*Note all times are Eastern Daylight Time.
Organizers: P. Ganesh, Bobby Sumpter, Thomas Maier, Maxim Ziatdinov (ORNL CNMS)
With the scalability presented by complex machine-learning (ML) and artificial intelligence (AI) approaches on modern hybrid computer architectures, there is a growing interest in applying such AI methodologies to ‘learn’ new physical and chemical insights from both big and small datasets. At the same time, the classical (correlative) AI/ML solutions may not be adequate for meaningful applications in the domain sciences as they typically do not take into account the causal relationships as well as the prior domain knowledge. This workshop will focus not only on leveraging well established AI/ML-based approaches but also latest developments in physics-guided AI/ML to enable novel technical solutions to computational bottlenecks, that need to be overcome to solve some of the current grand-challenge scientific problems in materials discovery and control, for future energy and computing applications. Specifically, the focus will be on using AI/ML-based methods in:
(i) Multi-fidelity simulation of materials: Simulations of materials at the atomic-scale encompass a wide range of methods – from the most exact to the most approximate. A major computational bottleneck remains – how can we quantify uncertainties in predicted material properties due to varying approximations in the underlying ab initio based quantum-mechanical approaches? This is particularly relevant for materials with strong electron-electron correlations, where practical approaches are either limited to a full quantum-mechanical Hamiltonian with approximate treatments of the correlations or to an exact treatment of correlations in restricted model Hamiltonians with very few degrees of freedom. The challenge lies not only in accelerating these methods, but also assessing the uncertainties in measured observables in both of these approaches due to their underlying approximations so as to make reliable predictions of experimentally measured properties. Probabilistic AI/ML-approaches that incorporate physical priors and constraints can potentially play a crucial role in both accelerating costlier exact methods, while at the same time quantifying the predictive uncertainties and propagating them across these complimentary approaches to make trustworthy predictions.
(ii) Multi-scale simulation of materials: Material properties fundamentally emerge at the atomic-scale (~few Å), where a quantum-mechanical description of the electronic-structure is required, while their functionalities due to these fundamental interactions start to emerge at larger length-scales. A major computational bottleneck remains – how can we bridge the scale from few atoms to hundreds of nano-meters so as to predict emergent functionalities – both static as well as in-operando? This is particularly relevant for reactive systems, where reactions involve few atoms, but influence the functionality of the whole nano-scale system. Examples include redox-reactions under bias at solid-solid and solid-liquid interfaces of catalytic, energy-storage and memory-materials, autocatalytic reactions in polymeric materials, synthesis of materials from few-atoms to a nanostructure, among others. Active and transfer learning based AI-approaches can potentially help with bridging the length scales, with uncertainties computed from multi-fidelity ab initio calculations.
(iii) Multi-physics simulation of materials: Heterogeneities in ordered systems come with varying characteristic length-scales and end up eventually dictating the response of meso-scale materials. A major computational bottleneck remains – how can we discover the governing equations of a material with characteristic heterogeneities that determine its response to external stimuli at the mesoscale? While quantum-mechanical evolution of the electronic-wavefunction is applicable to describe the response at the very smallest length-scales, Landau- or Euler-Lagrangian or phase-field formalisms are more appropriate to describe response to external fields at the mesoscale, say in the case of topological defects in ferroelectric materials, skyrmions in magnetic materials or ordering in charged polymeric systems. A combination of variational autoencoders and symbolic regression-based AI/ML techniques, specifically for dynamical systems, can potentially extract relevant ‘coarse-grained’ physical laws to treat dynamical response of heterogeneous meso-scale systems from multi-scale and multi-fidelity simulations.
Organizers: Xiaoping Wang (ORNL NScD), Zachary Morgan (ORNL NScD), Feng Ye (ORNL NScD) and Christina Hoffmann (ORNL NScD)
The two-day workshop will showcase the structural study of Bragg and diffuse scattering from single crystal neutron diffraction using advanced software tools in JANA2020 and rmc-discord. Topical lectures will cover concepts and applications of symmetry relationships to solve and refine complex crystal structures from Bragg diffraction, including twinning, modulated structure associated with local ordering, charge and magnetic phase transitions using the JANA2020 program. A new program rmc-discord recently developed for Reverse Monte Carlo refinement of diffuse scattering from three-dimensional single crystals data will be introduced for the first time. Hands-on tutorials on JANA2020 and 3D RMC refinements will be provided. Additional instructions will show how data can be further analyzed in other programs. Both software programs will be available for download.
Confirmed Speakers: Vaclav Petricek (Institute of Physics, Czech Republic); Margarida S. Henriques (Institute of Physics, Czech Republic); Zachary Morgan, Neutron Scattering Division, ORNL; Feng Ye, Neutron Scattering Division, ORNL; Xiaoping Wang, Neutron Scattering Division, ORNL
Organizers: Chris Tulk (ORNL NScD) and Ilia Ivanov (ORNL CNMS)
ORNL, its user facilities, and its user community are becoming uniquely positioned to study both the in-situ synthesis and in-situ characterization of materials that can only form under extreme experimental conditions. Indeed, a new extreme environments inelastic instrument-TITAN, which will complement capabilities elsewhere within NScD, has been proposed for the SNS - STS. It is the goal of this workshop to bring together scientists to discuss current and near future possibilities of conducting chemistry under exotic conditions. For example, we foresee a dynamic discussion on a broad range of topics where realization of metastable states through extreme conditions opens opportunities to access metastable states or intermediates which were not accessible in traditional chemistry, which will lead to a new range of products, new technologies, and new research break throughs. Our intent is to include chemistry of energy related materials, environment chemistry, including CO2, and chemistry enabling sustainable, recyclable, upcycling polymers enabling cycling economy of the future. Sessions could include, Chemistry in Plasma (hot and cold), Chemistry and Electrochemistry in critical CO2, Mechanochemistry, Chemistry Under Mixed Radiation Fields, and Chemistry Under Extreme Pressure (a relatively new field being pioneered at CNMS and NScD where chemical properties of the reactants are dramatically changed as they are brought within the distance of atomic interactions, enabling room temperature reactions which usually happen under very high temperatures).
Organizers: Matt Stone (ORNL NScD), An-Ping Li (ORNL CNMS), Wonhee Ko (ORNL CNMS), Zheng Gai (ORNL CNMS), Clarina Dela Cruz (ORNL NScD), Allen Scheie (ORNL NScD)
This workshop is a partnership between the Center for Nanophase Materials Sciences (CNMS) and the Quantum Materials Initiative (QMI) of the Neutron Scattering Division (NSD) of Oak Ridge National Laboratory (ORNL). The interactions involved in generating quantum states in materials span multiple length, energy, and time-scales. These are effectively probed using spectroscopy probes such as the neutron scattering facilities available at the High Flux Isotope Reactor (HFIR) and the Spallation Neutron Source (SNS), as well as the scanning tunneling microscopy (STM), electron energy loss spectroscopy, optical and laser spectroscopy available at the CNMS. These complementary user facilities co-located at ORNL provide powerful tools to understand and manipulate quantum materials. This workshop aims to bring together the user communities of these facilities with the local expertise to showcase recent quantum materials research from SNS, HFIR, and CNMS.
The workshop will be a two-day VIRTUAL event aiming to:
- Highlight the world class science being done by the quantum materials research community.
- Promote the collaborations of quantum materials researchers who use the CNMS, SNS, and HFIR facilities.
- Provide an opportunity for open discussions between facility users and ORNL staff.
Confirmed Speakers: M. Zahid Hasan (Princeton University); Cui-Zu Chang (The Penn State University); Hari Manoharan (Stanford University); Robert Klie (University of Illinois Chicago); Brian LeRoy (University of Arizona); Pengcheng Dai (Rice University); Chris Leighton (University of Minnesota); Kemp Plumb (Brown University); Stephen Wilson (University of California, Santa Barbara); Shan Wu (University of California, Berkeley); Bo Yuan (University of Toronto); Xiojian Bai (Oak Ridge National Laboratory); Shang Gao (Oak Ridge National Laboratory)
Organizers: Yan Chen (ORNL NScD), Alicia Manjon Sanz (ORNL NScD), Jue Liu (ORNL NScD), Jong K. Keum (ORNL NScD), Thomas Proffen (ORNL NScD), Ke An (ORNL NScD)
The advances of materials synthesis science shape future technologies of energy, environment, transportation and so on with broad and deep impacts on daily life. Expedited understanding of the underlying physical and chemical principles in materials synthesis is a key to the synthesis design to accelerate the predict-synthesize-measure loop for new materials. But the reactive, dynamic, and non-equilibrium conditions in most synthesis processes make it difficult to fully determine the mechanism and synthesis pathways via post-mortem characterizations. The in-situ methods by using advanced tools are uniquely suited to deliver the hidden information, such as metastable states, intermediate step, reversible process, dynamic evolution etc. That aids the understanding the mechanisms and thus the thermodynamic and kinetic controls.
In this workshop, the capability of in-situ tools at the user facilities SNS/HFIR/CNMS for materials synthesis research on broad energy and engineering applications is demonstrated, including neutron and X-ray diffraction, pair distribution function and scattering. Having instrument specifics to match the requirement on length scales, time scales, sensitivities as well as the synthesis environment, different instruments are used to reveal in-situ atomic occupancy and rearrangement, phase transition, morphology evolution etc. Through the workshop, the material synthesis science along with the in-situ tools are broadcasted to the NScD/CNMS user communities in various fields from academics to industry, in order to benefit the communities from synergistic utilization of those in-situ tools in synthesis science. Potential high-impact directions of material synthesis science will be discussed, and the demands on instrumentation, and sample environment will be collected to direct future development of the in-situ tools at the user facilities. The needs and development of the advanced computation and software will also be discussed, which enable both the automation of fast process of the high-throughput in-situ data and the real-time data analytics, with theory and modeling, to guide the productive synthesis control for desired materials.
Confirmed speakers: Jian Luo (U. of California San Diego), Peter Liaw (U. of Tennessee), Irene Peterson (Corning Inc.), Enyuan Hu (Brookhaven National Lab), Lu Cai (Idaho National Lab), Kai Xiao (ORNL), Carlos Marti-Gastaldo (Universidad de Valencia), YQ Cheng (ORNL), Efrain Rodriguez (U. of Maryland), Pengfei Cao (ORNL)
Organizers: Barry Winn (ORNL NScD), Masa Matsuda (ORNL NScD), Huibo Cao (ORNL NScD), Peter Jiang (ORNL NScD), Ovi Garlea (ORNL NScD), Josh Pierce (ORNL NScD)
This workshop showcases the polarized neutron diffraction and inelastic scattering community in the United States. Science emphasis is on quantum materials, hydrogenous systems, and innovative combinations thereof. The range of techniques includes half polarized, local susceptibility, longitudinal polarization analysis and 3D polarization analysis, spherical neutron polarimetry, dynamic nuclear polarization and enhanced resolution. Users of the polarization capabilities of PTAX, DEMAND and HYSPEC will present their findings for a day and a half. During the final afternoon the user community is briefed about emerging polarization capabilities and compatible sample conditions.
Organizers: Wellington Leite (ORNL NScD) and Haden Scott (ORNL NScD)
Small-angle neutron scattering (SANS), combined with contrast variation and/or deuteration, is particularly well suited to studying the structure of biomacromolecules under physiologically relevant conditions. This workshop will provide a practical guide on how to carry out, analyze, and interpret data from SANS experiments on biomembranes as well as biomacromolecular complexes. A practical tutorial section will demonstrate how to model SANS data for biomembranes using the EZ-SDP model and the use of appropriate modeling approaches for the analysis multicomponent protein complexes.
Organizer: Nhan Tran (Fermilab)
Machine Learning techniques executed in FPGAs can be very powerful tools for the realtime data processing. The hls4ml package allows to translate trained neural network models into synthesizable FPGA firmware. The firmware library targets efficient, ultrafast inference for its original application in real-time processing in particle physics. However, the generality of the package makes it applicable to a wide range of scientific and industry areas in which real-time processing on-device is needed.
In this tutorial we will give hands on experience with the workflow, including:
- Demonstration of the easy to use, yet deep customisation options hls4ml provides, including tunable parallelism and quantization.
- Model pruning, observing the impact on the resource usage of the inference.
- Quantization-aware training, resulting in low precision weights and activations and enabling very lightweight inference without loss of model accuracy.
- Synthesizing the FPGA firmware and evaluating the relevant metrics.
Attendees should have basic familiarity with Python, machine learning concepts, and ideally hands on experience with ML frameworks. Knowledge of FPGAs is advantageous, but not essential. Prerequisites: We will authenticate participants to our interactive tutorial notebooks using Github accounts. If you intend to take part in the tutorial, and do not already have a Github account, please sign up for one: https://github.com/
Confirmed Speakers: Ben Hawks and Jovan Mitrevski (Fermilab)
Organizers: Lilin He (ORNL NScD) and Yangyang Wang (ORNL NScD)
Small-angle scattering (SAS) of neutron or X-ray is an important characterization technique for soft materials. This workshop aims to train graduate students, postdoctoral researchers, and early-career materials scientists who want to get an in-depth knowledge on how to analyze SAS data. The workshop will cover lectures on the fundamentals of SAS data analysis and hands-on practical exercises. The topics include model-free and model-dependent data analysis methods. The use of SasView will be introduced. Scattering data from materials with hierarchical structures and anisotropic scattering will be touched on as well.
Confirmed speakers: Dr. Otto Glatter, Dr. Gregory Beaucage, Dr. Stephen King
Organizer: Andrei Savici (ORNL NScD)
Understanding single crystal diffuse neutron scattering (both elastic and inelastic) begins with data reduction (transformation from time of flight to momentum and energy transfer), and visualization. This tutorial is primarily targeted towards users of direct geometry inelastic instruments (ARCS/CNCS/HYSPEC/SEQUOIA) and for those looking at diffuse scattering on TOPAZ/CORELLI. It aims to introduce the new workflows of generating multi-dimensional datasets, taking cuts and slices, and making plots using Mantid and the matplotlib python library. Following a brief theoretical description, users will have an opportunity to perform hands on data processing and visualization. A basic knowledge of Python programming is a prerequisite for this tutorial. Example datasets will be provided. The analysis.sns.gov compute facility will be available for the users to work through the tutorial.
Organizers: John Katsaras (ORNL NScD), Rana Ashkar (Virginia Tech), Ben Doughty (ORNL Chemical Sciences Div.), Zach Liu (ORNL CNMS), Haden Scott (ORNL NScD), Charles Collier (ORNL CNMS), Jacob Kinnun (ORNL NScD), Mu-Ping Nieh (U. of Connecticut), Lilin He (ORNL NScD), Flora Meilleur (ORNL NScD), Hugh O'Neill (ORNL NScD), Kunlun Hong (ORNL CNMS), Jan Carrillo (ORNL CNMS), Volker Urban (ORNL NScD)
The nanoscale enables unique phenomena that lead to novel materials and new applications. However, to access this length scale requires special measuring, modeling and manipulation tools. During this one day workshop, speakers will describe different experimental, computational and chemical synthesis techniques available at ORNL that allow users to access the nanoscale of soft and biological materials.