BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.2.3.1//EN
TZID:Asia/Kolkata
X-WR-TIMEZONE:Asia/Kolkata
BEGIN:VEVENT
UID:197@cds.iisc.ac.in
DTSTART;TZID=Asia/Kolkata:20260525T153000
DTEND;TZID=Asia/Kolkata:20260525T163000
DTSTAMP:20260518T114150Z
URL:https://cds.iisc.ac.in/events/m-tech-research-thesis-colloquium-cds-da
 ta-driven-and-low-precision-methods-for-improving-and-accelerating-computa
 tional-fluid-dynamics-solvers/
SUMMARY:M.Tech Research Thesis {Colloquium}: CDS: "Data-driven and low-prec
 ision methods for improving and accelerating computational fluid dynamics 
 solvers."
DESCRIPTION:DEPARTMENT OF COMPUTATIONAL AND DATA SCIENCES\nM.Tech Research 
 Thesis Colloquium\n\n\n\nSpeaker : Mr. Surya Datta Sudhakar\nS.R. Number :
  06-18-01-10-22-24-1-24309\nTitle : "Data-driven and low-precision methods
  for improving and accelerating computational fluid dynamics solvers."\nRe
 search Supervisor : Dr. Konduri Aditya\nDate &amp\; Time : May 25\, 2026\,
  03.30 PM\nVenue : # 102 CDS Seminar Hall\n\n\n\nABSTRACT\n\nAdvancing sim
 ulation of turbulent and reacting flows poses two distinct challenges\, th
 e accurate representation of unresolved physical processes and the efficie
 nt utilization of the modern computing hardware. This thesis addresses bot
 h directions through data-driven subgrid-scale modeling and low-precision 
 solver development.\n\nThe first part of this thesis investigates the tran
 sferability of data-driven subgrid-scale closures for passive scalar trans
 port in non-equilibrium turbulence\, motivated by the central importance o
 f passive scalar mixing in a wide range of engineering and environmental f
 lows\, including combustion\, atmospheric transport\, pollutant dispersion
 \, and heat and mass transfer. Accurate modeling of unresolved scalar tran
 sport remains a major challenge in large-eddy simulation\, while training 
 separate neural-network closures for every flow condition is computational
 ly expensive and impractical. This study considers decaying two-dimensiona
 l turbulence with passive scalar transport as a stringent testbed due to i
 ts non-stationary dynamics\, evolving spectra\, intermittency\, and strong
  nonlocal interactions across scales.\nNeural-network-based closure models
  are trained to predict unresolved scalar transport from filtered high-fid
 elity simulation data\, with the Schmidt number governing scalar mixing be
 havior. Transfer learning across Schmidt number regimes is examined throug
 h selective fine-tuning of pretrained models and compared against conventi
 onal closure models\, demonstrating the superior adaptability of learned c
 losures. The results reveal a strong directional asymmetry\, with models t
 rained at higher Schmidt numbers transferring more effectively to lower Sc
 hmidt regimes than the reverse. Layer-wise analysis shows that predictive 
 improvements arise primarily from adapting shallow network layers\, while 
 modifications to deeper layers often reduce performance. Loss landscape an
 alysis further provides insight into the optimization behavior underlying 
 successful transfer. Spectral analysis shows that the fine-scale behavior 
 of the learned scalar closures exhibits greater universality across Schmid
 t numbers\, whereas large-scale closure behavior remains more regime-depen
 dent. These findings establish a computationally efficient and physically 
 interpretable framework for transferable machine-learned scalar closures i
 n large-eddy simulation.\nThe second part of this thesis investigates low-
 precision (FP16) computing for accelerating chemically reacting flow simul
 ations. Although modern CPU and GPU architectures offer substantially high
 er throughput and lower memory and communication costs for reduced-precisi
 on arithmetic\, direct application to reacting-flow solvers is challenging
  due to the extreme dynamic range of species concentrations\, reaction rat
 es\, and thermodynamic variables\, which introduce significant risks of nu
 merical underflow and overflow. To address this\, a dynamically scaled sol
 ver framework for lean hydrogen–air autoignition with detailed chemical 
 kinetics is developed\, incorporating adaptive scaling\, exponent bookkeep
 ing\, and selective higher-precision treatment of numerically sensitive op
 erations. A risk-band analysis framework is further introduced to systemat
 ically identify variables and computational regions susceptible to precisi
 on-induced numerical failure. The computational performance of the develop
 ed implementations\, spanning reference formulations\, templated C++\, and
  GPU-capable kernels\, is characterized through roofline analysis using In
 tel VTune and NVIDIA NSight Compute. The results establish the feasibility
  of reduced-precision reacting-flow solvers while revealing the interplay 
 between numerical stiffness and dynamic range constraints.\nOverall\, this
  thesis investigates two complementary strategies for improving computatio
 nal fluid dynamics solvers by developing transferable machine-learned clos
 ures for improved large-eddy simulation modeling and by reducing communica
 tion cost through low-precision numerical computation.\n\n\n\nALL ARE WELC
 OME
CATEGORIES:Events,MTech Research Thesis Colloquium
END:VEVENT
BEGIN:VTIMEZONE
TZID:Asia/Kolkata
X-LIC-LOCATION:Asia/Kolkata
BEGIN:STANDARD
DTSTART:20250525T153000
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
TZNAME:IST
END:STANDARD
END:VTIMEZONE
END:VCALENDAR