Approximating the Square Function: Mathematical Tricks and Other Methods

PeteSampras
Messages
43
Reaction score
2
Homework Statement
Is there some way to approximate this function, for x=1?
Relevant Equations
The function is ##(x-1)^{1/n}##, with ##n## integer and n>1 and n=odd.
Obviously, a priori it is not possible tu use the Taylor series because the derivative ##\sim (x-1)^{1/n-1}## is not well defined in x=1.

Is there any mathematical trick? or, other approximation?
 
  • Wow
Likes nuuskur
Physics news on Phys.org
PeteSampras said:
Homework Statement:: Is there some way to approximate this function, for x=1?
Relevant Equations:: The function is ##(x-1)^{1/n}##, with ##n## integer and n>1 and n=odd.

Obviously, a priori it is not possible tu use the Taylor series because the derivative ##\sim (x-1)^{1/n-1}## is not well defined in x=1.

Is there any mathematical trick? or, other approximation?
Do you mean for ##x \approx 1##?

At ##x = 1## you have ##0^{1/n} = 0##.
 
f'(x_0) = \lim _{x\to x_0} \frac{f(x)-f(x_0)}{x-x_0} \Rightarrow f'(x_0)(x-x_0) +f(x_0) = f(x) +o(|x-x_0|),\ x\to x_0 .
Take x_0 =1 + \delta. If |x-x_0| is small, then it's a pretty good linear approximation.
 
Last edited:
Thanks. I don't understand your idea, because ##f'(x_0)## does not exist for x_0=1
 
PeroK said:
Do you mean for ##x \approx 1##?

At ##x = 1## you have ##0^{1/n} = 0##.

Thanks. This is my doubt.

As for example, if is there any approximation of ##x^{1/3}## near x=0, where Taylor is not possible because the function is not differentiable
 
PeteSampras said:
Thanks. I don't understand your idea, because ##f'(x_0)## does not exist for x_0=1
Indeed, but if you can calculate value of derivative exactly just a bit further from 1, then you could approximate linearly around a very small neighborhood of x_0>1.
 
PeteSampras said:
Thanks. This is my doubt.

As for example, if is there any approximation of ##x^{1/3}## near x=0, where Taylor is not possible because the function is not differentiable
If ##x## is small, then powers of ##x## are smaller. But, ##x^{1/3} > x##, so you are not going to get a power series approximation to ##x## in powers of ##x##.
 
For the record, non-smooth approximation is hard. Most of the theory I've come across always assumes differentiability or absolute continuity or something "nice". Non-smooth continuous functions need not be nice at all.
 
nuuskur said:
Indeed, but if you can calculate value of derivative exactly just a bit further from 1, then you could approximate linearly around a very small neighborhood of x_0>1.

I don't understand :(
PeroK said:
If ##x## is small, then powers of ##x## are smaller. But, ##x^{1/3} > x##, so you are not going to get a power series approximation to ##x## in powers of ##x##.

This mean that, in my problem ##(1-x)^{1/n}## is not possible to approximate the function for ##x_0 \approx 1##?
 
  • #10
PeteSampras said:
This mean that, in my problem ##(1-x)^{1/n}## is not possible to approximate the function for ##x_0 \approx 1##?

I don't see how you can do it with a Taylor series. One problem is that the derivatives are unbounded near ##x = 1##.
 
  • #11
For an approximation, I suggest you expand using the binomial theorem and take the limit as x approaches 1. For ##x\gt 1## $$(x-1)^{\frac{1}{n}}=(x)^{\frac{1}{n}}(1-\frac{1}{x})^{\frac{1}{n}}=(x)^{\frac{1}{n}}\sum_{k=0}^{\infty} \begin{pmatrix}
\frac{1}{n} \\
k
\end{pmatrix} (\frac{1}{x})^k $$Use the identity$$
\begin{pmatrix}
\frac{1}{n} \\
k
\end{pmatrix}=\frac{(-1)^k}{k!} \prod_{j=0}^{k-1} (j-\frac{1}{n})$$ for each term in the series.
 
Last edited:
  • #12
You might as well work in terms of ##x^{\frac 1n}##, for x near zero,
It depends what you want from an approximation. Generally you want such as an easy way to generate numerical solutions. What form will x be in? If in scientific notation, ##x=a\times 10^m##, you have ##a^{\frac 1n}\times 10^{\frac mn}##. For large n you can ignore ##a^{\frac 1n}##, and it’s easy to reduce m so that it is less than n.
This leaves you with ##10^y##, 0<y<1. For that, you can use ##e^{y\ln(10)}## or ##e^{y\ln(10)+\ln(a)}##.

But is the problem as stated in post #1 as it was given to you? If not, please post the entire original problem.
 
Back
Top