On the least action principle in cosmology

Given the present distribution of mass tracing objects in an expanding universe, we develop and test a fast method for recovering their past orbits using the least action principle. In this method, termed FAM for fast action minimization, the orbits are expanded in a set of orthogonal time basis functions satisfying the appropriate boundary conditions at the initial and final times. The conjugate gradient method is applied to locate the extremum of the action in the space of the expansion coefficients of the orbits. The treecode gravity solver routine is used for computing the gravitational field appearing in the action and the potential field appearing in the gradient of the action. The time integration of the Lagrangian is done using Gaussian quadratures. FAM allows us to increase the number of galaxies over previous numerical action principle implementations by more than one order of magnitude. For example, orbits for the 15 000 IRAS PSC z galaxies can be recovered in 12 000 CPU seconds on a 400-MHz DEC-Alpha machine. FAM can recover the present peculiar velocities of particles and the initial fluctuations field. It successfully recovers the flow field down to cluster scales, where deviations of the flow from the Zel'dovich solution are significant. We also show how to recover orbits from the present distribution of objects in redshift space by direct minimization of a modified action, without iterating the solution between real and redshift spaces.