Harder-Narasimhan theory for linear codes (with an appendix on Riemann-Roch theory) (Q1730858)

\textit{G. Harder} and \textit{M. S. Narasimhan} [Math. Ann. 212, 215--248 (1975; Zbl 0324.14006)] proposed a method to study the cohomology of moduli spaces over finite fields. They introduce notions such as slopes, canonical filtration or semistability. An analogue for Euclidean and Hermitian lattices was developed later, see [\textit{D. Grayson}, Comment. Math. Helv. 59, 600--634 (1984; Zbl 0564.20027)]. The present paper first develops Harder-Narasimhan (HN) theory for combinatorial lattices and then applies it to linear error-correcting codes. \par Section 1 studies HN theory for modular lattices $L$ of finite length. $L$ equipped with a semimodular function deg: $L\rightarrow\mathbb{R}$ is called a HN lattice (Definition 1). Then the paper gives the concepts of slopes, canonical filtration and semistability. Example 11 shows that the theory is applicable to matroids. Theorem 17 shows that a Galois connection between two lattices $L$ and $M$ induces a degree function making both $L$ and $M$ HN lattices. \par Section 2 provides HN structures on the lattice $L_C$ of a $(n,k)$ linear code $C$. For so doing the paper uses three approaches: the relation between linear codes and Euclidean lattices, see [\textit{J. H. Conway} and \textit{N. J. A. Sloane}, Sphere packings, lattices and groups. With additional contributions by E. Bannai, R. E. Borcherds, J. Leech, S. P. Norton, A. M. Odlyzko, R. A. Parker, L. Queen and B. B. Venkov. 3rd ed. New York, NY: Springer (1999; Zbl 0915.52003)], the algebraic-geometric codes point of view, see [\textit{M. A. Tsfasman} and \textit{S. G. Vlǎduţ}, Algebraic-geometric codes. Transl. from the Russian. Dordrecht etc.: Kluwer Academic Publishers (1991; Zbl 0727.94007)] and a Galois connection between $L_C$ and the lattice of the subsets of $\{1,\dots,n\}$. Proposition 19 proves that the three HN structures are the same. \par Section 3 studies the relationship between the canonical filtration and the slopes of a code $C$ and those of its dual code (Corollary 36). Corollary 37 states that $C$ is semistable if and only if its dual is. Section 4 shows that if two codes $A$ and $B$ are semistables their tensor product $A\otimes B$ is also semistable (Theorem 39). \par Finally an Appendix develops a Riemann-Roch theory for linear codes and matroids.