Berechnung genauer Thermochemischer Daten für flexible Moleküle
Chemische und Thermische Verfahrenstechnik
Zusammenfassung der Projektergebnisse
This project aimed at the development of new tools to predict the thermodynamic properties of flexible molecules, which are needed to optimize many chemical and engineering systems, e.g., for the energy transition. To do this, it would build on top of our Configuration Integral Monte Carlo Integration (CIMCI) method, at the core of which lies a numerical Monte Carlo (MC) integration called MISER that calculates the semi-classical configuration integral (CI). We realized that one of its most critical assumptions, the interchangeability of sample statistics with population ones, is not approriate for basing reliable stratification decisions on. This not only leads to it constantly underestimate its reported errors, but also fundamentally misguides and misshapes the way it decides on when and where to stratify space. Wanting to have a more mathematically sound basis for such a critical part of CIMCI, especially as MISER is also generally used in other fields that require high-dimensional, numerical MC integration, we decided to focus our efforts on creating a wholesale replacement for MISER. The result is a new MC integration method called Recursively Stratified with Iterative Pre-Sampling (ReSIP). For each recursive call, it iterates through multiple pre-sampling runs until it can obtain a high enough level of confidence in at least one trial stratification to continue. This is what fundamentally sets it apart from comparable state-of-the-art methods and what lets it outperform the original MISER.
