Multisorted Core: Theory and Applications

• Multisorted core is a generalized core notion defined over multiple coordinated domains or sorts, extending the classical one-sorted definition.
• It serves as a central concept in constraint satisfaction, multiplex network analysis, and minion theory, enabling refined algorithmic and structural insights.
• Applications reveal practical implications such as endomorphism rigidity, canonical representation via minion cores, and weighted integration in core-periphery partitions.

Searching arXiv for recent and foundational papers on "multisorted core" and closely related usages. A multisorted core is a generalization of the classical notion of a core to settings in which structure is distributed across several sorts, layers, or components rather than a single carrier set. In the cited literature, the expression appears in several technically distinct senses. In universal-algebraic CSP theory, it denotes an instance-induced multisorted relational structure that admits no non-surjective endomorphisms (Delic et al., 15 Aug 2025). In multiplex network analysis, it denotes a core-periphery organization obtained by aggregating layer-specific richness across multiple edge types (Battiston et al., 2017). Closely related work in clone and minion theory introduces the minion core as a canonical representative under minion homomorphisms, and applies it to multisorted Boolean clones determined by binary relations (Barto et al., 2024). A broader common theme is that “core” is no longer defined relative to a single undifferentiated domain, but relative to a structured collection of coordinated domains.

1. From ordinary cores to multisorted structures

In the standard finite-template CSP setting, one writes $\operatorname{CSP}(\mathbf A)$ for a finite relational structure $\mathbf A=(A,\Gamma)$. A finite relational structure $\mathbf A'$ is a core if every endomorphism of $\mathbf A'$ is surjective; equivalently, $\mathbf A'$ is a minimal retract of $\mathbf A$. Every finite structure has a unique core up to isomorphism, and $$\operatorname{CSP}(\mathbf A)\equiv \operatorname{CSP}(\mathbf A')$$ for the core $\mathbf A'$, so complexity may be analyzed on the core without changing the decision problem (Delic et al., 15 Aug 2025).

The multisorted setting replaces a single carrier by a family of carriers indexed by sorts. In the language of multisorted minions, a multisorted set is written $A=(A_s)_{s\in S}$, and a multisorted operation has the form

$$f:A_w\to A_s,\qquad w=s_1\cdots s_n,\quad A_w=A_{s_1}\times\cdots\times A_{s_n}.$$

Minors are formed by permuting, identifying, and deleting input variables in a sort-compatible way, and a minion is a set of such operations closed under taking minors (Lehtonen et al., 2020).

An analogous perspective appears in the model theory of multisorted modules. There, the underlying object is a family of abelian groups or $\mathbf A=(A,\Gamma)$0-modules indexed by sorts, with pp-definable sets living in finite products of sorts. The theory is presented as equivalent to representations of quivers and to additive functors on preadditive categories, so the multisorted viewpoint is not a marginal extension but a standard formalism for additive structures (Prest, 2018).

This suggests that a multisorted core is best understood not as a single definition shared across all subjects, but as a family of core notions adapted to structures whose coordinates, relations, or operations are intrinsically typed.

2. Multisorted cores in constraint satisfaction theory

The most explicit universal-algebraic use of the term appears in the study of CSP instances whose reduced form is a multisorted core. After binary reduction, an instance $\mathbf A=(A,\Gamma)$1 is viewed as a multisorted structure whose sorts are the variable domains $\mathbf A=(A,\Gamma)$2. The key assumption is that this associated multisorted structure is a core in the analogous sense: it admits no non-surjective endomorphisms (Delic et al., 15 Aug 2025).

The paper places this assumption inside the standard algebraic CSP framework. A polymorphism $\mathbf A=(A,\Gamma)$3 preserves every relation of $\mathbf A=(A,\Gamma)$4, and if $\mathbf A=(A,\Gamma)$5 is a core then, after adding singleton unary relations, one may assume the polymorphism algebra is idempotent: $\mathbf A=(A,\Gamma)$6 The discussion then proceeds through Maltsev and Taylor structure. A Maltsev operation satisfies

$\mathbf A=(A,\Gamma)$7

while a Taylor algebra has a term $\mathbf A=(A,\Gamma)$8 satisfying the Taylor identities

$\mathbf A=(A,\Gamma)$9

This is the algebraic setting in which tractability is analyzed (Delic et al., 15 Aug 2025).

The multisorted-core assumption is used to exclude values lying in proper absorbing subuniverses. If $\mathbf A'$0 is a proper absorbing subuniverse of a finite subalgebra $\mathbf A'$1, then there is no solution $\mathbf A'$2 with $\mathbf A'$3. The reason is that such a solution would induce an endomorphism collapsing $\mathbf A'$4 onto a non-surjective image, contradicting that the instance is a core. This absorption obstruction is one of the central rigidity consequences of the multisorted-core hypothesis (Delic et al., 15 Aug 2025).

The structural analysis then uses binary syntactically simple instances and subdirect products of Maltsev algebras. A syntactically simple binary instance has only binary constraints, exactly one constraint $\mathbf A'$5 for each ordered pair $\mathbf A'$6, equality on the diagonal, and symmetry $\mathbf A'$7. Rectangularity plays a central role: if $\mathbf A'$8 with $\mathbf A'$9 in a Maltsev variety, then from

$\mathbf A'$0

it follows that

$\mathbf A'$1

This is obtained by applying a Maltsev term coordinatewise (Delic et al., 15 Aug 2025).

The later stages of the proof eliminate yellow and red behavior in Bulatov’s colored graph of an idempotent algebra, leaving only affine/blue behavior. The main algorithmic device is $\mathbf A'$2-consistency. The quotient CSP that remains becomes a system of linear equations over a finite Abelian $\mathbf A'$3-group $\mathbf A'$4, with constraints of the form

$\mathbf A'$5

or full products. The main theorem states that for any finite relational template $\mathbf A'$6 and instance $\mathbf A'$7 of $\mathbf A'$8, if $\mathbf A'$9’s corresponding multisorted structure is a core, then $\mathbf A'$0 can be solved using a logspace Turing machine with access to an oracle for $\mathbf A'$1. By the theorem of Buntrock et al., $\mathbf A'$2, so the resulting corollary is

$\mathbf A'$3

The paper explicitly presents this as a refinement of the tractable side of the Bulatov–Zhuk dichotomy rather than a change in the dichotomy boundary itself (Delic et al., 15 Aug 2025).

3. Minion cores and canonical representatives

A closely related but formally distinct notion is the minion core. For function minions, if $\mathbf A'$4 is an $\mathbf A'$5-ary operation and $\mathbf A'$6, the corresponding minor is

$\mathbf A'$7

A minion homomorphism $\mathbf A'$8 is a family of maps $\mathbf A'$9 preserving minors. A minion core is then defined as a minion $\mathbf A$0 such that every minion homomorphism $\mathbf A$1 is a minion automorphism. A minion core of $\mathbf A$2 is a minion core $\mathbf A$3 with $\mathbf A$4, where $\mathbf A$5 means $\mathbf A$6 and $\mathbf A$7 under existence of minion homomorphisms (Barto et al., 2024).

The central existence theorem states that every minion $\mathbf A$8 with all $\mathbf A$9 finite has a minion core, unique up to minion isomorphisms. In this sense the minion core plays the same role as the core of a relational structure: it is a canonical representative of an equivalence class, obtained by eliminating nonessential self-collapses (Barto et al., 2024).

This notion is used to classify multisorted Boolean clones determined by binary relations. The paper shows that clones on finite sets determined by binary relations whose projections to both coordinates have at most two elements can be described, up to minion homomorphisms, as multisorted Boolean clones determined by binary relations. Every such clone is equivalent to exactly one of a finite list of explicit minion cores, including $$\operatorname{CSP}(\mathbf A)\equiv \operatorname{CSP}(\mathbf A')$$0 and $$\operatorname{CSP}(\mathbf A)\equiv \operatorname{CSP}(\mathbf A')$$1, and these minions are all proved to be minion cores (Barto et al., 2024).

A complementary relational treatment is given by the theory of multisorted minions closed under reflections, extensions, and direct powers. There the central Galois correspondence is

$$\operatorname{CSP}(\mathbf A)\equiv \operatorname{CSP}(\mathbf A')$$2

$$\operatorname{CSP}(\mathbf A)\equiv \operatorname{CSP}(\mathbf A')$$3

For finite carriers, the closed sets of this connection are exactly minions and minor-closed sets of relation pairs. The paper does not define a multisorted core explicitly, but it gives a “core-like” theory of canonical reduction via reflections and finite powers. Its central closure theorem states that, for finite $$\operatorname{CSP}(\mathbf A)\equiv \operatorname{CSP}(\mathbf A')$$4, the condition $$\operatorname{CSP}(\mathbf A)\equiv \operatorname{CSP}(\mathbf A')$$5 is equivalent to the existence of a minion homomorphism $$\operatorname{CSP}(\mathbf A)\equiv \operatorname{CSP}(\mathbf A')$$6, and it introduces the terminology that $$\operatorname{CSP}(\mathbf A)\equiv \operatorname{CSP}(\mathbf A')$$7 mc-constructs $$\operatorname{CSP}(\mathbf A)\equiv \operatorname{CSP}(\mathbf A')$$8 if $$\operatorname{CSP}(\mathbf A)\equiv \operatorname{CSP}(\mathbf A')$$9 can be obtained from a finite power of $\mathbf A'$0 by reflection and minor-closed relaxation (Lehtonen et al., 2020).

Taken together, these results place multisorted cores within a broader program of identifying canonical representatives under minor-preserving morphisms.

4. Multisorted cores in multiplex network analysis

In network science, the phrase “multisorted core” is used in a different sense: a core defined across multiple layers of connectivity rather than within a single graph. A multiplex network is represented as

$\mathbf A'$1

where layer $\mathbf A'$2 encodes one type of interaction. For each node $\mathbf A'$3, the layer-specific degree $\mathbf A'$4 is its richness at layer $\mathbf A'$5, and this is decomposed into links toward lower-richness and higher-richness nodes: $\mathbf A'$6 The multiplex richness is

$\mathbf A'$7

and the corresponding “upward” richness is

$\mathbf A'$8

In the simplest case $\mathbf A'$9, though the coefficients may also compensate for differences in layer density or encode exogenous importance (Battiston et al., 2017).

The extraction rule is a scoring-and-rank thresholding procedure rather than a global combinatorial optimization. Nodes are ranked by decreasing $A=(A_s)_{s\in S}$0, then $A=(A_s)_{s\in S}$1 is plotted as a function of rank, and the core boundary is chosen at the rank where $A=(A_s)_{s\in S}$2 attains its maximum. All nodes to the left of that maximum form the multiplex core, and the remaining nodes form the periphery. The paper stresses that the multiplex core is not simply the union or intersection of layer cores, but a weighted integration of them (Battiston et al., 2017).

Overlap between layer cores is quantified by the core-similarity index

$A=(A_s)_{s\in S}$3

Here $A=(A_s)_{s\in S}$4 means no overlap and $A=(A_s)_{s\in S}$5 means identical cores across layers. Synthetic experiments on multiplex stochastic block models with $A=(A_s)_{s\in S}$6, $A=(A_s)_{s\in S}$7, $A=(A_s)_{s\in S}$8, $A=(A_s)_{s\in S}$9, $$f:A_w\to A_s,\qquad w=s_1\cdots s_n,\quad A_w=A_{s_1}\times\cdots\times A_{s_n}.$$0, $$f:A_w\to A_s,\qquad w=s_1\cdots s_n,\quad A_w=A_{s_1}\times\cdots\times A_{s_n}.$$1, and $$f:A_w\to A_s,\qquad w=s_1\cdots s_n,\quad A_w=A_{s_1}\times\cdots\times A_{s_n}.$$2 show how the detected multiplex core changes when the planted layer cores are disjoint, partially overlapping, or identical (Battiston et al., 2017).

The principal empirical application is a two-layer multiplex brain network built from DTI and fMRI, averaged across 171 healthy individuals. At the representative threshold $$f:A_w\to A_s,\qquad w=s_1\cdots s_n,\quad A_w=A_{s_1}\times\cdots\times A_{s_n}.$$3, the structural and functional cores overlap only partially, with measured core similarity $$f:A_w\to A_s,\qquad w=s_1\cdots s_n,\quad A_w=A_{s_1}\times\cdots\times A_{s_n}.$$4. The resulting multiplex core emphasizes posterior medial regions, parietal cortex, cingulate cortex, insula, temporal regions, and sensori-motor areas such as the pre/postcentral gyri. The paper interprets this as evidence that core organization in the human connectome is shaped jointly by structural wiring and functional coupling rather than by anatomical connectivity alone (Battiston et al., 2017).

A frequent conceptual confusion is to identify this multisorted or multiplex core with a multicores-periphery partition. The latter is a different mesoscale model in which a network contains several dense cores and one general sparse periphery. Its optimization criterion is the $$f:A_w\to A_s,\qquad w=s_1\cdots s_n,\quad A_w=A_{s_1}\times\cdots\times A_{s_n}.$$5-score of the cores-periphery ratio

$$f:A_w\to A_s,\qquad w=s_1\cdots s_n,\quad A_w=A_{s_1}\times\cdots\times A_{s_n}.$$6

evaluated over partitions of an average-linkage dendrogram. That framework allows several dense cores plus one general periphery, whereas the multiplex-core framework integrates layer-specific node richness into a single core-periphery division (Yan et al., 2016).

5. Combinatorial generalizations: simultaneous core multipartitions

A different multisorted analogue occurs in partition theory, where a single partition is replaced by an $$f:A_w\to A_s,\qquad w=s_1\cdots s_n,\quad A_w=A_{s_1}\times\cdots\times A_{s_n}.$$7-multipartition together with sort-dependent residue data. For a multipartition datum $$f:A_w\to A_s,\qquad w=s_1\cdots s_n,\quad A_w=A_{s_1}\times\cdots\times A_{s_n}.$$8, with $$f:A_w\to A_s,\qquad w=s_1\cdots s_n,\quad A_w=A_{s_1}\times\cdots\times A_{s_n}.$$9 and $\mathbf A=(A,\Gamma)$00, the $\mathbf A=(A,\Gamma)$01-residue of a node $\mathbf A=(A,\Gamma)$02 is

$\mathbf A=(A,\Gamma)$03

and the $\mathbf A=(A,\Gamma)$04-content of a multipartition is the multiset of these residues. A multipartition is an $\mathbf A=(A,\Gamma)$05-core if no other multipartition has the same $\mathbf A=(A,\Gamma)$06-content. When $\mathbf A=(A,\Gamma)$07, this recovers the usual notion of an $\mathbf A=(A,\Gamma)$08-core partition (Fayers, 2017).

This construction generalizes the classical theory of simultaneous core partitions. For relatively prime positive integers $\mathbf A=(A,\Gamma)$09 and $\mathbf A=(A,\Gamma)$10, the number of ordinary $\mathbf A=(A,\Gamma)$11-cores is finite and equals

$\mathbf A=(A,\Gamma)$12

and these objects are naturally identified with $\mathbf A=(A,\Gamma)$13-Dyck paths (Armstrong et al., 2013). The multipartition theory extends this single-sorted picture by allowing several components with charge vector $\mathbf A=(A,\Gamma)$14 (Fayers, 2017).

For an arbitrary family $\mathbf A=(A,\Gamma)$15, the simultaneous core set is

$\mathbf A=(A,\Gamma)$16

and the paper defines

$\mathbf A=(A,\Gamma)$17

If some $\mathbf A=(A,\Gamma)$18, then $\mathbf A=(A,\Gamma)$19 is finite if and only if $\mathbf A=(A,\Gamma)$20. If all $\mathbf A=(A,\Gamma)$21, finiteness holds if and only if $\mathbf A=(A,\Gamma)$22 and $\mathbf A=(A,\Gamma)$23 satisfies Condition X, a maximal/minimal-index criterion defined in the paper (Fayers, 2017).

The bipartition case admits explicit enumeration in several regimes. For example, when $\mathbf A=(A,\Gamma)$24,

$\mathbf A=(A,\Gamma)$25

and when $\mathbf A=(A,\Gamma)$26, $\mathbf A=(A,\Gamma)$27, and $\mathbf A=(A,\Gamma)$28 are coprime,

$\mathbf A=(A,\Gamma)$29

These formulas show that the “multisorted” extension in partition theory is not only formal but also enumeratively tractable in important subcases (Fayers, 2017).

6. Scope, limitations, and conceptual boundaries

The literature does not provide a single universally shared definition of “multisorted core.” Instead, it offers a cluster of related notions adapted to different mathematical objects. In CSP theory, the emphasis is on endomorphism rigidity of an instance-dependent multisorted relational structure and on the algorithmic consequences of that rigidity (Delic et al., 15 Aug 2025). In minion theory, the emphasis is on canonical representatives under minor-preserving maps, with uniqueness up to isomorphism for finite-arity-finite minions (Barto et al., 2024). In multiplex networks, the emphasis is on aggregating layer-specific centrality into a single core-periphery partition by ranking and thresholding rather than by retract theory (Battiston et al., 2017).

The limitations are correspondingly domain-specific. The CSP theorem is explicitly conditional on the associated multisorted structure already being a core, and the paper notes that one cannot expect to recognize core-ness efficiently in general because the relevant multisorted structure depends on instance size (Delic et al., 15 Aug 2025). The minion-core existence theorem is stated for minions with all arities finite (Barto et al., 2024). The multisorted-minion framework based on reflections, direct powers, and invariant relation pairs supplies a theory of canonical reduction, but it does not itself define a core in the classical endomorphism sense (Lehtonen et al., 2020). The multiplex-core construction depends on a ranking-and-thresholding heuristic, on the choice of richness measure, on the layer weights $\mathbf A=(A,\Gamma)$30, and in the connectome application on thresholding weighted matrices over a range of average degrees (Battiston et al., 2017).

A plausible implication is that “multisorted core” functions less as a single theorematic object than as a recurring structural pattern. Across algebra, complexity theory, network science, and combinatorics, the term marks situations in which the ordinary one-sorted notion of a core is too coarse because the underlying system has irreducible type structure. The resulting theories differ in their technical content, but they converge on the same formal intuition: core behavior must be defined relative to several coordinated domains at once.