New Science from High performance Computing

October 24-25. There was a two day meeting held at The Royal Society (UK Academy of Science). It discussed the new science emanating from the use of high performance computing. The meeting was attended by around 130 research scientists from academia and industry. The presentations were a tour de force of the latest research in chemistry, biomolecular modelling, materials, engineering, lasers, the makeup of the earth?s core, the climate of the whole earth, elementary particles and the theory of everything in the makeup of matter, to cosmology, the makeup of the universe.

With such diverse programme it would be impossible to do all these subjects justice, so in this article I'll concentrate on the chemistry and biomedical chemistry talks.

Chemistry at the heart of all material.

Chemistry is at the heart of most materials and products common today. Societies rely on a diversity of materials, steel for cars, fibres for clothes, building materials for houses, paints, fertilisers, pest control chemicals, pharmaceuticals and other health care products, fuels and semiconductors for electronic devices. Many material in common use today are synthetic. They were designed by chemists, at first by testing related compounds using empirical trial-and-error methods, but more recently using high performance computers and the quantum mechanical approach.

Materials are described and classified according to magnetic properties, their hardness, melting point, optical appearance, electrical conductivity, and chemical reactivity. In theory, if we could construct mathematical equations which describe these properties, we could predict how any material behaves by simply solving these equations. But to find the equations we need first to understand the physical laws that govern the properties of material.

We now know that electrons play a crucial role in the properties of materials. Electrons create and break chemical bonds, and their distribution determines many chemically important material properties such as the shapes of molecules, the arrangement of atoms in solids, the forces between atoms, and their electrical properties, (whether they are insulators, semiconductors, or metals). Even the complex macro-biomolecules in living organisms are ultimately governed by the electrons of atoms forming these molecules.

Quantum Mechanical Model.

It is now over seventy years since scientists formulated the theory of quantum mechanics which allows a quantitative prediction of the distribution and energies of electrons. The fundamental equation of quantum mechanics, is the Schroedinger equation, (named after the Austrian physicist who proposed it), and contains the key to understanding the properties of matter. It gives a complete (statistical) description of the electrons in atoms, molecules and solids. Solving Schroedinger's equation for a given arrangement of atoms reveals the distribution of the electrons and the corresponding energy of the molecule or solid. From this, one can determine things such as the way atoms are arranged in a molecule or solid when they are in their lowest total energy state, (so called ground state), and how to calculate the forces binding the atoms together. The ground state is generally the most stable form of the molecule and thus, the form the molecule commonly assumes in nature. To determine the ground state of a molecule or solid, researchers choose an arbitrary atomic configuration (Ab Initio) and calculate the total energy. They then, change the geometry, recalculate the total energy and compare it with the previous value. If the new value is lower, this procedure is repeated until the structure of lowest total energy is found. This structure is taken as the actual structure of the molecule or solid.

One can then proceed to pull the molecule or solid apart, step by step, recalculating the total energy for each step until it falls apart. This series of calculations reveals how strongly the molecule or solid holds together and how much energy is needed to break it. Another approach is to start with isolated atoms or molecules and bring them together through a series of calculations to form a new molecule, thereby revealing the new molecule's stability. These two procedures - breaking and making chemical bonds and forming new compounds - are at the heart of chemistry.

The Challenges of Practical Difficulties.

All this sounds very easy, but in practice, there are several difficulties to overcome. Even small molecules have many electrons and each electron interacts with all the others as well as the atomic nuclei in a molecule. For example, water, a relatively small molecule contains 10 electrons and three atomic nuclei. The metal tungsten, (used as a filament in light bulbs), each of its individual atoms carries 74 electrons, and to form even a small piece of metal requires about 1023, (ten to the power of twenty-three), of these tungsten atoms.

A second difficulty is that the motion of electrons is not governed by classical mechanics, which have been used very successfully in engineering disciplines, but by the laws of quantum (statistical) mechanics and in some cases it also includes the theory of relativity.

For many years scientists struggled with these challenges and although they formulated many brilliant methods they were reduced to calculating the already known properties of small molecules. With the advent of supercomputers and parallel high performance computers, chemistry has entered an era where simulation of real-world chemical systems became a reality. Chemists can now tackle complex systems that are difficult or even impossible to study experimentally. Molecules, solids, and surfaces can be constructed and visualised graphically. By performing complex calculations and looking at intermediate steps not accessible experimentally, scientists can now study in detail and thus control chemical processes. In this way, the computer-aided design of new materials has become a reality and is revolutionising the way chemists develop new materials to satisfy human needs.

As far back as 1985, Du Pont de Nemours, the largest chemical company in the USA, started to use supercomputers for molecule manipulation and design of new materials. Now all major chemical manufactures are using these techniques. Applications packages such as AMBER, (Assisted Model Building with Energy refinement), the Ab Initio packages GAUSSIAN and GAMES, the UNICHEM package and so on, are providing a wealth of procedures to explore and understand the properties of new materials.

Assistance from High Performance Computing.

In his talk Professor Thom Dunning from the North Carolina Supercomputing Centre, USA, reviewed the supercomputer architectures from the Cray vector machines to the proposed specialised Blue Gene Petaflop/s machine. He illustrated how the computation and memory bandwidth of Cray and the parallel vector NEC SX series machines are designed to maximise sustained performance made available to the user. This is usually in the order of 50-60%, while massively parallel distributed systems such as those offered by IBM and other vendors with PC based systems, their computing to memory ratio is unbalanced and therefore, despite claims of high peak flop/s performance, in reality deliver very little (often 2-5% computing power), to large scale applications. The reason for this poor return to the user, is because they use a small cache supplemented by slow main memory and relatively slow communication switches.

He then discussed the challenges that must be addressed to fully realise the benefits offered by high performance computing. The task is not easy as it requires close collaboration among disciplinary computational scientists, computer specialists and applied mathematicians. He illustrated this in a brief review of the development of NWChem, a general purpose computational chemistry code designed from the ground up for parallel supercomputers. He then went on to discuss examples of the new science that became possible by using the capabilities of NWChem. These include water-carbon interactions and an explosion of size of problems which can be handled.

He concluded by saying that there are theoretical challenges, such as high fidelity description of complex systems, to be overcome. For example, perturbation theory in molecular modelling does not always converge and therefore is not good enough for getting a good approximation of the behaviour of real systems. In combustion systems sub-grid processes are not accounted by current grid sizes.

New Science in Medicine Applications.

Another important area of application is in medicine where action drugs are designed and used to block a specific site of a macro-molecule, termed the drug receptor. As the structures of more of the body's macro-molecules become known , the design of anti-viral drugs for specific functions, such as for example, blocking the HIV virus which causes AIDS, can be tackled.

In her talk, Professor Valerie Daggett, department of medicinal chemistry, university of Washington, discussed her work in understanding the prion protein (PrP), via computational molecular modelling. She went on to say that: as computer power increases, the range of interesting biomolecular phenomena and properties that can be simulated also increases. It is now possible to simulate complicated protein conformational changes at ambient or physiological temperatures". In this regard, she has been attempting to map the conformational transitions of the prion protein to its infectious scrapie isoform which is capable of causing neuro-degenerative diseases in many mammals.

In the UK we are familiar with its Bovine Spongiform Encephalopathus manifestation in cows, BSE, and its human variant Creutzfelt-Jakob disease, CJD. These two forms have identical sequences and are conformational isomers, with heightened formation of sheet structure in the scrapie form. Conversion can be triggered by low pH and in vivo it appears to take place in an endocytic pathway and/or caveolae-like domains. It has thus far been impossible to characterise the conformational change at high resolution using experimental methods. Therefore, to investigate the effect of acidic pH on PrP conformation, numerous molecular dynamics simulations of hamster, human and bovine forms of PrP in water neutral and low pH were performed. It was found that the core of the protein is well maintained at neutral pH. At low pH, however, the protein is more dynamic, and the sheet-like structure increases both by lengthening of the native b-sheet and by addition of a portion of the N-terminus to widen the sheet by another two strands. This simulation uses initial data from an NMR scan, and then uses 25 thousand atoms, mostly water. The modelling could benefit from a hundred-fold of computing resources, both power and memory as well as TBytes of storage space.

Both BSE and CJD are poorly understood because governments, especially in the UK, for a long time were in denial, afraid of the consequences on the agro-industry, and hence, were slow to fund vital research.

The Evolution of the Universe.

The meeting ended with a talk by professor Carlos Frenk, university of Durham, UK, discussing the latest on cosmology. He concluded that computational simulation models accurately describe our universe. He said: "we live in a dominantly flat universe; gravity is due to cold dark matter; inflation from initial conditions (big bang) describes the adiabatic Gaussian behaviour of galaxy formation; and the structure growth of the universe can be accounted by gravity instability. Cold dark matter and Baryon matter only account for about 33% of all matter in the universe, so the universe must be filled with elementary particles, not yet discovered.

The two day meeting gave a superb survey of the new science made possible in the last decade, by the use of high performance computing. Despite the diversity of science there was a lot of commonality of techniques used, mathematical modelling and HPC are common threads. After listening for two days it became apparent that high performance computing is indispensable for new science. Most speakers used computers with hundreds of Gflop/s peak performance, but all of them wanted access to Terascale systems to enable them to improve the capabilities of their models and obtain more realistic results and better scientific insights.