Writing grant proposals, reviewing grant proposals, writing papers, reviewing papers, organising symposia, attending symposia.... makes the days pass by. I can’t really complain seeing as I just came back from a week in Berlin and am now preparing for a month in Shanghai.

At the FHI-AIMS workshop in Berlin, I found that the Fritz-Haber-Institut really wasn’t what I had imagined. I had an image of German precision (glass, steel and digital clocks); instead, the place is like an old wealthy American suburb (grand elaborate mansions) except with real character. There are beautiful parks, streets named after physical chemists and an air of relaxation. At the meeting itself, progress in the FHI-AIMS code was quite impressive: electron transport, self-consistent GW, dynamic TDDFT, metadynamics are all either implemented or well on their way.

Publication update: We found that interstitial oxygen in Al2O3 exhibits a sensitive dependence on the spin and structural configuration: a singlet peroxide (dumbell) species is found to be the global minimum structure (in contrast to previous theoretical studies - it is easy to miss the ground-state if you don’t break the symmetry of the octahedral interstitial site). Chemical Physics Letters.

The visit was well timed, as some work on defects in quaternary solar cell materials (Cu2ZnSnS4) in collaboration with Fudan University was published in Applied Physics Letters this week: a succinct distillation of some complex defect chemistry.

My mind has a tendency to wander when I read novels, but it’s a nice way to try to wind down before sleep. This year I did relatively well with fiction, getting through The White Tiger, The Third Policeman and First Love and other Novellas. Back to non-fiction, last Christmas I received Cathedrals of Science as a gift, but it is quite weighty, so ended up staying in Dublin for nine months. I finally got around to reading it over the last few weeks, and it is quite wonderful: it provides simultaneous insights into the birth of physical chemistry and the personalities of the people behind the movement.

We (at least I) have become complacent about relying on the Nerst equation, computing an Arrhenius activation energy or using concepts of resonance, ignoring the history of the people behind them. I’m guessing that the case is worsened in organic chemistry where virtually everything is “named”. This book should be essential undergraduate reading, as it displays both the excitement of working in science as well as the humanity (fallibility) of scientists. There’s no shortage of egos, superegos and grudges that last decades. You also get some insights into the establishment of the Nobel prizes, and the politics involved makes you realize why Boris Yellnikov (Woody Allen’s charming quantum physicist) never won the prize.

On a related note, there is an equally interesting piece on Paul Dirac, in the latest issue of Physics Today.

After some head scratching trying to get VASP 5 installed on Hector (Cray XT3), I finally got a working binary. To be extended as issues arise:

Problem 1: VASP5 has a preference for the IFC compiler (once you use -heap-arrays) and the Cray machine doesn't support it.

Solution 1: Go with PGF90. On my machine "-fast -O4  -Mipa=fast,inline" runs quite well, although for some features (such as DFPT), you have to reduce the optimization level to O2 to avoid segmentation faults.

Problem 2: PGF90 (version 8.02) compiles the code, appears to run a full SCF, but dumps out with a segmentation fault before writing out the end result.

Solution 2: This problem has occurred before: http://cms.mpi.univie.ac.at/vasp-forum/forum_viewtopic.php?2.5588.10 ; the fix involves inserting half a dozen lines into the paw.F file to sort out some variable allocations that PGF90 doesn't appreciate. I have had no problems since. (Click for the updated paw.F file).

Problem 3: The general approach to obtain a band structure is to run an SCF calculation (on a uniform k-point grid) and then fix the potential for the band structure k-points to give a nice (and cheap) E vs k curve. With the addition of non-local exchange (hybrid DFT functionals), this doesn't work in VASP. Actually, better phrased, the calculation will run, but you will have a set of chaotic eigenvalues, as k-points are no longer independent.

Solution 3: This fix is not ideal, but it is possible to include the additional band structure k-points in the original SCF calculation, i.e. in the KPOINTS file, but with weighting zero. You pay the computational penalty of calculating the extra k-points explicitly, but they don't contribute to the potential or total energy.  It is possible to do it the traditional way in CASTEP, so hopefully this will be integrated into VASP in the near future.

Problem 4: While testing processor scaling, I noticed a strange energy drift which sometimes appears with hybrid functionals for large supercells. For a static cell SCF calculation, starting from a random wavefunction, you can converge to different total energies (> 0.1 eV per cell) at large numbers of processors (> 128 cores).

Solution 4: From looking at the total energy breakdown, the electrostatic and atomic energies are exact, but the drift arises from a combination of differences in the exchange-correlation and Hartree energies. Restarting the job, to re-converge the wavefunction appears to help. Just something to be aware of and to check, especially if you are just doing a single point energy reference, e.g. for a bulk supercell before you introduce a defect.