Center for Computing Research
Automated Ensemble Analysis in Support of Nuclear Deterrence Programs
Managing the modeling-simulation-analysis workflows that provide the basis for Sandia’s Nuclear Deterrence programs is a requirement for assuring verifiable, documented, and reproducible results. The Sandia Analysis Workbench (SAW) has now been extended to provide workflow management through the final tasks of ensemble analysis using Sandia’s Slycat framework. This new capability enhances multi-platform modeling-simulation workflows through the Next Generation Workflow (NGW) system. With the goal of providing end-to-end workflow capability to the nuclear deterrence programs, the new technology integrates Slycat and SAW. It fills the workflow gap between computational simulation results and post-processing tasks of ensemble analysis and the characterization of uncertainty quantification (UQ). This work compliments simulation data management while providing encapsulated, targeted sub-workflows for ensemble analysis, verification and validation, and UQ. The integration of Slycat management into SAW affords a common point of control and configuration. This connects analysis with modeling and simulation, and provides a documented provenance of that analysis. The heart of the work is a set of innovative NGW components that harvest ensemble features, quantities of interest (QoIs), simulation responses, and in situ generated images, videos, and surface meshes. These components are triggered on-demand by the workflow engine when the prerequisite data and conditions are satisfied. Executing from an HPC platform, the components apply those artifacts to generate parameterized, user-ready analyses on the Slycat server. These components can eliminate the need for analyst intervention to hand-process artifacts or QoIs. The technology automates data flow and evidence production needed for decision-support in quantification of margins and uncertainty. Finally, these components deliver an automated and repeatable shortcut to Slycat’s meta-analysis crucial for optimizing ensembles, evaluating parameter studies, and understanding sensitivity analysis.
Contact: Hunt, Warren L.
Credibility in Scientific Machine Learning: Data Verification and Model Qualification
The Advanced Simulation and Computing (ASC) initiative on Advanced Machine Learning (AML) aims to maximize near and long-term impact of Artificial Intelligence (AI) and Machine Learning (ML) technologies on Sandia’s Nuclear Deterrence (ND) program. In this ASC-AML funded project, the team is developing new approaches for assessing the quality of machine learning predictions based on expensive/limited experimental data, with focus on problems in the nuclear deterrence (ND) mission domain. Guided by the classical Predictive Capability Maturity Model (PCMM) workflow for Verification, Validation, and Uncertainty Quantification (V&V/UQ), the project aims to rigorously assess the statistical properties of the input features when training a Scientific Machine Learning (SciML) model and examine the associated sources of noise, e.g., measurement noise. This will, in turn, enable the decomposition of output uncertainty into unavoidable aleatory part versus reducible model-form uncertainty. To improve Sandia’s stockpile surveillance analysis capabilities, the team uses signature waveforms collected using non-destructive functioning and identify the most-discriminative input features, in order to assess the quality of a training dataset. By further decomposing uncertainty into its aleatoric and epistemic components, the team will guide computational/sampling resources towards reducing the treatable parts of uncertainty. This workflow will enhance the overall credibility of the resulting predictions and open new doors for SciML models to be credibly deployed to costly or high stakes ND problems.
Contact: Rushdi, Ahmad
Machine Learning for Xyce Circuit Simulation
The Advanced Simulation and Computing (ASC) initiative to maximize near and long-term Artificial Intelligence (AI) and Machine Learning (ML) technologies on Sandia’s Nuclear Deterrence (ND) program funded a project focused on producing physics-aware machine learned compact device models suited for use in production circuit simulators such as Xyce. While the original goal was only to make a demonstration of these capabilities, the team worked closely with Xyce developers to ensure the resulting product would be suitable for the already large group of Xyce users both internal and external to Sandia. This was done by extending the existing C++ general external interface in Xyce and adding Pybind11 hooks. The result is that with release 7.3 of Xyce, the ability to define machine learned compact device models entirely in Python (the most commonly used machine learning language) and use them with Xyce will be publicly available.
Contact: Kuberry, Paul Allen
QSCOUT / Jaqal at the Frontier of Quantum Computing
DOE/ASCR is investing over 5 years in Sandia to build and host the Quantum Scientific Computing Open User Testbed (QSCOUT): a quantum testbed based on trapped ions that is available to the research community (led by Susan Clark, 5225). As an open platform, it will not only provide full specifications and control for the realization of all high level quantum and classical processes, it will also enable researchers to investigate, alter, and optimize the internals of the testbed and test more advanced implementations of quantum operations. To maximize the usability and impact of QSCOUT, Sandia researchers in 1400 (Andrew Landahl, 1425) have led the development of the Jaqal quantum assembly language, which has been publicly released in conjunction with a QSCOUT emulator. QSCOUT is currently hosting external user teams from UNM, ORNL, IBM, the University of Indiana, and the University of California at Berkeley for scientific discovery in quantum computing.
For more information contact email@example.com or visit https://www.sandia.gov/quantum/Projects/QSCOUT.html :POC: Andrew Landahl
Contact: Landahl, Andrew J
2020 Rising Stars Workshop Supports Women in Computational & Data Sciences
Rising Stars in Computational & Data Sciences is an intensive academic and research career workshop series for women graduate students and postdocs. Co-organized by Sandia and UT-Austin’s Oden Institute for Computational Engineering & Sciences, Rising Stars brings together top women PhD students and postdocs for technical talks, panels, and networking events. The workshop series began in 2019 with a two-day event in Austin, TX. Due to travel limitations associated with the pandemic, the 2020 Rising Stars event went virtual with a compressed half-day format. Nonetheless, it was an overwhelming success with 28 attendees selected from a highly competitive pool of over 100 applicants. The workshop featured an inspiring keynote talk by Dr. Rachel Kuske, Chair of Mathematics at Georgia Institute of Technology, as well as lightning-round talks and breakout sessions. Several Sandia managers and staff also participated. The Rising Stars organizing committee includes Sandians Tammy Kolda (Distinguished Member of Technical Staff, Extreme-scale Data Science & Analytics Dept.) and James Stewart (Sr. Manager, Computational Sciences & Math Group), as well as UT Austin faculty Karen Willcox (Director, Oden Institute) and Rachel Ward (Assoc. Professor of Mathematics).
For more information on Rising Stars, see https://risingstars.oden.utexas.edu
Contact: Stewart, James R.
IDEAS PSE Computational Platform Wins 2020 R&D 100 Award
The IDAES Integrated Platform is a comprehensive set of open-source Process Systems Engineering (PSE) tools supporting the design, modeling, and optimization of advanced process and energy systems. By providing rigorous equation-oriented modeling capabilities, IDAES helps energy and process companies, technology developers, academic researchers, and the DOE to design, develop, scale-up, and analyze new PSE technologies and processes to accelerate advances and apply them to address the nation’s energy needs. The platform is based on and extends the Pyomo optimization modeling environment originally developed at Sandia. IDAES has taken the core optimization capabilities in Pyomo and not only built a domain-specific process modeling environment, but also expanded the core environment into new areas, including logic-based modeling, custom decomposition procedures and optimization algorithms, model predictive control, and machine learning methods.
The IDAES PSE Computational Platform is developed by the Institute for the Design of Advanced Energy Systems (IDAES) and was recently awarded a 2020 R&D 100 Award. Led by National Energy Technology Laboratory (NETL), IDAES is a collaboration with Sandia National Laboratories, Berkeley Lab, West Virginia University, Carnegie Mellon University, and the University of Notre Dame.
For more information on IDAES, see https://idaes.org
Contact: Siirola, John Daniel
Investigating Arctic Climate Variability with Global Sensitivity Analysis of Low-resolution E3SM.
As a first step in quantifying uncertainties in simulated Arctic climate response, Sandia researchers have performed a global sensitivity analysis (GSA) using a fully coupled ultralow-resolution configuration of the Energy Exascale Earth System Model (E3SM). Coupled Earth system models are computationally expensive to run, making it difficult to generate the large ensembles required for uncertainty quantification. In this research an ultralow version of E3SM was utilized to tractably investigate parametric uncertainty in the fully coupled model. More than one hundred perturbed simulation ensembles of one hundred years each were generated for the analysis and impacts on twelve Arctic quantities of interest were measured using the PyApprox library. The parameter variations show significant impact on the Arctic climate state with the largest impact coming from atmospheric parameters related to cloud parameterizations. To our knowledge, this is the first global sensitivity analysis involving the fully-coupled E3SM. The results will be used to inform model tuning work as well as targeted studies at higher resolution.
Ultra-low atmosphere grid (left) and ultra-low ocean grid (right).
Points of contact: Kara Peterson (firstname.lastname@example.org) Irina Tezaur (email@example.com)
For more information on E3SM: https://e3sm.org/
Contact: Peterson, Kara J.
Machine-Learned Interatomic Potentials Are Now Plug-And-Play in LAMMPS
Researchers at Sandia and Los Alamos National Laboratories have discovered a new way to implement machine learning (ML) interatomic potentials in the LAMMPS molecular dynamics code. This makes it much easier to prototype and deploy ML models in LAMMPS and provides access to a vast reservoir of existing ML libraries. The key is to define an interface (MLIAP) that separates the calculation of atomic fingerprints from the prediction of energy. The interface also separates the atomic descriptors and energy models from the specialized LAMMPS data structures needed for efficient simulations on massively parallel computers. Most recently, a new model class has been added to MLIAP that provides access to any Python-based library, including the PyTorch neural network framework. This advancement was made under the DOE SciDAC/FES FusMatML project for the application of machine learning to atomistic models for plasma-facing materials in fusion reactors.
More information on the LAMMPS package for machine-learned interatomic potentials can be found here: LAMMPS MLIAP Package
Contact: Thompson, Aidan P.