Suppose, given a matrix $\textbf{A} \in \mathbb{C}^{m \times n}$ and a vector $\textbf{b} \in \mathbb{C}^{n}$, I want to find the minimal norm solution of

$$\min_{\textbf{x}}\|\textbf{A}\textbf{x} - \textbf{b} \|_{2} $$

with the condition that $\textbf{A}^{*}\textbf{b} = \textbf{0}_{n}$.

If $rank(\textbf{A}) =n$, then the solution is

$$\textbf{y} = (\textbf{A}^{*}\textbf{A})^{-1}\textbf{A}^{*}\textbf{b}.$$

This would give $\textbf{y} = \textbf{0}$ with the stated condition. However, for $\textbf{A}$ of arbitrary rank, the the minimal norm solution for this problem is

$$\textbf{y} = \textbf{A}^{\dagger}\textbf{b}$$

Where $\textbf{A}^{\dagger}$ is the pseudoinverse.

Can we say anything about $\textbf{y}$ with stated condition in the general case, or is it arbitrary


It is known that a vector $\hat {\mathbf x}$ is a least squares solution of the system $$ A {\mathbf x}= {\mathbf b} $$ iff $$A^*A\hat {\mathbf x}=A^* {\mathbf b}.$$ Thus, if the condition $A^* {\mathbf b}={\mathbf 0}$ is satisfied, then $\hat {\mathbf x}= {\mathbf 0}$ is a least squares solution. Obviously, it has the minimum possible norm (zero).

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.