Abstract
In this chapter, we consider the most general convex optimization problems, which are formed by non-differentiable convex functions. We start by studying the main properties of these functions and the definition of subgradients, which are the main directions used in the corresponding optimization schemes. We also prove the necessary facts from Convex Analysis, including different variants of Minimax Theorems. After that, we establish the lower complexity bounds and prove the convergence rate of the Subgradient Method for constrained and unconstrained optimization problems. This method appears to be optimal uniformly in the dimension of the space of variables. In the next section, we consider other optimization methods, which can work in spaces of moderate dimension (the Method of Centers of Gravity, the Ellipsoid Algorithm). The chapter concludes with a presentation of methods based on a complete piece-wise linear model of the objective function (Kelley’s method, the Level Method).
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
- 2.
In the proof of Theorem 3.1.11, we worked with the â„“ 1-norm. However, the result remains valid for any norm in \(\mathbb {R}^n\), since in finite dimensions all norms are topologically equivalent.
- 3.
As compared with the standard version of this theorem, we replace the continuity assumptions by assumptions on closedness of the epigraphs.
- 4.
- 5.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2018 Springer Nature Switzerland AG
About this chapter
Cite this chapter
Nesterov, Y. (2018). Nonsmooth Convex Optimization. In: Lectures on Convex Optimization. Springer Optimization and Its Applications, vol 137. Springer, Cham. https://doi.org/10.1007/978-3-319-91578-4_3
Download citation
DOI: https://doi.org/10.1007/978-3-319-91578-4_3
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-91577-7
Online ISBN: 978-3-319-91578-4
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)