Cryptographic research often balances on the fine line between theoretical security guarantees and practical algorithmic efficiency. A new development in the study of coding theory addresses a long-standing theoretical bottleneck in linear code analysis. Researchers N.S. Kolesnikov and S.A. Novoselov from the Immanuel Kant Baltic Federal University have released a paper detailing new algorithmic approaches that dramatically improve the theoretical complexity bounds for block code basis reduction. Their work, titled "Quasi-polynomial time algorithms for block basis reduction of linear codes," introduces modifications and relaxations of existing frameworks, potentially shifting how cryptographers evaluate the complexity of decoding and finding low-weight codewords in linear codes.
The core of the research centers on the Block Code Basis Reduction (BKZ) algorithm, an important tool initially proposed by researchers Ghentiyala and Stephens-Davidowitz. In lattice cryptography and coding theory, reduction algorithms serve to simplify bases and uncover hidden structural properties, such as short vectors or low-weight codewords. However, under the weak complexity bounds established in the original formulation, the traditional BKZ algorithm for linear codes requires an exponential number of calls to a minimum weight codeword (MWC) oracle.
This theoretical limitation created a curious paradox within the field. In almost all practical scenarios, the original BKZ algorithm performs significantly better than its worst-case theoretical bounds suggest. In fact, these theoretical bounds are frequently worse than a plain enumeration of codewords, making the gap between theory and practice wide and difficult to bridge. For researchers attempting to rigorously bound the security of code-based cryptographic schemes, this discrepancy posed a persistent challenge. Kolesnikov and Novoselov’s work directly targets this gap by developing block code basis reduction algorithms that achieve polynomial and quasi-polynomial complexity for a fixed block size, aligning theoretical guarantees much more closely with observed practical behavior.
To achieve these improved complexity bounds, the researchers introduce several key modifications and relaxations to the standard algorithmic framework. First, the authors examine a truncated version of the BKZ algorithm. They demonstrate that this truncated variant requires a heuristic quasi-polynomial number of calls to the MWC-oracle, while preserving the same high basis quality achieved by the full, unrestricted BKZ algorithm. By strategically cutting off unnecessary computational paths without sacrificing the structural integrity of the resulting basis, the algorithm avoids the exponential blow-up predicted by earlier weak complexity bounds.
Building upon this truncated approach, the researchers introduce a second major refinement by placing weak, targeted restrictions on the code basis itself. These careful constraints ensure that the algorithm can operate within a polynomial number of calls to the oracle. By structuring the input basis under these mild conditions, the computational overhead is substantially reined in, providing a reliable polynomial-time path for specific classes of block reductions.
The third major contribution of the paper is the introduction of the DeepBKZ algorithm, a framework adapted directly from advanced lattice reduction techniques. In the study of lattices, algorithmic variants such as PotBKZ have successfully bridged theoretical efficiency with practical power by relaxing certain strict requirements. Kolesnikov and Novoselov adapt this concept to the domain of linear codes. Under comparable relaxations, their newly introduced DeepBKZ algorithm also operates in a quasi-polynomial number of calls to the MWC-oracle, expanding the toolkit available for analyzing code bases.
To validate their theoretical findings and demonstrate their practical viability, the authors developed a concrete implementation of the proposed algorithms. Their software implementation is capable of processing block sizes ranging from two up to 24 and beyond. This practical realization allowed the researchers to move beyond abstract asymptotic analysis and generate concrete complexity estimates. Furthermore, the implementation enabled a detailed empirical analysis of the approximation factors achieved for the low-weight codeword problem when applied to random linear codes, offering clearer insight into how these theoretical improvements behave in computational environments.
The paper, registered under the designation Paper 2026/2685, is publicly available through the Cryptology ePrint Archive. Through this contribution, Kolesnikov and Novoselov provide the academic and cryptographic communities with refined theoretical tools and practical data that help reconcile the long-standing divergence between the worst-case complexity bounds of block code basis reduction and its robust empirical performance.

