Weighted Least Squares Solution

AI Thread Summary
The discussion revolves around solving a weighted least squares problem where certain measurements are trusted more than others. The user seeks to create a weighting matrix that applies a factor of two to the first and third rows while leaving the second and fourth rows unchanged. They initially attempted using a diagonal matrix but found it ineffective for their specific requirements. Ultimately, they resolved their issue independently but noted the importance of sharing solutions for future reference. The conversation highlights the complexities of applying weights in matrix equations.
samgrace
Messages
27
Reaction score
0

Homework Statement


\begin{bmatrix}
3x_{1}& 7x_{2}& 4x_{3} \\
3x_{1}& 4x_{2}& 5x_{3} \\
x_{1}& 10x_{2}& 8x_{3} \\
8x_{1}& 8x_{2}& 6x_{3} \\
\end{bmatrix}
=
\begin{bmatrix}
26 \\
16 \\
33 \\
46 \\
\end{bmatrix}
the measurements represented by equations 1 and 3 above can be trusted more than those represented by equations 2 and 4 and are given twice the weight.

Write down an explicit matrix form for the system of equations.

Solve it using Matlab.

However all I really need is to find the weighting factor, I can do the rest from there, struggling to see how I can weight the first and third rows by a factor of two, whilst simultaneously leaving the first and fourth alone.

Homework Equations



I am going to use

Ax = b
e = W(Ax-b)
so

e^{T}e = (Ax-b)^T*W^T*W*(Ax-b)

so A^T*W^T*W*A*x = A^T*W^T*W*b

basically multiply that out and solve via guassian elimination for x

The Attempt at a Solution



e = W(Ax-b)

\begin{bmatrix}
e \\
e \\
e \\
e \\
\end{bmatrix}

=
\begin{bmatrix}
?& ?& ?& ?& \\
?& ?& ?& ?& \\
?& ?& ?& ?& \\
?& ?& ?& ?& \\
\end{bmatrix}

*

\begin{bmatrix}
3x_{1}& 7x_{2}& 4x_{3}& -26& \\
3x_{1}& 4x_{2}& 5x_{3}& -16& \\
x_{1}& 10x_{2}& 8x_{3}& -33& \\
8x_{1}& 8x_{2}& 6x_{3}& -46& \\
\end{bmatrix}

I have tried various combinations of 4x4 matrices for the ? matrix (weighting matrix) that will result in the weighting factor needed, e.g a diagonal 4x4 matrix of 2's, works however the other rows get multiplied by this as well. Please inform me of how to find this.
 
Physics news on Phys.org
samgrace said:
e.g a diagonal 4x4 matrix of 2's, works however the other rows get multiplied by this as well.
Then you should use "1" instead of "2" at some places.
 
Need all of the elements in rows 1 and 3 to be multiplied by two and this cannot be acheived if some are ones. Also need all the elements in rows 2 and 4 to remain the same and this cannot be acheied if some are twos. Tried an algebraic finding of the weighing coefficients but to no avail.
 
No need for an answer I have now solved it
 
samgrace said:
No need for an answer I have now solved it
You can help others with the same question in the future if you post the solution here.
 
samgrace said:
No need for an answer I have now solved it

Did you use 2 or √2 in your matrix W? Do you see why this is not a silly question?
 

Similar threads

Back
Top