Cool Graphs: Active Property Search Towards Quantum Nano-Refrigerators
Organizations: Schulich Faculty of Chemistry Technion – Israel Institute of Technology Haifa, Israel · Department of Molecular Chemistry and Materials Science Weizmann Institute of Science Rehovot, Israel · Schulich Faculty of Chemistry and Helen Diller Quantum Center Technion – Israel Institute of Technology Haifa, Israel
Abstract
Nanodevices that absorb energy from their thermal environment can act as quantum refrigerators. Finding suitable configurations, however, is a needle-in-a-haystack problem: the many-body state space scales exponentially with the size of the device, where each candidate configuration requires an expensive quantum-chemical computation. We introduce a search framework that couples a Graph Variational Autoencoder (GVAE) with an uncertainty-driven Active Learning (AL) loop to efficiently identify energy-absorbing configurations in electronic junctions with a 2D "working material" to be optimized. The GVAE maps physical parameters - site energies, coupling matrices, temperature - into a smooth latent space that distinguishes cooling from dissipative parameter regimes, achieving a classification discriminability of on held-out configurations and on unseen extrapolation topologies. An AL pipeline based on Monte Carlo dropout then targets the decision boundary, iteratively selecting the candidates where the model is least certain for simulation. Over 32 AL cycles, the uncertainty-driven strategy discovers energy absorbing systems at rates consistently exceeding the random baseline across all tested topologies (where only 1.9-8.4% of configurations exhibit the desired property), providing a data-efficient pathway for navigating a combinatorically large design space. The learned latent space organizes configurations along physically interpretable axes, that correlate with underlying microscopic model parameters. This hints at the existence of "cooling coordinates" - low-dimensional, physically interpretable quantities separating cooling from non-cooling regimes. Our framework, thus, paves a new way towards effective cooling of nanoscale electronic information processing devices, to reduce the environmental footprint of high performance computing.
Figures & tables
| Parameter | Range |
|---|---|
| Grid sizes (train) | , , , , , , , |
| Grid sizes (extrap.) | , , |
| K | |
| eV | |
| , eV |
| Topology | Samples | Hit rate | Topology | Samples | Hit rate |
| 5000 | 8.38% | 5000 | 3.52% | ||
| 5000 | 6.58% | 5000 | 1.86% | ||
| 5000 | 4.08% | 5000 | 2.08% | ||
| 5000 | 3.80% | 5000 | 3.66% | ||
| Extrapolation topologies, absent from training | |||||
| 6000 | 9.15% | 6000 | 1.18% | ||
| Random | MC-dropout | GP-BO (EI) | GP-BO (UCB) | Deep ensemble | |
|---|---|---|---|---|---|
| Positives found | |||||
| Oracle calls per positive | |||||
| Enrichment over random | |||||
| range across topologies | — | – | – | – | – |
| Final pool positive rate | |||||
| Interpolation ROC-AUC |
Appendix figures & tables8 assets
Supplementary material from the paper’s appendix.
Appendix
| Hyperparameter | Search Range | Selected |
|---|---|---|
| Latent dimension ( ) | 32 | |
| Hidden dimension ( ) | 128 | |
| Learning rate | ||
| Training batch size | 128 | |
| 2.674 | ||
| Change | PR-AUC | ROC-AUC |
|---|---|---|
| Latent dimension 64 | ||
| Latent dimension 128 | ||
| Latent dimension 64, GAT |
| Interpolation | Extrapolation | |||
|---|---|---|---|---|
| Model | ROC-AUC | PR-AUC | ROC-AUC | PR-AUC |
| GVAE | ||||
| AE | ||||
| GAT classifier | ||||
| GCN classifier | ||||
| Hit rate | Enrichment | AL | ||||||
|---|---|---|---|---|---|---|---|---|
| Topology | Random | EI | UCB | AL | EI | UCB | AL | budget |
| 0.044 | 0.060 | 0.097 | 0.309 | 17.3% | ||||
| 0.064 | 0.294 | 0.414 | 0.335 | 18.0% | ||||
| 0.078 | 0.434 | 0.397 | 0.327 | 23.9% | ||||
| 0.038 | 0.133 | 0.097 | 0.273 | 9.0% | ||||
| 0.033 | 0.203 | 0.216 | 0.256 | 7.8% | ||||
| System | Fock dimension | Median time per call | |
|---|---|---|---|
| 2 | 4 | 0.77 ms | |
| 3 | 8 | 1.5 ms | |
| 4 | 16 | 3.5 ms | |
| 6 | 64 | 32 ms | |
| 9 | 512 | 0.65 s | |
| 12 | 4096 | 76 s |