Car Parrinello

25 years ago two budding scientists, Roberto Car and Michele Parrinello, used their expert knowledge, coupled with enthusiasm and a healthy dose of naivety, to develop a groundbreaking method for computer simulation.

October 7, 2010 - By Simone Ulmer

Michele Parrinello: one of the fathers of the Car-Parrinello Method, which celebrated its 25th anniversary this year. (Photo: S. Ulmer / ETH Zurich)

During the 1980s, the intellectual environment on the Adriatic – which included the International Centre for Theoretical Physics (ICPT) and the Scuola Internazionale Superiore di Studi Avanzati (SISSA) in Trieste, for instance – was just the place to research freely and easily. This is where the scientific curiosity of the two professors, coupled with a healthy dose of naivety, came together 25 years ago. These circumstances were major factors in the development of the Car-Parrinello method named after them. The method fundamentally changed the approach to simulations and thus supported both theory and experimentation. Whether in materials science, biology or the geosciences, simulations are necessary when experiments are not feasible, either because they are too dangerous or it is impossible to recreate the necessary conditions.

Understanding material

The two physicists combined their respective specialisations in the method. During his time at SISSA, inspired by a sojourn in the USA, Parrinello traded his paper-and-pencil studies for the computer. With its help, he wanted to describe the material that surrounds us through the movement of the atoms, using classical molecular dynamics. For this to work, however, computers using particular codes and algorithms have to relentlessly solve quantum mechanical equations of varying complexity.

In the 1980s, the dynamics of the molecules was calculated empirically, even though it was theoretically possible to calculate in an approximate manner the actual quantum mechanical molecule conditions “ab initio”, i.e. without using any concrete measurements, by applying the so-called Born-Oppenheimer approximation (BO approximation). The BO approximation method, however, was time-consuming and needed a computer capacity that was not available at the time: in the 1980s the computing power of a “supercomputer” was a few gigaflops; nowadays, a conventional computer’s processor can manage up to 25 gigaflops (25 billion computer operations per second).

Actual image with “ab initio”

However, the only correct way to model the actual forces exerted on an atom is “ab initio”. This means that to illustrate how chemical compounds break up and new ones form, the structure of the electrons, their energy state, has to be calculated for every position of the ion with a quantum mechanical equation. Calculating the electron structure was Roberto Car’s specialist field.

“Sometimes, ignorance is bliss“, jokes Parrinello, who has been a professor at ETH Zurich since 2001. Car and Parrinello saw beyond the opinions of the experts, who dismissed their idea of calculating both the molecular dynamics and the electron structure in one go as impossible. That meant combining the BO approximation with so-called quantum mechanical Density Functional Theory (DFT), which determines the structure of the electrons. “We knew a bit about each other’s field, but nowhere near enough. Otherwise we wouldn’t have thought it possible to combine both in one approach”, recalls Parrinello.

Tricks of the trade

For Parrinello, the starting point was to describe the element silicon. Under certain temperature and pressure conditions, silicon displays the characteristics of both metals and non-metals, and is an important semi-conductor. As these changes brought on by chemical reactions could not be determined with the Born-Oppenheimer approximation and the computer capacities that were available at the time, the two scientists resorted to a trick: they expanded the so-called Lagrange function for describing a physical system so that not every position of the ion and the forces exerted had to be calculated one step at a time. Using the Car-Parrinello method, the forces have to be determined once at the beginning of the simulation, while later the ion propagates itself like a wave with its electrons. The electron follows the ion quasi-adiabatically, without an exchange with its environment.

And so the first quantum mechanical calculation of the molecular state – “ab initio” – was successful. The Car-Parrinello method has been developed and honed for different applications over the years and has produced a lot of important information in various research disciplines. For instance, the former ETH-Zurich professor Artem Oganov used the method during his time at ETH Zurich to show that the bottom 150 kilometres of the Earth’s mantle did not – as had always been assumed – consist of the mineral perovskite, but rather a modified, layered variety that Oganov christened post-perovskite. The properties of the new mineral were able to explain conclusively the seismic discontinuity of this field for the first time.

Over the years, the two scientists have won numerous prestigious awards for their development, including the Dirac Medal in 2009. According to the commemorative publication to mark Parrinello’s 60th birthday back in 2005, the article describing the method published 25 years ago in “Physical Review Letters” was one of the most frequently cited interdisciplinary papers of all time.

Method with a long-term effect

With the leaps and bounds that have been made in computer technology over the last decade, the method is no longer as essential as it was 20 years ago; however, it is still effective in research today. Alessandro Curioni, the head of the Computational Science Group at the IBM Research Lab in Rüschlikon, who uses the method and had a hand in developing it further together with Parrinello and other researchers, stresses the importance of the Car-Parrinello method for molecular dynamics. An important side effect was also to see how Car and Parrinello’s little “trick” solved a large problem, shaping the scientists’ mode of thinking and opening up new horizons for them. “In the long run, the many ideas the Car-Parrinello method has generated for new quantum mechanical approaches to solving a problem will probably be even more important than the method itself”, says Curioni.

Michele Parrinello:
The Sicilian-born physicist has been a professor of computational science at ETH Zurich since 1 July 2001 and was Director of the Swiss National Supercomputing Centre (CSCS) in Manno, Ticino, until March 2003. His scientific career began as a full professor at SISSA in Trieste, before switching to the IBM Research Lab in Rüschlikon and then becoming Director at the Max Planck Institute for Solid State Research in Stuttgart. He will hold a dual professorship at ETH Zurich and the Università della Svizzera italiana in Lugano from 2011.He has won numerous awards for his work, including the American Chemical Society Award in Theoretical Chemistry in 2001, the American Physical Society 1995 Rahman Award and the 1990 Hewlett-Packard Award from the European Physical Society. In 2009 he received the Dirac Medal. He is an external member of the Max Planck Institute for Solid State Research, a fellow of the American Physical Society and a member of the International Academy of Quantum Molecular Science and the Berlin-Brandenburg Academy of Science, he is also Fellow of the Royal Society, of the Accademia dei Lincei, and on the National Academy of Science.
Car R & Parrinello M: Unified Approach for Molecular Dynamics and Density-Functional Theory, Phys. Rev. Lett. 55, 2471–2474 (1985) DOI: 10.1103/PhysRevLett.55.2471