매우 중요

Theorem

Let \(V\) be a subspace of \(\mathbb{R}^n\) with orthogonal projection \(P\), and let \(V_1,\ldots,V_k\) be subspaces of \(V\) with orthogonal projections \(P_1,\ldots,P_k\).

If \(V=V_1\oplus\cdots\oplus V_k\) (orthogonal decomposition of \(V\)), then

  1. \(P=P_1+\cdots+P_k\)

    (so for \(y \in V\), \(y=Py=P_1y+P_2y+\cdots+P_ky=y_1+y_2+\cdots+y_k\).)

 

  1. \(\text{dim}(V)\)=\(\text{dim}(V_1)+\cdots+\text{dim}(V_k)\).

 

  

매우 중요

Theorem

If \(A_1,\ldots,A_k\) are symmetric matrices satisfying any two of the following, then they also satisfy the third.

  1. \(A_i\) is idempotent for all \(i\);

  2. A:= \(\sum_{i=1}^{k}A_i\) is idempotent;

  3. \(A_iA_j=0\) for \(i\ne j\).

In this case, let \(V_i=\text{range}(A_i)\). Then, \(V_i \perp V_j\) for \(i \ne j\), and \(\sum_{i=1}^kA_i\) is the orthogonal projection onto \(V=V_1\oplus\cdots\oplus V_k\).

  

   

Proposition

Iif \(V\) and \(W\) are two subspaces, then \[ V \perp W \iff P_V P_W=0. \]

  

Proposition

Let \(U\) and \(W\) be two subspaces. Then, \[ P_UP_W= P_{U\cap W} \iff P_UP_W= P_WP_U, \] where \(P_{U\cap W}\) is orthogonal projection onto \(U\cap W\).

  

Proposition

Let \(V, U, W\) be three subspaces. Then,

\(P_U=P_V-P_W\iff W\subset V \mbox{ and } U=V\cap {W^\perp}\).

  

Proposition

If \(P\) is a projection, then

  1. its eigenvalues are all \(1\) or \(0\).

 

  1. There exists an orthogonal \(\Gamma\) such that \(P=\Gamma D \Gamma'\) where \(D\) is diagonal \(0-1\) matrix.

 

  1. \(\text{tr}(P)=\text{tr}(D)=\text{rank}(D)=\text{rank}(P)\)

  

Proposition

If \(A_1,A_2,\ldots A_k\) are projection matrices whose parewise products are all 0, then there is a single orthogonal matrix \(\Gamma\) s.t. \[ D_i=\Gamma'A_i\Gamma\mbox{ }\mbox{ } \mbox{ for all } i. \] * 각각의 direct sum space \(V_i\)의 Projection \(A_i\)에 대해, 대각행렬 \(D_i\)가 같은 location에서 원소 1값을 동시에 갖지 않는다고 가정하자. 그렇다면 \(D=\sum_{i=1}^k D_i\)를 원소 1과 0만을 갖는 대각행렬이 되고, \(\sum_{i=1}^kD_i=D=\Gamma (\sum_{i=1}^k A_i) \Gamma'\)를 만족한다. 대각행렬 \(D_i\)가 같은 location에서 원소 1값을 동시에 갖지 않기 때문에, \(D_i=\Gamma A_i \Gamma'\)이다.

  

Theorem

If \(A_1,\ldots,A_k\) are symmetric matrices satisfying any two of the following, then they also satisfy the third.

  1. \(A_i\) is idempotent for all \(i\);

  2. A:= \(\sum_{i=1}^{k}A_i\) is idempotent;

  3. \(A_iA_j=0\) for \(i\ne j\).

In addition,

A. \(\text{rank}(A)=\sum_{i=1}^k \text{rank}(A_i)\);

B. There exists an orthogonal matrix \(\Gamma\) such that \(D=\Gamma'A\Gamma\), and \(D_i=\Gamma'A_i\Gamma\) for \(i=1,\ldots,k\) satisfy * \(D,D_1,\ldots,D_k\) are diagonal \(0-1\) matrices; * \(D=\sum_{i=1}^k D_i\); * \(D_iD_j=0\) for \(i\ne j\).