Prime Focus Spectrograph (PFS): Fiber assignment through min-cost/max-flow

The Prime Focus Spectrograph (PFS) survey will target the same patch of sky from several dozens to 100 times. The problem of allocating PFS' 2394 fibers to objects over many visits of a field is a highly non-trivial optimization problem. Our network flow approach models the fiber allocation as a generalized network min-cost/max-flow problem. This methodology is inspired by SDSS, but extends this to address the variety of requirements of the the PFS survey. Ultimately, we solve the network flow through linear programming. This generally provides a very good solution in reasonable amounts of time and can give a clear quantitative measure of just “how good it is”. It allows us to define an arbitrary number of target classes with different weights, to enforce constraints on the target distribution, and to put caps to the number of observed objects per class. We will present the methodology and the implementation of our approach.