At the end of our meeting this evening, Evan asked a very good question which I didn't immediately understand, and therefore I didn't really give him a good answer. I think I understand the question a little better now, so hopefully I can give a better answer.
Evan's question was basically, "how is it possible to simulate a real-world
system on a computer with any sort of accuracy?" The basic answer is that
"if you can quantify the system (represent it through numbers and
functions), you can simulate it on a computer."
For example, molecules can be represented with point charges that repel each
other attached by springs that pull them together. And you can quantify
both of these. The energy between two point charges is
E = q1 * q2 / r
and the energy of a spring is
E = 1/2 * k * x^2
Voil�. You now have a quantitative representation of a system, and you can
then simulate real-world behavior with it. If you take two really large
lists (male and female) of a 2-element vector containing dominant or
recessive genes, and you randomly pick a vector from each pool, "mate" it
with a probability distribution, then you have a simulation of population
genetics. If you sum together a whole bunch of sine waves of different
frequencies and amplitudes, then you simulate the sound of someone plucking
a guitar string.
Now these simulations will [almost] always be approximations. The trick is
getting the simulation to be "good enough." Point charges and springs is
actually a really poor approximation of a molecule for most purposes, so we
add in a whole bunch of other energy terms, but they're just functions too.
If you try to approximate a guitar string with only 4 sine waves, then it
won't sound very similar. But use 4096 sine waves and nobody will be able
to tell the difference.
Now there are some problems that we are just incapable of representing at
this point. Trying to predict economic behavior fails miserably because we
don't have a good enough set of functions to approximate it. Trying to
simulate the behavior of an entire cell isn't possible because we've got the
functions, but not enough computing power to calculate all of them on a
reasonable time scale.
How do you know if a simulation is accurate? You tune it match real-world
observations. The charges-and-springs model of molecules doesn't give
answers that match observed values, so we add functions until it does match.
Economic models don't predict accurately, so we scrap them. Simulations of
population genetics work really well, so we keep them.
So the basis of scientific computing is to represent everything with
numbers and functions. Max Cohen, the protagonist of "Pi" (a movie I highly
recommend seeing if you take the MPAA with a grain of salt), stated it as
such: "Restate my assumptions: 1. Mathematics is the language of nature. 2.
Everything around us can be represented and understood through numbers. 3.
If you graph these numbers, patterns emerge. Therefore: There are patterns.
Everywhere around us."
I'm going to shift gears just a little, and then come back to my main point.
In my quantum mechanics class, we have what we call "DFM problems." That
stands for "Dead French Mathematician" problems. During the 19th century,
Europeans (many of them French) investigated a whole lot of interesting, but
to them, useless math problems, and generated a lot of interesting but
useless solutions. Riemann sums, Fourier series, Taylor series, Hermite
polynomials, you take your pick. As quantum mechanics was being developed
in the early 20th century, time and time again the physicists and chemists
ran into these problems, and found out that some DFM had solved the problem
80 years before. I was talking to my sister (a math ed. major) a few weeks
ago and mentioned to her that I was learning about group theory. "But
that's totally abstract algebra," she said.
"Yeah, and it's totally the basis of modern physical chemistry. You
mathematicians have no idea how valuable this stuff is." All that useless
math suddenly became very useful. Quantum mechanics owes its development in
large part to people who died even before anyone had thought of the problem.
Computer Science started to emerge in the mid-20th century. And computer
scientists tackled a lot of interesting but (at the time) useless problems.
Like the most efficient method of finding the determinant of a matrix. Or
how to split a signal into component sine waves. Or how to split up a long
series of function evaluations into a lot of little chunks that can each be
done on a separate processor. Or how to search through a very large
"haystack" string and find the closest match to a "needle" string.
Interesting, but useless in and of itself.
Along come the physical scientists who want to represent their real-world
systems quantitatively. The problem is, one of these systems requires you
to find the determiant of a 20000x20000 matrix. Well, some computer
scientist already found the most efficient solution that problem. Someone
else wants to elimiate certain frequencies from a recorded sound. A
computer scientist has already figured out how to solve that problem too.
Someone else needs to compute the value of a function 5,000,000 times. A
computer scientist has already figured out how to split that problem over 50
processors so it runs a lot faster. Someone else wants to find a specific
genetic sequence in the genome of an entire animal -- like 100 nucleic acids
in a sequence of 10,000,000,000. A computer scientist has already written
an algorithm to do just that, even though he never thought of it in those
terms.
So as much as us in the physical sciences look down on the computer
scientists, most of our problems wouldn't be solvable without the work that
the CS majors have already put in to solving problems that they didn't think
would ever have an application. Since they've already done the
computational grunt work, the physical scientists can worry about the
higher-level refinement of their models.
--
Soren Harward <[EMAIL PROTECTED]>
http://theboard.byu.edu/
signature.asc
Description: Digital signature
____________________ BYU Unix Users Group http://uug.byu.edu/ ___________________________________________________________________ List Info: http://uug.byu.edu/cgi-bin/mailman/listinfo/uug-list
