Structured Decompositions Skill

Sheaves on tree decompositions with bidirectional navigation

Version: 1.1.0 Trit: 0 (Ergodic - coordinates decomposition)

bmorphism Contributions

"Compositional Algorithms on Compositional Data: Deciding Sheaves on Presheaves" — ACT 2023, Benjamin Merlin Bumpus et al.

"any computational problem which can be represented as a sheaf with respect to these topologies can be decided in linear time on classes of inputs which admit decompositions of bounded width" — arXiv:2302.05575

Key Insight: Structured decompositions define Grothendieck topologies on categories of data (adhesive categories). This leads to algorithms on objects of any C-set category - structures such as: symmetric graphs, directed graphs, hypergraphs, databases, simplicial complexes, port graphs.

Implementation: Concrete implementations in the AlgebraicJulia ecosystem.

Related to bmorphism's work on:

• plurigrid/act - cognitive category theory building blocks
• Towards Foundations of Categorical Cybernetics - cybernetic systems via parametrised optics

Core Concept

StrDecomp = Functor

d: ∫G → C

• ∫G = category of elements of shape graph
• C = target category (Graph, FinSet, etc.)

using StructuredDecompositions

# Create decomposition from graph
d = StrDecomp(graph)

# Access components
bags(d)           # Local substructures
adhesions(d)      # Overlaps (shared boundaries)
adhesionSpans(d)  # Span morphisms

The 𝐃 Functor

Lifts decision problems to decomposition space:

# Define problem as functor
k_coloring(G) = homomorphisms(G, K_k)

# Lift and solve
solution = 𝐃(k_coloring, decomp, CoDecomposition)
(answer, witness) = decide_sheaf_tree_shape(k_coloring, decomp)

Specter-Style Navigation for Decompositions

Bidirectional paths for navigating decomposition structures:

using SpecterACSet

# Navigate bags
select([decomp_bags, ALL, acset_parts(:V)], decomp)

# Navigate adhesions with bidirectional transform
transform([decomp_adhesions, ALL], 
          adh -> reindex_adhesion(adh, mapping), 
          decomp)

Decomposition Navigators

decomp_bags

decomp_adhesions

decomp_spans

adhesion_between(i,j)

FPT Complexity

Runtime: O(f(width) × n) where width = max adhesion size

The sheaf condition ensures local solutions glue to global:

# Sheaf condition: sections over overlaps must agree
function verify_sheaf_condition(decomp, local_solutions)
    for (i, j) in adhesion_pairs(decomp)
        adh = adhesion(decomp, i, j)
        s_i = restrict(local_solutions[i], adh)
        s_j = restrict(local_solutions[j], adh)
        s_i == s_j || return false
    end
    return true
end

Integration with lispsyntax-acset

Serialize decompositions to S-expressions for inspection:

# Decomposition → Sexp
sexp = sexp_of_strdecomp(decomp)

# Navigate sexp representation
bag_names = select([SEXP_CHILDREN, pred(is_bag), SEXP_HEAD, ATOM_VALUE], sexp)

# Roundtrip
decomp2 = strdecomp_of_sexp(GraphType, sexp)

Adhesion as Colored Boundary

With Gay.jl deterministic coloring:

using Gay

struct ColoredAdhesion
    left_bag::ACSet
    right_bag::ACSet
    adhesion::ACSet
    color::String  # Deterministic from seed + index
end

function color_decomposition(decomp, seed)
    [ColoredAdhesion(
        bags(decomp)[i],
        bags(decomp)[j],
        adhesion(decomp, i, j),
        Gay.color_at(seed, idx)
    ) for (idx, (i, j)) in enumerate(adhesion_pairs(decomp))]
end

GF(3) Triads

dmd-spectral (-1) ⊗ structured-decomp (0) ⊗ koopman-generator (+1) = 0 ✓
sheaf-cohomology (-1) ⊗ structured-decomp (0) ⊗ colimit-reconstruct (+1) = 0 ✓

Time-Varying Data (Brunton + Spivak Integration)

For DMD/Koopman analysis on decomposed data:

@present SchTimeVaryingDecomp(FreeSchema) begin
    Interval::Ob
    Snapshot::Ob
    State::Ob
    
    timestamp::Hom(Snapshot, Interval)
    observable::Hom(Snapshot, State)
    
    Time::AttrType
    Value::AttrType
end

# Colimit reconstructs dynamics
# DMD = colimit of snapshot diagram over intervals

Julia Scientific Package Integration

From

julia-scientific

Cross-Domain Decomposition Patterns

# Metabolic network decomposition
using StructuredDecompositions, COBREXA
model = load_model("ecoli.json")
decomp = tree_decomposition(reaction_graph(model))
local_fba = [fba(submodel) for submodel in bags(decomp)]

# Molecular graph decomposition for ML
using StructuredDecompositions, MolecularGraph, AtomicGraphNets
mol = smilestomol("c1ccccc1")  # benzene
mol_decomp = tree_decomposition(mol)
features = [featurize(bag) for bag in bags(mol_decomp)]

# Quantum tensor network
using StructuredDecompositions, ITensors
tn = tensor_network(circuit)
decomp = mps_decomposition(tn)

References

• Bumpus et al. "Structured Decompositions" arXiv:2207.06091
• algebraicjulia.github.io/StructuredDecompositions.jl
• Nathan Marz: Specter inline caching patterns

See Also

• julia-scientific

  - Full Julia package mapping (137 skills)

• acsets

  - Algebraic databases foundation

• specter-acset

  - Bidirectional navigation

Scientific Skill Interleaving

This skill connects to the K-Dense-AI/claude-scientific-skills ecosystem:

Tree Decomposition

• etetoolkit [○] via bicomodule

Bibliography References

• algorithms

  : 19 citations in bib.duckdb

Cat# Integration

This skill maps to Cat# = Comod(P) as a bicomodule in the equipment structure:

Trit: 0 (ERGODIC)
Home: Prof
Poly Op: ⊗
Kan Role: Adj
Color: #26D826

GF(3) Naturality

The skill participates in triads satisfying:

(-1) + (0) + (+1) ≡ 0 (mod 3)

This ensures compositional coherence in the Cat# equipment structure.