I represented NASA and the University of Arizona at the Supercomputing Conference (SC14). I presented research projects I have been involved in on behalf of Dr. Tami Rogers, the primary investigator of these projects. My primary involvement was to use VAPOR to help us visualize our data, although I was familiar with the code and helped improve its parallel algorithm slightly.
NASA created a web page for each of the exhibitors, complete with a brief description of the project and some pictures. The page related to my work is located here.
I gave two different presentations multiple times throughout the conference. One presentation was in the "Theater" section of the NASA booth (20-25 minutes). The other was at a demo booth just outside the Theater area, and was given to curious passers-by (8-10 minutes).
Introduction
We are living in an exciting time for astronomy. More observatories, satellites, space probes and landers are gathering an increasing amount of high quality data. The Large Synoptic Survey Telescope is expected to produce around 20 TB per night, and the Square Kilometer Array is expected to produce nearly an exabyte (a million terabytes) of data every day!
More planets are being discovered and documented than ever before. NASA's Kepler Space Telescope has helped us discover a variety of new and exotic planets, bringing our total candidate planet count to over 4000 and our total confirmed planet count to over 1000.
With this massive influx of data, is important to be able to interpret these observations correctly. It is useful to make predictions that help astronomers point their telescopes and other instruments to the most interesting objects in the sky and make the best use of their resources.
Our contribution is to develop self-consistent numerical models to test hypotheses, probe dynamics and inform future observations for two types of astronomical objects: massive stars and hot Jupiters. Our models are based on the DR. SPIN code developed by Dr. Rogers, which is based on the DYNAMO code developed by Dr. Gary Glatzmaier. These models solve the nonlinear Navier-Stokes equations (plus the magnetic induction equation) self-consistently throughout the entire computational domain and use the anelastic approximation to handle density/pressure stratification. This allows us to study the spectrum of generated internal gravity waves (IGWs) and the behavior of the magnetic field and flow structures with more confidence than could be done with parameterized models, for example.
Massive Star Model
We developed and ran 2D and 3D models of convectively-driven IGWs in a massive star. We used a Cambridge STARS 1D stellar evolution model for a hypothetical star three times the mass of our Sun as our reference state. The 2D models each ran for several hundred thousand processor-hours (about 10 years on a single processor), and the 3D models each ran for several million processor-hours (about 100 years on a single processor).
Here are a few of our results:
- IGWs are generated efficiently at the convective-radiative boundary.
- IGWs can cause the surface of a massive star to rotate differently than the interior bulk of the star.
- The surface of massive stars could show signs of rotating differently at different points in time, as a result of IGW action.
- Angular momentum transport by IGWs may explain the apparent misalignment of exoplanets around hot stars, chemical anomalies, non-synchronous orbits of interacting binary stars, and phenomena involving Be-class stars.
For more information about our massive star simulations, please see the following publications:
Hot Jupiter Model
"Hot Jupiters" are Jupiter-like planets in other solar systems that orbit very close to their host star. As a result, they receive lots of light from the star and are very hot compared to further-out gas giants such as Jupiter and Saturn.
Hot Jupiters generally experience significant tidal forces that cause them to experience synchronous rotation, or tidal locking. As a result, hot Jupiters have a hot day side and (relatively) cold night side.
The atmospheres of hot Jupiters are hot enough to be partially ionized. Strong stellar or planetary magnetic fields interact with these ionized winds in a complicated way that is described by magnetohydrodyamics (MHD). Because of the degree of ionization and strength of the magnetic fields, MHD is exected to play a major role in hot Jupiter atmospheres.
Our numerical models of hot Jupiters were similar to the massive star model, except they require significantly less resolution, we account for a variable magnetic diffusivity, the temperature equation explicitly adds Newtonian cooling and Ohmic dissipation, and the stellar burning term is dropped in favor of an insolation term. These 3D models each ran for several hundred thousand processor-hours.
Here are a few of our results:
- Ohmic dissipation is unlikely to explain the observed inflated radii affecting roughly half of hot Jupiters. Predictions estimate that Ohmic dissipation needs to supply the atmosphere with roughly 1019 Watts to explain the observed inflated radii, but our models indicate that only roughly 1017 Watts are actually being produced. High wind speeds and stronger magnetic fields should increase Ohmic dissipation, but are unlikely to close the gap.
- The nightside of hot Jupiters may experience intermittant field reversals and strong instabilities.
- In the hottest models, mean winds are dragged so much that they reverse direction. This can displace the hot spot towards the west.
- Magnetic effects on hot Jupiters are potentially observable via displaced hotspots. (Other signatures of magnetic activity are possible but have not yet been considered.)
For more information about these (and other) hot Jupiter simulations, please see the following publications:
Computing Resources
Our simulations made use of the NASA High End Computing Capability (HECC) resources at NASA Ames Research center. Our models generated over 60 TB of data, and we produced at least another 10 TB just for visualization. Without support from NASA, our models would have taken 10-100 years to run in serial on a desktop computer (and only then if that desktop had a massive amount of RAM).
Visualization
We used the VAPOR software package (www.vapor.ucar.edu), a product of the the Computational Information Systems Laboratory at the National Center for Atmospheric Research. At the time, VAPOR had very limited support for datasets in spherical coordinates, which was problematic for our simulations. As a workaround, I developed a software pipeline to transform and interpolate our data onto a Cartesian coordinate grid before feeding the transformed data into VAPOR. Due to the size of the simulation data, we had to purchase a separate hard drive with striping (RAID-0) in order to process the data.