Total suppression of a large spin tunneling barrier in quantum adiabatic computation

We apply a quantum adiabatic evolution algorithm to a combinatorial optimization problem where the cost function depends entirely on the of the number of unit bits in a n-bit string (Hamming weight). The solution of the optimization problem is encoded as a ground state of the problem Hamiltonian H_p for the z-projection of a total spin-n/2. We show that tunneling barriers for the total spin can be completely suppressed during the algorithm if the initial Hamiltonian has its ground state extended in the space of the z-projections of the spin. This suppression takes place even if the cost function has deep and well separated local minima. We provide an intuitive picture for this effect and show that it guarantees the polynomial complexity of the algorithm in a very broad class of cost functions. We suggest a simple example of the Hamiltonian for the adiabatic evolution: H(tau) = (1-tau) hat S_{x}^{2} + tau H_p, with parameter tau slowly varying in time between 0 and 1. We use WKB analysis for the large spin to estimate the minimum energy gap between the two lowest adiabatic eigenvalues of H(tau).