A real symmetric positive semi-definite matrix $A$ can be decomposed in the form

$A = P^TLDL^TP$,

where $P$ is a permutation matrix, $L$ is a lower unit triangular matrix and $D$ is a diagonal matrix.


(a) Is this decomposition unique?

(b) Let $B = \Pi^TA\Pi$, where $\Pi$ is a permutation matrix. It follows that

$B = \Pi^TP^TLDL^TP\Pi = Q^TLDL^TQ$,

where $Q = \Pi P$ is a permutation matrix, too. If the LDLT decomposition is unique, this means that it is invariant to simultaneous permutations of rows and columns of a matrix, right?


No, it is not unique. For instance, for $A=I$, $L=D=I$, and any $P$ work. Or $A=0$, and then $D=0$ and any $L,P$ work.

The version without the permutation matrix is unique for positive definite matrices, because it is essentially a diagonally scaled LU, and it is a classical result that LU is unique for matrices with nonsingular leading principal submatrices, and this class includes positive definite matrices.

  • $\begingroup$ Thanks. Actually, what is important to me is uniqueness up to a permutation. This is the purpose of question (b). $A = 0$ is kind of pathological example. Is $LDL^T$, or $L$ or $D$ unique for positive semi-definite matrices (except for $A = 0$)? I mean, $P$ is obviously not unique because of (b), so the question is what about the other matrices? $\endgroup$ – user3749105 Feb 14 '16 at 21:54
  • $\begingroup$ @user3749105 There are non-zero counterexamples as well. For instance, take $\begin{bmatrix}0 &0 & 0 \\ 0 & 0 & 0 \\ 0 & 0 & 1\end{bmatrix}$. $\endgroup$ – Federico Poloni Feb 14 '16 at 22:04
  • $\begingroup$ @Poloni: $D$ seems unique, or not? $\endgroup$ – user3749105 Feb 14 '16 at 22:17
  • $\begingroup$ @user3749105 No -- just take a random positive definite matrix, and different choices for $P$ will give different factors $D$. I don't really think there is any meaningful uniqueness result, apart from the one I have stated in my answer ($LDL^\top$ without permutation, positive definite $A$). $\endgroup$ – Federico Poloni Feb 14 '16 at 22:29

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.