15th EUROPT Workshop on Advances in Continuous Optimization

Montréal, Canada, 12 — 14 July 2017

15th EUROPT Workshop on Advances in Continuous Optimization

Montréal, Canada, 12 — 14 July 2017

Schedule Authors My Schedule

In Memory of Roger Fletcher: Large-Scale Optimization

Jul 13, 2017 09:45 AM – 11:00 AM

Location: Amphithéâtre Banque Nationale

Chaired by Sven Leyffer

3 Presentations

  • 09:45 AM - 10:10 AM

    Novel update techniques for the revised simplex method

    • Julian Hall, presenter, University of Edinburgh
    • Qi Huangfu, FICO

    The development of techniques for creating and updating the invertible representation of the simplex basis matrix were a feature of my PhD studies under Roger in 1987-90. So it is appropriate that this talk presents recent recent work in this area. Specifically, it will describe two variants of the product form update and the multiple Forrest-Tomlin update which are valuable in the context of a recently released high performance revised simplex solver.

  • 10:10 AM - 10:35 AM

    On the behavior of a Class of Symmetric Conjugate Gradient Methods for Large-Scale Optimization

    • Mehiddin Al-Baali, presenter, Sultan Qaboos University

    A recent class of symmetric conjugate gradient methods for large-scale unconstrained optimization will be considered. The class, with any line search, suggests strategies for enforcing the sufficient descent and other useful properties. Numerical results on a set of standard test problems will be described. They show that the proposed symmetrical-strategy improves the performance of several conjugate gradient methods substantially (in particular, those of the well-known Fletcher-Reeves, Polak-Ribi\'{e}re and Hestenes-Stiefel methods).

  • 10:35 AM - 11:00 AM

    Nonlinear optimization problems on the sphere

    • Robert Womersley, presenter, University of New South Wales

    The problem of placing N points uniformly on the unit sphere has many different solutions depending on the particular objective: Riesz s-energy, separation/packing, covering, numerical integration (including spherical t-designs) and approximation. Different parametrizations of the variables, either as points in Euclidean space, spherical coordinates or positive semi-definite matrices of inner products with a rank condition, each have their own advantages and disadvantages. The aim is to have many (millions) points, and a global minimizer, when these highly nonlinear problems are characterized by many different local minimizers. Here there are applications for a wide range of efficient scalable nonlinear optimzation algorithms - areas that greatly benefited from Roger Fletcher's pioneering work.