- Principal Member of Technical Staff
- Extreme-scale Data Science and Analytics Department
- Sandia National Laboratories – Livermore, CA
About
My research combines
computational mechanics,
machine learning,
numerical linear algebra,
and
high-performance computing. The objective is to discover structure
in data to drastically reduce the cost of simulating nonlinear
dynamical systems at extreme scale. This enables high-fidelity models to be used for many-query applications such as
design optimization and
uncertainty quantification.
In 2011, I received my PhD in
Aeronautics
and Astronautics
(with a PhD minor in
ICME) from
Stanford University, where
Charbel Farhat was my
research adviser. From 2011 to 2014, I was the
President
Harry S. Truman Fellow at Sandia National Laboratories in Livermore,
CA. Since 2014, I
have been a Principal Member of Technical Staff at Sandia.
Recent publication highlights
Our preprint on structure-preserving model reduction for
marginally stable LTI systems via symplectic projection is now
available on the arXiv. (Joint work with
L. Peng)
Our preprint on
space–time nonlinear model reduction with least-squares Petrov–Galerkin projection is now
available on the arXiv. (Joint work with
Y. Choi)
Our preprint on using
time-evolution data to enable efficient
time-parallel numerical simulations is now
available on the arXiv. (Joint work with L. Brencher,
B. Haasdonk,
A. Barth)
Our paper on analyzing
least-squares Petrov–Galerkin projection in
nonlinear model reduction has been published in the
Journal of Computational Physics. (Joint work with
H. Antil,
M. Barone)
News
I will be giving a talk at the
Department of Mathematics Smith Colloquium at the
University of Kansas, May 4, 2017.
I will be giving the keynote lecture at the
U2 can UQ showcase at
University of Arizona, April 28, 2017.
I gave a talk at the
USACM Workshop on Uncertainty Quantification and Data-Driven Modeling in Austin, TX, March 23–24, 2017. (
slides)
I attended the
2017 SIAM Conference on Computational Science and Engineering in Atlanta, GA as a minisymposium organizer, speaker, and poster presenter, February 27 to March 3, 2017. (
talk slides)
I gave a talk at the
Workshop on Data-Driven Methods for Reduced-Order Modeling and Stochastic Partial Differential Equations at
BIRS, Banff, Canada, January 29 to February 3, 2017.
Our research group celebrated the end of the
2016 ROM/UQ summer internship program with a
math potluck. Thanks to Kookjin Lee, Sofia Guzzetti, Jiahua Jiang, Wayne Uy, Cagan Ozen, and Zhe Bai for a great summer!
Current funded projects
- Rigorous surrogates for quantifying margins of uncertainty (PI).
This project aims to develop fundamental nonlinear model-reduction
techniques to enable orders-of-magnitude reduction in the computational
resources needed to simulate large-scale dynamical systems in an
uncertainty-quantification context.
Sandia collaborators: Ray Tuminaro, Drew Kouri, Brian Freno.
External collaborators: Kookjin Lee (Univ of Maryland), Howard Elman (Univ of Maryland), Wayne Uy (Cornell Univ), Matthew
Zahr (Stanford), Sumeet Trehan (Stanford), Louis Durlofsky (Stanford), Lukas Brencher (Univ Stuttgart), Bernard Haasdonk
(Univ Stuttgart), Matthias Morzfeld (Univ Arizona), Fei Lu (LBNL).
Keywords: nonlinear model
reduction; uncertainty quantification; time parallelism; high-performance
computing; adaptive refinement; Bayesian inference; machine learning; stochastic optimization
- Subsystem ROM and UQ for rapid, agile, extreme-scale simulation (PI).
The goal of this project is enable a divide-and-conquer strategy for uncertainty quantification and model reduction for decomposable systems at extreme-scale.
Sandia collaborators: Khachik Sargsyan, Mohammad Khalil, Youngsoo
Choi, Chi Hoang, Ray Tuminaro, Radoslav Bozinoski, Ali Pinar.
External collaborators: Sofia Guzzetti (Emory Univ), Jiahua Jiang (Univ of Massachusetts Dartmouth), H Cagan Ozen (Columbia Univ), Susie Sargsyan (Univ of Washington).
Keywords: nonlinear model
reduction; uncertainty quantification; domain decomposition; high-performance
computing; networks; multigrid; Bayesian inference
- Advanced ROM methods for thermal/mechanical responses (PI).
The objective of this project is to develop robust nonlinear model-reduction methods to simulate the thermal/mechanical failure of complex engineering systems in different configurations.
Sandia collaborators: Roy Hogan, Jeffrey Fike, Liqian Peng, Chi Hoang.
External collaborators: Zhe Bai (Univ of
Washington), Steven Brunton (Univ of
Washington)
Keywords: nonlinear model
reduction; uncertainty quantification; time parallelism; high-performance
computing; finite element analysis; structure preservation
- UQ of Structural Loading.
The goal of this project is demonstrate the effective, robust application of nonlinear model-reduction techniques to a large-scale compressible cavity-flow simulation.
Sandia collaborators: Irina Tezaur (PI), Jeffrey Fike, Matthew Barone, Youngsoo Choi.
External collaborators: Harbir Antil (George Mason Univ),
Maciej Balajewicz (Univ of Illinois).
Keywords: nonlinear model reduction; computational fluid dynamics;
uncertainty quantification; high-performance computing
Research interests
- Nonlinear model reduction
- Machine learning
- Computational mechanics
- High-performance computing
- Numerical linear algebra
- Uncertainty quantification
- Time-parallel methods
- Krylov-subspace methods
- Structure-preserving approximations
- Numerical optimization