My first encounter with David Blackwell was as a student in his course on dynamic programming at Berkeley in the fall of 1965. There were, as I recall, about forty or so students from various applied areas, together with a few math types like me. The class met once a week in the evening for about two hours. David always arrived right on time, nattily dressed and sporting a bow tie. He would take a small piece of paper from his shirt pocket, glance at it briefly, and then, with no additional notes, lecture for about an hour. There was then a short break, after which David would look at the other side of the piece of paper before lecturing for the second hour.
The lectures were so clear that the applied students could understand and we math types could easily see that the arguments were airtight. David would often give an intuitive explanation for why a result should be true and then follow it with a rigorous proof. I still have my notes from the course and consult them almost every year to remind myself of an argument or a key example.
David held office hours at 8 AM. Since few students showed up at this early hour, I was able to see him a number of times with questions about dynamic programming and later on about my thesis problem and other matters. These meetings were always fruitful for me. David could always see quickly to the heart of a problem. Sometimes he knew the solution and, if he did not, he always had a good idea about where to look.
My thesis adviserwas a good friend of David’s. Lester liked to work with finitely additive probability measures, and, following his lead, I worked with them, too. David was quite dubious of this because of the nonconstructive nature of purely finitely additive measures. He once remarked that he was impressed by all the interesting results we were able to prove about these measures that do not exist.
On another occasion, whenand I had been working a long time on an obscure measurability problem, we asked David whether he thought our endeavor was worthwhile. He said that when a problem arises naturally in a theory and is difficult to solve, its solution may well require new mathematical tools that will be useful for other purposes as well. Indeed, when, with the aid of Lester Dubins and , we finally found the answer to our problem, it did require new techniques that we were able to apply elsewhere.
David made seminal contributions to mathematical statistics, probability theory, measure theory, and game theory. He also found deep connections between game theory and descriptive set theory. As already suggested, he was a great teacher. His only failing, which I observed while serving on search committees at the University of Minnesota, was that he was too kind to ever write anything but a good letter of recommendation for a job candidate.