Vector Addition Systems (VAS) Overview

• Vector Addition Systems (VAS) are discrete-state computational models defined by non-negative counters and transition vectors, with VASS extending them via control states.
• They serve as foundational tools for analyzing concurrent, distributed, and resource-constrained systems by addressing key decision problems like reachability and coverability.
• Algorithmic complexity and structural decompositions in VAS offer deep insights into decidability, non-elementary complexity, and practical verification challenges in computer science.

A Vector Addition System (VAS) is a foundational discrete-state computational model, also known in the literature as a Petri net (when generalized to Vector Addition Systems with States, or VASS). VAS and its extensions serve as core models for analyzing concurrent, distributed, and resource-constrained systems, and have become central objects in the verification and computational complexity landscape. This article details the formal structure, canonical decision problems, complexity theory, structural subclasses, and modern directions for research on VAS and VASS.

1. Formal Structure and Definitions

A d-dimensional Vector Addition System (VAS) is defined by a finite subset $T \subseteq \mathbb{Z}^d$ of update vectors (steps). Its configurations are elements of $\mathbb{N}^d$, interpreted as the valuation of d non-negative counters. A transition $t \in T$ is enabled at configuration $u \in \mathbb{N}^d$ exactly when $u + t \ge 0$ componentwise; the one-step relation is $u \rightarrow u+t$.

A Vector Addition System with States (VASS) generalizes VAS by adding a finite control-state set. Formally, a $d$-dimensional VASS is a tuple $M=(Q, T)$, where $Q$ is a finite set of control states and $T \subseteq Q \times \mathbb{Z}^d \times Q$ is a finite transition set. A configuration is $\mathbb{N}^d$0 with $\mathbb{N}^d$1; a transition $\mathbb{N}^d$2 moves from $\mathbb{N}^d$3 to $\mathbb{N}^d$4 if $\mathbb{N}^d$5 (Czerwiński et al., 2023).

2. Core Decision Problems

Several central decision problems are formulated for VAS and VASS:

• Reachability: Given $\mathbb{N}^d$6, $\mathbb{N}^d$7, does there exist a finite sequence of transitions leading from $\mathbb{N}^d$8 to $\mathbb{N}^d$9? (Czerwiński et al., 2023).
• Coverability: Does there exist $t \in T$0 such that $t \in T$1? This problem generalizes reachability to allow "covering" the target configuration (Pilipczuk et al., 24 Nov 2025).
• Boundedness: Is the set of reachable configurations from $t \in T$2 finite?
• Termination: Is there no infinite run from a given initial configuration? This is often analyzed via the construction of ranking functions (Zuleger, 2017).

These problems underpin the verification of safety, liveness, and resource bounds in concurrent and distributed systems.

3. Algorithmic Complexity and Parameterized Results

Classical Complexity

• Reachability in general VASS is decidable, but Ackermann-complete, i.e., its complexity matches functions in the Ackermann hierarchy (Czerwiński et al., 2023). Precise upper and lower bounds have now been matched; reachability in $t \in T$3-VASS takes time $t \in T$4 and is $t \in T$5-hard in dimension $t \in T$6 (Czerwiński et al., 2023).
• Coverability (in the classical model) is EXPSPACE-complete under unary encoding, and para-PSPACE-complete under binary encoding, with XNL-completeness emerging as the precise landscape under dimensional parameterization (Pilipczuk et al., 24 Nov 2025).
• For fixed dimension $t \in T$7, reachability and coverability admit significantly lower complexity. In particular, reachability in 2-VASS is PSPACE-complete (Blondin et al., 2014), 4-VAS reachability is in PSPACE, and 5-VAS reachability is ELEMENTARY (Chen et al., 27 Apr 2026).
• In integer-valued VASS (ZVASS), i.e., systems where counters range over $t \in T$8, both reachability and coverability collapse to NP-completeness, even when reset operations are permitted (Haase et al., 2014).

Parameterized Complexity

A refined analysis highlights tractable regimes and complexity boundaries:

Parameterization	Encoding	Complexity Class	Complete for
Dimension $t \in T$9	Unary	XNL	XNL
Dimension $u \in \mathbb{N}^d$0	Binary	para-PSPACE	para-PSPACE
Size of $u \in \mathbb{N}^d$1	Binary	para-PSPACE	para-PSPACE
Size of $u \in \mathbb{N}^d$2	Unary	Open	Open

Notably, the existence of fixed-parameter tractable algorithms for coverability or reachability parameterized by the size of $u \in \mathbb{N}^d$3 is a major open problem (Pilipczuk et al., 24 Nov 2025).

4. Structural Extensions, Over-Approximations, and Variants

Affine Continuous VASS

Affine continuous VASS extend the classical model by permitting counters over $u \in \mathbb{N}^d$4 and transitions of affine form $u \in \mathbb{N}^d$5, where $u \in \mathbb{N}^d$6 is an integer matrix, $u \in \mathbb{N}^d$7 an integer vector, and $u \in \mathbb{N}^d$8. This over-approximation can strictly increase tractability; for example, with appropriate matrix classes, reachability and coverability can be in NP or NEXP, but undecidability appears as soon as matrices are sufficiently expressive (e.g., contain negative entries or zero rows/columns) (Balasubramanian, 2024).

Monus Semantics

Monus VASS modify the decrement operation: decrements on zero-valued counters leave the counter at zero (i.e., saturate at 0). Monus reachability remains Ackermann-complete, but zero-reachability and coverability become notably more tractable (EXPSPACE- and NP-complete, respectively) (Baumann et al., 2023).

Integer VASS

When counter values are allowed over $u \in \mathbb{N}^d$9 rather than $u + t \ge 0$0, classic undecidability and non-elementary complexity phenomena vanish: reachability and coverability become NP-complete, and even the addition of resets does not increase complexity beyond NP. Inclusion rises to coNEXP-completeness in full generality (Haase et al., 2014).

Parameterization by Geometric Dimension

The geometric dimension $u + t \ge 0$1 of a VASS, i.e., the dimension of the vector space spanned by cycles, provides strictly sharper run-length and complexity bounds for coverability and unboundedness than raw system dimension $u + t \ge 0$2. For many subclasses, this yields exponential or even polynomial-sized witness bounds (Czerwiński et al., 17 Feb 2026).

5. Structural Theory, Decompositions, and Invariant Synthesis

The mathematical structure of VAS reachability is governed by deep decomposition theorems:

• The classical Mayr-Kosaraju-Lambert-Sacerdote-Tenney (KLMST) decomposition yields an ideal-theoretic breakdown of runs via witness graph sequences, enabling both the decidability proof and subsequent complexity analysis. This decomposition is tightly connected with well quasi-ordering principles (Dickson’s and Higman's lemmas) and supports cubic Ackermann upper bounds for the reachability decision algorithm (Leroux et al., 2015).
• The geometry of VAS reachability sets is articulated through the finite decomposition into almost hybridlinear sets, each with amenable geometric and combinatorial properties. This geometric perspective enables effective algorithms for semilinearity checking, demonstrates the existence of infinite linear sets in reachability set complements, and underpins recent advances in the classification of VAS reachability sets (Guttenberg et al., 2022).
• Forward inductive invariants definable in Presburger arithmetic provide completeness: non-reachability is equivalent to the existence of such an invariant separating the initial from the target configuration. This dual certificate regime gives a principled semi-algorithm for the reachability problem and makes visible the certificate structure underlying (non-)reachability (jerome, 2010).

6. Quantitative and Long-run Analysis

VAS and VASS models also naturally capture long-run and mean-payoff properties:

• For integer-valued VASS, the multi-dimensional long-run average objective is NP-complete; for natural-valued (Petri net) VASS, the problem becomes undecidable. In probabilistic settings, expectation-based versions become tractable (PTime-complete) under certain configurations (Chatterjee et al., 2020).
• Ranking functions, especially lexicographic or affine ones, yield not only termination certificates but refined polynomial (or linear) upper and lower bounds on maximal trace length as a function of initial configuration norm (Zuleger, 2017).

7. Illustrative Examples and Frontier Problems

• The complexity landscape is sharply sensitive to the allowed operations: e.g., in continuous VASS, resets make coverability undecidable, while in discrete VASS, this is not the case (Balasubramanian, 2024).
• In two-dimensional VASS, reachability and coverability are PSPACE-complete (Blondin et al., 2014). Box-reachability — reachability with the additional constraint that all configurations visited stay below the target — coincides with standard reachability above some effectively computable threshold (Almagor et al., 18 Aug 2025).
• Key open questions persist, notably regarding the fixed-parameter tractability of coverability when parameterized by the size of the system (unary encoding), the complexity of reachability in higher-dimensional or pushdown-extended VASS, and the practical computation of semilinear decompositions and inductive invariants.

---

In summary, Vector Addition Systems and their extensions represent a keystone in infinite-state verification, concurrent systems modeling, and computational complexity theory. Their rich structural, algorithmic, and geometric theory continues to inform the boundaries of decidability, complexity, and algorithmic analysis in theoretical computer science.