Diagrammatic Compatibility Overview

• Diagrammatic compatibility is a concept where compatibility conditions are defined directly on diagrams, such as graphs or string diagrams, rather than through external algebra.
• It provides structural certificates via legal triangulations, minimal cuts, and equivalence notions to ensure consistency in fields like unrooted phylogenetics and discrete planar mechanics.
• The framework facilitates schema change, syntactic rewrites, and higher-dimensional equivalence, proving essential in bridging combinatorial, categorical, and algebraic approaches.

Diagrammatic compatibility denotes families of constructions in which compatibility is expressed directly on a graph, string diagram, pasting diagram, or geometric network, rather than only in external algebraic notation. In unrooted phylogenetics it is the equivalence between compatibility of a tree profile and the existence of a legal triangulation of the display graph, a restricted triangulation of the edge-label intersection graph, or a complete set of pairwise parallel legal minimal cuts (Vakati et al., 2012). In discrete planar mechanics it is the passage from local star and loop configurations to algebraic conditions on lengths or elongations, with continuum limits reproducing classical strain compatibility (Treibergs et al., 2018). In categorical, logical, and higher-dimensional settings it appears as twisted satisfaction under schema change, normalization and equivalence of string diagrams, and equivalence-preserving operations in diagrammatic sets and Gray products (Diskin, 2023, Cailler et al., 11 Feb 2026, Chanavat, 2 May 2025). This suggests a recurring pattern: the diagram is itself the locus where compatibility is defined, certified, and manipulated.

1. Phylogenetic graph characterizations

For a profile $\mathcal{P}=\{T_1,\dots,T_k\}$ of unrooted phylogenetic trees, compatibility means that there exists a supertree $S$ on $X=\bigcup_i L(T_i)$ that displays every input tree. The display graph $G(\mathcal{P})$ is obtained by taking the union of the vertices and edges of all input trees. The edge-label intersection graph $L(\mathcal{P})$ has one vertex for each input-tree edge, two such vertices being adjacent precisely when the corresponding edges share an endpoint in $G(\mathcal{P})$; equivalently, $L(\mathcal{P})$ is the line graph of $G(\mathcal{P})$ (Vakati et al., 2012).

The first characterization, due to Vakati and Fernández-Baca, states that $\mathcal{P}$ is compatible if and only if $G(\mathcal{P})$ has a legal triangulation. Legality is controlled by two conditions: if a clique contains an internal edge of the display graph, then it contains no other original edge, and no fill-in edge is incident to a leaf. The second characterization, due to Gysel, Stevens, and Gusfield, states that $S$0 is compatible if and only if $S$1 admits a restricted triangulation, equivalently a chordal sandwich solution between $S$2 and its augmentation by all valid fill-in edges. In this formulation, fill-ins are permitted only between ELIG vertices not both arising from a single input tree.

A central result of the 2012 synthesis is that the ELIG formulation can be translated into the display graph by way of minimal separators and minimal cuts. A set of ELIG vertices is a legal minimal separator if, inside any one input tree, every subset it contains forms a clique; the corresponding edge set in the display graph is then a legal minimal cut, meaning that within each input tree all cut edges are incident to a common vertex and each component of the cut graph retains at least one edge. Legal minimal separators in $S$3 correspond exactly to legal minimal cuts in $S$4, and parallelism is preserved under this correspondence. This yields the equivalence

$S$5

where completeness means that every internal edge of every input tree is isolated by some cut in the family (Vakati et al., 2012).

These cut families connect directly to split compatibility. Every legal minimal cut $S$6 induces a split $S$7 of the taxon set, and parallel legal minimal cuts induce compatible splits. Hence, if $S$8 is a complete set of pairwise parallel legal minimal cuts, then $S$9 is a compatible split system and any tree displaying these splits is a compatible supertree for the original profile. Buneman’s Splits Equivalence Theorem then supplies the constructive bridge from diagrammatic certificate to tree.

The graph-theoretic viewpoint also clarifies scope and limitations. The results are stated for unrooted trees; rooted settings involve directed constraints and different encodings, and extension is nontrivial. If the display graph is disconnected, compatibility reduces to the connected components. The paper further notes that tree compatibility is NP-complete but fixed-parameter tractable in the number of trees, so legal triangulations, restricted triangulations, and complete parallel cut families function primarily as structural certificates rather than generic polynomial-time algorithms (Vakati et al., 2012).

2. Discrete planar structures and continuum limits

In discrete planar structures, diagrammatic compatibility refers to the fact that stars of triangles around interior nodes and closed loops along boundaries encode the admissibility of prescribed lengths or infinitesimal elongations. A truss is a finite graph with planar vertices $X=\bigcup_i L(T_i)$0 and edges $X=\bigcup_i L(T_i)$1 carrying positive lengths $X=\bigcup_i L(T_i)$2. For a triangulated structure, the star around an interior node is the union of the triangles meeting that node; for a triangular structure it is a regular hexagonal ring. These local diagrams support localized compatibility constraints, termed wagon wheel conditions (Treibergs et al., 2018).

For the nonlinear prescribed-length problem, the fundamental local invariant is the curvature atom

$X=\bigcup_i L(T_i)$3

where the angles are computed from edge lengths by the cosine law. The basic compatibility condition is $X=\bigcup_i L(T_i)$4 at every interior node. For a triangulated disk with strict triangle inequalities, realizability of a planar configuration is equivalent to vanishing of all these curvature atoms, with uniqueness up to rigid motion and reflection. In low valence this flatness condition can be converted into a polynomial equation in the squared edge lengths; for a 3-valent star, eliminating radicals yields a sixth-degree polynomial (Treibergs et al., 2018).

In the linearized problem, infinitesimal displacements $X=\bigcup_i L(T_i)$5 induce edge elongations $X=\bigcup_i L(T_i)$6 through

$X=\bigcup_i L(T_i)$7

For a regular hexagonal star in the triangular lattice, the wagon wheel condition becomes

$X=\bigcup_i L(T_i)$8

balancing rim and spoke elongations. In a general triangulation the same condition persists with geometric weights involving support distances $X=\bigcup_i L(T_i)$9 and angles $G(\mathcal{P})$0. The rigidity-matrix formulation writes the linearized system as $G(\mathcal{P})$1, with compatibility matrix $G(\mathcal{P})$2 satisfying $G(\mathcal{P})$3. In a triangulated disk, the number of compatibility conditions is

$G(\mathcal{P})$4

the number of interior vertices; for a multiply connected triangulated truss with $G(\mathcal{P})$5 holes,

$G(\mathcal{P})$6

The paper interprets this count as structural resilience, because it is also the number of bars that may be removed while preserving infinitesimal rigidity (Treibergs et al., 2018).

Local star compatibilities aggregate into boundary-only conditions. Summing wagon wheel equations over the stars inside a loop $G(\mathcal{P})$7 cancels interior contributions and leaves a double-layer functional supported on boundary edges and nearby interior edges. In the nonlinear setting, the resulting boundary angle identity is a discrete Gauss–Bonnet analogue. Around holes, one obtains ring-girder compatibility conditions, and the loss of compatibility under drilling depends on both removed interior stars and boundary geometry.

A recurrent misconception in passing from continuum to discrete settings is that rigidity always comes with local compatibility equations of continuum type. The paper states the contrary: there are rigid trusses without compatibility conditions in contrast to continuous materials. Conversely, as the mesh refines, the wagon wheel functional converges to the classical two-dimensional St Venant condition

$G(\mathcal{P})$8

so the discrete and continuum notions coincide asymptotically rather than identically (Treibergs et al., 2018).

3. Diagrammatic constraints, generalized sketches, and schema change

In the setting of data and system modelling, diagrammatic compatibility is the law governing how instances and constraints transform when a carrier schema graph changes. The ambient structure is a category with pullbacks, or more generally a finitely complete category in the institution-with-evidence formulation. Objects of the base category are interpreted as graphs, instances are arrows into a schema graph, and constraints are diagrammatic patterns or labeled arrows whose satisfaction is tested by pullback (Diskin, 2023).

A generalized sketch over a constraint signature is given by a carrier graph $G(\mathcal{P})$9 together with a category of declared constraints over $L(\mathcal{P})$0. Abstractly, a constraint is a triple $L(\mathcal{P})$1, where $L(\mathcal{P})$2 is a constraint name, $L(\mathcal{P})$3 is its arity graph, and $L(\mathcal{P})$4 is the class of valid instances. Concrete examples include path equations, regular constraints via injectivity, lifting constraints in Spivak’s sense, multiplicities such as $L(\mathcal{P})$5 and $L(\mathcal{P})$6, keys, and subsetting constraints. Instances over $L(\mathcal{P})$7 form the slice category $L(\mathcal{P})$8, so a preinstance is a graph morphism $L(\mathcal{P})$9.

The defining satisfaction test is pullback-based: for a constraint $G(\mathcal{P})$0 over $G(\mathcal{P})$1, one has

$G(\mathcal{P})$2

This makes satisfaction diagrammatic rather than merely syntactic, because conformance is determined by the existence and shape of specific pullback squares. The institutional issue is then how this relation behaves under schema morphisms $G(\mathcal{P})$3.

The compatibility law is twisted variance. Instances transform contravariantly by pullback along $G(\mathcal{P})$4, whereas constraints transform covariantly by postcomposition of their binding maps. In the basic binary case this yields the DCL satisfaction axiom

$G(\mathcal{P})$5

with $G(\mathcal{P})$6. In the enriched formulation with evidence, the same law is required to preserve a witness $G(\mathcal{P})$7: $G(\mathcal{P})$8 with the same evidence element on both sides. The paper packages this as a categorical e-institution and proves that Cartesian DCL-logic is e-institutional and sound (Diskin, 2023).

This framework is designed to make interoperability mathematically explicit. Constraints move covariantly, instances contravariantly, and pullback universality guarantees that satisfaction commutes with the diagrammatic change of schema. The resulting perspective generalizes classic sketch theory by admitting rich semantics $G(\mathcal{P})$9, dependencies among constraints, and evidence-preserving satisfaction. It also narrows a common misunderstanding: diagrammatic constraints are not just informal schema annotations, but objects in a categorical semantics with a precise transformation law under refactoring and view change (Diskin, 2023).

4. Syntactic equivalence, nominal syntax, and digital-circuit rewriting

A different use of diagrammatic compatibility concerns whether two syntactic descriptions represent the same diagram. In the term-based verification framework for string diagrams, one works in the free strict symmetric monoidal category generated by a signature of primitives. Terms are built from identities, generators, sequential composition, tensor composition, and symmetries. Structural equivalence is quotiented by the usual coherence equations—associativity and units for composition and tensor, interchange, and symmetry axioms—and then oriented into rewrite systems $L(\mathcal{P})$0 and $L(\mathcal{P})$1. The paper proves termination and confluence of both systems in Isabelle/HOL, so equivalence is decided by normalizing to canonical forms and comparing the results syntactically (Cailler et al., 11 Feb 2026).

The scope of this normalization is explicitly structural. It decides equivalence modulo symmetric monoidal coherence and wire permutations; full gate-level circuit theories are not included. The paper notes, for example, that quantum identities such as controlled-gate fusion or conjugation by Hadamards lie beyond the structural PROP system and would require additional domain-specific rewrite layers. The compatibility claim is therefore precise: different terms are diagrammatically compatible when they denote the same string diagram up to deformation and structural coherence, not necessarily when they are semantically equal in a richer theory (Cailler et al., 11 Feb 2026).

A complementary line of work reconciles name-free combinator syntax with name-based graph descriptions. In Uniflow and Biflow syntax, diagrams are built from structural combinators, variables, and the link binder $L(\mathcal{P})$2, which connects the uniquely named endpoints $L(\mathcal{P})$3 and $L(\mathcal{P})$4 inside $L(\mathcal{P})$5 and then removes those names. The semantics is given by framed point DAGs modulo isomorphism and wire homeomorphism, and the paper establishes sound and complete equational theories for the unified language, including traced and Frobenius-style extensions. In this sense, compatibility means that structural and nominal descriptions preserve the same diagrammatic semantics and admit the same equational reasoning (Ghica et al., 2017).

Digital-circuit semantics provides a third, operational variant. Circuits form Cartesian traced monoidal categories with values, gates, fork, join, stub, and a delay morphism $L(\mathcal{P})$6. The associated graph-rewrite formalism uses trace-framed point graphs with explicit feedback markers, together with local rules for constant evaluation, fork copying, streaming, and stub annihilation. At the categorical level the key properties are Cartesianity, trace, and iteration axioms; operationally, one exploits global-trace and global-delay normal forms. For combinational circuits the rewrite system is sound, terminating, and strongly confluent to canonical values, while for circuits with feedback the paper proves productivity under a delay-guardedness criterion (Ghica et al., 2017).

Across these syntactic settings, the diagram is not treated as a post hoc visualization of an algebraic term. Rather, rewrite rules, binders, and canonical forms are chosen so that syntactic compatibility coincides with diagrammatic equivalence under the relevant structural laws.

5. Higher-dimensional diagrammatic sets and Gray-product compatibility

In higher-dimensional rewriting and homotopy theory, diagrammatic sets are presheaves on categories of shapes built from atoms or regular directed complexes. They carry structural face and degeneracy operations but no primitive composition. Compatibility is therefore shifted onto a coinductively defined notion of equivalence and onto existence of weak composites for admissible pasting diagrams (Hadzihasanovic, 2019, Chanavat et al., 2024).

Representable diagrammatic sets were introduced as presheaves in which every spherical pasting diagram is connected to an individual cell by an equivalence cell. An $L(\mathcal{P})$7-cell is an equivalence if every division horn built from it admits a filler that is itself an $L(\mathcal{P})$8-equivalence. In a representable diagrammatic set, every spherical diagram has a weak composite unique up to higher equivalence, degenerate cells are equivalences, and equivalence coincides with weak invertibility. The 2-truncated case recovers bicategories, while groupoidal Kan cases satisfy a homotopy hypothesis through geometric realization and comparison with classical homotopy groups (Hadzihasanovic, 2019).

The internal theory of equivalence in diagrammatic sets was later refined via a coinductive definition of $L(\mathcal{P})$9 as a greatest fixed point over round diagrams. The paper proves that equivalences coincide with coinductive weak invertibility and bi-invertibility, that all degenerate cells are equivalences, that equivalences satisfy a 2-out-of-3 property, and that weakly invertible round contexts satisfy a division lemma. The division lemma asserts, in effect, that enwrapping a diagram by equivalences on all sides is invertible up to higher equivalence, which is a precise compatibility theorem for boundary modification and weak units (Chanavat et al., 2024).

Model-categorical developments place these ideas in a homotopical setting. Diagrammatic sets are presheaves on a category of shapes closed under Gray products, joins, suspensions, and duals, and they admit a cofibrantly generated model structure Quillen equivalent to simplicial sets. The Gray product on the presheaf category is defined by Day convolution and the model structure is monoidal with respect to it (Chanavat et al., 2024). For diagrammatic $G(\mathcal{P})$0-categories, the later monoidality theorem proves that

$G(\mathcal{P})$1

so tensoring a round diagram with an equivalence again yields an equivalence. The same work shows that tensoring with the walking equivalence produces functorial cylinder objects and that the opposite functor is a Quillen self-equivalence (Chanavat, 2 May 2025).

A common misconception is that compatibility here should mean strict compositional well-formedness. The opposite is closer to the actual picture: these theories are built precisely to replace strict composition by weak composites, weak units, and equivalence-stable boundary operations, while preserving coherent behavior under Gray products and homotopical passage to higher structure.

6. Compatibility, incompatibility, and categorical semantics

In generalized compositional theories, compatibility is expressed through classical structures and observable interaction. A classical structure is a special commutative dagger Frobenius algebra, diagrammatically represented by spider nodes. A process is compatible with such a structure when it is a comonoid homomorphism, hence preserves copying and deletion. Two observable structures are fully compatible when their classical points coincide; in finite-dimensional Hilbert spaces this recovers co-diagonalisability. Complementarity is stronger and is expressed by a Hopf law, while strong complementarity adds a bialgebra law. The chapter states that strong complementarity implies complementarity, but not conversely, and uses this distinction to derive GHZ/Mermin non-locality in the diagrammatic calculus (Coecke et al., 2015).

A different categorical strand treats incompatibility itself as first-class data. In the framework of diagrammatic negative information, a norphism is identified with a predicate on a hom-set,

$G(\mathcal{P})$2

so negative information becomes “what is not possible” rather than an exotic morphism type. Functor string diagrams then display how such predicates move under hom-functors and natural transformations. Triangle inequalities, preorders on hom-sets, and co-design problems in $G(\mathcal{P})$3 are expressed diagrammatically by monotone propagation of bans and feasibility predicates (Abbott et al., 2024).

Lattice-theoretic semantics develops another notion of compatibility from a binary relation $G(\mathcal{P})$4 on a set $G(\mathcal{P})$5. The closure operator

$G(\mathcal{P})$6

yields a complete lattice of fixpoints $G(\mathcal{P})$7, and the induced negation-like operator

$G(\mathcal{P})$8

acts as orthocomplementation, pseudocomplementation, or protocomplementation under suitable frame conditions. Adding an accessibility relation $G(\mathcal{P})$9 produces compatibility-and-accessibility frames in which a unary modality $\mathcal{P}$0 preserves $\mathcal{P}$1-fixpoints, thereby generalizing possibility semantics from classical modal logic to non-classical modal logics (Holliday, 2022).

These semantic strands share a structural feature. Compatibility is not merely coexistence; it is a rule for how copying, exclusion, negation, accessibility, or complementarity propagates through a diagrammatic environment.

7. Graphical nested complexes and compatible digraphs

In combinatorial geometry, compatibility fans for graphical nested complexes transpose the Fomin–Zelevinsky compatibility-degree construction from cluster theory to tubes in a graph. A tube is a nonempty connected induced subgraph, two tubes are compatible when they are nested or disjoint and non-adjacent, and a tubing is a set of pairwise compatible proper tubes. Relative to a maximal tubing $\mathcal{P}$2, the compatibility degree $\mathcal{P}$3 is $\mathcal{P}$4 on equality, $\mathcal{P}$5 on compatible distinct tubes, and otherwise counts vertices of $\mathcal{P}$6 adjacent to $\mathcal{P}$7. The associated compatibility vectors

$\mathcal{P}$8

support a complete simplicial fan realizing the nested complex of $\mathcal{P}$9. For paths this recovers Santos’ type-$G(\mathcal{P})$0 family of associahedral fans; for cycles it reproduces the type-$G(\mathcal{P})$1 compatibility-degree picture (Manneville et al., 2015).

The same term appears in universal algebra, but now for digraphs compatible with term operations. A digraph $G(\mathcal{P})$2 is compatible with an algebra $G(\mathcal{P})$3 when $G(\mathcal{P})$4 is preserved by every term operation. The paper studies weak, strong, radical, and extreme connectivity on such reflexive digraphs and proves diagrammatic characterizations of Maltsev-type conditions: a variety is Hobby–McKenzie iff strong and extreme components coincide in every compatible reflexive digraph, and a variety is $G(\mathcal{P})$5-permutable for some $G(\mathcal{P})$6 iff weak components are extremely connected (Gyenizse et al., 17 Mar 2026).

These graph-based theories show that compatibility can function as a combinatorial invariant with strong algebraic consequences. In the fan setting it controls simplicial realizations and flip relations; in compatible digraphs it detects equational properties of varieties through the collapse of connectivity notions. In both cases the decisive information is encoded diagrammatically at the level of adjacency and reachability rather than by external algebraic certificates alone.