New Lower Bounds for Matching Vector Codes

A Matching Vector (MV) family modulo $m$ is a pair of ordered lists $U=(u_1,…,u_t)$ and $V=(v_1,…,v_t)$ where $u_i,v_j \in \mathbb{Z}_m^n$ with the following inner product pattern: for any $i$, $< u_i,v_i>=0$, and for any $i \ne j$, $< u_i,v_j> \ne 0$. A MV family is called $q$-restricted if inner products $< u_i,v_j>$ take at most $q$ different values. Our interest in MV families stems from their recent application in the construction of sub-exponential locally decodable codes (LDCs). There, $q$-restricted MV families are used to construct LDCs with $q$ queries, and there is special interest in the regime where $q$ is constant. When $m$ is a prime it is known that such constructions yield codes with exponential block length. However, for composite $m$ the behaviour is dramatically different. A recent work by Efremenko [STOC 2009] (based on an approach initiated by Yekhanin [JACM 2008]) gives the first sub-exponential LDC with constant queries. It is based on a construction of a MV family of super-polynomial size by Grolmusz [Combinatorica 2000] modulo composite $m$. In this work, we prove two lower bounds on the block length of LDCs which are based on black box construction using MV families. When $q$ is constant (or sufficiently small), we prove that such LDCs must have a quadratic block length. When the modulus $m$ is constant (as it is in the construction of Efremenko) we prove a super-polynomial lower bound on the block-length of the LDCs, assuming a well-known conjecture in additive combinatorics, the polynomial Freiman-Ruzsa conjecture over $\mathbb{Z}_m$.

💡 Research Summary

The paper investigates fundamental limitations of locally decodable codes (LDCs) that are built using matching vector (MV) families over the ring ℤₘ. An MV family consists of two ordered lists U = (u₁,…,u_t) and V = (v₁,…,v_t) with vectors in ℤₘⁿ such that the inner product ⟨u_i, v_i⟩ equals zero for each i, while ⟨u_i, v_j⟩ is non‑zero whenever i ≠ j. When the set of possible inner‑product values is limited to at most q distinct elements, the family is called q‑restricted. Such families are the combinatorial core of the construction of q‑query LDCs: each coordinate of the message is associated with a pair (u_i, v_i), and the decoder queries q positions of the codeword, using the inner‑product pattern to recover the desired symbol.

Historically, when the modulus m is a prime, any q‑restricted MV family yields LDCs whose block length N grows exponentially in the message length n. For composite m, however, Grolmusz showed that MV families of super‑polynomial size exist, and Efremenko leveraged this to obtain the first sub‑exponential LDC with a constant number of queries (the “Efremenko construction”). This breakthrough raised the question of whether the sub‑exponential regime is inherent or merely an artifact of the specific construction.

The authors answer this by proving two general lower bounds that apply to any black‑box reduction from MV families to LDCs.

1. Quadratic lower bound for constant q.
   If the MV family is q‑restricted with q a fixed constant (or sufficiently small relative to n), then any LDC derived from it must have block length N = Ω(n²). The proof proceeds by examining the t × t inner‑product matrix M whose (i,j) entry is ⟨u_i, v_j⟩. The q‑restriction forces M to have at most q distinct non‑zero entries off the diagonal. By interpreting M as the adjacency matrix of a bipartite graph and applying a matching‑size argument, the authors show that a decoder that makes only q queries cannot distinguish enough pairs (i,j) unless the graph contains a matching of size Θ(n). This forces the dimension n to be at most O(√N), i.e., N must be at least quadratic in n. The argument is robust: it does not rely on any specific algebraic structure of the MV family, only on the combinatorial restriction on inner‑product values.

2. Super‑polynomial lower bound for constant modulus m (assuming the Polynomial Freiman‑Ruzsa conjecture).
   When m is a fixed constant, the authors invoke the Polynomial Freiman‑Ruzsa (PFR) conjecture over ℤₘ, a central open problem in additive combinatorics that asserts that a set with small doubling is efficiently contained in a low‑dimensional generalized arithmetic progression. Under this conjecture, the set of inner‑product values {⟨u_i, v_j⟩ : i ≠ j}—which is small because of the q‑restriction—must have a highly structured form. The authors show that such structure forces the size t of any MV family to be at most exp(O((log n)^{c})) for some constant c, i.e., sub‑exponential but still super‑polynomial in n. Consequently, any LDC built from such a family must have block length N ≥ t = n^{Ω(log n)}. This matches the known upper bounds only up to polynomial factors and demonstrates that, assuming PFR, the Efremenko construction is essentially optimal for constant m.

The technical contributions include:

• A clean reduction from the combinatorial q‑restriction to a rank‑lower‑bound on the inner‑product matrix, using elementary linear‑algebra and matching theory.
• A novel application of additive‑combinatorial structure theorems (specifically, the PFR conjecture) to the analysis of MV families, bridging coding theory and additive number theory.
• A “black‑box” perspective that abstracts away the details of any particular LDC construction, thereby establishing universal limits that apply to all known MV‑based schemes and any future ones that follow the same paradigm.

In the discussion, the authors compare their results with prior work. Efremenko’s sub‑exponential LDC relies on a super‑polynomial MV family constructed by Grolmusz; the new lower bounds show that, without violating the PFR conjecture, one cannot improve the exponent dramatically. Moreover, the quadratic bound for constant q explains why attempts to reduce the query number below a small constant have not yielded linear‑size codes.

The paper concludes with several open directions. One is to remove the reliance on the PFR conjecture and obtain unconditional super‑polynomial lower bounds for constant m. Another is to explore alternative combinatorial objects beyond MV families that might bypass the quadratic barrier for constant q. Finally, the authors suggest investigating whether relaxing the q‑restriction (e.g., allowing a slowly growing number of inner‑product values) could lead to new trade‑offs between query complexity and block length.

Overall, the work deepens our understanding of the intrinsic trade‑offs in MV‑based locally decodable codes, showing that the remarkable sub‑exponential constructions are close to optimal under widely believed conjectures, and that any substantial improvement will likely require fundamentally new ideas.