On the least action principle in cosmology

TL;DR

This paper introduces FAM, a fast and efficient method based on the least action principle to reconstruct the past orbits of galaxies in an expanding universe, enabling large-scale cosmological flow analysis.

Contribution

The paper presents FAM, a novel, computationally efficient implementation of the least action principle that can handle significantly larger datasets than previous methods.

Findings

FAM accurately recovers present peculiar velocities.

It successfully reconstructs the flow field down to cluster scales.

The method is computationally efficient, handling ~15,000 galaxies in about 12,000 CPU seconds.

Abstract

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-base 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 fields appearing in the action and its gradients. The time integration of the Lagrangian is done using Gaussian quadratures. FAM allows us to increase the number of galaxies used in previous numerical action principle implementations by more than one order of magnitude. For example, orbits for the $\sim 15,000$ IRAS PSC$z$ galaxies can be recovered in $\sim 12,000$ CPU seconds on a 400MHz 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 clusters 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 as in redshift space by direct minimization of a modified action, without iterating the solution between real and redshift spaces.