Skip to main content
15 events
when toggle format what by license comment
Dec 18, 2010 at 9:18 vote accept Graviton
Dec 16, 2010 at 7:16 comment added mpiktas Matrix $A$ dimension is $n\times k$, matrix $Q$ dimension is $n\times n$, matrix $R$ dimension is $n\times k$. Vector $x$ dimension is $k\times 1$, vector $b$ dimension is $n\times 1$, so $Rx$ dimension is $n\times 1$, and $Q^Tb$ dimension is $n\times 1$. Everything is ok.
Dec 16, 2010 at 6:40 comment added Graviton @mpiktas, I think your above formulation is still wrong. Take this equation: $Rx=Q^Tb$. Note that the LHS the dimension is $n \times 1$, but the RHS the dimension is $k \times 1$. There is a dimension mismatch here.
Dec 15, 2010 at 11:43 comment added mpiktas @Ngu Soon Hui. When we get to the matrix $R_1$ and vector $b_1$ we get the square linear system. Since determinant of $R_1$ is non zero, the solution always exists. If $Q^Tb$ is not of the form I mentioned, then the solution of the original system does not exist.
Dec 15, 2010 at 11:41 history edited mpiktas CC BY-SA 2.5
added the explanation how to calculate the solution
Dec 15, 2010 at 11:40 comment added Graviton @mpiktas, that's the problem: without $x$, how to check whether the equation $Rx=Q^Tb$ is fulfilled?
Dec 15, 2010 at 11:34 comment added mpiktas @Ngu Soon Hui, you do not have to if you only want to check whether $x$ is the solution.
Dec 15, 2010 at 11:22 comment added Graviton @mpiktas, even with your fix, I still fail to see how to compute $x$.
Dec 15, 2010 at 9:36 comment added mpiktas OK, I fixed the answer. Actually there is a slight typo in the question, $b$ should be $n\times 1$ vector, not $k\times 1$.
Dec 15, 2010 at 9:33 history edited mpiktas CC BY-SA 2.5
added 460 characters in body
Dec 15, 2010 at 9:17 comment added mpiktas @Ngu Soon Hui, sorry my answer is incorrect. I will see if I can fix it.
Dec 15, 2010 at 9:16 comment added mpiktas Yes, you are right. You cannot even multiply $Q$ by $b$, since the dimensions do not match. The answer I think still lies in decomposition of $A$, but it should be rephrased better.
Dec 15, 2010 at 9:16 comment added Graviton @mpiktas, how to compute the $x$ in this case?
Dec 15, 2010 at 9:03 comment added J. M. ain't a mathematician mpiktas, that's for least squares. For instance: if you QR decompose the 3-by-2 example of Ngu (call the original matrix $\mathbf A$), and let $\mathbf b=(2\quad 3\quad 4)^T$ and $\mathbf x=\left(0\quad \frac1{10}\right)^T$ , your equation doesn't work, even if the $\mathbf x$ I gave minimizes $\|\mathbf A\mathbf x-\mathbf b\|_{\infty}$
Dec 15, 2010 at 8:09 history answered mpiktas CC BY-SA 2.5