Focus Topics

DCOMP Lead Focus Topics

APS March Meeting 2022

16.01.02 Building the Bridge to Exascale: Applications and Opportunities for Materials, Chemistry, and Biology (DCOMP, DCP, DMP, DPOLY)

The expanding landscape of high-performance scientific computing plays a critical role in modern scientific discovery through a merging of simulation, modeling, and experimental data workflows. Artificial intelligence (AI) is accelerating demand for capabilities at high-performance computing (HPC) facilities around the world.   Computers are at the dawning of the Exascale Era where exascale performance (i.e., 10^18 floating-point operations per second) is combined with AI, large data sets, and scalable simulation codes to extend performance achievements in the post-Moore’s law era. This focus session will bring together researchers with experience in using high-performance cyberinfrastructure, including supercomputers, communication networks, data resources, and scalable workflows to achieve breakthrough scientific results in materials, biological, and materials physics. This includes researchers at experimental facilities such as light sources, neutron sources, and microscopy facilities with extreme data-science requirements, including machine-learning approaches, and researchers in computational materials, computational chemistry and computational biophysics with experience in large-scale simulations.  Software-development projects preparing a variety of physics applications for exascale-class machines will also be presented.   This session will highlight forefront examples of the state-of-the-art in computational physics today leveraging large, national-scale infrastructure.


16.01.03 Electrons, Phonons, Electron Phonon Scattering, and Phononics (DCOMP, DMP)

Electron-phonon interactions play a central role in many phenomena, most classically the resistivity of metals at ordinary temperatures, and are important for electrical and thermal conductivity of thermoelectrics, the temperature dependence of the optical band gaps of semiconductors, and other phenomena such as phonon drag. This focus topic covers electron-phonon interactions emphasizing fundamental physics, direct computation, first principles and phenomenological theory, optical and phonon spectroscopy and novel effects in nanostructures, nanodevices, 2D materials, and bulk materials. This focus topic also includes the emerging area of phononics, in particular manipulating phonon eigenstates, coherent superpositions and non-linearities, for example for logical operations or to manipulate sound or heat in unconventional ways or topological acoustic materials, including active materials.


16.01.04 First-principles modeling of excited-state phenomena in materials (DCOMP, DCP, DMP)

Many properties of functional materials, including bulk and two-dimensional materials and their interfaces, as well as quantum and topological materials, derive from excited-state phenomena. These processes determine properties such as band gaps, excitonic effects, electron-phonon couplings, and out-of-equilibrium dynamics of charge, spin, orbital, and lattice degrees of freedom and their couplings. Gaining deeper insight into these properties will advance our fundamental understanding of electronic structure theory beyond ground-state phenomena, and will underpin the design of new and improved materials for applications in energy-efficient electronics, solid-state lighting, solar photovoltaics, photocatalysis, and quantum technologies.

Predictive calculations of electronic excitations require theoretical frameworks that go beyond ground state density functional theory (DFT). In recent years, Green’s function based many-body perturbation theory methods like RPA, GW, BSE and beyond GW/BSE have been adopted by a rapidly growing community of researchers in the field of computational materials physics. These have now become the de facto standard for the description of excited electronic states in solids and their surfaces. Ehrenfest dynamics and surface-hopping schemes, e.g., based on time-dependent DFT, are used to describe coupled electron-ion dynamics as the origin of interesting physics in photo- catalysis, surface chemical reactions, scintillators, or radiation shielding. Nonequilibrium Green’s Function methods with many-body interactions from first-principles can be promising to tackle complex ultrafast and out-of-equilibrium quantum dynamics of excitons, electrons, phonons and spin. The description of electron-phonon and spin-phonon interactions using many-body Green’s function and density-matrix methods is also becoming increasingly popular.

Advances in high performance computing and scalable implementations in several popular electronic structure packages enable further progress. Sophisticated calculations are accessible for many users and feasible for large, complex systems with up to a few hundred atoms. Coupling with machine learning methods, the computational cost of excited state calculations can be further lowered by orders of magnitude. These methods are increasingly applied to interpret experiments, such as spectroscopies and femto- second pump-probe measurements, and to computationally design functional materials, interfaces, and nano-structures.

This focus topic is dedicated to recent advances in many-body perturbation theory and the theory of electron-phonon interactions in the excited state: challenges, scalable implementations in electronic structure codes, new machine learning and data-driven approaches, and applications to functional materials, two-dimensional materials and their interfaces, and quantum and topological materials. It aims to attract researchers working on the nexus of electronic and optical properties of materials, electron-phonon interactions, as well as materials and device physics.


16.01.05 Machine Learning for Quantum Matter (DCOMP, DMP, GDS)

Quantum matter, the research field studying states of matter whose properties are intrinsically quantum mechanical, draws from areas as diverse as hard condensed matter physics, quantum information, quantum gravity, and large-scale numerical simulations. Recently, condensed matter, quantum information, material science, and atomic, molecular, and optical physics communities have turned their attention to the algorithms underlying modern machine learning, with an eye on making progress in quantum matter research. This has led to several breakthroughs where machine learning algorithms are used, for example, to study large-scale materials with unprecedented accuracy. Other examples concern the ab-initio simulation of electronic structure problems. As evidenced by the community embracing a wide array of activities related to research at the intersection of machine learning and quantum mechanics (e.g. KITP program on Machine Learning for Quantum Many-Body Physics, Machine Learning for Physics and the Physics of Learning at IPAM/UCLA, Machine learning for quantum design at Perimeter Institute, Machine Learning for Quantum Simulation at Flatiron Institute, ELLIS Quantum Machine Learning, NEURIPS workshop on machine learning and the physical sciences), as well as the continuous appearance of increasingly creative research activity in this area, it is clear that over the next few years, machine learning will become very important for the computational study of condensed matter, quantum information, and other areas of quantum physics.


  • Giuseppe Carleo (EPFL);
  • Juan Felipe Carrasquilla Alvarez (Vector Institute)
  • Eliska Greplova (Delft University of Technology)
  • Bingqing Cheng (Cambridge University)

16.01.06 Precision Many-Body Physics (DCOMP, DAMOP, DCMP)

Precise understanding of strongly correlated materials and models is a major goal of modern physics. Achieving this understanding normally requires four complementary ingredients and thus four distinct directions of research: (i) conducting experiments that aim at producing highly accurate data, (ii) developing effective theories addressing the relevant degrees of freedom and/or emergent phenomena characteristic of a given phase of matter; (iii) solving simplified strongly correlated microscopic models either numerically or analytically, and (iv) cross-validating theoretical predictions against empirical data qualitatively and, ultimately, quantitatively. The last decade has seen breakthroughs made in all the four directions. An impressive progress has been achieved, and more is anticipated, where models and methods from many-body physics can be tested with precision, and where entirely new systems are realized that still await their accurate description. For example, in the field of ultra-cold atoms it is now feasible to perform analog quantum simulations aiming at experimental realization of key many-body quantum models and engineer novel Hamiltonians. Controllable experimental platforms also started to address fundamental questions about non-equilibrium quantum dynamics, discovering new dynamical phases of matter with no equilibrium counterpart. The proposal is to organize focus sessions that will bring together researchers who share the goal of achieving controllable theoretical and experimental understanding of phenomena taking place in correlated many-body systems. The key topics of the session(s) may include exactly solvable models, dualities and correspondences between seemingly unrelated theories (enabling the transfer of results and ideas), first-principles numeric approaches (such as tensor network and density-matrix renormalization group methods; path-integral, stochastic-series, and diagrammatic Monte Carlo techniques, dynamic cluster approximations, linked-cluster expansions, etc.); effective coarse-grained description of quantum phases and phase transitions; analytical and numerical methods for topological phases (including quantum spin-liquids, topological insulators, fractional quantum Hall states, and Chern insulators, etc.), and precise experimental studies of strongly correlated bosonic, fermionic, and spin systems (both at and out of equilibrium).


16.01.07 Understanding Amorphous Matter Through Modeling and Simulation (DCOMP, DPOLY, DSOFT, GSNP)

Simulations of complex models of disordered matter such as glasses, poly-disperse colloidal aggregates, amorphous polymers, dense suspensions, and granular packings have provided much insight into their underlying physics. Surprisingly, robust structural features can be discerned to arise in systems that otherwise seem to be entirely random by nature. Concepts related to non-equilibrium thermodynamics, energy landscapes, polymer entanglement are some examples that have been elucidated by carefully constructed computational investigations. New algorithms for optimizing and exploring structural, mechanical, rheological properties of disordered systems continue to emerge, providing tools for addressing these ubiquitous materials systems. This session will discuss applications of established computational methods toward the study of these systems as well as development of novel algorithms that hold the promise of overcoming limitations of current simulation strategies.


16.01.08 Computational Statistical Mechanics: Advances and Applications (DCOMP, GSNP)

Systems with a large number of degrees of freedom are fundamental for describing macroscopic behavior in a wide area of physical sciences and beyond. Consequently, statistical mechanics is one of the foundational theories for describing systems with disorder, limited microscopic knowledge and at finite temperature. Computer simulations are indispensable to advance understanding in these areas. In conjunction with modern computer architectures, new and improved algorithms and methodologies are being developed to enable increased computational performance and accuracy and the study of more complex physical problems. The main focus of this session will be on new methods and capabilities of Monte-Carlo, Molecular-Dynamics and Spin-Dynamics methods and their application. This Focus Session aims to provide a platform to bring together researchers from different disciplines to discuss and showcase recent advancements in computational statistical physics, as well as their applications to research problems at the frontier of computational physics.

Topics include (but are not limited to): simulation algorithms or techniques in computational statistical mechanics and their related studies; implementation techniques for modern computer architectures (e.g. GPUs or many-core processors); theoretical studies and discoveries aided or enhanced by computer simulations; applications of computational statistical mechanics to the study of thermodynamics, phase stability and transitions, critical phenomena at equilibrium, disorder driven phenomena, non-equilibrium, or irreversible processes for physical systems such as spin models, solid state systems, polymers and biological systems.


16.01.09 Real Space Methods for the Electronic Structure Problem: New Algorithms and Applications (DCOMP)

Many interesting material properties can be understood and predicted by computation involving a solution of the electronic structure problem. The combination of new algorithms applied to high performance computing platforms promises a number of potential advances in the understanding of the theory of complex materials and in the analysis of new experimental work on advanced materials. Yet, solution of the electronic structure problem remains computationally challenging when the system of interest contains a large number (thousands) of atoms. Real-space numerical electronic structure methods are mathematically robust, accurate and ideally suited for contemporary massively parallel computational resources.  Real space methods have successfully been applied to both ground state and excited states, especially for localized systems such as nanoscale structures. New algorithms have been developed to optimize solutions to eigenvalue problems and expedite or circumvent the computation of empty states in excited state computations.  Topics in this focus session include but are not limited to: real space or grid-based methods using finite differencing, finite elements, or variations thereof; applications to large nanoscale systems, ab initio molecular dynamics, noncollinear magnetic systems, optical excitations, and molecular transport; new algorithms designed for expediting and applying these methods to state of the art computational platforms.


  • Jim Chelikowsky (UT Austin);
  • Leoor Kronik (Weizmann, Israel)
  • Angel Rubio (Germany)

16.01.10 Artificial Intelligence and Data Science for the Computational Design and Discovery of Novel Materials (DCOMP, DMP)

The availability and improvements in high-performance computing and the development of advanced computational algorithms and techniques accelerate the discovery and rational design of functional materials. They also enable high-throughput calculations for rapid screening of novel materials with desired functionalities and properties, building material databases to apply artificial intelligence and data science techniques. This focus topic will cover research efforts in the developments and applications of methodologies for materials discovery by using novel data-driven approaches and machine learning methods to design materials with specific and targeted functional properties from first principles data. The focus topic concentrates on computational materials design and discovery, development of accessible and sustainable data infrastructure, development of new data analytic tools and statistical algorithms, advanced simulations of material properties in conjunction with new device functionality, uncertainty quantification; advances in predictive modeling that leverage machine learning and data mining, algorithms for global structure and property optimizations, and computational modeling of materials synthesis. The technical applications include but are not limited to electronic and optoelectronic materials, magnetism and spintronics, quantum materials, energy conversion and storage, complex alloys, and low-dimensional materials.


16.01.11 Emerging Trends in Molecular Dynamics Simulations and Machine Learning (DCOMP, DBIO, DPOLY, DSOFT, GDS)

Recent advances in deep learning have created a mini-revolution in all areas of Physics. Deep Learning (DL) has accelerated determination of complex energy landscapes, and stimulated algorithm design and data analytics. Availability of exascale machines within 1-2 years will make it easier to model hard and soft materials and biological systems with deep learning in conjunction with molecular dynamics (MD) simulations. Multimillion-to-billion atom MD simulations with DL trained on ab initio quantum mechanical simulations can reliably describe charge transfer, bond breaking/bond formation, and chemical reactions in materials under normal and extreme operating conditions. Generative models such as variational autoencoder (VAE) and generative adversarial networks (GAN) are very powerful DL models and have shown great success in creating material atomic structure for a desired property. Reinforcement learning (RL) is another widely used technique in DL domain, which is more suitable for these problems involving an optimization task and a sequential decision making under uncertainty. For examples, RL models have been used to predict reaction pathways, optimal conditions for chemical reactions and even design polymers with desired physical properties. Combining coarse grained and atomistic modeling with DL methods enable high throughput screening of materials. Accelerated dynamics approaches have enabled MD simulations to reach sufficiently long-time scales to study rare events. Novel task parallel frameworks are emerging for the analysis of peta-to-exascale DL-MD simulations.

Invited and contributed presentations in focus sessions will cover a wide range of topics that include but are not limited to:

  1. Deep learning for energy landscapes and force field development
  2. On-the-fly coarse and fine graining of MD simulations
  3. Accelerated dynamics using reinforcement learning
  4. Materials design using VAE and GAN
  5. Peta-to-exascale algorithms for hybrid deep learning and MD simulations


16.01.12 Modeling the electrochemical interface and aqueous solutions (DCOMP, DCP)

This session will focus on understanding the microscopic mechanisms that drive the function of electrochemical and photoelectro-chemical interfaces. These include metallic and semiconducting interfaces. Specific topics that will be covered are: electro-catalysis, photocatalysis, electrochemistry and ionic transport at these interfaces. The session will welcome talks that focus both on method development and applications, including the search for more efficient functional interfaces.


16.01.13 Extreme-Scale Computational Science Discovery in Fluid Dynamics and Related Disciplines (DCOMP, DFD)

Computational Physics is a broad, interdisciplinary field of research, which involves the development of computational techniques and the analysis of large datasets combined with theory, experiments and modeling, in concert with the so-called domain sciences.  One important and itself broad field of science represented by APS is fluid dynamics, where researchers have long embraced scientific computing based on partial differential equations, but have may have been somewhat late in recognizing the potential of machine-learning and artificial intelligence, which appear to be paradigms of rapidly increasing importance. 

This focused section will cover three main themes, namely, (a) Progress in extreme-scale HPC techniques for fluid flow problems, (b) Challenges in enabling access to massive data sets for non-HPC researchers, and (c) Machine learning, artificial intelligence and data analytics as applied to fluid dynamics.