Say I have a $3 \times 3$ matrix $C$ with singular value decomposition $$ C = USV^T $$ where $U = [u_1 \hspace{1mm} u_2 \hspace{1mm} u_3], V = [v_1 \hspace{1mm} v_2 \hspace{1mm} v_3]$ are orthogonal matrices ($u_i,v_i$ column vectors) and $S$ is the singular values diagonal matrix.

I did an experiment using Matlab and found that the matrices $u_iv_i^T$ has norm ($\|\cdot\|_2$) approximately equal to $1$ for each $i=1,2,3$.

My question is that, mathematically, are the matrix 2-norms of $u_iv_i^T$ indeed equal to $1$?

Matlab code


Let $\mathbf{u}_i = \begin{bmatrix}u_{1i} \\ u_{2i} \\ u_{3i}\end{bmatrix}$. What you are computing is $$\mathbf{u}_i \mathbf{v}_j^T = \begin{bmatrix}u_{1i}\mathbf{v}_j^T \\ u_{2i}\mathbf{v}_j^T \\ u_{3i}\mathbf{v}_j^T\end{bmatrix},$$ which is a rank-1 matrix with only one non-zero singular value. Consequently, $$||\mathbf{u}_i \mathbf{v}_j^T||_2 = ||\mathbf{u}_i \mathbf{v}_j^T||_{F} = 1.$$

Note that $||\cdot ||_F$ denotes the Frobenius norm of a matrix. Also, $$||\mathbf{u}_i \mathbf{v}_j^T||_{F} = \sqrt{|u_{1i}|^2||\mathbf{v}_j||^2+|u_{2i}|^2||\mathbf{v}_j||^2+|u_{3i}|^2||\mathbf{v}_j||^2}=\sqrt{|u_{1i}|^2+|u_{2i}|^2+|u_{3i}|^2}=1.$$

For more information, visit https://en.wikipedia.org/wiki/Matrix_norm#Matrix_norms_induced_by_vector_norms


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.