# 8. Hilbert spaces and operators#

Recommended reference: Griffiths [Gri95], in particular Section 3.1.

So far, we have been considering vectors in an \(n\)-dimensional real or
complex space. In quantum mechanics, one encounters
infinite-dimensional vector spaces as well. In particular, the state
of a quantum system is represented mathematically by a (unit) vector
in a *Hilbert space*.

## 8.1. Vector spaces#

As is customary in quantum physics and quantum algorithms, we use
*bra-ket notation* for vectors in Hilbert spaces.

(Vector space)

A (real or complex) *vector space* consists of

a set of vectors, denoted by symbols like \(\ket\alpha\)

a distinguished vector \(\mathbf{0}\) called the

*zero vector*1a way of

*adding vectors*, i.e. given two vectors \(\ket\alpha\) and \(\ket\beta\) we can obtain a third vector \(\ket\alpha+\ket\beta\)a way of

*multiplying a vector by a scalar*: given a scalar \(c\) (a real or complex number, depending on whether we consider a real or complex vector space) and a vector \(\ket\alpha\) we can obtain a vector \(c\ket\alpha\)

such that a number of properties hold:

\(0\ket\alpha=\mathbf{0}\)

\(1\ket\alpha=\ket\alpha\)

\(\mathbf{0}+\ket\alpha=\ket\alpha\)

\(\ket\alpha+\ket\beta=\ket\beta+\ket\alpha\)

\((\ket\alpha+\ket\beta)+\ket\gamma=\ket\alpha+(\ket\beta+\ket\gamma)\)

\(c(\ket\alpha+\ket\beta)=c\ket\alpha+c\ket\beta\)

\((c+d)\ket\alpha = c\ket\alpha + d\ket\alpha\)

\(c(d\ket\alpha)=(cd)\ket\alpha\)

(Basis, dimension)

A *basis* for a vector space is a collection of vectors
\((\ket{e_i})_i\) such that every vector can be written in exactly one
way as a linear combination of the \(\ket{e_i}\). The number of
vectors in a basis is called the *dimension* of the space.

As in Basis transformations, any two bases contain the same number of vectors, so the dimension does not depend on the choice of bases.

The simplest examples are the \(n\)-dimensional vector spaces \(\RR^n\) and \(\CC^n\) with coordinatewise addition and scalar multiplication.

An important special case is the two-dimensional space \(\CC^2\). In bra-ket notation, various symbols are used for the standard basis vectors \(\binom{1}{0}\) and \(\binom{0}{1}\) in this space, such as

\(\ket+\) and \(\ket-\);

\(\ket0\) and \(\ket1\) (in the context of quantum computing);

\(\ket\uparrow\) and \(\ket\downarrow\) (in the context of spins).

## 8.2. Hermitian inner products and Hilbert spaces#

(Hermitian inner product)

A *Hermitian inner product* assigns to vectors \(\ket\alpha\) and
\(\ket\beta\) a complex number \(\braket{\alpha|\beta}\) such that the
following properties hold:

\(\braket{\alpha|\beta}=\overline{\braket{\beta|\alpha}}\);

\(\braket{\alpha|\alpha}\ge0\), and \(\braket{\alpha|\alpha}=0\) holds precisely when \(\ket\alpha=\mathbf{0}\) (note that \(\braket{\alpha|\alpha}\) is a real number because of the first property);

\(\bra\alpha(\ket\beta+\ket\gamma) = \braket{\alpha|\beta} + \braket{\alpha|\gamma}\);

\(\bra\alpha(c\ket\beta)=c(\braket{\alpha|\beta})\);

(Norm of a vector)

The *norm* or *length* of a vector \(\ket\alpha\) in a Hilbert space
is defined as the real number

Note

The properties above imply that for any scalar \(c\), the inner product
of \(c\ket{\alpha}\) with \(\ket\beta\) equals \(\bar
c\braket{\alpha|\beta}\). In other words, while the inner product is
(complex) linear in its second argument, it is only *conjugate linear*
in its first argument. This is the usual convention in physics. It
differs from the convention in mathematics, where inner products are
usually linear in the first argument and conjugate linear in the
second argument.

(Hilbert space)

A *Hilbert space* is a complex vector space \(\HH\) equipped with a
Hermitian inner product \(\braket{\enspace|\enspace}\) and satisfying a
condition called *completeness*.

We will not make the notion of completeness precise in these notes. Roughly speaking, it means that we do not have to worry too much about convergence of sequences or series in our Hilbert space, and there is a notion of infinite linear combinations.

We take \(\HH=\CC^n\) and define an inner product by

Check for yourself that this satisfies all the properties of a Hermitian inner product.

Consider the space \(L^2(\RR)\) of square-integrable functions on the real line, i.e.\ functions \(f\colon\RR\to\CC\) such that the integral \(\int_{-\infty}^\infty|f(x)|^2 dx\) exists and is finite. This is a Hilbert space with inner product

Consider the space \(\ell^2\) of sequences \(z=(z_0,z_1,\ldots)\) such that

Then \(\ell^2\) is a Hilbert space with inner product

## 8.3. The dual Hilbert space#

With any Hilbert space \(\HH\), we can associate a *dual* Hilbert space
\(\HH^*\). In the bra-ket formalism used in quantum mechanics, for
every “ket vector” \(\ket\alpha\) in \(\HH\) we have a dual “bra vector”
\(\bra\alpha\) in \(\HH^*\). This correspondence has the following
properties:

vector in \(\HH\) |
corresponding vector in \(\HH^*\) |
---|---|

ket vector \(\ket\alpha\) |
bra vector \(\bra\alpha\) |

\(\ket\alpha+\ket\beta\) |
\(\bra\alpha+\bra\beta\) |

\(c\ket\alpha\) |
\(\bar c\bra\alpha\) |

In mathematical language, \(\HH^*\) is the space of continuous linear maps \(\HH\to\CC\). The bra vector \(\bra\alpha\) then represents the linear map that sends \(\ket\beta\) to \(\braket{\alpha|\beta}\).

\(\CC^n\))

(The dual ofIn the case \(\HH=\CC^n\) we can identify \(\HH^*\) with \(\CC^n\) as well; a vector \(\ket\alpha\) in \(\CC^n\) is then simply the same vector. However, it can also be convenient to distinguish \(\HH\) from \(\HH^*\) by denoting a ket vector \(\ket\alpha\) as a column vector and the corresponding bra vector \(\bra\alpha\) by the conjugate transpose of \(\ket\alpha\), so \(\bra\alpha\) is the row vector whose entries are the complex conjugates of those of \(\alpha\). In \(\CC^2\), for example, we have

This allows us to view \(\bra\alpha\) as a linear map sending \(\ket\beta\) to the scalar obtained by multiplying the row vector \(\bra\alpha\) by the column vector \(\ket\beta\).

Note

If you know about continuous linear maps of Hilbert spaces, here is a
useful connection between the mathematical definition of \(\HH^*\) and
the bra-ket formalism: the *Riesz representation theorem* states that
every continuous linear map \(\HH\to\CC\) is of the form
\(\braket{\alpha|\enspace}\) for some \(\alpha\) in \(\HH\).

## 8.4. Orthonormal bases#

(Orthonormal basis)

An *orthonormal basis* of a Hilbert space \(\HH\) is a collection of
vectors \((\ket{e_i})_{i\in I}\), where \(I\) is some index set, such that

the system is orthonormal: \(\braket{e_i|e_j} = \begin{cases} 1&\quad\text{if }i=j,\\ 0&\quad\text{if }i\ne j;\end{cases}\)

the system spans \(\HH\): every vector in \(\HH\) can be written as a (possibly infinite) linear combination of the \(\ket{e_i}\).

Consider the space \(L^2([0,1])\) of square-integrable functions on the unit interval. An orthonormal basis for this space consists of the functions \(e_n(x)=\exp(2\pi i n x)\) for all integers \(n\). This reflects the fact that every “nice” function on \([0,1]\) can be expressed as a Fourier series.

## 8.5. Operators#

An *operator* or *linear map* on a Hilbert space \(\HH\) is a map

such that

\(T(c\ket\alpha) = c(T\ket\alpha)\)

\(T(\ket\alpha+\ket\beta) = T\ket\alpha+T\ket\beta\)

Note

It is also possible to define linear maps between different Hilbert spaces, but we will not go into this.

Note

In physics, operators are often denoted with a hat, so what we call \(T\) would be written as \(\hat T\). One reason for this is to distinguish a physical quantity (like momentum or energy) from the corresponding mathematical operator. Since we do not consider physical quantities in this module, we do not use the hat notation.

Operators form a vector space: the sum of two operators \(T\) and \(U\) is defined by the formula

and multiplication of an operator \(T\) by a scalar \(c\) is defined by the formula

Like matrices (but unlike vectors), operators can be *composed*: if
\(T\) and \(U\) are two operators, then \(TU\) is the operator defined by

## 8.6. Matrix entries#

In physics, quantities of the form

are often of special interest. These are called *matrix entries* of
\(T\), especially if \(\ket\alpha\) and \(\ket\beta\) are basis vectors in
some fixed orthonormal basis of \(\HH\).

Consider a linear map \(A\) on \(\CC^n\), viewed as a matrix. Taking the matrix entries as above, letting \(\alpha\) and \(\beta\) range over the standard basis vectors, we obtain the usual matrix entries of \(A\); see Exercise 8.2.

Note

A remark on notation: you will often see notation like \(\braket{\alpha|T\beta}\). This is literally meaningless: the operator \(T\) can be applied to vectors, which we denote by \(\ket\beta\), while the symbol \(\beta\) is just a label. However, in practice it is very convenient and completely unambiguous to write \(\braket{\alpha|T\beta}\) instead of the strictly speaking more correct \(\bra\alpha(T\ket\beta)\).

Similarly, we will write \(\braket{T\alpha|\beta}\) to mean the inner product of \(T\ket\alpha\) with \(\ket\beta\). Extending this notation to scalars \(c\), we can also write

## 8.7. The adjoint of an operator#

Given an operator \(T\) on a Hilbert space \(\HH\), there exists an operator \(T^\dagger\) with the property

The operator \(T^\dagger\) is called the *adjoint* of \(T\).

(Properties of the adjoint)

For all operators \(T\), \(U\) and scalars \(c\), we have

\((cT)^\dagger = \bar c T^\dagger\)

\((TU)^\dagger = U^\dagger T^\dagger\)

\((T^\dagger)^\dagger = T\)

Take \(\HH=\CC^n\) with the standard inner product. For a matrix \(A\), we write \(A^\dagger\) for the conjugate transpose of \(A\). We extend this to vectors by viewing column vectors as \(n\times 1\)-matrices and row vectors as \(1\times n\)-matrices. For all \(\ket\alpha,\ket\beta\in\HH\) we then have

This shows that the adjoint of the operator \(A\) corresponds to the conjugate transpose of \(A\) viewed as a matrix.

## 8.8. Hermitian operators#

(Hermitian operator)

An operator \(T\) on a Hilbert space \(\HH\) is called *Hermitian* if it
is equal to its own adjoint, i.e.

It is clear from the definition that all real symmetric matrices are Hermitian.

The matrix

is Hermitian.

The following two results are of fundamental importance in quantum mechanics. Together, they explain why in the mathematical formalism of quantum mechanics, Hermitian operators correspond to physical observables.

The eigenvalues of a Hermitian operator \(T\) are real.

Proof. Suppose \(\lambda\) is an eigenvalue of \(T\) and \(\ket\alpha\) is a corresponding eigenvector. Then we have

Since \(\braket{\alpha|\alpha}\) is non-zero, we can divide by it and obtain \(\lambda=\bar\lambda\).

(Spectral theorem)

If \(T\) is a Hermitian operator on a Hilbert space \(\HH\), then there exists an orthonormal basis of \(\HH\) that consists of eigenvectors for \(T\).

## 8.9. Unitary operators#

(Unitary operator)

An operator \(T\) on a Hilbert space \(\HH\) is called *unitary* if the
adjoint \(T^\dagger\) is a two-sided inverse of \(T\), i.e.

Alternatively, \(T\) is unitary whenever \(T\) preserves inner products; see Exercise 8.3 for a precise statement.

For any angle \(\phi\), the matrix

is unitary.

## 8.10. Exercises#

Check that the inner product defined in Example 8.2 is indeed a Hermitian inner product on \(\CC^n\).

Consider a matrix \(A=(A_{i,j})_{i,j=1}^n\) and the standard basis \((\ket{e_i})_{i=1}^n\) of \(\CC^n\). Check that the matrix entry \(\braket{e_i|A|e_j}\) is simply the actual matrix entry \(A_{i,j}\).

Show that an operator \(T\) is unitary precisely when it satisfies \(\braket{T\alpha|T\beta}=\braket{\alpha|\beta}\) for all vectors \(\ket\alpha\), \(\ket\beta\).

The *Cauchy–Schwarz inequality* states that for vectors
\(\ket\alpha,\ket\beta\) in a Hilbert space we have

Show this using the basic properties of the inner product.

The *triangle inequality* states that for vectors
\(\ket\alpha,\ket\beta\) in a Hilbert space we have

Derive the triangle inequality from the Cauchy–Schwarz inequality.

Check that the *Pauli matrices*

are both Hermitian and unitary.

- 1
We avoid denoting the zero vector by \(\ket0\) because this symbol is often used for a distinguished unit vector, as in Example 8.1.