CERN-TH-2023-086
Ferromagnetic phase transitions in SU ( N )
Alexios P. Polychronakos1,2 and Konstantinos Sfetsos3,4
arXiv:2306.01051v3 [hep-th] 24 Oct 2023
1 Physics
Department, the City College of the New York
160 Convent Avenue, New York, NY 10031, USA
apolychronakos@ccny.cuny.edu
2 The
Graduate School and University Center, City University of New York
365 Fifth Avenue, New York, NY 10016, USA
apolychronakos@gc.cuny.edu
3 Department
of Nuclear and Particle Physics,
Faculty of Physics, National and Kapodistrian University of Athens,
Athens 15784, Greece
ksfetsos@phys.uoa.gr
4 Theoretical
Physics Department, CERN,
1211 Geneva 23, Switzerland
October 26, 2023
Abstract
We study the thermodynamics of a non-abelian ferromagnet consisting of "atoms"
each carrying a fundamental representation of SU ( N ), coupled with long-range twobody quadratic interactions. We uncover a rich structure of phase transitions from
non-magnetized, global SU ( N )-invariant states to magnetized ones breaking global
invariance to SU ( N − 1) × U (1). Phases can coexist, one being stable and the other
metastable, and the transition between states involves latent heat exchange, unlike in
usual SU (2) ferromagnets. Coupling the system to an external non-abelian magnetic
field further enriches the phase structure, leading to additional phases. The system
manifests hysteresis phenomena both in the magnetic field, as in usual ferromagnets,
and in the temperature, in analogy to supercooled water. Potential applications are in
fundamental situations or as a phenomenological model.
Contents
1
Introduction
2
2
A system of interacting SU ( N ) "atoms"
3
2.1
The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
2.2
Decomposition of n fundamentals of SU ( N ) into irreps . . . . . . . . . .
7
2.3
The thermodynamic limit of the model . . . . . . . . . . . . . . . . . . . .
9
3
4
Phase transitions with vanishing magnetic fields
3.1
Stability analysis and critical temperatures . . . . . . . . . . . . . . . . . 15
3.2
Phase transitions and metastability . . . . . . . . . . . . . . . . . . . . . . 18
Turning on magnetic fields
4.1
4.2
5
11
25
Small fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.1.1
The singlet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1.2
The symmetric representation . . . . . . . . . . . . . . . . . . . . . 26
Finite fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2.1
One-row and conjugate one-row states . . . . . . . . . . . . . . . 30
4.2.2
Two-row states and their (N − 1)-row conjugates . . . . . . . . . 35
Conclusions
38
A Analysis of one-row states with a magnetic field
41
A.1 Resolving metastability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
B Analysis of two-row states in a magnetic field
45
B.1 The case of low temperatures . . . . . . . . . . . . . . . . . . . . . . . . . 49
1
1 Introduction
Magnetic materials are of considerable physical and technological interest, and their
properties have long been the subject of theoretical research. Ferromagnets, the first
type of magnetism ever observed, hold a special place among them, as they manifest
nontrivial properties and symmetry breaking.
All known ferromagnets consist of interacting localized magnetic dipoles and break
rotational invariance below the Curie temperature. Each (quantum) dipole provides
a representation of the group of rotations, that is, of SU (2). Although this is a nonabelian group, it is of a particularly simple type: it has a unique Cartan generator, and
SU (2) dipoles can interact with external abelian magnetic fields that couple to their
Cartan generator. Nevertheless, the exact quantitative properties of physical ferromagnets remain an active topic of research [1].
Independently, nonabelian unitary groups SU ( N ) of higher rank play a crucial role
in particle physics and, indirectly through matrix models, in string theory and gravity.
Ungauged and gauged SU (3) groups are the most common, representing "flavor" and
"color" degrees of freedom, respectively. A collection of nucleons, or the constituents
of the quark-gluon plasma, are physical systems of components carrying representations of SU (3). This raises the obvious question of the properties that large collections
of such SU (3) or, more generally, SU ( N ) entities would have if they interacted with
each other as well as with external nonabelian magnetic fields.
In this work we investigate the properties of such systems in the ferromagnetic
regime, that is, in the regime where the mutual interaction of its components would
tend to "align" their SU ( N ) charges, in a way that we will make precise. The results
on the decomposition of the direct product of an arbitrary number of representations
of SU ( N ) into irreducible components that we derived in a recent publication [2] will
be a crucial tool in our calculations. We will also study the effects of an external nonabelian magnetic field coupled to the system. As we shall demonstrate, the properties
of nonabelian (N > 2) ferromagnets are qualitatively different from these of ordinary
(N = 2) ferromagnets. They display a rich phase structure involving various critical
temperatures, hysteresis both in the temperature and in the magnetic field, coexistence
of phases, and latent heat transfer during phase transitions.
In the sequel we will present the basic SU ( N ) model, consisting of distinguishable
2
quantum components in the fundamental representation, and will review the relevant
group theory results of [2]. We will proceed to study the thermodynamic phases of the
model in the absence, and subsequently in the presence, of external magnetic fields,
and will derive its symmetry breaking patterns, critical temperatures, and magnetization. Stability issues will be crucial and will determine the pattern of SU ( N ) breaking
in the various phases. We will further study the nontrivial situation of a magnetic field
inducing an enhanced breaking of SU ( N ), and will conclude with some speculations
about the phenomenological relevance of the model.
2 A system of interacting SU ( N ) "atoms"
Magnetic systems with SU ( N ) symmetry have been considered in the context of ultracold atoms [3–7] or of interacting atoms on lattice cites [8–14] and in the presence
of SU ( N ) magnetic fields [15–17].
In this section we lay out the basic structure of any model of interacting SU ( N )
atoms, specify its ferromagnetic regime, and review the group theory results necessary
for its analytic treatment.
2.1 The model
To motivate the basic model, consider a set of n atoms (or molecules) on a lattice,
interacting with two-body interactions. Each atom is in one of N degenerate or quasidegenerate states |s⟩, s = 1, 2, . . . , N. The generic two-body interaction between atoms
1 and 2 with states |s1 ⟩ and |s2 ⟩ would be
N
H12 =
∑′
′
s1 ,s1 ,s2 ,s2 =1
hs1 s2 ;s′ s′ |s1 ⟩ ⟨s1′ | ⊗ |s2 ⟩ ⟨s2′ | ,
1 2
hs1 s2 ;s′ s′ = h∗s′ s′ ;s1 s2 .
1 2
1 2
(2.1)
Define ja , a = 0, 1, . . . , N 2 − 1, the generators of U ( N ) in the fundamental N-dimensional
representation, with j0 the identity operator (the U (1) part). Using the fact that the ja
form a complete basis for the operators acting on an N-dimensional space, the above
interaction can also be written as
H12 =
N 2 −1
∑
a,b=0
h ab j1,a j2,b ,
3
h ab = h∗ab ,
(2.2)
where
j1,a = ja ⊗ 1 ,
j2,a = 1 ⊗ ja ,
(2.3)
are the fundamental U ( N ) operators acting on the states of atoms 1 and 2.
We now make the physical assumption that the above interaction is invariant under a change of basis in the states |s⟩, that is, under a common unitary transformation
of the states |s1 ⟩ of atom 1 and |s2 ⟩ of atom 2. This implies two equivalent facts: first,
the interaction will necessarily be, up to trivial additive and multiplicative constants
(proportional to the identity), the operator exchanging the states of the atoms,
′
+ C12
H12 = C12
N
∑
′
s,s =1
|s⟩ ⟨s′ | ⊗ |s′ ⟩ ⟨s| ,
(2.4)
′ , C being real constants. Second, the interaction will necessarily be of the
with C12
12
form
H12 =
′
c12
+ c12
N 2 −1
∑
a =1
j1,a j2,a .
(2.5)
′ (due to the U (1) part), we obtain a unique two-body
Omitting the trivial constant c12
interaction depending on a single coupling constant c12 . Note that the group U ( N )
emerges from the requirement of invariance under general changes of basis of the N
states, and leads to interactions linear in the operators in each atom. Using, instead,
an N-dimensional representation of a smaller group would require the inclusion of
higher polynomial terms in j1,a and j2,a .
The interaction Hamiltonian of the full system will be of the form
n
H=
∑
crs
N 2 −1
∑
jr,a js,a ,
(2.6)
a =1
r,s=1
where crs = csr is the strength of the interaction between atoms r and s (and crr = 0).
This Hamiltonian involves an isotropic quadratic coupling between the fundamental
generators of the n commuting SU ( N ) groups of the atoms.
Reasonable physical assumptions restrict the form of the couplings crs . We assume
that the interaction is homogeneous, that is, crs is translationally invariant under the
shift of both r and s by the same lattice translation (away from the boundary of the
lattice). In terms of the lattice positions of the atoms ⃗r,
c⃗r,⃗s = c⃗r−⃗s ,
c0 = 0 .
4
(2.7)
Therefore, each atom couples to a fixed weighted average of the SU ( N ) generators
of its neighboring atoms. We will also assume that interactions are reasonably longrange, that is, each atom couples to several of its neigboring atoms. This technical
assumption justifies the mean field condition that, in the thermodynamic limit, the
weighted average of the neighboring atoms is well approximated by their average
over the full lattice. That is
∑ c⃗s j⃗r+⃗s,a ≃
⃗s
∑ c⃗s
⃗s
1 n
c
js,a = − Ja ,
∑
n s =1
n
(2.8)
where we defined the total SU ( N ) generator
n
Ja =
(2.9)
∑ js,a
s =1
and the effective mean coupling1
c = − ∑ c⃗s .
(2.10)
⃗s
The minus sign is introduced such that ferromagnetic interactions, driving atom states
to align, correspond to positive c. Altogether, the full effective interaction assumes the
form
c
H=−
n
N 2 −1
∑
a =1
Ja2
n
−
∑
s =1
2
js,a
,
(2.11)
where the second term in the parenthesis eliminates the terms r = s. The first part of
H is proportional to the quadratic Casimir of the total SU ( N ) group C (2) = ∑ a Ja2 . The
second part is proportional to the sum of the quadratic Casimirs of each individual
atom. Since all js,a are in the fundamental representation, their quadratic Casimir is
a (common) constant, independent of their state. So the second term contributes a
trivial constant and can be discarded.
In addition to the atoms’ mutual interaction, we can couple the states of the atoms
to a global external field, contributing an additional term
n
HB =
N
∑ ∑
′
r =1 s,s =1
1 The
Bss′ |s⟩r ⟨s′ |r .
(2.12)
validity of the mean field approximation is strongest in three dimensions, since every atom
has a higher number of near neighbors and the statistical fluctuations of their averaged coupling are
weaker, but is expected to hold also in lower dimensions.
5
Parametrizing this one-atom operator in terms of the complete set of operators jr,a and
omitting the trivial constant terms corresponding to jr,0 , it becomes
HB =
n N 2 −1
∑ ∑
Ba jr,a =
r =1 a =1
N 2 −1
∑
Ba Ja .
(2.13)
a =1
We see that Ba acts as a global nonabelian magnetic field on the SU ( N ) "spins" jr,a .
Finally, making use of the fact that the interaction Hamiltonian is invariant under
global SU ( N ) transformations, we may choose a basis of states in which the sum
∑ a Ba Ja is rotated to the Cartan subspace spanned by the commuting generators Hi ,
i = 1, 2, . . . , N − 1. The full Hamiltonian of the model then emerges as
c (2) N −1
H = − C − ∑ Bi Hi .
n
i =1
(2.14)
We will assume that c is positive, so that the model is of the ferromagnetic type.
For N = 2 the above model reduces to the ferromagnetic interaction of spin-half
components. For higher N, the model has the same number of states per atom as a
spin-S SU (2) model with 2S + 1 = N. The dynamics of the two models, however,
are distinct: the SU (2) model is invariant only under global SU (2) transformations,
which cannot mix the N states of the atoms in an arbitrary way, unlike the SU ( N )
case. The enhanced symmetry of the SU ( N ) model leads, as we shall see, to a richer
structure and to qualitatively different thermodynamic properties.
Finding the eigenstates of the above model and determining its thermodynamics
involves decomposing the full Hilbert space of states into irreducible representations
(irreps) of the total SU ( N ), evaluating the quadratic Casimir C (2) and the magnetic
sum ∑i Bi Hi in each irrep, and calculating the partition function as a sum over these
irreps. This requires determining the decomposition of the direct product of a large,
arbitrary number n of SU ( N ) fundamentals into irreps and the multiplicity of each
irrep in the decomposition, as well as calculating the Casimir and the magnetic sum
for large irreps of SU ( N ). This task was performed in a recent publication [2], and the
relevant results will be reviewed in the next subsection.
6
2.2 Decomposition of n fundamentals of SU ( N ) into irreps
We summarize the group theory results pertaining to the decomposition of the direct
product of n fundamentals of SU ( N ) into irreps, as presented in [2] (results on the
simpler case of SU (2) were previously derived in [18, 19] and were applied in [18] to
regular ferromagnetism).
The setting and results become most tractable and intuitive in the momentum
representation, in which irreps of SU ( N ) are labeled by a set of distinct integers k i ,
i = 1, 2, . . . , N ordered as
k1 > k2 > · · · > k N .
(2.15)
Each irrep corresponds to a given set {k i }, for which we will use the symbol k. The
corresponding Young Tableaux (YT) of the irrep may be described by its lengths ℓi ,
i.e., number of boxes per row, for i = 1, 2, . . . , N − 1. The correspondence with k i is
ℓi = k i − k N + i − N ,
ℓ1 ⩾ ℓ2 ⩾ · · · ⩾ ℓ N −1 ⩾ 0 .
(2.16)
Note that the k i representation is redundant, since a shift of all k i by a common constant k i → k i + c leaves ℓi invariant and leads to the same irrep of SU ( N ) (the shift
changes the U (1) charge of the irrep, which equals the sum of the k i ). This freedom can
be used to simplify relevant formulae. In our situation, where irreps will arise from
the direct product of n fundamentals, it will be convenient to choose the convention
N
∑ ki = n +
i =1
N ( N − 1)
.
2
(2.17)
For the singlet representation (n = 0) all ℓi are zero, which in the above convention
corresponds to k i = N − i, i = 1, 2, . . . , N. The fundamental (n = 1) has a single box,
and corresponds to k1 = N and the rest of the k i as above.
In SU ( N ) there are N − 1 Casimir operators which, for the irrep k, can be expressed
in terms of the k i ’s. For our purposes we need the quadratic Casimir, which is given
in terms of the k i by
C
(2)
1 N 2
N ( N 2 − 1)
1
(k) = ∑ ki −
.
[n + N ( N − 1)/2]2 −
2 i =1
2N
24
7
(2.18)
Note that, using (2.16), (2.17), C (2) takes the more familiar form
C (2) (ℓ) =
1
2
N −1
∑
i =1
ℓi (ℓi + N + 1 − 2i ) −
N −1
1
2N
∑
ℓi
i =1
!2
.
(2.19)
For the singlet C (2) = 0, while for the fundamental C (2) = ( N − N −1 )/2.
For our purposes we also need the trace of the exponential of the magnetic term in
a giver irrep k,
N
Trk exp β ∑ Bj Hj ,
(2.20)
j =1
which will appear in the calculation of the partition function of our model. This was
calculated in [2]. To express it, define the Slater determinant
z1k1 z1k2 · · ·
ψk (z) = (z1 · · · z N
1
) − N ∑i k i
z2k1 z2k2 · · ·
..
..
..
.
.
.
zkN1 zkN2 · · ·
k
z1N −1 z1k N
k
z2N −1 z2k N
..
..
.
.
,
z = { zi ∈ C } ,
(2.21)
k
z NN −1 zkNN
which is antisymmetric under the interchange of any two zi ’s and of any two k i ’s. Also
define the Vandermonde determinant
∆ ( z ) = ( z1 · · · z N ) −
N −1
2
z1N −1 z1N −2 · · ·
z2N −1 z2N −2 · · ·
..
..
..
.
.
.
N −1
−2
zN
zN
···
N
z1
z2
..
.
1
1
..
.
,
(2.22)
zN 1
which is the Slater determinant (2.21) for the singlet irrep. Then
N
ψ (z)
,
Trk exp β ∑ Bj Hj = k
∆(z)
j =1
z j = e βBj .
(2.23)
The prefactors involving the product z1 · · · z N in (2.21) and (2.22) eliminate the U (1)
part of the irrep, which couples to the trace of the magnetic field ∑i Bi . If B is traceless,
then the U (1) charge decouples and we can ignore these prefactors. As a check of
(2.23), we can take the limit zi → 1 and verify that the ratio of determinants goes to
Trk 1 = dim(k) =
N
ki − k j
∏ j−i =
j > i =1
8
N
ℓi − ℓ j + j − i
,
j−i
j > i =1
∏
(2.24)
which is the standard expression for the dimension of the irrep.
The last nontrivial element needed for our purposes is the multiplicity dn,k of each
irrep k arising in the decomposition of n fundamental representations. This was also
calculated in [2], and the result is
N
dn,k = δk1 +···+k N ,n+ N ( N −1)/2
Dn,k =
n!
∏rN=1 kr !
∏
j > i =1
(Si − S j ) Dn,k ,
(2.25)
,
where Si is a shift operator acting on the right by replacing k i by k i − 1. Note that Dn,k
and dn,k are manifestly symmetric and antisymmetric, respectively, under exchange
of the k i . In [2] the action of the operator ∏ N
j>i =1 ( Si − S j ) on Dn,k was performed and
an explicit combinatorial formula for dn,k was obtained, but it will not be needed for
our purposes.
2.3 The thermodynamic limit of the model
We now have all the ingredients to study the statistical mechanics of our SU ( N ) ferromaget. The partition function is
Z=
∑
e
states
− βH
∑ dn;k e
=
βc (2)
n C (k)
N
Trk exp β ∑ Bj Hj ,
⟨k⟩
j =1
(2.26)
where β is the inverse of the temperature T and ⟨k⟩ denotes distinct ordered integers
k1 > k2 > · · · > k N satisfying the constraint (2.17). Using the results (2.18), (2.23)
and (2.25), and removing the trivial (k i -independent) terms in the Casimir (2.18), the
partition function becomes
1
δk +···+k ,n+ N ( N −1)
Z=
N
1
N! ∑
2
k
= ∑ δk
k
N ( N −1)
1 +···+ k N ,n +
2
= ∑ δk1 +···+k N ,n
k
N
∏
j > i =1
1
∆(z)
1
n!
∆(z) ∏rN=1 kr !
( Si − S j )
n!
∏rN=1 kr !
N
∏
j > i =1
( Si − S j )
N
∏
j > i =1
9
n!
∏rN=1 kr !
ψk (z) βc ∑s k2s
e 2n
∆(z)
βc
2
e 2n ∑s ks + βBs ks
βc ∑ k2 + βB k
1
s s
Si−1 − S−
e 2n s s
.
j
(2.27)
In the first line above we made the sum unrestricted, since the summand is symmetric
under permutation of the k i and vanishes for k i = k j , and introduced the constraint
explicitly. The second line follows since dn,k (the expression in the parenthesis) is
antisymmetric in the k i , and thus it picks the fully antisymmetric part of z1k1 · · · zkNN ,
reproducing ψk (z). The third line is obtained by shifting summation variables. In
doing so, the term N ( N − 1)/2 in the Kronecker δ is absorbed.
The above holds for arbitrary n. We now take the thermodynamic limit n ≫ 1.
The typical k i is of order n, and thus the exponent in the expression is of order n,
and any prefactor polynomial in n is irrelevant, as is the factor ∆(z). Similarly, the
1
action of ∏(Si−1 − S−
j ) produces a subleading factor that can be ignored. (One way
j >i
to see this is to note that in the large n limit the shift operators act as derivatives
1
(Si−1 − S−
≃ ∂ki − ∂k j ) and bring down subleading terms). Further, we apply to kr !
j
the Stirling approximation. Altogether we obtain
Z=
0
∑ δk1 +···+k N ,n e−β F(k)+O(n ) ,
(2.28)
k
where the free energy of the system is, up to a trivial overall constant,
N
c 2
k i − Bi k i .
F(k) = − Tn ln n + ∑ Tk i ln k i −
2n
i =1
(2.29)
In the large-n limit, quantities k i and F are extensive variables of order n. We will
now transition to intensive variables, that is, quantities per atom. To this end, we
define rescaled variables xi as
k i = nxi ,
i = 1, 2, . . . , N .
(2.30)
satisfying the constraint
N
∑ xi = 1 .
(2.31)
i =1
In terms of the xi , the non-extensive term − Tn ln n in the free energy cancels and F
becomes properly extensive,
N
NT0 2
xi − Bi xi = nF (x) ,
F(x) = n ∑ Txi ln xi −
2
i =1
10
(2.32)
where we have defined
c = NT0 ,
(2.33)
introducing a temperature scale T0 . From now on we will work with the intensive
quantities xi (magnetization per atom) and F (free energy per atom) and will omit the
qualifier "per atom".
In the large-n limit the sum in (2.28) can be obtained by a saddle-point approximation, as the exponent is of order n, by minimizing the free energy F (x) while respecting
the constraint ∑iN=1 xi = 1. This can be done with a Lagrange multiplier. Adding the
term λ(1 − ∑iN=1 xi ) to (2.32) and varying with respect to xi we obtain
∂i Fλ = T ln xi − NT0 xi − Bi − λ = 0 ,
i = 1, 2, . . . , N .
(2.34)
The Lagrange multiplier λ can be eliminated by subtracting one of the relations, say
for i = N, from the rest (which is equivalent to solving the constraint and expressing
one of the xi , say x N , in terms of the others). We obtain
T ln
xi
− NT0 ( xi − x N ) − ( Bi − BN ) = 0 ,
xN
i = 1, 2, . . . , N − 1 ,
(2.35)
where x N is determined from the constraint (2.31). Also, from (2.34) we obtain the
second derivatives
∂i ∂ j Fλ =
T
xi
− NT0 δij ,
i, j = 1, 2, . . . , N ,
(2.36)
subject to (2.31). The above Hessian will be needed later in order to investigate the
stability of the solutions. The simpler form of the equations (2.34) involving λ will
also be useful in determining the nature of solutions and in the stability analysis.
3 Phase transitions with vanishing magnetic fields
We now put Bi = 0 (setting all of the Bi equal is equivalent, as this would be a U (1)
field and would contribute a trivial constant to the energy) and examine the phase
structure of the system. We can collectively write equations (2.34) for Bi = 0 as
T ln x − NT0 x = λ ,
11
(3.1)
dropping the index i in xi to emphasize that it is the same equation for all xi ’s, unlike
the case with generic non-vanishing magnetic fields. The value of λ is fixed by the
summation condition (2.31).
We note that (3.1) always admits the trivial solution xi = 1/N (for an appropriate
λ), corresponding to the singlet irrep and an unbroken SU ( N ) phase. Generically,
however, the above equation has two solutions (see fig. 1). So, each xi can have one of
two fixed values, x− or x+ > x− . This means that the dominant irreps are those with
M equal rows, where M is the number of xi having the large stvalue x+ in the solution,
and gives SU ( M) × SU ( N − M) × U (1) as the possible a priori spontaneous breaking
of SU ( N ), the subgroup that preserves a matrix with M equal and N − M different
and equal diagonal entries. We will see, however, that stability of the configuration
requires that at most one xi value in the full solution be x = x+ ; that is, either M = 0,
corresponding to the singlet, or M = 1, corresponding to a one-row YT, a completely
symmetric representation.
T ln x - N T0 x
λ
x-
x0
x+
x
T
. The intersection
NT0
with some constant value of the Lagrange multiplier λ occurs at x = x± , with x− < x0 < x+ .
Figure 1: Plot of the LHS of (3.1), with its maximum occurring at x0 =
Then equations (2.35) with Bi = 0 become
T ln
xi
= NT0 ( xi − x N ) ,
xN
i = 1, 2, . . . , N − 1 ,
(3.2)
where x N = 1 − x1 − · · · − x N −1 is determined by the constraint in (2.31).
As argued before from (3.1), each xi can have one of two possible values. Hence, take
M of the xi to be equal, and the remaining N − M also equal and different. The integer
M can take any value from 0 to N, but the values M = 0 and M = N correspond to
the singlet configuration xi = 1/N that trivially satisfies (3.1). For M ̸= 0, N, taking
12
into account the summation to one condition, we set
1+x
,
N
1 − ax
xi =
,
N
i = 1, 2, . . . , M ,
xi =
i = M + 1, . . . , N ,
M
a=
.
N−M
(3.3)
Note that, according to (2.15), the xi ’s cannot strictly be equal for finite n. However, in
the large n limit, differences of O(1/n) are ignored. Further, the choice of the specific
xi that we set to each value is irrelevant, since the saddle point equations for Bi = 0 are
invariant under permutations of the xi . With the choice (3.3), N − M of the equations
are identically satisfied and the remaining M amount to
T ln
1+x
− (1 + a) T0 x = 0 .
1 − ax
(3.4)
This transcendental equation is invariant under the transformation
x → − ax ,
a → 1/a (equivalently M → N − M) .
(3.5)
Thus, without loss of generality we can choose
M ⩽ [ N/2] ,
or
0<a⩽1,
x ∈ (−1, 1/a) ,
(3.6)
where [ · ] denotes the integer part. Solutions with x > 0 specify an irrep with M equal
rows of length, using (2.16) and (3.3),
ℓi =
x
n + O(1) ,
N−M
i = 1, . . . , M ,
ℓi = O(1) ,
i = M + 1, . . . , N − 1 . (3.7)
Instead, an x < 0 specifies an irrep with N − M equal rows (corresponding to the
conjugate representation), with length given by (3.7) but with x replaced by x → − x.
The reason is that in this case the xi ’s in the second line of (3.3) are larger than those
of the first line and therefore the roles of M and N − M are reversed.
There are generically either one or three solutions to (3.4) depending on T (see fig. 2).
If the temperature is higher than a critical temperature Tc , then the only solution is
that with x = 0, that is, the singlet. If T < Tc , then there are two additional solutions.
For T = Tc these two solutions coalesce at x = xc , implying that the x-derivate of (3.4)
13
is zero as well at xc . These conditions are summarized as
1 + xc
Tc
ln
= (1 + a ) x c ,
T0 1 − axc
(3.8)
Tc
(1 + xc )(1 − axc ) =
≡t.
T0
Solving the first condition for t = Tc /T0 and substituting into the second we obtain a
transcendental equation that determines xc
(1 + a ) x c
1 + xc
= ln
(1 + xc )(1 − axc )
1 − axc
(3.9)
and from that and the first of (3.8) the critical temperature Tc . Alternatively, solving
the second equation in (3.8) for xc and substituting into the first one yields the transcendental equation for t = T/T0
ln
p
q
1 + a
1 1 + a + (1 + a)2 − 4at
p
−
1 − a + (1 + a)2 − 4at = 0 ,
a 1 + a − (1 + a)2 − 4at
2at
(3.10)
which assumes that a ⩽ 1. For a ⩾ 1 we simply replace a → 1/a.
For SU (2), a = 1 is the only possibility, and for a = 1 the solution of (3.10) is t = 1.
Hence, the critical temperature is just T0 , i.e. the one in (2.33). For generic a, it can
be checked that the left hand side of (3.10) is monotonically increasing in t and the
(1 + a )2
equation has a unique solution in the range 1 < t <
for all a ⩽ 1. For a near
4a
1 we have
1
(3.11)
t = 1 + (1 − a )2 + . . . ,
6
whereas for a near 0
1
= a − ln a + ln(− ln a) + . . .
t
(3.12)
and therefore t → ∞. Hence, the solution to (3.10) varies monotonically between these
two limiting cases. Note that for the case M = 1, a = 1/( N − 1) (which is particularly
relevant, as we shall see)
M=1:
Tc ≃ T0
N
≫ T0 ,
ln N
as
N ≫1.
(3.13)
In conclusion, Tc > T0 for any group SU ( N ) with N ⩾ 3. For T > Tc the only solution
is the trivial one, x = 0, while for T < Tc there are two additional solutions x1 , x2 : two
positive ones 0 < x1 < xc < x2 for T0 < T < Tc , and one positive and one negative
14
one x1 < 0 < x2 for T < T0 . A few generic cases are depicted in Fig. 1 where the left
hand side of (3.4) is plotted.
1.0
0.5
0.5
-1.0
-0.5
0.5
1.0
x
-1
1
2
3
4
5
6
x
-0.5
-1.0
-0.5
-1.5
Figure 2: Plots of the left hand side of (3.4) for N = 7. Left: M = 3, T = 0.7. Right: M = 1,
T = 1.6 (blue) and for T = Tc ≃ 1.72 (corresponding to x = xc ≃ 3.88) (orange). Temperature
is in units of T0 .
3.1 Stability analysis and critical temperatures
The solutions of (3.4) may be local extrema or saddle points of the action. To determine
their stability we examine the second variation of the free energy. From (2.36), δ2 F is
δ2 F =
1 N −1 2
Ci δxi ,
2 i∑
=1
Ci−1 =
T
− NT0 ,
xi
N
∑ δxi = 0 .
(3.14)
i =1
Note that this remains the same even in the presence of magnetic fields, so the stability
argument below is fully general.
If all coefficients Ci are positive at a stationary point, then clearly the solution is
stable. If two or more Ci ’s are negative, on the other hand, it is unstable. Indeed, we
can take, e.g., δxi1 + δxi2 = 0 for two of the negative coefficients Ci1 and Ci2 , and set
the rest of the δxi to zero in order to satisfy the constraint. Then, an obvious instability
arises. However, if only one Ci is negative and the rest of them positive, then the
solution could still be stable due to the presence of the constraint. A standard analysis
15
shows that the condition for stability in this case is2
N
∑ Ci < 0 .
(3.15)
i =1
For the case of vanishing magnetic field, xi satisfy the common equation (3.1)
T ln x − NT0 x = λ .
(3.16)
The function on the left hand side is plotted in fig. 1, repeated here as fig. 3, and has
a maximum at x0 =
T
NT0
for all i. For x = x0 , the coefficient Ci−1 determining the
perturbative stability at this value vanishes, while Ci−1 > 0 for x < x0 and Ci−1 < 0
for > x0 . Therefore, the left branch of the curve represents a priori stable points and
the right branch unstable ones.
T ln x - N T0 x
15
λ
10
5
5
10
15
20
x
T
. The intersection with some
NT0
constant value of the Lagrange multiplier λ occurs at the values xi = x± , with x− < x0 < x+ .
The left (green) part of the curve corresponds to stable solutions, whereas the right one (red)
to unstable ones.
Figure 3: Plot of (3.16), with its maximum occurring at x0 =
From the above general discussion, we understand that fully stable configurations
correspond either to choosing all xi ’s on the stable branch (all Ci > 0), or N − 1 of them
on the stable branch and one on the unstable branch (only one Ci < 0). This last con-
figuration can still be stable if it satisfies the condition (3.15). So the only potentially
stable configurations for zero magnetic field are the singlet (M = 0), corresponding
to a paramagnetic phase, and the fully symmetric single-row irrep (M = 1), corre2A
nice way to derive this is to view δxi as covariant coordinates on a space with metric gij = Ci δij
of Minkowski signature (−, +, . . . , +). Then δ2 F and the constraint become
δ2 F = gij δxi δx j , ui δxi = 0 with ui = 1 .
For the space spanned by the restricted δxi to be spacelike (with positive definite metric), ui must be
timelike, and ui ui = gij ui u j < 0 implies (3.15).
16
sponding to a specific ferromagnetic phase, with order parameter the variable x in
(3.3) determining the length of the single row.
For the remainder of the zero magnetic field discussion we will focus on the nontrivial solution M = 1 and set for the constant a = M/( N − M) the corresponding
value
a=
1
.
N−1
(3.17)
For this choice, and with xi as in (3.3), the coefficients Ci become
C1−1 = N A( x ) ,
Ci−1 = N A(− ax ) ,
where we defined
A( x ) =
i = 2, 3, . . . , N ,
T
− T0 .
1+x
(3.18)
(3.19)
The stability of the no magnetization solution x = 0 corresponding to the singlet is
easy to find. In that case A(0) = T − T0 , so that for T < T0 the solution x = 0 is a local
maximum and unstable, and for T > T0 it is a local minimum.
For M = 1, it is clear from fig. 3 that we must have C1 < 0, corresponding to the
single value x1 = (1 + x )/N on the unstable branch, and the remaining Ci > 0, so
A( x ) < 0 and A(− ax ) > 0. Also, 1 + x > 1 − ax, so that x > 0. Condition (3.15) must
also be satisfied,
A( x ) + aA(− ax ) > 0
=⇒
(1 + x )(1 − ax ) <
T
,
T0
(3.20)
where we used N − 1 = a−1 . Using (3.4) to eliminate T this rewrites as
(1 + a ) x
1+x
> ln
.
(1 + x )(1 − ax )
1 − ax
(3.21)
Note that this has the same form as (3.9) determining the critical xc . It can be seen that
it is satisfied for x > xc and violated for x < xc .
As analyzed in the previous section, the existence of solutions with M = 1 requires
T < Tc . For such temperatures, equation (3.4) has two solutions, one larger and one
smaller than xc . Only the solution with x > xc satisfies (3.21). Therefore, for temperatures T < Tc the solution with x > xc is stable and a local minimum and the one with
x < xc unstable. Referring again to Fig. 2, the solution to the left of x = xc for T < Tc
(blue) on the right plot is unstable, whereas the one to the right is stable.
17
We conclude by noting that at low temperatures T ≪ T0 , we expect the stable
configuration of the system to be the fully polarized one with x ≃ xmax = N − 1,
corresponding to the maximal one-row symmetric representation with ℓ1 ≃ n boxes.
Indeed, the solution of equation (3.4) in that case can by well approximated by
x ≃ ( N − 1) 1 − Ne
− NT0 /T
,
T ≪ NT0 ,
(3.22)
manifesting a nonperturbative behavior in T around T = 0.
3.2 Phase transitions and metastability
We saw in the previous section that for T0 < T < Tc both the completely symmetric
representation and the singlet are locally stable. The globally stable configuration is
determined by comparing the free energies of the two solutions. The free energy per
atom was found in (2.32). For zero magnetic fields, it takes the form
N
F (x, T ) =
∑
i =1
Txi ln xi −
NT0 2
x
2 i
(3.23)
and for the single-row zero magnetic field solution the free energy per atom becomes
Fsym ( x, T ) =
T
a(1 + x ) ln(1 + x ) + (1 − ax ) ln(1 − ax )
1+a
a
− T0 x2 − T ln N .
2
(3.24)
Variation of this expression with respect to x leads to (3.4).3 For the singlet we have
Fsinglet ( T ) = − T ln N .
(3.25)
For a specific temperature T1 and magnetization x1 the singlet and symmetric configurations will have the same free energy. Equating the two expressions and using (3.4)
3 Positivity of the second derivative gives the stability condition (3.20), but without any restriction on
a, that is, the number M of equal rows in the YT. The reason is that this expression only captures stability
under variations of the length of the YT. Taking also into account perturbations into configurations with
additional rows recovers the general stability condition requiring M = 0 or M = 1.
18
we obtain
T1 ln
1 + x1
= (1 + a) T0 x1 ,
1 − ax1
(3.26)
T0
T1 ln(1 + x1 ) =
x1 (2 − ax1 ) ,
2
where we used (3.4) to simplify the expression for Fsym . This system is solved by4
T1 =
N ( N − 2)
T0
,
2 ( N − 1) ln( N − 1)
x1 = N − 2 .
(3.27)
In general
T0 < T1 < Tc .
(3.28)
except for N = 2 where we have T0 = T1 = Tc , implying that for the SU (2) case there
is a single critical temperature. For large N
T1 ≃ T0
N
≫ T0 ,
2 ln N
as
N ≫1.
(3.29)
Recalling the limiting behavior of Tc in (3.13), we note that for large N, T1 ≃ Tc /2. For
T0 < T < T1 , Fsinglet > Fsym , while for T1 < T < Tc , Fsinglet < Fsym . The situation is
summarized in table 1.
irrep
T < T0
Singlet
unstable
Symmetric
stable
T0 < T < T1
metastable
stable
T1 < T < Tc
stable
metastable
Tc < T
stable
not a solution
Table 1: Phases in various temperature ranges for N ⩾ 3 and their stability characterization.
Hence, at high enough temperature the only solution is the singlet with no magnetization. At temperature Tc a magnetized state corresponding to the symmetric irrep
(one-row) also emerges, and is metastable until some lower temperature T1 . Between
these two temperatures the singlet is the stable solution. Below T1 and down to T0
the roles of stable and unstable solutions are interchanged. Below T0 the only stable solution is the one-row symmetric representation. Hence, we have a spontaneous
symmetry breaking as
SU ( N ) → SU ( N − 1) × U (1) .
(3.30)
Note that the free energy changes discontinuously at T = T0 and at T = Tc . The plot
4 In
the next section we will present a method for solving it even in the presence of a magnetic field.
19
of the free energy is at Fig. 4. The low temperature plateaux in Fig. 4 for the one-row
configuration (blue curve) is explained by the fact that due to (3.22) we have
Fsym
T − NT0 /T
N−1
T0 1 + 2 e
,
≃−
2
T0
T ≪ NT0 .
(3.31)
F
0.5
-2.0
1.0
T
1.5
-2.5
Fsinglet
Fsym
-3.0
-3.5
Figure 4: Plot of the free energy F for N = 7, for the one-row solution (blue) up to Tc ≃ 1.72
and for the singlet one (red) from T = 1, in units of T0 . These cover the temperature range
in which we have stability or metastability. The crossover behavior is at T = T1 ≃ 1.63, in
agreement with table 1.
To better understand the situation, we may use the thermodynamic relations between
the free energy F, the internal energy U and the entropy S
F = U − TS ,
F
,
U = −T ∂T
T
2
S = −∂T F ,
(3.32)
for Fsym given by (3.24) to obtain
a
Usym ( x ) = − T0 x2 ,
2
1
a(1 + x ) ln(1 + x ) + (1 − ax ) ln(1 − ax ) .
Ssym ( x ) = ln N −
1+a
(3.33)
These are readily recognized as the coupling energy and the logarithm of the number of states per atom for representations close to the dominant symmetric one (we
emphasize that x = x ( T ) via (3.4)). For the singlet (x = 0) we simply have
Usinglet = 0 ,
Ssinglet = ln N ,
(3.34)
that is, the maximal energy and maximal entropy.
Even though the free energy is discontinuous at T = Tc , we may still think of the transitions at T0 and Tc as a first order phase transitions, in the sense that the discontinuity
of U implies that latent heat has to be transferred for the phase transition to occur. In
20
0.5
1.0
1.5
2.0
T
Usinglet
Ssinglet
-0.5
1.5
-1.0
Ssym
1.0
Usym
-1.5
-2.0
0.5
-2.5
-3.0
0.5
1.0
1.5
T
Figure 5: Plots of U (left) and S (right) for N = 7. For the one-row solution (blue) up to
Tc ≃ 1.72 and for the singlet one (red) from T = 1, in units of T0 . The sharp rise for T → Tc− is
according to (3.38).
detail we have
T0
x2 ,
2( N − 1) c
T → Tc− :
Usym = −
T → Tc+ :
Usinglet = 0 ,
Ssym = ln N +
xc
− ln(1 + xc ) ,
1 + xc
(3.35)
Ssinglet = ln N ,
where we simplified Ssym by using relations (3.8). When we transition from below to
above Tc , we must give energy to the system, which also increases its entropy (recall
that, in the absence of volume effects, dU = TdS). The behavior of the energy and the
entropy with temperature is depicted in fig. 5.
At the intermediate temperature T = T1 , given in (3.27), there is the possibility of a
first order phase transition from a metastable to a stable phase. Near that temperature
Fsym = − T ln N +
Fsinglet = − T ln N .
N−2
ln( N − 1) T − T1 + . . . ,
N
(3.36)
This transition is typical in statistical physics where latent heat transfer is involved.
Hence we have that
T → T1− :
T→
T1+
:
Usym = −
1 ( N − 2)2
T0 ,
2 N−1
Usinglet = 0 ,
Ssym = ln N −
N−2
ln( N − 1) ,
N
(3.37)
Ssinglet = ln N ,
The transition from metastable to stable configurations near the temperature T1 will
not occur spontaneously under ideal conditions, leading to hysteresis. Only when
the system is perturbed, or given an exponentially large time such that large thermal
fluctuations occur, will it transition from a metastable to a stable configuration. This
is reminiscent of the hysteresis in temperature exhibited in certain materials and in
21
supercooled water [20]. The discontinuity in U implies that (latent) heat has to be
transferred for the phase transition to occur. When we transit T1 from above the system releases energy, which also lowers its entropy, the opposite happening when it
transit above T1 (The term "pseudo phase transition" has been used in the literature
for this kind of process).
To understand the nature of the phase transitions that the system can undergo, imagine that we start at a temperature T > Tc with the (paramagnetic) singlet state and
adiabatically lower the temperature by bringing the system into contact with a cooling agent (reservoir). If the system is not perturbed, it will stay at the paramagnetic
state until T = T0 , where this state becomes unstable, the (ferromagnetic) symmetric representation solution takes over, and the system undergoes a phase transition
releasing latent heat into the reservoir. Similarly, starting at a temperature below T0
with the (ferromagnetic) symmetric irrep state and raising adiabatically the temperature, the system will remain in this state if it is not perturbed and will transition to
the singlet at T = Tc , absorbing latent heat from the reservoir and undergoing a phase
transition. Clearly the system presents hysteresis, and no unique Curie temperature
exists, since in the range T0 < T < Tc the two phases coexist.
The situation is somewhat different if the system is isolated (decoupled from the
reservoir) while it is in a metastable state (that is, a ferromagnetic state at T1 < T < Tc
or a paramagnetic state at T0 < T < T1 ). A transition to the stable state would involve
exchange of latent heat, which can only be provided by, or absorbed into, the stable
phase. However, the heat capacity of the paramagnetic phase is zero (see figure 5), so
no such exchange can take place and the system is "trapped" in the unstable phase.
This is unlike, say, supercooled water, where perturbations nucleate the formation of
a stable solid state, releasing latent heat into the unstable liquid phase and raising the
temperature until the liquid phase is eliminated or the temperature reaches the point
at which the two phases become equally stable. This feature of our system is somewhat unrealistic, since we have ignored all other degrees of freedom except SU ( N )
spins. A realistic system would also have vibrational degrees of freedom of the atoms,
which would serve as a reservoir absorbing or receiving latent heat and thus enabling
transitions from metastable to stable states.
The parameter x is a measure of the spontaneous magnetization and constitutes
the order parameter for the phase transition. Recalling (3.7), x is a measure of the
22
length of the single row YT corresponding to the solution. From (3.4) and (3.8) we can
deduce that x ( T ) near Tc behaves as5
x − xc ≃
s
( N − 1) x c
xc + 1 − N/2
s
Tc − T
.
T0
(3.38)
Hence, the deviation from xc near the critical temperature follows Bloch’s law for
spontaneous magnetization of materials with an exponent of 1/2 and an N-dependent
coefficient. The behavior of x ( T ) is depicted in fig. 6.
4.0
x(T)
3.5
3.0
2.5
0.6
0.8
1.0
1.2
1.4
T
Figure 6: Typical plot of x ( T ) (here for N = 5) for the symmetric irrep for 0 < T < Tc ≃ 1.40.
For T = 0 it reaches the maximum value x (0) = N − 1 and for T → Tc− it goes sharply to xc
according to (3.38).
The SU (2) case: The above results are valid for a ̸= 1 that is for N ⩾ 3. When a = 1,
as in the SU (2) case, T0 = T1 = Tc and the stable-metastable range in table 1 does not
exist. The transition between the singlet and the symmetric representations occurs at
the unique Curie temperature T = T0 and at xc = 0. In this case we have
Symm. irrep ( T < T0 ):
Singlet ( T > T0 ):
x≃
√
3
x=0,
s
2
T0 − T
3T0
T0 − T ,
, F ≃ − T ln 2 −
T0
4T0
F = − T ln 2 ,
(3.39)
which shows that at T = T0 there is a second order phase transition.
Note that the naïve limit a → 1 of the result (3.38) for a < 1 would not recover the
above behavior. The reason is that the range of validity of (3.38) is T0 < T ≲ Tc , and
it shrinks to zero as Tc → T0 . The two first-order phase transition points at T0 and Tc
fuse into a single second-order transition in the limit a → 1. Fig. 7 depicts the free
5 Positivity
of xc + 1 − N/2 follows from the fact that between the unstable and the stable solutions
the left hand side of (3.4) as a function of x reaches a minimum (see the right plot in fig. 2.
23
energy against the temperature for the SU (2) case.
0.2
0.4
0.6
0.8
1.0
1.2
1.4
T
-0.2
-0.4
Fsym
-0.6
-0.8
Fsinglet
-1.0
Figure 7: Plot of the free F for N = 2. The continuity of the expression and its first derivative
between the one-row solution (blue) up to T = 1 and the singlet one (red) from T = 1, in units
of T0 , is manifest.
The behavior of x ( T ) near T = T0 follows Bloch’s law with an exponent 1/2, as in the
general SU ( N ) case, but with a different coefficient. The internal energy and entropy
are continuous but their first derivatives at T = T0 are not, as depicted in fig. 8.
U
0.7
T
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Ssinglet
0.6
Usinglet
-0.1
Ssym
0.5
Usym
0.4
-0.2
0.3
-0.3
0.2
-0.4
0.1
-0.5
0.2
0.4
0.6
0.8
1.0
1.2
1.4
T
Figure 8: Plots of U (left) and S (right) for N = 2 for T in units of T0 , for the magnetized (blue)
state up to T0 and the singlet (red) state above T0 .
For comparison between SU (2) and higher groups, we present in table 2 the values of the critical temperature in units of T0 , the critical magnetization as a fraction of the maximal magnetization xmax = N − 1 and the coefficient of x − xc in
Bloch’s law in (3.38) again as a fraction of the maximal magnetization, i.e. K =
p
( N − 1) xc /( xc + 1 − N/2)/xmax for the first few values of N, as well as for N ≫ 1.
N
Tc /T0
xc /xmax
K
2
1
0
1.7320
3
1.0926
0.3772
1.2177
4
1.2427
0.5071
0.9862
5
1.4027
0.5749
0.8480
N≫1
N/ ln N
√1
2/N
Table 2: Behavior near critical point for N = 2, 3, 4, 5, as well as for N ≫ 1 (the corresponding
entries are leading).
24
4 Turning on magnetic fields
We now turn to incorporating nonzero magnetic fields in the system. We recall the
equilibrium equation (2.34)
T ln xi − NT0 xi = Bi + λ ,
i = 1, 2, . . . , N ,
(4.1)
where the xi sum to 1 due to the constraint in (2.31).
4.1 Small fields
For small magnetic fields the solution of the coupled equations (4.1) can be found as
a perturbation of the solution for vanishing fields. In fact, we can do better than that.
We may consider the response of the system in the presence of generic magnetic fields
Bi under small perturbations δBi . The state will change as
xi → xi + δxi ,
i = 1, 2, . . . , N ,
(4.2)
where δxi is a perturbation to the solution of (4.1). To linear order we have that
Since
T
− NT0 δxi = Ci−1 δxi = δBi + δλ .
xi
(4.3)
N
∑ δxi = 0 we obtain the change of λ as
i =1
δλ = −
∑iN=1 Cj δBj
∑iN=1 Cj
.
(4.4)
Then, combining with (4.3) we get
δxi = Ci δBi −
∑N
j=1 C j δB j
∑N
j =1 C j
!
,
(4.5)
from which the magnetizability matrix mij obtains as
mij =
Ci Cj
∂xi
= Ci δij − N
,
∂Bj
∑k=1 Ck
25
i, j = 1, 2, . . . , N .
(4.6)
Note that mij is symmetric and satisfies
N
∑ mij
= 0 as a consequence of the fact that
i =1
the U (1) part decouples.
We may infer the signs of mij for general Ci ’s from the stability condition on the configuration. Recall that, for stability, either all Ci are positive (including infinity), or
only one of them is negative, say C1 , and the rest positive (x1 is the largest among the
xi ’s). In the latter case, stability further requires that (3.15) be satisfied. Using these
we can show that
C1 > 0 :
m11 , mii > 0 ,
m1i < 0 ,
mij < 0 ,
C1 < 0 :
m11 , mii > 0 ,
m1i < 0 ,
mij > 0 ,
C1−1 = 0 :
m11 , mii > 0 ,
m1i < 0 ,
mij = 0 ,
(4.7)
where i, j = 2, 3, . . . , N and i ̸= j. According to (3.14), the last case above happens
for T = NT0 x1 , which is possible only for T < NT0 , and corresponds to one of the
solutions of (4.1) reaching the top of the function in the LHS of (3.1), depicted in fig. 1.
4.1.1
The singlet
For the unmagnetized singlet configuration, stable for temperatures T > T0 , xi = 1/N
and Ci−1 = N ( T − T0 ). The magnetizability is
singlet :
1
1
mij =
δ −
.
N ( T − T0 ) ij N
(4.8)
As expected, it diverges as ( T − T0 )−1 at the critical temperature T0 where the con-
figuration destabilizes, and the signs of its components are in agreement with (4.7).
This determines the linear response of the system to small magnetic fields, of typical
magnitude B such that B ≪ T − T0 .
4.1.2
The symmetric representation
For the spontaneously magnetized configuration corresponding to the symmetric representation M = 1, stable for T < Tc , the xi are as in (3.3), with x the solution of (3.4)
that corresponds to a stable configuration. The components of the magnetization ma-
26
trix take the form
m11 =
N−1
,
N∆( x )
1
, i ̸= 1,
N∆( x )
A( x )
1
δ −
,
mij =
N A(− ax ) ij ∆( x )
m1i = −
(4.9)
i, j ̸= 1 ,
where a = 1/( N − 1) and A( x ) = T/(1 + x ) − T0 as before, and
1
T
∆( x ) = A(− ax ) + A( x ) = N
− T0 .
a
(1 + x )(1 − ax )
(4.10)
As T → Tc− , (3.8) shows that the magnetization diverges. To compute its asymptotic
behavior at T ≃ Tc we use (3.38) to obtain
Tc
−
(1 + x )(1 − ax ) =
T0
q
2xc (2axc + a − 1)
which implies
T0
∆=N
Tc
q
2xc (2axc + a − 1)
s
s
Tc − T
+... ,
T0
Tc − T
+... .
T0
(4.11)
(4.12)
Hence the entries of the magnetizability matrix as T → Tc− become
m11 ≃
N−1
Q
√
,
2
N
Tc − T
m1i ≃ −
mij ≃
1
Q
√
,
2
N
Tc − T
(4.13)
1
Q
√
,
N 2 ( N − 1) Tc − T
where
Q= p
i, j = 2, . . . , N ,
Tc /T0
2xc (2axc + a − 1) T0
.
(4.14)
So the magnetizability diverges as ( Tc − T )−1/2 , and the signs of its components are
in agreement with (4.7). We have depicted these in fig. 9 below.
27
0.03
m11
2.0
m22
m11
0.02
1.5
m23
0.01
1.0
0.6
0.8
1.0
1.2
1.4
T
1.6
0.5
-0.01
m12
0.4
0.5
0.6
0.7
0.8
0.9
1.0
T
Figure 9: Left: Plots of the four different types of magnetization matrix entries for the symmetric irrep. up to T = Tc for N ⩾ 3. Right: Same plot for SU (2).
The SU (2) case: This is special since Tc = T0 . For the singlet solution (T > T0 ), (4.8)
remains valid for N = 2, giving
m11 = m22 = −m12 =
1 1
,
4 T − T0
T → T0+ .
(4.15)
For the magnetized solution
m11 = m22 = −m12 =
1
1 − x2
,
4 T − T0 (1 − x2 )
T < T0 .
(4.16)
Using (3.39) we obtain the magnetizability
m11 = m22 = −m12 ≃
1 1
,
8 T0 − T
T → T0− .
(4.17)
We note that mij diverges as | T − T0 |−1 on both sides of the critical temperature, un√
like for N ⩾ 2. The reason is that the coefficient of Tc − T0 in the expansion (4.11)
vanishes for N = 2 and thus the next order in the expansion, O( T0 − T ), becomes the
leading one. We have depicted these in fig. 9.
4.2 Finite fields
We now consider the state of the system, given by (4.1), for general non-vanishing
magnetic fields.
The general qualitative picture can be obtained by the same considerations as in
the case of Bi = 0. For fixed λ, each xi satisfies (4.1) for a different effective Lagrange
multiplier λi = λ + Bi and can take two possible values xi± , the two solutions of (4.1)
for fixed i. The stability considerations of section 3.1, which remain valid for arbitrary
magnetic fields, determine that at most one of these solutions can lie on the unstable
28
branch of the curve x+ . So, fully stable configurations correspond either to choosing
all xi− on the stable branch, or N − 1 of them on the stable branch and one on the
unstable branch. The stability condition
∑ Ci
< 0 in (3.15) must also be satisfied in
i
the latter case.
To gain intuition on the behavior of the system, we focus on the case when only one
of the magnetic fields, say B1 , is different from the rest. We can absorb the equal terms
Bi , i > 1 in the Lagrange multiplier and call B = B1 − Bi . Then the set of equations
(4.1) becomes just two distinct equations: one for x1 , with RHS λ + B, and one for the
remaining N − 1 xi ’s, with RHS λ, as depicted in fig. 10.
18
T ln x - N T0 x
λ+B
16
14
λ
12
10
8
6
5
10
15
20
x
T
. We consider a magnetic
NT0
field B along one direction, say for i = 1. The intersection with the lines λ and λ + B occur at
two values of xi on each side of x0 for each line.
Figure 10: Plot of (3.16), with its maximum occurring at x0 =
Referring to fig. 10, denote the four solutions of (4.1) by x± (at the intersections of the
B (at the intersections with λ + B). For the choice of B > 0 in
curve with λ), and by x±
B < x < x B < x . For B < 0 the ordering
the figure we have the ordering x− < x−
+
0
+
B < x < x < x < x B . Since we have a magnetic field only in
would change to x−
−
+
0
+
B while each x (i > 1) can take values x .
direction 1, x1 can take values x±
±
i
B . Thus, we
For a stable configuration we may choose at most one value at x+ or x+
have the following possible cases:
B , corresponding to a one-row YT (if x B > x )
• a) N − 1 values x− and one value x−
−
−
B < x ). This is the deformation of the singlet for B = 0.
or its conjugate (if x−
−
B , corresponding, again, to a one-row YT.
• b) N − 1 values x− and one value x+
This is the deformation of the one-row solution for B = 0.
B , and one value x , corresponding to a two-row
• c) N − 2 values x− , one value x−
+
YT. This is a deformation of the one-row YT solution at B = 0 by increasing its
29
depth and breaking the SU ( N ) symmetry further. Following a similar analysis,
these would correspond either to irreps with two rows, or to irreps with N − 1
rows, N − 2 of which have equal lengths. We will encounter such cases below.
Although we will not examine in detail the more general configuration of a magnetic
field with M equal components and the remaining N − M equal and distinct, its qual-
itative analysis is similar. Fig. 10 remains valid, but now with M values of xi at λ + B
and N − M at λ. Implementing the stability criterion we have the following cases (for
B > 0):
B
• N − M values x− and M values x−i , corresponding to a YT with M rows. This is
the deformation of the singlet for B = 0.
B
B
• N − M values x− , M − 1 values x−i , and one value x+i , corresponding to a YT
with M rows. This is the deformation of the one-row solution for B = 0.
B
• N − M − 1 values x− , M values x−i , and one value x+ , corresponding to a YT
with M + 1 rows. This is the deformation of the one-row solution for B = 0 by
increasing its depth and breaking the SU ( N ) symmetry further.
Overall, equality of magnetic field components results in states with equal rows in
their YT.
4.2.1
One-row and conjugate one-row states
We proceed to investigate quantitatively the effect of a magnetic field in one direction
(say, 1) in the case where the system is in a one-row solution in the same direction, or
the corresponding conjugate representation. That is, we will examine the cases (a) and
(b) above (case (c) will be examined in the next subsection). Then the equation for the
30
system becomes6
T ln
1+x
= T0 (1 + a) x + B ,
1 − ax
−1 < x <
1
= N−1.
a
(4.20)
Solutions to this equation with x > 0 correspond to a single row YT, whereas solutions
with x < 0 to its conjugate, that is, a YT with N − 1 rows of equal lengths. These are
related by observing that (4.20) is invariant under x → − x/a, B → − B, and a → 1/a,
which maps symmetric irreps to their conjugate.
The stability conditions for the solution are determined by the general discussion in
the previous subsection. As before, according to (3.18) C1−1 = N A( x ), with A( x )
defined in (3.19), and thus Ci−1 = N A(− ax ) for i > 1. By its definition, A( x ) satisfies
x>0:
A( x ) < A(− ax ) ,
x<0:
A( x ) > A(− ax ) .
(4.21)
The general stability argument requires Ci > 0 for i > 1, that is, A(− ax ) > 0. For
C1 < 0 we must also have ∑i Ci < 0, or A( x ) + aA(− ax ) > 0, as it was shown in
(3.15) and (3.20). Altogether, combined with (4.21), the stability conditions that cover
all cases are
x>0:
A( x ) + aA(− ax ) > 0 ,
(4.22)
which guarantees positivity of A(− ax ) no matter what the sign of A( x ), and
x<0:
A(− ax ) > 0 .
(4.23)
The condition (4.22) above will be satisfied for
x < x− or x > x+ ,
6 Note
x± =
N−2±
p
that for a = 1 (the SU (2) case) and after setting x =
y=
N 2 − 4( N − 1) T/T0
,
2
B
T
T0 y − 2T0
T
B
+ 0 tanh y ,
2T
T
(4.24)
equation (4.20) becomes
(4.18)
which is the standard expression in phenomenological investigations of ferromagnetism [21]. For a ̸= 1
2Ty− B
(the SU ( N ) case with N ⩾ 3) this is modified by setting x = (1+ a)T and results to the generalization
0
y=
B
T
1
+ 0
2T
T coth y −
31
N −2
N
.
(4.19)
while (4.23) will be satisfied for
T
x0 = ( N − 1) 1 −
T0
x0 < x < 0 ,
.
(4.25)
The existence of x± , and the condition that x0 > −1, introduce two more temperatures
T+ =
N2
T0 > T0 ,
4( N − 1)
T− =
N
T0 > T0 .
N−1
(4.26)
For T > T+ , (4.22) is satisfied for all x > 0, while for T > T− , (4.23) is satisfied for all
x < 0. Note that both x± ∈ (−1, N − 1), and that x+ > 0, while x− > 0 for T > T0
and x− < 0 for T < T0 . In terms of relative ordering of temperature scales,
N=3:
N=4:
N>4:
3T
9T0
< T− = 0 ,
8
2
4T0
,
T+ = T− =
3
T+ =
(4.27)
T+ > T− .
We proceed to the analysis of the states of the system. It is most convenient to keep x
and T as the free variables and consider the magnetic field B as a function of x with T
as a parameter. Then (4.20) implies
B( x ) = T ln
1+x
− T0 (1 + a) x .
1 − ax
(4.28)
Note that
dB
a(1 + a) T0
=
( x − x+ )( x − x− ) = T0 A( x ) + aA(− ax ) .
dx
(1 + x )(1 − ax )
(4.29)
Hence, dB/dx is proportional to the stability condition (4.22) for x > 0. Therefore,
B( x ) is an increasing function of x for T > T+ and T < T− , and a decreasing one for
x− < x < x+ , with x− and x+ as local maxima and minima. The function B( x ) is
plotted in figure (11) for various values of the temperature. The intersection of these
graphs with the horizontal at B determines the solutions for the configuration of the
system.
The above allow us to determine the stability of solutions for various values of the
temperature and magnetic field. We consider two cases, according to (4.22 and 4.23).
Positive x: The constraint (4.22) is relevant. Therefore, when T > T+ , all solutions with
32
B(x)
B(x)
B(x)
B
B
B
x
Figure 11: Plot of B( x ) for T > T+ (left), T0 < T < T+ (middle) and T < T0 (right).
x > 0 are stable. For T0 < T < T+ , stability singles out solutions with 0 < x < x− and
x > x+ . Finally, for T < T0 , since then x− < 0, stability requires that x > x+ .
Negative x: The constraint (4.23) is now relevant, or equivalently x > x0 . Since x < 0,
we conclude that no stable solutions exist for x0 > 0, or T < T0 . For x0 < −1,
or T > T− , all x < 0 solutions are stable. Finally, for intermediate temperatures
T0 < T < T− , we have stability for −1 < x0 < x < 0.
The above are tabulated in table 3 below (we assume that N > 4 so that T+ > T− ):
x
x<0
x>0
T < T0
none
x > x+
T0 < T < T−
−1 < x0 < x
x < x− & x > x+
T− < T < T+
all
x < x− & x > x+
T+ < T
all
all
Table 3: Stable solutions for various ranges of T and x and for N > 4.
We can now investigate the existence of stable solutions for the full range of values of the temperature and the magnetic field. The complete analysis is relegated to
appendix A. The results are summarized in the temperature-magnetic field phase diagram of figure 12, presented for a generic value for N > 4. The phase diagram is
qualitatively the same for N = 3 and N = 4, changing only for N = 2. The only
difference is that, for N = 4, T+ = T− , while for N = 3 T+ < T− . This does not affect
the general features of the diagram, simply shifting the critical point vertex ( T+ , B+ )
to the left of the bottom asymptote T = T− , for N = 3, or on top of it, for N = 4.
Each region in the T−B plane depicted in the figure corresponds to a discrete phase
of the system. Moving within these regions without crossing any of the critical lines
interpolates continuously between configurations. In the connected regions C1 , C2 ,
and C3 there is a unique one-row configuration at each point ( T, B). In regions A
and B inside the curvilinear triangle there are two locally stable configurations at each
point, one absolutely stable and the other metastable, with the line separating A and
33
B
B+
C2
0
T0
T
A
T1
Tc
T-
T+
N T0
B
(Tb ,Bb )
C3
C1
D
Figure 12: The phase diagram of the system for generic N > 4. The (orange) curve separating
regions A and C2 is B( x− ); the (blue) curve separating regions A, B, C1 and C3 , D is B( x+ ),
intersecting the T-axis at the B = 0 critical temperature T = Tc ; and the (red) curve separating
regions A, B, C3 and C1 , D is B( x0 ) and it asymptotes to the vertical T = T− . Crossing any of
these lines precipitates a discontinuous change in the magnetization, i.e. the order parameter
x. The (green) straight line separating regions A and B is the metastability frontier of the two
coexisting phases inside these regions (see appendix A.1); crossing it exchanges the metastable
and the absolutely stable states, and its intersection with the T-axis is the B = 0 critical temperature T = T1 . Regions C1,2,3 constitute one continuous phase with nonzero magnetization
(except at B = 0 and T > Tc ), all points being accessible through continuous paths in the B − T
space, while regions A, B and D are separated from C1,2,3 by discontinuous transitions in the
order parameter. The (gray) dashed curve from T0 to its vertical asymptote at T = NT0 separates "broken-like" and "unbroken-like" configurations but otherwise mark no sharp phase
transition. The shaded region D corresponds to a two-row (double magnetization) phase. For
N = 3 the phase diagram remains qualitatively the same with the order of T+ and T− interchanged, while for N = 4, we simply have T+ = T− .
B being the border of metastability where the two phases have equal free energy. In
region D there are no stable one-row solutions, signifying that a two-row solution
must exist there. The lines separating regions C1,2,3 and the other regions are phase
boundaries, the configuration changing discontinuously as we cross a boundary.
The dashed curve for T0 < T < NT0 represents configurations with C1 = ∞, that
is, A( x ) = 0. This corresponds to points where x1 = (1 + x )/N reaches the top of the
curve in fig. 10, transiting from the unstable to the stable branch of the curve or vice
34
versa. For such points, x = T/T0 − 1, and (4.28) gives B on this curve as
B( T ) = T ln
( N − 1) T N ( T − T0 )
−
.
NT0 − T
N−1
(4.30)
Configurations to the left of this curve are in a "broken-like" SU ( N ) phase, with one
of the solutions of (4.1) in the unstable branch of the curve in fig. 10, while those
to the right of the curve are in an "unbroken-like" phase, with all solutions on the
stable branch. For B = 0 these are the true spontaneously broken or unbroken phases
of the system. A nonzero magnetic field breaks SU ( N ) explicitly, and the dashed
line represents a soft phase boundary, which must be crossed to transit between the
two phases as we move on the T − B plane. The physical signature of crossing this
boundary is that the off-diagonal elements of the magnetizability mij , with i ̸= j ̸= 1,
change sign, vanishing on the boundary (see (4.7)).
The line separating regions B and C3 intersects the T axis at the critical temperature Tc
found in the B = 0 section. Points ( T0 , 0) and ( T+ , B+ ), with
N
B+ = B( x+ ( T+ )) =
2( N − 1)
N
ln( N − 1) − N + 2
2
.
(4.31)
are critical points, while point ( Tb , Bb ) at the lower tip of region B, satisfying the transcendental equation
B( x0 ( Tb )) = B( x+ ( Tb )) ≡ Bb
(4.32)
is a multiple critical point, connecting several different configurations: one one-row
state in C1 , two in B, and one in C3 , as well as a two-row state in D, and possible tworow states in the other neigboring regions (see next section). One of the states in B,
and possibly other one-row or two-row states, are metastable.
4.2.2
Two-row states and their (N − 1)-row conjugates
As we have seen, for T < T− and for sufficiently negative magnetic fields there is no
stable solution to (4.20), and thus no state corresponding to the one-row YT symmetric
representation. From the general analysis of subsection 4.2, we expect the solution to
be the only other allowed configuration, that is, a state corresponding to a two-row
YT. In this subsection we recover this solution and check its stability.
We consider a configuration with two (generally unequal) lengths x1 and x2 and an
35
applied magnetic field in the x1 -direction, representing the generic breaking pattern
SU ( N ) → SU ( N − 2) × U (1) × U (1) ,
(4.33)
of the SU ( N ) symmetry. This includes a spontaneous breaking of SU ( N ) in addition
to the dynamical breaking SU ( N ) → SU ( N − 1) × U (1) due to the magnetic field.
We note that in the special case x1 = x2 the symmetry breaking pattern would be
SU ( N ) → SU ( N − 2) × SU (2) × U (1), but as we shall demonstrate this pattern is
never realized in the present case of a magnetic field in a single direction.
The xi must satisfy the system of equations
x1
= NT0 ( x1 − x N ) + B ,
xN
x
T ln i = NT0 ( xi − x N ) , i = 2, 3, . . . , N − 1 ,
xN
T ln
(4.34)
where x N is determined from the constraint in (2.30). We write
x1 =
1+x
,
N
x2 =
1+y
,
N
1 − α( x + y)
x3 = · · · = x N =
,
N
1
α=
.
N−2
(4.35)
For y = −αx/(1 + α) = − ax = − x/( N − 1) this ansatz reduces to the one for the
one-row solution. (4.34) gives rise to the system of transcendental equations
T ln
1+x
= T0 αy + (1 + α) x + B ,
1 − α( x + y)
1+y
T ln
= T0 αx + (1 + α)y .
1 − α( x + y)
(4.36)
The free energy of the configuration is
T
T0
2
2
2
(1 + x ) ln(1 + x )
( x − y) − N ( x + y ) +
F ( x, y, T ) =
2N ( N − 2)
N
x+y
B
+ (1 + y) ln(1 + y) + ( N − 2 − x − y) ln 1 −
− x − T ln N .
N−2
N
(4.37)
The transcendental equations (4.36) will be solved numerically. The full analysis
of solutions and their stability is relegated to the Appendix. Here we simply state the
results and present relevant plots.
36
We consider temperatures T < T0 for which spontaneous magnetization exists. As
discussed earlier, for such temperatures and negative magnetic fields we expect the
state to be in a two-row stable state, which includes the posssility of "antirows," that
is, N − 2 equal rows plus an additional row. Further, such states may coexist with a
one-row state and be either globally stable or metastable.
All cases refer to plots in Fig. 13. The blue and orange curves represent the solutions
of the first and second equation in (4.36), resp. The orange line y = − ax, in particular,
solves the second equation in the system (4.36) while the first one reduces to the one
for the one-row configuration (4.20). Intersections of blue and orange lines represent
the solutions the (4.36). Only locally stable solutions are considered.
• B > 0: We recover the known one-row solution on the y = − ax line for x > 0.
There is also a metastable two-row solution with x < 0, y > 0.
6
B0
6
B>0
5
5
4
4
3
x+y
2
3
x+y
2
1
B0 < B < 0
5
4
1
α
6
3
α
2
1
y -a x
0
y -a x
0
1
2
3
4
5
-1
-1
6
y -a x
0
-1
-1
1
α
1
0
-1
x+y
1
0
1
2
3
4
5
-1
6
0
1
2
3
4
5
6
6
6
B(x+ ) < B < B0
B < B(x+ )
5
5
4
4
3
x+y
2
3
x+y
1
α
2
1
α
1
1
y -a x
0
y -a x
0
-1
-1
-1
0
1
2
3
4
5
-1
6
0
1
2
3
4
5
6
Figure 13: Typical contour plots for T < T0 in the x − y plane of the two eqs. in (4.36), in blue
(1st eq.) and orange (2nd eq., for which one of the branches is a straight line). The intersection
points of the blue curves with the orange curves represent solutions of (4.36). Stable and
metastable solutions are indicated by black and magenta colored dots, respectively. B0 is the
value of B for which the central bulges of the curves would touch (bet. 3rd and 4th plot).
Plots are for N = 7 and T = 0.9, and for B = 0.2, 0, −0.4, −2, −3 (note that B0 ≃ −1.01 and
B( x+ ) ≃ −2.62) in units of T0 .
37
• B = 0: The system is symmetric under x ↔ y and we recover the known onerow solution on the y = − ax line for x > 0. There is also a stable solution on the
y = − x/a line for x < 0, which is equivalent to the previous one, representing
spontaneous magnetization in direction x2 .
• B( x+ ) < B < 0: We recover the known one-row solution on the y = − ax line
for x > 0, but now it is metastable. There is a stable two-row solution with
x < 0, y > 0.
• B < B( x+ ): There are no stable one-row solutions, in accordance with region
D of the one-row phase diagram of fig. 12. There is a unique stable two-row
solution.
We note that there are no solutions with x = y since this is not a consistent truncation of the system (4.36), unless B = 0 in which case we already know that the
corresponding two-row solution is unstable. Thus the symmetry breaking pattern
SU ( N ) → SU ( N − 2) × SU (2) × U (1) is never realized.
Recalling fig. (12), the picture that emerges, at least for T < T0 , is that a tworow solution coexists with the one-row solution in both regions C1 , C2 . The two-row
solution is metastable in C2 (B > 0) and becomes absolutely stable in C1 (B < 0). The
one-row solution is absolutely stable in C2 , becomes metastable in C1 , and ceases to
exist in region D, leaving the two-row state as the only stable solution there. The line
B = 0 is a metastability frontier between one-row and two-row solutions for T < T0 .
We expect this picture to extend for a range of temperatures T > T0 , with a two-row
state coexisting with the one-row one outside of region D, although for high enough
temperatures the two-row solution should cease to exist.
5 Conclusions
The thermodynamic properties and phase structure of the SU ( N ) ferromagnet emerge
as surprisingly rich and nontrivial, manifesting qualitatively new features compared
to the standard SU (2) ferromagnet. The phase structure of the system, in particular,
is especially rich and displays various phase transitions. Specifically, at zero magnetic
field the system has three critical temperatures (vs. only one for SU (2)), one of them
signaling a crossover between two metastable states. Spontaneous breaking of the
38
global SU ( N ) group in the ferromagnetic phase at zero external magnetic fields happens only in the SU ( N ) → SU ( N − 1) × U (1) channel. In the presence of a nonabelian
magnetic field with M nontrivial components (M < N), the explicit symmetry break-
ing (paramagnetic state) is SU ( N ) → SU ( N − M ) × U (1) M , while the spontaneous
breaking (ferromagnetic state) is SU ( N ) → SU ( N − M − 1) × U (1) M+1 . Finally, due
to the presence of metastable states, the system exhibits hysteresis phenomena both in
the magnetic field and in the temperature.
The model studied in this work, and its various generalizations described below,
could be relevant in a variety of physical situations. It could serve as a phenomenological model for physical ferromagnets, in which the interaction between atoms is
not purely of the dipole type and additional states participate in the dynamics. In
such cases, the SU ( N ) interactions could appear as perturbations on top of the SU (2)
dipole interactions, leading to modified thermodynamics. The model could also be
relevant to the physics of the quark-gluon plasma [22], which can be described as a
fluid of particles carrying SU (3) degrees of freedom, assuming their SU (3) states interact. Exotic applications, such as matrix models and brane models in string theory,
can also be envisaged (see, e.g. [23, 24]).
Various possible generalizations of the model, relevant to or motivated by potential applications, and related directions for further investigation suggest themselves.
They can be organized along various distinct themes: starting with atoms carrying a
higher representation of SU ( N ), generalizing the form of the two-atom interaction, or
including three- and higher-atom interactions.
The choice of fundamental representations for each atom was imposed by the
physical requirement of invariance of their interaction under common change of basis for the atom states. Its effect on the thermodynamics is to "bias" the properties
towards states with a large fundamental content. This manifests, e.g., in the qualitatively different properties of the system under positive and negative magnetic fields
(with respect to the system’s spontaneous magnetization). Starting with atoms carrying a higher irrep of SU ( N ) would modify these properties. In particular, starting
with atoms in the adjoint of SU ( N ) would eliminate this bias altogether. It might also
eliminate phases of spontaneous magnetization, and this is worth investigating.
The interaction of atoms jr,a js,a was isotropic in the group indices a, an implication
of the requirement of invariance under change of basis. Anisotropic generalizations
39
of the form (2.2) can also be considered, involving an "inertia tensor" h ab in the group.
Clearly this generalization contains the higher representation generalizations of the
previous paragraph as special cases. E.g., SU (2) interactions with the atoms in spin-1
states can be equivalently written as SU (3) fundamental atoms with a tensor h ab equal
to δab when a, b are in the SU (2) subgroup of SU (3) that admits the fundamental of
SU (3) as a spin-1 irrep, and zero otherwise. The more interesting special case in which
h ab deviates from δab only along the directions of the Cartan generators, in the presence
of magnetic fields along these directions, seems to be the most motivated and most
tractable, and is worth exploring. The phase properties of the model under generic
h ab is also an interesting issue.
Including higher than two-body interactions between the atoms is another avenue
for generalizations. Physically, such terms would arise from higher orders in the perturbation expansion of atom interactions, and would thus be of subleading magnitude, but the possibility to include them is present. Insisting on invariance under
common change of basis and a mean-field approximation would imply that such interactions appear as higher Casimirs of the global SU ( N ) and/or as higher powers
of Casimirs, the most general interaction being a general function f (C (2) , · · · , C ( N −1) )
of the full set of Casimirs of the global SU ( N ). These can be readily examined using
the formulation in this paper and may lead to models with richer phase structure. An
interesting extension of this study is in the context of topological phases nonabelian
models. Such topological phases have been proposed in one dimension [25–28] and it
would be interesting to see if they exist in higher dimensions.
Another independent direction of investigation is the large-N limit of the model.
This could be conceivably relevant to condensed matter situations involving interacting Bose condensates, or to more exotic situations in string theory and quantum
gravity. The presence of two large parameters, n and N, presents the possibility of
different scaling limits. These will be explored in an upcoming publication.
Finally, the nontrivial and novel features of this system offer a wide arena for experimental verification and suggest a rich set of possible experiments. The experimental realization of this model, or the demonstration of its relevance to existing systems,
remain as the most interesting and physically relevant open issues.
40
Acknowledgements
We would like to thank I. Bars for a very useful correspondence, D. Zoakos for help
with numerics, and the anonymous Reviewer for comments and suggestions that
helped improve the manuscript.
A.P. would like to thank the Physics Department of the U. of Athens for hospitality
during the initial stages of this work. His visit was financially supported by a Greek
Diaspora Fellowship Program (GDFP) Fellowship.
The research of A.P. was supported in part by the National Science Foundation under
grant NSF-PHY-2112729 and by PSC-CUNY grants 65109-00 53 and 6D136-00 02.
The research of K.S. was supported by the Hellenic Foundation for Research and Innovation (H.F.R.I.) under the “First Call for H.F.R.I. Research Projects to support Faculty members and Researchers and the procurement of high-cost research equipment
grant” (MIS 1857, Project Number: 16519).
A
Analysis of one-row states with a magnetic field
In this appendix we present the details of the analysis that we have summarized in
the main text.
T > T+ (fig. 14): B( x ) is increasing and all values of x are stable, so there exsits a
unique stable solution for all B, one-row for B > 0 and its conjugate for B < 0.
B(x)
6
4
B
2
-1
1
2
3
4
5
6
x
-2
Figure 14: Plot of B( x ) for T > T+ .
T− < T < T+ (fig. 15): We have stability for −1 < x < x− and x+ < x < N − 1,
the regions of increasing B( x ). Further, note that x− > 0 and B( x− ) > B( x+ ), and
that B( x− ) > 0 while B( x+ ) can be positive or negative. The critical temperature Tc
satisfies the condition
T = Tc
⇐⇒
41
B( x+ ) = 0 ,
(A.1)
which is precisely the condition (3.8). Note that, for N > 4, T− < Tc < T+ . For T > Tc ,
B( x+ ) > 0 and for T < Tc , B( x+ ) < 0.
B(x)
B(x)
1.5
1.0
1.0
B
B
0.5
0.5
-1
1
2
3
4
5
6
x
-0.5
-1
1
2
3
4
5
6
x
-1.0
-1.5
Figure 15: Plot of B( x ) for Tc < T < T+ (Left) and for T− < T < Tc (Right).
So within this temperature range we distinguish two sub-cases:
Tc < T < T+ (left plot in fig. 15): we have for various values of the magnetic field:
• For B < 0, x varies from −1 to 0 and there is a unique stable solution corresponding to the conjugate one-row YT.
• For 0 < B < B( x+ ), x varies from 0 to some x < x− and there is a unique stable
solution corresponding to a one-row YT.
• For B( x+ ) < B < B( x− ) we have two locally stable solutions, one for some
0 < x < x− and one for some x > x+ (a third solution in between is unstable).
The first one corresponds to an unbroken phase, as it represents a continuous
deformation of the singlet for B = 0, and the second one to a broken phase.
One is absolutely stable and the other metastable. To decide which, we need to
compare their free energies.
• For B > B( x− ), x varies from some value greater than x+ to N − 1 and there is a
unique stable solution.
Note that for B = 0 there is only the solution x = 0, as expected.
T− < T < Tc (right plot in fig. 15): we have for various values of the magnetic field:
• For B < B( x+ ), x varies from −1 to some negative value x obtained from
B( x+ ) = B( x ) and there is a unique stable solution corresponding to the con-
jugate one-row YT.
42
• For B( x+ ) < B < B( x− ) we have two locally stable solutions, one for some
x < x− and one for some x > x+ . The first one represents an unbroken phase
and the second one a broken phase, as they map to the singlet and the one-row
solutions for B = 0. One is absolutely stable and the other metastable. To decide,
we need to compare their free energies.
• For B > B( x− ), x varies from some x > x+ to N − 1 and there is a unique stable
solution corresponding to a one-row YT.
Note that B = 0 is in the range B( x+ ) < B < B( x− ) and we recover the expected
two solutions, x = 0 (singlet) and x > 0 (one-row).
T0 < T < T− (fig. 16): The situation is as in case T− < T < Tc , except now x cannot
be more negative than x0 = ( N − 1)(1 − T/T0 ) defined in (4.25). The relative size of
B( x0 ) and B( x+ ) will also play a role:
• For B < min{ B( x0 ), B( x+ )} there is no stable one-row solution and the stable
solution must necessarily have more rows. According to the general stability
analysis, it must be one with N − 1 rows out of which N − 2 have equal length.
• For min{ B( x0 ), B( x+ )} < B < max{ B( x0 ), B( x+ )} there is one stable solution,
for x < 0 (x > x+ ) if B( x0 ) < B( x+ ) (B( x0 ) > B( x+ )).
• For max{ B( x0 ), B( x+ )} < B < B( x− ) there are two stable solutions, one abso-
lutely stable and the other metastable. To decide, we need to compare their free
energy.
• For B > B( x− ), x there is a single stable solution varying from some x > x+ to
N − 1.
T < T0 : Only x > x+ solutions are stable.
• For B < B( x+ ) there is no stable one-row solution, and the solution must again
be one with ( N − 1)-rows.
• For B > B( x+ ) there is one stable solution for x > x+
43
2
B(x)
1
B
-1
1
2
3
4
5
6
x
B(x0 )
-1
-2
Figure 16: Plot of B( x ) for T0 < T < T− . The dashed lines refer to the value B( x0 ) which
could be higher (green) or lower (red) than B( x+ ).
A.1
Resolving metastability
To determine which configuration is absolutely stable and which is metastable when
there are two locally stable solutions, we need to compare their free energies. The free
energy of the system is given by (3.24) with the addition of the magnetic field term,
Fsym ( x, T ) =
T
a(1 + x ) ln(1 + x ) + (1 − ax ) ln(1 − ax )
1+a
B( x )
a
x − T ln N ,
− T0 x2 −
2
N
(A.2)
where a = 1/( N − 1) and where B = B( x ) is expressed in terms of x, T via (4.28).
To facilitate the comparison, define the modified free energy Φ (we use F ( x ) instead
of Fsym ( x, T ) for notational convenience)
Φ( x ) = F ( x ) +
N−2
B( x ) .
2N
(A.3)
We can show that Φ( x ) and B( x ) satisfy
Φ( N − 2 − x ) = Φ( x )
and
B( x ) + B( N − 2 − x ) = 2T ln( N − 1) − T0
(A.4)
N ( N − 2)
.
N−1
(A.5)
For two solutions with different x, x ′ but the same B, F ( x ) − F ( x ′ ) = Φ( x ) − Φ( x ′ ), so
we can compare their Φ to resolve metastability. At the transition point, when the two
44
solutions have the same free energy, (A.4) implies
Φ( x ) = Φ( x′ )
=⇒
x′ = N − 2 − x
(A.6)
and from (A.5) with B( x ) = B( x ′ ) = B( N − 2 − x ) the magnetic field Bt at which this
happens is
Bt = T ln( N − 1) −
T0 N ( N − 2)
.
2 N−1
(A.7)
For fixed B, this gives the transition temperature Tt at which the two solutions will
transit from stable to metastable as
1
Tt =
ln( N − 1)
N ( N − 2)
B + T0
2( N − 1)
.
(A.8)
For B = 0 this reproduces the temperature T1 and magnetization x1 that we determined before in (3.27).
B Analysis of two-row states in a magnetic field
In this appendix we investigate in detail two-row solutions, including their ( N − 1)-
row conjugates. We analyze the conditions for their stability, present the corresponding YT of their irreps, and derive numerical results for the case of temperatures T < T0 .
To proceed, we write the coefficients Ci defined in (3.14) in terms of the variables x
and y of (4.35). We obtain
C1−1 = N A( x ) ,
Ci−1
C2−1 = N A(y) ,
= N A − α( x + y) ,
(B.1)
i = 3, 4, . . . , N ,
with our usual A( x ) defined in (3.19). The variables x and y are restricted by the
conditions 0 < xi < 1 to the range
x, y > −1 ,
x+y <
1
= N−2.
α
(B.2)
Thus, the allowed solutions are within the triangle in the ( x, y)-plane with corners
at the points (1/a, −1), (−1, 1/a) and (−1, −1), depicted in fig. 17. This triangle is
further subdivided by the curves y = x, y = − ax and y = − x/a into six regions repre-
senting the possible ordering of x1 , x2 , xi (i ⩾ 3) and thus the various YT renditions of
45
the two-row solution. These regions are accordingly labelled by (ijk) for xi > x j > xk .
Assuming N > 3, most of the Ci ’s are proportional to 1/A(−α( x + y)), so we choose
solutions with A(−α( x + y)) > 0. Then at most one of the functions A( x ) and A(y)
can be negative. We list below the various possibilities together with conditions for
stability:
y
(−1, 1/a)
y=-x/a
231
213
x+y=1/α
y=x
123
321
x
y=-ax
132
312
(−1, −1)
(1/a, −1)
Figure 17: The domain of x, y. Coordinate axes do not create subdivisions.
Region A, or (123): x1 > x2 > x3 , with
− ax < y < x
=⇒
A( x ) < A(y) < A(−α( x + y)) .
(B.3)
The stability condition is
A(y) > 0 ,
α[ A( x ) + A(y)] A(−α( x + y)) + A( x ) A(y) > 0 .
(B.4)
In this region all Ci ’s are positive, excect C1 which may have either sign. We define
ℓ1 =
(1 + α) x + αy
,
N
ℓ2 =
(1 + α)y + αx
,
N
ℓ1 > ℓ2 > 0 .
(B.5)
The YT has the partition
(ℓ1 , ℓ2 ) .
(B.6)
Hence, it represents a two-row YT with ℓ1 and ℓ2 boxes, respectively. This follows
from the fact that x3 is the smallest among the three xi ’s and appears N − 2 times.
46
Region B, or (132): x1 > x3 > x2 , with
− x/a < y < − ax ,
A( x ) < A(−α( x + y)) < A(y) .
(B.7)
The stability condition is
A(−α( x + y) > 0 ,
α[ A( x ) + A(y)] A(−α( x + y)) + A( x ) A(y) > 0 .
(B.8)
In this region all Ci ’s are positive expect C1 which may have either sign. We define
ℓ1 =
x−y
,
N
ℓ2 = −
(1 + α)y + αx
,
N
ℓ1 > ℓ2 > 0 .
(B.9)
The ( N − 1)-row YT has the partition
(ℓ1 , ℓ2 , ℓ2 , . . . , ℓ2 ) .
|
{z
}
(B.10)
N −2
Hence, it represents a YT with ℓ1 boxes in the first line and ℓ2 boxes in the following N − 2 lines. This follows from the fact that the smallest among the three xi ’s, x2
appears only once.
Region C, or (213): x2 > x1 > x3 , with
− ay < x < y ,
A(y) < A( x ) < A(−α( x + y)) .
(B.11)
The stability condition is
A( x ) > 0 ,
α[ A( x ) + A(y)] A(−α( x + y)) + A( x ) A(y) > 0 .
(B.12)
In this region all Ci ’s are positive expect C2 which could be either positive or negative.
Defining
ℓ1 =
(1 + α)y + αx
,
N
ℓ2 =
(1 + α) x + αy
,
N
ℓ1 > ℓ2 > 0 ,
(B.13)
this case represents a two-row YT with the partition (B.6).
Region D, or (231): x2 > x3 > x1 , with
− ax < y < − x/a ,
A(y) < A(−α( x + y)) < A( x ) .
47
(B.14)
The stability condition is
A(−α( x + y) > 0 ,
α[ A( x ) + A(y)] A(−α( x + y)) + A( x ) A(y) > 0 .
(B.15)
In this region all Ci ’s are positive expect C2 which could be either positive or negative.
Defining
ℓ1 =
y−x
,
N
ℓ2 = −
(1 + α) x + αy
,
N
ℓ1 > ℓ2 > 0 .
(B.16)
this case represents a YT with the partition (B.10).
Region E, or (312):
with
x3 > x1 > x2 , with y < x < − ay, equivalently x3 > x1 > x2 ,
y < x < − ay ,
A(−α( x + y)) < A( x ) < A(y) .
(B.17)
The stability condition is
A(−α( x + y)) > 0 .
(B.18)
In these two regions all Ci ’s are positive. Defining
ℓ1 = −
(1 + α)y + αx
,
N
ℓ2 =
x−y
,
N
ℓ1 > ℓ2 > 0 ,
(B.19)
this case represents a ( N − 1)-row YT with the partition
(ℓ1 , ℓ1 , . . . , ℓ1 , ℓ2 ) ,
{z
}
|
(B.20)
N −2
Region F, or (321): x3 > x2 > x1 , with
x < y < − ax
A(−α( x + y)) < A(y) < A( x ) .
(B.21)
The stability condition is
A(−α( x + y)) > 0 .
(B.22)
In these two regions all Ci ’s are positive. Defining
ℓ1 = −
(1 + α) x + αy
,
N
ℓ2 =
y−x
,
N
ℓ1 > ℓ2 > 0 .
this case represents a ( N − 1)-row YT with the partition (B.20).
48
(B.23)
B.1 The case of low temperatures
We consider temperatures T < T0 for which spontaneous magnetization exists. Various cases arise depending on the sign of B and, if negative, on the value B( x+ ) < 0
with x+ defined in (4.24) (recall that B( x+ ) is negative for T < T0 ) and some other
intermediate value B0 to be defined shortly. In the plots below all points on the red
line y = − ax solve identically the second equation in the system (4.36), whereas the
first one reduces to the equation for the one-row configuration (4.20). This line can be
approached for x > 0 from the regions A and B and for x < 0 from the regions D and
F. In addition, the stability conditions for these regions reduce to those in (4.22). All
cases below map to one of the plots in Fig. 13.
• B > 0: We know that the one-row configuration has a stable solution which in
this two-parameter plot is on the y = − ax red line for x > 0. The intersection
point in the middle (region C in Fig. 17) is unstable, whereas that on the upper
left corner is, having a higher value for the free energy, metastable (region C).
• B = 0: We have included the case with B = 0, which is symmetric with respect
to the y = x line. We know that the one-row configuration has a stable solution
which in this two-parameter plot is on the y = − ax red line for x > 0. Due to
the above symmetry there is stable solution also on the y = − x/a line for x < 0,
which however is equivalent to the above. The other three intersecting points
are unstable.
• B0 < B < 0: The value of the magnetic field B0 arises when the two curves in the
plot meet tangentially. The one-row configuration has a stable solution which
in this two-parameter plot is on the y = − ax red line for x > 0. However,
this becomes now metastable, as the stable intersection point is on the upper left
corner (region D in fig. 17), corresponding to an ( N − 1)-row YT with partition
(B.10). The other intersection points are unstable.
• B( x+ ) < B < B0 : There are typically three intersection points along the y = − ax
line corresponding to the one-row configuration, the far right is now metastable
and the other two unstable. However, there are two additional intersecting
points in the far left of that plot, the lower one being unstable and the upper
one stable (region D in Fig. 17) corresponding to an ( N − 1)-row YT with parti-
tion (B.10).
49
• B < B( x+ ): There is one intersection points along the y = − ax line corresponding to the one-row configuration. This has x < 0 and is, as we have shown un-
stable. In fact, there is no stable one-row configurations for B < B( x+ ). Among
the other two intersection points the lower one is unstable and the upper one is
stable (region D in Fig. 17) and again it corresponds to a ( N − 1)-row YT with
partition (B.10).
References
[1] I.I. Lungu, A.M. Grumezescu and C. Fleaca, Unexpected Ferromagnetism − A Review, Appl. Sci. 2021, 11(15), 6707.
[2] A.P. Polychronakos and K. Sfetsos, Composing arbitrarily many SU ( N ) fundamentals, Nucl. Phys. B 994 (2023), 116314, 2305.19345 [hep-th].
[3] M.A. Cazalilla, A.F. Ho and M. Ueda, Ultracold gases of ytterbium: ferromagnetism
and Mott states in an SU (6) Fermi system, 2009 New J. Phys. 11 103033.
[4] A.V. Gorshkov et al., Two-orbital SU(N) magnetism with ultracold alkaline-earth
atoms, Nature Physics 6 (2010) 289-295.
[5] X. Zhang et al., Spectroscopic observation of SU ( N )-symmetric interactions in Sr orbital magnetism, Science Vol 345, Issue 6203 (2014) 1467.
[6] M. A. Cazalilla and A.M. Rey, Ultracold Fermi gases with emergent SU ( N ) symmetry, 2014 Rep. Prog. Phys. 77 124401.
[7] S. Capponi, P. Lecheminant, and K. Totsuka, Phases of one-dimensional SU(N) cold
atomic Fermi gases - From molecular Luttinger liquids to topological phases,
Ann. Phys. 367 (2016) 50-95.
[8] H. Katsura and A. Tanaka, Nagaoka states in the SU (n) Hubbard model,
Phys. Rev. A87, 013617 (2013).
[9] E. Bobrow, K. Stubis and Y. Li, Exact results on itinerant ferromagnetism and the
15-puzzle problem, Phys. Rev. B98, 180101(R) (2018)
[10] C. Romen and A.M. Läuchli, Structure of spin correlations in high-temperature
SU ( N ) quantum magnets, Phys. Rev. Research 2 (2020) 043009.
50
[11] D. Yamamoto, C. Suzuki, G. Marmorini, S. Okazaki and N. Furukawa, Quantum and Thermal Phase Transitions of the Triangular SU (3) Heisenberg Model under
Magnetic Fields, Phys. Rev. Lett. 125, 057204 (2020).
[12] K. Tamura and H. Katsura, Ferromagnetism in d-Dimensional SU (n) Hubbard Models with Nearly Flat Bands, Journal of Stat. Phys. volume 182, 16 (2021).
[13] K. Totsuka, Ferromagnetism in the SU ( N ) Kondo lattice model: SU ( N ) double exchange and supersymmetry Phys. Rev. A107, 033317 (2023).
[14] K. Tamura, H. Katsura Flat-band ferromagnetism in the SU(N) Hubbard and Kondo
lattice models 2303.15820 [cond-mat].
[15] D. Yamamoto et al., Quantum and Thermal Phase Transitions of the Triangular SU (3)
Heisenberg Model under Magnetic Fields, Phys. Rev. Lett. 125 (2020)057204
[16] Y. Miyazaki et al., Linear Flavor-Wave Analysis of SU (4)-Symmetric Tetramer Model
with Population Imbalance, J. Phys. Soc. Jpn. 91 (2022) 073702
[17] H. Motegi et al., Thermal Ising transition in two-dimensional SU (3) Fermi lattice gases
with population imbalance, 2209.05919 [cond-mat].
[18] A.P. Polychronakos and K. Sfetsos, Composition of many spins, random walks and
statistics, Nucl. Phys. B913 (2016), 664-693, 1609.04702 [hep-th].
[19] T. Curtright, T.S. Van Kortryk and C.K. Zachos, Spin Multiplicities,
Phys. Lett. A381 (2017), 422-427, 1607.05849 [hep-th].
[20] V. Holten, C.E. Bertrand, M.A. Anisimov, and J.V. Sengers, Thermodynamics of
supercooled water, J. Chem. Phys. 136, 094507 (2012), 1111.5587 [physics].
[21] See, e.g., Feynman lectures (ch. 36).
[22] R. Pasechnik and Michal Šumbera, Phenomenological Review on Quark-Gluon
Plasma: Concepts vs. Observations, Universe 3 (2017) 7, 1611.01533 [hep-ph].
[23] See, e.g., E. Martinec, Matrix Models and 2D String Theory, Application of random
matrices in physics, NATO Advanced Study Institute, Les Houches, France, June
6-25, 2004, 403-457, hep-th/0410136.
51
[24] G.J. Turiaci, M. Usatyuk, and W.W. Weng, Dilaton-gravity, deformations of the
minimal string, and matrix models,Classical and Quantum Gravity 38 (20) 2021,
2011.06038 [hep-ph].
[25] A. Roy and T. Quella Chiral Haldane phases of SU ( N ) quantum spin chains,
Phys. Rev. B 97 (2018) 155148.
[26] S. Capponi, P. Fromholz, P. Lecheminant and K. Totsuka, Symmetry-protected topological phases in a two-leg SU(N) spin ladder with unequal spins,
Phys. Rev. B 101 (2020) 195121.
[27] K. Totsuka, P. Lecheminant, and S. Capponi, Semiclassical approach to competing
orders in a two-leg spin ladder with ring exchange, Phys. Rev. B 86 (2012) 014435.
[28] K. Remund, R. Pohle, Y. Akagi, J. Romhanyi and N. Shannon, Semi-classical simulation of spin-1 magnets, Phys. Rev. Res. 4 (2022) 033106.
52