josh@cs.rutgers.edu (04/20/91)
+---------------------------------------------------------------------+ | The following material is reprinted *with permission* from the | | Foresight Update No 11, 4/15/91. | | Copyright (c) 1991 The Foresight Institute. All rights reserved. | +---------------------------------------------------------------------+ Computational Nanotechnology at PARC by Ralph Merkle Research in nanotechnology continues to grow: the latest indicator is the recent interest in computational nanotechnology here at the Xerox Palo Alto Research Center. In December we bought a Silicon Graphics 4D/35 workstation (6 megaflops) and the Polygraf molecular modeling software from BioDesign. This lets us model chemically stable structures with as many as 20,000 atoms, including proposed bearings, mechanical molecular logic elements, molecular structural elements, etc. In the future we expect to get software that will model transition states and reactive structures. Such quantum-mechanical techniques are far more computationally intensive, restricting analysis to ten or twenty atoms, but providing greater accuracy. What does all this mean? There is an accelerating trend towards modeling new designs and new concepts on the computer before building them. GM has found that "computational car crashes" on a CRAY are cheaper, more flexible, and provide more information than real car crashes. Pharmaceutical companies are investing heavily in molecular modeling to investigate new drugs for similar reasons. Xerox, at several different sites within the company and for diverse reasons, is also pursuing this trend by modeling a range of chemical systems. Seen against this backdrop, work in computational nanotechnology (at PARC or anywhere else) is simply a continuation of the trend: before you build a car, a copier, or an assembler, you should first model it on a computer. This lets you review more designs more quickly and more cheaply before actually building (expensive) physical systems; it reduces the lag time from product conception to product delivery; and it improves the quality of the final product. While itUs not entirely clear how long it will be until we achieve a flexible molecular manufacturing capability, it *is* clear that we will get there more quickly and with fewer false starts if we model the components of such a system on a computer before actually building them. Computational Chemistry Oversimplifying somewhat, there are two classes of molecular modeling software: molecular mechanics systems and quantum mechanical systems. Molecular mechanics usually treats the nuclei of atoms as classical Newtonian point masses moving in a potential energy function (or conservative force field) defined by the electron cloud around them. There is no attempt to determine where the electrons actually are, or even to worry about the electrons at all. Rather, the positions of the nuclei directly define the forces acting between them. As an example, consider two hydrogen atoms bonded together to form a molecule. As the nuclei move closer together, they repel each other. As they move farther apart, they attract each other. In equilibrium, the two nuclei will stay at a characteristic distance. While this repulsion and attraction is actually the result of a complex quantum mechanical interaction, it can be summarized simply by noting the attractive or repulsive force acting between the two hydrogen nuclei as a function of their distance. A complex quantum mechanical interaction can be accurately summarized by a simple graph. We donUt know the actual electron distribution that produced the forces acting on the two nuclei, and we donUt care. This is known more formally as the BornPOppenheimer approximation: the nuclei swim in a sea of electrons, but if all we are concerned about is the positions of the nuclei, then we don't actually care where the electrons are: all we really care about is the force field acting between the nuclei. The electrons disappear from the computation and from our thinking, and are replaced by the force field. The Polygraf software from BioDesign uses the BornPOppenheimer approximation to greatly simplify the problem of modeling the interactions between nuclei. By using structural data, heats of formation, and vibrational frequencies determined experimentally for many different compounds, it is possible to deduce a fairly accurate representation of the force field that must be acting between the nuclei. A carbon-carbon bond prefers to be a certain length, while two hydrogens bonded to a single carbon have a certain preferred angle between them. These and other similar interactions form the building blocks of the force field. Once this field is known, any structure can be modeled (with greater or lesser accuracy), whether or not it was already known experimentally. Empirically derived force fields have been available for many years. The better ones provide quite good results within the broad range of compounds they were designed to handle. By using this method, the geometry and interactions of chemically stable structures (rods within a matrix, a molecular bearing on a molecular shaft) can be modeled quite accurately. This method has the great strength that a direct solution of SchrdingerUs wave equation is not required. The empirically derived force field is used in its stead. It is this which allows modeling of structures with tens of thousands of atoms and more. Of course, because the force field is based on data derived from chemically stable structures it does not provide information about unstable structures or transition states. For this, it is usual to compute an approximate solution to Schrodinger's equation (including the electronic structure). This requires more computational effort, but allows analysis of chemically unstable species (e.g., free radicals) and transition states where bonds are in the process of being made or broken. Taken together, these two methods from computational chemistry can model the mechanical interactions of large structures with tens of thousands of atoms, and the chemical interactions of one or two dozen atoms when bonds are being made and broken. These are precisely the interactions that must be understood if we are to build complex structures with atomic precision. As we apply the methods of computational chemistry, a more detailed picture of molecular manufacturing will emerge: a picture that will shorten the path from todayUs limited abilities to the more general abilities of the future. Dr. Merkle's interests range from neurophysiology to computer security; he is a researcher at Xerox Palo Alto Research Center. +---------------------------------------------------------------------+ | Copyright (c) 1991 The Foresight Institute. All rights reserved. | | The Foresight Institute is a non-profit organization: Donations | | are tax-deductible in the United States as permitted by law. | | To receive the Update and Background publications in paper form, | | send a donation of twenty-five dollars or more to: | | The Foresight Institute, Department U | | P.O. Box 61058 | | Palo Alto, CA 94306 USA | +---------------------------------------------------------------------+