One Flexible Mathematician Covers the Career Spectrum

April 14, 2012

Careers in the Math Sciences
Ross Lippert

A few friends and colleagues have pointed out that my career is uncommon for having spanned all three typical spheres of employment for applied mathematicians: industry, the academy, and national laboratories. "Which is better?" more than one candidate has asked in interviews I've conducted in my current position. Of course, the answer depends on an individual's temperament and its fit to the styles of these areas and their work environments. I hope in this essay to give you a feel for what these environments are like. I feel compelled to begin by stating the obvious, which is that I can give only my own impressions, through my own filters. Other people's impressions will surely disagree with mine in various ways, and I don't mean to offend anyone by being candid about mine.

In chronological order, I have worked at Sandia National Labs, Celera Genomics, the MIT mathematics department, and now D.E. Shaw Research. In this article I compare these environments on a number of axes: How much autonomy does an employee have? What are the main tasks of the job? What is the nature of the funding? Where do the problems come from?

Let's start with a bit about my own temperament. I am a computational dabbler, more in the sense of "flexible" than "shallow," I like to think. Something about watching the math I do on paper play out on a computer thrills me, and it doesn't matter to me much beyond that what the particular subject is. My graduate experience was a struggle to focus, which was only partially successful in that my PhD thesis consisted of three separate bodies of work within numerical analysis, two of them arguably related to each other. One year I was all about wavelets, then I was all about differential geometry in linear algebra problems; after that I moved on to Jordan structures and eigenvalue sensitivity problems (one area that I keep coming back to).

For my first position, I ruled out academics. Being too scattered is a minus for an academic, unless you are Gauss. Also, watching my adviser manage multiple students and their projects while hustling after grants to keep his flock fed, I didn't think that was for me. I do not hustle well.

First Stop: A National Lab

My first postdoctoral job was as a research scientist at Sandia National Labs in Albuquerque, New Mexico. I was initially expected to continue work on one topic in my thesis that had applications to certain solid-state physics applications---a differential geometric approach to block Rayleigh quotient optimization, essentially, which had improved convergence properties. Jobs in labs and industry generally use some specific initial task as a starting point. From there you branch out, finding needs in the organization. I began at Sandia by working toward a better understanding of the scientific application and collaborating with computational physicists to produce a paper on the improvement that these ideas should deliver in theory. In the branching out, I found myself knee deep in a major software re-engineering project intended to demonstrate the theory in a production application. Refactoring and parallelizing a legacy production code was what I did in exchange for a serious "real-world" test of the mathematics.

This underscores a general characteristic of national labs. Within the academic sphere (at least in mathematics), software development, creating a widely usable embodiment of an idea, is not well rewarded by the prevailing system of review and recognition. Your goal is to write just enough to support the numerical test section of your journal article, publish the article, and get cited. The only academics who work hard to release polished software are graduate students, because they don't know any better, and tenured professors developing a side business or scratching an itch. In contrast, my main task at the lab was to turn academic expertise and research ideas into a lasting resource. Creating broad general (often public) computational frameworks is a big deal in the labs. This means doing the extra work and developing the auxiliary skills needed to create or improve a "deliverable" (a word I learned at Sandia). It also means that most work is interdisciplinary, done in collaboration with others. Managers work to facilitate the exposure of their departments and their problems to each other, to see if something results. The funding structure of the national labs is tied to recognized projects, and employee time is allocated fractionally across the budgets of such projects, which are ultimately funded somehow by the U.S. government. This gives you a moderate amount of autonomy as long as you are doing something that contributes to at least one of the many ongoing agendas. The national labs are not a bad arrangement. Many people view them as roughly a mid-point between academics and industry along several axes, and I think there is truth to this.

Of course, there were some negatives, mostly centered on funding. While good managers facilitate collaboration, bad managers can be territorial about projects and the funding they bring. National funding shifts with national priorities (as determined by Congress and the executive branch), which can cause lurches in project priorities. The powers that be got excited about projects related to safe nuclear waste disposal one year, defense against biological weapons another year. I got the impression that a function of the lab's bureaucracy was to buffer these shifts. As in academia, a bit of hustle goes on at labs, but it is mostly at the management level, to keep streams of money coming into the projects. Public funding also entails various conditions of oversight, to avoid impropriety, which apply not only to serious concerns about foreign espionage but also to what websites you can view from your office.Though it is not uniformly required, I had to sign a statement consenting to restrict my outside associations to obtain my security clearance. The scrutiny can be off-putting.

Initial Foray into Industry
After a couple of years, a colleague left the lab for a management position in research and development at an exciting small company (about 300 employees), Celera Genomics. Celera, led by the renegade biologist Craig Venter, was then competing with the Human Genome Project to deliver a finished sequence of the human genome. My colleague initiated a small brain drain from our department of which I was a part. I had not done any major work in computational biology, but there were a few analogies with computational linear algebra that made me think I might have something to contribute. When I arrived, the infrastructure for genome assembly was in place, though the final results were not complete; the research emphasis was shifting from assembly and verification toward analysis and search. I started out as "the guy who makes any code run 10x faster," mostly by applying blocking techniques common in linear algebra to various matching problems. There were also a few linear algebra heuristics for combinatorial problems that proved useful and robust. While justifying my employment in these ways, I was learning computational biology from my colleagues. I became "the guy who understands suffix trees," and later a few other "guys" too, but my point is that switching from what you know to something you are willing to learn is both possible and fun. I was not the only person with a fairly unrelated background, or even the only one coming from computational linear algebra or the physical sciences.

The most distinctive feature of the industrial environment was the concentration of activity around a narrow range of goals. Regardless of where anyone came from, everyone was working on the same project.Those who did come from computational biology were superstars in the field, and we learned from them by osmosis. Expertise was usually a shout over the cubicle wall away. I did have to put aside my previous research interests for something new, but I learned a lot in the process. As in national labs, you must translate academic expertise into some sort of deliverable, though some of your customers (possibly the only ones) are just down the hall. Because you always have customers within the organization, your manager can keep track of how you spend your time without as much regimentation and paperwork. There was also the opportunity to translate that research into journal articles, though that is not the main objective. The shared goals, expertise, and camaraderie made it a very exciting place to be, as did working on one of the really cool problems of the time.

(A concern I often hear about industry with respect to academic publication is whether a firm "owns your brain," i.e., whether your intellectual and academic life is entirely controlled by your employer under the terms of some obnoxious intellectual property contract. I have had no problem, either at Celera or in my current industrial position, publishing on subjects entirely outside the company's agenda (for example, a linear algebra article) or maintaining ongoing collaborations on my own time. Even where such contracts exist, I doubt they are worth enforcing against actions that do no damage to the company.)

Such a situation is great as long as it lasts. The downside is that it doesn't always last. In this case, Craig Venter, the CEO and source of a great deal of the company's inspiration, was ousted in a bit of upper-level business politics that I still do not really grasp. Culturally, the transition to new management was like passing from day into night. Though it wasn't as bad as Dilbert, I began to understand the comic much better. Directives from on high turned like a weathervane according to short-term trends. It was a depressing time.

Eventually, I found that the trick to flourishing under fluctuating priorities and dysfunctional upper management is to ignore your boss to a great degree, and improve your auxiliary skills: learn a new programming language or read a book on advanced string-matching algorithms or write some clever program. Every so often, current priorities align to give you an opportunity to turn one of these side-projects into a contribution, and your boss will be impressed with your foresight. In retrospect, I think that my boss adopted this strategy with his boss. Anyone considering an industrial position should seriously try to assess the stability of the research agenda, and try to avoid outfits where you might end up spinning your wheels for nothing.

Meanwhile, it was not a bad time to look at exit options. Some co-workers were jumping ship for a new institute founded by Venter, others for academic positions. I submitted my resume to the Venter institute and my CV to MIT, on the encouragement of my PhD adviser. Venter didn't get back to me, but MIT did. I was offered a limited-term instructorship in the mathematics department. Time to give that a try.

Life in an Academic Math Department
Rejoining academic life, on the faculty side, was a lot like being dropped into the deep end of a pool. After a brief dialog with the department head about the courses I would cover for the term, I was shown my office, given a key, and told where the systems administrator and course secretary were. That was pretty much it, and I was on my own for two and a half years. No starting projects here, which was fine, as I had my own. There were almost no meetings and no group identity. I knocked on doors and made my own collaborations. I sat in on various groups and worked on my own research, on whatever problems I wanted, and learned some new math (a bit of convex analysis and some machine learning). The strange thing about all that autonomy is that I spent much more time working. I took work home far more than ever before (having a relatively open computer network for the first time helped).

This new solitude, however, was balanced by teaching responsibilities. While at MIT, I did two undergraduate classes in numerical analysis and two graduate classes in computational biology, and led several sections of the "intro to differential equations" class. Sections are easy to manage, but putting a good course together takes time. I devoted more than half my time during the first year to planning and administering my lecture classes. This was worth the time: As people say, running a course really is great intellectual exercise. I examined areas of my subjects that had never come up in my experience and looked at familiar areas more deeply, filling in gaps. I also had to ask questions about how the subject was evolving, with a critical eye to topics in need of revision for current students. And there were the interactions with students, who often did surprising things with the course material. The students had a low tolerance for BS answers. I think I got a lot sharper in the subjects I taught.

I had summers off (two in this case), which meant I needed to find a source of funding for the time or see a stiff pay cut as my nine-month salary was spread across twelve. My past collaborations were valuable here. I spent one summer working with folks at Celera creating a novel compressed string index and another at Sandia on an interesting problem in spectral function theory related to quantum chemistry. Three journal articles resulted from that work. Get paid and write papers---that's what summer is for.

There was a disturbingly negative sentiment toward the industrial sector from some people in the department. At a department tea, a senior faculty applied mathematician told me he thought that I had fallen behind for having "wasted time in industry." I was advised by another that if I wanted to stay in the game I would have to get more publications in the "top journals and conferences." Top journals in computational biology, physics, and machine learning do not count in this score, however, and the referee cycle for Linear Algebra and its Applications was tediously slow. (Here I remind you to take into account my bias.) The name of the game is to publish articles in your area, having made many friends in the area who will read those articles.

The academic game is such that the last year of any appointment is best spent exclusively on the search for the next job. To succeed, you should be someone who wants to build a research empire, not someone who prefers getting lost in a puzzle. You need more enthusiasm for social networking and travel than I had. I had enjoyed playing, but I was probably not going to win. Adding to this already bad recipe for an academic job search, my wife was not in a position to move very far from the New England area. So, contrary to the normal form, I had a lot of fun my last year, working on interesting problems and papers. I submitted my CV to nearby universities, with the results that you might expect.

An Independent Research Lab
Over that second summer, however, I was contacted by a headhunter who was recruiting for a firm in the New York area that was attempting to revolutionize drug discovery. A former colleague from Sandia had given him my name. His description of the firm sounded like the sort of market-speak I had heard when I was in biotech at Celera. I would learn later that the firm was a privately funded research foundation headed by a pretty brilliant entrepreneur, but at the time, the contact just meant a free trip to New York. Never be so put off by market-speaking headhunters that you miss seeing what lies behind them. The firm was D.E. Shaw Research (DESRES), where a serious, ambitious, high-risk/high-impact project was under way in molecular dynamics and protein folding. Although New York did not count as sufficiently "nearby Boston" for my wife, the salary was enough to make weekend flights back and forth affordable until we were both able to move.

(To clarify, D.E. Shaw and Company is a well-known investment firm, distinct from D.E. Shaw Research. I still have a few close friends who think I am working at a hedge fund, despite my attempts to correct them.)

DESRES is an independent research lab. The emphasis is more on basic scientific research and less on bringing products to market or pursuing a profit. Still, it has many of the characteristics of my previous industry job, with a highly concentrated population of smart people from various fields working in a highly collaborative fashion on a relatively narrow goal. A great many of the researchers here, like me, have come from other fields (though I did have experience in related areas) and have retrained and refocused their interests along the lines of the project. Technical ideas are rendered into complete software and hardware deliverables, to be used mainly by our own researchers and collaborators, but also in part to be released freely to the academic public; the work requires academic expertise as well as a range of auxiliary skills. Academic publication (on computational topics) is a side goal, intended to explain technical innovations and to develop our reputation; because quality is paramount, the level of output is relatively low.

After five years, there has yet to be a "Dilbert transition." The leadership of the organization has been consistent, and growth has been slow enough to weaken any encroaching bureaucracy. I'm always on the lookout, though.


To offer some summary contrasts, I'll return to the questions I initially asked. Academic life definitely gives the most autonomy. In either the national labs or industry, you have a boss of some sort. National labs provide a bit more autonomy than industry, as you can become your own boss by securing funding for your own project, so long as there is buy-in from the powers that be. The main tasks in academic life are teaching, publication, and publicity. The main tasks in nationals labs and industry are similar: producing usable resources with some publication (perhaps with higher emphasis on publication in the labs). In both academia and the labs, you get funding for your work by hustling after grants or other centralized pools of money; the national labs have a bureaucracy that helps you along and even insulates you from funding concerns. Funding in an industrial research position is controlled by a single source with which you sustain an ongoing relationship.

As an academic, you select the problems you work on, while trying to pick those that will interest your peers. In national labs, the problems you work on must fit somewhere into the needs of one of many ongoing projects; a fit with your own research interests is certainly possible. In industry, there is usually one (or very few) projects, and you will probably spend your time learning about someone else's problems.

Sue Minkoff ([email protected]), of the University of Maryland Baltimore County, is the editor of the Careers in the Math Sciences column.

Donate · Contact Us · Site Map · Join SIAM · My Account
Facebook Twitter Youtube linkedin google+