At the end of your "for" loop, alpha_trial is going to be the last value it was in going through the loop, so it is going to be (alpha + alpha_reduce(B_steps)). Are you sure that is what you want?
Your code could be vectorized as
zeeman = -applied_field .* Msi .* cos(alpha + alpha_reduce);
But after your "for" loop, you do not use the value of zeeman(). So you might as well not calculate it at all.