The recently proposed statistical temperature molecular dynamics (STMD) algorithm [Kim et al., Phys. Rev. Lett. 97, 050601 (2006)] is used as the core of an optimization algorithm, statistical temperature annealing (STA), for finding low-lying energy minima of complex potential energy landscapes. Since STMD realizes a random walk in energy, the idea is simply to initiate repeated minimizations from configurations in the low-energy segments of STMD trajectories. STA is tested in structural optimization of various off-lattice $AB$ and extended $AB$ protein models in two and three dimensions with different chain lengths. New putative ground states were found for the two- and three-dimensional $AB$ 55-mer, and for the three-dimensional extended $AB$ 21-mer and 55-mer. The distinct folding features of the models are analyzed in terms of the statistical temperature and other representations of the structure of the potential energy landscape. It is shown that the characteristic behavior of the statistical temperature undergoes a qualitative change with the inclusion of a torsional potential in the extended $AB$ model, as the more rigid backbone makes the potential energy landscape more funnel-like.

• Received 11 January 2007

DOI:https://doi.org/10.1103/PhysRevE.76.011913