Abstract
We introduce a new algorithm, the Constant Constraint Matrix Approximation (CCMA), for constraining distances in molecular simulations. It combines the best features of many existing algorithms while avoiding their defects: it is fast, stable, can be applied to arbitrary constraint topologies, and can be efficiently implemented on modern parallel architectures. We test it on a protein with bond length and limited angle constraints, and find that it requires less than one sixth as many iterations as SHAKE to converge.