Research Interests
NAVIGATION

Academic Links
Homepage
Research Interests
Personal Information
Mathematical Department
University Salzburg
Conferences
Publications
Approximation Theorists
S.N. Bernstein
Mathworld

Personal Links
My flight aviation site
Alpamayo and Huascaran
Mallory and Irvine
Tennengebirge Mountains
The Grossvenediger
The Maldives

Research Interests
My research interest in interpolation theory is mainly focused in the investigation of pointwise divergence properties of equidistant Lagrange interpolation to the continuous functions on [-1,1]. We can illustrate the equidistant divergence by a well known example, discorverd by S.N. Bernstein in 1918.
 
The classical result of S.N. Bernstein

Bernstein Divergence
An "animated" running subsequence of
equidistant interpolants to the simple function Abs[x].

On the left-hand side you can see an "animated" running subsequence of equidistant interpolation polynomials for the function Abs[x] on the interval [-1,+1].

In 1918, S.N. Bernstein proved the surprising result that for each choice of x in [-1,+1], apart from the values -1, 0, +1 the sequence of Lagrange polynomials is unbounded and thus does not converge to the correct function value.

For the points -1, +1 divergence cannot occur in the equidistant interpolation array. However, for the point 0 the situation is more complicated:

It was proved in 1939 by D.L. Berman that the Lagrange polynomials at zero converge to its true function value.
S.M. Lozinskii established an upper bound for the approximation error and showed that the error tends to 0 with O(1/n).

In 1999 Bernstein's classical result was extended to the following expression:


An extension of Bernstein's result (red polynomial)..
Regarding to the above mentioned results of D.L. Berman and S.M. Lozinskii, I was able to amplify the bounds for the approximation error for Abs[x] at the point zero to the following expression:


Approximation error for the point zero (green line).
For Abs[x] it was shown that the approximation constant at the point zero is exactly 2/Pi.
It is further conjectured that the Bernstein result reflects a more general situation. Here is the conjecture:

I succeeded in proving the following (special) situation (which gives a slight extension of a result from 1990 by Byrne/Mills/Smith (see "On Lagrange interpolation with equidistant nodes", Bull. Austral. Math. Soc. Vol. 42 (1990), 81-89):

Regarding the approximation error at the point zero I have established the expression:
It is surprising that the approximation constant is again 2/Pi. It is further conjectured that the only integer values for the exponent alpha for which the constant equals 2/Pi are the values 1 and 3.