Cosm: orchestrated collective dynamics give rise to rapid computational search at the solution frontier
Abstract
We introduce Collective Switched Motion (Cosm), a heuristic optimization framework based on orchestrated collective dynamics. Cosm organizes the objective into conflict-minimized subobjectives and activates them sequentially, temporally separating competing local influences. The interplay of switching and local interactions among variables, supplemented by structured correlated perturbations, gives rise to collective search behavior including coordinated cluster motion. Using this framework, we develop and test a solver for sparse Ising optimization. On G64, a large sparse instance with a highly heterogeneous degree distribution from the longstanding Gset Max-Cut benchmark suite, Cosm establishes a new best-known cut of 8753. On large random-graph instances G61 and G70, whose best-known cuts of 5799 and 9595 have been difficult for prior heuristics to attain, a vectorized CPU implementation of Cosm reliably reaches these cuts with times-to-target of 15 s and 2.4 s, respectively. Cosm also heuristically attains the certified optimum of G81, the largest Gset instance (a 20,000-variable two-dimensional spin glass), along with those of G72 and G77. Tests on planted-solution spin-glass benchmarks suggest more favorable empirical scaling than that previously reported for seven compared solvers. Cosm demonstrates that, from simple interacting elements and a new form of orchestration, effective search can emerge.
Related
- Fractional Laplace Neural Operators: Exact Architectures, an Expressivity Frontier at Criticality, and Certified Stability for Memory-Driven Network Dynamics
- Optimal Graphon Control of Collective Dynamics
- Learning PDE solution operators with variable initial conditions via Latent Dynamics Networks
- Computational fluid dynamics assessment of upper airway morphology and flow dynamics in infants with Robin sequence treated with the Stanford Orthodontic Airway Plate
- On the computational cost of Stochastic Gradient Langevin Dynamics
BibTeX
@misc{zick2026cosm,
title = {{Cosm: orchestrated collective dynamics give rise to rapid computational search at the solution frontier}},
author = {Kenneth M. Zick and Nikhil Shukla and Alexander Marakov},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2605.30355},
eprint = {2605.30355},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2605.30355},
}
RIS
TY - UNPB
TI - Cosm: orchestrated collective dynamics give rise to rapid computational search at the solution frontier
AU - Kenneth M. Zick
AU - Nikhil Shukla
AU - Alexander Marakov
PY - 2026
DA - 2026/10/11
PB - arXiv
DO - 10.48550/arxiv.2605.30355
UR - https://arxiv.org/abs/2605.30355
AB - We introduce Collective Switched Motion (Cosm), a heuristic optimization framework based on orchestrated collective dynamics. Cosm organizes the objective into conflict-minimized subobjectives and activates them sequentially, temporally separating competing local influences. The interplay of switching and local interactions among variables, supplemented by structured correlated perturbations, gives rise to collective search behavior including coordinated cluster motion. Using this framework, we develop and test a solver for sparse Ising optimization. On G64, a large sparse instance with a highly heterogeneous degree distribution from the longstanding Gset Max-Cut benchmark suite, Cosm establishes a new best-known cut of 8753. On large random-graph instances G61 and G70, whose best-known cuts of 5799 and 9595 have been difficult for prior heuristics to attain, a vectorized CPU implementation of Cosm reliably reaches these cuts with times-to-target of 15 s and 2.4 s, respectively. Cosm also heuristically attains the certified optimum of G81, the largest Gset instance (a 20,000-variable two-dimensional spin glass), along with those of G72 and G77. Tests on planted-solution spin-glass benchmarks suggest more favorable empirical scaling than that previously reported for seven compared solvers. Cosm demonstrates that, from simple interacting elements and a new form of orchestration, effective search can emerge.
ER -