|
The
DV Story - Past, Present, and Future?
Disclaimer The following account is highly subjective and subject to revision. Comments and additions/corrections are welcome. Send to don-ellis@northwestern.edu
Prehistory In the late 50's and early 60's, John Slater's Solid State and Molecular Theory Group at MIT was a crossroads for chemists, physicists and computer types interested in developing methods for using the new computer technology to solve the big problems of quantum mechanics. Undergrads, grad students, postdocs and visitors mixed freely and interacted under the sometimes critical supervision and commentary of the boss. Among the students, Jules Moskowitz, John Wood, and Al Switendick had exceptional influence, because of their knowledge, ideas, and willingness to share. Visitors like Brian Sutcliffe, Jean-Paul Desclaux and Per-Olov Löwdin brought a European cultural and scientific perspective; we students learned to appreciate drinking shandy and non-California wines, among other adventures. Entering this ambient as an undergradaute, I was hired by Michael Barnett, a visionary quantum chemist out of the Coulson school, to work on the molecular integrals project. Despite my attempt to escape to Urbana, I remained at MIT for graduate school, held by Slater's vision of the future. Graduate school was a lot more fun than the undergraduate grind, and the vacuum tube computers began to give way to transistorized machines. No more late nights sitting next to the main frame, listening to the matrix-diagonalization routine on an FM radio. Trips up to Bill Lipscomb's lab at Harvard, where the wall clock ran backwards, and serious chemistry was being done by Bill Palke and Russ Pitzer caused some thoughts. The gap between real world problems of chemistry and what we could project in the near future with our Hartree-Fock and HF-CI schemes was too large. Slater's ideas about density functional theory, following up on work he and Dirac had done long ago was thought by all of us to be hopelessly approximate, relative to the demands of chemistry. Nevertheless, some of the guys began to get surprisingly good results on solid-state band structures, so maybe DF wasn't totally trash. Meanwhile Ben Woznick showed me his experiments on using Monte Carlo sampling for molecular integrals, and Isaiah Shavitt gave a talk about pseuodorandom and elegant multidimensional integration schemes. Maybe I could escape from the algebraic drudgery, recursion formulas, and infinite sums of my daily work and nightmares? The thesis got finished, job hunts were conducted, visits were made, and once more John Slater took a hand in matters. He had moved to the of Florida at Gainesville upon reaching the mandatory retirement age at MIT. There he built up the Quantum Theory Project with Per-Olov, again mixing together chemists and physicists indiscriminantly. Yngve Ohrn and other fugitives from snowy Scandinavia imposed a distinctly non-Floridian atmosphere in Gatorland. Seduced away from the industrial salary, or the heady intellectual atmosphere of Princeton, here I was, drinking beer in Hogtown with the likes of Frank Smith and Keith Johnson! Slater had suggested to them that using the Multiple Scattering technique to implement DF theory for molecules could be interesting. Indeed it was - but there was that awful Muffin Tin approximation. I wondered whether the pseudorandom integration schemes could be used to get around that problem, so Gayle Painter and I decided to write a band code. In those days metallic lithium and semimetallic graphite were considered significant challenges, so we were really pleased when the band structures came out well. DV-Xa was born.
Moving On After two years at Gainesville, the provincial legislature was revealing its true agricultural roots, and busily reneging on promises for support of research at the universities. Times were looking bleak, and an old collaborator from MIT, Art Freeman, appeared with an interesting proposition. It seemed that Northwestern U. was constructing a kind of paradise for physics in the suburbs of Chicago, and there were brave new worlds of research opening up there. Besides offering splendid research facilities, there was the Symphony, and winters weren't that bad. The bait was taken, the job accepted, and I moved just in time to witness the great Chicago riots at the Democratic Convention that year. Then it rained for three months. While following up on the band approach, I decided to see if we could do something with molecules. The prognosis was not very good, since everyone 'knew' that DF simply wasn't accurate enough to answer the kind of questions traditional quantum chemists had decreed as central. But maybe there were other questions, of lesser precision, that could be resolved. Anyway, CI and many-body perturbation theory was looking more and more tedious from the computational point of view. Thangam Parameswaran and I found that molecules like TiCl4 were reasonably well described with our first DV-Xa molecular code, using a simple analytical basis set. Since this was quite beyond the capabilities of the HF/CI codes of the day, this was pretty exciting. A Fulbright grant took me to the Netherlands, where an old MIT friend Piet Ros, and his student Evert-Jan Baerends decided to play with the scheme. Dutch thoroughness, acute chemical understanding, and hard work resulted in a stable procedure and some very interesting results on a series of small molecules. The methodology now appeared not only viable, but worthy of further development. The Amsterdam 'node' went on from there, in a series of independent developments worthy of another story. Perhaps it will be told elsewhere. Back at NU, a Sloan grant gave me a bit of freedom to work on the DV project, and several visitors made key contributions. Jim Waber was Prof. of Materials Science, and his lab was host to several scientists of great skill. It was our good luck to have Frank Averill, Hirohiko Adachi, and Arne Rosen in the same place at the same time. I had learned how to do numerical atomic calculations on a stay at Rudy Mössbauer's lab in Munich (thanks, Gopal), and Frank, Hiro, and Arne were also well acquainted with numerical methods. So we decided to make a Great Leap Forward, introducing numerically optimized basis sets, and to implement the Dirac equation to get full relativistic DF wavefunctions. Since this had been done years previously in the band-structure realm, we knew it could be accomplished. Arne also understood relativistic double groups, so we made a big step, and it all worked really well! All too soon, everyone had to return home. It had the good consequence that new nodes began to grow: Hiro's in Japan, Arne's in Sweden, and in Germany, Burkhardt Fricke began some very interesting work on highly relativistic collisions requiring further technology developments. Cross-pollination began to occur, with people travelling back and forth between the different labs, and new people beginning to interact as the possibilities for chemistry and materials science became evident. Other researchers had followed up on the traditional Gaussian basis sets, now using the DF Hamiltonian, and we were on the road to respectability. There was even a rigorous theoretical basis, thanks to Lou Sham and Walter Kohn, for the success of DF and an outpouring of improved exchange and correlation potentials.
The NU Node Back at the ranch, Art Freeman had made some Swiss contacts and brought in Bernard Delley, who had a marvelous background in both experiment and rigorous mathematical methods. Bernard and I realized that the main limitations in the molecular DV approach were due to the precision of the Coulomb potential, not in the approximations of the formal DF potentials. Various efforts had been made to improve upon the least-squares analytic scheme developed at A'dam, and finally on the third (or fourth) try, Bernard's approach to localized numerical expansion functions made a real advance possible. The availability of supercomputers (thanks, Seymour) and emerging graphical techniques led us rapidly into treating more complex systems. The embedded cluster technique, which I had been developing with Greg Benesh, Ernie Byrom, and Vladimir Gubanov, and others could now be applied to magnetic impurities in metals and sizable metallic particles. Linear chain systems like polyacetylene soon came within our reach, and we began to think about 'real' materials with defects, surfaces, and interfaces. Gordon Goodman suggested that we try to make a program version which would be accessible to experimentalists - something just a little more 'user friendly' than our typical lab development toolkits. He volunteered to work with us, bringing a powerful knowledge of group theory and atomic and molecular spectroscopy - the target of our first real software effort. That package formed the basis for many collaborations with experimentalists at Argonne Natl Lab and elsewhere. The Brazilian node came into existence. Diana Guenzburger spent some time at NU, and realized that the DF-DV approach fit well with analysis needs of the Mössbauer group at Rio de Janeiro, as well as with her own chemical interests. Some special development efforts were required to obtain the precision near the nucleus needed for hyperfine interactions. This work was begun on another visit to A'dam, and completed at NU and Rio. We noticed that global delocalization of peoples, ideas, and programs had occurred, but that collaboration was still taking place. Chang-Xin Guo, Pei-Lin Cao, Li-Bin Lin and other senior scientists came, made their indelible mark, and started sending us their best students. Chinese friends and visitors established their own network, with principal nodes at Hangzhou, Chengdu, Hefei, and Beijing. The first commercial DV code was offered to the public, not by us, and at a high price. Academics debated the merits of selling software versus scientific collaboration as a way of doing 'business' or doing 'research.' The NU mode of proselytizing and training visitors, and giving away codes continued. The DF-Gaussian basis codes reached a high level of sophistication, with core potentials making applications to fairly large systems feasible. Was DV about to be superseded by the good old Gaussians? The work at NU was pulled more and more toward the challenges facing materials science, partly through the influence of the Materials Research Center and the collaborations it fostered. Oxide ceramics with their nonstoichiometry, intrinsic and critical defect structures, and messy interfaces changed from being ugly and unapproachable, to being interesting and attractive. Greg Olson was preaching the gospel of Quantum Steel, and we believed. Kobe sent us Yoshi Itsumi, to teach about the industrial approach to steel, and to learn what we could do to help with the next generation of structural metals. The old tools suddenly seemed very inadequate, both for the scale of the problems, and for the necessity of dealing with temperature, diffusion, and mechanical stresses.
Toward the Mesoscale After flirting with quasi-dynamical DF approaches, we realized that our problems could be solved with existing/projected computers most rapidly by combining classical and quantum methodologies. Electron microscopy and energy loss data could set the boundary conditions and give the broad outlines of interface/surface/composite, which classical dynamics and Monte Carlo approaches could refine down to the atomic scale. Then our DV tools, in the embedded cluster framework, could focus on the electronic structure in a series of windows or view-frames. With a little intuition and luck, the electronic results could be used to refine the classical interatomic potentials, thus closing the loop. Attempts to get other people to do our work for us generally fail; however, we first tried to get some dynamicists to do our simulations, leaving us to do the DF work. This idea turned out to be poor, because of the iterative nature of the process, the need for results NOW, and uncertainties in both classical and quantum models which are revealed as the research is done. So, a series of students (thanks Hai-Ping, Qing, Richard, and XianHong) learned to do MD and MC simulations, to generate reasonable potentials, and to interface that data to the DF schemes, then relating the data to experimental facts. Finally, it became clear that a real interface is necessary, so that DF can be a subroutine of MD/MC and vice versa. This is our principal scientific challenge, now: to produce a useful hybrid classical and quantum approach to complex materials problems extending to the mesoscale. Several recent visitors have made major contributions to this ongoing, only partially functioning, modality of research. Kleber Mundim has brought a knowledge of protein structure simulations and dynamics, coupled to a fundamental knowledge of the mathematics and computational machinery. Along with special food, music, and art, perhaps Bahia will soon advertise itself as a center of materials simulation research. Oliver Warschkow has brought a fresh approach to the parallel-execution/concurrent analysis part of the problem, which is entering an exciting phase. It has become clear that advanced graphics and visualization techniques are essential for rapid advancement in this research area. This doesn't mean merely (!) designing and executing nice GUI's for users, but also thinking deeply about how to project multidimensional data bases onto axes which will permit us to understand and make decisions about the next steps to take. Anyone want to get involved?
Last updated: September, 2001 |