| name | fixed-point-compositionality-low-rank-gluings |
| description | Mathematical framework for compositional dynamics in threshold-linear networks via low-rank gluing rules. Use when studying modular network assembly, fixed point decomposition, compositional limit cycles, or engineering networks with predictable attractor repertoires. |
| license | MIT |
Fixed Point Compositionality via Low-Rank Gluing Rules
Mathematical theory of compositional dynamics in inhibition-dominated threshold-linear networks (TLNs) through structured modular assembly.
Core Concept: Compositionality
Brains generate complex behaviors from stable structures with limited resources via compositionality - decomposing complex tasks into reusable primitives.
This work provides first rigorous mathematical characterization linking structural modularity to functional compositionality in nonlinear networks.
Key Innovation: Low-Rank Gluing Rules
Novel modular network assembly connecting component subnetworks via specific low-rank couplings:
Network Architecture
- Component subnetworks: arbitrary internal connectivity
- Inter-module coupling: low-rank connections (rank-1, rank-k)
- Inhibition-dominated dynamics: threshold-linear units
Main Theorems
Theorem 1: Fixed Point Compositionality
Global fixed points constrained to combinations of local fixed points of constituent modules.
For low-rank gluings:
FixedPoints(Global) ⊆ Combinations(FixedPoints(Module₁) × FixedPoints(Module₂) × ...)
Theorem 2: Rank-1 Gluing Characterization
Complete classification determining which combinations yield global fixed points:
- Explicit construction rules for compositional attractors
- Predictable assembly of global dynamics from local motifs
Theorem 3: gCTLN Extension
Fixed point decomposition rules extended from CTLNs to generalized CTLNs (gCTLNs):
- Structural rules more robust than initially posited
- Wider applicability to biological network architectures
Applications
1. Combinatorial Attractor Engineering
Construct networks with combinatorially large repertoire of predictable attractors:
- Understanding from simpler component motifs
- Systematic design of complex dynamics
2. Compositional Limit Cycles
Beyond fixed points: compositional limit cycles emerge from gluing rules:
- Periodic dynamics from module oscillations
- Predictable timing from structural assembly
3. Graph-Based Networks
Extension to graph structures:
- Network topology → fixed point constraints
- Module structure → functional composition
Mathematical Framework
Threshold-Linear Networks (TLNs)
Dynamics governed by:
dx_i/dt = -x_i + [∑_j W_ij x_j + b_i]_+
where:
[·]_+ = threshold-linear nonlinearity (ReLU-like)
W_ij = synaptic weights (inhibition-dominated)
b_i = external inputs
Low-Rank Coupling Structure
Inter-module connections:
W_inter = UV^T (rank-k coupling)
where:
U, V = low-rank factors
- Specific structure constrains global dynamics
Fixed Point Decomposition
For rank-1 gluing W_inter = uv^T:
x_global = combination of {x_local(Module₁), x_local(Module₂), ...}
with explicit membership rules.
Biological Relevance
Compositional Brain Dynamics
- Modular cortical circuits → compositional computation
- Stable structure + flexible combinations
- Limited resources → efficient reuse of primitives
Inhibition-Dominated Networks
- Realistic cortical dynamics
- Winner-take-all competition
- Fixed point stability through inhibition
Network Assembly Rules
- Development: modules assembled via specific coupling rules
- Learning: modify low-rank factors → new compositional capabilities
- Evolution: reusable motifs across behavioral repertoire
Implementation Guidance
When to Use This Framework
Trigger conditions:
- Modeling modular neural circuits
- Engineering predictable attractor dynamics
- Studying compositionality in biological/artificial networks
- Analyzing fixed point structure of TLNs
- Designing networks with combinatorial dynamics
Construction Workflow
- Identify component modules - subnetworks with known fixed points
- Design low-rank coupling - specify
U, V factors
- Apply rank-1 theorem - determine valid combinations
- Construct global network - assemble with gluing rules
- Validate attractor repertoire - check combinatorial predictions
Graph-Based Application
- Define graph topology - network structure
- Apply gCTLN rules - fixed point decomposition
- Extend to generalized networks - beyond CTLN constraints
- Validate robustness - structural rule preservation
Theoretical Significance
First rigorous proof that:
- Modularity → Compositionality in nonlinear networks
- Low-rank structure constrains global attractors
- Combinatorial dynamics emerge from simple motifs
- Engineering recipe for predictable complex networks
Bridges gap between:
- Structural modularity (observed in brain)
- Functional compositionality (behavioral flexibility)
- Mathematical characterization (predictable assembly)
Paper Reference
arXiv:2606.07336 (q-bio.NC)
- Author: Juliana Londono Alvarez
- 39 pages, 18 figures
- Submitted: 2026-06-05
Related Work
- Combinatorial Threshold-Linear Networks (CTLNs)
- Attractor dynamics in recurrent networks
- Network assembly theory
- Modular circuit design
Activation: compositional dynamics, threshold-linear network, TLN, low-rank gluing, fixed point decomposition, modular network, attractor engineering, combinatorial dynamics, inhibition-dominated, gCTLN, network assembly, compositional limit cycle