Accelerating the DC algorithm for smooth functions (Q1749446)

The authors introduce the concept of DC-functions as functions, which can be expressed as the difference of two convex functions. They consider minimization problems, the objective function of which is a DC-function. Two new algorithms for solving such problems are proposed. The algorithms accelerate the convergence of the algorithms under an additional condition that the objective function has the so-called Łojasiewicz property. The convergence of the proposed algorithms is proved and the corresponding rate of convergence is analyzed. The methods are applied to problems, which occur in the study of biochemical reaction networks. Numerical tests show that the algorithms are faster than the known methods in both computational time and the number of iterations.