Interview with Bill Goddard
Few scientists would disagree that theory should play a more central role in manufacturing. When William Goddard III says that within the next few decades quantum mechanics will drive industry, however, they may raise their eyebrows.
Quantum chemist Goddard, Ferkel Professor of Chemistry and Applied Physics at California Institute of Technology in Pasadena, CA, believes that faster computers and programs coupled with advances in atomistic simulations -- based in quantum mechanics and molecular dynamics -- are positioning quantum mechanics to move to the heart of manufacturing. It is a hefty claim.
Quantum mechanics deals with molecules at the subatomic level -- calculating forces governing the behavior of electrons around a nucleus. These calculations are so massive that accurate simulations based on quantum mechanics are usually limited to 10 to 30 atoms. Given that engineering and technology deal with large systems -- trillions upon trillions of atoms -- applying quantum mechanics to industrial problems seems to many like driving a car by peering through a microscope. The scales are wrong.
Hierarchical strategy for coupling atomistc simulations to engineering
design.
Using hierarchical modeling
Goddard and other scientists who practice the "hierarchical" approach to
modeling systems believe it is possible to link these scales. Hierarchical
modeling progresses from the finest, most detailed level of understanding about
a system -- the level of quantum mechanics -- to coarser representations by
approximating interactions so that larger systems can be modeled for longer
times. It broadens the scales by simulating bigger systems with fewer, faster
calculations. Goddard's hierarchy has five levels (see chart above).
Although other researchers may have more steps, all culminate with engineering
processes. "It is a telescoping whereby you can transcend from quantum
mechanics to engineering with just a few levels of simulation," said Goddard.
Telescoping from one level of the hierarchy to the next is not easy. Each attempt at simplification increases the chances for inaccuracy in a simulation. Moving quantum mechanics simulations to those based on molecular dynamics demands that researchers establish boundary conditions and approximate the multitude of interactions in a system in ways that will not cause distortions. Often a methodology that works for one process may not work for another.
Because it is so difficult to simplify forces in order to broaden scales, Goddard has been moving back and forth between the lower levels of his hierarchy -- quantum mechanics and molecular dynamics -- for the past 30 years. He has focused on all five steps of his hierarchy in the last 6 years. Recently he and his team at Caltech's Materials and Process Simulation Center enjoyed a breakthrough that may propel them a step higher. His team developed several hierarchical modeling methodologies that increased the size, length, and time scales for molecular systems at least 1,000 times. As part of a study to understand how drugs help prevent the common cold, they simulated all 480,000 atoms in rhino virus 14, the cold virus. Now they are poised to expand to larger systems.
Last August Goddard and his colleagues ported their codes for simulating the virus, plus several others, to NCSA to optimize them on the SGI POWER CHALLENGEarray. The system's innate scalability attracted the researchers as much as did its long-term viability. Developing modeling methodologies is only half of the group's work. The other half is optimizing methodologies for high-performance computers they think researchers will use in the future. A look at how they are using their codes in the virus simulation provides insight into how they attack problems via the hierarchical approach to modeling.
What they are optimizing
The codes used in simulating the rhino virus 14 are two of many codes Goddard
is now optimizing with NCSAÕs resources. Most of Goddard's work focuses on
applications in materials science, often funded by industry; however, the virus
work is Grand Challenge research funded by NSF. Although applicable to
molecular biology, the program and its two underlying codes are being adapted
for other uses, such as in industrial polymers and electronic materials.
The MPsim program, or Massively Parallel Simulator, was developed specifically by Goddard's group to simulate systems with greater than a million atoms. MPsim performs molecular dynamics calculations using two breakthroughs developed at Caltech: the Cell Multipole Method (CMM) for fast calculations of nonbonded interactions and the Newton-Euler Inverse Mass Operator Method (NEIMO) for calculating the response of a system to nonbonded forces.
CMM overcomes a bottleneck in large systems that arises from having to sum long-range Coulomb interactions. Coulomb forces are the electric charges between particles. Unlike other nonbonded interactions (van der Waals forces and hydrogen bonds), they do not fall off rapidly with distance. The electrostatic interaction between every particle must be calculated for all pairs of atoms. For a million-atom system, the number of calculations for Coulomb forces reaches into the trillions. Goddard's group adapted fast multipole strategies (originally done by Rokhlin and Greengard at Yale University) to large periodic systems using a hierarchy of meshes to reduce the number of interactions to be calculated from trillions to millions.
Ribbon display of the asymmetric structural unit (center) made up of four
individual viral proteins (corners).
Testing the methodology
The rhino virus simulation presented an ideal test for CMM. Scientists know
that the virus transmits its RNA to healthy cells by binding to receptors on
the surface of the target cell. Once bound it releases its RNA through
microscopic holes at the receptor sites. Scientists have speculated that drugs
lessen or prevent colds by stiffening the walls of the receptor sites so that
they cannot expand to accommodate the much larger viral RNA as it attempts to
slither into the cell. Because stiffness is determined by the interactions of
all the atoms in a protein coat, it was essential to simulate every atom in
order to confirm the action of drugs.
To look at collections of viruses or other proteins, Goddard's group will need to enlarge their simulation by coarse graining the virus structure. Coarse graining refers to looking at collections of atoms rather than individual atoms. Rather than simulating all 516,000 atoms of the protein coat, the group will employ the NEIMO tool to represent only the virus's 240 globular proteins as independent entities.
NEIMO was first developed by Abhinandan Jain at NASA and Caltech's Jet Propulsion Laboratory for guiding robotic arms in space satellites. It is a means of simplifying motion by restricting the degrees of freedom within a system. (Jain developed a way to calculate the relationship of the matrices between movement and torque -- the fundamental physics behind these ordinary gestures -- that scales linearly rather than to the third power.) Nagarajan Vaidehi of Goddard's group adapted this methodology to proteins so that the bulk of a protein can be treated as rigid and only some of the surface elements as flexible. In their simulation of the rhino virus, the team will ignore the small changes at the bonds and angles within its 240 proteins and calculate only the degrees of freedom at those bonds that work like door hinges.
Filling in the Gaps
Simultaneous with these developments, Goddard's group has been backtracking to
develop more accurate force fields, which will be required for drug design. The
force fields available to researchers in this area are nearly a decade old.
While adequate for studying drug effects, they may not provide the detailed
information critical for predicting molecular interactions. Since Goddard's
group hopes to move further into drug design, they are collaborating with
Richard Friesner of Columbia University to develop a new quantum mechanics code.
The new code would enable them to include solvents while handling very large
molecules.
This process is not a concern to Goddard who knows that climbing the modeling hierarchy often requires taking a step back. What is important is keeping one's eyes on the target, he said. "The whole point," said Goddard, "is that theory can solve real problems. In the past there have been gaps between fundamental theory and manufacturing, but, gradually, we are filling in those gaps."
Space-filled model of a pentagonal section of the rhino virus surface made
up of five asymmetric structural units.
Return to the Table of Contents.