Achievement
Preliminary performance of the Parallel Debye Summation Engine (PARADYSE), whose development is being led by Sudip K. Seal (CSMD), shows at least two to three orders of magnitude improvement in the turnaround times of end-to-end structural loop refinement analysis for SNS users. This speedup promises far greater work throughput and an unprecedented knowledge discovery capability for SNS users.
Significance and Impact
SNS users use DISCUS (http://discus.sourceforge.net/) as the de-facto software suite for structural loop refinement. To compute diffraction intensities, DISCUS uses conventional Debye summation algorithms which become computational prohibitive as the number of atoms in the diffraction sample increases. Even systems with modest sizes demand days to weeks for loop refinements analyses that are ultimately meant to reveal the atomic structure of the diffraction samples. Using PARADYSE to compute the Debye summations, performed thousands of times in iterative refinement methods, reduces such analyses times from days/weeks to minutes/hours. PARADYSE not only dramatically improves analysis turnaround times but also handles system of atoms that are far larger than has ever been possible. Together, these new capabilities are expected to enable neutron diffraction studies with unparalleled scope and complexity.
Research Details
- Scalable spatial data structures have been built to enable fast parallel atomic indexing and complex tree-based operations in support of faster parallel Debye summation methods.
- This computational engine, called PARADYSE, is being integrated with the loop refinement workflow in the DISCUS software suite for deployment to SNS users.
Overview
PARADYSE, a breakthrough capability for the analysis of neutron scattering data, is under active development by Sudip K. Seal (CSMD). This software engine is the main deliverable for the LDRD project “Taking N2 to N: Developing a Highly Efficient, Multiscale Modeling Framework for Hierarchical Materials” (Project ID: 8420, PI: Lilin Ne). PARADYSE supports fast multipole method-based Debye summation kernels, being developed by George Fann (CSMD). The overall speed of PARADYSE stems both from parallelization as well as from algorithmically faster numerical kernels. PARADYSE will not only reduce neutron scattering data analysis times dramatically, but will also enable neutron diffraction studies at unprecedented scale and scope.
Last Updated: May 28, 2020 - 4:04 pm