BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Memento EPFL// BEGIN:VEVENT SUMMARY:MEchanics GAthering -MEGA- Seminar: Talk1 - Hydropower research an d testing at EPFL\; Talk2 - Machine learning potentials for molecular liqu ids DTSTART:20190926T161500 DTEND:20190926T173000 DTSTAMP:20260511T120845Z UID:98e8e973f59bd7e411cf0eeb2118360c1fac61be9fba4b70c44ab6af CATEGORIES:Conferences - Seminars DESCRIPTION:Andres Mueller (LMH\, EPFL) & Max Veit (COSMO\, EPFL)\nHydr opower research and testing at EPFL by Andres Mueller (LMH\, EPFL)\nHyd ropower Research and Testing at EPFL. The stability of our electrical grid is challenged daily by the massive integration of inherently intermittent renewable energy sources (RES)\, such as solar or wind power. However\, i n order to reach the ambitious goals for the decarbonization of our electr icity supply\, the share of these RES is yet to be significantly increased . In addition to being a highly reliable and clean source of electricity\, hydropower also plays a crucial role in providing the grid with the neces sary flexibility in terms of load balancing and frequency regulation. This requires an extension of the hydropower plants' operating range\, occasio nally giving rise to cavitation induced system critical flow instabilities inside the machines. The research aiming at understanding\, simulating an d predicting these limiting factors\, in order to ultimately improve the p ower system's capacity to accommodate larger shares of RES\, is outlined i n this talk. Finally\, EPFL's role as an independent provider of reduced s cale model testing services in large hydropower projects around the globe is briefly introduced.\n\nMachine learning potentials for molecular liquid s by Max Veit (COSMO\, EPFL)\nMachine learning potentials for molecular liquids. The reliable prediction of the macroscopic properties of molecul ar liquids requires potential energy surface (PES) models that are not onl y accurate\, but computationally efficient enough to handle large systems and reach long time scales typically inaccessible to explicit quantum-mech anical methods. First\, I will introduce a new approach to the systematic approximation of the first-principles PES of a molecular liquid using the GAP machine learning method [A. Bartók\, M. Payne\, R. Kondor\, and G. Cs ányi\, Phys. Rev. Lett. 104\, 136403 (2010)]. By applying machine learnin g to separately approximate each physical component of the interaction ene rgy in a full many-body framework and with high and controllable accuracy\ , we can simulate the liquid accurately across a wide range of temperature s and pressures (with the inclusion of quantum nuclear effects) while gain ing physical insight into the inner workings of the fluid. Following the r ecent success of this approach on predicting the equation of state of comp ressed fluid methane [M. Veit\, S. K. Jain\, S. Bonakala\, I. Rudra\, D. H ohl\, and G. Csányi\, arXiv:1810.10475]\, I will discuss how recent impro vements in the efficiency of calculating representations of the atomic str uctures\, and from there\, the potential energy and force necessary to run simulations\, will foster further development of this type of PES for oth er molecular liquids\, as well as more demanding applications of these PES s to problems previously considered inaccessible to simulations with quant um-mechanical accuracy.\n  LOCATION:MED 2 2423 https://plan.epfl.ch/?room==MED%202%202423 STATUS:CONFIRMED END:VEVENT END:VCALENDAR