Is the Error O(1/n) in Linear Interpolation Correct?

  • Context: Graduate 
  • Thread starter Thread starter AxiomOfChoice
  • Start date Start date
  • Tags Tags
    Error
Click For Summary
SUMMARY

The discussion centers on the accuracy of the error term O(1/n) in linear interpolation for a function f, given specific conditions. The user presents a scenario where f(k/n) and f((k+1)/n) are defined in relation to i and n, leading to the conclusion that the error in interpolation is actually bounded by 1/√n rather than O(1/n). This discrepancy arises from the calculations showing that the errors depend on the specific values of k and t, indicating that the original claim in the book is incorrect.

PREREQUISITES
  • Understanding of linear interpolation in numerical analysis
  • Familiarity with Big O notation and error analysis
  • Knowledge of floor functions and their properties
  • Basic concepts of functions and limits in calculus
NEXT STEPS
  • Study the implications of error bounds in numerical methods
  • Research linear interpolation techniques and their accuracy
  • Learn about the properties of floor functions in mathematical analysis
  • Explore advanced topics in asymptotic analysis and error estimation
USEFUL FOR

Mathematicians, computer scientists, and engineers involved in numerical analysis, particularly those interested in interpolation methods and error estimation techniques.

AxiomOfChoice
Messages
531
Reaction score
1
Consider the following situation. You know, for a given function [itex]f[/itex] and [itex]i,k \in \mathbb N \cup \{ 0 \}[/itex], [itex]n\in \mathbb N[/itex], that

[tex] f \left( \frac kn \right) = \frac{i}{\sqrt n}.[/tex]

In addition, you know that either

[tex] f \left( \frac{k+1}{n} \right) = \frac{i-1}{\sqrt n}[/tex]

-OR-

[tex] f \left( \frac{k+1}{n} \right) = \frac{i+1}{\sqrt n}.[/tex]

And suppose that you linearly interpolate between k/n and (k+1)/n. The book I'm reading claims that, for any t between k/n and (k+1)/n, we have

[tex] f(t) - f \left( \frac{\lfloor tn \rfloor}{n} \right) = O(1/n).[/tex]

where of course I have referred to the floor function. (Incidentally, if [itex]\frac kn \leq t \leq \frac{k+1}{n}[/itex] as assumed, doesn't it follow that [itex]\lfloor nt \rfloor = k[/itex]?) Question: IS THIS (the part about O(1/n)) RIGHT? I don't think it is.
 
Last edited:
Physics news on Phys.org
I've calculated, and it seems to me that if the FIRST option obtains, we have

[tex] f(t) - f \left( \frac{\lfloor tn \rfloor}{n} \right) = \frac{k - tn}{\sqrt n},[/tex]

whereas if the SECOND option obtains, we have

[tex] f(t) - f \left( \frac{\lfloor tn \rfloor}{n} \right) = \frac{tn - k}{\sqrt n}.[/tex]

So I know EXACTLY what the errors are as a function of [itex]t[/itex], and in either case they are bounded by [itex]1/\sqrt n[/itex], simply because [itex]|tn - k| \leq 1[/itex]. Have I done something wrong?
 

Similar threads

  • · Replies 5 ·
Replies
5
Views
3K
  • · Replies 10 ·
Replies
10
Views
3K
  • · Replies 3 ·
Replies
3
Views
4K
  • · Replies 5 ·
Replies
5
Views
3K
  • · Replies 3 ·
Replies
3
Views
3K
  • · Replies 1 ·
Replies
1
Views
3K
  • · Replies 13 ·
Replies
13
Views
3K
  • · Replies 11 ·
Replies
11
Views
3K
  • · Replies 21 ·
Replies
21
Views
4K
  • · Replies 2 ·
Replies
2
Views
4K