In the first-principle molecular dynamics dynamics and stable configurations of atoms are calculated through electronic state based on quantum mechanics. To perform computing of this kind, we developed dedicated computer code called STATE. STATE is a powerful computer program with high-speed parallel processing, particularly, for calculations of high-precision surface state, the strength of vibration spectra and magnetic-field. Using STATE, the decomposition reaction of formic acid on titanium oxide (Figure 1) was possible. And reaction heat and frequency and its strength of adsorbate in some other basic catalytic reactions were also calculated. These calculations accorded closely with actual experimental results. STATE is already used in joint research by a number of groups, and was released within AIST in July as part of TACPACK, an integrated software package developed by Tsukuba Advanced Computing Center
(TACC) and RICS.

Although STATE provides large-scale, high-precision computing power, computing time rises by a power of between 2 and 3 as the number of atoms increases, so computation becomes more difficult as the size of the system increases. To address this problem, the Institute developed, for the first time in the world, a unique order (N) method based (ABRED) on the recursion method, which reduced the computing load to a power of 1 to the number of atoms.

Although parallel programming of this method has not yet been conducted, it is now possible to compute electronic state of systems for a few hundred atoms on a single PC. The method has already been applied to problems in carbon nanotubes and manganese polynuclear complexes. One difference between this method and other order (N) approaches is that it can be applied to metals. The program has been optimized to enable calculations on up to the fourth row of atoms. After further parallelization of the code, the Institute plans to release it for general use.

To provide methods of electronic state computing, we are currently developing a density functional method using finite element bases (FEMTECK) as well as a fragment molecular orbital method (FMO) for large molecules such as proteins. All of these calculations are performed using AIST's Hitachi SR8000 supercomputer, which has been proven to offer highly efficient parallel processing using 512 CPUs. Within two or three years, we expect to be able to use the fragment molecular orbital method, on a large PC cluster, to calculate the structures of proteins composed of thousands of atoms.

The density functional method is extremely effective and widely used. Because it generally uses local density approximation, however, this method is unable to address some problems of electronic correlation. This shortcoming introduces many qualitative and quantitative problems into the results of calculation of such practically important features as luminescence characteristics, light-absorption, band gap and magnetic field. We are currently developing a theoretical method for these problems of electronic correlation. The required computing time is still too long as a practical computing technique, but we have already been able to reproduce some important experimental values using this approach. The method has proven an important tool in running computer simulations of the electronic and optical properties in the fields of strongly correlated electronic system, spin electronics and optoelectronics.

To create organic field-effect transistors (FETs) and molecular sensors, the molecules must first be lined up on the surface in a certain pattern. One method of doing this is to make the molecules line up automatically by self-organization. Such methods have been developed in areas of "wet chemistry" such as supramolecular chemistry. It is, however, difficult to know what structures are obtained from what molecular structures. In this case, molecular dynamics is applied to predict structures and functions of the resulting aggregations. To apply classical molecular dynamic to structural prediction, however, several problems must first be solved; sufficiently accurate intermolecular forces must be obtained. Efficient sampling methods are required. And the accurate time-integration method for a long-time calculation must be developed. The Institute was able to use these methods to develop high-speed, high-precision code. Figure 2 illustrates an example in which this code was applied: a description of the dynamics of different molecule transportation in lipid bilayers. Using these technologies for the molecular assemblies, we are collaborating with experiment groups with the aim of developing systems for the design of nanostructures.

If we apply the term “nanoscale” a little large system where atomic and molecular interactions are replaced by a mean field, we can use the method of calculating continuum media. Examples include the phase field method and vertex model, which are used in computing the organization in metals and ceramics. If we “zoom out” a little further, in simulations of microscopic electronic machines (MEMs) and the machines to construct nanostructures, simulations of continua in solid and fluid mechanics become most useful. In these microscopic domains, however, computing requirements are usually too rigorous to be handled by commercially available software. For example, in the super-inkjet now being developed by the Nanotechnology Research Institute, the problem of high-speed two-phase flow occurs when the ink is ejected from the nozzle. Our Institute have developed an alternative solution; A shown in Figure 3, the simulation impact spot of the ink droplets closely matches the actual experimental results.