You are on page 1of 13

How is spontaneous symmetry breaking possible?

David John Baker

and Hans Halvorson

Abstract
We pose and resolve a seeming paradox about spontaneous symmetry breaking in
the quantum theory of infinite systems. For a symmetry to be spontaneously broken,
it must not be implementable by a unitary operator. But Wigner’s theorem guarantees
that every symmetry is implemented by a unitary operator that preserves transition
probabilities between pure states. We show how it is possible for a unitary operator
of this sort to connect the folia of unitarily inequivalent representations. This result
undermines interpretations of quantum theory that hold unitary equivalence to be
necessary for physical equivalence.
1 Introduction
The precise mathematical definition of spontaneous symmetry breaking (SSB) in quantum
theory is somewhat up for grabs. But all hands agree that, in the case of infinitely many
degrees of freedom, unitarily inequivalent representations are needed.
In physics more generally, SSB occurs when a ground state is not invariant under a
symmetry of the laws. This means that a symmetry transformation will take a ground
state to another (mathematically distinct) ground state. But in quantum field theory, an
(irreducible) Hilbert space representation of the commutation relations can include only a
single vacuum state. So a spontaneously broken symmetry must map between different
unitarily inequivalent representations.
Thus for SSB to occur in infinite quantum theory, the broken symmetry must not be
implemented by a unitary operator. This is generally agreed to be a necessary condition
(Emch and Liu, 2005) and is sometimes taken to be both necessary and sufficient (Earman,
2003; Strocchi, 2008).
Put this way, it can be difficult to see how a symmetry can possibly be spontaneously
broken. The difficulty arises from an apparent conflict with Wigner’s unitary-antiunitary
theorem, a foundational result that applies to all quantum theories. Since a symmetry
ought to preserve all the empirical predictions of a quantum state, it must not change the
transition probabilities between pure states, which are represented by the inner products

djbaker@umich.edu

hhalvors@princeton.edu
1
between vectors in a Hilbert space representation. Wigner’s theorem shows that any mapping
that preserves these probabilities for all vector states in a Hilbert space must be a unitary
mapping.
All symmetries preserve transition probabilities in this way. Besides being physically
intuitive, this can be proven rigorously even in paradigm cases of SSB. But this seems to
lead to paradox. We know SSB is possible in infinite quantum theory — there are well-
known examples. But we also know that any symmetry, even a broken one, must satisfy
the premises of Wigner’s theorem. Since SSB requires unitary inequivalence, these two facts
appear inconsistent.
This inconsistency must be only apparent. Our task is to explain why. We’ll begin by
explaining some general features that apply in all cases of quantum SSB. We will show that in
such cases, Wigner’s theorem applies. The seeming paradox therefore threatens. To resolve
it, we’ll show that the existence of a unitary symmetry in Wigner’s sense does not entail the
unitary equivalence of the Hilbert space representations it connects. There remains a sense
in which the symmetry is not (strictly speaking) unitarily implementable.
2 Quantum SSB
We begin by recalling some general properties of quantum theories on the algebraic approach.
At the broadest level of generality, a quantum theory is described by a C

-algebra A obeying
the canonical commutation or anticommutation relations (CCRs or CARs respectively) in
their bounded form. This is either an algebra of observables or (as in the present case) a
field algebra. The self-adjoint operators in A stand for physical quantities and are often
called observables. The states of the algebra are the possible assignments of expectation
values to the operators in A. These are given by normed linear functionals ω : A →C. The
expectation value of A in state ω is written ω(A).
A clear connection exists between this algebraic formalism and the better-known Hilbert
space formalism. The abstract algebraic states and observables can be concretely realized
by a Hilbert space representation of A (also called a representation of the CCRs/CARs).
Such a representation is a mapping π from A into the algebra of bounded operators B(1)
on a Hilbert space 1. The representation map is not usually a bijection; Hilbert space
representations will include more operators, and in particular more observables, than the
C

-algebra. Some of the states ω of A will be representable by density operators on 1 that
agree with their expectation values for all A in A. Collectively these states are called the
folium of the representation π.
Every algebraic state ω has a unique “home” representation in which it is given by a
cyclic vector. This is established by the
GNS Theorem. For each state ω of A, there is a representation π of A on a Hilbert space
1, and a vector Ω ∈ 1 such that ω(A) = ¸Ω, π(A)Ω¸, for all A ∈ A, and the vectors
¦π(A)Ω : A ∈ A¦ are dense in 1. (Call any representation meeting these criteria a GNS
representation.) The GNS representation is unique in the sense that for any other represen-
tation (1

, π

, Ω

) satisfying the previous two conditions, there is a unique unitary operator
2
U : 1
//
1

such that UΩ = Ω

and Uπ(A) = π

(A)U, for all A in A (see Kadison and
Ringrose, 1997, 278–279).
The definition of “same representation” presumed in this statement of uniqueness is called
unitary equivalence. We call representations π and π

unitarily inequivalent, and treat them
as distinct,
1
if there is no unitary operator U between their Hilbert spaces which relates the
representations by
Uπ(A) = π

(A)U. (1)
When equation (1) does hold, we say that the unitary U intertwines the representations π
and π

.
A useful source on the representation of symmetry in this framework is Roberts and
Roepstorff (1969). They posit (very reasonably) that any symmetry of a quantum system
must at a minimum consist of two bijections, α from the algebra of physical quantities A onto
itself and α

from the space of states of A onto itself. These must preserve all expectation
values, so that
α

(ω(α(A))) = ω(A). (2)
They then show that any such α is a ∗-automorphism of A, a bijection α : A
//
A which
preserves its algebraic structure and commutes with the adjoint mapping ()

. Furthermore,
α

can be defined in terms of this ∗-automorphism as it acts on states. Clearly α

(ω) is given
by ω ◦ α
−1
= ω(α
−1
(A)). We therefore have a justification, from physical principles, of the
oft-cited fact that symmetries in quantum theory are given by ∗-automorphisms.
Clearly if α is a symmetry and π(A) is a representation of A on a Hilbert space 1,
π ◦ α(A) = π(α(A)) is also a representation of A on 1. In this case α will act as a bijective
mapping from π to π ◦ α. We call α unitarily implementable in the representation π when
there is a unitary mapping U : 1
//
1 such that
π

(A) = π(α(A)) = Uπ(A)U

. (3)
This means the symmetry α is unitarily implementable in π iff π and π ◦ α are unitarily
equivalent representations of A.
This is where spontaneous symmetry breaking comes in. In general, a state ω breaks the
symmetry α only if α is not unitarily implementable in ω’s GNS representation.
2
When this
occurs, ω’s GNS representation and the GNS representation of the symmetry-transformed
state α

(ω) = ω ◦ α
−1
will be unitarily inequivalent.
In one example, the CAR algebra (so-called because it obeys the canonical anticommu-
tation relations) is the field algebra for a system of interacting spin-1/2 systems. We may
use the infinite version of the algebra to represent an infinitely long chain of spins confined
to a one-dimensional lattice, as in the Heisenberg model of a ferromagnet. This infinite CAR
algebra possesses a non-unitarily implementable automorphism which represents a symmetry
1
Although we always treat inequivalent representations as formally or mathematically distinct, note that
they may not always be physically inequivalent. As we shall see, they are sometimes related by symmetries,
which are normally assumed to preserve all the physical facts.
2
On the approach shared by Strocchi and Earman, this is both a necessary and sufficient condition.
3
of the ferromagnet: namely, a 180-degree rotation which flips all of the spins in the chain.
The rotation is therefore a spontaneously broken symmetry. See Ruetsche (2006) for a de-
tailed study of this case; we present only a few general features it shares in common with
other examples of SSB.
The lowest-energy states available to an infinite spin chain are ones in which all of the
spins align in the same direction. The Heisenberg ferromagnet has two such ground states:
ω, in which all of the spins point along +x (where x is the axis of the one-dimensional chain)
and ω

, in which they all point along −x. These states each define a GNS representation
(π, π

respectively) on Hilbert spaces 1 and 1

. If α is the automorphism of the CAR algebra
representing a 180-degree rotation along an axis perpendicular to x, ω

= α(ω). But since α
is spontaneously broken, π must be unitarily inequivalent to π

. This is where the seeming
paradox comes in.
3 Wigner’s theorem and the paradox
The expectation values of observables aren’t the only important quantities in quantum
physics. We should also expect a symmetry to preserve the transition probabilities be-
tween (pure) states of any quantum theory. In the Hilbert space formalism these are given
by inner products: ¸ψ, ψ

¸ represents the likelihood of a spontaneous transition from vector
state ψ to ψ

.
3
It seems obvious that no symmetry worth its salt will alter any of the transition proba-
bilities. This led Wigner to conclude that any symmetry worth its salt is given by a unitary
operator.
4
For the following can be proven (see Bargmann, 1964):
Wigner’s Theorem. Any bijection from the unit rays (vectors states) of a Hilbert space
1 to the unit rays of 1

which preserves the inner product is given by a unitary mapping
W : 1
//
1

.
This is puzzling. A spontaneously broken symmetry should still map bijectively between
the vector states of the two representations, and it would be very strange if it failed to
preserve transition probabilities. But we also know that broken symmetries are not unitarily
implementable. This seems impossible, but since well-known examples of SSB exist, paradox
threatens.
We are not the first to notice this problem. Considering the C

-algebra B(1) of all
bounded operators on a Hilbert space, Earman presents the problem as follows:
Obviously, if U is unitary then [the map A → UAU

] is an automorphism of
B(1). Conversely, any automorphism of B(1) takes this form for a unitary
U – ... T is an inner automorphism
5
of the C

-algebra B(1). How then can
3
Only on collapse interpretations do such transitions actually occur, of course. But we should nevertheless
expect other interpretations to retain the statistical predictions codified in transition probabilities.
4
Or by an antiunitary operator; for present purposes we ignore the difference.
5
An automorphism α is inner just in case there is a unitary V ∈ A such that α(A) = V AV

for all A ∈ A
4
a symmetry in the guise of an automorphism of a C

-algebra A fail to be an
unbroken or Wigner symmetry [one which preserves transition probabilities]?
(Earman, 2003, 341)
Earman provides an answer, but not a satisfactory one. He notes correctly that in cases of
SSB, the automorphism that represents the broken symmetry is not an automorphism of all
of B(1). Instead it is defined only on the field algebra or algebra of observables (the CAR
algebra in the ferromagnet example). In a Hilbert space representation this distinguished
algebra is given by a much smaller subalgebra of B(1). So since the symmetry is not an
automorphism of B(1), there is no reason to assume it should be unitary in the first place.
This only appears to resolve the problem because Earman has presented it in a limited
form. He considers the symmetry solely as it acts on operators, but the paradox can only
be properly motivated by considering the symmetry’s action on states. Whether α is an
automorphism of B(1) or not, α

is still a bijection between the states of representations 1
and 1

. If it’s not unitary, it must fail to meet one of the premises of Wigner’s theorem.
The problem is that the premises of Wigner’s theorem are satisfied by all symmetries.
Consider α

as it maps between the GNS representations of two ground states. Since a
symmetry always takes pure states to pure states, and the vector states of a GNS represen-
tation are all and only its pure states, α

is indeed a bijection of vectors states (unit rays).
Furthermore, as Roberts and Roepstorff (1969, 335) prove, α

must preserve all transition
probabilities. Wigner’s theorem therefore applies.
Earman is not alone in struggling with this seeming paradox. Other authors have sug-
gested (contrary to the theorem of Roberts and Roepstorff) that perhaps spontaneously bro-
ken symmetries do fail to preserve transition probabilities. For example, Arageorgis (1995,
302 fn 111) claims that, because “the preservation of probabilities in the Hilbert space for-
mulation implies the existence of a unitary (or antiunitary) operator,” no mapping between
the folia of inequivalent representations – presumably not even a symmetry transformation
– can preserve transition probabilities. Even Ruetsche (2006, 483) makes a rare slip on this
point:
A state ω breaks a dynamical symmetry α iff α is not unitarily implementable
on ω’s GNS representation. Unlike rotations on ground states of the finite spin
chain, dynamical symmetries do not conserve transition probabilities in the GNS
representation of states that break them. (Emphasis ours)
But to the contrary, by the very definition of a symmetry, all symmetries — whether broken
or not — preserve transition probabilities. Therefore, broken symmetries satisfy the premises
of Wigner’s theorem.
So the answer to Earman’s question — how can a symmetry fail to be a Wigner-type
unitary symmetry? — is that it cannot. And yet, there are unitarily non-implementable
symmetries. Paradox threatens.
5
4 The resolution
Since Wigner’s theorem applies to all symmetries, a spontaneously broken symmetry must
in some sense give a unitary mapping between the states of unitarily inequivalent represen-
tations. So there must be some wiggle room in the definition of unitary equivalence that
makes this possible. To resolve the paradox, we must look again at the definition and find
the wiggle room.
In effect, we have two data points to work with. First, as Roberts and Roepstorff prove,
Wigner’s theorem applies to all algebraic symmetries. This means that any symmetry α

,
as it acts on states, must be implemented by some unitary operator. Since the existence of
this operator is guaranteed by Wigner’s theorem, we’ll call it the Wigner unitary W. Since
for any state ω, α

(ω) = ω ◦ α
−1
, Wigner’s theorem is telling us that W must take the state
vector that represents ω to a vector that represents ω ◦ α
−1
.
Our second data point is the fact that spontaneous symmetry breaking is possible. This
implies that some symmetries are not unitarily implementable, in the sense that they fail
to satisfy Eq. (3) for any unitary U. When Eq. (3) is not satisfied, the symmetry’s action
on operators (which takes each operator A to α(A)) cannot be implemented by a unitary
operator. Combining our two data points, this means that spontaneously broken symmetries
are unitarily implementable as they act on states, but not as they act on operators.
This means that the unitary W must map the vector states of 1 to those of 1

without
satisfying Eq. 1 for U = W. In such a case, the representations are unitarily inequivalent even
though their Hilbert spaces are related by a unitary operator. There is nothing contradictory
about this, since the existence of a unitary operator implies unitary equivalence only if the
operator intertwines the representations. Their respective representations π, π

of the C

-
algebra may nonetheless be preserved by W, as long as W does not map the representation
π pointwise to the representation π

. So it may still be that
Wπ(A) = π

(A)W (4)
although there are individual operators A ∈ A for which (1) does not hold.
In fact, an operator meeting these criteria exists whenever the states of two GNS repre-
sentations are connected by a symmetry. We will show in steps that in every such case a
unitary W exists which satisfies Eq. (4), and which implements the symmetry as it acts on
states without implementing it as it acts on operators. First, we establish its existence:
Representation Wigner Theorem. Let ¸1, π, Ω¸ be a GNS representation for ω, and let
¸1

, π

, Ω

¸ be a GNS representation for ω ◦ α
−1
. Then there is a unique unitary operator
W = W
π,π
: 1
//
1

such that WΩ = Ω

, and Wπ(α
−1
(A)) = π

(A)W for all A ∈ A.
(Proof in Appendix 1.)
Since α
−1
(A) = A and W intertwines π ◦ α
−1
and π

, Eq. (4) follows. This means
that when π and π

are not unitarily equivalent, W maps between these two representations
without mapping π pointwise to π

, which is what we expected. In other words, W acts as a
bijection between the operators of these representations but does not, in general, implement
the symmetry as it acts on operators.
6
We have established that a unitary mapping preserving transition probabilities can exist
even between unitarily inequivalent representations. Indeed, such a mapping always exists
in cases of SSB. This is not yet enough to ensure that the conclusion of Wigner’s theorem
is true (as it must be). Wigner’s theorem ensures, not just that such a mapping exists, but
that every mapping which preserves transition probabilities must be unitary. This includes
every symmetry (whether spontaneously broken or not) as it acts on states.
So we must also show that a spontaneously broken symmetry can be given by a unitary
operator in the sense just discussed, without being unitarily implemented — that is, without
satisfying Eq. 3. To establish this, we will show that W itself implements the symmetry as
it acts on states.
Keep in mind that any representation π : A
//
B(1) of a C

-algebra gives rise to a map
π

of unit vectors of 1 into the state space of A. In particular,
π

(x)(A) = ¸x, π(A)x¸, (A ∈ A).
We now use this map to show that W implements the symmetry α

as it acts between the
states of representations π and π

.
Corollary 1. Let (1, π, Ω) be a GNS representation for ω. Then the Wigner unitary W for
α implements the action of α on vectors in 1. That is, (π

)

(Wx) = π

(x) ◦ α
−1
for any
unit vector x in 1. (Proof in Appendix 1.)
In other words, when we apply W to the state vector x ∈ 1 which represents the algebraic
state π

(x) in the GNS representation π, the result is the vector Wx ∈ 1

which represents
the state α



(x)) = π

(x) ◦ α
−1
in the representation π

. This is just what it means for W
to implement the symmetry as it acts on states.
Finally, we confirm that W does not in general implement the symmetry α as it acts on
operators. This is just to say that it does not in general intertwine the representations π
and π

. In fact, we can show that W intertwines these representations only if it is trivial:
Corollary 2. If the Wigner unitary W : 1
//
1

also induces a unitary equivalence between
π and π

, then α

◦ π

= π

. (Proof in Appendix 1.)
That is, every vector state in π’s Hilbert space is invariant under the symmetry α

, and
hence left unchanged by W. This means W must be the identity.
The astute reader may be puzzled by some of the properties we ascribe to the Wigner
unitary. In particular, we’ve shown that the Wigner unitary, which implements a symmetry
as it acts on states, never implements that same symmetry as it acts on operators unless
it is the trivial identity operator. How, then, can a non-trivial symmetry be unitarily im-
plemented on both states and operators (and hence unbroken)? This sort of puzzle is best
resolved by looking at concrete examples of Wigner unitaries in the case of both broken and
unbroken symmetries, which examples we provide in Appendix 2.
We’ve shown that whenever two states are related by a symmetry, a unitary mapping
exists between the Hilbert spaces of their GNS representations and has the properties we
would expect. This is the Wigner unitary. Its existence vindicates Wigner’s theorem, in that
7
it shows how the theorem can be true even when spontaneous symmetry breaking prevents
a symmetry from being unitarily implemented as it acts on operators. The seeming paradox
is no paradox at all.
5 Foundational significance
Besides the dissolution of a confusing paradox, are there foundational implications of this
result? We believe so. To underscore the foundational importance of our resolution of
the paradox, let’s briefly explore how it bears on one vexed question in the philosophy of
quantum field theory. What are the necessary conditions for physical equivalence between
field-theoretic states? In AQFT, the representations of the field algebra separate the states
into natural “families:” the folia of states given by density operators in each representation.
We may therefore ask what conditions must be met for two such families of states – two folia
– to represent the same set of physical possibilities.
For two folia to be physically equivalent, they must at least be empirically equivalent.
The paradoxical line of reasoning suggests that unitary equivalence is necessary if we want
to preserve transition probabilities. Since a quantum theory’s transition probabilities are
part of its empirical content, it would seem to follow that the folia of unitarily inequivalent
representations cannot predict the same empirical consequences – making them physically
inequivalent by the above reasoning.
This is why Arageorgis, while attempting “to clarify the connection between ‘intertrans-
latability’ [a necessary condition for physical equivalence] and ‘unitary equivalence,’” writes
in his seminal dissertation,
Intertranslatability requires a mapping between theoretical descriptions that pre-
serves the reports of empirical findings. These are couched in terms of proba-
bilities in quantum theory. And as Wigner has taught us, the preservation of
probabilities in the Hilbert space formulation implies the existence of a unitary
(or antiunitary)
6
operator. (Arageorgis, 1995, 302 fn 111)
He takes this point to establish that folia must belong to unitarily equivalent representations
if they are to count as physically equivalent. But his argument includes a false premise:
the assumption that the existence of a unitary operator connecting the folia of two repre-
sentations implies a unitary equivalence between those representations. As we have shown,
though, there is no such implication if the unitary operator is what we’ve called a Weyl
unitary. Arageorgis’s argument is unsound.
This means that at least one significant part of the empirical content of a quantum
theory – its transition probabilities – can be preserved by a mapping between the folia of
two inequivalent representations. If we further assume (as conventional wisdom dictates)
that a quantum theory’s symmetries preserve all empirical content, then the folia of at least
some pairs of inequivalent representations must be empirically equivalent if spontaneous
6
Recall that for purposes of this paper we ignore the distinction between unitary and antiunitary.
8
symmetry breaking is possible. The notion that unitary equivalence is a necessary condition
for physical equivalence should now appear quite suspect. Insofar as the so-called “Hilbert
space conservative” interpretation of quantum field theory identifies physical equivalence
with unitary equivalence (see Ruetsche, 2002), that interpretation must come into question
as well.
Appendix 1: Representation Wigner Theorem
Here we prove the results mentioned in the main text.
Definition. Let ω be a state on a C

-algebra A, let 1 be a Hilbert space with Ω a unit
vector in 1, and π : A
//
B(1) a representation of A. We say that the triple ¸1, π, Ω¸ is a
GNS representation for ω just in case:
1. ¸Ω, π(A)Ω¸ = ω(A), for all A ∈ A, and
2. ¦π(A)Ω : A ∈ A¦ is dense in the Hilbert space 1.
The GNS theorem shows that for each state ω, there is a GNS representation; and that
any two GNS representations of ω are unitarily equivalent.
Let A be a C

-algebra, let ω be a state of A, and let α be a ∗-automorphism of /. Let
(1, π, Ω) be the GNS triple of A induced by ω, and let (1

, π

, Ω

) be the GNS triple of A
induced by ω◦α
−1
. For brevity, we sometimes just use π and π

to denote the corresponding
triples.
Representation Wigner Theorem. Let ¸1, π, Ω¸ be a GNS representation for ω, and let
¸1

, π

, Ω

¸ be a GNS representation for ω ◦ α
−1
. Then there is a unique unitary operator
W = W
π,π
: 1
//
1

such that WΩ = Ω

, and Wπ(α
−1
(A)) = π

(A)W for all A ∈ A.
Proof. Let (1, π, Ω) be a GNS representation of A for the state ω, and let (1

, π

, Ω

) be a
GNS representation of A for the state ω ◦ α
−1
. Define W : 1
//
1

by setting
Wπ(A)Ω = π

(α(A))Ω

, ∀A ∈ A.
Since α(I) = I, it follows that WΩ = Ω

. Since

(α(A))Ω

|
2
= ¸Ω

, π

(α(A

A))Ω

¸ = ω(α
−1
(α(A

A))) = ω(A

A) = |π(A)Ω|
2
,
it follows that W is well defined and extends uniquely to a unitary operator from 1 to
1

. Note that since π(A)Ω = W

Wπ(A)Ω = W

π

(α(A))Ω

, it follows that W

π

(B)Ω

=
π(α
−1
(B))Ω for all B ∈ A. Therefore,
W

π

(A)Wπ(B)Ω = W

π

(Aα(B))Ω

= π(α
−1
(A)B)Ω = π(α
−1
(A))π(B)Ω,
for all A, B ∈ A. Since the vectors π(B)Ω, for B ∈ A, are dense in 1, it follows that
W

π

(A)W = π(α
−1
(A)) for all A ∈ A. That is, W implements a unitary equivalence from
π ◦ α
−1
to π

.
9
To show the uniqueness of W, it suffices to note that Ω is a cyclic vector for π ◦ α, Ω

is a cyclic vector for π

, and WΩ = Ω

. Thus, there is at most one unitary intertwiner from
π ◦ α
−1
to π

that maps Ω to Ω

.
Note that if α = ι is the identity automorphism, and if we take ¸1

, π

, Ω

¸ = ¸1, π, Ω¸,
then I satisfies the conditions of the theorem, hence by uniqueness W
π,π
= I.
For the following corollary, recall that any representation π : A
//
B(1) of a C

-algebra
gives rise to a map π

of unit vectors of 1 into the state space of A. In particular,
π

(x)(A) = ¸x, π(A)x¸, (A ∈ A).
Corollary 1. Let (1, π, Ω) be a GNS representation for ω. Then the Wigner unitary W for
α implements the action of α on vectors in 1. That is, (π

)

(Wx) = π

(x) ◦ α
−1
for any
unit vector x in 1.
Proof. By the Theorem, a Wigner unitary W intertwines π ◦ α
−1
and π

, that is
π(α
−1
(A)) = W

π

(A)W,
for all A ∈ A. Hence

)

(Wx)(A) = ¸Wx, π

(A)Wx¸ = ¸x, W

π

(A)Wx¸ = ¸x, π(α
−1
(A))x¸ = π

(x)(α
−1
(A)),
for all A ∈ A.
The preceding corollary can be conveniently pictured via a commuting diagram:
1 1

W
//
S(A)
1
OO
π

S(A) S(A)
α

//
S(A)
1

OO

)

where S(A) is the state space of A, and α

: S(A)
//
S(A) is the symmetry ω → ω ◦ α
−1
.
Corollary 2. If the Wigner unitary W : 1
//
1

also induces a unitary equivalence between
π and π

, then α

◦ π

= π

.
Recall that α

: S(A)
//
S(A) is defined by α

(ω) = ω ◦ α
−1
.
Proof. If W induces a unitary equivalence between π and π

then
π(A)W

= W

π

(A) = π(α
−1
(A))W

,
for all A ∈ A. Canceling the unitary operator W

on the right gives π(A) = π(α
−1
(A)) for all
A ∈ A, that is π = π ◦ α
−1
. From the latter equation it clearly follows that π

= α

◦ π

.
10
Appendix 2: Properties of the Wigner unitary operator
We now give a special case of the Representation Wigner Theorem which will illustrate some
properties of the Wigner unitary. But first we need a lemma.
Lemma. If ¸1, π, Ω¸ is a GNS representation for the state ω then ¸1, π ◦ α
−1
, Ω¸ is a GNS
representation for the state ω ◦ α
−1
.
Proof. Since ¸Ω, π(α
−1
(A))Ω¸ = ω(α
−1
(A)) and since Ω is cyclic under ¦π(α
−1
(A)) : A ∈ A¦,
it follows that ¸1, π ◦ α
−1
, Ω¸ is a GNS representation for ω ◦ α
−1
.
We can now apply the Representation Wigner Theorem to the representations ¸1, π, Ω¸
and ¸1, π ◦ α
−1
, Ω¸
Specialized Representation Wigner Theorem. If W : 1
//
1 is the Wigner operator
for the representations ¸1, π, Ω¸ and ¸1, π ◦ α
−1
, Ω¸, then W = I.
Proof. By RWT, WΩ = Ω and Wπ(α
−1
(A)) = π(α
−1
(A))W for all A ∈ A. Since α is an
automorphism, Wπ(A) = π(A)W for all A ∈ A. Hence
Wπ(A)Ω = π(A)WΩ = π(A)Ω,
for all A ∈ A. Since the set ¦π(A)Ω : A ∈ A¦ of vectors is dense in 1, it follows that Wx = x
for every vector in 1; that is, W = I.
We now apply SRWT to the case of symmetries in elementary quantum mechanics. Let
1 be a finite-dimensional Hilbert space, and let U : 1
//
1 be a unitary operator that
induces a symmetry. Then we have the following transformations:
ϕ −→ Uϕ transformed state
A −→ UAU

transformed observable
Of course, B(1) is a C

-algebra, and each (pure) state of B(1) is represented uniquely by
a ray in 1. The unitary U induces the automorphism α(A) = UAU

of B(1), as well as
the corresponding state mapping.
In order to apply SRWT, we need to find representations. The first representation is
¸1, ι, ϕ¸, where ι : B(1)
//
B(1) is the identity, and ϕ is an arbitrarily chosen unit vector.
The second representation is ¸1, ι◦α
−1
, ϕ¸. By GWT, there is a Wigner unitary W : 1
//
1,
and by SRWT, W = I.
So, in what sense does W induce the symmetry U on states? Should we not have W = U?
No, because a vector ϕ in 1 names different states on B(1) according to which representation
we consider, either ι or α
−1
. Relative to the first, ϕ represents the state A → ¸ϕ, Aϕ¸, and
relative to the second, ϕ represents the state A → ¸ϕ, α
−1
(A)ϕ¸.
What W does is to map a vector representing some state ω relative to π to a vector
representing the state ω◦α
−1
relative to π◦α
−1
. In the way we have set things up, W = 1
H
,
11
which just means that if ϕ represents ω relative to π, then ϕ represents ω ◦ α
−1
relative to
π

= π ◦ α
−1
. So, indeed, the identity map implements the symmetry ω → ω ◦ α
−1
of states!
Let us look now, more generally, at the case of an unbroken symmetry. By hypothesis,
the symmetry α is unbroken just in case the representations (1, π, Ω) and (1, π◦α
−1
, Ω) are
unitarily equivalent. That is, there is a unitary operator V : H
//
H such that V π(α
−1
(A)) =
π(A)V . In fact, in the most interesting case where ω is a pure state, V can be chosen such
that V = π(U) for some unitary operator U ∈ A, hence
π(α(A)) = V π(A)V

= π(UAU

),
for all A ∈ A. (To verify the existence of such a U ∈ A, see Kadison and Ringrose (1997,
730).)
Of course, we are still guaranteed the existence of the Wigner Unitary W : 1
//
1.
(In fact, we know that W = I; but ignore that fact for now.) Which operator, W or V ,
implements the symmetry α on states? The answer is that they both do, but in different
senses.
Compare the following two diagrams:
1 1
W
//
S(A)
1
OO
π

S(A) S(A)
α

//
S(A)
1
OO

)

1 1
V
//
S(A)
1
OO
π

S(A) S(A)
α

//
S(A)
1
OO
π

The square on the left shows the action of the Wigner unitary W for the special case of the
GNS representation (1, π ◦α
−1
, Ω) for ω◦α
−1
. The square on the right shows that action of
the unitary V that implements the equivalence between π and π◦α
−1
. The key difference, of
course, is that V implements the symmetry in such a way that the correspondence between
vectors and states can be held invariant (the vertical arrows are the same), whereas W’s
implementation requires a change of correspondence (π

versus (π

)

). But a state is a
way to map observables to numbers, so changing the correspondence between vectors and
states is equivalent to leaving this correspondence fixed and instead changing the labels of
observables. In equation form:

)

= α

◦ π

,
i.e. the correspondence (π

)

matches vectors with an observable A in exactly the way that
the correspondence π

matches vectors with the observable α
−1
(A).
Acknowledgements
Thanks to Wayne Myrvold and Giovanni Valente for detailed comments on an earlier draft.
DJB would like to thank Laura Ruetsche for several valuable discussions of the seeming
paradox and John Earman for helpful correspondence. HPH thanks Jeremy Butterfield,
Joe Rachiele, and Noel Swanson for helping him clarify the contrast between Wigner and
intertwining unitaries.
12
References
Arageorgis, A. (1995). Fields, Particles and Curvature. unpublished dissertation, University
of Pittsburgh.
Bargmann, V. (1964). Note on wigner’s theorem on symmetry operations. Journal of Math-
ematical Physics, 5:862–868.
Earman, J. (2003). Rough guide to spontaneous symmetry breaking. In Brading, K. and
Castellani, E., editors, Symmetries in Physics: Philosophical Reflections, pages 335–346.
Cambridge UP, Cambridge.
Emch, G. and Liu, C. (2005). Explaining quantum spontaneous symmetry breaking. Studies
in History and Philosophy of Modern Physics, 36:137–163.
Kadison, R. V. and Ringrose, J. R. (1997). Fundamentals of the theory of operator algebras.
Vol. II. American Mathematical Society, Providence, RI.
Roberts, J. E. and Roepstorff, G. (1969). Some basic concepts of algebraic quantum theory.
Communications in Mathematical Physics, 11:321–338.
Ruetsche, L. (2002). Interpreting Quantum Field Theory. Philosophy of Science, 69:348–378.
Ruetsche, L. (2006). Johnny’s so long at the ferromagnet. Philosophy of Science (Proceed-
ings), 73:473–486.
Strocchi, F. (2008). Symmetry Breaking. Springer, Berlin/Heidelberg.
13

(Call any representation meeting these criteria a GNS representation. Such a representation is a mapping π from A into the algebra of bounded operators B(H) on a Hilbert space H. A clear connection exists between this algebraic formalism and the better-known Hilbert space formalism. there is a unique unitary operator 2 . for all A ∈ A. there is a representation π of A on a Hilbert space H. must satisfy the premises of Wigner’s theorem. Collectively these states are called the folium of the representation π. The representation map is not usually a bijection. Wigner’s theorem applies. At the broadest level of generality. We will show that in such cases. Hilbert space representations will include more operators.between vectors in a Hilbert space representation. and a vector Ω ∈ H such that ω(A) = Ω. This is either an algebra of observables or (as in the present case) a field algebra. The expectation value of A in state ω is written ω(A). We’ll begin by explaining some general features that apply in all cases of quantum SSB. Ω ) satisfying the previous two conditions. For each state ω of A. Every algebraic state ω has a unique “home” representation in which it is given by a cyclic vector. a quantum theory is described by a C ∗ -algebra A obeying the canonical commutation or anticommutation relations (CCRs or CARs respectively) in their bounded form. To resolve it. Since SSB requires unitary inequivalence. π(A)Ω . and in particular more observables. we’ll show that the existence of a unitary symmetry in Wigner’s sense does not entail the unitary equivalence of the Hilbert space representations it connects. Some of the states ω of A will be representable by density operators on H that agree with their expectation values for all A in A. The self-adjoint operators in A stand for physical quantities and are often called observables.) The GNS representation is unique in the sense that for any other representation (H . this can be proven rigorously even in paradigm cases of SSB. Wigner’s theorem shows that any mapping that preserves these probabilities for all vector states in a Hilbert space must be a unitary mapping. than the C ∗ -algebra. All symmetries preserve transition probabilities in this way. 2 Quantum SSB We begin by recalling some general properties of quantum theories on the algebraic approach. and the vectors {π(A)Ω : A ∈ A} are dense in H. We know SSB is possible in infinite quantum theory — there are wellknown examples. these two facts appear inconsistent. even a broken one. Our task is to explain why. This is established by the GNS Theorem. These are given by normed linear functionals ω : A → C. π . This inconsistency must be only apparent. The abstract algebraic states and observables can be concretely realized by a Hilbert space representation of A (also called a representation of the CCRs/CARs). Besides being physically intuitive. But we also know that any symmetry. There remains a sense in which the symmetry is not (strictly speaking) unitarily implementable. But this seems to lead to paradox. The seeming paradox therefore threatens. The states of the algebra are the possible assignments of expectation values to the operators in A.

of the oft-cited fact that symmetries in quantum theory are given by ∗-automorphisms. the CAR algebra (so-called because it obeys the canonical anticommutation relations) is the field algebra for a system of interacting spin-1/2 systems. α from the algebra of physical quantities A onto itself and α from the space of states of A onto itself. π ◦ α(A) = π(α(A)) is also a representation of A on H. In this case α will act as a bijective mapping from π to π ◦ α. Clearly if α is a symmetry and π(A) is a representation of A on a Hilbert space H. Furthermore. 2 On the approach shared by Strocchi and Earman. (2) / A which They then show that any such α is a ∗-automorphism of A.2 When this occurs. they are sometimes related by symmetries. We therefore have a justification. 1997. These must preserve all expectation values. 278–279). We may use the infinite version of the algebra to represent an infinitely long chain of spins confined to a one-dimensional lattice. a state ω breaks the symmetry α only if α is not unitarily implementable in ω’s GNS representation. and treat them as distinct. We call α unitarily implementable in the representation π when / H such that there is a unitary mapping U : H π (A) = π(α(A)) = U π(A)U ∗ . Clearly α (ω) is given by ω ◦ α−1 = ω(α−1 (A)). note that they may not always be physically inequivalent. In one example. from physical principles. so that α (ω(α(A))) = ω(A). In general. for all A in A (see Kadison and U :H Ringrose.1 if there is no unitary operator U between their Hilbert spaces which relates the representations by U π(A) = π (A)U. This infinite CAR algebra possesses a non-unitarily implementable automorphism which represents a symmetry Although we always treat inequivalent representations as formally or mathematically distinct. a bijection α : A preserves its algebraic structure and commutes with the adjoint mapping (·)∗ . We call representations π and π unitarily inequivalent. (1) When equation (1) does hold. The definition of “same representation” presumed in this statement of uniqueness is called unitary equivalence. as in the Heisenberg model of a ferromagnet. This is where spontaneous symmetry breaking comes in. (3) This means the symmetry α is unitarily implementable in π iff π and π ◦ α are unitarily equivalent representations of A. we say that the unitary U intertwines the representations π and π . 1 3 . As we shall see. α can be defined in terms of this ∗-automorphism as it acts on states./ H such that U Ω = Ω and U π(A) = π (A)U . They posit (very reasonably) that any symmetry of a quantum system must at a minimum consist of two bijections. which are normally assumed to preserve all the physical facts. A useful source on the representation of symmetry in this framework is Roberts and Roepstorff (1969). this is both a necessary and sufficient condition. ω’s GNS representation and the GNS representation of the symmetry-transformed state α (ω) = ω ◦ α−1 will be unitarily inequivalent.

See Ruetsche (2006) for a detailed study of this case.3 It seems obvious that no symmetry worth its salt will alter any of the transition probabilities. But since α is spontaneously broken. These states each define a GNS representation (π. 3 Wigner’s theorem and the paradox The expectation values of observables aren’t the only important quantities in quantum physics. Considering the C ∗ -algebra B(H) of all bounded operators on a Hilbert space. 1964): Wigner’s Theorem.. ω = α(ω). W :H This is puzzling. paradox threatens. Conversely. any automorphism of B(H) takes this form for a unitary U – . if U is unitary then [the map A → U AU ∗ ] is an automorphism of B(H). and it would be very strange if it failed to preserve transition probabilities. in which they all point along −x. Any bijection from the unit rays (vectors states) of a Hilbert space H to the unit rays of H which preserves the inner product is given by a unitary mapping /H. T is an inner automorphism5 of the C ∗ -algebra B(H). but since well-known examples of SSB exist. This seems impossible. The lowest-energy states available to an infinite spin chain are ones in which all of the spins align in the same direction. 4 Or by an antiunitary operator. π must be unitarily inequivalent to π . The rotation is therefore a spontaneously broken symmetry. We are not the first to notice this problem. in which all of the spins point along +x (where x is the axis of the one-dimensional chain) and ω .of the ferromagnet: namely. The Heisenberg ferromagnet has two such ground states: ω. In the Hilbert space formalism these are given by inner products: ψ. How then can Only on collapse interpretations do such transitions actually occur. A spontaneously broken symmetry should still map bijectively between the vector states of the two representations. We should also expect a symmetry to preserve the transition probabilities between (pure) states of any quantum theory. we present only a few general features it shares in common with other examples of SSB. 5 An automorphism α is inner just in case there is a unitary V ∈ A such that α(A) = V AV ∗ for all A ∈ A 3 4 . ψ represents the likelihood of a spontaneous transition from vector state ψ to ψ . But we also know that broken symmetries are not unitarily implementable. If α is the automorphism of the CAR algebra representing a 180-degree rotation along an axis perpendicular to x.. This led Wigner to conclude that any symmetry worth its salt is given by a unitary operator. a 180-degree rotation which flips all of the spins in the chain.4 For the following can be proven (see Bargmann. Earman presents the problem as follows: Obviously. But we should nevertheless expect other interpretations to retain the statistical predictions codified in transition probabilities. of course. for present purposes we ignore the difference. This is where the seeming paradox comes in. π respectively) on Hilbert spaces H and H .

dynamical symmetries do not conserve transition probabilities in the GNS representation of states that break them. And yet. α is indeed a bijection of vectors states (unit rays). but not a satisfactory one. because “the preservation of probabilities in the Hilbert space formulation implies the existence of a unitary (or antiunitary) operator. (Emphasis ours) But to the contrary. the automorphism that represents the broken symmetry is not an automorphism of all of B(H). Instead it is defined only on the field algebra or algebra of observables (the CAR algebra in the ferromagnet example). but the paradox can only be properly motivated by considering the symmetry’s action on states. Arageorgis (1995. as Roberts and Roepstorff (1969. 335) prove. α is still a bijection between the states of representations H and H . Furthermore. So the answer to Earman’s question — how can a symmetry fail to be a Wigner-type unitary symmetry? — is that it cannot. all symmetries — whether broken or not — preserve transition probabilities. 5 . broken symmetries satisfy the premises of Wigner’s theorem. Consider α as it maps between the GNS representations of two ground states. He notes correctly that in cases of SSB. He considers the symmetry solely as it acts on operators. So since the symmetry is not an automorphism of B(H). Other authors have suggested (contrary to the theorem of Roberts and Roepstorff) that perhaps spontaneously broken symmetries do fail to preserve transition probabilities. Whether α is an automorphism of B(H) or not. α must preserve all transition probabilities. there is no reason to assume it should be unitary in the first place. Wigner’s theorem therefore applies. it must fail to meet one of the premises of Wigner’s theorem. 341) Earman provides an answer. by the very definition of a symmetry.” no mapping between the folia of inequivalent representations – presumably not even a symmetry transformation – can preserve transition probabilities.a symmetry in the guise of an automorphism of a C ∗ -algebra A fail to be an unbroken or Wigner symmetry [one which preserves transition probabilities]? (Earman. The problem is that the premises of Wigner’s theorem are satisfied by all symmetries. For example. Therefore. Earman is not alone in struggling with this seeming paradox. Paradox threatens. and the vector states of a GNS representation are all and only its pure states. there are unitarily non-implementable symmetries. If it’s not unitary. 2003. In a Hilbert space representation this distinguished algebra is given by a much smaller subalgebra of B(H). 483) makes a rare slip on this point: A state ω breaks a dynamical symmetry α iff α is not unitarily implementable on ω’s GNS representation. 302 fn 111) claims that. Even Ruetsche (2006. Since a symmetry always takes pure states to pure states. Unlike rotations on ground states of the finite spin chain. This only appears to resolve the problem because Earman has presented it in a limited form.

and W π(α−1 (A)) = π (A)W for all A ∈ A. we must look again at the definition and find the wiggle room. So there must be some wiggle room in the definition of unitary equivalence that makes this possible. There is nothing contradictory about this. an operator meeting these criteria exists whenever the states of two GNS representations are connected by a symmetry. Eq. a spontaneously broken symmetry must in some sense give a unitary mapping between the states of unitarily inequivalent representations. This means that any symmetry α . 1 for U = W . First. W maps between these two representations without mapping π pointwise to π . we’ll call it the Wigner unitary W . α (ω) = ω ◦ α−1 . This means that when π and π are not unitarily equivalent. (4). the representations are unitarily inequivalent even though their Hilbert spaces are related by a unitary operator. but not as they act on operators. π of the C ∗ algebra may nonetheless be preserved by W . which is what we expected. Wigner’s theorem is telling us that W must take the state vector that represents ω to a vector that represents ω ◦ α−1 . Wigner’s theorem applies to all algebraic symmetries. W acts as a bijection between the operators of these representations but does not. In other words. in the sense that they fail to satisfy Eq. (3) for any unitary U . must be implemented by some unitary operator. To resolve the paradox. as Roberts and Roepstorff prove. W = Wπ. (4) follows. This means that the unitary W must map the vector states of H to those of H without satisfying Eq. In fact. Their respective representations π. we establish its existence: Representation Wigner Theorem. In effect. Since for any state ω.π : H (Proof in Appendix 1. Combining our two data points. First. in general. π. as long as W does not map the representation π pointwise to the representation π . 6 . as it acts on states. Ω be a GNS representation for ω. (3) is not satisfied. Our second data point is the fact that spontaneous symmetry breaking is possible. the symmetry’s action on operators (which takes each operator A to α(A)) cannot be implemented by a unitary operator. In such a case. π . implement the symmetry as it acts on operators.) Since α−1 (A) = A and W intertwines π ◦ α−1 and π . We will show in steps that in every such case a unitary W exists which satisfies Eq. Since the existence of this operator is guaranteed by Wigner’s theorem. and let H . this means that spontaneously broken symmetries are unitarily implementable as they act on states.4 The resolution Since Wigner’s theorem applies to all symmetries. and which implements the symmetry as it acts on states without implementing it as it acts on operators. So it may still be that W π(A) = π (A)W (4) although there are individual operators A ∈ A for which (1) does not hold. Let H. Then there is a unique unitary operator / H such that W Ω = Ω . This implies that some symmetries are not unitarily implementable. we have two data points to work with. When Eq. since the existence of a unitary operator implies unitary equivalence only if the operator intertwines the representations. Ω be a GNS representation for ω ◦ α−1 .

Indeed. when we apply W to the state vector x ∈ H which represents the algebraic state π ∗ (x) in the GNS representation π. To establish this. Corollary 1. 3. Ω) be a GNS representation for ω. We’ve shown that whenever two states are related by a symmetry. which examples we provide in Appendix 2. This is the Wigner unitary. Let (H. In particular. That is. This is just to say that it does not in general intertwine the representations π and π . but that every mapping which preserves transition probabilities must be unitary. we will show that W itself implements the symmetry as it acts on states. π. This includes every symmetry (whether spontaneously broken or not) as it acts on states. we can show that W intertwines these representations only if it is trivial: / H also induces a unitary equivalence between Corollary 2. and hence left unchanged by W . In particular. (Proof in Appendix 1. Finally. without being unitarily implemented — that is. the result is the vector W x ∈ H which represents the state α (π ∗ (x)) = π ∗ (x) ◦ α−1 in the representation π . not just that such a mapping exists. which implements a symmetry as it acts on states. In fact. (A ∈ A). without satisfying Eq. in that 7 . (Proof in Appendix 1.) That is. then α ◦ π = π . can a non-trivial symmetry be unitarily implemented on both states and operators (and hence unbroken)? This sort of puzzle is best resolved by looking at concrete examples of Wigner unitaries in the case of both broken and unbroken symmetries. The astute reader may be puzzled by some of the properties we ascribe to the Wigner unitary. This means W must be the identity. π(A)x . We now use this map to show that W implements the symmetry α as it acts between the states of representations π and π . / B(H) of a C ∗ -algebra gives rise to a map Keep in mind that any representation π : A ∗ π of unit vectors of H into the state space of A. If the Wigner unitary W : H ∗ ∗ π and π . Then the Wigner unitary W for α implements the action of α on vectors in H. a unitary mapping exists between the Hilbert spaces of their GNS representations and has the properties we would expect. never implements that same symmetry as it acts on operators unless it is the trivial identity operator. then. This is just what it means for W to implement the symmetry as it acts on states. (π )∗ (W x) = π ∗ (x) ◦ α−1 for any unit vector x in H. How.) In other words. Its existence vindicates Wigner’s theorem. So we must also show that a spontaneously broken symmetry can be given by a unitary operator in the sense just discussed. such a mapping always exists in cases of SSB. we’ve shown that the Wigner unitary. we confirm that W does not in general implement the symmetry α as it acts on operators. π ∗ (x)(A) = x.We have established that a unitary mapping preserving transition probabilities can exist even between unitarily inequivalent representations. every vector state in π’s Hilbert space is invariant under the symmetry α . This is not yet enough to ensure that the conclusion of Wigner’s theorem is true (as it must be). Wigner’s theorem ensures.

let’s briefly explore how it bears on one vexed question in the philosophy of quantum field theory. And as Wigner has taught us. If we further assume (as conventional wisdom dictates) that a quantum theory’s symmetries preserve all empirical content. are there foundational implications of this result? We believe so. while attempting “to clarify the connection between ‘intertranslatability’ [a necessary condition for physical equivalence] and ‘unitary equivalence. But his argument includes a false premise: the assumption that the existence of a unitary operator connecting the folia of two representations implies a unitary equivalence between those representations. Since a quantum theory’s transition probabilities are part of its empirical content. then the folia of at least some pairs of inequivalent representations must be empirically equivalent if spontaneous 6 Recall that for purposes of this paper we ignore the distinction between unitary and antiunitary. they must at least be empirically equivalent. 8 . there is no such implication if the unitary operator is what we’ve called a Weyl unitary. though. the representations of the field algebra separate the states into natural “families:” the folia of states given by density operators in each representation. These are couched in terms of probabilities in quantum theory. This means that at least one significant part of the empirical content of a quantum theory – its transition probabilities – can be preserved by a mapping between the folia of two inequivalent representations. it would seem to follow that the folia of unitarily inequivalent representations cannot predict the same empirical consequences – making them physically inequivalent by the above reasoning. To underscore the foundational importance of our resolution of the paradox. 302 fn 111) He takes this point to establish that folia must belong to unitarily equivalent representations if they are to count as physically equivalent. The paradoxical line of reasoning suggests that unitary equivalence is necessary if we want to preserve transition probabilities. The seeming paradox is no paradox at all. For two folia to be physically equivalent. 1995.’” writes in his seminal dissertation. As we have shown. (Arageorgis. 5 Foundational significance Besides the dissolution of a confusing paradox. We may therefore ask what conditions must be met for two such families of states – two folia – to represent the same set of physical possibilities. Arageorgis’s argument is unsound.it shows how the theorem can be true even when spontaneous symmetry breaking prevents a symmetry from being unitarily implemented as it acts on operators. This is why Arageorgis. Intertranslatability requires a mapping between theoretical descriptions that preserves the reports of empirical findings. the preservation of probabilities in the Hilbert space formulation implies the existence of a unitary (or antiunitary)6 operator. What are the necessary conditions for physical equivalence between field-theoretic states? In AQFT.

π : H Proof. Ω be a GNS representation for ω ◦ α−1 . Insofar as the so-called “Hilbert space conservative” interpretation of quantum field theory identifies physical equivalence with unitary equivalence (see Ruetsche. Since π (α(A))Ω 2 ∀A ∈ A. Ω) be the GNS triple of A induced by ω. We say that the triple H. Representation Wigner Theorem. and let (H . Ω be a GNS representation for ω. for all A. π . 9 . Define W : H W π(A)Ω = π (α(A))Ω . That is. Let (H. Let ω be a state on a C ∗ -algebra A. there is a GNS representation. let H be a Hilbert space with Ω a unit / B(H) a representation of A. = Ω . Ω is a vector in H.symmetry breaking is possible. and let α be a ∗-automorphism of A. Note that since π(A)Ω = W ∗ W π(A)Ω = W ∗ π (α(A))Ω . for B ∈ A. Ω ) be the GNS triple of A induced by ω ◦ α−1 . Therefore. are dense in H. π. Let (H. π(A)Ω = ω(A). π (α(A∗ A))Ω = ω(α−1 (α(A∗ A))) = ω(A∗ A) = π(A)Ω 2 . and let H . π. Appendix 1: Representation Wigner Theorem Here we prove the results mentioned in the main text. W implements a unitary equivalence from π ◦ α−1 to π . W ∗ π (A)W π(B)Ω = W ∗ π (Aα(B))Ω = π(α−1 (A)B)Ω = π(α−1 (A))π(B)Ω. Ω) be a GNS representation of A for the state ω. Ω. Let A be a C ∗ -algebra. let ω be a state of A. it follows that W ∗ π (B)Ω = π(α−1 (B))Ω for all B ∈ A. Ω ) be a / H by setting GNS representation of A for the state ω ◦ α−1 . Definition. π . π . and let (H . it follows that W ∗ π (A)W = π(α−1 (A)) for all A ∈ A. it follows that W Ω = Ω . and π : A GNS representation for ω just in case: 1. it follows that W is well defined and extends uniquely to a unitary operator from H to H . B ∈ A. π. and 2. Since the vectors π(B)Ω. we sometimes just use π and π to denote the corresponding triples. Let H. Then there is a unique unitary operator / H such that W Ω = Ω . π. that interpretation must come into question as well. for all A ∈ A. and W π(α−1 (A)) = π (A)W for all A ∈ A. 2002). Since α(I) = I. The GNS theorem shows that for each state ω. The notion that unitary equivalence is a necessary condition for physical equivalence should now appear quite suspect. {π(A)Ω : A ∈ A} is dense in the Hilbert space H. For brevity. and that any two GNS representations of ω are unitarily equivalent. W = Wπ.

To show the uniqueness of W . 10 . From the latter equation it clearly follows that π ∗ = α ◦ π ∗ . and if we take H . and W Ω = Ω . π . (π )∗ (W x) = π ∗ (x) ◦ α−1 for any unit vector x in H. π ∗ (x)(A) = x. Corollary 1. that is π = π ◦ α−1 . then I satisfies the conditions of the theorem. Ω) be a GNS representation for ω. hence by uniqueness Wπ. Proof. π (A)W x = x. π(α−1 (A))x = π ∗ (x)(α−1 (A)). Ω = H. If W induces a unitary equivalence between π and π then π(A)W ∗ = W ∗ π (A) = π(α−1 (A))W ∗ . That is. W ∗ π (A)W x = x. By the Theorem. Thus. Let (H. Hence (π )∗ (W x)(A) = W x. In particular. π(A)x . it suffices to note that Ω is a cyclic vector for π ◦ α. for all A ∈ A. and α : S(A) Corollary 2. Proof. recall that any representation π : A ∗ gives rise to a map π of unit vectors of H into the state space of A. Note that if α = ι is the identity automorphism.π = I. for all A ∈ A. π. there is at most one unitary intertwiner from π ◦ α−1 to π that maps Ω to Ω . If the Wigner unitary W : H π and π . Ω is a cyclic vector for π . π. then α ◦ π ∗ = π ∗ . / B(H) of a C ∗ -algebra For the following corollary. Recall that α : S(A) / S(A) /H is the symmetry ω → ω ◦ α−1 . (A ∈ A). The preceding corollary can be conveniently pictured via a commuting diagram: S(A) O π∗ α / S(A) O (π )∗ H W / H / S(A) where S(A) is the state space of A. Ω . that is π(α−1 (A)) = W ∗ π (A)W. Canceling the unitary operator W ∗ on the right gives π(A) = π(α−1 (A)) for all A ∈ A. Then the Wigner unitary W for α implements the action of α on vectors in H. also induces a unitary equivalence between is defined by α (ω) = ω ◦ α−1 . for all A ∈ A. a Wigner unitary W intertwines π ◦ α−1 and π .

Since Ω. In order to apply SRWT. π. α−1 (A)ϕ . where ι : B(H) / H. because a vector ϕ in H names different states on B(H) according to which representation we consider. Ω is a GNS representation for the state ω then H. W = 1H . and ϕ is an arbitrarily chosen unit vector. The unitary U induces the automorphism α(A) = U AU ∗ of B(H).Appendix 2: Properties of the Wigner unitary operator We now give a special case of the Representation Wigner Theorem which will illustrate some properties of the Wigner unitary. π(α−1 (A))Ω = ω(α−1 (A)) and since Ω is cyclic under {π(α−1 (A)) : A ∈ A}. We can now apply the Representation Wigner Theorem to the representations H. Ω / H is the Wigner operator Specialized Representation Wigner Theorem. Ω is a GNS representation for the state ω ◦ α−1 . Let / H be a unitary operator that H be a finite-dimensional Hilbert space. By GWT. Ω and H. it follows that W x = x for every vector in H. Then we have the following transformations: ϕ −→ U ϕ transformed state ∗ A −→ U AU transformed observable Of course. Ω and H. for all A ∈ A. Proof. W Ω = Ω and W π(α−1 (A)) = π(α−1 (A))W for all A ∈ A. Relative to the first. The first representation is / B(H) is the identity. in what sense does W induce the symmetry U on states? Should we not have W = U ? No. π ◦ α . Since the set {π(A)Ω : A ∈ A} of vectors is dense in H. B(H) is a C ∗ -algebra. If H. and let U : H induces a symmetry. π ◦ α−1 . In the way we have set things up. ι. W = I. and each (pure) state of B(H) is represented uniquely by a ray in H. there is a Wigner unitary W : H and by SRWT. We now apply SRWT to the case of symmetries in elementary quantum mechanics. Ω . ϕ . W = I. The second representation is H. ϕ . Hence W π(A)Ω = π(A)W Ω = π(A)Ω. either ι or α−1 . H. Proof. it follows that H. that is. Ω is a GNS representation for ω ◦ α−1 . Aϕ . By RWT. Lemma. What W does is to map a vector representing some state ω relative to π to a vector representing the state ω ◦ α−1 relative to π ◦ α−1 . So. 11 . we need to find representations. π ◦ α−1 . π. and relative to the second. π. π ◦ α−1 . W π(A) = π(A)W for all A ∈ A. then W = I. ϕ represents the state A → ϕ. as well as the corresponding state mapping. But first we need a lemma. If W : H −1 for the representations H. Since α is an automorphism. ι◦α−1 . ϕ represents the state A → ϕ.

see Kadison and Ringrose (1997. whereas W ’s implementation requires a change of correspondence (π ∗ versus (π )∗ ). Ω) are /H such that V π(α−1 (A)) = unitarily equivalent. is that V implements the symmetry in such a way that the correspondence between vectors and states can be held invariant (the vertical arrows are the same). In equation form: (π )∗ = α ◦ π ∗ . 730). π ◦ α−1 . In fact.which just means that if ϕ represents ω relative to π. the identity map implements the symmetry ω → ω ◦ α−1 of states! Let us look now. so changing the correspondence between vectors and states is equivalent to leaving this correspondence fixed and instead changing the labels of observables. indeed. but in different senses. π ◦ α−1 . So. That is. HPH thanks Jeremy Butterfield. Of course. the symmetry α is unbroken just in case the representations (H. of course. more generally. (To verify the existence of such a U ∈ A. 12 . DJB would like to thank Laura Ruetsche for several valuable discussions of the seeming paradox and John Earman for helpful correspondence. i. Compare the following two diagrams: S(A) O π∗ α / S(A) O (π )∗ S(A) O π∗ α / S(A) O π∗ H W / H H V / H The square on the left shows the action of the Wigner unitary W for the special case of the GNS representation (H. hence π(α(A)) = V π(A)V ∗ = π(U AU ∗ ). The key difference. Ω) and (H. but ignore that fact for now. Acknowledgements Thanks to Wayne Myrvold and Giovanni Valente for detailed comments on an earlier draft.) Which operator. implements the symmetry α on states? The answer is that they both do. we are still guaranteed the existence of the Wigner Unitary W : H (In fact. By hypothesis. π. Ω) for ω ◦ α−1 . V can be chosen such that V = π(U ) for some unitary operator U ∈ A. W or V . then ϕ represents ω ◦ α−1 relative to π = π ◦ α−1 .e. we know that W = I. for all A ∈ A. at the case of an unbroken symmetry. Joe Rachiele. the correspondence (π )∗ matches vectors with an observable A in exactly the way that the correspondence π ∗ matches vectors with the observable α−1 (A). in the most interesting case where ω is a pure state.) / H. and Noel Swanson for helping him clarify the contrast between Wigner and intertwining unitaries. But a state is a way to map observables to numbers. The square on the right shows that action of the unitary V that implements the equivalence between π and π ◦ α−1 . there is a unitary operator V : H π(A)V .

. Ruetsche. II. F. and Roepstorff. University of Pittsburgh. 73:473–486. Johnny’s so long at the ferromagnet. Particles and Curvature. J. C. (2005). J. In Brading. RI. Symmetry Breaking. L. 11:321–338. (2003). (2006). Emch. Kadison. J. American Mathematical Society. V. Symmetries in Physics: Philosophical Reflections. Fields. Providence. Roberts. pages 335–346. (1964). L. R. R. G. Communications in Mathematical Physics. Ruetsche. A. (2002).References Arageorgis. 36:137–163. Earman. Some basic concepts of algebraic quantum theory. Philosophy of Science. Philosophy of Science (Proceedings). E. 13 . E. (1997). Cambridge. Vol. Fundamentals of the theory of operator algebras. Studies in History and Philosophy of Modern Physics. (2008). 5:862–868. Strocchi. and Castellani. Springer. Rough guide to spontaneous symmetry breaking. Journal of Mathematical Physics. K. (1995). G. Explaining quantum spontaneous symmetry breaking. Note on wigner’s theorem on symmetry operations. Bargmann. V. and Ringrose. Cambridge UP. 69:348–378. editors. (1969). and Liu. Berlin/Heidelberg. Interpreting Quantum Field Theory. unpublished dissertation.