Unconstrained real valued optimization based on stochastic differential equations

A new approach of finding optimal variables to continuous, real valued functions is outlined. The technique is simple to implement, and does not require multiple instances of the variables or knowledge of the cost function derivative. The variable update is governed by the numerical solution to a specific stochastic differential equation. A number of parameters can be changed to alter the manner in which the variables are updated. Different parameter choices lead to changes in the rate and manner of convergence to the optimal solution. Test functions are used to establish method performance. Results show that the method is competitive with established approaches.