It is more than four years since two of Anirban DasGupta’s close colleagues, Jayanta Ghosh and Herman Rubin, passed away. He knew both well, beyond the level of being “just” colleagues, and he shares a few cherished personal memories that we hope will be of interest to readers:
I first met Professor Ghosh in 1975 when he returned from Pittsburgh and taught my class a methods course out of Mood and Graybill. I quickly realized that his assignments and exam problems were atypical, not bookish problems. He assigned a very small number of problems that forced us to think. He was an extraordinarily generous grader. If I understood an epsilon amount on some problem, he would give me full credit. If I deserved a C, he would give me an A+. At that point, I just knew him from a distance. I started to get to know him personally when he first started to visit Purdue for one semester each year, toward the end of his Directorship at the ISI. We dined together frequently. He was not fussy, at all, about what he would eat. In the early ’90s, when I was a more active cook, I often made a Parsi dish called dhansak, a dish combining various lentils, leafy vegetables and potatoes, and some meat, chicken or lamb. Professor Ghosh loved dhansak so much that when Rabi Bhattacharya came to visit him, he introduced me by saying that I make a great Parsi dish. When the 6.9 earthquake struck San Francisco in 1989, and we were watching it together, he was so worried about the safety and wellbeing of Peter Bickel, and, as I recall, he called Peter the next day. I saw that tender and soft side again when he got tearful in front of me when Ashim Mallik passed away.
I remember at least two occasions when he kept me from doing stupid things. In the mid-90s, I had done some work on proposing a new default prior methodology. I was very pleased with myself at the mathematical scope of the proposal. But at the end, the prior had to be found by solving a variational calculus problem that was hard to solve explicitly. Professor Ghosh read that manuscript and advised me, rightly, that it would not be well received. What good is a prior that cannot be computed, he said. I realized that the work was no good. I also recall that around the same time I wanted to give a test for whether a distribution on the line has a finite mean. I consulted Andrew Rukhin and Professor Ghosh about it. Of course, it could not be tested consistently against general alternatives. I remember an evening phone conversation when I pressed him saying that it should be possible to find tests that work well in many cases. He told me that one can always do something on any problem. Again, I dropped that problem, rightly, on his advice. It was not easy to make a friendship with him, but I was lucky that we developed one in the 1990s. He made several contributions to the Purdue statistics department, not the least of which was to teach a PhD level course on Bayesian Nonparametrics out of his book with R.V. Ramamoorthi. He supervised the dissertations of numerous PhD students at Purdue and was a seasoned leader in writing survey articles. One example is the survey of sequential design and allocation with Arup Bose and Atasi Basu. His office used to be the office of Shanti Gupta and Mary Ellen Bock. Undeservedly, the department Head Hao Zhang gave me his office after he passed away. Now, dear reader, I assure you that my desk is a lot cleaner than his was.
I did not know Herman Rubin personally before I joined Purdue. I knew him only through some his fundamental papers. I was incredibly fortunate that I became professionally close to him within a very short time. Borrowing a line from Jim Berger, I quickly started using Herman as my “intellectual filter.” I realized very soon that two questions about Herman are unanswerable: what does Herman Rubin not know, and which problems can Herman Rubin not solve? I found it astonishing that when a complex calculation was needed, he rarely came to the board and wrote. He was an inexplicable master of doing calculations in his head, most of the time correctly, and with unfathomable speed.
Herman was prodigiously unparalleled in many other ways. Mohan Delampady told me that in the ’80s, Herman came to work at 6:30 AM. He used to walk around the departmental floor twice a day. He would walk into the offices of a few specific faculty members or those of random graduate students. He started his conversation by saying, “By the way…”. It was always about a math problem. He would stay for about 35 minutes, and most times I would learn something new, either a clever technique or a theorem.
Herman had unorthodox views on mundane things. He viewed cooking as a decision theory problem: what is cooking except a set of optimal actions? Herman and Jean Rubin had me over for Thanksgiving dinner several times: the turkey was always cooked by Herman. Herman often confided in me. On the paper with Herman Chernoff on estimating an unknown discontinuity point, Herman said to me that he thought most of the work was done by Chernoff. I had concluded that Herman admired Gauss more than Euler. He had tremendous respect, among others, for Paul Halmos, Charles Stein, David Blackwell, Jerzy Neyman, Paul Lévy, Larry Brown and Lucien Le Cam, and some others that are still living. One of them was C.R. Rao, who gave the first Pillai memorial lecture of the Purdue statistics department, about ℒp inequalities on eigenvalues. I was a witness to the interaction between two masters at that lecture, one giving the seminar and the other sitting in the audience.
Herman was almost incapable of surviving without being engaged in mathematics. When he had his first spinal stenosis surgery, our systems director Doug Crabill went to the hospital and made special wifi arrangements so Herman could answer the accumulated math questions.
Herman was a blunt truth-teller. I was promoted to associate professorship in 1989. I asked the department in 1992, through Jim Berger, that I be put up for promotion to full professorship. I now realize that my proposition was premature. Jim explained to me very patiently after the primary committee meeting why it was risky to do it that year. I recall going to Herman’s office some days after that and tell him in some way that I was hoping for the other decision. Herman was doing something on his tiny little computer. He looked at me and said, “Oh yes. I voted against you.” Shanti Gupta told me the next year, in 1993, that Herman voted for me this time.
If I hadn’t come to Purdue, I would have never known Herman so closely: it would have been an inestimable loss.
You can read Anirban’s obituaries of Professors J.K. Ghosh and Herman Rubin, which were printed in the December 2017 and June/July 2018 IMS Bulletins, respectively, at https://imstat.org/2017/11/16/obituary-jayanta-kumar-ghosh-1937-2017/ and https://imstat.org/2018/05/15/obituary-herman-rubin-1926-2018/.