Difference between revisions of "Interpolation"

Revision as of 17:58, 1 September 2008

1 Linear Interpolation
- 1.1 Example
  - 1.1.1 Implementation in Matlab
  - 1.1.2 Implementation in Excel
2 Polynomial Interpolation
- 2.1 Example
  - 2.1.1 Implementation in Matlab
3 Cubic Spline Interpolation
- 3.1 Example
  - 3.1.1 Implementation in Matlab
4 Lagrange Polynomial Interpolation
- 4.1 Example

Linear Interpolation

A linear function may be written as $f(x) = a_0 + a_1 x.$ Given any two data points, $( x_1, y_1 )$ , and $(x_2,y_2)$ , we can determine a linear function that exactly passes through these points. We can do this by solving for $a_0$ and $a_1$ . Since there are two unknowns we require two equations. They are:

\begin{align} y_1 &= a_0 + a_1 x_1 \\ y_2 &= a_0 + a_1 x_2 \end{align}

These may be written in matrix form as

\left[ \begin{array}{cc} 1 & x_1 \\ 1 & x_2 \end{array} \right] \left( \begin{array}{c} a_0 \\ a_1 \end{array} \right) = \left( \begin{array}{c} y_1 \\ y_2 \end{array} \right)

Given values for $( x_1, y_1 )$ , and $(x_2,y_2)$ , these equations may be easily solved in MATLAB. But since they are so simple, we can easily by hand to obtain a general solution,

\begin{align} a_0 &= \frac{x_1 y_2 -x_2 y_1}{x_1-x_2} \\ a_1 &= \frac{y_1-y_2}{x_1-x_2} \end{align}

This gives the coefficients, which may then be substituted into the original equation $f(x) = a_0 + a_1 x$ and simplified to obtain

f(x) = \left( \frac{y_2-y_1}{x_2-x_1} \right) \left( x-x_1 \right) + y_1.

This is a very convenient equation for performing linear interpolation.

Example

The density of air at various temperatures and atmospheric pressure is given in the following table

Temperature (°F)	1.39	1.16	0.99	0.87	0.78	0.69	0.63	0.58	0.54	0.50	0.46	0.44	0.41
Density (kg/m³)	-10	80	170	260	350	440	530	620	710	800	890	980	1070

Estimate the density of air at 32 °F. Using linear interpolation, we have $\rho(T)=a_0 + a_1 T$ . We define $(T_1,\rho_1)=(-9.7,1.39)$ and $(T_2,\rho_2)=(80,1.161)$ . Now we can calculate

\begin{align} \rho(32) &\approx \frac{\rho_2 -\rho_1}{T_2-T_1}\left( T - T_1 \right) + \rho_1 \\ &= \frac{1.16-1.39}{-9.7-80} \left( 32 + 9.7 \right) + 1.39 \\ &= 1.28. \end{align}

Implementation in Matlab

There are three ways to do interpolation in MATLAB.

Solve the linear system.

 T1=-9.7;  T2=80;
 rho1=1.39; rho2=1.16;
 T = 32;
 A = [ 1 T1; 1 T2 ];
 b = [ rho1; rho2 ];
 a = A\b;
 rho = a(1) + a(2)*T;

Use the general equation we derived above:

 T1=-9.7; T2-80;
 rho1=1.39; rho2=1.16;
 T = 32;
 rho = (rho2-rho1)/(T2-T1) * (T-T1) + rho1;

Use MATLAB's built-in interpolation tool

 Ti = [-9.7 80];
 rhoi = [1.39 1.16];
 rho = interp1( Ti, rhoi, 32 );

All three of these methods provide exactly the same answer. However, using MATLAB's interp1 function allows you to use a vector for the points that you want to interpolate to. For example, if we wanted the density at 30, 50, 70, 10, and 110 °F, we could accomplish this by:

  Ti = [-9.7 80 170 ];
  rhoi = [1.39 1.16 0.995];
  T = 30:20:110;
  rho = interp1( Ti, rhoi, T );

Implementation in Excel

This section is a stub and needs to be expanded.

If you can provide information or finish this section you're welcome to do so and then remove this message afterwards.

Polynomial Interpolation

Polynomial interpolation is a simple extension of linear interpolation. In general, an n^th degree polynomial is given as

f(x)=\sum_{i=0}^n a_i x^i.

If n=1 then we recover a first-degree polynomial, which is linear. The formula gives $f(x)=\sum_{i=0}^1 a_i x^i = a_0 + a_1 x$ .

In general, if we want to interpolate a set of data using an n^th degree polynomial, then we must determine n+1 coefficients, $a_0 \ldots a_n$ . Therefore, we require n+1 points to interpolate using an n^th degree polynomial.

The equations that must be solved are given as

\left[\begin{array}{ccccc} 1 & x_{1} & x_{1}^2 & \cdots & x_{1}^{n}\\ 1 & x_{2} & x_{2}^2 & \cdots & x_{2}^{n}\\ \vdots & \vdots & \vdots & \cdots & \vdots \\ 1 & x_{n+1} & x_{n+1}^2 & \cdots & x_{n+1}^{n}\end{array}\right]\left(\begin{array}{c} a_{0}\\ a_{1}\\ \vdots\\ a_{n}\end{array}\right) = \left(\begin{array}{c} y_{1}\\ y_{2}\\ \vdots\\ y_{n+1}\end{array}\right)

Solving these equations provides the values for the coefficients, $a_0 \ldots a_n$ . Once we know these coefficients, we can evaluate the polynomial at any point.

Example

Using a third degree polynomial and the data in the table above, approximate the density of air at 50 °F and 300 °F.

We require 4 points to fit a third degree polynomial, $a_0, \, a_1, \, a_2, \, a_3$ . The equations to be solved are:

\begin{align} \rho_1 &= a_0 + a_1 T_1 + a_2 T_1^2 + a_3 T_1^3 \\ \rho_2 &= a_0 + a_1 T_2 + a_2 T_2^2 + a_3 T_2^3 \\ \rho_3 &= a_0 + a_1 T_3 + a_2 T_3^2 + a_3 T_3^3 \\ \rho_4 &= a_0 + a_1 T_4 + a_2 T_4^2 + a_3 T_4^3 \end{align}

Once we have solved these equations we have the polynomial coefficients and we can then evaluate the resulting polynomial at any temperature that we wish.

We still must choose what 4 points in the table to use in the above equations. We will choose the 4 points in the table that are closest to the temperature that we wish to interpolate to. Note that if we want to evaluate the density at different temperatures, we may want to use different points in the table to determine the coefficients!

Our procedure to perform the interpolation is:

Determine which points we need to use.
Set up and solve the linear system to obtain the polynomial coefficients $a_0 \ldots a_3$ .
Evaluate the polynomial to obtain the temperature.

Implementation in Matlab

The MATLAB code to obtain the density at 50 °F is

% Set the data points from the table.
Ti = [ -9.7   80  170  260  350  440  530 ...
        620  710  800  890  980 1070 ]';
rhoi=[ 1.39 1.16 0.99 0.87 0.78 0.69 0.63 ...
       0.58 0.54 0.50 0.46 0.44 0.41 ]';

% Set the temperature we want to evaluate the density at.
T = 50;

% determine the set of points that we will use to construct
% the interpolant. We want to use points that are closest
% to the temperature that we are interested in.
points = 1:4;

% set up the linear system
A = [ ones(4,1), ones(4,1).*Ti(points), ones(4,1)*Ti(points).^2, ones(4,1)*Ti(points).^3 ];
b = rhoi(points);

% solve for the polynomial coefficients
a = A\b;

% evaluate the density.
rho = a(0) + a(1)*T + a(2)*T^2 + a(3)*T^3;

For T=300 we would probably want to choose a different set of points to determine the polynomial coefficients. Therefore, we would repeat the above with the only change being:

T=300;
points=4:8;

Cubic Spline Interpolation

Cubic spline interpolation uses cubic polynomials to interpolate datasets. The concept is illustrated in the following figure:

The data points are connected with cubic functions, and on each interval the coefficients must be determined. If we have $n$ data points then we can define $n-1$ cubic functions, one between each data point. For each spline we have four coefficients to determine. Therefore, we require $4(n-1)$ constraints to determine the $4(n-1)$ coefficients.

The constraints may be summarized in the following table:

Constraints for Cubic Splines
Description	Mathematical Statement	Number of Constraints
At each data point, two splines meet. Here the splines must be equal to the data values.	$f_{i}(x_{i+1}) = y_{i+1} \quad 2 \le i \le n-2$ $f_{i+1}(x_{i+1}) = y_{i+1} \quad 2 \le i \le n-2$	$2(n-2)$
At the first and last point, the function must equal the data.	$f_1(x_1) = y_1$ $f_{n-1}(x_n)=y_n$	$2$
At each interior point the first derivatives of adjacent splines must be equivalent.	$\left. \frac{\partial f_{i}}{\partial x} \right\|_{i+1} = \left. \frac{\partial f_{i+1}}{\partial x} \right\|_{i+1}$	$n-2$
At each interior point, the second derivatives of adjacent splines must be equivalent.	$\left. \frac{\partial^2 f_{i}}{\partial x^2} \right\|_{i+1} = \left. \frac{\partial^2 f_{i+1}}{\partial x^2} \right\|_{i+1}$	$n-2$
We specify the curvature at the end points. By choosing the curvature to be equal to zero, we obtain a natural spline.	$\left. \frac{\partial^2 f_1}{\partial x^2} \right\|_{1} = 0$ $\left. \frac{\partial^2 f_{n-1}}{\partial x^2} \right\|_{n} = 0$	$2$

Adding up all of the constraints, we see that we have $4(n-1)$ constraints - precisely the number that we need. We could formulate equations for all of these constraints and solve them all to determine the coefficients for the cubic function on each interval.

Example

Using the data in the table above, determine the density of air at the following temperatures: 60, 120, 180, 240, 300, 360, 420, 480, 540 °F.

Implementation in Matlab

Matlab has built-in functions for cubic spline interpolation:

y = interp1( xi, yi, x, 'spline' );

(xi,yi) are the points at which we have data defined.

x is the point(s) where we want to interpolate.

'spline' tells Matlab to interpolate using cubic splines.

Using this built-in function, we may obtain the interpolated values as follows:

% Set the data points from the table.
Ti = [ -9.7   80  170  260  350  440  530 ...
        620  710  800  890  980 1070 ]';
rhoi=[ 1.39 1.16 0.99 0.87 0.78 0.69 0.63 ...
       0.58 0.54 0.50 0.46 0.44 0.41 ]';

% set the temperatures that we want to interpolate to.
T = 60:60:540;

% determine the density at the desired temperatures.
rho = interp1( Ti, rhoi, T, 'spline' );

Lagrange Polynomial Interpolation

This section is a stub and needs to be expanded.

If you can provide information or finish this section you're welcome to do so and then remove this message afterwards.

Given $n_p$ points $(x_k,y_k),$ the $n^\mathrm{th}$ order Lagrange polynomial that interpolates these function values, $f(x)$ are expressed as

f(x)=\sum_{k=0}^{n} y_{k} L_{k}(x),

where $L_{k}(x)$ is the Lagrange polynomial given by

L_{k}(x)=\prod_{\stackrel{i=0}{i\ne k}}^{n}\frac{x-x_{i}}{x_{k}-x_{i}},

and $n_p = n+1$ . In other words, for an $n^\mathrm{th}$ order interpolation, we require $n_p=n+1$ points.

The $\prod$ operator represents the continued product, and is analogous to the $\sum$ operator for summations. For example, $\prod_{i=1}^{3} (x+i) = (x+1)(x+2)(x+3)$ .

@@ Line 172: / Line 172: @@
 points=4:8;
 </source></ul>
 == Cubic Spline Interpolation ==
-{{Stub|section}}
+Cubic spline interpolation uses cubic polynomials to interpolate datasets.  The concept is illustrated in the following figure:
+[[image:spline1.png|center|500px]]
+The data points are connected with cubic functions, and on each interval the coefficients must be determined.  If we have <math>n</math> data points then we can define <math>n-1</math> cubic functions, one between each data point.  For each spline we have four coefficients to determine.  Therefore, we require <math>4(n-1)</math> constraints to determine the <math>4(n-1)</math> coefficients.
+The constraints may be summarized in the following table:
+{| border="1" cellpadding="3" cellspacing="0" align="center" style="text-align:center"
+|+ Constraints for Cubic Splines
+|-
+! width="375"|Description
+! Mathematical Statement
+! Number of Constraints
+|- valign="top"
+| At each data point, two splines meet.  Here the splines must be equal to the data values.
+| <math>f_{i}(x_{i+1}) = y_{i+1} \quad 2 \le i \le n-2</math>
+<math>f_{i+1}(x_{i+1}) = y_{i+1} \quad 2 \le i \le n-2</math>
+| <math>2(n-2)</math>
+|- valign="top"
+| At the first and last point, the function must equal the data.
+| <math>f_1(x_1) = y_1</math>
+<math>f_{n-1}(x_n)=y_n</math>
+| <math>2</math>
+|- valign="top"
+| At each interior point the first derivatives of adjacent splines must be equivalent.
+| <math>\left. \frac{\partial f_{i}}{\partial x} \right|_{i+1} = \left. \frac{\partial f_{i+1}}{\partial x} \right|_{i+1}</math>
+| <math>n-2</math>
+|- valign="top"
+| At each interior point, the second derivatives of adjacent splines must be equivalent.
+| <math>\left. \frac{\partial^2 f_{i}}{\partial x^2} \right|_{i+1} = \left. \frac{\partial^2 f_{i+1}}{\partial x^2} \right|_{i+1}</math>
+| <math>n-2</math>
+|- valign="top"
+| We specify the curvature at the end points.  By choosing the curvature to be equal to zero, we obtain a natural spline.
+| <math>\left. \frac{\partial^2 f_1}{\partial x^2} \right|_{1} = 0</math>
+<math>\left. \frac{\partial^2 f_{n-1}}{\partial x^2} \right|_{n} = 0</math>
+| <math>2</math>
+|}
+Adding up all of the constraints, we see that we have <math>4(n-1)</math> constraints - precisely the number that we need.  We could formulate equations for all of these constraints and solve them all to determine the coefficients for the cubic function on each interval.
+<br clear="all" />
 === Example ===
+Using the data in the table above, determine the density of air at the following temperatures: 60, 120, 180, 240, 300, 360, 420, 480, 540 &deg;F.
 ==== Implementation in Matlab ====
+Matlab has built-in functions for cubic spline interpolation:
+<source lang="matlab">
+y = interp1( xi, yi, x, 'spline' );
+</source>
+: <tt>(xi,yi)</tt> are the points at which we have data defined.
+: <tt>x</tt> is the point(s) where we want to interpolate.
+: <tt>'spline'</tt> tells Matlab to interpolate using cubic splines.
+Using this built-in function, we may obtain the interpolated values as follows:
+<ul><source lang="matlab">
+% Set the data points from the table.
+Ti = [ -9.7   80  170  260  350  440  530 ...
+  710  800  890  980 1070 ]';
+rhoi=[ 1.39 1.16 0.99 0.87 0.78 0.69 0.63 ...
+.58 0.54 0.50 0.46 0.44 0.41 ]';
+% set the temperatures that we want to interpolate to.
+T = 60:60:540;
+% determine the density at the desired temperatures.
+rho = interp1( Ti, rhoi, T, 'spline' );
+</source></ul>

Difference between revisions of "Interpolation"

Revision as of 17:58, 1 September 2008

Contents

Linear Interpolation

Example

Implementation in Matlab

Implementation in Excel

Polynomial Interpolation

Example

Implementation in Matlab

Cubic Spline Interpolation

Example

Implementation in Matlab

Lagrange Polynomial Interpolation

Example

Navigation menu

Views

Personal tools

Navigation

Search

Tools