Optimization Research in Other Divisions

Accelerating the Discovery of New Batteries, Thermoelectrics, and Solar Cells with Optimized Materials Screening

Anubhav Jain: ajain@lbl.gov

New energy technologies are limited by the fundamental physical properties of the materials they are composed of. Thus, discovering and identifying new materials with improved functional properties is critical to the advancement of modern photovoltaic, thermoelectric, and battery technologies. Our group employs supercomputers in combination with high-throughput materials design (an accelerated design procedure akin to 'rapid computational prototyping') to simulate the properties of new materials for energy applications. However, among the billions of possible materials, only a handful may meet the technical and economic requirements necessary to be truly revolutionary. The large search space makes the high-throughput screening process impossible to conduct using exhaustive search, and one must be able to use past information (e.g., expert knowledge or the results of prior calculations) to guide the selection of materials candidates for computation.

We have already written a high-throughput computation and workflow engine called FireWorks to conduct simulations on a massive scale. Next, we are developing an optimization framework that integrates machine learning with FireWorks to find more good materials with fewer computations. Within a given computational budget, such a framework will maximize the chances of finding one or more revolutionary new materials and in essence provide a method such that, given a simulation procedure and a search space of candidates, can automatically determine the best materials candidates in that search space for a given application by conducting large numbers of ever-improving simulations automatically at supercomputing centers.

Click here for more info

Predicting optimal parameters for new compact particle accelerators

Remi Lehe: rlehe@lbl.gov

Laser-wakefield acceleration is a new technology which could potentially shrink the size of particle accelerators from hundreds of meters to only a few tens of centimeters! Being able to build compact and cheap particle accelerators with this technology would have a strong impact on high-energy physics and fundamental science, but also on applications in medicine and homeland security.

In a laser-wakefield accelerator, an intense femtosecond laser pulse (represented in red below) propagates through a gas jet and generates an accelerating plasma structure in its wake (represented in blue below). Under the right parameters (in terms of gas density and laser intensity), some of the electrons “break off” from the plasma structure and get accelerated to high energy (represented with yellow arrows below) – a phenomenon known as injection.

Determining the region of parameter space in which injection does happen is of paramount importance, since this is the regime which is required for a number of applications of laser-wakefield acceleration. In this context, we used surrogate optimization in order to reduce the uncertainty on the threshold between injection and no injection (represented by a wide red line below) by concentrating numerical simulations around the region of interest.

Click here for more info

Particle Accelerator Multi-Objective Global Optimization Using Stochastic Evolutionary Algorithms

Ji Qiang: jqiang@lbl.gov

Particle accelerators are one of the most important scientific instruments and have important applications in science, industry, and national security. For example, since 2009, the free electron laser (FEL) accelerator Linac Coherent Light Source (LCLS) has produced the world’s brightest X-ray pulses that help reveal fundamental processes in materials and living things. In 2012, the Large Hadron Collider (LHC) accelerator at the European Organization for Nuclear Research (CERN) discovered the Higgs boson, which explains the origin of mass in fundamental particles. Meanwhile, particle accelerators are also one of the most complex and expensive instruments that can employ several ten thousands of elements and cost multiple billion dollars. In order to improve the performance and reduce the cost of those accelerators, machine control parameter optimization is needed. The number of control parameters can vary from a few dozen to a few thousand, and the objective functions are highly nonlinear. In our study, we have developed a parallel multi-objective unified differential evolution algorithm with variable population and external storage for global parameter optimization [1,2]. This optimizer, together with our particle accelerator beam dynamics simulator has been successfully applied to the LCLS-II accelerator design [3,4].

[1] J. Qiang, C.E. Mitchell, S. Paret, R. Ryne, Y. Chao, “A Parallel Multi-objective Differential Evolution Algorithm for Photoinjector Beam Dynamics Optimization”, 2013, Proc. of International Particle Accelerator Conference

[2] J. Qiang, C. Mitchell, A. Qiang, “Tuning of an Adaptive unified differential evolution algorithm for global optimization”, 2016, Proc. of 2016 Congress on Evolutionary Computation (CEC)

[3] J. Qiang, C. Mitchell, “Electron beam dynamics optimization using a unified differential evolution algorithm”, 2014, Proc. of Free Electron Laser

[4] J. Qiang, "Start-to-End Beam Dynamics Optimization of X-Ray FEL Light Source Accelerators", 2016, Proc. North America – Particle Accelerator Conference

Click here for more info

PySMO & IDAES

Mayo Amusat: ooamusat@lbl.gov

Python-based Surrogate Modeling Objects (PySMO) is an open-source tool for generating accurate algebraic surrogates that are directly integrated with an equation-oriented (EO) optimization platform, specifically IDAES and its underlying optimization library, Pyomo. PySMO leverages the optimization capabilities of Pyomo to generate algebraic surrogates using a wide range of methods, with minimal manual parameter tuning. A key strength of PySMO is that the algebraic nature of its surrogates ensure that first and second order derivatives required by advanced gradient-based optimization algorithms can easily be computed. PySMO includes implementations of five sampling methods (uniform, Latin hypercube, centroidal voronoi tessellation, Hammersley, and Halton), and three surrogate methods (polynomial regression, Kriging, and radial basis functions), providing a breadth of capabilities suitable for a variety of engineering applications.

PySMO surrogates have been demonstrated to be very useful for gradient-based optimization in the Institute for Design of Advanced Energy Systems (IDAES), particularly in application areas such as (i) improving tractability and reducing computational expense of large-scale optimization problems (ii) enabling conceptual design of large search spaces, and (iii) enabling the algebraic representation of external simulation codes and complex phenomena in IDAES for process optimization and design.

IDAES brings the most advanced modeling and optimization capabilities to challenges around the reliable, environmentally sustainable and cost-efficient transformation and decarbonization of the world’s energy systems. The resulting IDAES integrated platform utilizes state-of-the-art equation-oriented optimization solvers and algorithms to enable the design, optimization and operation of optimization of complex, innovative steady state and dynamic processes. IDAES's process modeling framework and other tools provide advanced computational and simulation capabilities to examine and support decision-making on a complete range of technology options, and to support their rapid and effective implementation.

For more information about PySMO, see https://idaes-pse.readthedocs.io/en/1.5.1/surrogate/pysmo/index.html

For more information about IDAES, visit https://idaes.org/

Stochastic Processes for Function Approximation and Autonomous Data Acquisition at Large-Scale Experimental Facilities

Marcus Noack: MarcusNoack@lbl.gov

The execution and analysis of ever more complex experiments are increasingly challenged by the vast dimensionality of the parameter spaces that underlie investigations in the biological, chemical, physical, and materials sciences. While an increase in data-acquisition rates should allow broader querying of the parameter space, the complexity of experiments and the subtle dependence of the model function on input parameters remains daunting due to the sheer number of variables. To meet these challenges, new strategies for autonomous data acquisition are rapidly coming to fruition, and are being deployed across a spectrum of scientific experiments. One promising direction that is being explored by the community is the use of stochastic processes, especially Gaussian process regression (GPR). GPR is a quick, non-parametric, and robust approximation and uncertainty quantification method that can directly be applied to autonomous data acquisition.

This project aims to develop new stochastic-process-based mathematical and computational methods to achieve high-quality, domain-aware function approximation, uncertainty quantification, and, by extension, autonomous experimentation. The product being developed as part of this project is gpCAM, a simple-to-use, flexible, and HPC-ready Python-based software tool for Gaussian-process-based function approximation and autonomous experimentation.

More info: gpcam.lbl.gov