The night shift was my lot in life, and I didn’t mind. For some months in the spring and summer of 1952, I rented a basement room just off Connecticut Avenue, not far from the Bureau of Standards, wrestled with the SEAC at night, and slept part of the day. (I had to keep a notepad by the phone, which was next to my head. Had I not scribbled a note after every call that woke me, I would have gone right back to sleep without remembering a thing about the call when I got up in the afternoon—even though, to the caller, I probably sounded wide awake.) I hired a young man with no scientific credentials as an assistant. His job was to operate the computer and keep track of its input paper tapes and its output pages printed by a teletype machine. I trained him to convert the computer’s hexadecimal (base-16) output into decimal numbers so that we could summarize our results in graphs and tables. He had no clearance and at best (I hope) a very hazy idea of what we were doing.
The purpose of the SEAC calculations was simple: Follow the thermonuclear burning and the related fission for “steady-state” flame propagation in the Mike device and calculate Mike’s total energy release (its “yield”). A few “runs” had to be duplicated by hand to make sure that the computer was doing what we thought it was doing. (The time is now long past when checking a computer’s work by hand is possible.) Also most runs had to be repeated to be sure that two calculations with identical input gave identical answers. And we had to do run after run after run with altered input data, altered assumptions about the physical processes, and altered calculational approximations.
Even though the SEAC was, at the time, probably the best computer in the world, its limitations still dictated a calculational approach that was highly simplified. We assumed that the fission trigger had done its job, creating a bath of super-hot radiation; that this radiation had compressed the deuterium (and its central “sparkplug”) to a targeted compression; that thermonuclear burning had been initiated; and that a flame was now spreading through the material. These starting “assumptions” were really the end points of other calculations—calculations that were intended to assess the early stages of the explosion. Some of these were carried out in Los Alamos, some at Matterhorn. My SEAC calculations were designed to find out if the flame, once started, would continue to propagate through the material (in a “steady state”) or would fizzle, and, if it didn’t fizzle, what total energy would be released.
The calculations that I labored over night after night and the equations and input assumptions that I tinkered with on many a day, when added to all of the other theoretical and calculational efforts being conducted by a very small army of physicists and mathematicians (no more than a few dozen of them), convinced us finally that Mike would be a success in the multi-megaton range. I save our final predicted number and its comparison with the actual yield for the next chapter.
As I mentioned at the end of Chapter 11, when I got back to Princeton from Los Alamos in 1951, I did what other young people tend to do with accumulated money: I spent it. The two-seat British Singer Roadster that I bought had the look and feel, if not exactly the power, of a genuine sports car. It “cornered” very nicely, and it had a feature especially suitable to Washington’s hot, humid summer. Not only did the canvas top come down, but also the windshield folded forward. I could speed about the streets of Washington with the wind in my face. One girlfriend at the time liked to ride in the Singer crying out “faster, faster.” We both survived and she married a doctor who drove a sedan.
The Singer was handy for the frequent trips that I had to make back to Princeton. (The Carryall was relegated to the status of “second car.”) John Wheeler was the unquestioned leader of the physics we were exploring, and I needed to check in with him often. Occasionally he came to Washington, but mostly I went to Princeton. John Toll was also heavily involved. He and I assisted Wheeler in developing the equations that we wanted to use. Toll also worked with me in writing programs for the SEAC. Wheeler took no direct role in the programming but was keenly interested in whatever approximations we had to make.
The SEAC was closely modeled after John von Neumann’s MANIAC. (Indeed every computer since owes some features of its architecture and operation to that progenitor machine.{1}) The main difference was that the SEAC had two parallel memory banks, whereas the MANIAC had only one. The MANIAC’s memory consisted of what were called Williams tubes[79]—forty of them, These were cathode ray tubes in which an electron beam scattered charge away from spots on the screen (a 32 × 32 array of 1024 spots) and then could determine by a complex dance that included a redirected electron beam and an adjacent metal plate whether or not that spot’s charge had been scattered—thus whether it represented a zero or a one. (And every time a spot was read, it had to be instantly rewritten.) This memory was fast but it was balky. The electrons didn’t always do what they were supposed to do.
A little arithmetic will tell you that the MANIAC’S memory amounted to about 40,000 bits, or about 5,000 bytes. Now, for around ten dollars, you can purchase a “thumb drive” with a memory capacity greater than that of a million MANIACs.
The SEAC also had a bank of Williams tubes (with about half the memory capacity of the MANIAC) and in addition a set of what are called mercury delay lines. Each delay line looks a bit like a fluorescent tube, two feet long and a little more than an inch in diameter. It doesn’t emit light. Instead it sends little acoustic pulses from one end to the other and then feeds these pulses electrically back to the starting point. So zeros and ones are constantly running along the tubes: 360 of these bits chasing each other down each tube. As they emerge from the ends of the tubes, they can be “read” and copied into the computational part of the computer. For better or worse, the two kinds of memory could not be used at the same time. There was a simple switch on the SEAC: Williams tubes in one position, mercury delay lines in the other position. We usually had the switch in the delay-line position, to assure more reliable, albeit slower, operation. Every once in a while we would throw the switch to the Williams tube position and hope for the best.
Every reader of this book is likely to use a computer—desktop, laptop, tablet, or smart phone. I will therefore take a little more space to describe features of the SEAC, to contrast it with modern computers.{3} (I gained a true affection for the SEAC, which was, after all, my nighttime companion for the many weeks of a long, hot summer.) It had 512 “words” of memory. Each word, 45 bits in length, could accommodate either one number of up to 13 digits or one command. In practice, the large majority of the words were given over to commands. It was normally sufficient to store no more than a few dozen numbers. The SEAC—unlike nearly all computers since—was a “four-address” machine. This means that a single command line might contain instructions such as the following: Take a number from address A and a number from address B, combine them in some way or compare them, send the result to address C, and then go to address D to get your next command. (Forty-five bits is enough for all of this.) Needless to say, programming was in “machine language.” I would sit down with a few sheets of paper containing 512 numbered lines, and write the commands and the designated numerical storage locations on the numbered lines.
