REVIEW 6 minor 28 references
Some new Constructions of Coded Caching Schemes with Reduced Subpacketization
T0 review · 0 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read A placement-delivery array exists precisely when three binary matrices satisfy five local conditions, and this equivalence yields caching schemes with linear subpacketization.
desk verdict A competent, verified set of combinatorial constructions for low-subpacketization coded caching; the core equivalence is a reformulation, the new families are real, and the main caveat is honest conditionality on design existence. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the placement-delivery array (PDA): an $F\times K$ array whose entries are stars or integers $1,\dots,S$, with each column containing exactly $Q$ stars, each integer appearing at least once, and no integer appearing in the same row or column twice with the other position starred. The load-bearing machinery of the paper is Theorem 1, which replaces the PDA by three binary matrices $C_{X,Y}$, $C_{X,Z}$, and $C_{Y,Z}$ satisfying E1–E5; this rewrites the existence of a caching scheme as a purely local consistency problem. A relaxed sufficient condition (E1–E3 plus E6) is then used as the construction tool: E6 makes an associated bipartite graph regular, so a perfect matching, supplied by a standard graph-theory lemma, selects the third matrix. The paper feeds this machine with incidence matrices coming from projective geometries, configurations, and $t$-designs, whose regularity properties automatically provide the required matching conditions.
What would settle it
To test the central equivalence, enumerate all $(K,F,Q,S)$ PDAs for small parameters, enumerate all binary-matrix triples satisfying E1–E5, and check that the two lists match; any mismatch would refute Theorem 1. For the design-based constructions, take a concrete parameter triple, say a configuration $(v_r,b_k)$ with $k$ near the bound $(v-1)/(r+1)$, and check whether such a configuration exists; if it does not, that instance of the scheme is vacuous even though the conditional theorem remains true. For Theorem 7, instantiate a specific $t$-$ (v,k,\lambda)$ design and verify that the three computed parameter sets give integer subpacketization and rate values.
Extended reading notes
Core claim
On its own terms, the paper proves that there exists a $(K,F,Q,S)$ placement-delivery array if and only if there exist three binary matrices $C_{X,Y}$, $C_{X,Z}$, and $C_{Y,Z}$, with $X$ an $F$-set, $Y$ an $S$-set, and $Z$ a $K$-set, satisfying conditions E1–E5. Condition E1 fixes the number of non-star entries in each column, E2 says every delivery symbol is used at least once, and E3–E5 state that each occurrence of a 1 in one matrix is matched to exactly one 1 in each of the other two matrices, so the three matrices are mutually consistent. Given such matrices, the PDA is recovered by placing the symbol $y\in Y$ at position $(x,z)$ exactly when $C_{X,Z}(x,z)=C_{X,Y}(x,y)=C_{Y,Z}(y,z)=1$. Armed with this characterization, the paper constructs matrices from the incidence structure of projective geometries over finite fields, from $(v_r,b_k)$ configurations, and from $t$-designs; each construction yields three families of PDAs, hence three caching schemes, with explicit parameters for memory fraction and rate. In the configuration-based construction, a $(v,k,1)$-BIBD gives a scheme with $K=F=v$, i.e., linear subpacketization, and the paper computes how far its rate sits above the optimal baseline. The final construction shows that the direct product of a $(K_1,F_1,Q_1,S_1)$ PDA and a $(K_2,F_2,Q_2,S_2)$ PDA is a $(K_1K_2,F_1F_2,F_1Q_2+F_2Q_1-Q_1Q_2,S_1S_2)$ PDA, giving a corresponding combined caching scheme.
Load-bearing premise
The caching schemes in Theorems 4–7 are conditional on the existence of the combinatorial designs they name—a $(v_r,b_k)$ configuration or a $t$-$ (v,k,\lambda)$ design with the stated parameters—and the paper does not prove that the particular parameter ranges needed for linear subpacketization are actually realized; it relies on standard design-theory existence results.
Editorial extensions
If this is right
- From any configuration $(v_r,b_k)$ one obtains three caching schemes; when the configuration is a $(v,k,1)$-BIBD, one of them has $K=F=v$ (linear subpacketization) and rate approximately $K(1-M/N)^2/(2-M/N)$, which grows linearly with $K$.
- From any $t$-$ (v,k,1)$-design with $t\le k/2+1$, and from any $t$-$ (v,k,\lambda)$-design with $t_1+t_2\le t$, one obtains three explicit caching schemes whose parameters are given by binomial coefficients and the design's lambda counts.
- The direct-product rule means any two known PDA-based schemes can be combined into a scheme with user count $K_1K_2$, file size $F_1F_2$, memory fraction $Q_1/F_1+Q_2/F_2-(Q_1/F_1)(Q_2/F_2)$, and rate $(S_1/F_1)(S_2/F_2)$.
- The projective-geometry construction yields three families of PDAs parameterized by Gaussian binomial coefficients, one of which coincides with a previously known line-graph construction, so the new equivalence subsumes that result as a special case.
Reading between the lines
- The matrix equivalence turns PDA design into a constraint-satisfaction problem, so a natural next step is computational search over small binary-matrix triples to discover low-subpacketization schemes outside the named design families; the paper does not run such a search.
- Iterating the direct product yields schemes whose subpacketization is the product of the factors while the rate multiplies; this gives a concrete knob for trading rate against file size that the paper notes only in passing.
- The rate analysis for linear subpacketization is carried out for the BIBD case; an implicit testable question is whether other configurations with block size closer to the bound $(v-1)/(r+1)$ yield better rate at the same linear subpacketization, since the paper's parameter formulas would apply directly.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies centralized coded caching with reduced subpacketization within the placement delivery array (PDA) framework. Its main contributions are: (i) Theorem 1, an equivalence between the existence of a (K,F,Q,S) PDA and the existence of three binary matrices satisfying five conditions (E1-E5); (ii) Theorem 2 and Corollary 1, sufficient conditions based on regular bipartite graphs and perfect matchings that yield three PDA parameter sets from a single triple of matrices; (iii) concrete PDA constructions from projective geometries over finite fields (Theorem 3), combinatorial configurations (Theorem 4), t-(v,k,1)-designs (Theorems 5 and 6), and t-(v,k,lambda)-designs (Theorem 7); and (iv) a direct product operation on PDAs (Theorem 8) that produces new caching schemes from existing ones. The paper derives explicit parameters, memory fractions, and rates for each family, and shows that several known constructions appear as special cases.
Significance. The central claims are sound and the paper makes a solid contribution to the low-subpacketization coded caching literature. The matrix-based characterization of PDAs in Theorem 1 is a genuinely useful reformulation, and the subsequent design-based constructions are explicit and checkable. I verified the uniqueness arguments in the proofs of Theorems 1 and 2, the perfect-matching step in Theorem 2, the Gaussian-binomial counting in Section IV, and the parameter simplifications in Theorems 5-7; they are correct. The constructions are conditional on the existence of configurations or t-designs, but this is standard in combinatorial construction papers, and known infinite families such as Steiner triple systems realize the linear-subpacketization regime. The direct product construction is simple and useful. The paper generalizes several known schemes and enriches the available tradeoff between subpacketization and rate.
minor comments (6)
- [Section III, Corollary 1] The relabeling instructions in the proof of Corollary 1 appear to be misstated: as written, the permutations do not produce the parameter sets that are claimed. For example, parameter set 1) is obtained by the cyclic relabeling X->Y, Y->Z, Z->X, not by 'relabelling X by Z, Y by X, and Z by Y'. Please correct the stated relabelings or the resulting parameter formulas.
- [Theorem 8 proof] In the proof of Theorem 8, the expression 'Q = F1Q2 + F2Q1 - Q1Q1' should be 'Q = F1Q2 + F2Q1 - Q1Q2'; likewise '|X1Q2' should be '|X1|Q2'.
- [Section II] The configuration bound is displayed as 'k <= v-1 r +1', which is ambiguous; it should be written as 'k <= (v-1)/r + 1'.
- [References] Reference [25] misspells the second author's name as 'Denitz'; it should be 'Dinitz'.
- [Remark 1] In Remark 1, the phrase 'the loss in R = K(...' should likely be 'the rate R = ...' or 'the rate increase is ...'; the current wording is awkward.
- [Corollary 2] Corollary 2 says there exists a scheme 'for any (K,M,N) caching system' with the stated parameters, but K and M/N are fixed by the construction; the wording should be adjusted to avoid implying that arbitrary (K,M,N) are supported.
Circularity Check
No significant circularity: the constructions are conditional derivations from external combinatorial designs, and the PDA equivalence is proved self-containedly.
full rationale
The paper's central equivalence (Theorem 1) is proved directly in Appendix A from the PDA definition: the necessity direction constructs the three binary matrices from a PDA, and the sufficiency direction constructs a PDA from matrices satisfying E1–E5. No step of that proof assumes the conclusion or a cited version of the same theorem. The PDA-to-scheme translation is Lemma 2, taken from [15], which is an external prior result used as a black box; it is not a self-citation, and it is not fitted. The constructions in Theorems 3–7 apply Corollary 1 by explicitly verifying conditions E1', E2', E3, E6, and E7 for matrices derived from projective geometries, configurations, and t-designs. The resulting parameters (K, F, Q, S, M/N, R) are closed-form functions of the input design parameters, not fitted values or renamed targets. Theorems 4–7 are explicitly conditional on the existence of configurations or t-designs, with standard external references [24], [25] for existence; this is an external-support matter, not circularity, and known infinite families (e.g., Steiner systems) instantiate the hypotheses. The acknowledged difficulty in expressing (k choose t0) in terms of K and M/N is a limitation of the resulting rate formula, not a circular dependence. The direct-product construction in Theorem 8 also derives its parameters algebraically from two given PDAs, using the proved Theorem 1 rather than smuggling in the result. No load-bearing self-citation or fitted-input-as-prediction pattern appears.
Assumptions & free parameters
free parameters (3)
- Projective geometry parameters (q, k, m, t)
- Configuration parameters (v, r, k, b)
- t-design parameters (v, k, t, lambda, t0, t1, t2)
assumptions (5)
- standard math Lemma 1: every k-regular bipartite graph with k>0 has a perfect matching.
- domain assumption PDA-to-scheme translation (Lemma 2, from [15]): a (K,F,Q,S) PDA yields an F-division caching scheme with M/N=Q/F and rate R=S/F.
- standard math Counting identities for t-designs (Equation 1): number of blocks containing an s-subset is lambda_s = lambda * C(v-s,t-s)/C(k-s,t-s).
- standard math Gaussian binomial coefficient formulas (Lemma 3): numbers of subspaces with given dimensions over finite fields.
- domain assumption Existence of configurations and t-designs for the parameter sets used.
Cite this review
Pith. "Pith review of Some new Constructions of Coded Caching Schemes with Reduced Subpacketization." pith.science (2026). https://pith.science/paper/HCAEOAW7
@misc{pith2026190806570,
author = {Pith},
title = {Pith review of: Some new Constructions of Coded Caching Schemes with Reduced Subpacketization},
year = {2026},
howpublished = {\url{https://pith.science/paper/HCAEOAW7}},
note = {Machine review of arXiv:1908.06570}
}
read the original abstract
We study the problem of constructing centralized coded caching schemes with low subpacketization level based on the placement delivery array (PDA) design framework. PDA design is an efficient way to construct centralized coded caching schemes and most existing schemes, including the famous Maddah-Ali-Niesen scheme, can be described using PDA. In this paper, we first prove that constructing a PDA is equivalent to constructing three binary matrices that satisfy certain conditions. From this perspective, we then propose some new constructions of coded caching schemes using PDA design based on projective geometries over finite fields, combinatorial configurations, and t-designs, respectively. Our constructions achieve low subpacketization level (e.g., linear subpacketization) with reasonable rate loss and include several known results as special cases. Finally, we give an approach to construct new coded caching scheme from existing schemes based on direct product of PDAs. Our results enrich the coded caching schemes of low subpacketization level.
Reference graph
Works this paper leans on
-
[17]
Centralized Coded Ca ching Schemes: A Hypergraph Theoretical Approach,
C. Shangguan, Y . Zhang, and G. Ge, “Centralized Coded Ca ching Schemes: A Hypergraph Theoretical Approach,” IEEE Trans. Inf. Theory , vol. 64, no. 8, pp. 5755-5766, Aug. 2018
work page 2018
-
[19]
Coded caching via line graphs of bipartit e graphs,
P . Krishnan, “Coded caching via line graphs of bipartit e graphs,” in Proc. IEEE Information Theory W orkshop (ITW) , 2018, pp. 1-5. 30
work page 2018
-
[20]
Coded caching via projective geometry: A new low subpacketizatio n scheme,
C. Hari Hara Suthan, M. Bhavana, and P . Krishnan, “Coded caching via projective geometry: A new low subpacketizatio n scheme,” in Proc. IEEE Int. Symp. Inform. Theory (ISIT) , 2019, pp. 682-686
work page 2019
-
[1]
Web caching us ing access statistics,
A. Meyerson, K. Munagala, and S. Plotkin, “Web caching us ing access statistics,” in Proc. ACM-SIAM SODA , 2001, pp. 354-363
work page 2001
-
[2]
Fundamental limits of ca ching,
M. A. Maddah-Ali and U. Niesen, “Fundamental limits of ca ching,” IEEE Trans. Inf. Theory , vol. 60, no. 5, pp. 2856-2867, May 2014
work page 2014
-
[3]
On the optimali ty of uncoded cache placement,
K. Wan, D. Tuninetti, and P . Piantanida, “On the optimali ty of uncoded cache placement,” in Proc. IEEE Inf. Theory W orkshop (ITW), 2016, pp. 161-165
work page 2016
-
[4]
The exact r atememory tradeoff for caching with uncoded prefetching,
Q. Y u, M. A. Maddah-Ali, and A. S. Avestimehr, “The exact r atememory tradeoff for caching with uncoded prefetching,” in Proc. IEEE Int. Symp. Inf. Theory (ISIT) , Jun. 2017, pp. 1613-1617
work page 2017
-
[5]
Caching and delivery via interferen ce elimination,
C. Tian and J.Chen, “Caching and delivery via interferen ce elimination,” IEEE Trans. on Information Theory , vol. 64, no. 3, pp. 1548-1560, 2018
work page 2018
Show all 28 references
-
[6]
Coded caching with nonun iform demands,
U. Niesen and M. A. Maddah-Ali, “Coded caching with nonun iform demands,” IEEE Trans. Inf. Theory , vol. 63, no. 2, pp. 1146-1158, Feb. 2017
2017
-
[7]
Coded caching f or files with distinct file sizes,
J. Zhang, X. Lin, C.-C. Wang, and X. Wang, “Coded caching f or files with distinct file sizes,” in Proc. IEEE Intl. Symp. Inf. Theory , Sep. 2015, pp. 1686-1690
2015
-
[8]
Coded caching with het erogenous cache sizes,
S. Wang, W. Li, X. Tian, and H. Liu, “Coded caching with het erogenous cache sizes,” 2015, [Online]. Available: https://arxiv.org/ abs/1504.01123
2015 arXiv
-
[9]
Order optima l coded delivery and caching: Multiple groupcast index codi ng
M. Ji, A. M. Tulino, J. Llorca, and G. Caire, “Order optima l coded delivery and caching: Multiple groupcast index codi ng.” 2014, [Online]. Available: http://arxiv.org/abs/1402.4 572
2014
-
[10]
Decentralized coded ca ching attains order-optimal memory-rate tradeoff,
M. A. Maddah-Ali and U. Niesen, “Decentralized coded ca ching attains order-optimal memory-rate tradeoff,” IEEE/ACM Trans. Netw., vol. 23, no. 4, pp. 1029-1040, Aug. 2014
2014
-
[11]
Hierarchical coded caching,
N. Karamchandani, U. Niesen, M. A. Maddah-Ali, and S. N. Diggavi, “Hierarchical coded caching,” in Proc. IEEE Int. Symp. Inf. Theory , Jun. 2014, pp. 2142-2146
2014
-
[12]
Finite-length analysis of caching-aided coded multicasting,
K. Shanmugam, M. Ji, A. M. Tulino, J. Llorca, and A. G. Dim akis, “Finite-length analysis of caching-aided coded multicasting,” IEEE Trans. Inf. Theory , vol. 62, no. 10, pp. 5524-5537, Oct 2016
2016
-
[13]
Coded Caching Sc hemes with Low Rate and Subpacketizations,
M. Cheng, Q. Yan, X. Tang, and J. Jiang, “Coded Caching Sc hemes with Low Rate and Subpacketizations,” 2017, availabl e online at https://arxiv.org/abs/1703.01548
2017 arXiv
-
[14]
Coded cac hing with linear subpacketization is possible using Ruzsa- Szemer´edi graphs,
K. Shanmugam, A. M. Tulino, and A. G. Dimakis, “Coded cac hing with linear subpacketization is possible using Ruzsa- Szemer´edi graphs,” in Proc. IEEE Int. Symp. Inform. Theory (ISIT) , 2017, pp. 1237-1241
2017
-
[15]
On the placement d elivery array design for centralized coded caching scheme,
Q. Yan, M. Cheng, X. Tang, and Q. Chen, “On the placement d elivery array design for centralized coded caching scheme, ” IEEE Trans. Inf. Theory , vol. 63, no. 9, pp. 5821-5833, Sep. 2017
2017
-
[16]
Constructions o f Coded Caching Schemes With Flexible Memory Size,
M. Cheng, J. Jiang, Q. Yan, and X. Tang, “Constructions o f Coded Caching Schemes With Flexible Memory Size,” IEEE Trans. Communications, vol. 67, no. 6, pp. 4166-4176, Jun. 2019
2019
-
[18]
Placement delive ry array design through strong edge coloring of bipartite graphs,
Q. Yan, X. Tang, Q. Chen, and M. Cheng, “Placement delive ry array design through strong edge coloring of bipartite graphs,” IEEE Communications Letters , vol. 22, no. 2, pp. 236-239, Feb 2018
2018
-
[21]
Coded Caching Schemes With Reduced Subpacketization From Linear Block Codes,
Li Tang and Aditya Ramamoorthy, “Coded Caching Schemes With Reduced Subpacketization From Linear Block Codes,” IEEE Trans. Inf. Theory , vol. 64, no. 4, pp. 3099-3120, Apr 2018
2018
-
[22]
Coded Cachin g based on Combinatorial Designs,
S. Agrawal, K. V . S. Sree, and P . Krishnan, “Coded Cachin g based on Combinatorial Designs,” in Proc. IEEE Int. Symp. Inform. Theory (ISIT) , 2019, pp. 1227-1231
2019
-
[23]
A Generalized Gr ouping Scheme in Coded Caching,
M. Cheng, J. Jiang, Q. Wang, and Y . Yao, “A Generalized Gr ouping Scheme in Coded Caching,” IEEE Trans. Communications, vol. 67, no. 5, pp. 3422-3430, May 2019
2019
-
[24]
D. R. Stinson, Combinatorial designs: constructions and analysis . Springer Science and Business Media, 2007
2007
-
[25]
C. J. Colbourn and J. H. Denitz, Handbook of Combinatorial Designs , 2nd ed. Chapman and Hall/CRC, 2006
2006
-
[26]
Pisanski and B
T. Pisanski and B. Servatius, Configurations from a graphical viewpoint , Birkhauser, Basel, 2013
2013
-
[27]
J. A. Bondy and U. S. R. Murty, Graph Theory with Applications . New Y ork: North Holland, 1979
1979
-
[28]
Hirschfeld, Projective Geometries Over Finite Fields
J. Hirschfeld, Projective Geometries Over Finite Fields. Oxford Mathemat ical Monographs. Oxford University Press New Y ork, 1998
1998
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.