SNAP Overview:

The Supernova Acceleration Probe uses type Ia supernovae as standard candles to make precision measurements of brightness (or magnitude) and speed (or redshift). The relation between brightness and recession speed is directly affected by the geometry of the universe and its evolution over time. The first measurement of this distribution was performed by Hubble in the 1930s, who discovered that the further away (dimmer) a galaxy is, the more red-shifted its spectrum. He concluded from this that the universe is expanding with a constant of expansion that now bears his name, the Hubble constant.

Since that time, improved measurements have pinned the value of Hubble's constant to about 65. Measurements of the brightness/speed relation are compared to calculations of the cosmos that include the attractive and repulsive forces from the energy distribution of the cosmos and the effect of these on the expansion and curvature of space from general relativity.

In 1998 two new measurements were published within a few weeks of each other that showed a small deviation from the current theory. Today the leading explanation of this deviation is that Dark Energy contributes about 70% to the energy budget of the universe. It is a repulsive force emanating from every spot in the universe that accelerates its expansion. Understanding the magnitude and characteristics of Dark Energy is one of the most interesting problems in physics today.

SNAP Data-Mining:

The SNAP satellite will beam about 50 MBytes of data to earth each second during its 3 year mission. The data will be archived, and processed through several parallel analysis pipelines. The resulting data sets will contain catalogs of images and astronomical objects as well as the raw images and spectra themselves. The data will need a highly optimized physical organization to facilitate its retrieval for detailed study. The archive is estimated at the 50 TByte level. It is estimated that just under 200 CPUs will be needed for the primary supernova discovery pipeline. With some fine-tuning of the network load, gigabit Ethernet is suitable.

Physically, computing architectures for this scale project are either high-performance parallel computing clusters connected with gigabit Ethernet to multi-TByte RAID arrays and tape archives, OR they are the GRID which relies on a massive resources accessible via the Internet. The choice between these two options will depend on bandwidth, multi-platform algorithm reliability, and portability of software and licenses.

An interesting engineering challenge for SNAP results from deviations from a smoothly operating pipeline. We find that if the CPU cluster is down for 10 minutes, it will take ~4 hours to catch up again. The problem results from network bandwidth and is not solved by adding processors. Technological advances in networks will be needed to engineer SNAP with sufficient safety margin. Other solutions are possible, but would require custom software using shared memory options. These are less desirable, because they make various analysis pipelines interdependent on the physical location of memory and basically result in a far less flexible environment.

The pipeline will be implemented within a framework suitable for the analysis. Several pipelines are currently used by the astronomy community for the Hubble Space Telescope and the Sloan Digital Sky Survey. Reusing their software will be cost effective if it meets the SNAP requirements.

SN Image Decomposition:

Supernovae are discovered by comparing recently acquired images with an older (deeper) reference image. If there is a source of light in the new image that did not exist in the old image, it is a supernova. It is easiest to identify new sources in an image created by subtracting the new image from the reference image. Since supernovae usually occur in galaxies, it is important to get both the zero-level subtraction identical as well as the host galaxy subtraction. Aligning the images, matching the point-spread functions, matching the photometry and bias all require precise calibration. We are working on better automated signal processing software to facilitate supernovae discovery.

Most supernovae are not type Ia, suitable for cosmological standard candles. Type Ia supernovae are identified by their spectral properties. We are working on selection algorithms based on the characteristic rise of the type Ia light-curves and the coarse (R,B,V,I band) color characteristics to reduce the backgrounds from type Ib, Ic and type II supernovae at early epochs.

Simulated type Ia lightcurve for SNAP (postscript)
Simulated type II lightcurve for SNAP (postscript)