pith. sign in

arxiv: 2406.19516 · v2 · submitted 2024-06-27 · 🧮 math.CO · cs.DM· math.OC

Almost Orthogonal Arrays: Search Three Ways

Pith reviewed 2026-05-23 23:53 UTC · model grok-4.3

classification 🧮 math.CO cs.DMmath.OC
keywords almost orthogonal arraysinteger programminglocal searchalgebraic constructionsnon-orthogonality measurescombinatorial designsexperimental design
0
0 comments X

The pith

Three search methods produce almost orthogonal arrays that improve on existing constructions for multiple non-orthogonality measures.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Orthogonal arrays are difficult to build exactly for many parameter choices, so the paper studies their relaxation to almost orthogonal arrays that permit controlled departures from perfect orthogonality. The authors generate new examples by formulating the search as an integer program, by running local-search meta-heuristics, and by using algebraic constructions. When the resulting arrays are scored on several different non-orthogonality measures, many of them match or beat the best previously published arrays. All newly found arrays are placed in a public repository. The work therefore supplies a practical route to obtaining usable near-orthogonal designs without requiring exact orthogonality.

Core claim

Integer programming, local-search meta-heuristics, and algebraic methods each locate almost orthogonal arrays whose quality, measured by several non-orthogonality criteria, is competitive with or strictly better than the arrays already recorded in the literature; the complete collection of new arrays is deposited in a public repository.

What carries the argument

Almost orthogonal arrays, obtained by relaxing the pairwise balance conditions of orthogonal arrays and located through integer-programming models, local-search heuristics, and algebraic constructions.

If this is right

  • Researchers needing near-orthogonal designs now have additional explicit arrays for many parameter sets.
  • When one of the three search techniques fails to improve an existing array, the other two remain available.
  • The public repository supplies a growing baseline that future constructions can be compared against.
  • The same three techniques can be applied immediately to parameter combinations not yet examined.

Where Pith is reading between the lines

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

  • Hybrid algorithms that combine integer programming with local search or algebraic insight could locate still better arrays.
  • Direct benchmarking inside a specific application domain would test whether lower non-orthogonality scores translate into measurable gains in that domain.
  • Algebraic constructions may expose structural patterns that purely numerical search methods overlook.

Load-bearing premise

The chosen non-orthogonality measures are reliable indicators of how useful the arrays will be in downstream applications.

What would settle it

A concrete experimental-design task in which an array with a worse score on every non-orthogonality measure used in the paper nevertheless produces better statistical results than the improved arrays would falsify the practical significance of the reported gains.

read the original abstract

Orthogonal arrays play a fundamental role in many applications. However, constructing orthogonal arrays with the required parameters for an application usually is extremely difficult and, sometimes, even impossible. Hence there is an increasing need for a relaxation of orthogonal arrays to allow a wider flexibility. The latter has lead to various types of arrays under the name of ``nearly-orthogonal arrays'', and less often ``almost orthogonal arrays''. In this paper, we explore how to find almost orthogonal arrays three ways: using integer programming, local search meta-heuristics and algebraic methods. We compare all our search results with the ones existing in the literature, and we show that they are competitive, improving some of the existing arrays for many non-orthogonality measures. All our found almost orthogonal arrays are available at a public repository.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The paper investigates constructions of almost orthogonal arrays via three distinct search paradigms: integer programming, local-search metaheuristics, and algebraic techniques. It reports that the resulting arrays are competitive with, and in several cases improve upon, previously tabulated examples when evaluated under multiple non-orthogonality measures; all constructed arrays are deposited in a public repository.

Significance. If the numerical comparisons hold, the work supplies immediately usable arrays for experimental-design settings where exact orthogonality is unattainable, together with a reproducible data release that supports downstream verification and reuse. The multi-method strategy itself is a modest methodological contribution.

major comments (2)
  1. [§4, Tables 2–5] §4 (Results), Tables 2–5: the improvement claims rest on direct numerical comparison of non-orthogonality measures between newly constructed arrays and literature references. The manuscript does not supply the exact evaluation scripts or the precise floating-point implementations used for the reference arrays; an undetected difference in measure definition or rounding would falsify the reported gains. Because the arrays themselves are public, this verification gap is fixable but currently load-bearing for the central empirical claim.
  2. [§3.1] §3.1 (Integer Programming formulation): the objective function and the encoding of the non-orthogonality penalties are stated at a high level; the precise linearization or big-M constants employed for each measure are not given. Without these details, independent reproduction of the IP solutions (and therefore of the claimed improvements) cannot be confirmed.
minor comments (3)
  1. The repository URL is mentioned only in the abstract; it should also appear in the main text (e.g., at the end of §4) with a permanent identifier or commit hash.
  2. Notation for the various non-orthogonality measures is introduced piecemeal; a single consolidated table or subsection collecting all definitions would improve readability.
  3. A few figure captions (e.g., Figure 3) omit the precise parameter tuple (v,k,λ,…) of the displayed array, forcing the reader to cross-reference the tables.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and the recommendation of minor revision. The comments correctly identify two reproducibility gaps that we will close by expanding the manuscript and supplementing the public repository. We address each point below.

read point-by-point responses
  1. Referee: [§4, Tables 2–5] §4 (Results), Tables 2–5: the improvement claims rest on direct numerical comparison of non-orthogonality measures between newly constructed arrays and literature references. The manuscript does not supply the exact evaluation scripts or the precise floating-point implementations used for the reference arrays; an undetected difference in measure definition or rounding would falsify the reported gains. Because the arrays themselves are public, this verification gap is fixable but currently load-bearing for the central empirical claim.

    Authors: We agree that the absence of the evaluation scripts creates a verification gap. In the revised manuscript we will add an appendix containing the complete Python scripts used to compute every non-orthogonality measure, including the exact floating-point implementations, rounding conventions, and measure definitions employed for both our arrays and the literature references. These scripts, together with the already-public arrays, will be deposited in the repository so that every tabulated comparison can be reproduced exactly. revision: yes

  2. Referee: [§3.1] §3.1 (Integer Programming formulation): the objective function and the encoding of the non-orthogonality penalties are stated at a high level; the precise linearization or big-M constants employed for each measure are not given. Without these details, independent reproduction of the IP solutions (and therefore of the claimed improvements) cannot be confirmed.

    Authors: We accept that the current presentation of the IP model is insufficient for independent reproduction. The revised §3.1 will contain the fully linearized objective function, the explicit big-M constants chosen for each penalty term, the complete set of auxiliary variables and constraints, and the Gurobi parameter settings used to obtain the reported solutions. revision: yes

Circularity Check

0 steps flagged

No circularity: purely computational search with external comparisons

full rationale

The paper reports results from three independent search algorithms (integer programming, local search, algebraic methods) applied to construct almost orthogonal arrays, followed by direct numerical comparison against published literature examples using explicitly defined non-orthogonality measures. No derivation chain, fitted parameters renamed as predictions, self-referential definitions, or load-bearing self-citations appear; the central claims rest on the correctness of the search outputs and the reproducibility of the measure calculations, which are external to any internal reduction. This matches the default expectation for computational enumeration papers.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The work rests on the standard combinatorial definitions of orthogonal arrays and non-orthogonality measures drawn from prior literature; no new free parameters, axioms, or invented entities are introduced.

axioms (1)
  • domain assumption Standard definitions of orthogonal arrays and almost orthogonal arrays from combinatorial design theory
    The paper presupposes these definitions to frame the search problem.

pith-pipeline@v0.9.0 · 5674 in / 1067 out tokens · 16075 ms · 2026-05-23T23:53:13.437153+00:00 · methodology

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

68 extracted references · 68 canonical work pages

  1. [1]

    ARINA. (2021). Computational cluster from IZO-SGI, SGIker, University of the Basque Country

  2. [2]

    Bolboac˘ a & Lorentz J¨ antschi

    Sorana D. Bolboac˘ a & Lorentz J¨ antschi. (2007). Design of experiments: Useful orthog- onal arrays for number of experiments from 4 to 16. Entropy, 9(4), 198–232. https : //doi.org/10.3390/e9040198

  3. [3]

    Kathleen H. V. Booth & D. R. Cox. (1962). Some systematic supersaturated designs. Technometrics, 4, 489–495. https://doi.org/10.2307/1266285

  4. [4]

    R. C. Bose & K. A. Bush. (1952). Orthogonal arrays of strength two and three. Ann. Math. Statistics, 23, 508–524. https://doi.org/10.1214/aoms/1177729331

  5. [5]

    Brent, Judy-anne H

    Richard P. Brent, Judy-anne H. Osborn, & Warren D. Smith. (2015). Note on best possible bounds for determinants of matrices close to the identity matrix. Linear Algebra Appl., 466, 21–26. https://doi.org/10.1016/j.laa.2014.09.041

  6. [7]

    Enrico Carlini & Giovanni Pistone. (2007). Hilbert bases for orthogonal arrays. Journal of Statistical Theory and Practice , 1(3-4), 299–309. https://doi.org/10.1080/15598608. 2007.10411842

  7. [8]

    Charles Colbourn & Jeffrey Dinitz (Eds.). (2007). Handbook of combinatorial designs (Second). Chapman & Hall/CRC, Boca Raton, FL

  8. [9]

    Cox, John Little, & Donal O’Shea

    David A. Cox, John Little, & Donal O’Shea. (2015). Ideals, varieties, and algorithms (Fourth) [An introduction to computational algebraic geometry and commutative alge- bra]. Springer, Cham. https://doi.org/10.1007/978-3-319-16721-3

  9. [10]

    Lih-Yuan Deng, Dennis K. J. Lin, & Jiannong Wang. (1999). A resolution rank crite- rion for supersaturated designs. Statist. Sinica, 9(2), 605–610. 37

  10. [11]

    Delphine Dupuy, C´ eline Helbert, & Jessica Franco. (2015). DiceDesign and DiceEval: Two R packages for design and analysis of computer experiments. Journal of Statistical Software, 65(11), 1–38. https://doi.org/10.18637/jss.v065.i11

  11. [12]

    Kaitai Fang, Gennian Ge, & Minqian Liu. (2004). Construction of optimal super- saturated designs by the packing method. Sci. China Ser. A , 47(1), 128–143. https: //doi.org/10.1360/02ys0271

  12. [13]

    Kai-Tai Fang, Gennian Ge, Min-Qian Liu, & Hong Qin. (2004). Combinatorial con- structions for optimal supersaturated designs. Discrete Math., 279(1-3), 191–202. https: //doi.org/10.1016/S0012-365X(03)00269-3

  13. [14]

    Kai-Tai Fang, Dennis K. J. Lin, & Min-Qian Liu. (2003). Optimal mixed-level super- saturated design. Metrika, 58(3), 279–291. https://doi.org/10.1007/s001840300266

  14. [15]

    Kai-Tai Fang, Dennis K. J. Lin, & Chang-Xing Ma. (2000). On the construction of multi-level supersaturated designs. J. Statist. Plann. Inference , 86(1), 239–252. https: //doi.org/10.1016/S0378-3758(99)00163-9

  15. [16]

    Kai-Tai Fang, Dennis K. J. Lin, Peter Winker, & Yong Zhang. (2000). Uniform design: Theory and application. Technometrics, 42(3), 237–248. https : / / doi . org / 10 . 2307 / 1271079

  16. [17]

    Kai-Tai Fang, Xuan Lu, Yu Tang, & Jianxing Yin. (2004). Constructions of uniform designs by using resolvable packings and coverings. Discrete Math. , 274(1-3), 25–40. https://doi.org/10.1016/S0012-365X(03)00100-6

  17. [18]

    Kai-Tai Fang, Xuan Lu, & Peter Winker. (2003). Lower bounds for centered and wrap- around L2-discrepancies and construction of uniform designs by threshold accepting. J. Complexity, 19(5), 692–711. https://doi.org/10.1016/S0885-064X(03)00067-0

  18. [19]

    Kai-Tai Fang, Chang-Xing Ma, & Peter Winker. (2002). Centered L2-discrepancy of random sampling and Latin hypercube design, and construction of uniform designs. Math. Comp., 71(237), 275–296. https://doi.org/10.1090/S0025-5718-00-01281-3

  19. [20]

    Kai-Tai Fang, Yu Tang, & Jianxing Yin. (2005). Lower bounds for wrap-around L2- discrepancy and constructions of symmetrical uniform designs. J. Complexity , 21(5), 757–771. https://doi.org/10.1016/j.jco.2005.01.003

  20. [21]

    A. J. Geyer, D. A. Bulutoglu, & S. J. Rosenberg. (2014). The LP relaxation orthogonal array polytope and its permutation symmetries. J. Comb. Math. Comb. Comput. , 91, 165–176

  21. [22]

    Geyer, Dursun A

    Andrew J. Geyer, Dursun A. Bulutoglu, & Kenneth J. Ryan. (2019). Finding the symmetry group of an LP with equality constraints and its application to classifying orthogonal arrays. Discrete Optim., 32, 93–119. https://doi.org/10.1016/j.disopt.2019. 01.001

  22. [23]

    Gopalakrishnan, Douglas R

    K. Gopalakrishnan, Douglas R. Stinson, & David Cheriton. (2006, December). Ap- plications of orthogonal arrays to computer science. In R. Balakrishnan & C. E. Veni Madhavan (Eds.), Discrete Mathematics. Proceedings of the international conference on discrete mathematics, Indian Institute of Science, Bangalore, December 2006 (pp. 149– 164, Vol. 7). Interna...

  23. [24]

    Ulrike Gr¨ omping. (2018). R package DoE.base for factorial experiments. Journal of Statistical Software, 85(5), 1–41. https://doi.org/10.18637/jss.v085.i05 38

  24. [25]

    Samad Hedayat, Neil J.A

    A. Samad Hedayat, Neil J.A. Sloane, & John Stufken. (1999). Orthogonal Arrays: Theory and Practice. Springer Science+Business Media, LLC

  25. [26]

    Hickernell

    Fred J. Hickernell. (1998a). A generalized discrepancy and quadrature error bound. Math. Comp., 67(221), 299–322. https://doi.org/10.1090/S0025-5718-98-00894-1

  26. [27]

    Fred J Hickernell. (1998b). Lattice rules: How well do they measure up? In Random and quasi-random point sets (pp. 109–166). Springer

  27. [28]

    Hickernell

    Fred J. Hickernell. (2000). What affects the accuracy of quasi-Monte Carlo quadrature? In Monte Carlo and quasi-Monte Carlo methods 1998 (Claremont, CA) (pp. 16–55). Springer, Berlin

  28. [29]

    Hickernell & Min-Qian Liu

    Fred J. Hickernell & Min-Qian Liu. (2002). Uniform designs limit aliasing.Biometrika, 89(4), 893–904. https://doi.org/10.1093/biomet/89.4.893

  29. [30]

    Klaus Hinkelmann & Oscar Kempthorne. (2008). Design and analysis of experiments. Vol. 1. introduction to experimental design (Second). Wiley-Interscience [John Wiley & Sons], Hoboken, NJ

  30. [31]

    A Kaveh & M Ilchi Ghazaan. (2018). A new hybrid meta-heuristic algorithm for optimal design of large-scale dome structures. Engineering Optimization , 50(2), 235– 252

  31. [32]

    Ali Kaveh. (2014). Advances in metaheuristic algorithms for optimal design of struc- tures. Springer

  32. [33]

    (2018).Meta-heuristic algorithms for optimal design of real-size structures

    Ali Kaveh & Majid Ilchi Ghazaan. (2018).Meta-heuristic algorithms for optimal design of real-size structures. Springer

  33. [34]

    Xiao Ke, Rong Zhang, & Hua-Jun Ye. (2015). Two- and three-level lower bounds for mixture L2-discrepancy and construction of uniform designs by threshold accepting. J. Complexity, 31(5), 741–753. https://doi.org/10.1016/j.jco.2015.01.002

  34. [35]

    Jianfa Lai, Kai-Tai Fang, Xiaoling Peng, & Yuxuan Lin. (2024). Construction of uni- form designs over continuous domain in computer experiments. Comm. Statist. Simula- tion Comput., 53(1), 130–146. https://doi.org/10.1080/03610918.2021.2011924

  35. [36]

    Dennis K. J. Lin. (1993). A new class of supersaturated designs.Technometrics, 35(1), 28–31. Retrieved March 2, 2024, from http://www.jstor.org/stable/1269286

  36. [37]

    Dennis K. J. Lin. (1995). Generating systematic supersaturated designs. Technomet- rics, 37(2), 213–225. Retrieved March 2, 2024, from http://www.jstor.org/stable/ 1269622

  37. [38]

    Yuan-Lung Lin, Frederick Kin Hing Phoa, & Ming-Hung Kao. (2017). Optimal design of fMRI experiments using circulant (almost-)orthogonal arrays. Ann. Statist. , 45(6), 2483–2510. https://doi.org/10.1214/16-AOS1531

  38. [39]

    Hickernell

    Min-Qian Liu & Fred J. Hickernell. (2002). E(s2)-optimality and minimum discrep- ancy in 2-level supersaturated designs. Statist. Sinica, 12(3), 931–939

  39. [40]

    Xuan Lu, William Li, & Miao Xie. (2006). A class of nearly orthogonal arrays.Journal of quality technology, 38(2), 148–161

  40. [41]

    Luenberger & Yinyu Ye

    David G. Luenberger & Yinyu Ye. (2008).Linear and Nonlinear Programming (Third, Vol. 116). Springer, New York

  41. [42]

    Chang-Xing Ma & Kai-Tai Fang. (2001). A note on generalized aberration in factorial designs. Metrika, 53(1), 85–93. https://doi.org/10.1007/s001840100112 39

  42. [43]

    Chang-Xing Ma, Kai-Tai Fang, & Erkki Liski. (2000). A new approach in constructing orthogonal and nearly orthogonal arrays. Metrika, 50, 255–268

  43. [44]

    Fran¸ cois Margot. (2002). Pruning by isomorphism in branch-and-cut.Math. Program., 94, 71–90. https://doi.org/10.1007/s10107-002-0358-2

  44. [45]

    Fran¸ cois Margot. (2003a). Exploiting orbits in symmetric ILP [Integer programming (Pittsburgh, PA, 2002)]. Math. Program., 98, 3–21. https://doi.org/10.1007/s10107- 003-0394-6

  45. [46]

    Fran¸ cois Margot. (2003b). Small covering designs by branch-and-cut [The Aussois 2000 Workshop in Combinatorial Optimization]. Math. Program., 94, 207–220. https: //doi.org/10.1007/s10107-002-0316-z

  46. [47]

    Fran¸ cois Margot. (2007). Symmetric ILP: Coloring and small integers. Discrete Op- tim., 4(1), 40–62. https://doi.org/10.1016/j.disopt.2006.10.008

  47. [48]

    Luca Mariot, Stjepan Picek, Domagoj Jakobovic, & Alberto Leporati. (2018). Evolu- tionary search of binary orthogonal arrays. Parallel Problem Solving from Nature–PPSN XV: 15th International Conference, Coimbra, Portugal, September 8–12, 2018, Proceed- ings, Part I 15 , 121–133

  48. [49]

    Luis Mart´ ınez, Mar´ ıa Merino, & Juan Manuel Montoya. (2023). An integer program- ming model for obtaining cyclic quasi-difference matrices.Oper. Res. Perspect., 10, Paper No. 100260, 6. https://doi.org/10.1016/j.orp.2022.100260

  49. [50]

    Michel Minoux. (1986). Mathematical programming: Theory and algorithms. John Wi- ley & Sons, Ltd., Chichester

  50. [51]

    Art B. Owen. (1992). Orthogonal arrays for computer experiments, integration and visualization. Statist. Sinica, 2(2), 439–452

  51. [52]

    R Core Team. (2020). R: A Language and Environment for Statistical Computing . R Foundation for Statistical Computing. Vienna, Austria. https://www.R-project.org/

  52. [53]

    Rosenberg

    Steven J. Rosenberg. (1995). A large index theorem for orthogonal arrays, with bounds. Discrete Math., 137(1-3), 315–318. https://doi.org/10.1016/0012-365X(95) 91428-S

  53. [54]

    Siegfried M. Rump. (2018). Estimates of the determinant of a perturbed identity matrix. Linear Algebra Appl., 558, 101–107. https://doi.org/10.1016/j.laa.2018.08.009

  54. [55]

    The Uniform Design. (2004). The Uniform Design Association of China

  55. [56]

    (2021, November)

    User’s Manual for IBM ILOG CPLEX Optimization Studio 20.1.0 [Available at https: //www.ibm.com/docs/en/icos/20.1.0]. (2021, November). IBM

  56. [57]

    H´ elcio Vieira Jr., Susan Sanchez, Karl Heinz Kienitz, & Mischel Carmen Neyra Belder- rain. (2011). Generating and improving orthogonal designs by using mixed integer pro- gramming. European J. Oper. Res. , 215(3), 629–638. https://doi.org/10.1016/j.ejor. 2011.07.005

  57. [58]

    J. C. Wang & C. F. Jeff Wu. (1992). Nearly orthogonal arrays with mixed levels and small runs. Technometrics, 34(4), 409–422

  58. [59]

    P. C. Wang & H. W. Jan. (1995). Designing two-level factorial experiments using orthogonal arrays when the run order is important. Journal of the Royal Statistical Society. Series D (The Statistician) , 44(3), 379–388. 40

  59. [60]

    Yuan Wang & Kai Tai Fang. (1981). A note on uniform distribution and experimental design. Kexue Tongbao (English Ed.) , 26(6), 485–489

  60. [61]

    Wayne L. Winston. (1987). Operations Research: Applications and Algorithms . Duxbury Press, Boston, MA

  61. [62]

    Wolsey & George L

    Laurence A. Wolsey & George L. Nemhauser. (1999). Integer and combinatorial opti- mization (Vol. 55). John Wiley & Sons

  62. [63]

    C. F. J. Wu. (1993). Construction of supersaturated designs through partially aliased interactions. Biometrika, 80(3), 661–669. Retrieved March 2, 2024, from http://www. jstor.org/stable/2337185

  63. [64]

    C. F. Jeff Wu, Run Chu Zhang, & Renguan Wang. (1992). Construction of asym- metrical orthogonal arrays of the type OA(sk, sm(sr1)n1 · · ·(srt)nt). Statist. Sinica, 2(1), 203–219

  64. [65]

    Hongquan Xu. (2002). An algorithm for constructing orthogonal and nearly- orthogonal arrays with mixed levels and small runs. Technometrics, 44(4), 356–368

  65. [66]

    Hongquan Xu & C. F. J. Wu. (2001). Generalized minimum aberration for asymmet- rical fractional factorial designs. Ann. Statist., 29(2), 549–560. https://doi.org/10.1214/ aos/1009210552

  66. [67]

    Shu Yamada & Dennis K. J. Lin. (2002). Construction of mixed-level supersaturated design. Metrika, 56(3), 205–214. https://doi.org/10.1007/s001840100173

  67. [68]

    Shu Yamada & Tomomi Matsui. (2002). Optimality of mixed-level supersaturated designs. J. Statist. Plann. Inference , 104(2), 459–468. https://doi.org/10.1016/S0378- 3758(01)00248-8

  68. [69]

    Yong-Dao Zhou, Kai-Tai Fang, & Jian-Hui Ning. (2013). Mixture discrepancy for quasi-random point sets. J. Complexity , 29(3-4), 283–301. https://doi.org/10.1016/j. jco.2012.11.006 A Proofs for Section 2 and Section 3 Proof of Proposition 2.4. By its definition, n(A, x, j) counts the number of times that x ∈ St appears as a row in A[j] := A[j1, . . . , jt]...