In this paper,we use the gradient descent algorithm with adaptive search to study the low rank solution of a class of minimization problem on Stiefel manifold which arises in the atomic chemistry.The eigenvalues of the matrix in the objective function are estimated firstly.Then by constructing the Lagrangian function of the equivalent problem,a necessary condition for the local minimum of the problem is given.Finally,comparing with the existed algorithm numerical implementations show the efficiency of the new algorithm.