Extreme Optimization > Mathematics Library for .NET > User's Guide > Current Page > Chebyshev Expansions - Curves

Extreme Optimization Mathematics Library for .NET

Chebyshev Expansions

Chebyshev polynomials form a special class of polynomials whose properties make them especially suited for approximating other functions. As such, they form an essential part of any numerical library. The Extreme Optimization Mathematics Library for .NET supports Chebyshev approximations through the the ChebyshevSeries class.

The ChebyshevSeries class

A Chebyshev expansion is a linear combination of Chebyshev polynomials. The Chebyshev polynomials are never formed explicitly. All calculations can be performed using only the coefficients.

The Chebyshev polynomials provide an alternate basis for representating general polynomials. Two characteristics make Chebyshev polynomials especially attractive. They are mutually orthogonal, and there exists a simple recurrence relation between consecutive polynomials.

Chebyshev polynomials are defined over the interval [-1, 1]. Using Chebyshev expansions outside of this interval is usually not meaningful and is to be avoided. To allow expansions over any finite interval, transformations are applied wherever necessary.

The ChebyshevSeries class inherits from PolynomialBase This class defines a number of properties shared by all polynomial classes. PolynomialBase is itself derived from LinearCombination.

The parameters of a Chebyshev expansion are the coefficients of the polynomial.

The Degree of a Chebyshev expansion is the highest degree of a Chebyshev polynomial that appears in the sum. The number of parameters of the expansion equals the degree plus one.

Constructing Chebyshev Expansions

The ChebyshevSeries has four constructors. The first two variants let you specify the degree of the highest order Chebyshev polynomial in the expansion. You can also specify the lower and upper bound of the interval. If the lower and upper bound are omitted, the standard interval [-1,1] is assumed:

// Chebyshev expansion of degree 5:
ChebyshevSeries c1 = new ChebyshevSeries(5);
// Chebyshev expansion of degree 5 over the interval [0,2]:
ChebyshevSeries c2 = new ChebyshevSeries(5, 0.0, 2.0);

' Chebyshev expansion of degree 5:
Dim c1 As ChebyshevSeries = New ChebyshevSeries(5)
' Chebyshev expansion of degree 5 over the interval [0,2]:
Dim c2 As ChebyshevSeries = New ChebyshevSeries(5, 0.0, 2.0)

The third and fourth variants let you specify the coefficients of the expansion, and optionally the lower and upper bound of the interval. The kth element of the coefficient array should be the coefficient of the Chebyshev polynomial of degree k.

Double [] coefficients = new Double[] {1, 0.5, 0.25, 0.125, 0.0625};
// Chebyshev expansion of degree 4 with the above coefficients:
ChebyshevSeries c3 = new ChebyshevSeries(coefficients);
// Chebyshev expansion of degree 4 over the interval [0, 2]:
ChebyshevSeries c4 = new ChebyshevSeries(coefficients, 0.0, 2.0);

Dim coefficients As Double() = New Double() {1, 0.5, 0.25, 0.125, 0.0625}
' Chebyshev expansion of degree 4 with the above coefficients:
Dim c3 As ChebyshevSeries = New ChebyshevSeries(coefficients)
' Chebyshev expansion of degree 4 over the interval [0, 2]:
Dim c4 As ChebyshevSeries = New ChebyshevSeries(coefficients, 0.0, 2.0)

In addition, you can also create a Chebyshev expansion using one of several static methods.

The LeastSquaresFit method constructs a ChebyshevSeries that is the least squares fit of a specified degree through a given set of points. If the boundaries of the approximation interval are not specified, the smallest and highest x-coordinates are used as the lower and upper bound. The result is mathematically equivalent to the polynomial obtained from the LeastSquaresFit method of the Polynomial class. The calculation is much more stable, however, leading to good accuracy even for high degree approximations.

The GetInterpolatingPolynomial function takes a RealFunction delegate as its first parameter. The second parameter is an integer indicating the degree of the polynomial. This method calculates the interpolating polynomial (in the form of a Chebyshev expansion) through the so-called Chebyshev points . This technique ensures that the approximation deviates as little as possible from the actual function.

The following examples illustrate these methods. We will approximate the function sin(x) over the interval [-pi/2,pi/2] in two ways. First, we will find a least squares fit through 7 points on this curve. We will then calculate the Chebyshev interpolating polynomial of degree 4.

// x and y values contain the data points:
double[] xValues = new double[13];
double[] yValues = new double[13];
int index;

for(index = 0; index < 13; index++)
{
	xValues[index] = (index-6) * Constants.Pi / 12;
	yValues[index] = Math.Sin(xValues[index]);
}

// Now we can find the least squares approximation:
ChebyshevSeries lsqApproximation = 
    ChebyshevSeries.LeastSquaresFit(xValues, yValues,
        -Constants.Pi/2, Constants.Pi/2, 4);

// Note that, as expected, the odd coefficients are close to zero.
Console.WriteLine("Least squares fit:");
for(index = 0; index <= 4; index++)
    Console.WriteLine("c({0}) = {0}", 
        index, lsqApproximation[index]);

// Now let's calculate the interpolating Chebyshev expansion:
RealFunction f = new RealFunction(Math.Sin);
ChebyshevSeries interpolant = 
    ChebyshevSeries.GetInterpolatingPolynomial(f, 
        -Constants.Pi/2, Constants.Pi/2, 4);

// This expansion is close to the previous one:
Console.WriteLine("Interpolating polynomial:");
for(index = 0; index <= 4; index++)
    Console.WriteLine("c({0}) = {0}", 
        index, interpolant[index]);

' x and y values contain the data points:
Dim xValues As Double() = New Double(13)
Dim yValues As Double() = New Double(13)
Dim index As Integer
For index = 0 To 12
	xValues(index) = (index-6) * Constants.Pi / 12
	yValues(index) = Math.Sin(xValues(index))
Next

' Now we can find the least squares approximation:
Dim lsqApproximation As ChebyshevSeries = _
    ChebyshevSeries.LeastSquaresFit(xValues, yValues, _
        -Constants.Pi/2, Constants.Pi/2, 4)

' Note that, as expected, the odd coefficients are close to zero.
Console.WriteLine("Least squares fit:")
For index = 0 To 4
    Console.WriteLine("c({0}) = {0}", _
        index, lsqApproximation.Coefficients(index))
Next

' Now let's calculate the interpolating Chebyshev expansion:
Dim f As RealFunction = New RealFunction(AddressOf Math.Sin)
Dim interpolant As ChebyshevSeries = _
    ChebyshevSeries.GetInterpolatingPolynomial(f, _
        -Constants.Pi/2, Constants.Pi/2, 4)

' This expansion is close to the previous one:
Console.WriteLine("Interpolating polynomial:")
For index = 0 To 4
    Console.WriteLine("c({0}) = {0}", _
        index, interpolant.Coefficients(index))
Next

Note that, even though the interpolating polynomial was calculated using only a third of the data, the resulting approximation is still very close to the least squares solution.

Working with Chebyshev expansions

The ChebyshevSeries class implements all methods and properties of the Curve class.

The ValueAt method returns the value of the polynomial at a specified point. SlopeAt returns the derivative. For expansions over an interval different from the standard [-1,1], appropriate transformations are applied automatically.

Console.WriteLine("p1.ValueAt(2) = {0}", c1.ValueAt(2));
Console.WriteLine("p1.SlopeAt(2) = {0}", c1.SlopeAt(2));

Console.WriteLine("p1.ValueAt(2) = {0}", c1.ValueAt(2))
Console.WriteLine("p1.SlopeAt(2) = {0}", c1.SlopeAt(2))

The GetDerivative method returns the Chebyshev expansion that is the derivative of a Chebyshev expansion.

Curve derivative = c1.GetDerivative();
Console.WriteLine("Slope at 2 (derivative) = {0}",
    derivative.ValueAt(2));

Dim derivative As Curve = c1.GetDerivative()
Console.WriteLine("Slope at 2 (derivative) = {0}", _
    derivative.ValueAt(2))

Integral evaluates the definite integral over a specified interval. This value is calculated directly using the coefficients of the expansion. No numerical approximation is used.

Copyright 2004-2008, Extreme Optimization. All rights reserved.
Extreme Optimization, Complexity made simple, M#, and M Sharp  are trademarks of ExoAnalytics Inc.
Microsoft, Visual C#, Visual Basic, Visual Studio, Visual Studio.NET, and the Visual Studio Logo are registered trademarks of Microsoft Corporation