October 2013

Supernova revelations

Supercomputer models point to twists in a star’s death throes.

Left: About half a second after ignition, the buoyant flame (red) has traveled nearly 900 kilometers through the white dwarf (blue). Center: A detailed look at the flame’s surface at this point shows it to be complex and highly wrinkled due to the vorticity and turbulence generated by its own rise. Right: Just a quarter of a second later and the flame breaks through the surface of the now slightly expanded white dwarf.

Supernovae used to be simple. Just ask Stan Woosley.

Almost 50 years ago, when the renowned astrophysicist began contemplating the explosive deaths of stars, there were just two kinds: core collapse and exploding white dwarfs, the latter now better known as Type 1a supernovae.

The physics of these cosmic firecrackers was simple: They ignited and blew up.

“Now I think we’re reaching a compromise that Type 1a supernovae are more than one thing and each of them is very complicated,” says Woosley, 68, a world-leading computational astrophysicist at the University of California, Santa Cruz.

“This is what happens in science: It starts simple and then gets messy.”

With the Jaguar and now Titan supercomputers at Oak Ridge National Laboratory’s Leadership Computing Facility, Woosley and colleagues are getting the most detailed look ever at what’s happening inside white dwarfs in the critical moments before they blow.

Petascale computers, capable of quadrillions of calculations per second, “are giving us a new view of the stars,” Woosley says.

What computational astrophysicists are seeing through the 100,000-processor lens is changing our understanding of thermonuclear combustion in space – and maybe even on Earth.

The shift in perspective is even more dramatic in Woosley’s supercomputer simulations.

Woosley is quick to note that the coupled powers of faster machines and customized codes drive his stellar computational insights.

His team’s recent landmark work relied on two made-to-order petascale combustion codes – MAESTRO and CASTRO – developed by Lawrence Berkeley National Laboratory mathematicians John Bell and Anne Almgren, who specialize in turbulent combustion at the extremes.

“Without those codes we couldn’t have done anything at Oak Ridge,” Woosley says.

Supernova Revalations 2

Left: Near the center of the white dwarf (blue dot), nuclear burning drives convection which causes hot plumes (orange/red) to rise and expand. After about a century, the burning is so vigorous that the temperature approaches a billion degrees and ignites a flame at a single point. Right: This hot flame buoyantly rises towards the surface as it burns, creating vorticity (increasing from blue to yellow/white) and turbulence that affects the flame’s further evolution.

Both codes use the same BoxLib software framework. That means Woosley and colleagues could merge MAESTRO output into CASTRO to run the first end-to-end, first principles three-dimensional simulation of a Type 1a supernova from ignition to observables – the light curves and spectra telescope astronomers can use to validate the simulation.

What these supercomputer simulations show, Woosley says, is that what was once thought to be a single type of white-dwarf-based supernova is at least three kinds.

That finding could have broad implications for cosmological studies because Type 1a supernovae are used as so-called standard candles – identical light sources – for determining cosmic distances and rates of expansion.

The canonical Type 1a supernova representation, first described in the 1960s and used in cosmological models, is the Chandrasekhar mass model. In this template, a white dwarf – a carbon-oxygen cinder of an old sun-like star, smaller in radius than the moon – is in a binary pair with a much younger star.

Over millions of years, the white dwarf pulls matter from its neighbor, i growing until it reaches 1.38 times the sun’s mass – the Chandrasekhar limit. This mass is a stellar tipping point: The gravitational pressure exerted by the heavy dwarf ignites the carbon oxygen core, triggering a Type 1a supernova.

“It’s always a white dwarf that’s blowing-up,” Woosley says. “But now we think the white dwarf can be a Chandrasekhar mass or short of the Chandrasekhar mass and accreting from a star or can come from two white dwarfs merging. We’re not sure at this point which model is the main one, so at Oak Ridge we’re modeling all three.”

These simulations have revealed that the Chandrasekhar mass model itself is lopsided.

Over the past decade there’s been intense debate among supernova physicists about the specifics of Type 1a ignition. Does the nuclear flashpoint in these cosmic chunks of super dense carbon and oxygen start in the middle, at a single point or simultaneously at several points?

“Our calculations at Oak Ridge have shown that the Chandrasekhar model white dwarf only ignites one time and never at the middle of the star,” Woosley says. The research was published last year with colleagues including UC-Santa Cruz post-doctoral astrophysicist Chris Malone and Andy Nonaka, a research scientist at Berkeley Lab’s Center for Computational Science and Engineering. “Even though the off-center variance is only 30 to 100 kilometers, this makes a very different supernova than if you ignite in the middle, or five places at once.”

The shift in perspective is even more dramatic in Woosley’s supercomputer simulations, which clearly show that sub-Chandrasekhar mass white dwarf’s can experience a two step detonation process – similar to that envisioned for terrestrial reactions at DOE’s NIF, the National Ignition Facility. NIF’s powerful lasers generate X-rays that should compress a BB-sized capsule of frozen hydrogen isotopes and spark a nuclear fusion reaction.

In this case, a carbon-oxygen dwarf in a binary accretes an onion-skin-like layer of helium. Nuclear burning is sparked not in the dwarf’s carbon heart but at the helium layer’s lower boundary. The heated helium shell compresses the underlying matter, raising the temperature and triggering a detonation in the carbon core.

“One of the things we’ve done at Oak Ridge is to show that it takes just the right amount of helium. Too little and the star doesn’t blow up and too much and the spectrum is all wrong,” Woosley says, referring to work with UC-Santa Cruz postdoc Rainer Moll. “If you have just the right amount, the helium detonation wraps around the star and it becomes like a successful NIF pellet.”

Working with Moll and Berkeley Lab’s Daniel Kasen, Woosley is simulating supernovae resulting from merging white dwarfs. General relativity predicts the two orbiting stars will, in time, gravitationally fall into one another.

Woosley says their initial modeling shows this can indeed occur and produce spectra and light curves that match those from observed Type 1a supernovae.

From these different results, astronomers now have a bounty of data with which to analyze observed stellar explosions.

And five decades into his research career Woosley’s supernova interest shows no signs of abating, with “retirement” a code word for more time to do research.

With next-generation exascale computers coming, he’s already contemplating the secrets that might be digitally revealed about the more common Type II, or core collapse, supernovae. These occur with the collapse of a massive star’s iron core, releasing a flood of neutrinos that blow off all the matter outside a newly formed neutron star.

“Building codes where realistic radiation and neutrino transport scale well on more than 100,000 (processor) cores will be a challenge,” Woosley says. “Not impossible. It just needs a lot of work.”