In the study of quantum properties of many-body interacting systems to ideal extremes may be represented by Density Functional Theory (DFT) on the one side and by Hubbard- or Heisenberg- like models on the other side, the former being a widely used framework for quantitative calculations in realistic materials with moderate correlations and the latter being a route to investigate the role of strong correlations at the price of focusing mainly on universal or qualitative features. A celebrated computational method in this second case is the so-called Density Matrix Renormalization Group (DMRG). During the training worskhop on application porting to the GRID the use case of a popular DFT parallel code - Quantum Espresso - was presented. There is evidence that for systems with tens of atoms and periodic boundary conditions or for problems in which the computational load can be splitted in an almost parallel way, such as the calculation of the entries of a vibrational dynamic matrix, the GRID approach proves to be successful. However, when the number of atoms reaches O(100) a massively parallel treatment is required and the overall performance of the GRID approach may be significantly affected by intercommunication latencies and priority policies. On the DMRG side, instead, the execution is typically serial but an advantageous use of the GRID may be the istance of many independent runs for different system's parameters. For example, using the DIRAC submission tool within portal.italiangrid.it we could collect thousands of runs to compute the ground-state entanglement entropy of a class of spin-1 quantum Hamiltonians with spatial anisotropy, whose phase diagram is still controversial in some parts.
Studiesof many-body quantum systems on the italian GRID
C Degli Esposti Boschi;
2014
Abstract
In the study of quantum properties of many-body interacting systems to ideal extremes may be represented by Density Functional Theory (DFT) on the one side and by Hubbard- or Heisenberg- like models on the other side, the former being a widely used framework for quantitative calculations in realistic materials with moderate correlations and the latter being a route to investigate the role of strong correlations at the price of focusing mainly on universal or qualitative features. A celebrated computational method in this second case is the so-called Density Matrix Renormalization Group (DMRG). During the training worskhop on application porting to the GRID the use case of a popular DFT parallel code - Quantum Espresso - was presented. There is evidence that for systems with tens of atoms and periodic boundary conditions or for problems in which the computational load can be splitted in an almost parallel way, such as the calculation of the entries of a vibrational dynamic matrix, the GRID approach proves to be successful. However, when the number of atoms reaches O(100) a massively parallel treatment is required and the overall performance of the GRID approach may be significantly affected by intercommunication latencies and priority policies. On the DMRG side, instead, the execution is typically serial but an advantageous use of the GRID may be the istance of many independent runs for different system's parameters. For example, using the DIRAC submission tool within portal.italiangrid.it we could collect thousands of runs to compute the ground-state entanglement entropy of a class of spin-1 quantum Hamiltonians with spatial anisotropy, whose phase diagram is still controversial in some parts.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.