---
title: Network Satisfaction Problem (NSP)
url: https://www.emergentmind.com/topics/network-satisfaction-problem-nsp
type: topic
---

# Network Satisfaction Problem (NSP)

Searching arXiv for recent papers on network satisfaction problems for relation algebras and the specific algebra \(56_{65}\).
The **Network Satisfaction Problem** (NSP) is a decision problem attached to a fixed finite relation algebra \(A\). In the relation-algebraic formulation, the input is an \(A\)-network \((V,f)\), and the question is whether the network is satisfiable, that is, whether there exist a representation \(B\) of \(A\) and an assignment \(s:V\to B\) such that every constraint \(f(x,y)\) is realized by the pair \((s(x),s(y))\). In this sense, NSP is the satisfiability problem for qualitative binary constraint networks whose composition table is encoded by a finite relation algebra. The problem arises in the study of qualitative reasoning calculi, including interval and spatial calculi, and it is closely connected to constraint satisfaction over infinite templates [2512.06878].

## 1. Relation algebras and atomic constraints

A relation algebra is a Boolean algebra equipped with additional operations intended to abstract binary relations under complement, converse, and relational composition. In the formulation used for NSP, a relation algebra is an algebra
\[
A=(A;\sqcup,\bar{\phantom{o}},\bot,\top,id,\breve{\phantom{o}},\circ)
\]
such that the Boolean reduct is a Boolean algebra, \(\circ\) is associative and distributes over \(\sqcup\), \(id\) is a two-sided identity for composition, \(\breve{\phantom{o}}\) is an involution that distributes over \(\sqcup\), converse reverses composition, and Tarski’s residuation axiom holds. For finite relation algebras, the central combinatorial data are the **atoms**, namely the minimal nonzero elements in the Boolean order, together with the atomic composition table [2512.06878].

For atoms \(x,y,z\in A_0\), the triple \((x,y,z)\) is called an **allowed triple** if
\[
z\le x\circ y,
\]
and otherwise it is a **forbidden triple**. Thus the allowed triples encode exactly which local configurations are compatible with composition. In symmetric relation algebras, every element is self-converse, so allowed triples are invariant under permutation of coordinates. This atomic viewpoint is fundamental throughout the NSP literature because finite networks are checked locally against the allowed-triple relation, while global satisfiability depends on the existence of an appropriate representation [2512.06878].

A representation of a relation algebra \(A\) is a relational structure \(B\) that interprets each algebra element \(a\in A\) as a binary relation \(a^B\subseteq B^2\) in such a way that Boolean operations, converse, identity, and relational composition are interpreted exactly as in the set-theoretic algebra of binary relations. A relation algebra is **representable** if it has at least one such representation. This distinction is already complexity-relevant: for non-representable finite relation algebras, no network is satisfiable in any representation, so \(\NSP(A)\) is trivial [2512.06878].

## 2. Networks, consistency, and satisfiability

An \(A\)-network is the relation-algebraic analogue of a binary constraint network. One formulation uses a finite set of variables \(V\) and a labeling function \(f:V^2\to A\); another allows \(f\) to be a partial function on a subset \(E\subseteq V^2\). In both formulations, \(f(x,y)\) is the binary constraint assigned to the ordered pair \((x,y)\). A network is **atomic** if every label is an atom of \(A\) [2512.06878] [2008.11943].

Given a representation \(B\) of \(A\), an \(A\)-network \((V,f)\) is **satisfiable in \(B\)** if there exists a map \(s:V\to B\) such that
\[
(s(x),s(y))\in f(x,y)^B
\]
for all relevant pairs \(x,y\). The network is **satisfiable** in the general NSP sense if it is satisfiable in some representation of \(A\). This distinction is crucial: NSP quantifies not only over assignments into a fixed structure, but also over the choice of representation itself [2512.06878] [2008.11943].

A network is **consistent** when its labels satisfy the local algebraic conditions imposed by identity and composition. In one standard form, for all \(x,y,z\in V\),
\[
f(x,y)\le f(x,z)\circ f(z,y)
\quad\text{and}\quad
f(x,x)\le id.
\]
For atomic closed networks, this is the path-consistency condition induced by the atomic composition table. Consistency is necessary for satisfiability, but in general it is not sufficient. A major theme in the literature is to identify classes of relation algebras for which local consistency already guarantees global realizability, or for which every satisfiable network is realized inside one canonical representation [2512.06878] [2507.09324].

A representation \(B\) is called **universal** if every satisfiable \(A\)-network is satisfiable in \(B\). It is **fully universal** if every consistent atomic \(A\)-network is satisfiable in \(B\). It is **square** if \(\top^B=B^2\). A **normal representation** is fully universal, square, and homogeneous. These notions organize much of the modern theory because they determine when NSP can be recast as an ordinary CSP over a fixed template [2008.11943] [2507.09324].

## 3. CSP formulations, atom structures, and polymorphisms

When a suitable universal representation is available, NSP becomes a standard constraint satisfaction problem. If \(B\) is a universal representation of \(A\), then \(\NSP(A)\) is the same problem as \(\CSP(B)\): the input atomic network is viewed as a finite relational structure, and satisfiability becomes the existence of a homomorphism into \(B\). If \(A\) has a normal representation \(\mathfrak B\), then \(\NSP(A)\) and \(\CSP(\mathfrak B)\) coincide up to the standard translation between networks and primitive positive sentences [2512.06878] [2008.11943].

For complexity analysis, a second finite template is often associated with \(A\): its **atom structure**. The domain is the set \(A_0\) of atoms, equipped with unary relations recording which atoms lie below a given algebra element, a binary relation for converse, and a ternary relation for allowed triples. In the flexible-atom setting, the atom structure is conservative, and polymorphisms of the normal representation induce conservative polymorphisms on the finite atom structure. This enables the import of finite-domain CSP methods into the analysis of infinite-domain NSP templates [2008.11943].

In that framework, the existence of certain polymorphisms becomes the central tractability criterion. For finite symmetric representable relation algebras with a flexible atom, the decisive condition is the existence of a conservative 6-ary operation on atoms that preserves allowed triples and satisfies the **Siggers identity**
\[
f(x,x,y,y,z,z)=f(y,z,x,z,x,y).
\]
If such an operation exists, \(\NSP(A)\) is in \(P\); otherwise it is NP-complete. This is a complete classification for symmetric relation algebras with a flexible atom, and it solves Hirsch’s “Really Big Complexity Problem” for that subclass [2008.11943].

The model-theoretic side is equally important. Normal representations are homogeneous, and for homogeneous templates primitive positive definability is controlled by polymorphism preservation. Canonical polymorphisms, obtained via Ramsey-theoretic canonisation, transfer algebraic information between the infinite representation and the finite atom structure. This suggests that NSP is best understood as a meeting point of relation algebra, Fraïssé theory, and the universal-algebraic approach to CSPs [2008.11943].

## 4. Complexity landscape

The broad classification problem asks for the computational complexity of \(\NSP(A)\) for finite relation algebras \(A\). The landscape is heterogeneous. There exist finite relation algebras with NSP in \(P\), others with NP-complete NSP, and even finite relation algebras with undecidable NSP. For algebras with normal representations, however, \(\NSP(A)\) is always in NP, because the problem becomes a CSP over a fixed finitely bounded homogeneous structure [1912.08482] [2008.11943].

Two general NP-hardness criteria are known for relation algebras with a normal representation \(B\). First, if \(B\) contains a non-trivial equivalence relation with finitely many equivalence classes, then \(\CSP(B)\), and hence \(\NSP(A)\), is NP-complete. Second, if \(\Aut(B)\) is primitive, \(|B|>2\), and \(A\) contains a symmetric atom \(a\) with a forbidden triple \((a,a,a)\), equivalently \(a\not\le a\circ a\), then \(\NSP(A)\) is NP-hard. These criteria explain hardness for several small relation algebras via structural properties of their normal representations rather than by direct encoding of a classical NP-complete problem [1912.08482].

For relation algebras with at most four atoms, a systematic case analysis has been completed. The 2025 classification shows that for every finite relation algebra \(A\) with at most four atoms, \(\NSP(A)\) is always either in \(P\) or NP-hard. The proof uses a combination of representability results, constructions of universal, fully universal, or normal representations, reductions to finite conservative CSPs on atom structures, bounded-size representation arguments, and explicit hardness reductions, including PCSP reductions in some symmetric cases [2507.09324].

The algebra \(56_{65}\) was the last unresolved four-atom case. The 2025 paper on circular chromatic numbers and signed graphs proves that \(56_{65}\) has a finitely bounded universal square representation and concludes that \(\NSP(56_{65})\in\mathrm{NP}\). This settles the final open case in the classification of the existence of universal square representations, as well as the complexity of the corresponding NSP, for relation algebras with at most four atoms [2512.06878].

## 5. The algebra \(56_{65}\) and signed-graph semantics

The four-atom relation algebra \(56_{65}\) is symmetric and has atom set
\[
A_0=\{id,N,0,1\}.
\]
Here \(id\) is identity, \(N\) plays the role of a non-edge relation, and \(0,1\) correspond to edge labels in a signed-graph representation. The forbidden atomic triples are
\[
\{(N,N,N),(1,1,1),(0,0,1),(0,1,0),(1,0,0)\}
\]
together with the triples enforcing the usual behavior of \(id\). Because the algebra is symmetric, forbidden and allowed triples are permutation-invariant. This algebra is designed so that its square representations correspond exactly to certain signed graphs with a strong extension property [2512.06878].

The key correspondence is as follows. Given a representation \(B\) of \(56_{65}\), one forms a signed graph \((G,\sigma)\) on vertex set \(B\), where \(\{x,y\}\) is an edge exactly when \((x,y)\in 0^B\cup 1^B\), and the edge receives label \(0\) or \(1\) according to whether it lies in \(0^B\) or \(1^B\). Conversely, from a consistent signed graph satisfying the required extension conditions, one recovers a representation by declaring non-edges to realize \(N\) and labeled edges to realize \(0\) or \(1\). In this way, satisfiable atomic \(56_{65}\)-networks are essentially finite signed graphs whose embeddability determines satisfiability [2512.06878].

The signed-graph side is governed by **anti-even-balancing**. A signed graph is anti-even-signable if there exists a labeling \(\sigma:E(G)\to\mathbb Z_2\) such that every triangle has label sum \(0\), while every induced cycle of length at least \(4\) has label sum \(1\). The crucial structural result is the existence of a universal anti-even-balancing labeling on the complement \(\overline C\) of the generic circular triangle-free graph \(C\), where
\[
V(C)=\{z\in\mathbb C\mid |z|=1,\ \arg(z)\in \mathbb Q\cdot\pi\}
\]
and \((z,w)\) is an edge iff
\[
|\arg(z/w)|>\tfrac{2}{3}\pi.
\]
The resulting signed graph \((\overline C,\sigma_C)\) is universal for finite anti-even-balanced signed triangle-free graphs and has the 3-extension property needed for the relation-algebraic construction [2512.06878].

From \((\overline C,\sigma_C)\), the authors construct a representation \(B\) of \(56_{65}\) that is square, finitely bounded, and universal. This yields the theorem:
\[
\text{The relation algebra }56_{65}\text{ has a finitely bounded universal square representation. Consequently, }\NSP(56_{65})\text{ is in NP.}
\]
The proof also ties NSP to circular chromatic number: graphs of circular chromatic number less than \(3\) are characterized via anti-even-signability, and this characterization is what makes the universal signed-graph construction possible [2512.06878].

## 6. Local consistency, decidability, and open directions

A distinct line of work studies when NSP can be solved by local consistency methods. For arbitrary finite relation algebras, the meta-problem is highly negative: it is undecidable whether \(\NSP(A)\) is in \(P\), undecidable whether it is solved by \(k\)-consistency for some \(k\), and undecidable whether it is solved by path-consistency. This is proved by reduction from the undecidable representability problem using direct products with a fixed relation algebra whose NSP is itself undecidable [2304.12871].

For finite symmetric integral relation algebras with a flexible atom, the situation is much more rigid. In that class, NSP can be solved by \(k\)-consistency for some \(k\) if and only if the algebra admits a Siggers behavior on atoms, and in that case the \((4,6)\)-consistency procedure suffices. Moreover, the corresponding meta-problem is decidable in polynomial time in the number of atoms. The proof uses a sufficient condition for bounded width in terms of symmetry, normal representations, all \(1\)-cycles, and Siggers behavior, together with a result of Alexandr Kazda on finite binary conservative structures [2304.12871].

Several open problems remain. For finite relation algebras with a normal representation, the decidability of whether NSP is solved by \(k\)-consistency remains open in general. For the signed-graph template associated with \(56_{65}\), the complexity of \(\CSP(\overline C,\sigma_C)\) is posed explicitly as an open question. The 2025 paper notes that the representation \(B\) for \(56_{65}\) pp-constructs \(K_3\), which strongly suggests NP-hardness for appropriate CSPs, but it does not prove NP-hardness or NP-completeness for \(\NSP(56_{65})\) itself. It also leaves open questions about homogenizability and Ramsey expansions of the universal signed graph [2304.12871] [2512.06878].

In the current state of the subject, NSP occupies a precise position between algebraic logic and CSP theory. Its inputs are finite labeled networks, its semantics are given by representations of finite relation algebras, and its complexity is controlled by a mixture of allowed triples, amalgamation properties, model-theoretic universality, and polymorphism identities. The development from normal representations and flexible atoms to signed graphs and circular chromatic number shows that the notion is not merely a reformulation of CSP, but a framework in which representation theory and complexity theory constrain one another in a highly structured way.

Source: https://www.emergentmind.com/topics/network-satisfaction-problem-nsp