To: P.

Rungta
From: C. M. Caves
Subject: Antilinear operators
2001 May 6

The following notes are an elaboration of Chap. XV.I of Messiah and of CMC notes
dated 00–7–12.

I. General considerations
An antilinear operator K : |ψ⟩ → K|ψ⟩ acts on linear combinations according to
( ) ( ) ( )
K α1 |ψ1 ⟩ + α2 |ψ2 ⟩ = α1∗ K|ψ1 ⟩ + α2∗ K|ψ2 ⟩ . (1)

A product of linear and antilinear operators is linear if it has an even number of antilinears
and antilinear if it has an odd number of antilinears.
The left action of an antilinear operator, i.e., ⟨ϕ| → ⟨ϕ|K, is given by
( ) [ ( )]∗
⟨ϕ|K |ψ⟩ = ⟨ϕ| K|ψ⟩ . (2)

The complex conjugation is present so that

[( ) ] [( )( )]∗
⟨ϕ1 |α1∗ + ⟨ϕ2 |α2∗ K |ψ⟩ = ⟨ϕ1 |α1∗ + ⟨ϕ2 |α2∗ K|ψ⟩
[ ( )]∗ [ ( )]∗
= α1 ⟨ϕ1 | K|ψ⟩ + α2 ⟨ϕ2 | K|ψ⟩
( ) ( )
= α1 ⟨ϕ1 |K |ψ⟩ + α2 ⟨ϕ2 |K |ψ⟩ (3)
[( ) ] [( ) ]
= ⟨ϕ1 |K α1 |ψ⟩ + ⟨ϕ2 |K α2 |ψ⟩
[( ) ( )
= ⟨ϕ1 |K α1 + ⟨ϕ2 |K α2 ]|ψ⟩ ,

i.e., so that K is antilinear to the left:

( ) ( ) ( )
⟨ϕ1 |α1∗ + ⟨ϕ2 |α2∗ K = ⟨ϕ1 |K α1 + ⟨ϕ2 |K α2 . (4)

The presence of the complex conjugation in Eq. (2) means that in a matrix element one
must always indicate explicitly whether an antilinear operator acts to the right or to the
left.
The adjoint (Hermitian conjugate) of an antilinear operator is defined in the same
way as for a linear operator, i.e.,
( )† ( )
K|ψ⟩ = ⟨ψ|K† , (5)

but this means that

( ) [( ) ]∗ ( )
⟨ϕ| K|ψ⟩ = ⟨ψ|K† |ϕ⟩ = ⟨ψ| K† |ϕ⟩ . (6)

1
 This definition of adjoint means that the adjoint works the same way on products as
for linear operators. To verify this, we need to check two cases, keeping in mind that the
product of two antilinear operators is a linear operator and the product of an antilinear
operator and a linear is antilinear:

⟨ψ|(KL)† |ϕ⟩ = ⟨ϕ|(KL)|ψ⟩∗ = (⟨ϕ|K)(L|ψ⟩) = [(⟨ψ|L† )(K† |ϕ⟩)]∗ = ⟨ψ|(L† K† |ϕ⟩) ,

(7)
⟨ψ|[(KA)† |ϕ⟩] = ⟨ϕ|(KA|ψ⟩) = [(⟨ϕ|K)(A|ψ⟩)]∗ = (⟨ψ|A† )(K† |ϕ⟩) = ⟨ψ|(A† K† |ϕ⟩) .

Thus we have that (KL)† = L† K† and (KA)† = A† K† . Notice that the linear operators
K† K and KK† are Hermitian and, indeed, positive.
An antilinear operator is specified by its “matrix elements” in any orthonormal basis
|ej ⟩:
( ) [( ) ]∗ ( )
⟨ej | K|ek ⟩ = ⟨ej |K |ek ⟩ = ⟨ek | K† |ej ⟩ . (8)
Given an orthonormal basis, there is an associated linear operator A that acts the same
way as K on the basis vectors,
A|ej ⟩ = K|ej ⟩ , (9)
and thus has the same matrix elements in this basis:
( )
⟨ej |A|ek ⟩ = ⟨ej | K|ek ⟩ . (10)

The relation between K and A becomes clear when K acts on an arbitrary vector
( ) ( )
∑ ∑ ∑ ∑
K|ψ⟩ = K cj |ej ⟩ = c∗j K|ej ⟩ = c∗j A|ej ⟩ = A c∗j |ej ⟩
j j j j (11)
= A|ψ ∗ ⟩ = AC|ψ⟩ = CA∗ |ψ⟩ .

Here the complex conjugations on |ψ⟩ and A stand for complex conjugation in the chosen
basis |ej ⟩, and C is the antilinear operator that complex conjugates in this basis. The
antilinear operator can be written simply as

K = AC = CA∗ . (12)

The matrix elements of C are

( ) [( ) ]∗ ( )
δjk = ⟨ej | C|ek ⟩ = ⟨ej |C |ek ⟩ = ⟨ek | C † |ej ⟩ . (13)

We can conclude that ( )

⟨ej |C = ⟨ej | , (14)
which means that when acting to the left, C complex conjugates in the same basis. We
can also conclude that
C † |ej ⟩ = |ej ⟩ , (15)

2
from which it follows that C † = C. Since it is obvious that C −1 = C, we can summarize the
important properties of complex conjugation by

C = C −1 = C † . (16)

Notice that
K† = CA† = AT C , (17)
where AT ≡ (A∗ )† denotes transposition in the chosen basis.
There are two lessons here. The first is that all antilinear operators can be constructed
from the linear operators and a single antilinear operator, which can be chosen to be
complex conjugation in a particular basis. The second is that antilinear operators turn
the basis-independent operations of complex conjugation and transposition into basis-
independent operations on antilinear operators.
An antiunitary operator U is an antilinear operator that preserves the absolute value
of inner products, i.e.,
( )( )  ( )
|⟨ϕ|ψ⟩| =  ⟨ϕ|U † U|ψ⟩  = ⟨ϕ| U † U|ψ⟩  (18)

for all |ϕ⟩ and |ψ⟩. Since C complex conjugates inner products, it is antinunitary. There are
two useful characterizations of an antiunitary operator: the antiunitary of U is equivalent
to (i) U † U = UU † = 1 and (ii) U = U C = CU ∗ , U being unitary, in any basis (and then
all bases). It is obvious that if U † U = 1, then U is antiunitary. Suppose, then, that U is
antiunitary. The definition (18) of antiunitarity, applied to an orthonormal basis, implies
that U † U = eiΛ , where Λ is Hermitian and diagonal in the chosen basis. But, since U † U is
positive, we have that U † U = 1. The second characterization is trivially equivalent to the
first.
Consider now a change of orthonormal basis:
∑ ∑
|fj ⟩ = V † |ej ⟩ = |ek ⟩(V † )kj = ∗
|ek ⟩Vjk . (19)
k k

The matrix elements of an antilinear operator K change from one basis to another according
to
( )[ ( )]
( ) ∑ ∑ ∑ ( )
∗
⟨fj | K|fk ⟩ = Vjl ⟨el | K Vkm |em ⟩ = Vjl ⟨el | K|em ⟩ (V T )mk . (20)
l m l,m

We need to say a word about how linear operators transform under antiunitaries.
Consider a linear operator A and an antiunitary U:

(⟨ψ|U † )A(U|ψ⟩) = [⟨ψ|(U † AU|ψ⟩)]∗ = ⟨ψ|U † AU|ψ⟩∗ = ⟨ψ|U † A† U|ψ⟩ . (21)

Thus the appropriate antiunitary transformation of a linear operator A is U † A† U. If

A is Hermitian, this reduces to the same form as for a unitary transformation, but for
nonHermitian operators, it different by having to take the adjoint of A.

3
II. Polar decompositions
An extremely useful tool is the polar decomposition of an antilinear operator K. First
diagonalize the positive operator K† K:
∑
K† K = λ2j |ej ⟩⟨ej | , λj ≥ 0 . (22)
j

Relative to the eigenbasis |ej ⟩, K can be written as

K = AC = CA∗ , (23)

where C is complex conjugation in the eigenbasis. Notice that

K† K = CA† AC = AT A∗ . (24)

Since K† K is real in the eigenbasis, however, we can simplify this to

∑
A† A = CK† KC = K† K = AT A∗ = λ2j |ej ⟩⟨ej | . (25)
j

The polar-decomposition theorem for A says that there exists a unitary operator U such
that √ √ ∑
A = U A† A = AA† U = λj |fj ⟩⟨ej | , (26)
j

where
U |ej ⟩ = |fj ⟩ =⇒ U ∗ |ej ⟩ = |fj∗ ⟩ (27)
and ∑
AA† = U A† AU † = λ2j |fj ⟩⟨fj | . (28)
j

Notice that AA† is not real in the eigenbasis:

∑
(AA† )∗ = A∗ AT = U ∗ A† AU T = λ2j |fj∗ ⟩⟨fj∗ | . (29)
j

We also have √ √ ∑
A† = A† AU † = U † AA† = λj |ej ⟩⟨fj | ,
j
√ ∑
∗ ∗
A = U A† A = λj |fj∗ ⟩⟨ej | , (30)
j
√ ∑
AT = A† AU T = λj |ej ⟩⟨fj∗ | .
j

4
 Using the polar decomposition of A, we can obtain a number of useful forms for the
polar decomposition of K, summarized in the following:
 
√ √ ∑
K = AC = U K† KC = KK† U C =  λj |fj ⟩⟨ej | C ,
j
  (31)
√ ∑
K = CA∗ = CU ∗ K† K = C  λj |fj∗ ⟩⟨ej | .
j

Other useful relations include

∑
K† K = A† A = AT A∗ = U † KK† U = λ2j |ej ⟩⟨ej | ,
j
∑ (32)
KK† = AA† = U K† KU † = λ2j |fj ⟩⟨fj | .
j

Notice that U † KK† U , not KK† , is real in the eigenbasis.

III. Symmetric antilinear operators

We are now ready to consider antilinear Hermitian (or symmetric) antilinear oper-
ators, i.e., antilinear operators satisfying K = K† . That K is antilinear Hermitian is
equivalent to each of the following:
( ) ( )
(i) K is symmetric, i.e., ⟨ϕ| K|ψ⟩ = ⟨ψ| K|ϕ⟩ for all |ϕ⟩ and |ψ⟩.
(ii) K is symmetric in any one orthonormal basis (and then all orthonormal bases).
(iii) K can be diagonalized in an orthonormal basis; i.e., there exists an orthonormal basis
|ej ⟩ such that K|ej ⟩ = λj |ej ⟩. The “eigenvalues” λj can be made real and positive.
The first two of these are trivial consequences of the definition of the adjoint. Before
considering the third, notice that K2 |ej ⟩ = K† K|ej ⟩ = |λj |2 |ej ⟩. Moreover, notice that the
reason that the “eigenvalues” can be made real and positive is that under a rephasing of
the eigenvectors, i.e., |e′j ⟩ = eiδj |ej ⟩, the “eigenvalues” undergo a phase change,
( )
K|e′j ⟩ = K eiδj |ej ⟩ = e−iδj K|ej ⟩ = e−iδj λj |ej ⟩ = λj e−2iδj |e′j ⟩ . (33)
The real and positive “eigenvalues” λj can be characterized uniquely as the square roots
of the eigenvalues of K† K = K2 .
Now consider item (iii). One direction is trivial: if there is such a basis, then K,
being diagonal in that basis, is also symmetric in that basis, and hence K = K† . So now
suppose that K = K† , and show that the desired basis exists. We proceed by using the
polar decomposition of K, being careful now to label explicitly any degeneracies:
  ( )
∑ (∑ )
∑ ∑
K= λα |fαj ⟩⟨eαj |  C , K2 = K† K = λα |eαj ⟩⟨eαj | ,
α j α j
  ( )
∑ (∑ )
†
∑ ∑
K = K† =  λα ∗
|eαj ⟩⟨fαj | C , K = KK =
2
λα |fαj ⟩⟨fαj | .
α j α j

(34)

5
The expressions on the right imply that within each eigensubspace of K2 , labeled by an
index α, we have ∑ (α)
|fαj ⟩ = |eαk ⟩Vkj , (35)
k

(α)
where Vkj is a unitary matrix. Notice that for the null subspace (λα = 0), we can assume
(α)
that Vjk is the unit matrix, i.e., |fαj ⟩ = |eαj ⟩. This being the result we want to get to (for
this subspace), we can afford to assume that λα ̸= 0 in the following. Plugging Eq. (35)
into the expressions on the left in Eq. (34) gives
   
∑ (∑ ) ∑ (∑ )
K= Vjk |eαj ⟩⟨eαk |  C =  Vkj |eαj ⟩⟨eαk |  C ,
(α) (α)
λα λα (36)
α j,k α j,k

(α) (α)
thus implying that Vjk = Vkj is symmetric. A symmetric unitary matrix can be diago-
nalized by an orthogonal transformation:

(α)
∑ (α) (α)
Vjk = Ojl Okl eiϕαl . (37)
l

(To see this, write V = eiH , where H = H † is Hermitian. The symmetry of V implies that
H = H T is also symmetric and, hence, that H = H ∗ is real. As a real, symmetric matrix,
H—and, hence, V —can be diagonalized by an orthogonal transformation. Another way
to see this, which we use below in the antisymmetric case, is to write V = A + iB, where
A = (V + V † )/2 = (V + V ∗ )/2 = A∗ = AT and B = −i(V − V † )/2 = −i(V − V ∗ )/2 =
B ∗ = B T are real and symmetric. The unitarity of V gives I = V † V = A2 + B 2 + i[A, B],
which implies that [A, B] = 0. Thus A and B—and, hence, V —can be diagonalized
simultaneously by an orthogonal transformation.)
Now define a new orthonormal basis by defining new vectors within each λα ̸= 0
eigensubspace: ∑
|e′αl ⟩ =
(α)
|eαj ⟩Ojl . (38)
j

For the null subspace, simply define |e′αj ⟩ = |eαj ⟩. Since the new basis is related to the old
by a (real) orthogonal transformation, complex conjugation in the two bases is the same,
i.e., C = C ′ . Thus we have
(  
∑ (∑ )) ∑ ∑
K= λα eiϕαl |e′αl ⟩⟨e′αl | C′ =  λj eiϕj |e′j ⟩⟨e′j | C ′ = C ′ λj e−iϕj |e′j ⟩⟨e′j | ,
α l j j
(39)

6
where in the second expression, we abandon explicit labeling of the degeneracies. We now
have K in diagonal form with K|e′j ⟩ = λj eiϕj |e′j ⟩. To bring K into the standard form
that has real, positive “eigenvalues,” we can rephase the eigenvectors appropriately, i.e.,
|e′′j ⟩ = eiϕj /2 |e′j ⟩,
( )
∑
K = C ′′ (C ′′ C ′ ) λj e−iϕj |e′j ⟩⟨e′j |
j
( )
∑ ( ) ′
′′ −iϕj ′′ ′
=C λj e C C |e′j ⟩
⟨ej |
j
| {z }
= eiϕj |e′j ⟩
( )
∑ (40)
= C ′′ λj |e′j ⟩⟨e′j |
j
( )
∑
′′
=C λj |e′′j ⟩⟨e′′j |
j
( )
∑
= λj |e′′j ⟩⟨e′′j | C ′′ .
j

A trivial consequence of our general results for symmetric antilinear operators is that
an antiunitary operator U is diagonalizable in an orthonormal basis if and only if U = U †
(or U 2 = 1), which means that U is complex conjugation in that basis.
It is worth emphasizing the freedom that is available in the eigenvectors of K = K† . For
this purpose, we drop the primes in the last form of Eq. (40) and again label degeneracies,
writing
 
∑ (∑ )
( )
K= λα |eαj ⟩⟨eαj |  C ⇐⇒ ⟨eαj | K|eβk ⟩ = λα δαβ δjk . (41)
α j

Allowed basis changes are restricted to unitary transformations within each eigensubspace,
∑ (α)∗
|fαj ⟩ = |eαk ⟩Vjk , (42)
k

which gives ( ) ( )
⟨fαj | K|fβk ⟩ = λα δαβ V (α) V (α)T jk . (43)

In the null subspace (λα = 0), any unitary transformation is allowed, but to maintain
a( diagonal )form in the nonnull subspaces (λα ̸= 0) requires that the unitary matrix
V (α) V (α)T be diagonal, i.e.,
( (α) (α)T )
V V jk
= eiϕj δjk . (44)

7
Defining a unitary matrix O(α) by

Ojk = e−iϕj /2 Vjk ,

(α) (α)
(45)

we see that ( ) ( )
O(α) O(α)T jk
= e−i(ϕj +ϕk )/2 V (α) V (α)T jk = δjk , (46)

which means that O(α) is a (real) orthogonal matrix.

We can summarize the freedom in the eigenvectors as follows. Starting from a set of
eigenvectors that put K in the standard form that has real, positive “eigenvalues,” we can
transform to new eigenvectors in each eigensubpace,
∑
|fαj ⟩ = e−iϕj /2
(α)
|eαk ⟩Ojk . (47)
k

In the null subspace, any unitary transformation is allowed. In the nonnull subspaces, these
(α)
new eigenvectors maintain the diagonal form of K provided that Ojk is an orthogonal
matrix. The general transformation is thus an orthogonal transformation followed by
rephasings of the eigenvectors. If we wish to retain real, positive “eigenvalues,” then we are
restricted to orthogonal transformations without the phases. In each nonnull eigensubspace
of K2 , K picks out a unique real subspace of the same dimension, which puts K in the
standard form with real, positive “eigenvalues.” For nondegenerate eigensubspaces, this
real subspace is simply a unique (up to a sign change) phase choice for the eigenvector.
The proof of item (iii) shows that to diagonalize a symmetric antilinear operator K, one
first diagonalizes the linear operator K2 and then in each nonnull eigensubspace of K2 ,
one finds the unique real subspace that puts K in the standard form with real, positive
eigenvalues.
I still want to include the proof of minimal “trace” for symmetric antilinears.

IV. Antisymmetric antilinear operators

We turn now to antilinear antiHermitian (or antisymmetric) antilinear operators, i.e.,
those satisfying K = −K† . Much of the discussion is nearly identical to that for symmetric
antilinear operators, but we can afford just to repeat it, there being little point in trying
to save space. It might be useful to consult the companion report entitled Antisymmetric
operators on a real vector space before reading this section.
That K is antilinear antiHermitian is equivalent to each of the following:
( ) ( )
(i) K is antisymmetric, i.e., ⟨ϕ| K|ψ⟩ = −⟨ψ| K|ϕ⟩ for all |ϕ⟩ and |ψ⟩.
(ii) K is antisymmetric in any one orthonormal basis (and then all orthonormal bases).
(iii) There exists an orthonormal basis |ejσ ⟩ where for each j, σ = ±1 (pairs of basis
vectors) or σ = 0 (single basis vectors), such that K|ejσ ⟩ = −σλj |ej,−σ ⟩; i.e., K has
matrix elements ( )
⟨ejσ | K|ekτ ⟩ = λj δjk σδσ,−τ .
| {z }
= ϵστ
The quantities λj can be made real and positive.

8
The first two of these are trivial consequences of the definition of the adjoint. Before
considering the third, notice that K2 |ejσ ⟩ = −K† K|ejσ ⟩ = −σ 2 |λj |2 |ejσ ⟩. Moreover, notice
that the reason that the λj ’s can be made real and positive is that under a rephasing of
the basis vectors, i.e., |e′jσ ⟩ = eiδjσ |ejσ ⟩, the λj ’s undergo a phase change,
( )
K|e′jσ ⟩ = K eiδjσ |ejσ ⟩
= e−iδjσ K|ejσ ⟩
( ) (48)
= e−iδjσ −σλj |ej,−σ ⟩
= −σλj e−i(δjσ +δj,−σ ) |e′j,−σ ⟩ .

For a single basis vector, the rephasing is irrelevant because K|ej0 ⟩ = 0; for a pair of
eigenvectors, λj is rephased by e−i(δj,+1 +δj,−1 ) . The real and positive quantities λj can be
characterized uniquely as the square roots of the eigenvalues of K† K = −K2 .
Now consider item (iii). One direction is trivial: if there is such a basis, then K is
clearly antisymmetric in that basis and, hence, K = −K† . So now suppose that K = −K† ,
and show that the desired basis exists. We proceed by using the polar decomposition of
K, being careful now to label explicitly any degeneracies:
  ( )
∑ (∑ )
∑ ∑
K= λα |fαj ⟩⟨eαj |  C , −K2 = K† K = λα |eαj ⟩⟨eαj | ,
α j α j
  ( )
∑ (∑ ) ∑ ∑
−K = K† =  λα ∗
|eαj ⟩⟨fαj | C , −K2 = KK† = λα |fαj ⟩⟨fαj | .
α j α j

(49)
The expressions on the right imply that within each eigensubspace of K , labeled by an 2

index α, we have ∑ (α)

|fαj ⟩ = |eαk ⟩Vkj , (50)
k
(α)
where Vkj is a unitary matrix. Notice that for the null subspace (λα = 0), we can assume
(α)
that Vjk is the unit matrix, i.e., |fαj ⟩ = |eαj ⟩. These null-subspace basis vectors serve
as the single basis vectors (σ = 0) in the ultimate basis we seek. Having dealt with the
λα = 0 case, we can afford to assume that we are dealing with the λα ̸= 0 subspaces in
the following. Plugging Eq. (50) into the expressions on the left in Eq. (49) gives
   
∑ (∑ (α) ) ∑ ( ∑ (α) )
K= λα Vjk |eαj ⟩⟨eαk |  C =  λα − Vkj |eαj ⟩⟨eαk |  C , (51)
α j,k α j,k

(α) (α)
thus implying that Vjk = −Vkj is antisymmetric. An antisymmetric unitary matrix can
be brought into the following form by an orthogonal transformation:
(α)
∑ (α) (α)
Vjk = Oj,lσ Ok,lτ eiϕαl ϵστ . (52)
l,σ,τ

9
(To see this, write V = iA+B, where A = −i(V +V † )/2 = −i(V −V ∗ )/2 = A∗ = −AT and
B = (V − V † )/2 = (V + V ∗ )/2 = B ∗ = −B T are real and antisymmetric. The unitarity
of V gives I = V † V = A2 + B 2 − i[A, B], which implies that [A, B] = 0. Thus A and
B—and, hence, V —can be simultaneously brought into the desired form by an orthogonal
transformation, i.e.,
∑
Ajk = Oj,lσ Ok,lτ µl ϵστ
l,σ,τ ∑
∑ =⇒ Vjk = Oj,lσ Ok,lτ (iµl + νl )ϵστ .
Bjk = Oj,lσ Ok,lτ νl ϵστ l,σ,τ
l,σ,τ

The unitarity of V guarantees, first, that the σ = 0 case does not occur and, second, that
iµl + νl = eiϕl is a phase, thus bringing Vjk into the desired form.)
Now define a new orthonormal basis by defining new vectors within each λα ̸= 0
eigensubspace: ∑
|e′α,lσ ⟩ =
(α)
|eαj ⟩Oj,lσ . (53)
j

For the null subspace, simply define |e′α,j0 ⟩ = |eαj ⟩ (σ = 0). Since the new basis is related
to the old by a (real) orthogonal transformation, complex conjugation in the two bases is
the same, i.e., C = C ′ . Thus we have
 
∑ (∑ )
K= λα eiϕαl ϵστ |e′α,lσ ⟩⟨e′α,lτ |  C ′
α l,σ,τ
 
∑ (54)
= λj eiϕj ϵστ |e′jσ ⟩⟨e′jτ | C ′
j,σ,τ
∑
= C′ λj e−iϕj ϵστ |e′jσ ⟩⟨e′jτ | ,
j,σ,τ

where in the second expression, we abandon explicit labeling of the degeneracies. We now
have K in the desired form with
∑
K|e′jσ ⟩ = λj eiϕj ϵτ σ |e′jτ ⟩ = −σλj eiϕj |e′j,−σ ⟩ .
τ

10
To bring K into the standard form that has real, positive λj ’s, we can rephase the eigen-
vectors appropriately, i.e., |e′′jσ ⟩ = ei(ϕj +σθj )/2) |e′jσ ⟩. This final basis change brings K into
the form ( )
∑
K = C ′′ (C ′′ C ′ ) λj e−iϕj ϵστ |e′jσ ⟩⟨e′jτ |
j,σ,τ
( )
∑ ( ′′ ′ ′ )
= C ′′ λj e−iϕj ϵστ C C |ejσ ⟩ ⟨e′jτ |
j,σ,τ
| {z }
= ei(ϕj +σθj )/2 |e′′jσ ⟩
( )
∑ (55)
= C ′′ λj ϵστ ei(σ+τ )θj /2 |e′′jσ ⟩⟨e′′jτ |
j,σ,τ
( )
∑
= C ′′ λj ϵστ |e′′jσ ⟩⟨e′′jτ |
j,σ,τ
( )
∑
= λj ϵστ |e′′jσ ⟩⟨e′′jτ | C ′′ .
j,σ,τ

A trivial consequence of our general results for antisymmetric antilinear operators is

that an antiunitary operator U can be brought to the form of item (iii) in an orthonormal
basis if and only if U = −U † (or U 2 = −1).
It is worth emphasizing the freedom that is available in the eigenvectors of K =
†
−K . For this purpose, we drop the primes in the last form of Eq. (55) and again label
degeneracies, writing
 
∑ (∑ )
( )
K= λα ϵστ |eα,jσ ⟩⟨eα,jτ |  C ⇐⇒ ⟨eα,jσ | K|eβ,kτ ⟩ = λα δαβ δjk ϵστ .
α j,σ,τ
(56)
Allowed basis changes are restricted to unitary transformations within each eigensubspace,
∑ (α)∗
|fα,jσ ⟩ = |eα,kτ ⟩Vjσ,kτ , (57)
k,τ

which gives ∑ (α)

( ) (α)
⟨fα,jσ | K|fβ,kτ ⟩ = λα δαβ Vjσ,lµ ϵµν Vkτ,lν . (58)
l,µ,ν

In the null subspace (λα = 0), any unitary transformation is allowed, but to maintain the
desired matrix elements in the nonnull subspaces (λα ̸= 0) requires that
∑ (α) (α)
Vjσ,lµ ϵµν Vkτ,lν = eiϕj δjk ϵστ . (59)
l,µ,ν

11
Defining a unitary matrix M (α) by

Mjσ,kτ = e−iϕj /2 Vjσ,kτ ,

(α) (α)
(60)

we see that ∑ (α) (α)

Mjσ,lµ ϵµν Mkτ,lν = δjk ϵστ . (61)
l,µ,ν

We need a little more efficient notation to appreciate the consequences of this require-
ment. Define the symplectic form

ϵjσ,kτ = δjk ϵστ , (62)

and let ϵ stand for this antisymmetric matrix, which satisfies ϵ = ϵ∗ = −ϵT = −ϵ† and
ϵ2 = −1. Equation (61), written in matrix notation, becomes

M (α) ϵM (α)T = ϵ . (63)

This requirement means that M (α) is a complex symplectic matrix, in addition to being
unitary. We can put the requirement (63) in an explicit form by first taking the adjoint,
M (α)∗ ϵM (α)† = ϵ, multiplying by ϵ on the right, ϵM (α)∗ ϵM (α)† = −1, and and then
comparing with M (α) M (α)† = 1 to get M (α) = −ϵM (α)∗ ϵ. Writing out this last equation
in index form gives
(α)
∑ (α)∗ (α)∗
Mjσ,kτ = ϵσµ ϵτ ν Mjµ,kν = στ Mj,−σ;k,−τ , (64)
µ,ν

which gives the explicit relations

(α) (α)∗ (α) (α)∗
Mj,+1;k,+1 = Mj,−1;k,−1 , Mj,+1;k,−1 = −Mj,−1;k,+1 . (65)

We can summarize the freedom in the basis vectors as follows. Starting from a set of
orthonormal vectors that put K in the standard form that has real, positive λj ’s, we can
transform to new eigenvectors in each eigensubpace,
∑
|fα,jσ ⟩ = e−iϕj /2
(α)∗
|eα,kτ ⟩Mjσ,kτ . (66)
k,τ

In the null subspace, any unitary transformation is allowed. In the nonnull subspaces,
(α)
these new orthonormal vectors maintain the diagonal form of K provided that Mjσ,kτ is
a complex symplectic matrix. The general transformation is thus a complex symplectic
transformation followed by rephasings of the basis vectors. If we wish to retain real, positive
λj ’s, then we are restricted to complex symplectic transformations without the phases. For
nondegenerate (two-dimensional) subspaces, the matrix M can be any unit-determinant
unitary matrix ( iθ )
e cos θ eiχ sin θ
M = ||Mστ || = . (67)
−e−iχ sin θ e−iθ cos θ

12