$X_{n \times p}$ is a real, thin ($n>p$) rectangular matrix of rank $p$, so $X^T X$ is full rank. The Moore-Penrose pseudoinverse of $X$ is given by $X^+=(X^TX)^{-1}X^T$.

Let's now define $W=XA$

($X$ is linearly transformed by $A_{p \times k}$ to produce a $W_{n \times k}$). $W$ is of rank $k$ (so $W^TW$ is full rank).

I need to compute the Moore-Penrose Pseudoinverse of $W$ for many different $A$ matrices efficiently. Is there an identity or a decomposition that allows expressing $W^+$ as some function of $X^+$ or $(X^TX)^{-1}$?


due to distributive order reversal rule for transposing multiple and associativity of multiplication , $$W^+=(A^T (X^TX) A)^{-1} A^T X^T$$ not too much opportunity for pre-calculation it seems


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.