Recollections by Richard Bleil
Today was pretty productive. For the first time since before the holidays, I’ve cleaned off my tables and can actually use them, the result of a deep and dark depression that has lasted for months. As per usual, my house has been reflecting my mental state of mind, meaning it is in chaos. One of the biggest issues I face is lack of confidence. Living alone and isolated as I do, there is nobody here to tell me that they have confidence in me and believe in my abilities. I can believe in myself for only so long, and it comes and goes in waves. When my self-confidence is low, that’s when I need somebody to tell me that they believe in me, but there is nobody here.
One of the decisions I made was to try to build my own research computer. Back in the ‘90’s, I actually had grants to work on the national Cray supercomputer, but I can’t help but wonder what modern computers can do relative to the Cray. It was the fastest and most powerful computer of its day, but that was also about thirty years ago. Thirty years after the Apollo missions, our cell phones have more computer power than those that landed on the moon.
The question is if I can actually do anything with it. The impetus was a quantum mechanical calculation on my laptop (the one I’m currently writing on). It ran for roughly a month and a half straight when, suddenly, it decided to update and I lost all of the work it had done. And, yes, I know how to turn off automatic updates, but the problem is that laptops and tablets are not designed to run continuously. What I need for work like this is a tower computer, but the question becomes if it would be a waste of money for me to get one.
Ultimately, I decided that I need one if for no other reason than belief in myself. I decided to build my own system because I’m a little bit different. It may sound odd, but most people who want tower systems do not use them for high-level statistical thermodynamics quantum mechanical calculations, but rather they want gaming computers. Gaming systems are often high in hard drive space, have amazing video cards and maybe memory, but not so much CPU power. I need very high CPU speed and power, memory, and disc space but I really don’t care about pretty flashing lights or video card speed. I purchased the components that I needed to build the fastest and most powerful research computer that I could and put it together myself.
When I finally put it together, I apparently did just about everything correctly, which is kind of a surprise considering the number of components involved. Unfortunately, in building it, I apparently bent some pins, and the system won’t boot. Now I’m waiting on a new motherboard to try again.
As I’m waiting for it to arrive, I’m thinking, as you might imagine, about the computer research that I did as a post-doctoral research assistant, which inevitably makes me start to think about the Pivot method. It’s an optimization algorithm that I conceptualized and wrote. The optimization program everybody believed in then (as now) is called the “Tabu Search”. The idea is that, if you have a very complicated mathematical equation and landscape, how do you find the global minimum for that function without getting trapped in a local minimum (or maximum). For example, if the military had a multiple warhead cruise missile, say with five warheads, and say twenty-eight military targets, how can you inflict the most damage?
The mathematical expression that results includes the military value of each target, the possibility that the cruise missile, with all remaining warheads, will be lost and some other considerations. The equation will be very complicated, with a lot of great flightpaths that would inflict considerable damage but not necessarily the most it could be. Algorithms like the Tabu search will find the absolute most destructive path.
My algorithm, Pivot, performed significantly better (about six times better) on every test function that we could find than even Tabu. Unfortunately, as we completed the research for the first paper, wrote it, submitted it and had it accepted for a regular publication, a new post-doc started in our research group who was from the same country as my adviser. Unbeknownst to me, and finding out about it the hard way, my adviser gave my idea to the new post-doc who tried to improve it, wrote a paper, and submitted it as a “rapid letter” publication. Had it been accepted, it would have been published before mine giving him full credit, especially since my name was nowhere on or in the paper. It was an attempt, plain and simple, to steal my work.
Eventually, he did get his work published, this time at least with my paper in the references but not as a co-author. The problem is that he muddied the work by basically lying about its efficacy. The efficiency of these algorithms is determined by “function calls”, that is, how often does it have to use the function. The fewer function calls, the more efficient. He, however, was only counting about half of the function calls. As a result, anybody who tested the algorithm went with his (presumably) more efficient version, but his results were never reproducible. That sunk any recognition of the work for me, for him, and for my dishonest adviser. If I ever get my monster computer running, I’m planning on re-writing and testing the algorithm, and using it in new research based on quantum theory. It’s not uncommon for disreputable research advisers to try underhanded techniques like this. It’s unfortunate that the people actually doing the work don’t get credit when this happens and, in this case, the research is lost altogether because of published lies from within the same research group.