Again I have to apologize for the sparseness of posting lately, but I’ve got two research projects going full blast and time has not been something I have a lot of. I’ll still be writing at least a few times a week, and you can’t beat the price. ðŸ˜‰ In any case once things cool down just a little I should be back to a more regular schedule.

Today’s function isn’t interesting because of the function itself, the interest comes from what we’ll do with it. Let’s say we have a function like this:

If we want to see where the function is equal to zero, it’s clear that 0 = x^2 – 2 is solved by x equal to the square root of two, either positive or negative. That’s the snappy analytic solution, but let’s say we want to use this function to compute a decimal approximation to whatever accuracy we feel like. The brute force method is just to try various decimal numbers and see what gets us closest, refining our guess each time. But this is ugly, slow, and requires a lot of continual work. We’d prefer a faster method where we can just turn the crank easily and get a good answer. To do this, let’s plot the function and (for reasons I’ll explain in the next paragraph) its tangent line at the point x = 4.

Let’s say we picked x = 4 for our initial guess as the value of the square root of two. It’s an awful guess, obviously, but that’s ok since we’re looking for a procedure that can turn any terrible guess gradually into better and better approximations of the actual answer. So we pick our initial point and draw the tangent line. We notice that it cuts the x-axis pretty close to the square root of two (which is exactly where the parabola cuts the x-axis, since the square root of two is by definition the number that makes the function equal to zero). So why don’t we take the location where the straight line cuts the x-axis, and use that as our second guess, and draw the tangent line there:

This line cuts the axis even closer to the point x = the square root of 2. If we take that point as our next guess and repeat the process, we should be closer still. So first we need to actually work this procedure into mathematical language so we can actually get some numbers out of it. The equation of a line is:

Where m is the slope and b is the y-intercept. We know the slope of a tangent line is just the derivative of the function at that point, and y is just the value of the function at that point. That means we can solve for b:

Where the prime denotes differentiation, and we’re subscripting the x so we know it’s the particular value of the guess we’re using. Our new x for the next iteration of the guess will be the value that makes our line equation y = mx + b equal zero, i.e., 0 = m x + b. Substituting all the previous stuff into this equation and solve for x. After a little simplification, we get:

Now this is a very general expression that works to find the zeros of an arbitrary function f, whatever it happens to be. Our particular function can be plugged in (for us, f'(x) = 2x by a little bit of calculus), which gives us the complete procedure:

Let’s give this a try. Plug in our initial guess of 4 and the procedure tells us our next guess is 2.25. Plug that in and the procedure gives us 1.56944. Plug that in and we get 1.42189. Repeat again and get 1.41423. So on and so forth closing in on the rounded-off real value of 1.41421, and if we didn’t round off at 5 decimal places as I’m doing eventually we’d get as many digits of the square root as we wanted to arbitrary accuracy.

This procedure is called Newton’s Method, and it’s a fine way of calculating the zeros of a function if you know its derivative. The method is not quite perfect – if a function has more than one zero the one you get will depend on your initial guess in a not-always-predictable way. And while the method is quite general, there are certain functions that don’t fulfill the relatively generous convergence conditions. Still, it’s a great method with a long and continuing history of use. It’s pretty likely that your pocket calculator uses a very similar method when you hit the square root button. And now if you ever find yourself without such a button, you can do it yourself if you’re patient.