Let’s look at a case where the LU factorization breaks down. This happens when a pivot becomes equal to 0.

Consider:

Why does the LU algorithm break down?

  • We have because is lower triangular with 1 on the diagonal.
  • det() = 0 if and only if det() = 0. This is equivalent to saying that one of the diagonal entries of is 0. LU and determinant.
  • Furthermore, because and the triangular form of and we have that
  • Assume that all the pivots are non-zero up to step . Then is singular if and only if is zero. This can be stated equivalently as if and only if . LU and determinant.
  • So, if , then all the pivots remain non-zero, and the algorithm completes.

In our example, at step . This is because the top left block of is singular. So we cannot proceed to step .

Triangular factorization, LU algorithm, LU and determinant

Existence of the LU factorization

Theorem: the LU factorization exists if

for all .

This was proved above. A more precise version of this result is as follows.

Theorem: The LU factorization exists if and only if

for all .

More details

We make a few observations.

A matrix is non-singular if and only if , . For non-singular matrices, , , is a necessary and sufficient condition for the existence of an LU factorization.

Assume that has an LU factorization. If , then is a linear combination of the columns , . If and for , then is a linear combination of the columns , .

When , column of is not uniquely defined, and an infinite number of LU factorizations satisfy .

We also have:

Theorem: There exists a unique and if and only if for all .

Variant:

Theorem: There exists a unique non-singular and if and only if for all .