# Reversible and irreversible bracket-based dynamics for deep graph neural networks

Source: [Reversible and irreversible bracket-based dynamics for deep graph neural networks](<https://www.sota2.com/research/paper/68fa71861e209131cea51c30>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-04T04:33:19.795Z

Authors: Anthony Gruber, Kookjin Lee, Nathaniel Trask

Published: 2023-05-24T23:24:18.000Z

## Abstract

Recent works have shown that physics-inspired architectures allow the training of deep graph neural networks (GNNs) without oversmoothing. The role of these physics is unclear, however, with successful examples of both reversible (e.g., Hamiltonian) and irreversible (e.g., diffusion) phenomena producing comparable results despite diametrically opposed mechanisms, and further complications arising due to empirical departures from mathematical theory. This work presents a series of novel GNN architectures based upon structure-preserving bracket-based dynamical systems, which are provably guaranteed to either conserve energy or generate positive dissipation with increasing depth. It is shown that the theoretically principled framework employed here allows for inherently explainable constructions, which contextualize departures from theory in current architectures and better elucidate the roles of reversibility and irreversibility in network performance.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Node Classification on CORA 80/10/10 random split (test)](<https://www.sota2.com/research/sota/node-classification-on-cora-80-10-10-random-split-test.md>): Node Classification; CORA 80/10/10 random split (test); Accuracy: 83.

- [Node Classification on CiteSeer 80/10/10 random split (test)](<https://www.sota2.com/research/sota/node-classification-on-citeseer-80-10-10-random-split-test.md>): Node Classification; CiteSeer 80/10/10 random split (test); Accuracy: 0.742.

- [Node Classification on Computer 80/10/10 random split (test)](<https://www.sota2.com/research/sota/node-classification-on-computer-80-10-10-random-split-test.md>): Node Classification; Computer 80/10/10 random split (test); Accuracy: 84.8.

- [Trajectory Prediction on Damped Double Pendulum (test)](<https://www.sota2.com/research/sota/trajectory-prediction-on-damped-double-pendulum-test.md>): Trajectory Prediction; Damped Double Pendulum (test); MAE (Angle q): 0.00364.

- [Trajectory prediction on MuJoCo HalfCheetah (test)](<https://www.sota2.com/research/sota/trajectory-prediction-on-mujoco-halfcheetah-test.md>): Trajectory Prediction; MuJoCo HalfCheetah (test); Relative Error: 0.0566.

- [Trajectory prediction on MuJoCo Hopper (test)](<https://www.sota2.com/research/sota/trajectory-prediction-on-mujoco-hopper-test.md>): Trajectory Prediction; MuJoCo Hopper (test); Relative Error: 0.0279.

- [Trajectory prediction on MuJoCo Swimmer (test)](<https://www.sota2.com/research/sota/trajectory-prediction-on-mujoco-swimmer-test.md>): Trajectory Prediction; MuJoCo Swimmer (test); Relative Error: 0.0122.

- [Node Classification on PubMed (80/10/10 random split, test)](<https://www.sota2.com/research/sota/node-classification-on-pubmed-80-10-10-random-split-test.md>): Node Classification; PubMed 80/10/10 random split (test); Accuracy: 78.2.

- [Node Classification on Photo 80/10/10 random split (test)](<https://www.sota2.com/research/sota/node-classification-on-photo-80-10-10-random-split-test.md>): Node Classification; Photo 80/10/10 random split (test); Accuracy: 92.4.

- [Node Classification on Coauthor-CS (test)](<https://www.sota2.com/research/sota/node-classification-on-coauthor-cs-test.md>): Node Classification; Coauthor-CS (test); Accuracy: 92.5.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2305.15616>)

[PDF](<https://arxiv.org/pdf/2305.15616.pdf>)

[Implementation](<https://github.com/natrask/BracketGraphs>)
