Abstract
Pushing all-electron DFT past some old limits: Large-scale surface structure, molecules, and some serial predictions
Volker Blum
Duke University
Predicting and understanding structure and materials properties from first principles benefits heavily from the reach and relative computational ease of density fuctional methods today. For few-atom structures, serial structure screening, exhaustive structure predictions and meaningful time scales in molecular dynamics are now routinely possible; the ongoing challenges are at least two-fold:
(i) the affordability and accuracy of the physical approximations [in density functional theory (DFT), primarily the exchange-correlation functional], (ii) the affordability and accuracy of all other numerical approximations that lead to a total energy for a given structure. This talk covers some of our efforts to push the reach of first-principles methods past these limits. On the implementation side, the work is based on the numeric atom-centered orbital code package FHI-aims [1], which implements all-electron DFT and perturbative methods "beyond". The code is currently aimed at system sizes from a few up to a few thousand of atoms, handled efficiently in parallel from few-core computers up to massively parallel architectures. In particular, a goal has been to create a robust, conventional (matrix based) eigensolver framework [2] that does not become a botteneck for the goals mentioned above. In a nutshell, we are thus able to address structure and dynamics starting from small systems (e.g., few atom peptide molecules) but in large numbers (exploring conformational space) all the way up to large-scale nanostructured systems. For example, well-defined large-scale surface reconstructions can be effectively prepared and have a long history of use in experiment; in particular, we address Au(100 and Pt(100) [3]. We then show how biomolecular structure and dynamics can be accurately predicted for individual, large peptide molecules, here Polyalanine in direct comparison to gas-phase experiments [4,5]. Finally, we explore the potential of a first-principles "serial conformer screening" for systems which are not conclusively covered by faster coarse-grained models (force-fields): the case of ion-peptide interactions.
[1] V. Blum, R. Gehrke, F. Hanke, P. Havu, V. Havu, X. Ren, K. Reuter and M. Scheffler, Comput. Phys. Commun. 2009, 180, 2175-2196; http://www.fhi-berlin.mpg.de/aims/ .
[2] "Eigensolvers for Petaflop Applications (ELPA)". http://elpa.rzg.mpg.de .
[3] P. Havu, V. Blum, V. Havu, P. Rinke and M. Scheffler, Phys. Rev. B 2010, 82, 161418(R).
[4] A. Tkatchenko, M. Rossi, V. Blum, J. Ireta and M. Scheffler, Phys. Rev. Lett. 2011, 106, 118102.
[5] M. Rossi, V. Blum, P. Kupser, G. von Helden, F. Bierau, K. Pagel, G. Meijer and M. Scheffler, J. Phys. Chem. Lett. 2010, 1, 3465-3470.
(i) the affordability and accuracy of the physical approximations [in density functional theory (DFT), primarily the exchange-correlation functional], (ii) the affordability and accuracy of all other numerical approximations that lead to a total energy for a given structure. This talk covers some of our efforts to push the reach of first-principles methods past these limits. On the implementation side, the work is based on the numeric atom-centered orbital code package FHI-aims [1], which implements all-electron DFT and perturbative methods "beyond". The code is currently aimed at system sizes from a few up to a few thousand of atoms, handled efficiently in parallel from few-core computers up to massively parallel architectures. In particular, a goal has been to create a robust, conventional (matrix based) eigensolver framework [2] that does not become a botteneck for the goals mentioned above. In a nutshell, we are thus able to address structure and dynamics starting from small systems (e.g., few atom peptide molecules) but in large numbers (exploring conformational space) all the way up to large-scale nanostructured systems. For example, well-defined large-scale surface reconstructions can be effectively prepared and have a long history of use in experiment; in particular, we address Au(100 and Pt(100) [3]. We then show how biomolecular structure and dynamics can be accurately predicted for individual, large peptide molecules, here Polyalanine in direct comparison to gas-phase experiments [4,5]. Finally, we explore the potential of a first-principles "serial conformer screening" for systems which are not conclusively covered by faster coarse-grained models (force-fields): the case of ion-peptide interactions.
[1] V. Blum, R. Gehrke, F. Hanke, P. Havu, V. Havu, X. Ren, K. Reuter and M. Scheffler, Comput. Phys. Commun. 2009, 180, 2175-2196; http://www.fhi-berlin.mpg.de/aims/ .
[2] "Eigensolvers for Petaflop Applications (ELPA)". http://elpa.rzg.mpg.de .
[3] P. Havu, V. Blum, V. Havu, P. Rinke and M. Scheffler, Phys. Rev. B 2010, 82, 161418(R).
[4] A. Tkatchenko, M. Rossi, V. Blum, J. Ireta and M. Scheffler, Phys. Rev. Lett. 2011, 106, 118102.
[5] M. Rossi, V. Blum, P. Kupser, G. von Helden, F. Bierau, K. Pagel, G. Meijer and M. Scheffler, J. Phys. Chem. Lett. 2010, 1, 3465-3470.
No video available