Convergence and efficiency of angular momentum projection for many-body systems

Abstract

In many so-called "beyond-mean-field" many-body methods, one creates symmetry-breaking states and then projects out states with good quantum number(s); the most important example is angular momentum. Motivated by the computational intensity of symmetry restoration, we investigate the numerical convergence of two competing methods for angular momentum projection with rotations over Euler angles, the textbook-standard projection through quadrature, and a recently introduced projection through linear algebra. We find well-defined patterns of convergence with increasing number of mesh points (for quadrature) and cut-offs (for linear algebra). Because the method of projection through linear algebra requires inverting matrices generated on a mesh of Euler angles, we discuss two methods for robustly reducing the number of required evaluations. Reviewing the literature, we find our inversion involving rotations about the $z$-axis is equivalent to trapezoidal "quadrature" commonly used as well as Fomenko projection used for particle-number projection. The efficiency depends upon the number of angular momentum $J$ to be projected, but in general inversion methods, including Fomenko projection/trapezoidal "quadrature" dramatically improve the efficiency.