We present a Markov-chain Monte Carlo algorithm of worm type that correctly
simulates the fully-packed loop model on the honeycomb lattice, and we prove
that it is ergodic and has uniform stationary distribution. The fully-packed
loop model on the honeycomb lattice is equivalent to the zero-temperature
triangular-lattice antiferromagnetic Ising model, which is fully frustrated and
notoriously difficult to simulate. We test this worm algorithm numerically and
estimate the dynamic exponent z = 0.515(8). We also measure several static
quantities of interest, including loop-length and face-size moments. It appears
numerically that the face-size moments are governed by the magnetic dimension
for percolation.Comment: 31 pages, 10 figures. Several new figures added, and some minor typos
corrected. Uses the following latex packages: algorithm.sty, algorithmic.sty,
elsart-num.bst, elsart1p.cls, elsart.cl