[Home]

Table of contents


Numerical Analysis
Last updated on: FRI MAR 26 IST 2021

Numerical Analysis

Numerical analysis is all about solving equations (approximately) using a computer. Here the term "equation" is used in a very general sense: You'll be given a set up, and some partial information about it, and you have to find some other information based on that. Here are some examples: While all these sound nice, many students feel uncomfortable about solving equations only approximately. For them, numerical analysis seems to be much less useful than the exact techniques learnt in school. Let us first understand why this is not so.

Approximate versus exact

We shall discuss some scenarios where an approximate solution is a desirable thing.

When an exact solution is impossible/difficult to find

Take for example the integral $$ \int_a^b e^{-x^2}\, dx. $$ We need to find this integral for given $a,b$ in many problems of probability, statistics and physics. The high-school technique of first finding an antiderivative $F(x)$ of $e^{-x^2},$ and then computing $F(b)-F(a),$ does not work here. Liouville has proved that $F(x)$ cannot be expressed in terms of any combination of trigonometric, exponential, logarithmic or algebraic functions!

In some situation involving real data, we may get a rather complicated function (e.g., motion of the moon after taking into account the attraction of Jupiter), and we may need to find when that function takes a certain value. (See the moonpos procedure in Astrolib if you are astronomically inclined.) Even finding out if the equation may be solved exactly, is a difficult problem.

When an "exact solution" is not a solution at all!

Suppose that you are trying to make a telescope, and find that the distance between two lenses must be $x$ cm, where $x^2-2 = 0.$

The exact solution is of course $x = \sqrt 2$. But what does this $\sqrt 2$ really mean? It means a positive number whose square is 2. Thus, this "exact solution" is a mere restatement of the original problem! However, $x\approx 1.414$ is of much more use here.

Similarly, if the exact answer to some problem is $\pi$, then you are basically reducing that problem to the problem of finding the ratio of the circumference and diameter of a circle. Many "exact solutions" are just saying that the problem at hand has the same solution as some other problem.

When the exact solution is less reliable than the approximate one

In mathematics, we are fond of certain frequently used functions, like sine, cosine, logarithm, exponential, square root, etc. We are generally happy if we can express the answer to some problem in terms of these favourite functions. For example, most students of mathematics will consider $$ \frac{-10 + \sqrt{10^2-4\times 7\times\frac{2}{3}}}{2\times 7} $$ to be an exact root of the equation $7x^2 + 10x + \frac 23 = 0.$

When we need the exact numerical value, we simplify the expression down to $$ \frac{-10+\sqrt{\frac{244}{3}}}{14}. $$ No approximation so far. Now we have to find $\frac{244}{3},$ which cannot be stored exactly in a computer (recurring). So we approximate it to $81.3333.$ Next we feed this into the square root. Even $\sqrt{81.3333}$ cannot be stored exactly in a computer. So another layer of approximation is needed: $9.0185.$ We have again used a 4-decimal place approximation. Thus, But that does not guarantee that $\sqrt{\frac{244}{3}}$ and $9.0185$ match up to 4 decimal places! Our favourite functions may not behave nicely with approximations!

Next we add the $-10$ (this does not entail any extra approximation). But when you finally divide by 14, yet another layer of approximation is inevitable. So all that we can say about the final answer: $-0.0701$ is that it may contain an error of unknown order of magnitude!

A numerical method will not try to express answers in terms of the favourite functions. It will consider the input to output as a single function: (coefficients mapped to roots). Then it will approximate this single function directly. Thus, typically, you will have a more reliable error estimate here.

Who uses numerical analysis

Modern engineering and physics would hardly be possible without numerical analysis. Even computer-generate animation movies (e.g., Shrek) or computer games involving collisions (e.g., simulated soccer) make heavy use of numerical analysis.

In statistics, we often com across equations where the constants involved are based on observed data (and hence may be nasty beyond control). So statistician cannot survive without numerical analysis.

Comments

To post an anonymous comment, click on the "Name" field. This will bring up an option saying "I'd rather post as a guest."