Algebraic geometry codes — constructed from rational points on curves over finite fields — have rich structure exploitable for both construction and decoding. “Linearized” algebraic geometry codes generalize these to the sum-rank metric via division algebras over function fields, extending the framework to network coding and distributed storage. But can they be decoded efficiently?
Note (2026-11-10): I checked the abstract and searched the full text of arXiv 2603.11826 (Berardini, Caruso and Drain). The polynomial-time decoding algorithm for linearized algebraic geometry codes with unramified evaluation places, the Serre duality and Riemann-Roch theorem for the division algebras, and the coincidence of the dual codes (linearized Differential codes) with linearized AG codes over the adjoint algebra are the paper's. The term self-dual does not appear in the paper. Its decoder works by first determining a space that localizes the error and then solving the syndrome equations restricted to that space, and it decodes errors up to a stated sum-rank weight bound under the hypothesis that a non-split non-evaluation rational place exists; the essay's account that self-duality is the structural reason the decoder runs in polynomial time, and its omission of the decoding radius and the hypothesis, are not what the paper states.
Yes: polynomial-time decoding exists for unramified evaluation places (arXiv:2603.11826). The key is Serre duality and Riemann-Roch theory, adapted to division algebras. The dual codes — linearized Differential codes — turn out to coincide with the linearized AG codes themselves over the adjoint algebra.
The structural insight: the codes are self-dual in a precise algebraic-geometric sense. The dual of a linearized AG code is another linearized AG code on the adjoint algebra, and this coincidence is not accidental but forced by Serre duality. The decoding algorithm works because the syndrome computation, which requires understanding the dual code, reduces to understanding the code itself. Self-duality is not just an aesthetic property; it's the structural reason the decoder achieves polynomial time. The algebra knows its own dual, and this knowledge is computationally exploitable.