Yesterday at AIMS we were fortunate enough to have two seminars to start off the week. The first was from Brent Miszalski who is a SALT Research Fellow at the SAAO. He came to tell us about an optimization problem in multi-object spectroscopy.
Brent began by motivating the need for large galaxy redshift surveys. These surveys measure the redshifts of many galaxies over a large area of the sky. As a result , by using the Hubble Law (which relates distance to redshift) we can measure the cosmological properties of the universe as well as the distribution of galaxies.
Brent then explained that multi-object spectroscopy (MOS) is a technique important in carrying out such large galaxy redshift surveys, as in spectra are needed to measure redshifts, and large numbers need to be taken at once in order to survey large numbers of galaxies.
MOS involves taking a known distribution of objects in a particular part of the sky and then using either a mask, robotic fiber optic positioning device or some other technique to ensure that spectra are only taken for those parts of the sky where there are objects of interest.
Because very large numbers of of objects are involved, and because the systems involved have several constraints, the setting up of the telescope for different parts of the sky (which involves moving fibers etc.) can be a challenging optimization problem.
Brent explained to us his work on designing a more optimal routine for the instrument used in the 2F survey. This telescope gathers spectra via optical fibers which are repositioned between observations using a robotic arm.
The constraints in the system are to have the fibers as straight as possible, pick mostly high priority targets and not to collide fibers and sensors.
The algorithm that Brent discussed involved using simulated annealing, a computational technique that explores configurations through random while simulating the slow cooling of the system being studied. As the simulated system “cools” fewer and fewer “bad” changes are accepted.
After explaining how he implemented the various constraints, Brent showed as some results that illustrated the improvement the simulated annealing algorithm made on optimizing the fiber positioning.
In the last part of his talk, Brent discussed MOS optimization problem for SALT. Here, instead of robotic positioning, the sky is masked using a carbon fiber plate, with slits cut it in order to collect spectra from the relevant objects.
Brent explained that the constraints here involve maximizing the number of objects each slit covers, while avoiding overlap of the spectra that the slits produce.
The talk finished with an appeal for students to begin working on the problem. A problem which is not only fascinating, but one whose solution could be very useful for the next several years of SALT operation.