Problem Description
Prove that for any matrix there exists a unitary matrix and a lower-triangular matrix such that .
Source: Stepic Linear Algebra: Problems and Methods.
Problem Dissection
The proof it is almost equal to the QR decomposition:
Case 1: M is non degenerate
Let Q be a matrix of the normalized columns in reverse order of M using Gram-Schmidt orthogonalization algorithm.
In other words: If where is a column vector, after Gram-Schmidt algorithm, we get an ONB .
Remember that in the i-th iteration of Gram-Schmidt algorithm, we subtract from the i-th vector the projections of all the previous vectors, so the transition matrix S’ from M to Q’ is upper triangular
If we reverse the order of the columns in S’ we get a lower triangular matrix S. However, now S will transform the matrix to
So since S is lower triangular and non degenerate, it’s inverse will be lower triangular too: .
It just remains to prove that Q is unitary. If Q is unitary, then ( is the identity matrix). note that the cell of is the inner product of the column i with the conjugate of the column j (both columns from matrix Q).
Note that each pair of columns of Q are perpendicular (because it is an ONB) so if , then
.
Also note that if , then is just the modulo of the column i, and it is one (again by definition of an ONB). So is the identity matrix and therefore Q is unitary.
Case 2: M is degenerate
Let’s apply the same algorithm, but just with the linearly independent columns of M, and complete the remaining columns in order to get a base of