BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260421T090514Z
LOCATION:Plenary Room (Bldg. 6 - 001)
DTSTART;TZID=Europe/Stockholm:20260629T193900
DTEND;TZID=Europe/Stockholm:20260629T194000
UID:submissions.pasc-conference.org_PASC26_sess124_pos127@linklings.com
SUMMARY:Graph Neural Network Potentials for Million-Atom Molecular Dynamic
 s Simulations of Aluminum Solidification
DESCRIPTION:Ian Störmer and Julija Zavadlav (Technical University of Munic
 h)\n\nSolidification is ubiquitous in the fabrication of metal parts. Mole
 cular dynamics simulations can predict the microstructure and the correspo
 nding mechanical properties. However, both high accuracy of interatomic po
 tential energy and scalability to millions of atoms are required to captur
 e physically relevant grain and microstructure evolution. Classical potent
 ials for metals are limited in accuracy and chemical complexity. Machine-l
 earning interatomic potentials approach first-principles accuracy, but the
 ir applicability to large-scale systems remains an open challenge due to h
 igh computational cost. In this work, we train equivariant graph neural ne
 twork potentials (GNNPs) for pure aluminum, assess their performance, and 
 compare them to existing potentials. We find that classical potentials per
 form well in low-energy solid states but deteriorate in high-energy liquid
  states. Contrarily, our GNNPs remain accurate across all phases, improvin
 g the accuracy of the conducted simulations. We employ the developed model
  to predict key solidification properties and to conduct a million-atom so
 lidification simulation, demonstrating a pathway for feasible simulations 
 with near-first-principles-level accuracy at experimentally relevant syste
 m sizes, benefiting multiscale materials design.\n\n
END:VEVENT
END:VCALENDAR