"Stairway to heaven": Piz Daint validates method
Initial simulations on the supercomputer “Piz Daint” have validated a method that aims to solve the key problems in simulating chemical reactions involving liquid water. Two professors from ETH and the University of Zurich have developed the method over the last few years.
March 18, 2014 - by Simone Ulmer
More than 70 per cent of the Earth’s surface is covered with water. Water is essential to life and to many important physicochemical processes in living organisms as well as in and on the Earth itself. Although the element of water is omnipresent and of vital importance, there is little agreement on the structure of the hydrogen bonding network in liquid water. This, however, is of key importance, for instance for the development of drugs or in materials research. A new method, which has been successfully tested on the CSCS supercomputer “Piz Daint”, could soon provide a solution.
Water, the mystical substance
Water is the only chemical compound on earth that occurs naturally in a solid, liquid and gaseous state. “We are familiar with the crystal structure of ice and know what the hydrogen bonds look like, but there are doubts about whether they are exactly the same in liquid water”, says Joost VandeVondele, Professor of Nanoscale Simulations at ETH Zurich. These doubts are based on the unusual properties of water, which can flow as a solid (i.e. ice) and reaches its highest density at just under 4 degrees Celsius before it freezes – ice floats as its density is lower than that of liquid water, a property that fundamentally shapes our planet’s climate and ability to support life.
The key role of water is the reason why researchers like VandeVondele would like to simulate water in a realistic manner. The scientists want to understand the structural properties of water including its electronic structure. As soon as water becomes active in a chemical reaction, the electronic properties and the forces at play in the water molecule have to be taken into account. This ab initio calculation for approximately determining real quantum mechanical molecular states with the help of supercomputers has been possible for about 30 years. “Simulating liquid water ab initio was a real breakthrough”, says VandeVondele. But scientists quickly realised that the specific hydrogen bonds cannot be simulated so easily. A sobering experience for VandeVondele was when he calculated that a litre of water weighed 800 instead of 1000 grams using Density Function Theory (DFT), which is normally applied for these purposes. “A serious error that worried us greatly”, says VandeVondele, “as many of our simulations are based on the quality of these models”.
From solid to “heavenly” precision
Science has come up with empirical tricks to correct the error, says VandeVondele, but this is not very satisfactory. Thus, for several years now, the scientists have been working on a suitable theory which is all-encompassing, so to speak. They speak of a Jacob’s ladder, which they would like to climb with their molecular simulation in order to move from “solid to heavenly precision”. One of the key problems of simulation is taking into account the van der Waals’ forces. These are only weak forces that act between the atoms or molecules, but they seem to be essential for molecular simulation.
The method that helps a gecko stick to the ceiling
With the new simulations, they have been able to climb another rung on the ladder, says VandeVondele. The research group of Jürg Hutter, a professor at the University of Zurich, and VandeVondele’s group created the foundations for this by implementing a method called MP2 in the CP2K code they used. According to the researchers, MP2 is able to describe very precisely the interactions between molecules. This is because MP2, in contrast to the standard DFT, can take into account virtual orbitals in the system and therefore make the simulations that much more precise, says VandeVondele. Figuratively speaking, “if a gecko’s adhesive properties are described with DFT, it falls off the ceiling; with MP2 it will stick.”
MP2 is based on the Møller-Plesset Perturbation Theory (MP) and was published in 1934 by the scientists of the same name. With the new method, MP2 implemented in CP2K, the researchers have recently succeeded in correctly determining the density of water without resorting to any “tricks”1. However, the calculations of MP2 are extremely complex and require considerable computing power – around a thousand times more than the standard DFT. According to VandeVondele, only the efficient graphics processors of "Piz Daint" will make calculations of this kind routinely possible.
Whether MP2 can also be applied to other liquids and solid matter was tested by the researchers shortly after the supercomputer was upgraded with graphics processors. On supercomputer “Monte Rosa” the tests would have taken about a year, while on “Piz Daint” they took just a few weeks. VandeVondele stresses that the benchmark calculations for determining the quality of the model are good and that they now have a kind of gold standard to validate DFT results in cases of doubt.
Monitoring molecular motion development over time
“The new process has recently enabled us to monitor molecular dynamics and to determine the evolution of systems over time”, says VandeVondele. This is important, as many experimental methods measure dynamic properties. For instance, infrared spectroscopy examines time correlations in molecular vibrations. “Determining this with greater precision is the new kind of simulation we are now carrying out on ‘Piz Daint’”.
The researcher stresses that the procedure is a fundamental approach with a high-quality theory. After water, the scientists would thus like to study several other systems on the same theoretical level of accuracy, since it is now possible to simulate chemical reactions in solutions and make predictions about crystal structures.
|1 Del Ben M, Schönherr M, Hutter J & VandeVondele J: Bulk Liquid Water at Ambient Temperature and Pressure from MP2: J. Phys. Chem. Lett. 2013, 4, 3753-3759. Dx.doi.org/10.1021/jz401931f|