Fitting 2nd Order multivariate quadratic with matrices

1.1k Views Asked by At

Hopefully you at least entertain this question as it took forever to construct the below matrix using TeX. Any ways, so I have a list of data points ($X_1$,$X_2$,Y), with the X's being independent variables and Y being dependent. What I am attempting to find is the best fit multivariate quadratic using sum of the least squares.

The general 2nd order polynomial I am attempting to find is in the form of: $$ y = B_0 + B_1 X_1 + B_2 X_2 + B_{11} X_1^2 + B_{22} X_2^2 + B_{12}X_1X_2 $$

1) Can anyone tell me if the below matrix is correct? In comparison, a traditional uni-variate quadratic fit can be seen on page 9.

$$ \begin{vmatrix} n & \sum_{} X_1 & \sum_{} X_2 & \sum_{} X_1 X_2 & \sum_{} X_1^2 & \sum_{} X_2^2 \\ \sum X_1 & \sum_{} X_1^2 & \sum_{} X_1 X_2 & \sum_{} X_1^2X_2 & \sum_{} X_1^3 & \sum_{} X_1 X_2^2 \\ \sum X_2 & \sum_{} X_1 X_2 & \sum_{} X_2^2 & \sum_{} X_1 X_2^2 & \sum_{} X_1^2 X_2 & \sum_{} X_2^3 \\ \sum X_1 X_2 & \sum_{} X_1^2 X_2 & \sum_{} X_1 X_2^2 & \sum_{} X_1^2 X_2^2 & \sum_{} X_1^3 X_2 & \sum_{} X_1 X_2^3 \\ \sum X_1^2 & \sum_{} X_1^3 & \sum_{} X_1^2 X_2 & \sum_{} X_1^3X_2 & \sum_{} X_1^4 & \sum_{} X_1^2 X_2^2 \\ \sum X_2^2 & \sum_{} X_1 X_2^2 & \sum_{} X_2^3 & \sum_{} X_1 X_2^3 & \sum_{} X_1^2X_2^2 & \sum_{} X_2^4 \\ \end{vmatrix} \begin{vmatrix} B_0\\ B_1\\ B_2\\ B_{12}\\ B_{11}\\ B_{22}\\ \end{vmatrix} = \begin{vmatrix} \sum_{} Y_i\\ \sum_{} X_1 Y_i\\ \sum_{} X_2 Y_i\\ \sum_{} X_1 X_2 Y_i\\ \sum_{} X_1^2 Y_i\\ \sum_{} X_2^2 Y_i\\ \end{vmatrix} $$

2) When I solve for the $B_{12}$ which is the $X_1X_2$ coefficient, if that is ~ 0 or very small, then that means there is very little interaction between X & Y?

1

There are 1 best solutions below

0
On BEST ANSWER

Yes, your derivation is correct.

$B_{12}$ being small (compared to $B_{11}$ and $B_{22}$) indeed indicates that the axis of the paraboloid are close to the coordinate axis.