Efficient hybrid algorithms for finding zeros of convex functions.
We consider two hybrid algorithms for finding an {var_epsilon}-approximation of a root of a convex real function that is twice differentiable and satisfies a certain growth condition on the interval [0, R]. The first algorithm combines a binary search procedure with Newton's method. The binary search produces an interval contained in the region of quadratic convergence of Newton's method. The computational cost of the binary search, as well as the computational cost of Newton's method, is of order O(log log(R/{var_epsilon})). The second algorithm combines a binary search with the secant method in a similar fashion. This results in a lower overall computational cost when the cost of evaluating the derivative is more than .44042 of the cost of evaluating the function. Our results generalize same recent results of Ye.
- Research Organization:
- Argonne National Lab. (ANL), Argonne, IL (United States)
- Sponsoring Organization:
- ER
- OSTI ID:
- 937620
- Report Number(s):
- ANL/MCS/JA-2435; TRN: US200905%%347
- Journal Information:
- J. Complexity, Vol. 10, Issue 1994
- Country of Publication:
- United States
- Language:
- ENGLISH
Similar Records
A parallel gradient distribution algorithm for large-scale optimization
A constrained load flow using convex structure (modified algorithm)