Block Matrix Formulas
Block Matrix Formulas
Block Matrix Formulas
Abstract
We derive a number of formulas for block matrices, including the block matrix
inverse formulas, determinant formulas, psuedoinverse formulas, etc.
If you find this writeup useful, or if you find typos or mistakes, please let me
know at John.Gubner@wisc.edu
1. Preliminary Observations
Given a block matrix
A B
Φ := ,
C D
if D is invertible, then the Schur complement of D is
Σ := A − BD−1C.
I BD−1
Σ 0 I 0
Φ= . (1)
0 I 0 D D−1C I
Then det Φ = det Σ det D, and we see that given D is invertible, Φ is invertible if and
only if Σ is invertible. When Φ is invertible, taking the inverse of (1) yields
0 Σ−1 0 I −BD−1
I
Φ−1 =
−D−1C I 0 D−1 0 I
" #
Σ−1 −Σ−1 BD−1
= . (2)
− D−1CΣ−1 D−1CΣ−1 BD−1 + D−1
1
BlockMatrixFormulas.tex July 7, 2015
and
I A−1 B
I 0 A 0
Φ= . (3)
CA−1 I 0 Θ 0 I
Then det Φ = det A det Θ, and we see that given A is invertible, Φ is invertible if and
only if Θ is invertible. If Φ is invertible,
I −A−1 B A−1 0
I 0
Φ−1 =
0 I 0 Θ−1 −CA−1 I
" #
A−1 + A−1 BΘ−1CA−1 −A−1 BΘ−1
= . (4)
− Θ−1CA−1 Θ−1
2. Results
Using (1)–(4), there are several identities that can easily be found.
2
BlockMatrixFormulas.tex July 7, 2015
Then
G∗ G G ∗ H
∗
M M= ,
H ∗G H ∗H
and by (8),
" #
∗ −1 (G∗ PH⊥ G)−1 −(G∗ PH⊥ G)−1 G∗ H(H ∗ H)−1
(M M) = , (9)
− (H ∗ PG⊥ H)−1 H ∗ G(G∗ G)−1 (H ∗ PG⊥ H)−1
where PH⊥ := I − H(H ∗ H)−1 H ∗ and PG⊥ := I − G(G∗ G)−1 G∗ . It now easily follows
that " # " #
∗ −1 ∗ (G∗ PH⊥ G)−1 G∗ PH⊥ (PH⊥ G)†
(M M) M = = . (10)
(H ∗ PG⊥ H)−1 H ∗ PG⊥ (PG⊥ H)†
A − BD−1 BT 0 I −BD−1
A B I 0
= .
0 D 0 I BT D −D−1 BT I
Suppose the center matrix on the right is known to be positive semidefinite. Then
since the matrices on either side are transposes of each other, it is easy to see that
the matrix on the left-hand side is also positive semidefinite. It then follows that
A − BD−1 BT is positive semidefinite; in symbols, A BD−1 BT .
3
BlockMatrixFormulas.tex July 7, 2015
Given a 2 × 2 block matrix U, we denote its upper-left block by {U}ULB . For the
matrix Φ defined earlier, {Φ}ULB = A. We show below that if Φ = Φ∗ , then
More specifically, if
P Q
U := ,
R S
then
{U ∗ Φ−1U}ULB P∗ A−1 P.
To derive the result, we first use (4) to compute the left-hand block column of Φ−1U.
We get " #
{A−1 + A−1 BΘ−1CA−1 }P − A−1 BΘ−1 R
.
− Θ−1CA−1 P + Θ−1 R
It follows that
" #
{A−1 + A−1 BΘ−1CA−1 }P − A−1 BΘ−1 R
∗ −1
{U Φ U}ULB = P∗ R∗
− Θ−1CA−1 P + Θ−1 R
= P∗ A−1 P + P∗ A−1 BΘ−1CA−1 P
− P∗ A−1 BΘ−1 R − RΘ−1CA−1 P + R∗ Θ−1 R
= P∗ A−1 P + (B∗ A−1 P − R)∗ Θ−1 (CA−1 P − R).
References
[1] W. W. Hager, “Updating the inverse of a matrix,” SIAM Review, vol. 31, no. 2, pp. 221–239, 1989.
[2] Wikipedia, “Schur complement — Wikipedia, The Free Encyclopedia,” [Online]. Available:
https://en.wikipedia.org/w/index.php?title=Schur_complement&oldid=
658902757, accessed July 6, 2015.
[3] Wikipedia, “Woodbury matrix identity — Wikipedia, The Free Encyclopedia,” [Online].
Available: https://en.wikipedia.org/w/index.php?title=Woodbury_matrix_
identity&oldid=661757721, accessed July 6, 2015.