Are the two different Forms of the damped pseudo-inverse (damped Moore–Penrose inverse) equal?

405 Views Asked by At

To generalize the concept of the inverse to non square matrix the pseudo inverse (also called Moore–Penrose inverse) can be used:

For tall matrices with full column rank: $$A^+ = (A^T \cdot A)^{-1} A^T$$

For wide matrices with full row rank: $$A^+ = A^T(A \cdot A^T)^{-1}$$

If A has not full rank(row or column) and is therefore singular the inverse of in the above equation is not defined anymore. To arrive at the damped Moore-Penrose pseudoinverse we can therefore introduce a damping term into the two equations above to make sure the inverse is defined $$A_d^+ = (A^T \cdot A + \lambda^2 \mathbb{I})^{-1} A^T$$ $$A_d^+ = A^T(A \cdot A^T + \lambda^2 \mathbb{I})^{-1}$$

Are the two formulas above equal for equal lambda and a general real matrix A? After trying out several examples it seems to me that this could be the case. If it is true, how would you formally proof this? $$A_d^+ = (A^T \cdot A + \lambda^2 \mathbb{I})^{-1} A^T= A^T(A \cdot A^T + \lambda^2 \mathbb{I})^{-1}$$