Distance-Optimal Sum-Rank Codes

• Distance-optimal sum-rank codes are error-correcting codes that achieve the Singleton bound under the sum-rank metric by unifying Hamming and rank metrics.
• Explicit constructions, including linearized Reed–Solomon, cyclic, and block-lifted methods, demonstrate their practical efficiency in network coding, distributed storage, and space-time applications.
• Advanced bounds and decoding algorithms, such as strong Singleton-like and sphere-packing bounds, validate their optimality and stimulate ongoing research in robust error correction.

Distance-optimal sum-rank codes are a class of codes that maximize the minimum sum-rank distance subject to constraints on the code size, dimension, field, and block structure. The sum-rank metric, which generalizes both Hamming and rank metrics, finds widespread use in network coding, distributed storage, space-time coding, and universal error correction, driving fundamental and explicit research on distance optimality, Singleton-type bounds, code constructions, and strong nonexistence results.

1. Sum-Rank Metric, Singleton-like Bound, and Distance Optimality

The sum-rank metric is defined as follows. Let $\mathbb{F}_q$ be a base field, $m\ge1$ the extension degree, and length partition $n = n_1 + \cdots + n_\ell$. For $$c = (c^{(1)},\dots,c^{(\ell)}) \in (\mathbb{F}_{q^m})^{n_1} \times \cdots \times (\mathbb{F}_{q^m})^{n_\ell}$$, choose an $\mathbb{F}_q$-basis $A$ of $\mathbb{F}_{q^m}$ and set $M_A(c^{(i)}) \in \mathbb{F}_q^{m \times n_i}$ by expanding coordinates. The sum-rank weight is $$\mathrm{wt}_{sr}(c) = \sum_{i=1}^{\ell} \mathrm{rank}_q M_A(c^{(i)})$$ and the associated metric (distance) is $d_{sr}(c, d) = \mathrm{wt}_{sr}(c - d)$ for $c, d \in \mathbb{F}_{q^m}^n$. For a linear code $C \subseteq \mathbb{F}_{q^m}^n$ with dimension $k$, redundancy $r = n - k$, the classical sum-rank Singleton bound states $d_{sr}(C) \le n - k + 1 = r + 1$ and a code attaining $d_{sr}(C) = n-k+1$ is called MSRD (maximum sum-rank distance) (Martínez-Peñas, 2017, Martínez-Peñas, 2019).

2. Explicit Constructions of Distance-Optimal and MSRD Sum-Rank Codes

Explicit constructions feature prominently in several families:

• Linearized Reed-Solomon codes (LRS): LRS codes developed over division rings achieve the Singleton bound for arbitrary block partitions and general fields. Their construction involves operator-polynomials and an explicit block decomposition according to conjugacy classes, providing an explicit generator matrix for every set of blocks. They unify Hamming-metric (RS) and rank-metric (Gabidulin) codes and always produce MSRD codes for any field and block sizes (Martínez-Peñas, 2017).
• Sum-rank Hamming codes: These are the maximal-length one-error-correcting (distance $d_{sr} \geq 3$) codes. For $m=1$ (Hamming-metric case with block structure), sum-rank Hamming codes correspond bijectively to maximal partial spreads and are perfect codes—balls of radius $1$ perfectly partition the space. Efficient syndrome decoders exist with $O(n)$ complexity, and duals (simplex codes) admit spread-based lower bounds on their minimum sum-rank distance (Martínez-Peñas, 2019).
• Block-lifted constructions: By lifting Hamming-metric codes over $\mathbb{F}_{q^u}$ via $q$-polynomial maps to the space $(\mathbb{F}_q^{(u,u)})^n$, one constructs codes with minimum sum-rank distance $d_{sr} \geq \min_{i}\{(i+1)d_i\}$ where $d_i$ are the constituent Hamming distances (Chen, 2024, Chen, 2022).

Recent work provides infinite families of distance-optimal cyclic and binary sum-rank codes with $d_{sr}=4$, the first such result for the sum-rank metric (Liu et al., 9 Jan 2026, Chen et al., 2024).

3. Bounds, Strong Singleton-like, and Nonexistence Results

Classical and recent bounds constrain existence and parameters of distance-optimal sum-rank codes:

• Classical Singleton-like bound: For square blocks $m \times m$, codes obey $|C| \leq q^{m^2(t - d_{sr} + 1)}$ for $t$ blocks, a code meeting equality is MSRD (Chen, 2023, Liu et al., 9 Jan 2026).
• Strong Singleton-like and covering-code bounds: By importing covering codes and list-decoding techniques from the Hamming metric, strong Singleton-like bounds have been established that are strictly tighter than the classical Singleton bound for large block lengths and small minimum distances. For example, a binary $(2 \times 2)^t$ code with $d_{sr}=4$ cannot exist for block length $t$ above certain thresholds depending on the code's dimension (Chen, 2023, Liu et al., 9 Jan 2026). These strong bounds have forced a dichotomy—explicit MSRD codes exist only for $t < q$ or $t < q^m$; for $t$ beyond, nonexistence is provable.
• Sphere-packing bound: For $d_{sr}=4$, precise sphere-packing volume calculations demonstrate that certain cyclic sum-rank codes with specific parameters are distance-optimal in the packing sense (Chen, 2024, Liu et al., 9 Jan 2026).

4. Cyclic, Negacyclic, and Constacyclic Distance-Optimal Codes

The connection between cyclicity and distance optimality in sum-rank codes is well-established:

• Cyclic and negacyclic sum-rank codes: These are defined by block-wise cyclic (for $\lambda=1$) or negacyclic (for $\lambda=-1$) shifts, and can be realized by lifting classical cyclic or negacyclic codes via $q$-polynomial methods (Chen et al., 2024). The minimal distance of such lifted codes enjoys lower bounds via constituent code distances (Theorem 3.1), with tightness if certain conditions on the constituent distances are satisfied.
• Bounds for cyclic constructions: BCH and Hartmann–Tzeng bounds adapt to the sum-rank setting, yielding minimum distance estimates for codes constructed by lifting BCH and cyclic codes (Chen et al., 2024, Liu et al., 9 Jan 2026).
• Explicit infinite families: There now exist infinite families of binary cyclic sum-rank codes with matrix size $2 \times 2$, block length $t=4^e-1$ for $e \geq 3$, dimension $4t-6$, and $d_{sr}=4$ that are distance-optimal in both Singleton and sphere-packing senses (Chen et al., 2024). The construction uses single-parity check and BCH codes over $\mathbb{F}_4$.

5. Convolutional and Systematic MSRD Sum-Rank Codes

Distance-optimality in convolutional and systematic settings is characterized via matrix properties:

• Maximum column sum-rank convolutional codes: These codes are constructed using superregular block Hankel matrices, and achieve $(n-k)(m+1)+1$ minimum sum-rank distance up to memory $m$ (Mahmood et al., 2015). Systematic MSRD block and convolutional encoders are characterized by superregular parity matrices that remain superregular after base-field-induced transformations. This algebraic perspective allows explicit MSRD code constructions over small fields and underlies many practical codes (Almeida et al., 2020).
• Design principles: For systematic block MSRD codes over $\mathbb{F}_{q^M}$, necessary and sufficient superregularity criteria for parity matrices are established; for convolutional codes, block Toeplitz matrices must remain superregular for all relevant field operations.

6. Quasi-Perfect Codes, Plotkin Sums, and Families with Small Defect

Quasi-perfect and almost-MSRD codes (small Singleton defect) are constructed using precise algebraic and combinatorial techniques:

• Quasi-perfect codes: Families of codes with minimum sum-rank distance $3$ or $4$, covering radius $2$, and block structure $(2,m)^t$ or $(2,2)^n$, are shown to exist via lifting perfect Hamming codes and additive codes (Chen, 2024, Liu et al., 9 Jan 2026). Binary quasi-perfect sum-rank codes can be constructed from known perfect codes and charge the covering radius via the block-wise decoder reduction.
• Almost-MSRD codes: Codes achieving the Singleton bound up to a small fixed defect (e.g., $2$ for block length up to $q^2$, or $4$ for length $q^4-1$) are obtained using cyclic and Reed–Solomon constructions over extension fields (Chen, 2024).
• Plotkin sum constructions: As in classical coding theory, Plotkin's $(u | u + v)$ construction carries over to the sum-rank metric, yielding distance-optimal codes of doubled length and controlled dimension (Liu et al., 9 Jan 2026, Chen, 2024).

7. Decoding, Algorithmic Aspects, and Applications

Distance-optimal sum-rank codes admit diverse fast decoding methods and have numerous applications:

• Syndrome decoding and reduction to Hamming decoding: Explicit syndrome-based algorithms with $O(n)$ complexity for perfect sum-rank Hamming codes (Martínez-Peñas, 2019), as well as fast reduction to Hamming-metric decoders for binary codes built from BCH, Goppa, and additive codes (Chen et al., 2023).
• Network coding and distributed storage: MSRD and distance-optimal sum-rank codes are optimal for multishot matrix-multiplicative channels, partial-MDS locally repairable storage codes, universal error correction in mixed error-erasure models, and space-time block coding for MIMO (Martínez-Peñas, 2019, Chen, 2022, Liu et al., 9 Jan 2026).
• Algebraic geometry codes: Linearized AG codes generalize classical AG codes and approach the Singleton bound asymptotically, sometimes outperforming the Gilbert–Varshamov bound for sufficiently large fields (Berardini et al., 2023).

---

Principal References:

Construction / Bound Type	Main arXiv Papers	Key parameters
Linearized Reed-Solomon, MSRD	(Martínez-Peñas, 2017, Martínez-Peñas et al., 2021, Chen, 2022)	Arbitrary blocks, $(n-k+1)$ distance
Hamming/Simplex, perfect/duals	(Martínez-Peñas, 2019)	Maximal length, $d_{sr}=3$
Cyclic, negacyclic, constacyclic	(Chen et al., 2024, Liu et al., 9 Jan 2026, Chen, 2024)	Infinite families with $d_{sr}=4$
Strong Singleton-like/nonexist	(Chen, 2023, Liu et al., 9 Jan 2026)	Nonexistence for large $t$
Quasi-perfect, Plotkin sum	(Chen, 2024, Liu et al., 9 Jan 2026)	$d_{sr}=3$/$4$, covering radius $2$
Convolutional/systematic MSRD	(Mahmood et al., 2015, Almeida et al., 2020)	Superregular matrix conditions

The current state-of-the-art encompasses fully explicit MSRD constructions for moderate block lengths, infinite families of distance-optimal codes with fixed distance, precise upper bounds on code dimension and existence, novel quasi-perfect and almost-MSRD families, and diverse decoding and application domains. Remaining open problems include bridging exponential gaps in block-length (existence/nonexistence), constructing large-parameter superregular matrices for systematic encoders, and extending fast decoding algorithms for general sum-rank metrics.