Constrained Monte Carlo Approach to Modeling Disordered Materials Parthapratim Biswas*, Raymond Atta-Fynn, and D.A.Drabold Department of Physics and Astronomy, Ohio University, Athens, OH 45701 We discuss the application of a constrained Monte Carlo (CMC) algorithm to generate models of amorphous silicon. The constrained Monte Carlo, which is also known as ‘reverse’ Monte Carlo in the literature, is essentially a technique for generating structural configurations of disordered materials based on experimental data such as structure factor or radial distribution of the atoms. Originally developed by McGreevy and Pusztai [1,2,3], the method has been widely used in recent years not only to model liquid and glassy materials but also to model crystalline systems. Unlike most of the traditional approaches, an appealing feature of CMC is that it can be used without the knowledge of atomic potentials but experimental data only. At its simplest, one starts with a suitable configuration which may or may not be completely random, and prescribes a set of rules for the evolution of the atomic positions. The atoms are displaced randomly using periodic boundary conditions until the input experimental data match the data obtained from the generated configuration. In addition to experimental data, one can also include further information in the form of a number of constraints describing the characteristic features of the structure to be generated. Mathematically, this is a constrained optimization problem and its success depends upon the ability of the algorithm to find a minimum cost configuration by exploring the multi-dimensional search space. Using a number of judicious constraints along with the static structure factor, we have recently demonstrated how one can generate useful new models of amorphous silicon [4]. Here we summarize the main results for two models of amorphous silicon obtained in our work. The first model consists of 500 atoms of silicon containing in a box of length 21.2 A while the second one is a 525-atom model of amorphous silicon with 5% hydrogen atoms. The second model is generated from the first by hydrogenation of the dangling bond defects and relaxing the configuration using a first principles density functional calculations. In figures 1a-1d, we have plotted the results of our constrained Monte Carlo simulation. The structure factor obtained from a 500-atom model of a-Si is plotted in Fig. 1a along with the experimental data reported in Ref. [5]. Here we have started with a completely random configuration and used the static structure factor of a-Si obtained in the experiment of Laaziri et al. as input data [5]. Apart from the static structure factor, we have also included information about the bonding geometry between the silicon atoms and the average coordination number as constraints in the cost function. The bond angle distribution in Fig. 1b also appears to be consistent with the models obtained from molecular dynamics simulation and other methods. The average bond angle for the 500-atom model is found to be 108.6 o with a root mean square deviation of 12.8 o . The number of bond angles below 75 o and above 150 o for this model are 5 and 4 respectively, indicating that the configuration obtained from this CMC approach is reasonably relaxed. In Figs. 1c-1d, we have plotted the electronic density of states (EDOS) of the two models using a first principles density functional Hamiltonian. Microsc Microanal 10(Suppl 2), 2004 Copyright 2004 Microscopy Society of America DOI: 10.1017/S1431927604884460 804