Obituary: Andrew R. Conn

By Michael L. Overton

Andrew R. Conn, known to everyone as Andy, was born in London on August 6, 1946, and passed away in Westchester County, NY, on March 14, 2019. He played a pioneering role in nonlinear optimization and remained very active in the field until shortly before his death.

Andy obtained his B.Sc. with honors in mathematics at Imperial College London in 1967 and then moved to Canada, earning a master’s degree in computer science from the University of Manitoba in 1968. Opting to study in one of the world’s coldest places turned out to be an extraordinarily happy choice for Andy because he met Barbara, his wife of nearly 50 years, in Winnipeg. Andy continued his studies at the University of Waterloo and received his Ph.D. from the Department of Applied Analysis and Computer Science in 1971. He completed a yearlong postdoctoral fellowship at the Hebrew University of Jerusalem before returning to Waterloo in 1972 as a faculty member, first in the department that awarded his Ph.D. (later renamed the Department of Computer Science), and then in the Department of Combinatorics and Optimization. In 1990, after 18 years at Waterloo, Andy moved to IBM’s Thomas J. Watson Research Center in Yorktown Heights, NY. The two halves of his research career at these respective institutions were completely different; he detailed his career trajectory in a fascinating article entitled "My Experiences as an Industrial Research Mathematician," which appeared in a previous issue of SIAM News [4].

Andy’s research interests were extensive but centered on nonlinear optimization. His work was typically motivated by algorithms and included convergence analysis, applications, and software. According to Google Scholar, Andy had more than 100 publications with at least 10 citations — and over 14,400 citations in total. His first published paper, “Constrained Optimization Using a Nondifferentiable Penalty Function” [3], appeared in the SIAM Journal on Numerical Analysis in 1973 and was quite influential in paving the way for the widespread use of exact penalty functions in nonlinear programming. Andy’s work with Richard H. Bartels and James W. Sinclair on linear  approximation [1] was another important early-career paper. He had 10 Ph.D. students at the University of Waterloo, including Paul Calamai, Thomas F. Coleman, and Yuying Li — all of whom are currently professors there.

In the late 1980s and 1990s, Andy maintained an extraordinarily productive collaboration with Nicholas I.M. Gould and Philippe L. Toint that resulted in more than 20 publications by these three authors alone. Trust Region Methods [6], the first book in the Mathematical Optimization Society (MOS)-SIAM Series on Optimization, was a highlight of this partnership. Other noteworthy contributions include the LANCELOT software for large-scale constrained optimization [5]—for which Andy, Gould, and Toint received the 1994 Beale – Orchard-Hays Prize for Excellence in Computational Mathematical Programming—and the influential Constrained and Unconstrained Testing Environment (CUTE) [2].

In recent years, Andy became interested in derivative-free optimization (DFO). Together with Katya Scheinberg and Luis Nunes Vicente, he published several papers on this topic and a book titled Introduction to Derivative-Free Optimization [7], for which the trio was awarded the 2015 MOS-SIAM Lagrange Prize in Continuous Optimization. The citation states that the book “includes a groundbreaking trust region framework for convergence that has made DFO both principled and practical.”

Although Andy’s research was predominantly devoted to continuous nonlinear optimization, he had an ongoing interest in applications involving mixed integer nonlinear programming and differential equations. He collaborated with several colleagues at IBM to apply nonlinear optimization to circuit-tuning problems, ultimately developing a system that has significantly impacted IBM’s circuit design. Along with Ruud Haring and Chandu Visweswariah, he won an IBM Outstanding Technical Achievement Award for this work.

Andy was also no stranger to SIAM. He served as chair of the SIAM Activity Group on Optimization from 1992 to 1994, and was named a SIAM Fellow in 2013.

Andy was an extraordinary person, admired and respected by all who knew him. He had a passion for fairness, honesty, and openness, to the extent that he signed all of his referee reports — even the most negative. He loved sharing his knowledge through collaboration, teaching, and conversation; was continually supportive of friends and colleagues; and had a wonderful sense of humor. Lior Horesh, Andy’s manager at IBM, spoke highly of him. “Beyond his remarkable scholastic accomplishments, Andy was a dear friend with great empathy and sense of humor, and a charming personality,” Horesh said. “He was an incredible colleague with extraordinary insights, curiosity, and eagerness to solve complex problems.” Many more tributes to Andy are available on the blog set up by his family.

Andy was diagnosed with melanoma in 2013. Despite his serious illness and aggressive treatments, he remained very brave and amazingly cheerful during the last five years of his life. Andy celebrated his 70th birthday in 2016 with many colleagues at the Workshop on Nonlinear Optimization Algorithms and Industrial Applications, which was held in his honor at the Fields Institute for Research in Mathematical Sciences. Naturally, he attended every talk and viewed most of the posters.

Just last summer, Andy attended the 23rd International Symposium on Mathematical Programming in Bordeaux, France, racing from session to session in his usual fashion and enjoying fine food and wine with friends and colleagues in the evenings. Three months before he died, he gave a wonderful talk at New York University entitled “An \(\ell_1\)-Augmented Lagrangian algorithm and why, at least sometimes, it is a very good idea” — a perfectly Andy-esque title that conveyed content, enthusiasm, and a caveat all at once. Less than two weeks before his passing, he was explaining a math problem to one of his grandchildren with equally energetic enthusiasm.

Andy is survived by his wife Barbara, his mother and four siblings, his daughter Leah, his son Jeremy, and three grandchildren. He will be greatly missed by his family, friends, and the optimization community.

References
[1] Bartels, R.H., Conn, A.R., & Sinclair, J.W. (1978). Minimization Techniques for Piecewise Differentiable Functions: The  Solution to an Overdetermined Linear System. SIAM J. Numer. Anal., 15(2), 224-241.
[2] Bongartz, I., Conn, A.R., Gould, N.I.M., & Toint, P.L. (1995). CUTE: Constrained and unconstrained testing environment. ACM Trans. Math. Soft. (TOMS), 21(1), 123-160. 
[3] Conn, A.R. (1973). Constrained Optimization Using a Nondifferentiable Penalty Function. SIAM J. Numer. Anal., 10(4), 760-784.
[4] Conn, A.R. (2011, May 17). My Experiences as an Industrial Research Mathematician. SIAM News, 44(4), pp. 1-3.
[5] Conn, A.R., Gould, N.I.M., & Toint, P.L. (1992). LANCELOT: A Fortran package for large-scale nonlinear optimization (Release A). In Springer Series in Computational Mathematics. New York, NY: Springer-Verlag.
[6] Conn, A.R., Gould, N.I.M., & Toint, P.L. (2000). Trust-Region Methods. In MOS-SIAM Series on Optimization. Philadelphia, PA: Society for Industrial and Applied Mathematics.
[7] Conn, A.R., Scheinberg, K., & Vicente, L.N. (2009). Introduction to Derivative-Free Optimization. In MOS-SIAM Series on Optimization. Philadelphia, PA: Society for Industrial and Applied Mathematics.

Michael L. Overton is a professor of computer science and mathematics at New York University. He first met Andy Conn in 1979.