MSGID: 1:317/3 64a8e666   
   PID: hpt/lnx 1.9.0-cur 2019-01-08   
   TID: hpt/lnx 1.9.0-cur 2019-01-08   
    Machine learning takes materials modeling into new era    
    Deep learning approach enables accurate electronic structure calculations   
   at large scales    
      
     Date:   
         July 7, 2023   
     Source:   
         Helmholtz-Zentrum Dresden-Rossendorf   
     Summary:   
         The arrangement of electrons in matter, known as the electronic   
         structure, plays a crucial role in fundamental but also applied   
         research such as drug design and energy storage. However, the   
         lack of a simulation technique that offers both high fidelity and   
         scalability across different time and length scales has long been   
         a roadblock for the progress of these technologies. Researchers   
         have now pioneered a machine learning- based simulation method   
         that supersedes traditional electronic structure simulation   
         techniques. Their Materials Learning Algorithms (MALA) software   
         stack enables access to previously unattainable length scales.   
      
      
         Facebook Twitter Pinterest LinkedIN Email   
      
   ==========================================================================   
   FULL STORY   
   ==========================================================================   
   The arrangement of electrons in matter, known as the electronic structure,   
   plays a crucial role in fundamental but also applied research such as drug   
   design and energy storage. However, the lack of a simulation technique   
   that offers both high fidelity and scalability across different time   
   and length scales has long been a roadblock for the progress of these   
   technologies.   
      
   Researchers from the Center for Advanced Systems Understanding (CASUS)   
   at the Helmholtz-Zentrum Dresden-Rossendorf (HZDR) in Go"rlitz, Germany,   
   and Sandia National Laboratories in Albuquerque, New Mexico, USA, have now   
   pioneered a machine learning-based simulation method (npj Computational   
   Materials), that supersedes traditional electronic structure simulation   
   techniques. Their Materials Learning Algorithms (MALA) software stack   
   enables access to previously unattainable length scales.   
      
   Electrons are elementary particles of fundamental importance. Their   
   quantum mechanical interactions with one another and with atomic nuclei   
   give rise to a multitude of phenomena observed in chemistry and materials   
   science.   
      
   Understanding and controlling the electronic structure of matter provides   
   insights into the reactivity of molecules, the structure and energy   
   transport within planets, and the mechanisms of material failure.   
      
   Scientific challenges are increasingly being addressed through   
   computational modeling and simulation, leveraging the capabilities of   
   high-performance computing. However, a significant obstacle to achieving   
   realistic simulations with quantum precision is the lack of a predictive   
   modeling technique that combines high accuracy with scalability across   
   different length and time scales. Classical atomistic simulation methods   
   can handle large and complex systems, but their omission of quantum   
   electronic structure restricts their applicability. Conversely, simulation   
   methods which do not rely on assumptions such as empirical modeling and   
   parameter fitting (first principles methods) provide high fidelity but   
   are computationally demanding. For instance, density functional theory   
   (DFT), a widely used first principles method, exhibits cubic scaling with   
   system size, thus restricting its predictive capabilities to small scales.   
      
   Hybrid approach based on deep learning The team of researchers now   
   presented a novel simulation method called the Materials Learning   
   Algorithms (MALA) software stack. In computer science, a software stack   
   is a collection of algorithms and software components that are combined   
   to create a software application for solving a particular problem.   
      
   Lenz Fiedler, a Ph.D. student and key developer of MALA at CASUS,   
   explains, "MALA integrates machine learning with physics-based approaches   
   to predict the electronic structure of materials. It employs a hybrid   
   approach, utilizing an established machine learning method called deep   
   learning to accurately predict local quantities, complemented by physics   
   algorithms for computing global quantities of interest."  The MALA   
   software stack takes the arrangement of atoms in space as input and   
   generates fingerprints known as bispectrum components, which encode the   
   spatial arrangement of atoms around a Cartesian grid point. The machine   
   learning model in MALA is trained to predict the electronic structure   
   based on this atomic neighborhood. A significant advantage of MALA is its   
   machine learning model's ability to be independent of the system size,   
   allowing it to be trained on data from small systems and deployed at   
   any scale.   
      
   In their publication, the team of researchers showcased the remarkable   
   effectiveness of this strategy. They achieved a speedup of over 1,000   
   times for smaller system sizes, consisting of up to a few thousand   
   atoms, compared to conventional algorithms. Furthermore, the team   
   demonstrated MALA's capability to accurately perform electronic structure   
   calculations at a large scale, involving over 100,000 atoms. Notably,   
   this accomplishment was achieved with modest computational effort,   
   revealing the limitations of conventional DFT codes.   
      
   Attila Cangi, the Acting Department Head of Matter under Extreme   
   Conditions at CASUS, explains: "As the system size increases and more   
   atoms are involved, DFT calculations become impractical, whereas MALA's   
   speed advantage continues to grow. The key breakthrough of MALA lies in   
   its capability to operate on local atomic environments, enabling accurate   
   numerical predictions that are minimally affected by system size. This   
   groundbreaking achievement opens up computational possibilities that   
   were once considered unattainable."  Boost for applied research expected   
   Cangi aims to push the boundaries of electronic structure calculations   
   by leveraging machine learning: "We anticipate that MALA will spark a   
   transformation in electronic structure calculations, as we now have a   
   method to simulate significantly larger systems at an unprecedented   
   speed. In the future, researchers will be able to address a broad   
   range of societal challenges based on a significantly improved baseline,   
   including developing new vaccines and novel materials for energy storage,   
   conducting large-scale simulations of semiconductor devices, studying   
   material defects, and exploring chemical reactions for converting   
   the atmospheric greenhouse gas carbon dioxide into climate-friendly   
   minerals."  Furthermore, MALA's approach is particularly suited for   
   high-performance computing (HPC). As the system size grows, MALA enables   
   independent processing on the computational grid it utilizes, effectively   
   leveraging HPC resources, particularly graphical processing units. Siva   
   Rajamanickam, a staff scientist and expert in parallel computing at the   
   Sandia National Laboratories, explains, "MALA's algorithm for electronic   
   structure calculations maps well to modern HPC systems with distributed   
   accelerators. The capability to decompose work and execute in parallel   
   different grid points across different accelerators makes MALA an   
   ideal match for scalable machine learning on HPC resources, leading to   
   unparalleled speed and efficiency in electronic structure calculations."   
   Apart from the developing partners HZDR and Sandia National Laboratories,   
   MALA is already employed by institutions and companies such as the   
   Georgia Institute of Technology, the North Carolina A&T State University,   
   Sambanova Systems Inc., and Nvidia Corp.   
      
       * RELATED_TOPICS   
             o Matter_&_Energy   
                   # Physics # Electronics # Materials_Science # Construction   
             o Computers_&_Math   
                   # Computer_Modeling # Software # Computers_and_Internet   
                   # Spintronics_Research   
       * RELATED_TERMS   
             o Circuit_design o Supercomputer o Computer_software o   
             Computer_simulation o Metal o Time_in_physics o Capacitor   
             o Artificial_intelligence   
      
   ==========================================================================   
      
    Print   
      
    Email   
      
    Share   
   ==========================================================================   
   ****** 1 ****** ***** 2 ***** **** 3 ****   
   *** 4 *** ** 5 ** Breaking this hour   
   ==========================================================================   
       * Six_Foods_to_Boost_Cardiovascular_Health   
       * Cystic_Fibrosis:_Lasting_Improvement *   
       Artificial_Cells_Demonstrate_That_'Life_...   
      
       * Advice_to_Limit_High-Fat_Dairy_Foods_Challenged   
       * First_Snapshots_of_Fermion_Pairs *   
       Why_No_Kangaroos_in_Bali;_No_Tigers_in_Australia   
       * New_Route_for_Treating_Cancer:_Chromosomes *   
       Giant_Stone_Artefacts_Found:_Prehistoric_Tools   
       * Astonishing_Secrets_of_Tunicate_Origins *   
       Most_Distant_Active_Supermassive_Black_Hole   
      
   Trending Topics this week   
   ==========================================================================   
   SPACE_&_TIME Asteroids,_Comets_and_Meteors Big_Bang Jupiter   
   MATTER_&_ENERGY Construction Materials_Science Civil_Engineering   
   COMPUTERS_&_MATH Educational_Technology Communications   
   Mathematical_Modeling   
      
      
   ==========================================================================   
      
   Strange & Offbeat   
   ==========================================================================   
   SPACE_&_TIME   
   Quasar_'Clocks'_Show_Universe_Was_Five_Times_Slower_Soon_After_the_Big_Bang   
   First_'Ghost_Particle'_Image_of_Milky_Way   
   Gullies_on_Mars_Could_Have_Been_Formed_by_Recent_Periods_of_Liquid_Meltwater,   
   Study_Suggests MATTER_&_ENERGY Holograms_for_Life:_Improving_IVF_Success   
   Researchers_Create_Highly_Conductive_Metallic_Gel_for_3D_Printing   
   Artificial_Cells_Demonstrate_That_'Life_Finds_a_Way' COMPUTERS_&_MATH   
   Number_Cruncher_Calculates_Whether_Whales_Are_Acting_Weirdly   
   AI_Tests_Into_Top_1%_for_Original_Creative_Thinking   
   Growing_Bio-Inspired_Polymer_Brains_for_Artificial_Neural_Networks Story   
   Source: Materials provided by Helmholtz-Zentrum_Dresden-Rossendorf. Note:   
   Content may be edited for style and length.   
      
      
   ==========================================================================   
   Journal Reference:   
      1. Lenz Fiedler, Normand A. Modine, Steve Schmerler, Dayton J. Vogel,   
         Gabriel A. Popoola, Aidan P. Thompson, Sivasankaran Rajamanickam,   
         Attila Cangi. Predicting electronic structures at any length scale   
         with machine learning. npj Computational Materials, 2023; 9 (1)   
         DOI: 10.1038/s41524- 023-01070-z   
   ==========================================================================   
      
   Link to news story:   
   https://www.sciencedaily.com/releases/2023/07/230707111625.htm   
      
   --- up 1 year, 18 weeks, 4 days, 10 hours, 50 minutes   
    * Origin: -=> Castle Rock BBS <=- Now Husky HPT Powered! (1:317/3)   
   SEEN-BY: 15/0 106/201 114/705 123/120 153/7715 218/700 226/30 227/114   
   SEEN-BY: 229/110 112 113 307 317 400 426 428 470 664 700 291/111 292/854   
   SEEN-BY: 298/25 305/3 317/3 320/219 396/45 5075/35   
   PATH: 317/3 229/426