Short history of GPAW and latest news about PAW setups
Short history of GPAW and latest news about PAW setups
Short history of GPAW and latest news about PAW setups
- No tags were found...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
More trivia<strong>G<strong>PAW</strong></strong> used to be called Grid<strong>PAW</strong>The code started its life in CVS, then moved to SVN at berlios.de<strong>and</strong> then finally to our own SVN serverThe compiled part <strong>of</strong> <strong>G<strong>PAW</strong></strong> was written in C++ in the beginning<strong>G<strong>PAW</strong></strong>’s web-pages used to be in a MoinMoin wikiWe’ve used both Numeric, numarray <strong>and</strong> numpy as our Pythonarray package
More numberscpu hours <strong>of</strong> tests: 4150number <strong>of</strong> committers: 43gpaw-users subscribers: 278gpaw-developers subscribers: 135gpaw-svncheckins subscribers: 22
Homepage from 2004
Download page from 2004
Testing <strong>G<strong>PAW</strong></strong>’s <strong>PAW</strong> <strong>setups</strong>Challenges:We need good benchmark results we can trust.Getting correct lattice constants or bond-lengths does notguarantee that the setup is good.Getting correct cohesive energy or atomization energy is a bettertest, but this involves atoms: not fun!Formation energies for bulk oxides would be a good test, butsome <strong>of</strong> the oxide structures are quite complicated <strong>and</strong> containmany atoms.FHI-aims:ELK:All-electron full-potential density functional theory codeusing a numeric local orbital basis set.https://aimsclub.fhi-berlin.mpg.deAll-electron full-potential linearised augmented-planewave (FP-LAPW) code.http://elk.sourceforge.net/
StrategyOptimize volume for fcc <strong>and</strong> rock-salt (oxygen + X) for allelements (from hydrogen to nobelium) using AIMS <strong>and</strong> ELK.Do non-relativistic calculations.Compress fcc <strong>and</strong> rock-salt structures to 90 % <strong>of</strong> equilibriumlattice constants.Use oxide formation energies <strong>and</strong> fcc <strong>and</strong> rock-salt compressionenergies as benchmark numbers.Simple tests make iterations faster.NoteThis kind <strong>of</strong> work is extremely boring
AIMS reference energies
Oxide formation energies relative to AIMS
Ruthenium exampleAdsorption energies in eV: Ru(001) + 1 2 X 2 - X/Ru(001)0.8 0.9 0.10 Gajdošet al.O/Ru(001) 2.51 2.78 2.78 2.67N/Ru(001) 0.46 0.88 0.93 0.94H/Ru(001) 0.54 0.57 0.59NO - 1 2 N 2 - 1 2 O 2 0.96 0.95 0.96 0.95M. Gajdoš, J. Hafner <strong>and</strong> A. Eichler, J.Phys.: Condens. Matter 18(2006) 41-54
FCC compression energies relative to AIMS
Rock-salt compression energies relative to AIMS
Iron example8e 16e 16e+ FHI-aims ELKoxide formation 3.58 3.56 3.59 3.60 3.60fcc compression 1.35 1.32 1.26 1.27rock-salt compression 2.25 2.18 2.04 2.04
Logarithmic derivatives at r=2.3 Bohr (8e <strong>and</strong> 16e+)
Plane-wave ModeAdvantages:Fast for not too large systemsFast convergence with respect to number <strong>of</strong> plane-wavesSmaller memory footprintSimpler density mixing metric <strong>and</strong> preconditioningNo egg-box errorSimple implementation <strong>of</strong> stress tensorPoisson equation is solved in reciprocal space. Can also be used inlcao <strong>and</strong> fd mode:calc = <strong>G<strong>PAW</strong></strong>(..., realspace=False, ...)
The futureImportant work:New <strong>PAW</strong> <strong>setups</strong>Robust eigensolver/density mixer<strong>G<strong>PAW</strong></strong>-1.0?Also important work, but not as important as the above:Better basis sets for lcaoParallelize plane-wave calculations over plane-wavesGeneralized multigrid solvers that can h<strong>and</strong>le grid sizes <strong>of</strong> theform: 2 a 3 b 5 c 7 d
Thursday (May 23)Activities for <strong>G<strong>PAW</strong></strong> developers (we start at 9:00):Coordination <strong>of</strong> code development <strong>and</strong> discussions <strong>about</strong> thefuture: Quick tour <strong>of</strong> ongoing projects - what’s the current status?Introduction to Sphinx <strong>and</strong> reStructuredTextIntroduction to testing <strong>of</strong> <strong>G<strong>PAW</strong></strong>H<strong>and</strong>s on: Write new documentation/tutorials <strong>and</strong> how to makesure they stay up to dateLunchStatus <strong>of</strong> unmerged branches: rpa-gpu-expt, cuda, lcaotddft,lrtddft_indexed, aep1, libxc1.2.0Questions open for discussion:When do we drop support for Python 2.4 <strong>and</strong> 2.5?Strategy for porting <strong>G<strong>PAW</strong></strong> to Python 3?Switch from SVN to Bazaar <strong>and</strong> Launchpad?H<strong>and</strong>s on: Write new documentation/tutorials --- continuedPresentations <strong>of</strong> today’s work (we stop at 15:00)
Finally ...Thank you for your attention,<strong>and</strong> don’t forget to give me your pdf file from your talk