FOCUS ON…

Simulating the Universe

How do you find out what LSST will see and how it will see it? Simulations. Although astronomers and optical engineers have been simulating images and operations for space telescopes and adaptive optics systems for years, no one has simulated how data will be used — until LSST. On other telescopes, observers do the data reduction. Because the LSST data will be available to all, LSST’s data management group will do the data reduction to make sure that data are correct before “observers” have access to them for study. The science studies that can be done are dependent on the design of the sky survey and the attendant operations of LSST — the subject of current simulation activities. “Working in concert with science collaborations, the data management group and design engineers, the simulations will help us design the robotic brain (the scheduler) of LSST,” promises Philip Pinto, LSST scientist leading the Simulation and Data effort.

Why Simulate?

The LSST is designed to make very precise measurements of a wide range of astrophysical phenomena — to measure the unknown. To understand how well current designs can perform this daunting task, they must be tested and refined before the final design is actually built. The fact that the LSST is an all-purpose survey just makes this task more difficult — the LSST must achieve the best compromise among the many competing demands of a wide variety of measurements.

The light which arrives at the top of Earth’s atmosphere carries with it the subtle signals LSST is designed to detect. These signals include the distortions in the images of distant galaxies caused by intervening Dark Matter, the “fingerprint” of the chemical makeup of exploding stars, or the trajectories of potentially hazardous asteroids.

To understand if LSST can detect these signals, the image simulation team makes computer models of how this light is produced and how it is altered by its long passage to Earth. The next phase of simulations traces the passage of light through the Earth’s atmosphere and the telescope itself to the camera where it is detected and converted to “raw” digital images. Finally, this simulated image data is passed on to the LSST’s automated data-processing system. The end result is a database containing measurements. The important question is then: how close are the measurements to the original signals?

While these simulations test how well the LSST’s optics, electronics, and software can detect subtle signals, the operations simulation team is learning how to use this powerful new tool. How does one look often enough and in the right places to track moving objects? How uniform a map of the sky can LSST make in the face of ever-changing weather conditions? How much time should be spent moving the telescope from place to place on the sky and how much should be spent taking images? LSST will take roughly five million images during its ten-year survey. Operations simulation answers the question: which one should we do next? The algorithm for making this choice will become the brain of the robotic observatory.

While these simulations work in concert to test the design of LSST, it’s helpful to look at the separate activities and goals of each.

Creating an image

The main purpose of image simulations is to provide high fidelity simulations of what the actual signal will be. Using a model of the Universe and assumptions about the light travel through the Universe for given distances, atmospheric turbulence and other imperfections of conversion into electronic signal and bytes, the team creates facsimiles of observations.

The questions facing LSST’s image simulators are different from those of other projects. For example, simulations for designing control systems, such as adaptive optics systems, for telescopes usually deal with the response for a single or very few point source(s) as images. LSST must consider objects across a very large field and is focused on second order effects of object ellipticity, not just axi-symmetric PSFs. Most projects are not trying to generate the whole data flow of images to produce a final answer as LSST is. Also, consider the quantity of data to be simulated: the space telescope Gaia (launch date: 2011) will catalogue on the order of 1 billion stars with completeness to about 20th magnitude to make the largest, most precise 3-D map of the Galaxy. LSST will catalogue to four magnitudes fainter than Gaia. Only the proposed Pan-STARRS rivals LSST’s data collection goals. Finally LSST is interested in a different “flavor” of data: whereas a project such as the Large Hadron Collider generates far more data than LSST, it looks for rare event, and so a large portion of its data is not interesting to investigators. “LSST doesn’t know what’s interesting or what’s not; so we need to consider everything interesting,” says Pinto.

The image simulation team is gradually ramping up from providing small pieces of the focal planes to producing a larger and larger sample of data. This summer the team provided a simulation of ½ night’s data to the data management team. Work is still at the stage of inputing basic physics into the simulation models. The next steps are further model refinement and increased data amounts.

Operating LSST

“Operating the LSST is simply a ‘giant traveling salesmen problem’ with deadlines,” declares Pinto. Just as the traveling salesman has to determine how to get to as many cities as possible in a certain amount of time using the shortest route possible without arriving too early or too late, LSST has to complete observations in a way that optimizes the sometimes competing science goals. Striking the balance of all users’ types of science is an exercise in conflict resolution. Avoiding open warfare and preparing a roadmap for operations and observations is the goal of simulations.

The OpSim Team’s role is to develop tools to determine how the observing strategy and telescope design affect the science outcomes of LSST Survey. “What if?” simulations help different groups study their own scheduling problems to find solutions that provide science that is just as good but within a schedule that accommodates other science goals. For example, scientists studying near-Earth objects (NEOs) would like to observe ½ hour apart every night because this is the traditional cadence of such observations. But other science collaborations require observing time under differing conditions. Tension occurs among conflicting needs: the time available to sample, to achieve areal coverage and to reach a depth in sky. Add to mix the choice of filter and non-observing time and it becomes a challenge to satisfy (or simulate) all observing desires. The computational effort needed to work out conflicting inputs and desired outcomes quickly becomes too large unless limits are imposed — so the simulations cannot consider every possible alternative but they can provide a very good basis for reviewing operating options.

Operations simulation is based on a detailed design model of LSST: the telescope points to object “A”, accelerates to move to object “B”, decelerates, stops moving, opens shutter. Information available to the model includes the distance and angle between A and B as well as the time it takes to move between the two. As many such simulations are done, the work provides data management simulations, which predict how LSST will populate the database.

Simulations also help refine the telescope site design. Originally the design called for the dome slit to move in front of the telescope and have it stop at each observation. Running the simulation of this operation showed its inefficiency. Redesigning the dome to move continuously as the telescope moved to observations decreased the power consumption.

Putting it all together for the science

Simulating images and operations of the LSST is a tremendous task with computational, scientific, engineering and even political aspects. The image and operational simulations create a reference survey. This survey is not the LSST survey but a tool for use by the science collaboration teams and engineering teams to determine how best to optimize the work that will be done with LSST. Management and scientists will resolve the issues highlighted in the simulations and prioritize observations.

As simulations continue, they will provide the data management (DM) team with information to populate a database. Using information about individual observations of objects, how often LSST observes a portion of the sky, site information such as seeing and cloud cover and other variables, the image simulations will eventually become the simulated data in the database. The DM team will be able to further test the data pipelines and the database access and manipulation with these data.

The LSST simulations are more extensive and of a different kind than any previous telescope project. Scientists and data managers are learning more from these efforts than even they expected at the start of the project. With each refinement and expansion of the simulations the LSST brain is closer to completion.