Quantum Mechanics 321

Notes on Vector Notations and Vector Representations

Spring 2014

When you first encounter vectors in physics, usually vectors for displacement, velocity, or
force, you likely encounter the visual representation of a vector as an arrow in space, with
the point of the arrow providing information about the direction of the vector and the
length of the arrow providing information about the magnitude of the
vector. The vector is typically denoted by a bold font letter or letter
with an arrow over it: or ⃗, with the scalar magnitude of the vector
symbolized by | | or just . One could continue to label each vector
with a unique letter, but once you learned how vectors summed, it made
it much easier to catalog and combine them by imposing a coordinate
system, finding the projection of each vector onto the coordinate axes,
and expressing each vector as a sum of these projected vector
components: . The magnitude of these vector components
are known as the scalar components, or just “components” , of the vector. From the geometrical picture
the magnitudes of the vector and its scalar components have the relations | | and
| | .

A further improvement was made by introducing unit vectors for

each axis ̂ and ̂ (sometimes denoted ̂ and ̂ or and ), so that
̂ ̂ ̂ ̂. The choice of coordinate axes
and corresponding unit vectors, or basis vectors, was arbitrary
(though practice in the art of choosing the “right” coordinates made
solving problems easier). One could have equally well chosen a
different coordinate system with different unit vectors (e.g. ̂ and ̂ )
and a different decomposition of each vector (e.g. ̂ ).

Since the specification of a vector required specifying a scalar

multiple of each unit vector, it became possible to represent a vector
by only specifying the scalar components multiplying each vector.
Instead of writing ̂ ̂ ̂ ̂, one can
write or ( ). These sets
of ordered scalars constitute a representation of the vector .

There is a subtle, but important difference between the statement

̂ ̂ and ( ). The former explicitly specifies the basis vectors being used,
while the latter omits any information about the coordinate system. This distinction is clear when
comparing how the vector is decomposed in different bases: it is true that ̂ ̂ ̂ , but
it is not true that ( ) ( ); the latter are manifestly different column vectors.

1
Strictly speaking it is incorrect to write the column vector representation as an equality, “ ( )”,

since it can lead to contradictions such as ( ) ( ) if vectors are being represented in more than
one coordinate system. However, many authors and teachers will still use an equality symbol, while
either explicitly or implicitly specifying the coordinate basis that the column vector corresponds to.
While the omission of basis information in the column vector representation would lead one to believe
that explicitly writing the linear combination ̂ ̂ is a superior notation, there is a
serious shortcoming with this notation as well, once one starts multiplying vectors.

Inner Products

There are many possible rules one can come up with for combining two vectors through “multiplication”,
the most familiar of these are the dot product, which combines two vectors to produce a scalar, and the
cross product, which combines two vectors to produce a vector. Of these two, the dot product is the more
ubiquitous as a cross product can only be well-defined for vectors in 3 and 7 dimensions, while the dot
product can be defined for vectors of any dimension. The dot product is just one type of inner product,
which is a rule, subject to a few mathematical axioms, for combining two vectors to make a scalar. For
real space vectors, the rule for finding a dot product is to multiply the corresponding components of each
vector and sum these products: with ̂ ̂ and ̂ ,̂ the dot product is
. When the dot product is a vector dotted with itself, the scalar result has the geometrical
interpretation of being the length of the vector squared: . In more abstract vector spaces one
used this fact to define the “length” or norm of a vector as the square root of the inner product of a vector
with itself: | | √ . There is also a straightforward geometrical interpretation of a vector’s
inner product with a unit vector, since the result is the scalar projection, i.e. the length of the of the
vector on the unit vector’s axis: ̂ . A vector’s vector projection is thus ̂̂ .

Since calculating dot products requires multiplying components of vectors, and the components change
depending on which coordinate system one uses, one might ask which representation should one use to
calculate the dot product? The answer is it doesn’t matter. As long as the components of both vectors in
the dot product are written in the same representation, the scalar result is always the same. If

̂ ̂ ̂ ̂
and ̂ ̂ ̂ ̂,
then .

For a scalar-producing combination of vectors to qualify as an inner product, one of the central axioms it
has to adhere to is that the product is invariant under a change of basis. When changing bases or
coordinate systems almost all the numbers used to describe and specify vectors and their relations change,
but the inner products are the unchanging scalars to hold onto.

2
In the column vector representation, one calculates dot products using a combination of row and column
vectors: ( ). Note that while the equality “ ( )” is strictly speaking incorrect,
this dot product equality is exact, with both sides of the equation yielding the same scalar. The row-
column combination here nicely uses the usual linear algebra multiplication rules to specify the dot
product calculation, but also reveals a peculiarity of this type of representation. It makes sense to
multiply ( ) to get a scalar, but not if the row and column vector were switched:

( ) ( ) is a matrix not a scalar, even though, for real vectors, is

the same scalar in either order. Moreover, while and are both meaningful expressions, one
can calculate ( ), but not ( ).

Dual Space Vectors

At first, the lack of commutativity of the row and column vector representations, and the restriction that
one can multiply rows and columns but not add them, appear to be further shortcomings of this form of
vector representation. In fact, both reveal a hidden truth about inner products that is not at all apparent
from working with dot products in introductory physics: all inner products combine vectors from two
different vector spaces. In the product the vector (in physics contexts often termed the covariant
vector) and the vector (the contravariant vector) are not taken from the same set of vectors. The sum
is between either two covariant or two contravariant vectors, so that the symbols and
technically do not signify an identical pair of vectors in the expressions and . This distinction
is entirely obscured by the notation and glossed over in introductory physics since conflating
contravariant vectors and their corresponding covariant vectors or dual space vectors rarely caused
trouble. But in pursuing more advanced physics one must eventually become aware of the distinction,
particularly in learning quantum mechanics and general relativity.

A first introduction to the covariant/contravariant distinction is often implicitly encountered in

expressions for 3-dimensional waves such as . The scalar result of the dot product of
the wavevector and the position vector is well defined, but it makes no sense to try and combine
and by adding them – the vectors do not even have the same units.

Since the vectors in the product are not from the same space, the inner product of vectors does not even
have to be the same if the order of the vectors are reversed. In the complex vector spaces used in
quantum mechanics, where linear combinations of vectors can be made using complex coefficients, the
inner product is defined so that . The row and column representations of inner products
are then ( ) and ( ). Note that the inner product of a vector
with itself (strictly speaking with its dual space counterpart) is always real and positive, preserving the
notion of the “length” of a vector even when it is complex.

3
In our examination of vector notations, we have found that while the bold-faced vectors could be used to
identify and manipulate vectors regardless of coordinate system, the arbitrary choice of coordinate system
is hidden when using row and column vectors. Conversely, when considering inner products, the row and
column vectors make a clear distinction between the different species of vectors entering into the product,
while this distinction is hidden when using bold-faced vectors. What is needed is a vector notation that
combines the virtues of each of these notations.

Bra ket Notation

Devised by Paul Dirac in 1939 to deal with vector spaces in quantum mechanics, the bra ket notation
preserves both notational virtues. The contravariant vector is now represented by the ket | ⟩, its
covariant counterpart by the bra |, and the dot product by | ⟩. (Note that the dot product is just the
covariant multiplying the contravariant from the left: “ || ⟩”, but it is customary to only write one “|”
without ambiguity.) One can still write decompositions of a vector specifying the basis:
| ⟩ | ̂⟩ | ̂⟩ | ̂⟩ | ̂ ⟩, while also making a distinction between the different products:
| ⟩ = | ⟩ . Like the bold-faced vector notation, the symbols within the | ⟩ or | are just labels used
to denote the vector, so one could write | ⟩ to signify the vector | ⟩ | ⟩. Keep in mind the plus
sign inside the ket is not a mathematical operation, it’s just part of a label that helps to specify what vector
is under discussion. One could equally well, though somewhat less compactly, have written
| ⟩ to specify this vector.

The dual space counterpart of a vector is found using the Hermitian conjugate operation, signified by a
superscript “dagger”: | ⟩ | and | | ⟩. This operation is also used on with the row-column
representations, where it signifies the complex conjugate of the transpose of a matrix:

( ) , ( ), and ( ) ( ).

As with the bold-faced vector notation, one often writes “| ⟩ ( )” and “ | ”, though,
again, this is strictly speaking not correct, since the row and column representations of the vectors are
basis-dependent, while the bra’s and ket’s are not. The correspondence with the row and column vectors
is also manifest in the fact that the bra’s and ket’s do not commute | ⟩ | ⟩ |. The inner product on
the left is a scalar, while the “outer product” on the right is a more complicated mathematical object.

To add some intuition to this object, consider the outer product of a unit vector and its Hermitian
conjugate: | ̂⟩ ̂|. We first note that, since in the {| ̂⟩ | ̂⟩} basis | ̂⟩ is represented by ( ), ̂| is

represented by and | ̂⟩ ̂| by ( ) ( ). So while bra’s and kets are represented by

column vectors and row vectors, outer products are represented by square matrices, whose entries depend
on the particular coordinate system or basis being used to represent it. This particular outer product
we’ve encountered earlier, since | ⟩ | ̂⟩ | ̂⟩, ̂| ̂⟩ , and ̂| ⟩̂ , we have

4
 | ̂⟩ ̂| | ⟩ | ̂⟩ ̂| ( | ̂⟩ | ̂⟩) | ̂⟩ ̂| | ̂⟩ | ̂⟩ ̂| | ̂⟩ | ̂ ⟩ ̂| ̂⟩ | ̂⟩ ̂| ̂⟩ | ̂⟩,

which is just | ⟩, the vector projection of | ⟩ onto the unit vector’s axis. | ⟩ | ̂⟩ ̂| ⟩ is the bra ket
analogue of ̂ ̂ . Similarly| ⟩ | ̂⟩ |̂ ⟩ and | ⟩ | ̂ ⟩ ̂ | ⟩. The outer product of a unit
vector is known as the “projection operator” ̂ for that vector axis, since using it on any vector yield the
vector projection of the vector along this axis: ̂ | ̂⟩ ̂|.

Linear Operators

Projection operators are part of a class of objects which act to turn one vector into another, e.g. there
exists some linear operator ̂ that can change | ⟩ into | ⟩: ̂ | ⟩ | ⟩. The “linear” adjective refers
to the fact that these operators are distributive ̂ | ⟩ | ⟩ ̂ | ⟩ ̂ | ⟩. As we’ve just seen, for
finite vector spaces linear operators can be represented by square matrices, which leads to the definition
of the Hermitian conjugate (or adjoint) of an operator. Representing | ⟩ ̂ | ⟩ as

( ) ( )( ) ( ),

we take the complex conjugate transpose of each side:

( )

Or | | ̂ . Thus, when dealing with matrix representations, the Hermitian conjugate of an operator
is the expected complex conjugate transpose of the operator. Note also from this example that taking the
Hermitian conjugate of a product is equal to taking the Hermitian conjugate of each term in the product in
reverse order:

( ̂ | ⟩) | ⟩ ̂ |̂ or ( ̂ ̂ ) ̂ ̂ or ( ̂ ̂ | ⟩) | ⟩ ( ̂ ̂) |̂ ̂

Using this property of Hermitian conjugates, one can formulate the definition of the Hermitian conjugate
of an operator in a basis independent way. Since | ̂| ⟩ ( | ̂ | ⟩) ( ̂ | ⟩) | ⟩ is equal to the
same scalar no matter the basis, one defines ̂ as that operator which satisfies ( ̂ | ⟩) | ⟩ | ̂| ⟩
for all vectors | ⟩ and | ⟩. Often one denotes an operator-vector product ̂ | ⟩ by | ̂ ⟩. Again note that
as with | ⟩ there is no mathematical operation going on inside the ket, the “ ̂ ” label is to let the
reader know that this is the vector produced by taking operator ̂ times ket | ⟩. With this notation the
definition of the Hermitian conjugate can be made even more compact:

̂ is the operator which satisfies ̂ | ⟩ | ̂ ⟩ for all vectors | ⟩ and | ⟩.

5
 Orthonormal Bases and Vector Representations

As we have seen, to find a representation of vectors using rows or columns one needs to specify a set of
basis vectors such as | ̂⟩ and | ̂⟩. While there are many different sets of basis vectors, it is almost always
best to use sets which are orthonormal. To qualify as an orthonormal basis a (possibly infinite) set of
vectors {| ⟩ | ⟩ | ⟩ } must satisfy the following conditions

1) The set must span the space – one must be able to write any vector | ⟩ in the vector space as a linear
combination of this set: | ⟩ ∑ | ⟩, using some set of scalar coefficients { }.

2) The set should be linearly independent – no vector in the set should be able to be written as a linear
combination of the others.

3) Each vector in the set should be normalized – the length (and length squared) of each vector in the set
should be 1, meaning that the inner product of a vector and its dual space counterpart is 1: | ⟩ .

4) The vectors in the set should all be orthogonal to each other: | ⟩ for .

The first two conditions qualify the set as a basis, the third condition is the “normal” part, and the fourth
condition the “ortho” part of the orthonormal basis. The last two conditions can be more compactly
written

| ⟩

where is the Kronecker delta, a function of two integer indices, that equals 1 when the integers are
the same and equals 0 when they are not. The scalar coefficients { } that specify a vector are
known as the “components” of the vector and are the representation of that vector in the orthonormal
basis, i.e.

if | ⟩ ∑ | ⟩, then ( ) is the representation of | ⟩ in the {| ⟩| ⟩| ⟩ } basis.

To find the th components of a vector in this basis, one can multiply the vector by the bra | and
exploit the orthonomality of a basis:

| ⟩ |∑ | ⟩ ∑ | ⟩ ∑ .

| ⟩
| ⟩
So the representation of | ⟩ in this basis can also be written ( ). One can use this same method to
| ⟩

find the representations of the basis vectors | ⟩, | ⟩, etc. in the {| ⟩| ⟩| ⟩ } basis: the

6
 | ⟩ | ⟩
| ⟩ | ⟩
representation of | ⟩ is ( ) ( ), the representation of | ⟩ is ( ) ( ), etc. The
| ⟩ | ⟩

linear combination | ⟩ ∑ | ⟩, which holds in any basis, looks like

( ) ( ) ( )

in the {| ⟩| ⟩| ⟩ } basis.

Represented in a different basis, the components of | ⟩ are entirely different, for the

{| ⟩| ⟩| ⟩ } basis, if | ⟩ ∑ | ⟩, | ⟩ is represented by ( ).

What is the relation between the representations ( ) and ( )? Since

| ⟩ ∑ | ⟩ ∑ | ⟩

if we multiply by the bra | we find

∑ | ⟩ ∑ | ⟩, or, using orthonomality

∑ | ⟩.

This method can be used to find each of the components of the {| ⟩| ⟩| ⟩ } basis representation
in terms of linear combinations of the components of the {| ⟩ | ⟩| ⟩ } basis. These relations
can be compactly written as a linear algebra equation:

| ⟩ | ⟩ | ⟩
| ⟩ | ⟩ | ⟩
( ) ( )( ) ( ).
| ⟩ | ⟩ | ⟩

Where is the matrix of inner products of different basis vectors. Similarly,

| ⟩ | ⟩ | ⟩
| ⟩ | ⟩ | ⟩
( ) ( )( ) ( ).
| ⟩ | ⟩ | ⟩

7
Since | ⟩ | ⟩ it is clear why the matrix is the Hermitian conjugate of the matrix .

This conjugate pair is of a special type. Since ( ) ( ) ( ), is the identity

matrix. Thus is the inverse of , a condition that categorizes the basis-transforming as a

unitary matrix.

Another relation between the two representations relies on the invariance of the inner products. Since the
norm2 | ⟩ is the same in all bases,

| ⟩ ∑ | ∑ | ⟩ ∑ | ⟩ ∑ ∑ | | .

Similarly, | ⟩ ∑ | | .

Note that in squaring a sum, we have changed the dummy index letter in one of the sums to avoid
confusing which parts of the product are being summed over by which sum. A failure to do so can lead to
errors. For example, the sum ∑ , so its square should be 9. If one squares it by writing
∑ ∑ , it is likely that in combining the terms of the sums, one erroneously writes
“∑ ”, rather than ∑ ∑ ∑ .
This is a more advanced version of the error “ ” frequently made when first learning
algebra.

Linear Operator Representations

In the previous section, we saw how to find a column representation of a vector, but how do we find the
matrix representation of a linear operator? For any vector | ⟩ and operator ̂ , let | ⟩ be the
components of the vector | ⟩ in the {| ⟩ | ⟩ | ⟩ } basis and ̂
⟨ | ⟩ be the components of
̂
the vector | ⟩ ̂ | ⟩ in the same basis. Since

| ⟩ ∑ | ⟩,

|̂ ⟩ ̂∑ | ⟩ ∑ ̂| ⟩, and

⟨ |̂ ⟩ |∑ ̂| ⟩ ∑ | ̂| ⟩.

From this equality, the relations between the representations of | ⟩ and | ̂ ⟩ can be written as a linear
algebra expression:

| ̂| ⟩ | ̂| ⟩ | ̂| ⟩
( ) | ̂| ⟩ | ̂| ⟩ | ̂| ⟩ ( )
| ̂| ⟩ | ̂| ⟩ | ̂| ⟩
( )

8
The matrix in the expression is the operator representation of ̂ in the {| ⟩ | ⟩ | ⟩ } basis. The
basis-independent scalar entries | ̂ | ⟩ are known as matrix elements even when the matrix is not
explicitly being written.

We have focused on going from the bra ket notation to representations in a basis, but how does one go
from a representation and a basis to a basis-independent bra ket expression? In the case of vectors this

process is straightforward: if a vector has the representation ( ) in the {| ⟩| ⟩| ⟩ } basis, then

| ⟩ ∑ | ⟩. The operator case is a bit less transparent. We start by noting that in the
{| ⟩ | ⟩ | ⟩ } basis representation the outer product of any two basis vectors | ⟩ | are matrices
th th
with a 1 in the row, column and 0’s everywhere else:

th
entry

th
row
th
entry
( )
( )
th
column

Hence any matrix in this basis representation can be written as a linear combination of these matrices.
For example an operator represented by the 2×2 matrix ( ) in the {| ⟩| ⟩} basis can be
written in a basis-independent manner as:

| ⟩ | | ⟩ | | ⟩ | | ⟩ |.

Similarly, the representation of the operator ̂ above can be used to write

̂ ∑ | ̂| ⟩| ⟩ |,

where the sum is over both indices.

As with vector representations, one often wants to change the representation of an operator from one basis
to another. To find the algorithm for doing this, we exploit the fact that inner products are invariant under
a change of basis. Writing the inner product | ̂ | ⟩ in one representation:

( ) ( ).

We know that the vectors transform using a unitary matrix:

9
 ( ) ( ) and ( ) ,

so using unitarity we write the inner product

( )( )

( ) ( )

( ) ( ) ( )

Shows that the matrix representation of the operator ̂ in one representation is in the
representation calculated by multiplying vector representations by .

Projection, Hermitian, and Unitary Operators

Three commonly encountered operators in quantum mechanics are projection, Hermitian, and unitary
operators. We have already encountered projection operators above: any operator that can be written
̂| ⟩ | ⟩ |, where | ⟩ is a normalized vector, is a projection operator. Note that in a basis that has
| ⟩ as a member, the representation of ̂| ⟩ is a matrix of all zeroes except for a single 1 on the diagonal
of the matrix. A Hermitian operator ̂ is one that satisfies ̂ ̂ . It can be shown that one can
always choose the eigenvectors of these operators to be an orthonormal set, and that the eigenvalues are
always real. In the standard quantum formalism Hermitian operators correspond to measureable
properties of a quantum state and the eigenvalues are possible results of the measurement. A unitary
operator ̂ is one that satisfies ̂ ̂ ̂ ̂ ̂, where ̂ is the identity operator, represented by the
identity matrix, that leaves a vector unchanged. Unitary operators transform one vector to another
without changing their norm, and thus can be used to describe the transformation of one normalized
quantum state to another, while their Hermitian conjugate describes the inverse transformation.
Unsurprisingly, they are represented by unitary matrices.

We have already discussed unitary matrices in transforming representations from one basis to another.
However, there is a subtle distinction between those matrices and the representations of unitary operators
that transform one state to another.

10
Consider the 2-D real-space vector | ⟩ shown above. The representation of this vector in the –
coordinate system is ( ). Its representation in the – coordinate system is ( ). To transform

the column vector from one representation to another, one uses the unitary matrix ( ),

so that ( )( ) ( ). This changes the representation but does not change the
vector. Now consider what happens if we transform the vector from | ⟩ to | ⟩ by rotating it clockwise
by an angle . In the – coordinate system representation, this changes the vector from ( ) to

( ). In both cases the column vector is altered in the same way, but in the first case it is the
representation which is changed while the vector is unchanged, while in the second case the vector is
changed while the representation is unchanged. In the second case there must exist a unitary
transformation that takes | ⟩ to | ⟩: ̂| ⟩ | ⟩. The representation of the operator ̂ in the –
coordinate system is ( ) . Thus the matrix that transforms the representation of | ⟩
from the – coordinate system to the – coordinate system is the Hermitian conjugate of the
representation in the – coordinate system of the operator that changes | ⟩ to | ⟩. Geometrically the
first case corresponds to rotating the coordinate counterclockwise by and the second case rotating the
vector clockwise by . The operator ̂ has many representations besides the one in the – coordinate
system, but the matrix is not the representation of any operator in a basis, since its purpose is to
transform column vectors from one representation to another. A student re-reading this paragraph until its
distinctions are clear will have made great progress towards understanding vectors, operators, and their
representations.

Continuous Basis Sets

The preceding sections dealt with finite and denumerably infinite (countable) vector bases. One can
extend the formalism to vector spaces with an “uncountable”, continuous basis set as follows. Vectors
that previously could be decomposed into the sum of an infinite set of countable components are now
decomposed into a continuous integral where the coefficients for each of the basis vectors are accounted
for in a continuous function:

11
 | ⟩ ∑ | ⟩ | ⟩ ∫ | ⟩

As indicated, the members of the basis set, rather than being indexed by the integer are indexed by a
continuous variable such as . Unlike most discrete cases, the variable chosen in the continuous case
often has physical meaning, here | ⟩ is written as a linear combination of the position basis states {| ⟩},
which are descriptions of objects with definite position. Whereas before the representation of | ⟩ in the

{| ⟩} basis consisted of a column vector enumeration of the coefficients : ( ). The coefficients of

the continuous basis are contained in the function . Rather than specifying an integer value to find
a particular coefficient for the basis vector | ⟩, one specifies a position to find the
coefficient of the basis vector | ⟩.

Consider the notation of a quantum state written as a linear combination of the position basis states:
| ⟩ ∫ | ⟩ . The state is labeled by “ ”, a label that indicates the state may be
different at different times. For each time there is an expansion of the state in the position basis. The
coefficients of this expansion are contained in the function in this context known as the
“wavefunction” of the object being described. The variable is necessary because each | ⟩ basis state
requires a different coefficient in the expansion. The variable is necessary because the set of
coefficients are different at different times. A common mistake is to conflate the state | ⟩ with the
wavefunction , which the same mistake as writing | ⟩ ( ). The state | ⟩ and vector | ⟩

are basis independent. ( ) is the representation of | ⟩ in the – coordinate basis, and is the
representation of | ⟩ in the {| ⟩} position state basis.

It is not necessary to use the same letter “ ” for both the state and the wavefunction, but it is a way of
letting the reader know which set of coefficients corresponds to which state. Consider the expansion of
the same state in a different basis, for example, the states {| ⟩} which have definite wavenumbers and
definite momentum : | ⟩ ∫ | ⟩ . | ⟩ is represented by the function in
this momentum basis.

A subtle difference between the sum and the integral case is that, for the sum, the coefficients are
dimensionless scalars, while the integral, since it often over a variable with dimensions, may necessitate
the coefficient functions and have dimensions themselves. In quantum mechanics the
state | ⟩, being related to probabilities, is dimensionless. The infinitesimal and have units of
length and inverse length (the unit of the wavenumber) respectively. For the units to work out, one
chooses and the states | ⟩ to have the peculiar units of the inverse square root of length while
and the states | ⟩ to have units of the square root of length. It is rare that one is mindful of the
dimensions of quantum states and wavefunctions, but as with all calculations in physics keeping track of
units can be a useful way of spotting errors in equations and calculations.

12
 Normalization of Continuous Basis Sets

In the discrete case, one could exploit the orthonormality of the basis to find a particular coefficient or the
general formula for a vectors representation in a basis. To do the same operation in a continuous basis
one should have

| ⟩ ∑ | ⟩ ∑ | ⟩ ∫ | ⟩

For the final step to work, the inner product | ⟩ cannot be a Kronecker delta, since this would make
| ⟩ zero everywhere except a single value of where it is equal to the infinitesimally small
. Instead one needs a function that is infinite at so that the product | ⟩ . This
“function” is the Dirac delta function | ⟩ , which is 0 everywhere , infinite when
and satisfies ∫ and ∫ . From a rigorous
mathematical perspective the Dirac delta function is not a function, since it is not well defined at the point
of interest. However there exists an extension of the concept of a function known as a “distribution”
which can be used to handle Dirac deltas in a formal manner. Physicists tend to ignore these
mathematical subtleties.

Physicists also sometimes find it useful to write down an explicit functional form for the Dirac delta
function which again ignores such subtleties. To do so one starts from the relation (rigorously provable,
but beyond the scope of this text)

∫ ∫

Since the position integral is over all space, one can shift the origin without affecting the result:
∫ ( ∫ ) . When the term in parenthesis is clearly infinite, when
the integral is continuously adding up all the complex numbers with a norm of one. Adding all these
together once equals zero, and this integral repeats this addition over and over again. Since the limits are
infinite, strictly speaking this integral is bounded but does not converge, but physicists typically ignore
this lack of true convergence and call this integral 0 for , meaning that it has all the requisite
properties to identify it as a Dirac delta function

∫ .

Similarly one can find an expression for a Dirac delta ∫ with wavenumber
arguments. Note also from these expressions and the definition of the Dirac delta function that it is
“symmetric” with respect to its arguments:

From this investigation one sees that the inner product of a vector with itself in an “orthonormal”
continuous basis set is not 1 but infinity. Because of this result these basis vectors are not normalizable,
though one sometimes states that they are Dirac-delta normalizable: | ⟩ .

13
 Finding Representations in Continuous Basis Sets

As we have seen, the state | ∫ ⟩ | ⟩ ∫ | ⟩ has the representation

in the position basis and the representation in the wavenumber basis. We can use the
Dirac-delta orthonormality to find these representations:

| ⟩ ∫ | ⟩ ∫ , or

| ⟩ ∫ | ⟩ ∫

This operation has given us a particular component of the | ⟩ basis state, but as before it is a general
method for finding all coefficients and hence the value of for every , just as the algorithm
| ⟩ ∑ | ⟩ ∑ can be used for every coefficient in a countable basis.

There is a subtle notational difference in the continuous case in that rather than using different letters such
as or to distinguish different basis vectors being combined, one uses the same letter with subscripts
or superscripts such as or when distinguishing different continuous basis vectors being combined.
In sums over all basis vectors | ⟩ ∑ | ⟩ the discrete index was a “dummy index” that could
just as easily have been a different letter. For the integrals over all continuous states the variable being
integrated is now a dummy variable, so that one can equivalently write | ⟩ ∫ | ⟩ or
| ⟩ ∫ | ⟩ . One can then rewrite the calculation of a particular component of | ⟩
as | ⟩ ∫ | ⟩ ∫ . There is no difference in the
calculation, but the notation lends itself to a slightly different interpretation. The result before one would
interpret as the value of the function at a particular point , whereas this calculation yields
itself. Of course, the function is nothing more than its values for all individual points, but
the notational switch makes it clear that the result of this general inner product can be used for any point,
just as | ⟩ can be used for any value of .

With this notation, one sees that the representation of any state in the | ⟩ in the position basis is | ⟩,
and its representation in the wavenumber basis is | ⟩.

What are the representations of the basis vectors themselves? In the position basis a position state | ⟩ is
represented by | ⟩ , the Dirac delta function. To find the representation of the
wavenumber states in the position basis, | ⟩ , we again use the fact that inner products are the
same in all bases. With this in mind we look at how | ⟩ is calculated in the position
basis. Expanding | ⟩ in the position basis

| ⟩ ∫ | ⟩ . The corresponding bra is

| | ⟩ (∫ | ⟩ ) ∫ | .

Note that the Hermitian conjugate of the representation , just being a function that outputs scalars,
is its complex conjugate. For the same reason, these component functions commute with the bras and
kets. With these expansions we can take the inner product. As with the discrete case where

14
computational mistakes were avoided by making the dummy index of two sums being multiplied distinct,
similar mistakes are avoided here by making the dummy variables of the two integrals being multiplied
distinct.

| ⟩ (∫ | )( ∫ | ⟩ ) ∫ ∫ | ⟩

∫ ∫ ∫

( )
Above we noted that ∫ , so this equality will work if
√
is
the representation of | ⟩ in the {| ⟩} basis. Finding the representations of | ⟩ and | ⟩ in the {| ⟩} basis
are straightforward. A wavenumber state | ⟩ is represented by | ⟩ . A position state | ⟩
is represented by | ⟩ | ⟩ ( ) .
√

Normalization of States in the Continuous Case

While the continuous basis states are only Dirac delta normalizable it is possible for linear combinations
of these states to be normalizable to 1. Consider taking the inner product of | ⟩ ∫ | ⟩
with its dual counterpart

| ⟩ (∫ | ⟩ ) ∫ | ⟩

∫ | ∫ | ⟩ ∫ ∫ | ⟩

∫ ∫ ∫ | |

It is possible for this final integral to be finite. The class of functions for which this is true are called
square-integrable functions. So for a quantum state to be normalizable (necessary if one wants to
interpret it probabilistically) it must be represented by a square-integrable wavefunction, which can be
scaled so that the norm of the state is 1. This condition is true for the representation of | ⟩ in any
basis, and one can show that if ∫ | | then ∫ | | .

Changing Representations in the Continuous Case

In the discrete case one changed a column vector from one representation to another using a unitary
matrix. In the continuous case we would like the analogous algorithm for changing a representation
in one basis to in another. We start by using the basis-independent expansions of a state:

15
 | ⟩ ∫ | ⟩ ∫ | ⟩

To find in terms of we multiply both sides by |:

∫ | ⟩ ∫ | ⟩ ∫

The term on the left is the analogue of a unitary matrix multiplying a column vector, where is the
“column vector” representing the state and | ⟩ are the entries in the “matrix”. One can make this
integral more explicit by evaluating the inner product to get

√
∫ .

This operation is known as taking the Fourier transform of , and itself is said to be the
Fourier transform of . One can similarly show that

∫ ,
√

an operation known as the inverse Fourier transform, with being the inverse Fourier transform of
.

Linear Operators in the Continuous Case

Again looking to the discrete case for guidance, we recall that for a linear operator ̂ , we could write

̂ ∑ | ̂| ⟩| ⟩ |

and identify the representation of ̂ with the matrix elements | ̂ | ⟩. As a simple example, consider
the identity operator which satisfies | ⟩ | ⟩ for all vectors (or states) | ⟩. The matrix elements of ̂
̂ ̂
are | ̂| ⟩ | ⟩ . Unsuprisingly this means the matrix representation of ̂ has 1’ s along
it’s diagonal and 0’s for all “off-diagonal” entries, i.e. the identity matrix . The identity operator in
the discrete case is unique in that it’s representation is the same in all bases, as can be seen by repeating
this calculation for a different basis or noting that the unitary transformation of its matrix representation
leaves the matrix unchanged: . Note that with these matrix elements determined, one
can write basis-independent expressions for : ̂

̂ ∑ | ̂| ⟩| ⟩ | ∑ | ⟩ | ∑ | ⟩ | ∑ ̂| ⟩.

The last two expressions in particular often prove useful.

Turning to a continuous basis such as {| ⟩}, the expression for a linear operator becomes

16
 ̂ ∫ ∫ | ̂ | ⟩| ⟩ |

With the representation of ̂ in this basis determined by the “matrix elements” | ̂ | ⟩ (a term still used
despite the lack of an actual matrix). Examining the identity operator in this continuous case, we find the
matrix elements of this operator are now | ̂| ⟩ | ⟩ Dirac delta functions. Putting
these elements into the basis-independent expression yields

̂ ∫ ∫ | ̂| ⟩| ⟩ | ∫ ∫ | ⟩ | ∫ | ⟩ | ∫ ̂| ⟩ ,

again providing quite useful renderings of the identity operator. Note that unlike the discrete case where
the matrix representation of ̂ is the same for every basis, its representation in the continuous case is basis-
dependent. In the {| ⟩} basis its matrix elements are | ̂| ⟩ | ⟩ , again Dirac delta
functions, but not the same Dirac delta functions as the .

Turning to a slightly more interesting case, consider the the position operator ̂. By definition the
position states | ⟩ are eigenstates of this operator with eigenvalue

̂| ⟩ | ⟩.

Finding the matrix elements of ̂ in the position basis is straightforward

| ̂| ⟩ | | ⟩ | ⟩ .

meaning a basis-independent expression for the operator is

̂ ∫ ∫ | ̂| ⟩| ⟩ | ∫ ∫ | ⟩ | ∫ | ⟩ | .

We can now use this formulation to evaluate an inner product such as | ̂| ⟩. From the
expansions of the two states in the position basis

| ⟩ ∫ | ⟩ and | ⟩ ∫ | ⟩

The inner product is

| ̂| ⟩ (∫ | )∫ | ⟩ | (∫ | ⟩ )

∫ ∫ ∫ | ⟩ | ⟩

∫ ∫ ∫

Integrating out the delta functions has produced this is a rather compact way of writing the inner product
using representations in the position basis. For this reason, one usually considers the representation of ̂
in the position basis to simply be the variable rather than the matrix elements insofar as one

17
can identify each part of the inner product with a corresponding part of the integrand used to calculate it.
In doing so, the formal analogy with the discrete case is lost, which can sometimes cause confusion. In
the discrete case one has

| ̂| ⟩ ( )( ) ∑

where we can clearly identify | with the scalars used to represent is, ̂ with its matrix elements
, and | ⟩ with the scalars. The two sums over the indices and serve to carry out the two
linear algebra products. The formal analog in the continuous case is

| ̂| ⟩ ∫ ∫

with | represented by , ̂ by the matrix elements , and | ⟩ by .

The two sums in the discrete case have become two integrals. Carrying out one of the integrals makes the
expression to evaluate the inner product more compact, but obscures how each piece of the inner product
is represented. The equality | ̂| ⟩ ∫ is thus best thought of as a
recipe for calculating this type of inner product, without a one-to-one correspondence between terms on
each side.

To find a similar recipe for using the momentum operator ̂ in inner products such as | ̂| ⟩,
we recall that the states | ⟩ have definite momentum , so ̂ | ⟩ | ⟩. Using the same steps as were
employed in the position operator case, the momentum operator can be written

̂ ∫ ∫ | ̂ | ⟩| ⟩ | ∫ ∫ | ⟩ | ∫ | ⟩ |

The inner product | ̂| ⟩ is

| ̂| ⟩ (∫ | )∫ | ⟩ | (∫ | ⟩ )

∫ ∫ ∫ | ⟩ | ⟩

∫ ∫ ∫ ( )

∫ ∫ ∫ ( )
[ ]

The term in brackets without the could be integrated over to yield a Dirac delta function. To get
rid of this derivative, we integrate by parts for the variable and exploit the fact that the for
wavefunctions such as to be square-integrable one must have :

18
 | ̂| ⟩

∫ ∫ ( )
[ ]

∫ ∫ ∫ ( )
[ ][ ]

∫ ∫ [ ]

∫ [ ]

Again the relatively compact expression | ̂| ⟩ ∫ [ ] leads one

to identify the states with their respective representations and ̂ with the operation . Because of
this, it is often said that this operation is ̂ in the position basis. As with the case of ̂, making this
identification provides a useful recipe for calculating inner products with ̂ using wavefunctions in the
position basis, even while the formal analogy with the discrete case is obscured by doing so.

In a similar manner to the proofs above one can show that the suitable operation identifications in the
momentum basis are ̂ and ̂ . We incorporate these identifcations into the
following table, which summarizes position and wavenumber representations:

State/Operator Position Basis Momentum Basis

| ⟩
√
| ⟩
√

19