bernstein_approximation, a Python code which looks at some simple cases of approximation of a function f(x) by a Bernstein polynomial.
A Bernstein polynomial of degree n is a linear combination of the (n+1) Bernstein basis polynomials of degree n: BP(n,x) = sum ( 0 <= k <= n ) CP(n,k) * B(n,k)(x).
For 0 <= k <= n, the k-th Bernstein basis polynomial of degree n is:
B(n,k)(x) = C(n,k) * (1-x)^(n-k) * x^k
where C(n,k) is the combinatorial function "N choose K"
defined by
C(n,k) = n! / k! / ( n - k )!
For an arbitrary value of n, the set B(n,k) forms a basis for the space of polynomials of degree n or less.
Every basis polynomial B(n,k) is nonnegative in [0,1], and may be zero only at the endpoints.
Except for the case n = 0, the basis polynomial B(n,k)(x) has a unique maximum value at
x = k/n.
For any point x, (including points outside [0,1]), the basis polynomials for an arbitrary value of n sum to 1:
sum ( 1 <= k <= n ) B(n,k)(x) = 1
For 0 < n, the Bernstein basis polynomial can be written as a combination of two lower degree basis polynomials:
B(n,k)(x) = ( 1 - x ) * B(n-1,k)(x) + x * B(n-1,k-1)(x) +
where, if k is 0, the factor B(n-1,k-1)(x) is taken to be 0,
and if k is n, the factor B(n-1,k)(x) is taken to be 0.
A Bernstein basis polynomial can be written as a combination of two higher degree basis polynomials:
B(n,k)(x) = ( (n+1-k) * B(n+1,k)(x) + (k+1) * B(n+1,k+1)(x) ) / ( n + 1 )
The derivative of B(n,k)(x) can be written as:
d/dx B(n,k)(x) = n * B(n-1,k-1)(x) - B(n-1,k)(x)
A Bernstein polynomial can be written in terms of the standard power basis:
B(n,k)(x) = sum ( k <= i <= n ) (-1)^(i-k) * C(n,k) * C(i,k) * x^i
A power basis monomial can be written in terms of the Bernstein basis of degree n where k <= n:
x^k = sum ( k-1 <= i <= n-1 ) C(i,k) * B(n,k)(x) / C(n,k)
Over the interval [0,1], the n-th degree Bernstein approximation polynomial to a function f(x) is defined by
BA(n,f)(x) = sum ( 0 <= k <= n ) f(k/n) * B(n,k)(x)
As a function of n, the Bernstein approximation polynomials form a sequence
that slowly, but uniformly, converges to f(x) over [0,1].
By a simple linear process, the Bernstein basis polynomials can be shifted to an arbitrary interval [a,b], retaining their properties.
The computer code and data files described and made available on this web page are distributed under the MIT license
bernstein_approximation is available in a MATLAB version and an Octave version and a Python version.
bernstein_polynomial, a Python code which evaluates the Bernstein polynomials, useful for uniform approximation of functions;
chebyshev, a Python code which computes the Chebyshev interpolant/approximant to a given function over an interval.
vandermonde_approx_1d, a Python code which finds a polynomial approximant to a function of 1D data by setting up and solving an overdetermined linear system for the polynomial coefficients, involving the Vandermonde matrix.
