Pith. sign in

REVIEW 2 major objections 1 minor 45 references

DASH dimensionality reduction improves incumbent solutions from Gurobi on difficult large-scale MIQPs for subset portfolio selection.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

DASH reduces problem dimensionality in MIQP subset selection to improve MIP solver incumbent quality on hard portfolio instances.

T0 review reviewed 2026-06-26 challenge →

load-bearing objection DASH is a targeted reduction for convex MIQPs in subset portfolio selection that reports better Gurobi incumbents on hard instances, but the abstract supplies no derivation, guarantees, or independent check on the difficulty metric. the 2 major comments →

arxiv 2606.20141 v1 pith:6DH35EDR submitted 2026-06-18 stat.CO

DASH: A Dimensionality Reduction Method for Large-scale Convex MIQP with Applications in Subset Portfolio Selection

Pinzhang Cheng This is my paper
classification stat.CO
keywords MIQPdimensionality reductionsubset selectionportfolio optimizationmixed integer programmingincumbent solutionsGurobiconvex optimization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 introduces DASH, a method to reduce the dimensionality of large convex mixed-integer quadratic programs arising in best-subset selection. The approach is tested on portfolio optimization where the goal is to select a subset of assets under quadratic risk terms. When the covariance matrix is ill-conditioned or portfolio weights have box constraints, standard solvers like Gurobi struggle to find good solutions quickly. DASH consistently finds better incumbents in those cases, and the size of the gain grows with how hard the instance is.

Core claim

The central claim is that applying the Decreasing Active Set Hierarchy reduces the effective size of the MIQP while maintaining feasibility and objective quality, allowing MIP solvers to produce superior incumbent solutions for subset portfolio selection problems whose difficulty is governed by covariance condition number and weight bounds.

What carries the argument

DASH, or Decreasing Active Set Hierarchy, which hierarchically reduces the active set of variables in the MIQP to lower its dimensionality for faster solver progress.

Load-bearing premise

The assumption that problem difficulty for these MIQPs is reliably indicated by the condition number of the covariance matrix together with the box constraints on weights.

What would settle it

An experiment comparing DASH and plain Gurobi on portfolio instances with low condition number but added difficulty from other sources, such as very large cardinality constraints, to check if the improvement disappears.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Improvement is larger and lasts longer as problem difficulty increases.
  • DASH works best when the covariance matrix has a high condition number.
  • It provides practical value for large instances where global optimality cannot be reached.
  • The method targets convex MIQPs in subset selection.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • DASH might extend to other combinatorial subset problems beyond portfolios.
  • Preprocessing to estimate condition number could decide when to apply DASH.
  • Combining DASH with other cutting planes or heuristics could yield further gains.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 1 minor

Summary. The paper proposes DASH (Decreasing Active Set Hierarchy), a dimensionality reduction method for large-scale convex MIQPs in subset portfolio selection. It claims that DASH improves the quality of incumbent solutions obtained by Gurobi, with the improvement being consistent, significant, and scaling with problem difficulty, where difficulty is associated with the condition number of the covariance matrix and box constraints on portfolio weights.

Significance. If the method is shown to preserve solution quality and the experimental results hold under independent validation of the difficulty metric, this could offer a valuable approach for tackling computationally challenging MIQPs in finance. The focus on practical incumbent solutions for NP-hard problems is relevant.

major comments (2)
  1. [Abstract] The assertion that 'the magnitude and duration of improvement by DASH scale with the difficulty of the problem' is based on an untested assumption that condition number and box constraints are reliable predictors of Gurobi performance. The manuscript does not report separate experiments correlating these quantities with solver metrics in the absence of DASH, raising the risk that the scaling observation is circular.
  2. [Method] The description of DASH lacks sufficient detail on the dimensionality reduction procedure, including any pseudocode, mathematical formulation of how active sets are decreased, and proof or argument that optimality or feasibility is preserved for the original problem.
minor comments (1)
  1. [Abstract] No quantitative results, specific metrics, or problem dimensions are provided despite claiming 'extensive set of numerical experiments'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments and recommendation for major revision. We address each major comment below and commit to revisions that strengthen the manuscript's clarity and rigor.

read point-by-point responses

  1. Referee: [Abstract] The assertion that 'the magnitude and duration of improvement by DASH scale with the difficulty of the problem' is based on an untested assumption that condition number and box constraints are reliable predictors of Gurobi performance. The manuscript does not report separate experiments correlating these quantities with solver metrics in the absence of DASH, raising the risk that the scaling observation is circular.

    Authors: We acknowledge the concern regarding potential circularity. While our experiments vary the condition number and box constraints and demonstrate corresponding scaling in DASH's benefits, we did not include separate runs isolating these metrics' correlation with baseline Gurobi performance. In the revision, we will add a dedicated subsection with experiments that directly correlate condition number and box constraint tightness with Gurobi metrics (e.g., time to reach target incumbent quality) in the absence of DASH, providing independent validation of the difficulty predictors. revision: yes

  2. Referee: [Method] The description of DASH lacks sufficient detail on the dimensionality reduction procedure, including any pseudocode, mathematical formulation of how active sets are decreased, and proof or argument that optimality or feasibility is preserved for the original problem.

    Authors: We agree that the method description requires expansion for reproducibility and rigor. The revised manuscript will include: (i) pseudocode for the full DASH procedure, (ii) explicit mathematical formulation of the active-set hierarchy and dimensionality reduction steps, and (iii) a formal argument establishing that the reduced formulation preserves feasibility of the original MIQP (solutions remain feasible when lifted back) while noting that DASH is intended as a heuristic to improve incumbent quality rather than to guarantee global optimality of the original problem. revision: yes

Circularity Check

0 steps flagged

No circularity detected; empirical evaluation of DASH method contains no self-referential derivations or fitted predictions

full rationale

The paper introduces DASH as a dimensionality reduction heuristic for a subclass of convex MIQPs and reports empirical improvements over Gurobi on portfolio-selection instances. The abstract states that problem difficulty 'is related to' condition number and box constraints and that improvement 'scales with' difficulty, but supplies no equations, fitted parameters, or derivation chain. No self-citations, uniqueness theorems, or ansatzes are invoked in the provided text. The central claims rest on numerical experiments whose outcomes are not forced by any internal definition or fit, satisfying the criteria for a self-contained, non-circular evaluation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Review performed on abstract only; no explicit free parameters, axioms, or invented entities are stated in the provided text.

reviewed 2026-06-26 · how reviews work

0 comments
Cite this review

Pith. "Pith review of DASH: A Dimensionality Reduction Method for Large-scale Convex MIQP with Applications in Subset Portfolio Selection." pith.science (2026). https://pith.science/paper/6DH35EDR

@misc{pith2026260620141,
  author       = {Pith},
  title        = {Pith review of: DASH: A Dimensionality Reduction Method for Large-scale Convex MIQP with Applications in Subset Portfolio Selection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6DH35EDR}},
  note         = {Machine review of arXiv:2606.20141}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Subset selection problems as MIPs (Mixed Integer Programs) are NP-hard. For large scale problems, it is infeasible to find global optimal solutions in a reasonable time and good-quality incumbent solutions are sought after with MIP solvers in practice. This paper proposes DASH (Decreasing Active Set Hierarchy) -- a dimensionality reduction method that improves the MIP solver performance for a subclass of best subset selection problems that can be formulated as MIQPs (Mixed Integer Quadratic Programs). We develop and evaluate the performance of DASH in the subset portfolio selection problem with comparison to Gurobi, a commercial MIP solver. In addition to the problem size, the difficulty of a problem is related to the condition number of the covariance matrix and the box constraint on portfolio weights. An extensive set of numerical experiments with varying problem configurations shows that DASH offers consistent and significant improvement of incumbent solutions when the problem is difficult to solve by Gurobi. In particular, the magnitude and duration of improvement by DASH scale with the difficulty of the problem.

Figures

Figures reproduced from arXiv: 2606.20141 by Pinzhang Cheng.

Figure 1
Figure 1. Figure 1: Evolution of Value Function And Minimizer by [PITH_FULL_IMAGE:figures/full_fig_p014_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Attention Evolution via FW Steps Note: The Figure uses synthetic spiked covariance matrix with the following problem parameters: p = 100, k = 10, l = −1, u = 1, κ = 1e1, sample 1 in seed 42. to ℓ1-balls (Duchi et al., 2008), which is also efficient but requires further treatment to prevent violation of the unit hypercube constraint t ∈ [0, 1]p . In the convex feasible set T = conv(S), the FW method selects… view at source ↗
Figure 3
Figure 3. Figure 3: Difficulty of A Spectrum of Problem Configurations, [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Sensitivity Analysis of DASH Improvement [PITH_FULL_IMAGE:figures/full_fig_p024_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Attention Score Evolution via FW Steps Note: The Figure uses synthetic spiked covariance matrix with the following problem parameters: p = 100, k = 10, l = −1, u = 1, κ = 1e3, sample 1 in seed 42. 34 [PITH_FULL_IMAGE:figures/full_fig_p034_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Attention Score Evolution via FW Steps Note: The Figure uses synthetic spiked covariance matrix with the following problem parameters: p = 100, k = 10, l = −1, u = 1, κ = 1e6, sample 1 in seed 42. 35 [PITH_FULL_IMAGE:figures/full_fig_p035_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FWF Efficiency: Closed-form vs Envelope Gradient (10 samples per [PITH_FULL_IMAGE:figures/full_fig_p036_7.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

45 extracted references · 4 canonical work pages

  1. [1]

    The Journal of Finance , year =

    Markowitz, Harry , title =. The Journal of Finance , year =

  2. [2]

    2003 , type =

    Ledoit, Olivier and Wolf, Michael , title =. 2003 , type =

    2003

  3. [3]

    INFORMS Journal on Computing , volume =

    Bertsimas, Dimitris and Cory-Wright, Ryan , title =. INFORMS Journal on Computing , volume =

  4. [4]

    arXiv preprint arXiv:1907.02109 , year =

    Bertsimas, Dimitris and Cory-Wright, Ryan and Pauphilet, Jean , title =. arXiv preprint arXiv:1907.02109 , year =

    work page arXiv 1907

  5. [5]

    , title =

    Clarkson, Kenneth L. , title =. ACM Transactions on Algorithms (TALG) , volume =

  6. [6]

    Advances in Neural Information Processing Systems (NIPS) , year =

    Lacoste-Julien, Simon and Jaggi, Martin , title =. Advances in Neural Information Processing Systems (NIPS) , year =

  7. [7]

    Proceedings of the 30th International Conference on Machine Learning (ICML) , year =

    Jaggi, Martin , title =. Proceedings of the 30th International Conference on Machine Learning (ICML) , year =

  8. [8]

    2011 , type =

    Bai, Jushan and Shi, Shuzhong , title =. 2011 , type =

    2011

  9. [9]

    An Overview of Bilevel Optimization , journal =

    Colson, Beno. An Overview of Bilevel Optimization , journal =

  10. [10]

    Proceedings of the 25th International Conference on Machine Learning (ICML) , year =

    Duchi, John and Shalev-Shwartz, Shai and Singer, Yoram and Chandra, Tushar , title =. Proceedings of the 25th International Conference on Machine Learning (ICML) , year =

  11. [11]

    arXiv preprint arXiv:1705.06270 , year =

    Sinha, Ankur and Malo, Pekka and Deb, Kalyanmoy , title =. arXiv preprint arXiv:1705.06270 , year =

    work page arXiv

  12. [12]

    , title =

    Ross, Stephen A. , title =. Journal of Economic Theory , volume =

  13. [13]

    Optimization Letters , volume =

    Cococcioni, Marco and Fiaschi, Lorenzo , title =. Optimization Letters , volume =

  14. [14]

    Saffi, Pedro A. C. and Sigurdsson, Kari , title =. The Review of Financial Studies , volume =

  15. [15]

    , title =

    Floudas, Christodoulos A. , title =. Encyclopedia of Optimization , publisher =

  16. [16]

    arXiv preprint arXiv:2401.05080 , year =

    Boyd, Stephen and Johansson, Kasper and Kahn, Ronald and Schiele, Philipp and Schmelzer, Thomas , title =. arXiv preprint arXiv:2401.05080 , year =

    work page arXiv

  17. [17]

    Dolan, Elizabeth D. and Mor. Benchmarking Optimization Software With Performance Profiles , journal =

  18. [18]

    Financial Analysts Journal , year =

    Michaud, Richard , title =. Financial Analysts Journal , year =

  19. [19]

    Journal of Portfolio Management , year =

    Chopra, Vijay and Ziemba, William , title =. Journal of Portfolio Management , year =

  20. [20]

    Land, A. H. and Doig, A. G. , title =. Econometrica , volume =. 1960 , publisher =

    1960

  21. [21]

    Mathematical Programming , volume =

    Fischetti, Matteo and Lodi, Andrea , title =. Mathematical Programming , volume =. 2003 , publisher =

    2003

  22. [22]

    Journal of Multivariate Analysis , volume =

    Ledoit, Olivier and Wolf, Michael , title =. Journal of Multivariate Analysis , volume =. 2004 , publisher =

    2004

  23. [23]

    and Uppal, Raman , title =

    DeMiguel, Victor and Garlappi, Lorenzo and Nogales, Francisco J. and Uppal, Raman , title =. Management Science , volume =. 2009 , publisher =

    2009

  24. [24]

    The Journal of Finance , volume =

    Jagannathan, Ravi and Ma, Tongshu , title =. The Journal of Finance , volume =. 2003 , publisher =

    2003

  25. [25]

    arXiv preprint arXiv:2505.10099 , year=

    A Scalable Gradient-Based Optimization Framework for Sparse Minimum-Variance Portfolio Selection , author=. arXiv preprint arXiv:2505.10099 , year=

    work page arXiv

  26. [26]

    1996 , publisher=

    A first course in optimization theory , author=. 1996 , publisher=

    1996

  27. [27]

    Naval Research Logistics Quarterly , volume=

    An algorithm for quadratic programming , author=. Naval Research Logistics Quarterly , volume=. 1956 , publisher=

    1956

  28. [28]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume=

    Regression shrinkage and selection via the lasso , author=. Journal of the Royal Statistical Society: Series B (Methodological) , volume=. 1996 , publisher=

    1996

  29. [29]

    2009 , publisher=

    The elements of statistical learning: data mining, inference, and prediction , author=. 2009 , publisher=

    2009

  30. [30]

    International Conference on Learning Representations (ICLR) , year=

    Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding , author=. International Conference on Learning Representations (ICLR) , year=

  31. [31]

    Proceedings of Machine Learning and Systems , volume=

    What is the state of neural network pruning? , author=. Proceedings of Machine Learning and Systems , volume=

  32. [32]

    Econometrica , volume=

    Determining the number of factors in approximate factor models , author=. Econometrica , volume=. 2002 , publisher=

    2002

  33. [33]

    Journal of Banking & Finance , volume=

    Sparse portfolio selection via the sorted _1 -Norm , author=. Journal of Banking & Finance , volume=. 2020 , publisher=

    2020

  34. [34]

    The Journal of Portfolio Management , volume=

    The effect of errors in means, variances, and covariances on optimal portfolio choice , author=. The Journal of Portfolio Management , volume=. 1993 , publisher=

    1993

  35. [35]

    Proceedings of the National Academy of Sciences , volume=

    Sparse and stable Markowitz portfolios , author=. Proceedings of the National Academy of Sciences , volume=. 2009 , publisher=

    2009

  36. [36]

    The Journal of Finance , volume=

    When will mean-variance efficient portfolios be well diversified? , author=. The Journal of Finance , volume=. 1992 , publisher=

    1992

  37. [37]

    Computers & Operations Research , volume=

    Heuristics for cardinality constrained portfolio optimisation , author=. Computers & Operations Research , volume=. 2000 , publisher=

    2000

  38. [38]

    Mathematical Programming , volume =

    Bienstock, Daniel , title =. Mathematical Programming , volume =

  39. [39]

    and Lee, Sangkyun and Bogdan, Ma

    Kremer, Philipp J. and Lee, Sangkyun and Bogdan, Ma. Sparse portfolio selection via the sorted _1 -. Journal of Banking & Finance , volume =

  40. [40]

    Operations Research , volume =

    Gao, Jianjun and Li, Duan , title =. Operations Research , volume =

  41. [41]

    Matematički Vesnik , volume =

    Rakočević, Vladimir , title =. Matematički Vesnik , volume =

  42. [42]

    The Annals of Statistics , volume=

    Nonlinear shrinkage estimation of large-dimensional covariance matrices , author=. The Annals of Statistics , volume=. 2012 , publisher=

    2012

  43. [43]

    Physical Review Letters , volume =

    Laloux, Laurent and Cizeau, Pierre and Bouchaud, Jean-Philippe and Potters, Marc , title =. Physical Review Letters , volume =

  44. [44]

    Operations Research , volume =

    Hazimeh, Hussein and Mazumder, Rahul , title =. Operations Research , volume =. 2020 , doi =

    2020

  45. [45]

    2004 , publisher =

    Convex Optimization , author =. 2004 , publisher =

    2004

This paper was first reviewed by grok-4.3 on June 26, 2026.