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:20241120T082409Z
LOCATION:HG E 3
DTSTART;TZID=Europe/Stockholm:20240604T163000
DTEND;TZID=Europe/Stockholm:20240604T170000
UID:submissions.pasc-conference.org_PASC24_sess139_msa165@linklings.com
SUMMARY:Scalable and Consistent Mesh-Based Modeling of Fluid Flows with Di
 stributed Graph Neural Networks
DESCRIPTION:Minisymposium\n\nShivam Barwey, Riccardo Balin, Bethany Lusch,
  Saumil Patel, Ramesh Balakrishnan, and Pinaki Pal (Argonne National Labor
 atory); Romit Maulik (University of Pennsylvania, Argonne National Laborat
 ory); and Venkatram Vishwanath (Argonne National Laboratory)\n\nGraph neur
 al networks (GNNs) have shown considerable promise in accelerated mesh-bas
 ed modeling for applications like fluid dynamics, where models must be com
 patible with unstructured grids for practical simulation capability in com
 plex geometries. To realize the vision of robust mesh-based modeling, howe
 ver, the question of scalability to large graph sizes (O(10M) nodes and be
 yond) must be addressed, particularly when interfacing with unstructured d
 ata produced by high-fidelity computational fluid dynamics (CFD) codes. As
  such, we focus on the development of a distributed GNN that relies on nov
 el alterations to the baseline message passing layer to facilitate scalabl
 e operations with consistency. Here, consistency refers to the fact that a
  GNN trained and evaluated on one rank is arithmetically equivalent to eva
 luations on multiple ranks. Demonstrations are performed in the context of
  in-situ coupling of GNNs with NekRS, an exascale CFD code, using the Pola
 ris supercomputer at the Argonne Leadership Computing Facility. The crux o
 f the NekRS-GNN approach is to show how the same CFD domain-decomposition 
 strategy can be linked to the distributed GNN training and inference routi
 nes. Emphasis is placed on two modeling applications: (1) developing surro
 gates for unsteady fluid dynamics forecasts, and (2) mesh-based super-reso
 lution of turbulent flows.\n\nDomain: Chemistry and Materials, Climate, We
 ather, and Earth Sciences, Engineering, Computational Methods and Applied 
 Mathematics\n\nSession Chair: Alessandro Rigazzi (HPE)
END:VEVENT
END:VCALENDAR