There is a formula

The derivation is on the top of this page:

http://en.wikipedia.org/wiki/Linear_interpolation

It can be derived from what I call the point - point formula of Cartesian geometry. basically, if you want to linearly interpolate between two data points you fit a straight line between those two points and then just plug in.

There is also a method that can do this interpolation. Welcome to the forum!

]]>I have contacted a mathematics professor about it from a statistics class.

That is a good idea because as I understand it Thiele and Fisher ( the statistician ) were coworkers. I have never seen the derivation of it. So I thank you beforehand.

year ago when my math teacher refused to teach it to me.

Maybe he was right? From a Numerical Analyst point of view I am forced to agree. As a practical method it is a hopeless jumble and there are better ways.

]]>You know... a proof.

I have contacted a mathematics professor about it from a statistics class. When I obtain the derivation process I shall post it here to benefit others who may also be interested in learning about where the formula comes from.

Thanks...

]]>No, the derivatives would be the same.

Yes, in the fit, but a fit was impossible, I think because of the problem with the derivatives being different at the same x value for the two piecewise functions. That is why they could not be joined into one function. You were also asking for a interpolating fit ( exact ). For many discontinous functions that is not possible. Just explaining why there was no reply, I failed in the attempt even though I used thousands of points and 3 computers! Mathematically, I suspect the above reason will not allow the problem to be solved.

Now getting to the Thiele fit that would require reciprocal differences and a continued fraction.

That is how it is done. The p() are the reciprocal differences, very hard to compute by hand. This is clearly a job for a computer.

]]>Anyway, this question was about deriving the formula itself.

]]>I am pretty sure that the reason I did not get back to you is that I was unable to find a single smooth curve to join those 2 piecewise functions. Although they look fine graphed they are not one function but 2 very close together. You see interpolation like numerical integration is sensitive to the singularites or discontinuities in the derivative of the function. The two functions have different derivatives at the same point. That is why a fit was not possible in my opinion. I do not know of any rational or polynomial function that have 2 different derivative values at the same point.

]]>If anyone has a proof/derivation of this formula, I thank ye. I do not have a lot of books on statistics or physics.

]]>Remember this?

]]>I am curious and trying to learn more about Thiele's Interpolation Formula, the mechanics of how it works, and understanding, in general, how it is derived, among other things.

That which fascinates me about the formula is an odd inversion of the function within the function itself that seems to appear with any set of three points friendly enough with the function to not divide by zero. For example, if you solve the formula for the three symmetric points

you get the formula

which, on the domain from 0 to 100 creates a pretty graph whose derivatives are 9 when x = 0, 1 when x = 25, and 1/9 when x = 100, simply using the values of x from the points given above.

What's bizarre is that the function has an inverse of itself across the vertical asymptotic line and in the second quadrant which is shifted over by a value equal but negative to the second x value (-25)and up by the second y value (75). The inverse in the second quadrant therefore has derivatives equal to and equidistant to those of the function's curve in the first quadrant, proving they are the same.

How does this work? How can a function contain its own inverse?

My observation of this goings on stems from my attempts at curve fitting which can be found elsewhere in the "Help Me!" portion of the forum, though by now I believe it must be buried several pages back.

Thank you for your input!

]]>