Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Accelerating Eigenvalue Dataset Generation via Chebyshev Subspace Filter
Hong Wang, Jie Wang, Jian Luo, huanshuo dong, Yeqiu Chen, Runmin Jiang, Zhen huang
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add truncated FFT sorting before sequential Chebyshev solving (ablation) | SCSF · Generalized Poisson (dim 2500, tol 1e-12), L=20 | avg computation time (s): 8.248 -> 2.971 | Improved |
| add truncated FFT sorting before sequential Chebyshev solving (ablation) | SCSF · Generalized Poisson (dim 2500, tol 1e-12), L=400 | avg computation time (s): 42.60 -> 33.91 | Improved |
| add truncated FFT sorting (ablation) | SCSF · Helmholtz (dim 6400, tol 1e-8), L=400 | avg computation time (s): 66.66 -> 40.52 | Improved |
| use truncated FFT sorting instead of full greedy sorting (ablation) | SCSF · Helmholtz (dim 6400), dataset size 10^4 | sorting time (s): 592.7 -> 151.1 | Improved |
| use truncated FFT sorting instead of full greedy sorting (ablation) | SCSF · Helmholtz (dim 6400, tol 1e-8, L=400) | avg computation time (s): 40.52 -> 40.52 | No clear effect |
| initialize with eigenvectors of previous problem (no sorting) instead of random init (ablation) | ChFSI · main experiments | avg computation time (s): 1.2 to 1.5 times speedup | Improved |
| replace Chebyshev filter by other solvers using previous eigenvectors as initial subspace (ablation) | LOBPCG · Helmholtz (dim 6400, tol 1e-8), L=200 | avg computation time (s): 129.9 -> 95.9 | Improved |
| initialize with previous eigenvectors (ablation) | Eigsh and Krylov-Schur · Helmholtz (dim 6400, tol 1e-8), L=200 | avg computation time (s): 151.7 -> 150.2 (Eigsh); 98.34 -> 100.6 (KS) | No clear effect |
| initialize with previous eigenvectors (ablation) | Jacobi-Davidson · Helmholtz (dim 6400, tol 1e-8), L=200 | avg computation time (s): 489.6 -> 760.1 | Hurt |
| vary Chebyshev filter degree m from 12 to 40 (ablation) | SCSF · Helmholtz (dim 6400, tol 1e-8), L=400 | avg computation time (s): 43.92 (m=12), 40.52 (m=20), 43.50 (m=40) | No clear effect |
| increase inherited subspace size (ablation) | SCSF · Helmholtz (dim 6400, tol 1e-8), L=400 | avg computation time (s): 43.28 (dim 50) -> 37.43 (dim 100) | Improved |
| increase matrix dimension | SCSF vs Eigsh · Generalized Poisson, tol 1e-12, L=400 | avg computation time (s): SCSF 158.49 vs Eigsh 3162.28 at dim 10000 | Improved |
Closest papers by meaning
- STNet: Spectral Transformation Network for Solving Operator Eigenvalue Problem (NeurIPS 2025 · similarity 0.86)
- Operator SVD with Neural Networks via Nested Low-Rank Approximation (ICML 2024 · similarity 0.85)
- Learning to Discover Iterative Spectral Algorithms (NeurIPS 2026 · similarity 0.84)
- Accelerating Data Generation for Neural Operators via Krylov Subspace Recycling (ICLR 2024 · similarity 0.84)
- Explicit Eigenvalue Regularization Improves Sharpness-Aware Minimization (NeurIPS 2024 · similarity 0.83)
- On the Benefits of Active Data Collection in Operator Learning (ICML 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
