Show that the matrix [ A A^† O B^H A^† O ] is a projection but not an orthogonal

Show that the matrix [ A A^† O B^H A^† O ]...

Key Concepts

Adjoint/Hermitian Transpose

The adjoint or Hermitian transpose of a matrix (denoted A* or A^H) involves taking the transpose of the matrix and then the complex conjugate of each entry. In real vector spaces, this concept reduces to simply the transpose. The Hermitian transpose is crucial in defining self-adjoint matrices, as a matrix equals its own adjoint if and only if it is symmetric (or Hermitian), a necessary condition for an orthogonal projection.

Moore-Penrose Pseudoinverse

The Moore-Penrose pseudoinverse is a generalization of the matrix inverse which exists for any matrix, even when the matrix is not square or is singular. Denoted typically by A†, when multiplied by A (in the appropriate order), it produces a projection matrix that projects onto the column space of A. This construction is essential in least squares problems and in dealing with linear systems that do not have unique solutions.

Projection Matrix

A projection matrix is a matrix P that satisfies the idempotence property, meaning that P^2 = P. This property ensures that once a vector is projected onto a subspace, further applications of the matrix do not change it. Projection matrices are widely used in linear algebra to decompose vector spaces into subspaces and their complements.

Idempotence

Idempotence is a property of an operator or matrix whereby applying it twice is equivalent to applying it once. In the context of projection matrices, idempotence (P^2 = P) confirms that the operation of projecting a vector onto a subspace remains stable under repeated projections.

Orthogonal Projection

An orthogonal projection is a projection that is also self-adjoint (or symmetric in the real case), meaning that P = P*. This additional property guarantees that the projection minimizes the distance between the original vector and the target subspace, ensuring that the error vector is orthogonal to the subspace onto which the projection is made.

Transcript

00:01 We are given a special kind of matrix, and we are asked to prove statements about this matrix.

00:12 We are given that b is an n -by -ensometric matrix such that b squared is equal to b.

00:19 And we're told that this is a projection matrix, is what this is called, or an orthogonal projection matrix in some cases.

00:39 And we're told that if y is some vector in rn will denote by y -hat matrix b times the vector y -hat so y -hat is another vector and z will be y -minus y -hat which is another vector in part a we are asked to show that z is orthogonal to y -hat to do this we have z dotted with y hat is equal to well we can write both of these in terms of y so z is equal to y minus y hat dotted with and then y hat is equal to b y by definition and this is then equal to y minus b y dotted with b y again and we have by the addativity of the dot product, this is equal to y dotted with by minus b y dotted with b y dotted with b y.

03:59 So we have this is equal to y dotted with b y minus.

04:12 And then since b is an n by n symmetric matrix, we have that this is going to be equal to by a previous exercise, y dotted with b of b of y, or b times b times y.

04:32 This is equal to y dotted with b y minus y dotted with b squared y.

04:45 We recall that b squared is equal to b.

04:47 So this is equal to y dotted with b y minus y dotted with b y dotted with b y.

04:55 And these are both scalers, so this simply reduces to zero.

05:00 And so we've shown that z and y hat are orthogonal.

05:18 In part b, we are given that w is the column space of b, and we are asked to show that y is the sum of a vector in w and a vector in w perp.

05:55 Or we call this w complement.

06:00 So to prove this, we have that y is some vector in rn, and so y, y hat dotted with bu...