Code for Li, Z., van Gennip, Y. & John, V. "An MBO method for modularity optimisation based on total variation and signless total variation." European Journal of Applied Mathematics, 37(3), 743–825, 2026.
MMBO is a Merriman–Bence–Osher-type method for graph clustering (community detection) by
modularity optimisation. Leading eigenpairs are approximated with a Nyström extension and
QR decomposition (Nystrom_extension_QR.py), so the method scales to graphs with 70,000
nodes (MNIST).
| Dataset | Newman–Girvan null model | Erdős–Rényi null model |
|---|---|---|
| Zachary's karate club | Example1_ZKC_NG.py |
EX1_ZKC_ER_test.py |
| MNIST (70k nodes, 10-NN graph) | Example2_MNIST_NG.py |
EX2_MNIST_ER_test.py |
| Stochastic block model | Example3_SBM_NG.py |
EX3_SBM_ER_test.py |
MMBO_HU_Boyd.py contains MMBO and the two earlier MBO schemes used for comparison
(Hu et al., Boyd et al.); MNIST_ER_lou_CNM_sc.py runs the Louvain, CNM and spectral
clustering baselines. Results are evaluated by modularity, ARI, purity, NMI and runtime.
Install the dependencies listed in Required_packages.txt. MNIST is downloaded by
GraphLearning.
Implementation and experiments: Zijun Li. The research idea is due to Yves van Gennip; the theory was developed jointly by Zijun Li and Yves van Gennip; Volker John (FU Berlin) co-supervised the work.