Last time: | Next time: |
Today:
Some of you interpreted this differently from me. The authors ask "for which values of do tiny relative changes in produce relative change in much larger in magnitude?
You then talked about the derivative, essentially. This is legitimate in my view.
What's the difference?
The condition number is the "relative derivative": the ratio of the relative error in over the relative error in .
The derivative could be defined as the error in over the error in .
In particular, condition is "dimensionless": if we change the units of the function , or of its argument , the derivative changes its units, too. It scales with both. But the condition number does not.
A useful example to think about is problem #8, above. If we look at the condition number of the function cos(x), it gets larger as we approach each root of cos(x): small change in x, big change in cos(x). But the condition number gets relatively larger as x gets bigger (and the derivative of cos(x) at the roots doesn't change). For a fixed error, the relative error in x is going to get smaller, whereas the relative error in the function won't change.
But because 2 divides 10.... Consider terminating binary decimal for any integer .
And because the sum of terminating decimals is terminating....
Some of you commented on other bases. It's possible, for example, that a terminating decimal in base 3 will not have a terminating decimal expansion in base 10. Can you give an example?
We've pretty finished with chapters 1 and 2, and I think that we're ready to start into Chapter 3. We're going to be considering root-finding.
I will start, however, by asking what questions folks might have about chapter 2.
The most important equation in the world (Physicist Charles Shirkey, BGSU):
How would you do it?
(Put away your books!)
Today I want to lay out the problems and issues, and conditions under which we can solve this. Maybe even a strategy for solving some cases.
Then we'll see what our authors have to suggest. They have essentially three methods to propose. We'll look at those, and then one more -- Muller's method -- which will allow us to find complex roots of real-valued equations.