Dimitri Bertsekas, whose foundational work in optimization, control theory, and reinforcement learning shaped the intellectual architecture of modern artificial intelligence, died on June 3 at his home in Belmont, Massachusetts. He was 83. A professor emeritus at MIT and a full-time faculty member at Arizona State University, Bertsekas leaves behind a legacy measured not only in the 20-plus books he authored or co-authored, but in the generations of researchers, engineers, and practitioners who learned their craft from his lucid, mathematically rigorous expositions.
From MIT to Arizona State: A Career Defined by Intellectual Range
Bertsekas earned his PhD in system science from MIT in 1971, after undergraduate studies at the National Technical University of Athens and a master’s degree at George Washington University. He taught at Stanford and the University of Illinois before returning to MIT in 1979, where he remained until 2019. He then joined Arizona State University as the Fulton Professor of Computational Decision Making. His research spanned optimization, large-scale computation, reinforcement learning, dynamic programming, and artificial intelligence — fields he helped define and unify through his writing.
The Books That Shaped a Discipline
Bertsekas’ most enduring contribution may be his books. Titles such as Dynamic Programming and Stochastic Control, Neuro-Dynamic Programming (co-authored with John Tsitsiklis), Data Networks (co-authored with Robert Gallager), and Parallel and Distributed Computation became standard textbooks at MIT and beyond. Stephen Boyd, Samsung Professor at Stanford, described how generations of researchers learned optimization from Bertsekas’ exquisitely clear books, adding that he entered the field himself in no small part because of them.
The clarity of Bertsekas’ explanations was legendary. Colleagues noted that his writing felt like hearing him speak directly to the reader. He combined mathematical precision with an intuitive grasp of complex ideas, organizing entire subjects into coherent, understandable wholes. His 2025 essay “Academia, Art, and Life” explored his belief that research is a creative form combining craftsmanship and art — a philosophy he practiced daily.
Mentorship as a Defining Force
Bertsekas’ influence extended through the many students and young researchers he mentored. Asu Ozdaglar, now department head of EECS at MIT, recalled that taking his nonlinear optimization class changed the direction of her career. Jinane Abounadi, now executive director of the MIT Sandbox Innovation Fund Program, described learning optimization and dynamic programming from a true master. Steven E. Shreve, now professor emeritus at Carnegie Mellon, remembered how Bertsekas gave him a draft manuscript to proofread as a form of one-on-one instruction — a gambit that led to a lifelong collaboration and their co-authored book Stochastic Optimal Control: The Discrete Time Case.
John Tsitsiklis, his co-author and close collaborator, wrote that for Bertsekas, research was about discovering meaning — uncovering the right way to view a subject and conveying it in crystal-clear prose. Tsitsiklis described their joint work on Neuro-Dynamic Programming as a tour de force that anticipated and helped define ideas central to reinforcement learning and approximate dynamic programming.
Recognition and Legacy
Bertsekas received numerous honors, including the 2018 INFORMS John von Neumann Theory Prize (with Tsitsiklis), the 2014 Richard E. Bellman Control Heritage Award, the 2015 George B. Dantzig Prize, and the 2022 IEEE Control Systems Award. He was elected to the U.S. National Academy of Engineering in 2001. He also founded Athena Scientific, a publishing company, and served as chief scientific advisor to Bayforest Technologies.
Beyond the awards, those who knew him remember a humble, warm, and generous man. Angelia Nedich, a former student who became his colleague at Arizona State, recalled enjoying simple moments — a good coffee, a flavorful meal, a glass of margarita on road trips in the Southwest. Benjamin Van Roy, now a professor at Stanford, described Bertsekas as a treasure to humanity, a Renaissance man, and a phenomenal role model. Yuchao Li, a postdoc mentored by Bertsekas at Arizona State, remembered him as a loving father figure full of infinite wisdom, always eager to help at the slightest sign of difficulty.
What This Means for the AI and Optimization Community
Bertsekas’ work provides the theoretical backbone for much of today’s reinforcement learning and optimization-driven AI. For practitioners and researchers, his books remain the definitive reference for understanding dynamic programming, neuro-dynamic programming, and stochastic control. Anyone working in modern AI — particularly in areas such as reinforcement learning, control, and large-scale optimization — will benefit from studying his clear, rigorous treatments of these subjects. His legacy is not just in the algorithms we use, but in the standards of clarity and intellectual honesty he set for the entire field.
Bertsekas is survived by his wife Joanna, his son Telis, and three grandchildren. He was preceded in death by his son Costas.
