Lev Mezheritskiy Bregman (1941–2023) was a Soviet-born Israeli mathematician whose work provides the hidden scaffolding for much of modern data science, optimization, and information theory. While his name may not be a household word among the general public, his mathematical constructs—most notably Bregman Divergence—are foundational to the algorithms that power machine learning, image processing, and economic modeling today.
1. Biography: From Leningrad to the Negev
Lev Bregman was born on January 25, 1941, in Leningrad (now St. Petersburg), USSR. He grew up in a period of intense scientific rigor in the Soviet Union, coming of age during the "Golden Age" of Soviet mathematics.
Education and Early Career:
Bregman attended the Faculty of Mathematics and Mechanics at Leningrad State University, one of the world's premier institutions for mathematical research. He earned his candidate’s degree (the Soviet equivalent of a PhD) in the mid-1960s. His early research was deeply influenced by the Leningrad school of optimization and functional analysis, which was then grappling with the practicalities of linear and convex programming for industrial and economic planning.
The Israeli Chapter:
In 1991, following the collapse of the Soviet Union, Bregman immigrated to Israel as part of the massive wave of Soviet Jewish "refugee scientists." He joined the Department of Mathematics at Ben-Gurion University of the Negev (BGU) in Beer-Sheva. He spent the remainder of his career there, rising to the rank of Professor and becoming a vital member of the Israeli mathematical community until his death in February 2023.
2. Major Contributions: The Geometry of Distance
Bregman’s legacy rests on two pillars: one in the realm of optimization and the other in combinatorics.
A. Bregman Divergence (Bregman Distance)
In 1967, Bregman published a paper that introduced a new way to measure the "distance" between two points based on a convex function. Unlike the standard Euclidean distance (the "straight line" we learn in school), a Bregman Divergence is not necessarily symmetric. It measures how much a convex function deviates from its linear approximation.
This concept proved revolutionary because many common "distances" in science—such as Kullback-Leibler divergence (used in information theory) and Itakura-Saito distance (used in audio processing)—were discovered to be specific types of Bregman Divergences.
B. Bregman’s Theorem (The Minc Conjecture)
In 1973, Bregman solved a famous problem in matrix theory known as the Minc Conjecture. The problem concerned the "permanent" of a (0,1)-matrix (a matrix consisting only of zeros and ones). The permanent is similar to a determinant but much harder to calculate. Bregman provided a sharp upper bound for the permanent of these matrices, a result now known as Bregman’s Theorem. This discovery had profound implications for graph theory and combinatorics, particularly in counting perfect matchings in bipartite graphs.
3. Notable Publications
Bregman was a proponent of "quality over quantity," authoring papers that were often brief but mathematically dense and conceptually transformative.
- "The relaxation method of finding the common point of convex sets and its application to the solution of problems in convex programming" (1967): Published in the USSR Computational Mathematics and Mathematical Physics journal. This is his most cited work, introducing the Bregman Divergence.
- "Some properties of determinants of matrices with non-negative elements" (1973): Published in Doklady Akademii Nauk SSSR. This paper contained the proof of the Minc Conjecture.
- "Relatively additive convex functions and their applications" (2001): A later work co-authored with Yair Censor, refining the use of his divergence measures in practical optimization.
4. Awards and Recognition
While Bregman lived a relatively quiet academic life, the mathematical community recognized the magnitude of his work through the "naming" of his discoveries. In mathematics, having a "Divergence," a "Method," and a "Theorem" named after you is the ultimate indicator of lasting influence.
He was a frequent keynote speaker at international conferences on optimization and was highly revered at Ben-Gurion University, where he was honored for his contributions to both the university’s prestige and the integration of immigrant scientists into the Israeli workforce.
5. Impact and Legacy: The Engine of Machine Learning
Bregman’s work from the 1960s lay dormant for decades before becoming a cornerstone of 21st-century technology:
- Machine Learning & AI: Bregman Divergences are the standard tool for designing "loss functions"—the mathematical yardstick that tells an AI how far its guess is from the truth. The popular Clustering algorithm (k-means) has been generalized using Bregman distances to work on diverse types of data.
- Mirror Descent: His work is the basis for "Mirror Descent" algorithms, which are used to solve large-scale optimization problems where standard gradient descent fails.
- Image Reconstruction: In medical imaging (like CT scans) and astronomical imaging, "Bregman Iterations" are used to clean up noise and reconstruct sharp images from sparse data.
6. Collaborations
Throughout his career, Bregman was a bridge between the rigorous Russian school of analysis and the burgeoning Israeli high-tech and academic sectors.
- Yair Censor: Perhaps his most significant collaborator at Ben-Gurion University. Together, they explored "Projections onto Convex Sets" (POCS), which has applications in radiotherapy treatment planning and signal processing.
- The Leningrad School: In his early years, he worked closely with colleagues at the Research Institute of Mathematics and Mechanics, contributing to the Soviet efforts in cybernetics and economic optimization.
7. Lesser-Known Facts
- The "Delayed" Fame: For nearly 20 years, Bregman’s 1967 paper was rarely cited outside the Soviet Union. It wasn't until the 1980s and 90s, as computer science began to require more sophisticated optimization tools, that the global scientific community realized he had already provided the necessary framework.
- A Modest Scholar: Colleagues often described Bregman as a deeply humble man who was more interested in the elegance of a proof than in academic politics or fame. He was known for his "mathematical intuition," often seeing the geometric solution to a problem long before the algebra was written out.
- Permanent vs. Determinant: His proof of the Minc Conjecture is often taught in advanced combinatorics courses as a masterclass in using entropy-like arguments to solve problems that seem to have nothing to do with entropy.
Lev Bregman’s life spanned two countries and two eras of mathematics. He began by solving theoretical problems in the Soviet Union and ended as a foundational figure in the global digital revolution, proving that pure mathematical thought is truly timeless.