Catalytic Tree Evaluation From Matching Vectors

We give new algorithms for tree evaluation (S. Cook et al. TOCT 2012) in the catalytic-computing model (Buhrman et al. STOC 2014). Two existing approaches aim to solve tree evaluation in low space: on the one hand, J. Cook and Mertz (STOC 2024) give an algorithm for TreeEval running in super-logarithmic space $O(\log n\log\log n)$ and super-polynomial time $n^{O(\log\log n)}$. On the other hand, a simple reduction from TreeEval to circuit evaluation, combined with the result of Buhrman et al. (STOC 2014), gives a catalytic algorithm for TreeEval running in logarithmic $O(\log n)$ free space and polynomial time, but with polynomial catalytic space. We show that the latter result can be improved. We give a catalytic algorithm for TreeEval with logarithmic $O(\log n)$ free space, polynomial runtime, and subpolynomial $2^{\log^εn}$ catalytic space (for any $ε> 0$). Our result opens a new line of attack on putting TreeEval in logspace, and immediately implies an improved simulation of time by catalytic space, by the reduction of Williams (STOC 2025). Our catalytic TreeEval algorithm is inspired by a connection to matching-vector families and private information retrieval, and improved constructions of (uniform) matching-vector families would imply improvements to our algorithm.

💡 Research Summary

The paper introduces a new catalytic‑space algorithm for the Tree Evaluation problem (TreeEval), a central problem in complexity theory that asks to compute the value at the root of a binary tree whose leaves are labeled by ℓ‑bit strings and whose internal nodes are labeled by binary functions f : {0,1}^ℓ × {0,1}^ℓ → {0,1}^ℓ. Prior work offered two contrasting approaches: Cook‑Mertz (STOC 2024) achieved O(log n·log log n) free (non‑catalytic) space but required super‑polynomial time n^{O(log log n)}; a reduction to circuit evaluation combined with Buhrman et al.’s catalytic model (STOC 2014) gave O(log n) free space and polynomial time but needed polynomial‑size catalytic tape. The authors improve the latter result dramatically: for any constant ε > 0 they obtain an algorithm that runs in O(log n) free space, polynomial time, and only 2^{O(log^ε n)} catalytic space (sub‑polynomial).

The technical core is a novel connection between catalytic computation and information‑theoretic private‑information‑retrieval (PIR) built on matching‑vector families. A matching‑vector family is a collection of vectors whose inner products belong to a restricted set; such families underpin the best known locally‑decodable codes and the most communication‑efficient PIR schemes with a constant number of servers. The authors observe that the Cook‑Mertz algorithm can be viewed as a white‑box use of an s‑server PIR protocol: each internal node’s function f_u is evaluated by retrieving the appropriate entry from its truth table via PIR, while the index (the pair of child values) is masked by the content τ of the catalytic tape. By requiring a PIR scheme whose queries have the form τ + encode_j(i) and that supports concatenation of queries for the two children, the authors can interleave PIR Answer and Reconstruct steps with the recursive TreeEval calls, thereby never exposing the masked values.

Using matching‑vector‑based PIR (instead of the Reed‑Muller‑based PIR used by Cook‑Mertz) yields a constant number of servers (2^t for any constant t) and communication complexity CC = exp(O((log N_DB)^{1/t})), where N_DB is the size of the database (the truth table of f_u). Consequently, the catalytic space needed equals CC, which becomes 2^{O(log^ε n)} for an appropriate choice of t. The free space consists of O(h·log s) bits for the recursion stack (h is the tree height) plus O(log CC + ℓ) bits for streaming through the database, which simplifies to O(log n) overall. The runtime is polynomial in the input size because each level performs a constant number of PIR calls and the tree depth is O(log n).

The paper formalizes these ideas in three main theorems. Theorem 1.1 states the basic result (O(log n) free space, 2^{O(log^ε n)} catalytic space, polynomial time). Theorem 1.2 shows a smooth trade‑off: by allowing slightly more free space (O(log n·log log n·log log log n)) one can reduce catalytic space to exp(exp(O(log log n))) while improving the runtime to n^{O(√log log n)}. Theorem 1.4 (informal) connects the quality of matching‑vector families to the algorithm: if uniformly good families exist beyond current lower bounds, TreeEval could be solved in O(log n·log log n) free space and polynomial time, essentially matching Cook‑Mertz’s space bound while retaining polynomial time.

An immediate application is a new time‑space‑catalytic trade‑off via Williams’s reduction (STOC 2025). Corollary 1.3 shows that any Turing machine running in time T can be simulated with O(√T) free space, 2^{O(T^ε)} catalytic space, and 2^{O(√T)} time. Compared with Williams’s original result (which required O(√T·log T) free space and 2^{O(√T·log T)} time), the new algorithm reduces free space by a logarithmic factor and eliminates the extra log T in the runtime, at the cost of a sub‑polynomial catalytic tape. The authors also discuss analogous improvements for circuit evaluation.

Beyond the immediate results, the paper opens a new research direction: improving matching‑vector constructions (especially uniform families) would directly tighten the bounds on TreeEval and potentially lead to a full log‑space algorithm. The authors highlight the broader significance of linking catalytic computation with cryptographic primitives, suggesting that further cross‑pollination could yield advances in both fields.

In summary, the work presents a sophisticated catalytic algorithm for TreeEval that leverages matching‑vector‑based PIR to achieve logarithmic free space, sub‑polynomial catalytic space, and polynomial time, and demonstrates how this breakthrough yields stronger time‑space‑catalytic trade‑offs and points to promising future research avenues.