Abstract
In considering the nondifferentiable optimization problem, a new concept is introduced, known as the ball gradient. The ball-gradient magnitude is positive at any local minimum point, independently of whether the minimum point is well behaved, a cusp, or a sheet minimum. The ball-gradient magnitude is negative outside a ball of radius ε around a local minimum point and thus is usable as a terminating criterion on nondifferentiable functions.
Similar content being viewed by others
References
Dixon, L. C. W.,Nonlinear Optimization: A Survey of the State of the Art, Software for Numerical Mathematics, Edited by D. Evans, Academic Press, London, England, 1974.
Wolfe, P. A Method of Conjugate Subgradients for Minimizing Nondifferentiable Functions, Nondifferentiable Optimization, Edited by M. L. Balinski and P. Wolfe, North-Holland Publishing Company, Amsterdam, Holland, 1975.
Demjanov, V. J.,Algorithms for Some Minimax Problems, Journal of Computer and Systems Science, Vol. 2, pp. 342–380, 1968.
Shor, N. Z.,Generalized Gradient Methods of Minimization and Their Use in Problems of Mathematical Programming, Mathematical Methods, Vol. 12, No. 2, 1976.
Goldstein, A. A.,Optimization of Lipschitz Continuous Functions, Mathematical Programming, Vol. 13, No. 1, 1977.
Author information
Authors and Affiliations
Additional information
Communicated by D. F. Lawden
The author acknowledges many helpful discussions with Dr. M. Gaviano during the redrafting of this report.
Rights and permissions
About this article
Cite this article
Dixon, L.C.W. Reflections on nondifferentiable optimization, part 1, ball gradient. J Optim Theory Appl 32, 123–133 (1980). https://doi.org/10.1007/BF00934719
Issue Date:
DOI: https://doi.org/10.1007/BF00934719