A new research paper published on arXiv details findings regarding the efficiency of standard Strang splittings used in kinetic Langevin dynamics. The study proves that these methods, commonly employed in Markov chain Monte Carlo algorithms, do not achieve the theoretical acceleration for sampling targets with a condition number $\kappa$. The research establishes lower bounds for the mixing time of OBABO and BAOAB schemes, demonstrating that ballistic cold-start mixing fails uniformly over a class of smooth strongly convex functions. Complementary upper bounds suggest that for fixed-parameter OBABO tunings, the optimal condition-number dependence for total variation mixing is linear, up to logarithmic factors. AI
RANK_REASON Academic paper published on arXiv detailing theoretical findings in computational mathematics. [lever_c_demoted from research: ic=1 ai=0.1]
- BAOAB
- Dieuleveut
- Kinetic Langevin dynamics
- Leimkuhler
- Markov chain Monte Carlo
- OBABO
- Paulin
- Strang splittings
- Taylor
- Whalley
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →