In commonly used Monte Carlo algorithms for lattice gauge theories, the integrated auto-correlation time of the topological charge is known to increase exponentially as the continuum limit and/or the large-N limit are approached, causing the Monte Carlo Markov chain to remain trapped in a fixed topological sector. This topological freezing constitutes a major obstruction to the study of the topological properties of gauge theories on the lattice. In this seminar I will present the novel parallel tempering on boundary conditions algorithm that I have recently implemented for 4d SU(N) Yang-Mills theories, following the original proposal of M. Hasenbusch for 2d CP(N-1) models, which allows to dramatically reduce the auto-correlation time of the topological charge by up to two orders of magnitude. I will discuss improvements and results obtained with this algorithm for the topological susceptibility, for the quartic coefficient b_2 (related to <Q^4>) and for low-lying glueball masses.
Nobuyuki Matsumoto