Notes

Vector space

Requirements for a set of entities to form a vector space

  1. Closure under commutative and associative addition
  2. Closure under multiplication of a scalar
  3. The existance of a null vector
  4. Multiplication by unity leaves any vector unchanged
  5. Each vector has a corresponding negative vector

Linear equations

Row-echelon form

  1. All zero rows are at the bottom of the matrix
  2. If two sucessive rows are non-zero, the second row starts with strictly more zeros than the first

Reduced row-echelon form

  1. It’s in row-echelon form
  2. The leading left most non-zero entry in each non-zero row is 1
  3. All other elements of the column in which the leading entry 1 occurs are zeros

Row operations

  1. Interchanging rows
  2. Multiplying row by non-zero scalar
  3. Adding a multiple of one row to another row

Preserves row-equivalence

Gauss-Jordan algorithm / Gaussian elimination

You should know this

Homogenous equations

“Implicit equations with no non-zero constants”

A homogenous system of m linear equations in n unknowns always has a non-trivial solution if m < n

Matrices

Inner product

Hilbert space is basically inner product space for finite dimensions

Orthornormal: \(\braket{\hat{e}_i}{\hat{e}_j} = \delta_{ij}\) (i.e. orthogonal and normalised)

\begin{aligned} \braket{a}{b} & = \braket{b}{a}^\ast \\
& = \sum_i a_i^\ast b_i \text{ (On orthonormal basis)} \\
& = \sum_{ij} a_i^\ast b_j \braket{\hat{e}_i}{\hat{e}_j} \\
\end{aligned}

Inequalities and equalities

Cauchy-Schwarz’s inequality

\[ \vert\braket{a}{b}\vert \leq \Vert a\Vert \Vert b\Vert \]

In 2d, abcosx <= aba

equality when a is a scalar multiple of b

Triangle Inequality

\[ \Vert a+b \Vert \leq \Vert a \Vert + \Vert b \Vert \]

Bessel’s inequality

\[ \Vert a \Vert^2 \geq \sum_i \vert \braket{\hat{e}_i}{a} \vert^2 \] or

\[ \braket{a}{a} \geq \sum_i \vert a_i \vert^2 \]

where \(\hat{e}_i\) is an orthonormal basis

where the equality holds if the sum includes all N basis vectors or if remaining basis vectors have a_i = 0

Parallelogram equality

\[ \Vert a + b \Vert^2 + \Vert a - b \Vert^2 = 2(\Vert a \Vert^2 + \Vert b \Vert^2) \]

Functions of matrices

\[ \exp A = \sum_{n=0}^\infty \frac{A^n}{n!} \]

Derivatives exist.

Operations

Transpose

\[ (AB\ldots G)^T = G^T \ldots C^T B^T A^T \]

Hermitian Conjugate (Complex Transpose)

\[ A^\dagger = (A^\ast)^T = (A^T)^\ast \]

Trace

\[ \Tr A = \sum_i A_{ii} \]

\[ \Tr ABC = \Tr BCA = \Tr CAB \]

(Invariant under cyclic permutations)

\[ \Tr \Omega = \Tr U^\dagger \Omega U \]

Determinant

Minor:

\(M_{ij}\) = determinant of matrix obtained by removing row i and column j

Cofactor:

\(C_{ij} = (-1)^{i+j} M_{ij}\)

Thus a determinant is the sum of the products of the elements along any row or column and their corresponding cofactors

e.g.

\[ \sum_i A_{1i}C_{1i} \]

  • Properties

    1. \(\vert A^T \vert = \vert A \vert\)

    2. If rows of A are interchanged, its determinant changes sign but is unaltered in magnitude

    3. if all the elements of a row or column of A have a common factor, it can be factored out. Also, \(\vert \lambda A \vert = \lambda^N \vert A \vert\)

    4. If any two rows are identical or multiples of one another, det is 0

    5. Determinant is unchanged by adding to the elements of one row any fixed multiple of the elements of another row

    6. Determinant of the product of multiple matrices is invariant under permutations of the matrices

    \[ \vert A \vert = a \cdot (b \times c) \]

    \(\vert A^{-1} \vert = \vert A \vert^{-1}\)

    \[ \det \Omega = \det U^\dagger \Omega U \]

Inverse

Exists when determinant is not zero, i.e. the matrix is non-singular

\[ (A^{-1})_{ik} = \frac{C_{ki}}{\vert A \vert} \]

\[ (ABC)^{-1} = C^{-1}B^{-1}A^{-1} \]

A diagonally dominant matrices are invertible/non-singular

Rank

R(A) = R(A^T) = number of linearly independent vectors row-wise or column-wise

or

Size of the largest square submatrix of A whose determinant is non-zero

Special forms

Diagonal matrix

Having non-zero elements only on the leading diagonal

Also denoted by e.g. A = diag(1,2,-3)

\[ \vert A \vert = \prod_i A_{ii} \]

\[ A^{-1} = \text{diag} (\frac{1}{A_{11}},\frac{1}{A_{22}}, \ldots) \]

if A and B are diagonal, their product is commutative

Lower or upper triangular matrices

\[ \vert A \vert = \prod_i A_{ii} \]

Symmetric and antisymmetric matrices

Symmetric:

\[ A = A^T \]

Anti/Skew-symmetric:

\[ A = -A^T \]

Any matrix can be written as a sum of a symmetric and an antisymmetric matrix

\[ A = \frac12 (A + A^T) + \frac12 (A - A^T) \]

Note that if A is an NxN antisymmetric matrix, |A| = 0 if N is odd

Orthogonal matrix

\[ A^T = A^{-1} \]

\[ \vert A \vert = \pm 1 \]

Hermitian matrices (Complex Symmetric)

Hermitian

\[ A = A^\dagger \]

Anti-Hermitian

\[ A = -A^\dagger \]

\[ A = \frac12 (A+A^\dagger) + \frac12(A-A^\dagger) \]

Unitary matrix (complex orthogonal)

\[ A^\dagger = A^{-1} \]

The determinant and eigenvalues of a unitary matrix have unit modulus

It also represents a linear operator which leaves the norms (inner products) of complex vectors unchanged (i.e. orthonormal-ness)

An operator is unitary iff each of its matrix representations are unitary

The rows may be interpreted as components of n orthonormal vectors similar to how columns are components of n vectors.

Normal matrices

\[ AA^\dagger = A^\dagger A \]

Hermitian matrices and unitary matrices are normal

A Normal matrix is hermitian iff it has real eigenvalues

Vectors and spaces

Completeness relation for orthonormal vectors

\[ \sum_i \ket{i}\bra{i} = I \]

where |i> is any orthonormal basis for the vector space V

Which allows us to represent any operator in the outer product notation. Given A : V \(\rightarrow\) W, and |v_i> and |w_j> are orthornormal bases for V and W

\begin{aligned} A & = I_W A I_V \\
& = \sum_{ij} \bra{w_j}A\ket{v_i} \ket{w_j} \bra{v_i} \end{aligned}

Outer product representation of Unitary U

Define:

\[ \ket{w_i} \equiv U\ket{v_i} \]

Note that that implies:

\[ U = \sum_i \ket{w_i} \bra{v_i} \]

And that if |v_i> and |w_i> are any orthonormal bases, then U defined above is also unitary

Projectors

A class of Hermitian operators. Let W be a k-dimensional vector subspace of the d-dimensional vector space V. Using Gram-Schmidt procedure it is possible to construct an orthonormal basisc for V (|1>,…,|d>)such that |1>,…,|k> is an orthonormal basis for W. The projector by defintion:

\[ P \equiv \sum_{i=k}^k \ket{i} \bra{i} \]

\[ (P_i)_{kl} = \delta_{kl}\delta_{li} \]

The orthogonal complement of P is Q = I - P, which is a projector onto the vector space spanned by |k+1>…|d>

\[ P_i P_j = \delta_{ij} P_j \]

Note that P^2 = P

Operators

Commutator

Definition:

\[ \Omega \Lambda - \Lambda\Omega \equiv [\Omega,\Lambda] \]

Identities:

\[ [\Omega,\Lambda] = \Lambda[\Omega,\theta] + [\Omega,\Lambda]\theta \]

\[ [\Lambda\Omega,\theta] = \Lambda[\Omega,\theta] + [\Lambda,\theta]\Omega \]

Diagonalisable

A diagonal representation for an operator A on vector space V is a representation:

\[ A = \sum_i \lambda_i \ket{i}\bra{i} \] with corresponding eigenvalues \(\lambda_i\)

An operator is diagonalisable is it has a diagonal representation.

Any normal operator M on a vector space V is diagonal wrt some orthonormal basis for V. Also any diagonalisable operator is normal

Active and passive transformations

If all vectors in a space are subject to a unitary transform (Active Transformation)

\[ \ket{V} \to U\ket{V} \] Then the same change would be effected if subjected operators to the change (Passive Transformation) \[ \Omega \to U^\dagger\Omega U \]

Eigenvectors and eigenvalues

\[ Ax = \lambda x \]

\[ A^{-1}x^i = \frac{1}{\lambda_i}x_i \]

The eigenspace corresponding to a eigenvalue v is the set of vectors with eigenvalue v

Normal matrices:

\[ A^\dagger x^i = \lambda_i^\ast x^i \]

If \(\lambda_i \neq \lambda_j\), the eigenvectors \(x^i\) and \(x^j\) must be orthogonal

An eigenvalue corresponding to two or more different eigenvectors (i.e. they are not multiples) are degenerate. or in other words the eigenspace has more than 1 dimensions

Suppose \(\lambda_1\) is k-fold degenerate (for x^1 to x^k), then any linear combination of those x^1 also has an eigenvalue of \(\lambda_1\)

Gram-Schmidt orthogonalisation

\begin{aligned} z^1 & = x^1, \\
z^2 & = x^2 - \braket{\hat{z}^1}{x^2} \hat{z}^1, \\
z^3 & = x^3 - \braket{\hat{z}^2}{x^3} \hat{z}^2 - \braket{\hat{z}^1}{x^3} \hat{z}^1, \\
& \ldots \\
z^k & = x^k - \sum_{i=1}^{k-1} \braket{\hat{z}^i}{x^k}\hat{z}^k \\
\end{aligned}

Where \(\hat{z}^i = \frac{z^i}{\Vert z^i \Vert}\)

Eigen-everything of Hermitian and anti-hermitian matrices

Eigenvalues of hermitian matrices are real

Eigenvalues of anti-hermitian matrices are pure imaginary (-\(\lambda\))

To every hermitian operator, there exists at least a basis consisting of its orthonormal eigenvectors. It is diagonal in this eigenbasis and has its eigenvalues as its diagonal entries.

Eigen-everything of unitary matrix

Eigenvalues of a unitary matrix have unit modulus and are complex.

Assuming no degeneracy, the eigenvectors of a unitary operator are mutually orthogonal

Eigen-everything of a general square matrix

Any N x N matrix with distinct eigenvalues has N linearly independent eigenvectors.

If it has a degenerate eigenvalues, however, then it may or may not have N linearly independent eigenvectors.

Not having linearly independent eigenvectors makes it defective

Simultaneous eigenvectors

Two different normal matrices have a common set of eigenvectors that diagonalises them both if and only if they commute

If an eigenvalue of A is degenerate then not all of its possible sets of eigenvectors will also constitute a set of eigenvectors of B. i.e. If B is nondegenerate but A is degenerate, then all eigenvectors of B are eigenvectors of A.

Determination

Chracteristic function: c(lambda) = det | A - lambda I | Characteristic equation:

\[ \vert A - \lambda I \vert = 0 \]

LHS is the characteristic/secular determinant of A

When the determinant is written as a polynomial equation in \(\lambda\), the coefficient of \(-\lambda^{N-1}\) will be Tr A, i.e. the sum of the roots (recall p+q = -b/a, pq = c/a)

Change in basis

Given S where:

\[ x_i = \sum_{j=1}^N S_{ij}x'_j \]

\[ x = Sx' \]

where S is the transformation matrix associated with the change in basis

and

\[ x' = S^{-1} x \]

The components of x transform inversely to how the basis vectors transform, as the vector x itself much remain unchanged

This applies to linear operators of x:

\[ y = Ax, y' = A’x' \]

=>

\[ y' = S^{-1}ASx' \]

or in other words:

\[ A' = S^{-1}AS \]

which is an example of a similarity transformation

So any property of A which represents some basis independent property of its linear operator will be shared by A'

  1. If A = I, A' = I
  2. |A| = |A'|
  3. Eigenvalues are the same
  4. Trace is unchanged

Unitary S

  1. If the original basis is orthonormal and S is unitary then the new basis is also orthonormal
  2. If A is (anti) hermitian then A' is (anti) hermitian
  3. If A is unitary then A' is also unitary

Diagonalisation of matrices

Define a basis \(x^j\) given by: \[ x^j = \sum_{i=1}^N S_{ij}{\bf e_i}, \]

where \(x^j\) are chosen to be the eigenvectors of the linear operator \(\mathcal{A}\). i.e.

\[ \mathcal{A} x^j = \lambda_j x^j \].

In our new basis, \(\mathcal{A}\) is represented by the matrix \(A' = S^{-1}AS\). The columns of S are the eigenvectors of the matrix A, i.e. \(S_{ij} = (x^j)_i\). Therefore, A' is diagonal with the eigenvalues of \(\mathcal{A}\) as the diagonal elements.

Since we require S to be non-singular, the N eigenvectors of A must be linearly independent and form a basis for the N-dimensional vector space. Any matrix with distinct eigenvalues can be diagonalised by this procedure. If it doesn’t have N linearly independent eigenvectors, then it cannot be diagaonlised.

For normal matrices, the N eigenvectors are indeed linearly independent. For them, S is unitary.

Note:

\[ \det \exp A = \exp(\Tr A) \]

Infinite dimensions

Define a function

Inner product in (countable) infinite dimensions:

\[ \braket{f}{g} = \int_0^L f(x)g(x) \dd{x} \]

Dirac delta

In infinite dimensions,

\[ \braket{x}{x'} = \delta(x-x') \]

if two basis vectors are the same. \[ \delta ‘(x-x’) = \delta(x-x')\dv{x} \]

Dirac delta with fourier transform

\[ \delta(x-x') = \frac{1}{2\pi}\int_{-\infty}^\infty e^{ik(x-x')}\dd{k} \]

\[ \delta(f(x)) = \sum_i \frac{\delta(x_i-x)}{|\dv{f}{x_i}|} \] where x_i are the zeros of f(x)