended reading notes

Core claim

The central discovery is that the simultaneous-diagonalization idea, previously used for instance-specific linear relaxations of ordinary semidefinite programs, transfers to the binary setting: for binary semidefinite programs arising from binary QCQPs, take the constraint $X - xx^\top \succeq 0$, find two weighted aggregations of the original quadratic constraint matrices — one equal to the identity, one commuting with the objective — and use their common eigenvectors as cut directions $v^\top (X - xx^\top) v \geq 0$. Adding these spectral cuts to an outer-approximation loop (Algorithm 2) and to a lazy branch-and-bound variant (Algorithm 3) gives a solver that the paper shows outperforms the generic outer-approximation, cutting-plane, and branch-and-bound integer-semidefinite baselines on Boolean least squares and remains competitive on quadratic knapsack.

Load-bearing premise

The method needs two weighted combinations of the constraint matrices: one that equals the identity matrix and another, with no overlapping weights, that commutes with the objective matrix; if such combinations do not exist, the advertised eigenvector set reduces to eigenvectors of the objective alone.

Editorial extensions

If this is right

  • Integer semidefinite programming becomes a practically viable way to solve binary quadratically constrained quadratic programs, not just a theoretical reformulation.
  • The spectral cuts $v^\top (X - xx^\top) v \geq 0$ can be added to existing integer-semidefinite solvers as a strengthened formulation, independent of the outer-approximation loop.
  • The lazy single-tree variant avoids restarting the mixed-integer solver from scratch at every outer iteration, which is why it scales to larger Boolean least squares instances.
  • The method applies beyond BQCQP-derived problems to any integer semidefinite program where an objective and a constraint aggregation admit a common eigenbasis.
  • On quadratic knapsack, the same algorithm remains competitive for small instances but is beaten by a generic branch-and-bound approach for $n \geq 30$, so the spectral recipe is not universally dominant.

Reading between the lines

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

  • The identity aggregation $\sum_i q_{1,i} A_i = I$ cannot hold for the raw quadratic constraint matrices of the tested problems unless the affine constraints $\mathrm{Diag}(X) = x$ are silently folded into the matrices $A_i$; the paper never states this, and without it the eigenvector set collapses to eigenvectors of $C$ alone.
  • The cut-generation recipe may generalize to mixed-integer conic programs beyond semidefinite ones, wherever a simultaneous diagonalizer exists for the objective and one constraint aggregation, though the paper only demonstrates the semidefinite case.
  • The tables show spectral cuts sometimes hurt performance (for instance, on some quadratic knapsack settings), so a testable extension is to characterize when the common-eigenbasis cuts tighten the relaxation enough to justify the extra second-order cone burden.
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.

Referee Report

3 major / 8 minor

Summary. The paper proposes two outer-approximation algorithms for integer semidefinite programs arising from binary QCQPs: a spectral second-order outer approximation (OA_SOC, Algorithm 2) and a lazy-constraint branch-and-cut variant (LAZY_SOC, Algorithm 3). The key idea is to generate valid cuts v^T(X−xx^T)v ≥ 0 from eigenvectors of a matrix that simultaneously diagonalizes the objective C and an aggregation A(q2) of the constraint matrices, and to update the outer approximation with dual cuts, with finite termination guaranteed by Lemma 1. The paper reports computational experiments on cardinality-constrained Boolean least squares (Section 5.1) and quadratic knapsack (Section 5.2), comparing OA_SOC and LAZY_SOC with SCIP-SDP, PAJARITO, and the Kobayashi–Takano algorithms. Tables 1–3 show OA_SOC solving BLS instances up to n=100 in seconds and being competitive on QK for small n, while SCIP-SDP wins on larger QK instances. The central claim is that the spectral outer-approximation approach is competitive with, and sometimes faster than, state-of-the-art ISDP solvers.

Significance. If the algorithm were fully specified and the experimental results reproducible, this would be a useful empirical contribution to the emerging area of integer semidefinite programming for binary QCQPs. The study is comparatively broad: ten random instances per parameter setting, shifted geometric means, a one-hour time limit, and comparisons against several established solvers, with the code made available. The paper honestly notes that specialized MIQP solvers such as Gurobi can outperform ISDP-based approaches on QK. The main novelty—spectral cuts from simultaneous diagonalization—is clearly motivated, and Lemma 2 provides a neat disaggregation of dual PSD cuts into second-order constraints. However, the significance is contingent on clarifying how the spectral set S is actually constructed for the test problems; as written, the algorithm's description does not match the problems solved.

major comments (3)
  1. [Section 4, Algorithm 2, steps 1–2; Sections 5.1–5.2] Algorithm 2 requires a vector q1 with sum_i q1_i A_i = I and a vector q2, with support disjoint from q1, such that C and sum_i A_i q2_i commute. The paper never states which constraint matrices A_i are used (the original n×n matrices or the lifted (n+1)×(n+1) matrices of the BSDP reformulation), and for the test problems neither interpretation supports the existence of such q1. In the quadratic knapsack formulation QKSDP of Section 5.2, after encoding Diag(X)=x, the n×n constraint matrices are the single matrix diag(w), and no scalar multiple of diag(w) equals I_n for non-constant weights. In the lifted formulation, the matrices encoding Diag(X)=x introduce off-diagonal entries that cannot be canceled by any linear combination of the available constraint matrices, so no combination equals I_{n+1}. The same obstruction applies to the BLS problem of Section 5.1. Consequently, Algorithm 2 as written cannot be instantiated for the problems whose results are reported in Tables 1–3, and the OA_SOC results are not reproducible from the manuscript. The authors must specify the constraint set used in Algorithm 2, prove the existence of q1 and q2 for the test instances, or describe the construction actually used in the experiments.
  2. [Section 4, Eq. (6)] Program (6) is stated as min f(y) subject to C A(y) = A(y) C, and the text notes that the zero matrix always serves as a feasible solution. The constraints that q2 be nonzero and have support disjoint from q1, which are stated in Algorithm 2 step 2, do not appear in the displayed program. As written, q2=0 is always feasible, in which case the commuting condition is trivial and the set S in Algorithm 2 reduces to the eigenvectors of C alone; the method then degenerates to standard eigenvector cuts and is not the advertised simultaneous-diagonalization algorithm. The authors should present the full optimization problem including the missing constraints (with variables in R^r rather than R^n, since there are r constraint matrices) and discuss when feasible aggregations exist.
  3. [Section 6, first paragraph of experimental setup] The labels 'basic' and 'spectral' are used inconsistently. The text says that problems with the cuts v^T(X−xx^T)v ≥ 0 are 'basic' and those without are 'spectral', yet immediately afterward the same cuts are called 'spectral cuts', and the discussion says that 'spectral cuts reduce the performance' of most methods while 'OA_SOC with the basic formulation performs the best.' Since the tables and figures compare PAJARITO_OA-basic versus PAJARITO_OA-spectral, SCIPSDP-basic versus SCIPSDP-spectral, and related pairs, the reader cannot determine which formulation was used for each reported number. This ambiguity affects the interpretation of the central empirical comparison and must be resolved.
minor comments (8)
  1. [Section 1.1 (Notation)] The definition of [n] reads 'For an integer k∈N, [n] denotes the set of natural numbers {1,...,k}'; this should be 'For an integer n∈N, [n] denotes {1,...,n}.'
  2. [Section 4, Eq. (6)] The optimization variable in program (6) is written as y∈R^n, but the number of constraint matrices is r; it should be y∈R^r. The accompanying sentence 'the zero matrix always serves as a feasible solution' should refer to the zero vector.
  3. [Section 4.1] The definition of the standard second-order cone L^{1+n} is written as r ≥ ||t||_2^2; the standard second-order cone is r ≥ ||t||_2. The squared norm belongs to the rotated second-order cone defined just below.
  4. [Section 4.2, paragraph after Lemma 2] The suggested cut in Algorithm 2 step 8 is written as w_j^T X w_j ≥ (w_j z_j)^2; this is dimensionally inconsistent and should be w_j^T X w_j ≥ (w_j^T x)^2, matching the hypothesis of Lemma 2.
  5. [Section 3, Lemma 1] The sentence 'Let S′ be optimal for the latter program' should refer to the dual program DSDP(X_L); the variable S′ lives in the dual, so 'latter' should be 'former'. Also, 'Set T={S′}' should be 'Set T = T ∪ {S′}' to align with Algorithm 1 step 6.
  6. [Section 4, Algorithm 3] The last line of Algorithm 3 is garbled ('Go to 3. return X^'); the stopping criterion for the lazy-constraint loop (namely λ_n(X^) ≥ −ε) should be stated explicitly as a condition in the branch-and-bound procedure.
  7. [Section 6, experimental setup paragraph] The sentence 'Since CUTSDP is essentially the same algorithm as the cutting plane algorithm of these two authors, we do not test the latter algorithm' is unclear, because KOB_Cutting_Plane appears in the tables and figures; presumably the authors mean they do not test CUTSDP, not the Kobayashi–Takano cutting-plane algorithm.
  8. [Section 6.1] The paragraph beginning 'The results of the cardinality-constrained Boolean least squares problem are consistent...' appears twice with slightly different wording; one copy should be deleted.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the spectral outer-approximation algorithm reuses the authors' prior theoretical result, but the central empirical claim is an external solver comparison with no fitted inputs or renamed predictions.

full rationale

The paper's central claim is that OA_SOC and LAZY_SOC are competitive with, and sometimes faster than, external solvers such as SCIP-SDP and PAJARITO on binary semidefinite reformulations of BQCQPs. This claim is supported by direct computational experiments on randomly generated instances, with shifted geometric mean times reported against external solvers; no parameter is fitted to these outcomes and no predicted quantity is defined in terms of the data used to produce it. The spectral cut selection is imported from the authors' prior work (de Roux et al., 2025), which is a self-citation, but it is used as a mathematical design principle rather than as a fitted ingredient of the reported results, and the theorem that simultaneous diagonalizability makes the linear relaxation exact is an externally checkable published result rather than an assumption tailored to make the experiments come out favorably. The BQCQP-to-BSDP reformulation is also cited from independent work by de Meijer and Sotirov. The under-specification of the q1 and q2 construction in Algorithm 2 is a reproducibility and correctness concern, not a circularity concern, because it does not make any claimed output equivalent to an input by construction. Accordingly, no circular step is identified and the circularity score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or mathematical entities. Its method relies on a prior reformulation theorem, strong duality, and spectral theory. The main unstated assumption is that the spectral initialization is well-defined for the test problems.

assumptions (4)
  • domain assumption Strong duality holds for the SDP subproblems SDP(X_L) and DSDP(X_L), so Lemma 1's termination argument applies.
    Invoked before Lemma 1 and used in the convergence justification of Algorithms 1 and 2; not proven for the specific test instances.
  • ad hoc to paper The matrix family C, A_1,...,A_r admits a nonzero aggregation A(q_2) that commutes with C, and q_1 exists so that sum_i q1_i A_i = I.
    Algorithm 2 steps 1 and 2. The zero matrix is always feasible for program (6), so q2 may vanish; the existence of q1 is not established for BSDPs where Diag(X)=x is handled separately from the A_i constraints.
  • domain assumption The BQCQP to BSDP equivalence stated as Theorem 1 from de Meijer and Sotirov 2023 is correct.
    This equivalence is the foundation of the reformulations used in Sections 2.3 and 5.
  • standard math A set of symmetric matrices is simultaneously diagonalizable if and only if all matrices in the set commute pairwise.
    Standard linear algebra result used to justify the LP in program (6).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Spectral Outer-Approximation Algorithms for Binary Semidefinite Problems." pith.science (2026). https://pith.science/paper/BC4PYX6Z

@misc{pith2026250618265,
  author       = {Pith},
  title        = {Pith review of: Spectral Outer-Approximation Algorithms for Binary Semidefinite Problems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BC4PYX6Z}},
  note         = {Machine review of arXiv:2506.18265}
}
read the original abstract

Integer semidefinite programming (ISDP) has recently gained attention due to its connection to binary quadratically constrained quadratic programs (BQCQPs), which can be exactly reformulated as binary semidefinite programs (BSDPs). However, it remains unclear whether this reformulation effectively uses existing ISDP solvers to address BQCQPs. To the best of our knowledge, no specialized ISDP algorithms exploit the unique structure of BSDPs derived from BQCQPs. This paper proposes a novel spectral outer approximation algorithm tailored for BSDPs derived from BQCQP reformulations. Our approach is inspired by polyhedral and second-order representable regions that outer approximate the feasible set of a semidefinite program relying on a spectral decomposition of a matrix that simultaneously diagonalizes the objective matrix and an aggregation of the constraint matrices. Computational experiments show that our algorithm is competitive with, and in some cases outperforms, state-of-the-art ISDP solvers such as SCIP-SDP and PAJARITO, highlighting ISDP's potential for solving BQCQPs.

Figures

Figures reproduced from arXiv: 2506.18265 by the authors.

Figure 1
Figure 1. Time performance of different algorithms on the cardinality-constrained binary least squares problem with [PITH_FULL_IMAGE:figures/full_fig_p012_1.png] view at source ↗
Figure 2
Figure 2. Time performance of different algorithms on the cardinality-constrained binary least squares problem with [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗
Figure 3
Figure 3. Time performance of different algorithms on the quadratic knapsack problem [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 28 canonical work pages

  1. [1]

    Conic optimization: a survey with special focus on copositive optimization and binary quadratic problems

    Mirjam D \"u r and Franz Rendl. Conic optimization: a survey with special focus on copositive optimization and binary quadratic problems . EURO Journal on Computational Optimization, 9: 0 100021, 2021

    work page 2021

  2. [2]

    Quadratic optimization problems

    Naum Z Shor. Quadratic optimization problems . Soviet Journal of Computer and Systems Sciences, 25: 0 1--11, 1987

    work page 1987

  3. [3]

    A framework for solving mixed-integer semidefinite programs

    Tristan Gally, Marc E Pfetsch, and Stefan Ulbrich. A framework for solving mixed-integer semidefinite programs . Optimization Methods and Software, 33 0 (3): 0 594--632, 2018

    work page 2018

  4. [4]

    Exact and approximation algorithms for sparse principal component analysis

    Yongchun Li and Weijun Xie. Exact and approximation algorithms for sparse principal component analysis. INFORMS Journal on Computing, 2024

    work page 2024

  5. [5]

    Approximating k-means-type clustering via semidefinite programming

    Jiming Peng and Yu Wei. Approximating k-means-type clustering via semidefinite programming. SIAM journal on optimization, 18 0 (1): 0 186--205, 2007

    work page 2007

  6. [6]

    Sos-sdp: an exact solver for minimum sum-of-squares clustering

    Veronica Piccialli, Antonio M Sudoso, and Angelika Wiegele. Sos-sdp: an exact solver for minimum sum-of-squares clustering. INFORMS Journal on Computing, 34 0 (4): 0 2144--2162, 2022

    work page 2022

  7. [7]

    Computing restricted isometry constants via mixed-integer semidefinite programming

    Tristan Gally and MARC E Pfetsch. Computing restricted isometry constants via mixed-integer semidefinite programming . preprint, submitted, 2016

    work page 2016

  8. [8]

    On Integrality in Semidefinite Programming for Discrete Optimization

    Frank de Meijer and Renata Sotirov. On integrality in semidefinite programming for discrete optimization . arXiv preprint arXiv:2306.09865, 2023

    work page Pith review arXiv 2023

Show all 30 references
  1. [9]

    The max-cut problem and quadratic 0--1 optimization; polyhedral aspects, relaxations and bounds

    Endre Boros and Peter L Hammer. The max-cut problem and quadratic 0--1 optimization; polyhedral aspects, relaxations and bounds . Annals of Operations Research, 33 0 (3): 0 151--180, 1991

    1991

  2. [10]

    Community detection in graphs

    Santo Fortunato. Community detection in graphs . Physics reports, 486 0 (3-5): 0 75--174, 2010

    2010

  3. [11]

    General heuristics for nonconvex quadratically constrained quadratic programming

    Jaehyun Park and Stephen Boyd. General heuristics for nonconvex quadratically constrained quadratic programming . arXiv preprint arXiv:1703.07870, 2017

    2017 arXiv

  4. [12]

    Retrospective on optimization

    Lorenz T Biegler and Ignacio E Grossmann. Retrospective on optimization . Computers & Chemical Engineering, 28 0 (8): 0 1169--1192, 2004

    2004

  5. [13]

    On the equivalence of quadratic optimization problems commonly used in portfolio theory

    Taras Bodnar, Nestor Parolya, and Wolfgang Schmid. On the equivalence of quadratic optimization problems commonly used in portfolio theory . European Journal of Operational Research, 229 0 (3): 0 637--644, 2013

    2013

  6. [14]

    A branch-and-cut approach to portfolio selection with marginal risk control in a linear conic programming framework

    Zhibin Deng, Yanqin Bai, Shu-Cherng Fang, Ye Tian, and Wenxun Xing. A branch-and-cut approach to portfolio selection with marginal risk control in a linear conic programming framework . Journal of Systems Science and Systems Engineering, 22 0 (4): 0 385--400, 2013

    2013

  7. [15]

    Quadratic and higher-order unconstrained binary optimization of railway rescheduling for quantum computing

    Krzysztof Domino, Akash Kundu, \"O zlem Salehi, and Krzysztof Krawiec. Quadratic and higher-order unconstrained binary optimization of railway rescheduling for quantum computing . Quantum Information Processing, 21 0 (9): 0 337, 2022

    2022

  8. [16]

    Five Starter Problems: Solving Quadratic Unconstrained Binary Optimization Models on Quantum Computers

    Arul Mazumder and Sridhar Tayur. Five Starter Problems: Solving Quadratic Unconstrained Binary Optimization Models on Quantum Computers . arXiv preprint arXiv:2401.08989, 2024

    2024 arXiv

  9. [17]

    A copositive framework for analysis of hybrid Ising-classical algorithms

    Robin Brown, David E Bernal Neira , Davide Venturelli, and Marco Pavone. A copositive framework for analysis of hybrid Ising-classical algorithms . SIAM Journal on Optimization, 34 0 (2): 0 1455--1489, 2024

    2024

  10. [18]

    Polyhedral approximation in mixed-integer convex optimization

    Miles Lubin, Emre Yamangil, Russell Bent, and Juan Pablo Vielma. Polyhedral approximation in mixed-integer convex optimization . Mathematical Programming, 172: 0 139--168, 2018

    2018

  11. [19]

    Matrix Analysis

    Roger A Horn and Charles R Johnson. Matrix Analysis . Cambridge university press, 2012

    2012

  12. [20]

    Instance-specific linear relaxations of semidefinite optimization problems

    Daniel de Roux, Robert Carr, and R Ravi. Instance-specific linear relaxations of semidefinite optimization problems. Mathematical Programming Computation, 2025

    2025

  13. [21]

    Outer approximation with conic certificates for mixed-integer convex problems

    Chris Coey, Miles Lubin, and Juan Pablo Vielma. Outer approximation with conic certificates for mixed-integer convex problems . Mathematical Programming Computation, 12 0 (2): 0 249--293, 2020

    2020

  14. [22]

    On polyhedral and second-order cone decompositions of semidefinite optimization problems

    Dimitris Bertsimas and Ryan Cory-Wright. On polyhedral and second-order cone decompositions of semidefinite optimization problems . Operations Research Letters, 48 0 (1): 0 78--85, 2020

    2020

  15. [23]

    Polyhedral approximations of the semidefinite cone and their application

    Yuzhu Wang, Akihiro Tanaka, and Akiko Yoshise. Polyhedral approximations of the semidefinite cone and their application . Computational Optimization and Applications, 78 0 (3): 0 893--913, 2021

    2021

  16. [24]

    Second order cone programming relaxation of a positive semidefinite constraint

    Sunyoung Kim, Masakazu Kojima, and Makoto Yamashita. Second order cone programming relaxation of a positive semidefinite constraint . Optimization Methods and Software, 18 0 (5): 0 535--541, 2003

    2003

  17. [25]

    A branch-and-cut algorithm for solving mixed-integer semidefinite optimization problems

    Ken Kobayashi and Yuich Takano. A branch-and-cut algorithm for solving mixed-integer semidefinite optimization problems . Computational Optimization and Applications, 75: 0 493--513, 2020

    2020

  18. [26]

    The quadratic knapsack problem—a survey

    David Pisinger. The quadratic knapsack problem—a survey . Discrete Applied Mathematics, 155 0 (5): 0 623--648, 2007

    2007

  19. [27]

    The MOSEK optimization toolbox for MATLAB manual

    MOSEK ApS. The MOSEK optimization toolbox for MATLAB manual. Version 10.0. , 2022. URL http://docs.mosek.com/9.0/toolbox/index.html

    2022

  20. [28]

    Gurobi optimizer reference manual , 2020

    Gurobi Optimization et al. Gurobi optimizer reference manual , 2020

    2020

  21. [29]

    An adaptive linear approximation algorithm for copositive programs

    Stefan Bundfuss and Mirjam D \"u r. An adaptive linear approximation algorithm for copositive programs . SIAM Journal on Optimization, 20 0 (1): 0 30--53, 2009

    2009

  22. [30]

    Inner approximating the completely positive cone via the cone of scaled diagonally dominant matrices

    Jo \ a o Gouveia, Ting Kei Pong, and Mina Saee. Inner approximating the completely positive cone via the cone of scaled diagonally dominant matrices . Journal of Global Optimization, 76: 0 383--405, 2020

    2020

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.