Today and tomorrow, we're reporting on presentations at an important conference on Productive Nanosystems: Launching the Technology Roadmap. Chris Phoenix is providing live blog coverage for us...
Fifth talk: James Davenport, Director, Computational Science Center, Brookhaven National Laboratory.
We need computational tools which take account of atomistics. Atomic resolution is critical. We're not dealing with big enough systems! But massively parallel petaflop systems are becoming available.
A single 5-nm dot of material has about 5,000 atoms. A 7-nm cube has about 40,000 atoms. You can make cobalt single-atom lines on platinum with a shallowly stepped surface. Quantum dots have interesting optical properties, and changing the size in a small way changes the optical properties. Biological effects can depend on size. Proteins can have thousands of atoms: they're nanoscale systems. Note that to study a protein in simulation, you need to add water: tens of thousands of atoms.
A roadmap for simulation has to deal with mixed length and time scales. You will be dealing with petaflop systems, which the simulation community doesn't yet know how to deal with. Software needs to be interoperable. Data sharing needs standards. (This would also help with integrating experimental results.) Data storage and retrieval is a problem: large amounts of data are generate by both experiment and simulation.
Hierarchy of tools, smallest to largest scale:
- Quantum mechanical (ab-initio, DFT)
- Car-Parrinello or first principles molecular dynamics (MD)
- Force field MD (AMBER, CHARMM) [This is what's used in a lot of nanomachine simulations]
- Heisenberg magnets
- Nanoparticle-nanoparticle
- Continuum (elasticity, micromagnetics)
Quantum simulations using the Schrodinger equation are basically impossible for multi-atom simulations. But there are approximations (e.g. Hartree-Fock) that are not very expensive, but pretty accurate.
There's a lot of discussion about properties emerging from lower-level simulations, "You've probably all heard of the program Guassian," the ability to study magnetism but not the temperature dependence of magnetism... I'm not going to try to report on this.
Bulk gold is inert, but nanoscale gold is a useful catalyst. Relativistic effects are important since gold is heavy.
It's relatively rare that the quantum nature of atoms comes in. For example, for proteins, you use classical approximations. And in fact, you extrapolate parameters for the various different atoms.
For molecular dynamics, you want a time step of 10^-15 seconds. Protein folding takes milliseconds. The fastest folding protein that we know of takes a microsecond. That's a billion time steps! Until recently, we could do about 1/1000 of that. With Blue Gene/L, a few weeks of computer time might get you a microsecond: 10,000 processors, 10 particles per processor.
A recent simulation, of carbon nanotubes growing on iron, was done with forces computed on the fly from Hellmann-Feynman. (So the forces didn't have to be estimated.)
In high-end computing, the future is parallel. Clock speeds aren't getting faster. So we'll be getting multi-core. Blue Gene/P in 2008 will have 560 teraflop/s. Cray XT4 will be petaflop. 10 petaflop @ NCSA/Illinois in 2010. In 10 years, we may have exaflop machines! Petaflop and exaflop machines will have tens or hundreds of thousands of processors. They'll run slow for better reliability (less heat).
Tying it back to productive nanosystems: You will need computers; petaflop machines are around the corner; plan to use them on larger atomic systems; combine them with data repositories and experimental systems; think about multidisciplinary education (which may be corrosive to e.g. the idea of distinct physics departments).
Question: What's the current vogue for connections in multiprocessor machines? A: Current topology for interconnecting processors is a torus network. But with multi-cores (Blue Gene-P is a quad-core machine) communication will be fast among cores on a chip.
I didn't hear much relating to the Productive Nanosystems Roadmap in this talk.
Chris Phoenix
Tags: nanotechnology nanotech nano science technology ethics weblog blog
Comments