-
Colmena: Scalable Machine-Learning-Based Steering of Ensemble Simulations for High Performance Computing
Authors:
Logan Ward,
Ganesh Sivaraman,
J. Gregory Pauloski,
Yadu Babuji,
Ryan Chard,
Naveen Dandu,
Paul C. Redfern,
Rajeev S. Assary,
Kyle Chard,
Larry A. Curtiss,
Rajeev Thakur,
Ian Foster
Abstract:
Scientific applications that involve simulation ensembles can be accelerated greatly by using experiment design methods to select the best simulations to perform. Methods that use machine learning (ML) to create proxy models of simulations show particular promise for guiding ensembles but are challenging to deploy because of the need to coordinate dynamic mixes of simulation and learning tasks. We…
▽ More
Scientific applications that involve simulation ensembles can be accelerated greatly by using experiment design methods to select the best simulations to perform. Methods that use machine learning (ML) to create proxy models of simulations show particular promise for guiding ensembles but are challenging to deploy because of the need to coordinate dynamic mixes of simulation and learning tasks. We present Colmena, an open-source Python framework that allows users to steer campaigns by providing just the implementations of individual tasks plus the logic used to choose which tasks to execute when. Colmena handles task dispatch, results collation, ML model invocation, and ML model (re)training, using Parsl to execute tasks on HPC systems. We describe the design of Colmena and illustrate its capabilities by applying it to electrolyte design, where it both scales to 65536 CPUs and accelerates the discovery rate for high-performance molecules by a factor of 100 over unguided searches.
△ Less
Submitted 6 October, 2021;
originally announced October 2021.
-
Elucidating the Structure of the Magnesium Aluminum Chloride Complex electrolyte for Magnesium-ion batteries
Authors:
Pieremanuele Canepa,
Saivenkataraman Jayaraman,
Lei Cheng,
Nav Nidhi Rajput,
William D. Richards,
Gopalakrishnan Sai Gautam,
Larry A. Curtiss,
Kristin A. Persson,
Gerbrand Ceder
Abstract:
We present a rigorous analysis of the Magnesium Aluminum Chloro Complex (MACC) in tetrahydrofuran (THF), one of the few electrolytes that can reversibly plate and strip Mg. We use \emph{ab initio} calculations and classical molecular dynamics simulations to interrogate the MACC electrolyte composition with the goal of addressing two urgent questions that have puzzled battery researchers: \emph{i})…
▽ More
We present a rigorous analysis of the Magnesium Aluminum Chloro Complex (MACC) in tetrahydrofuran (THF), one of the few electrolytes that can reversibly plate and strip Mg. We use \emph{ab initio} calculations and classical molecular dynamics simulations to interrogate the MACC electrolyte composition with the goal of addressing two urgent questions that have puzzled battery researchers: \emph{i}) the functional species of the electrolyte, and \emph{ii}) the complex equilibria regulating the MACC speciation after prolonged electrochemical cycling, a process termed as conditioning, and after prolonged inactivity, a process called aging. A general computational strategy to untangle the complex structure of electrolytes, ionic liquids and other liquid media is presented. The analysis of formation energies and grand-potential phase diagrams of Mg-Al-Cl-THF suggests that the MACC electrolyte bears a simple chemical structure with few simple constituents, namely the electro-active species MgCl$^+$ and AlCl$_4^-$ in equilibrium with MgCl$_2$ and AlCl$_3$. Knowledge of the stable species of the MACC electrolyte allows us to determine the most important equilibria occurring during electrochemical cycling. We observe that Al deposition is always preferred to Mg deposition, explaining why freshly synthesized MACC cannot operate and needs to undergo preparatory conditioning. Similarly, we suggest that aluminum displacement and depletion from the solution upon electrolyte resting (along with continuous MgCl$_2$ regeneration) represents one of the causes of electrolyte aging. Finally, we compute the NMR shifts from shielding tensors of selected molecules and ions providing fingerprints to guide future experimental investigations.
△ Less
Submitted 8 November, 2015;
originally announced November 2015.
-
Transport properties of n-type ultrananocrystalline diamond films
Authors:
I. S. Beloborodov,
P. Zapol,
D. M. Gruen,
L. A. Curtiss
Abstract:
We investigate transport properties of ultrananocrystalline diamond films for a broad range of temperatures. Addition of nitrogen during plasma-assisted growth increases the conductivity of ultrananocrystalline diamond films by several orders of magnitude. We show that films produced at low concentration of nitrogen in the plasma are very resistive and electron transport occurs via a variable ra…
▽ More
We investigate transport properties of ultrananocrystalline diamond films for a broad range of temperatures. Addition of nitrogen during plasma-assisted growth increases the conductivity of ultrananocrystalline diamond films by several orders of magnitude. We show that films produced at low concentration of nitrogen in the plasma are very resistive and electron transport occurs via a variable range hopping mechanism while in films produced at high nitrogen concentration the electron states become delocalized and the transport properties of ultrananocrystalline diamond films can be described using the Boltzmann formalism. We discuss the critical concentration of carriers at which the metal to insulator transition in ultrananocrystalline diamond films occurs and compare our results with available experimental data.
△ Less
Submitted 26 December, 2006;
originally announced December 2006.
