快猫短视频

THE American Museum of Natural History in New York is known for its impressive collection of dinosaur bones and a high-tech planetarium. For astrophysicists, though, its claim to fame is as a computing powerhouse.

The fastest of the museum鈥檚 two computer clusters can carry out more than 5 trillion mathematical operations per second. This allows it to simulate aspects of the universe in unprecedented detail. Standing nearby, listening to the low hum of fans cooling the racks of processors, I can practically feel it recreating the astrophysical processes that shaped our solar system.

Harnessing supercomputers to simulate astrophysical events is not new. Two years ago, the largest, most detailed simulation of the evolution of the universe was performed at the Max Planck Institute for Astrophysics in Garching, Germany. In what was dubbed the Millennium Run, a supercomputer simulated the complex interplay of events that led to the formation of galaxies within a virtual cube with sides 2 billion light years long.

What is new is the software today鈥檚 giant machines are using. Their extra processing cycles are being coupled to a new set of algorithms that allow aspects of the universe to be looked at in much greater detail. Researchers have now managed to simulate the processes that power the sun, the life cycles of quasars and the initial fusion of particles that leads to planet formation. 鈥淭his is a very exciting time in astrophysics,鈥� says Saul Teukolsky of Cornell University in Ithaca, New York, who co-wrote a textbook on computer modelling.

While their colleagues in other disciplines can devise experiments to test their theories, astrophysicists are stuck with analysing the light that reaches us from celestial objects. To fill in the gaps they rely on simulations. 鈥淪imulations give us a completely new view of the universe from what is available by observation alone,鈥� says Stelios Kazantzidis of Stanford University in California. 鈥淚t is the closest thing to a physics lab that we have.鈥�

In the past, most simulations relied on a computing method called smoothed particle hydrodynamics (SPH) in which the whole universe is treated as if it were composed of solid particles, with dense areas such as galaxies having more particles. This lends itself to simulating gravity, which can be modelled as a force acting between the particles. SPH worked well for the Millennium Run, which simulated the gravitational forces that turned ripples in the early universe into the galaxies, quasars and black holes. It has also modelled the expansion and contraction of the universe.

But SPH models struggle to deal with the complex details of these processes such as viscosity, turbulence, drag and magnetism, which become more important than gravity at smaller scales. Now that is starting to change. Nick Gnedin of the University of Chicago is using an alternative to SPH known as adaptive mesh refinement (AMR). Instead of treating the universe as a collection of particles, AMR models it as a single mesh or fabric. That works well for modelling these other forces: pull a section of the fabric, and neighbouring sections are affected, rather like the effects of turbulence and viscosity in a fluid or a gas. Pull on a particle under SPH, and the only way it can affect other particles is via gravity.

AMR was first discussed 20 years ago, but only now, thanks to the latest supercomputers, is it being used for cosmology, says Gnedin. 鈥淔or the next 10 years, AMR will be the workhorse of cosmology.鈥�

Now his team has used AMR to add the effects of light to a key process in galaxy formation: . Crudely, the ripples are concentrations of gravity that draw in more matter. But the speed of the process is affected by the light produced as gravity fuses nuclei. The light exerts an outward pressure that is more like the action of a fluid or gas than the interaction with solid particles, so the process is more accurately modelled by AMR than SPH. 鈥淥riginally cosmology models just had gravity in them,鈥� says Ralph Roskies, the head of the Pittsburgh Supercomputing Center, a joint venture of Carnegie Mellon University in Pennsylvania, the University of Pittsburgh and Westinghouse Electric Company. 鈥淣ow there is gravity and radiation.鈥�

In future, cosmologists plan to add magnetism to models like Gnedin鈥檚, using a technique that takes into account the complex twists of magnetic field lines, which look rather like the turns of a screw and create a complex interplay of forces. In 2001, Dinshaw Balsara now at the University of Notre Dame in South Bend, Indiana, worked out how to add such detailed field lines to AMR simulations (Journal of Computational Physics, vol 174, p 614).

Others are using new supercomputer algorithms to study processes that occur closer to home. Using the Pittsburgh Supercomputing Center to run his simulation, Juri Toomre of the University of Colorado in Boulder is modelling the nuclear reactions and energy transfer that power the sun (see Picture). The nuclear energy produced by fusion reactions in the sun鈥檚 core is transported to the surface by convection currents, which are not modelled well by SPH. To make things worse, the currents are subject to turbulence caused by the sun鈥檚 rotation and the complicated magnetic field lines that snake through it. 鈥淭he sun is an in-your-face challenge,鈥� says Toomre. 鈥淚t is probably the most turbulent area in the universe.鈥�

He says that advances in supercomputing in the past few years have allowed his team to use a technique known as ASH, which is similar to AMR except it works on the scale of a single star, rather than a galaxy. They used it to simulate the interaction between convection, rotation and magnetic field lines, leading to the . Toomre hopes to apply a similar model to other stars in future.

Another turbulent system that was a mystery was the earliest processes of planet formation. 快猫短视频s knew that grains of interstellar dust coalesce to form larger bodies called planetesimals, which in turn form planets. But how exactly the grains coalesce was poorly understood. As well as gravity, the process depends on turbulence and drag caused when the central star turns on and blows the surrounding gas outwards. To model the process, Mordecai Mac Low of AMNH, and colleagues, used the museum鈥檚 computing cluster to run another alternative to SPH known as the Pencil Code (Nature, ). This too is similar to AMR and ASH, but works on an even smaller scale.

So what is next? Researchers would like to bring SPH and AMR-like techniques together, so that both large-scale and small-scale processes can be examined at once. Other systems are being developed that could trump SPH and AMR but have not yet been tested. What is clear is that increasing computing power (see 鈥淏rains to rent鈥�) is leading to ever more realistic simulations. 鈥淧eople aren鈥檛 willing any more to accept an impressionistic simulation that shows an interesting effect,鈥� Balsara says. 鈥淭hey want the whole enchilada.鈥�

鈥淧eople no longer want an impressionistic simulation. They want the whole enchilada鈥�

Cosmology 鈥� Keep up with the latest ideas in our .