Sunday, November 1, 2009

Physicists are Everywhere

For those of you who don't really know what goes on in the world of experimental physics, let me tell you about what I've been doing the last couple weeks.

So, I've posted about the KATRIN project before, but here's a recap: KATRIN is a project that aims to directly measure the neutrino mass by examining the beta-decay spectrum of tritium. The specific part I'm working on is the rear section, which is an important calibration piece of the experiment. For the output of the experiment to make sense, we need to know precisely how much tritium there is in the system at any given time. To accomplish this, we use a detector in the back wall to count the electrons that don't make it through the main spectrometer.

Now, I've been examining one potential problem with this picture- one having to do with electron scattering. You see, the beta decay electrons don't just travel around until they hit the detector- they can also scatter by interacting with the tritium in the column. These interactions make the electrons lose energy, and happen with more frequency when the tritium density increases. With less energy, the electrons are then less likely to be counted in the back end, and this introduces a systematic uncertainty into the measurement of the tritium density.

So, to explore the problem, I wrote a monte carlo program that shows the effect that electron scattering actually has. My program first randomly selects where the electron is emitted from, and at which angle. Then the program selects how far it gets before it scatters, and how much energy it loses. If the electron hasn't reached the wall, the process repeats until it does. Then, the program repeats the whole process again until it has simulated 100,000 electrons. Then, I ran my output through my professor's code, which does a similar thing with the detector design we're using. Altogether, these simulations show what we should expect as an output, should we see some variation in the tritium density, and gives a baseline for the systematic uncertainties we should expect. It also gives us some commentary on the feasibility of the detector design.

I have a point, but you'll have to read on.

At the beginning of baseball season every year, you can usually find some article online talking about computer algorithms that predict the outcomes of the coming season. These algorithms calculate how many games a team will win, how many runs they will score, their likelihood of making the playoffs, and much more. The way they do it is actually quite simple (in principle).

These algorithms merely take the rosters of each team, and uses each players' projected stats to calculate the likelihood of every possible outcome of any particular pitcher-batter match-up. Using these stats, the algorithms simulate every game of the season, one plate appearance at a time. They then repeat the algorithm about a thousand times to minimize uncertainties in random variations.

Hopefully you're starting to see a pattern. Here's another example:

Last fall, I watched a few episodes of a show called "Deadliest Warrior". The premise of the show is that they explore and analyze the weapons and fighting techniques of some of history's most famous warriors, and try to answer the question of who was more lethal. It's the perfect show for any guy who has ever sat around drinking with his buddies and asked the question, "Hey, if a samurai and a viking ever fought to the death, who would win?" The answer, of course- the samurai.

Anyway, so here's how the show went about answering this question: First, they invited experts and martial artists to showcase the weapons and techniques of each side, running tests on each weapon to determine its killing ability. Then, they put the data into a computer program that simulated the battle, blow by blow, a thousand times and tallied the wins for each side.

I'm not an expert in financial markets, but I'm pretty sure they do something similar there as well.

So, maybe my point is this: The work that goes on in the Physics world doesn't need to seem so distant and scary. There are many people out there who do very similar work in fields that are much more accessible to the general public.

As a matter of fact, my job is much easier than those that I highlighted earlier. My program is about 100 lines of Python code. Sports and battle simulations are much harder to do.

For example, those computer algorithms seem to predict the Yankees winning the world series every year. This year is the first since 2000 that they've been right.

...And I think we know what kind of track record the banking industry has...

1 comment:

  1. Mortgage traders use monte carlo simulation in projecting interest rate "paths" in order to derive a kind of "average" value for mortgage portfolios. This measure--often called Option Adjusted Spread--attempts to account for all the prepayment scenarios for each possible interest rate path.