Mathematical Theories of Liquid Crystals

II. The Maier-Saupe Theory

Epifanio G. Virga
Mathematical Institute
University of Oxford
[email protected]
on leave from
Department of Mathematics
University of Pavia, Italy
Summary
Introduction
Helmholtz Free Energy
Soft and Hard Interactions
Mean-Field Free Energy
Density Functional Theory
Maier-Saupe Free Energy

1
 Introduction

The Maier-Saupe theory of liquid crystals is perhaps the most success-

ful microscopic theory proposed so far to explain the condensation
of the nematic phase. Maier & Saupe (1958)

In essence, it rests on three conceptual pillars:

• it is a mean-field theory;
• it considers only attractive interactions between molecules;
• it applies only to homogeneous phases.

2
 Helmholtz Free Energy
We consider a dynamical system with N identical particles in the
Hamiltonian formalism.
Lagrangian

L(q, q̇) := K(q, q̇) − U (q)

q = (q1 , . . . , qN ) ∈ QN generalized coordinates
Q single particle configuration space
q̇ = (q̇1 , . . . , q̇N ) generalized velocities
K kinetic energy
U potential energy
Hamiltonian

H(q, p) := K + U
pj := ∂∂L
q̇j conjugate momenta

3
 equations of motion

∂H
q̇ =
∂p
∂H
ṗ = −
∂q
QN × RN ⊂ R2N phase space
canonical ensemble

Imagine a great many replicas of the system described by the same

Hamiltonian H, but differing in the occupation of the phase space, as a
result, for example, of different initial conditions. Different ensembles
differ in their probability density ̺. Gibbs(1902)

4
In the canonical ensemble,
1 − 1 H(q,p)
̺0 (q, p) :=e kT
Z
k Boltzmann constant
T absolute temperature
partition function
Z Z
1
Z := dp dq e− kT H(q,p)
RN QN
Helmholtz free energy

F := −kT ln Z.
The number of particles N , the volume V they occupy, and the tem-
perature T hidden in F are parameters. Equilibrium thermodynamics
results from differentiating F with respect to them. For example,
∂F
P = − ∂V is the pressure
5
 Amended configurational partition function

We reduce Z to a purely configurational quantity, by integrating out

the momenta components of H. For a canonical H, purely quadratic
in p, this amounts to extracting a factor (depending on T ) out of Z,
which in turn only affects F by an inessential additive constant.
Z
1 1
− kT U (q)
ZN := e dq FN := −kT ln ZN
N ! QN

A classical justification for introducing the correcting factor N1 ! is to

account for the indistinguishability of the particles comprising the sys-
tem.
It has long been known (apparently, to Gibbs himself) that failing to
introduce this correcting factor would make the theory vulnerable to
Gibbs’ paradox for the entropy of mixing ideal gases.

6
 Pairwise Interactions

We assume that the N particles only interact in pairs

N
1 X
U (q) = U (qi , qj )
2
i6=j=1

U :Q×Q→R
U (qi , qj ) = U (qj , qi )

In general, U (q1 , q2 ) comprises the whole energy involved in the in-

teraction between particles q1 and q2 . It includes both slowly varying,
long-range potentials, typically responsible for the attraction between
particles, and rapidly varying, short-range potentials, typically respon-
sible for the repulsion between particles, often steric in nature. We
call soft the former component of U and hard the latter.

7
 typical potential
5

/
UL J ε
4

1
0 0.5 1 rm/σ 1.5 2
0 r/σ0
8
 Soft and Hard Interactions
U can be decomposed into the sum of its attractive (soft) and repul-
sive (hard) components:

U (q1 , q2 ) = U (A) (q1 , q2 ) + U (R) (q1 , q2 )

Though, to some extent, such a decomposition is arbitrary, the dis-

tinctive feature of U (R) is its abrupt divergence when the interacting
particles tend to come in contact with one another.

U (R) is often assumed to be arbitrarily close to zero when the interact-

ing particles are close to one another, and to diverge to +∞ as they
touch.

The divergence of U (R) makes it questionable taking averages of

repulsive interactions.

9
 only attractions
An assumption of the Maier-Saupe theory is to ignore repulsive in-
teractions:
U (R) ≡ 0
computing the free energy
Z  1 1 PN (A) 
1 − 2 kT U (q ,q )
FN = − kT ln e j=2 1 j
dq1
N ! QN
× ···
 1 1 PN −1 (A)

× e− 2 kT j=1 U (qN ,qj )
dqN
Z N
1 Y
= − kT ln Ih dq1 dq2 . . . dqN
N! QN h=1
PN
− 12 1
U (A) (qh ,qj )
Ih (q1 , . . . , qN ) := e kT j=1,j6=h

10
 Mean-Field Free Energy

Mathematically, a mean-field theory is an approximation that re-

places complicated expressions, difficult to evaluate, which involve
many individual particles and their mutual interactions, with sim-
pler expressions, easier to evaluate, which involve a single, effective
particle.

Physically, it is as if the actions exchanged by particles were replaced

by an effective field that all particles produce and feel at the same
time.

≈ here means “replaced by”

11
 replacements

N
1 X
U (A) (qh , qj ) ≈ E(qh )
2
j=1,j6=h
1
− kT E(qh )
Ih (q1 , . . . , qN ) ≈ I(qh ) := e

E single-particle effective potential

mean-field free energy

Z N
1 Y
FN = −kT ln Ih dq1 dq2 . . . dqN
N! QN h=1
Z N
1 1
− kT E(q) (mf)
≈ −kT ln e dq =: FN
N! Q

12
 upper estimate

It follows from Gibbs-Bogoliubov inequality (and an equivalent ap-

proach to mean-field theory) that the mean-field free energy is larger
than the true free energy it replaces.
Gartland & Virga (2010)
remark

For a deformable particle the configuration space Q will also allow

for extra configurational degrees of freedom in addition to the usual
translational and orientational ones.

13
 Density Functional Theory
(mf)
• The mean-field free energy FN is a scalar depending on (N, V, T )
only in a parametric form.
• Considering, within the canonical ensemble, how particles
are distributed in the configuration space Q, we construct a
theory where the mean-field free energy becomes a functional
F of the distribution density.
distribution (number) density
Z
ρ := Q → R+ ρ(q)dq = N
Q
effective potential
N Z
1 X 1
U (A) (qh , qj ) ≈ E(qh ) → E(q) := ρ(q ′ )U (A) (q, q ′ )dq ′
2 2 Q
j=1,j6=h

14
 partitioning Q

M
[
Q= Q(i)
i=1
We split the system in a great number, say M , of subsystems
Ni number of particles in the ith subsystem
q (i) ∈ Q(i) core of the ith subsystem
Q(i) ⊂ Q region occupied by the ith subsystem
∆qi measure of Q(i)
discretizing ρ

  M
X
ρ : Q → R+ ρi := ρ q (i) Ni = ρi ∆qi Ni = N
i=1

15
 free-energy additivity

M
X
(mf) (mf)
FN = F Ni
i=1
Z  Ni
(mf) 1 1
− kT E(q)
F Ni = −kT ln e dq
Ni ! Q(i)
M   Ni
(mf)
X 1 Ni − 1 E (i)
FN = −kT ln e kT
i=1
Ni ! ρi
 
E (i) := E q (i)
Stirling’s approximation
 
1 Ni
ln Ni ≈Ni
Ni !

16
 Free-energy functional

M  (i)

(mf)
X 1 E
FN ≈ − kT Ni ln − − kT N
i=1
ρi kT
Disregarding the last addendum, as it does not depend on ρ, we justify
the definition
Z Z
1
F [ρ] := kT ρ(q) ln ρ(q)dq + ρ(q ′ )ρ(q)U (A) (q ′ , q)dq ′ dq
Q 2 Q2
Z
1
E [ρ] := ρ(q ′ )ρ(q)U (A) (q ′ , q)dq ′ dq
2 Q2
Z
ρ(q)dq = N
Q

17
 Maier-Saupe Free Energy
Here we apply the general mean-field theory presented so far to Lon-
don dispersion forces interactions between non-polar molecules.
London dispersion forces

• Liquid crystal molecules are thought of as being both neutral

and with no permanent dipole.
• However, instantaneous dipoles can arise in a molecule and
the electric field thus created can polarize a nearby molecule.
• Though the (time) average of dipoles vanishes in each molecule,
the average energy between induced dipoles does not vanish
and results in an attractive interaction.
• This basic interaction mechanism can easily be understood clas-
sically, but the evaluation of the induced dipole can only be
achieved via a quantum mechanics (perturbative) computa-
tion, relying on a number of approximations.

18
 Unsöld approximation

Assume that all states in the molecules that contribute to their dis-
persion interaction have excitation energies close to E0 .
Unsöld (1927), London (1930,1937)
dispersion energy

C 3 E0
Udisp = − 6 U(er )αmol U(er ) · α′mol C=
r 4 (4π)2
r distance between molecular centres
αmol , α′mol molecular polarizability tensors
U(er ) := er ⊗ er − 31 I
er unit vector along the line joining molecular centres

19
 degrees of freedom

Molecules are treated as rigid particles:

Q=B×Ω
B region in space of volume V
Ω orientational manifold
q = (x, ω)
x position vector of molecular centre
ω orientational degrees of freedom (e.g. Euler angles)
attractive potential


(A) ′
U
disp r≧R
U (q, q ) :=
0 r<R

20
 spatially homogeneous density

ρ(q) = ρ0 ̺(ω)
|{z}
N
V
Z Z Z
ρ(q)dq = N ̺(ω)dω = N ⇒ ̺(ω)dω = 1
Q Ω Ω

effective potential
Z
1
E(x, ω) = ρ0 ̺(ω ′ )U (A) (x, ω; x′ , ω ′ )dx′ dω ′
2 B×Ω
Z Z ∞ Z 
1 Cρ0 ′ ′ ′
=− U(e r )α mol (ω)U(e r )drdA(er ) · α mol (ω )̺(ω )dω
2 Ω R r 4 S2
Z
2πCρ0 ′ ′ ′
=− α mol (ω) · α mol (ω )̺(ω )dω + constant
135R3 Ω
in the thermodynamic limit, N, V → ∞ with ρ0 unchanged.

21
 uniaxial molecules

Ω = S2

1
αmol (ω) = (αk − α⊥ )q(ω) + (2α⊥ + αk )I
3
q(ω) = m(ω) ⊗ m(ω) − 31 I molecular tensor

Z
1 ′ ′ ′ π (αk − α⊥ )2
E(x, ω) = − U0 q(ω) · q(ω )̺(ω )dω U0 := 3 2
E0 ρ0
2 S2 45R (4π)
Z
1 ′ ′ ′ ′ ′ 2
= − U0 P2 (m(ω) · m(ω ))̺(ω )dω U0 := U0
2 S 2 3
E is independent of x
Z
1
E(ω) = − U0 q(ω) · q(ω ′ )̺(ω ′ )dω ′
2 S2

22
 Free-energy functional

Z Z
1
F [ρ] : = kT ρ(q) ln ρ(q)dq + ρ(q ′ )ρ(q)U (A) (q ′ , q)dq ′ dq
Q 2 Q2
Z Z
= kT ρ(q) ln ρ(q)dq + E(ω)ρ(q)dq
Q Q

q = (x, ω)
ρ(q) = ρ0 ̺(ω)
Z
FMS [ρ] = kT N ̺(ω) ln(ρ0 ̺(ω))dω
S2
Z
1
− U0 N ̺(ω ′ )̺(ω)q(ω ′ ) · q(ω)dω ′ dω
2 S2 ×S2
Z
̺(ω)dω = 1
S2

23
 (dimensionless) free-energy per particle

Z
1
FMS [̺] := FMS [ρ] = ̺(ω) ln ̺(ω)dω
N kT S2
Z
1
− β ̺(ω ′ )̺(ω)q(ω ′ ) · q(ω)dω ′ dω
2 S2 ×S2
U0
β := dimenionless reciprocal temperature
kT
first variation
Z  Z  
δFMS (̺)[δ̺] = (1 − λ) + ln ̺(ω) − β q(ω ′ )̺(ω ′ )dω ′ · q(ω) δ̺(ω)dω
S2 S2
R
λ Lagrange multiplier for S2
̺(ω)dω = 1

24
 equilibrium density

R
β( q(ω ′ )̺eq (ω ′ )dω ′ )·q(ω)
e S2 eβQ·q(ω)
̺eq (ω) = R R = R βQ·q(ω)
β ( S2 q(ω )̺eq (ω )dω )·q(ω)
′ ′ ′
e dω
S 2 e dω S2
Z
Q= q(ω ′ )̺eq (ω ′ )dω ′
S2

• The integral equilibrium equation for ̺eq becomes the self-

consistency equation for the order tensor Q.
1
• Q = 0, and ̺eq = ̺0 ≡ 4π , is a solution for all β.

25
 (rescaled) equilibrium free energy

 Z 
1 1 1 1
fMS (β, Q) := (FMS [̺eq ]+ln 4π) = Q·Q− ln eβQ·q(ω) dω
β 2 β 4π S2

fMS (β, 0) = 0 for all β

critical points

R
∂fMS S2
q(ω)eβQ·q(ω) dω
=Q− R
βQ·q(ω) dω
∂Q S 2 e
∂fMS
= 0 ⇔ Q = hqi̺eq
∂Q
The critical points of fMS are solutions to the self-consistency equa-
tion for Q and have necessarily admissible eigenvalues.

26
 critical point properties

• We regard fMS (β, ·) as defined in the whole space of symmetric

traceless tensors.
• But fMS (β, Q) represents the mean-field free energy only at its
critical points Q. Its non-critical values are physically irrele-
vant.
• For given β, the order tensor Q at which fMS (β, ·) attains its
least critical value represents the stable equilibrium phase.
• Q = 0 is a critical point of fMS (β, ·) for all values of β.
• fMS (β, ·) is an isotropic function:
fMS (β, RQRT ) = fMS (β, Q) ∀ R ∈ O(3).
• If Q 6= 0 is a critical point of fMS (β, ·), then Q is uniaxial.
Fatkullin & Slastikov (2005)
Liu, H. Zhang & P. Zhang (2005)

27
 scalar order parameters

 
1
Q = S ez ⊗ ez − I + T (ex ⊗ ex − ey ⊗ ey ) .
3
reduced free energy

fMS (β, Q) = fs (β, S, T )

Z 2π Z π
1 2 1 1
fs (β, S, T ) := S + T 2 − ln dϕ eβgs (ϑ,ϕ;S,T ) sin ϑdϑ
3 β 4π 0 0
 
1
2
gs (ϑ, ϕ; S, T ) = S cos ϑ − + T sin2 ϑ cos 2ϕ
3
coercivity

2 1 2 2 2
|gs | < |S| + |T | ⇒ fs (β, S, T ) > S + T − |S| − |T |
3 3 3

28
 • fs (β, ·, ·) enjoys a 6-fold symmetry in the admissible triangle.
• All critical points of fs (β, ·, ·) lie on the uniaxial lines.

0.6

U′+
0.4

U
0.2

T 0 U′
O U+
−0.2

U ′′
−0.4

U′′+
−0.6

−0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 1.2

S

We shall restrict the search for the critical values of fs to T = 0.

29
 restricted free energy

∂fs
(β, S, 0) = 0 ∀ β, S
∂T
⇓

All critical points (S0 , 0) of fs (β, ·, ·) are critical points of

Z π
1 2 1 1 βS (cos2 ϑ− 13 )
fu (β, S) := fs (β, S, 0) = S − ln e sin ϑdϑ
3 β 2 0
√
!
1 2 2 1 daw βS
fu (β, S) = S − S − ln √
3 3 β βS
Z x
2 2
daw(x) := e−x et dt
0

30
 Dawson’s integral

daw
0.6

0.4

0.2

x
-10 -8 -6 -4 -2 2 4 6 8 10
-0.2

-0.4

-0.6
31
 change of variables

x :=βS
    √ 
x 1 1 2 2 daw ( x)
fu β, = x − x − ln √
β β 3β 3 x
equilibria
 
∂fu 2x 1
= − G(x)
∂S 3 β
3 3 1
G(x) := √ √ − −
4x x daw ( x) 4x2 2x
2
lim G(x) =
x→0 15
1 1
G(x) ≈ for x → +∞ G(x) ≈ − for x → −∞
x 2x

32
 y

0.25

0.20

0.15 G∗

G(x)
0.10

G(x)
0.05

x∗
x
-10 -8 -6 -4 -2 2 4 6 8 10
33
 local uniaxial stability

.
• For β1 > G∗ = 0.149 there is only one equilibrium solution,
S = 0.
.
• For β1 < G∗ = 0.149 there are two extra equilibrium solutions.
1 2
• One of the extra solutions has S < 0 for β < 15 .

 

 

2

  

∂ fu 2β 1 ′
= − G(x) −xG (x)
∂S 2 3 
 β 


| {z } 

0 at equilibrium

34
 y

0.25

0.20

0.15 G∗

G(x)
0.10

G(x)
0.05

x∗
x
-10 -8 -6 -4 -2 2 4 6 8 10
35
 local biaxial stability

 
1 2 4


fs (β, S, T ) = fu (β, S) + β − g (βS) T + O T
β
| {z }
G(βS)−g(βS)

1 1 3 1 3
g(x) := + + − √ √ − √ √ .
4 4x 16x2 8 x daw ( x) 16x x daw ( x)

36
 y

0.25

0.20

0.15 G∗

G(x)
0.10

G(x)
0.05

g(x)
x∗
x
-10 -8 -6 -4 -2 2 4 6 8 10
37
 y

0.25

0.20

0.15 G∗

G(x)
0.10

G(x)
0.05

x∗
x
-10 -8 -6 -4 -2 2 4 6 8 10
38
 absolute minimizer

• The free energy vanishes on the isotropic phase:

fu (β, 0) = 0

• The free energy of the competing locally stable nematic phase

is negative whenever

G(x) < F (x)

 √ 
1 3 daw ( x)
F (x) := + 2 ln √
x x x

39
 y

0.25

0.20

0.15 G∗ G∗

F (x)

G(x)
0.10

G(x)
0.05

x∗ x∗ x
-10 -8 -6 -4 -2 2 4 6 8 10
40
 y

0.25

0.20

0.15 G∗ G∗

G(x)
0.10

G(x)
0.05

x∗ x∗ x
-10 -8 -6 -4 -2 2 4 6 8 10
41
 S
1.0

0.8

0.6
S∗
0.4

0.2

G∗
1/β
0.03 0.06 0.09 0.12 0.15 0.18
. . .
x∗ = 2.923 G∗ = 0.147 S ∗ = x∗ G∗ = 0.429

42
 Sources

• I. Fatkullin & V. Slastikov, Critical points of the Onsager

functional on a sphere, Nonlinearity, 18, 2005, 2565–2580.
• E.C. Gartland, Jr., & E.G. Virga, Minimum principle for
indefinite mean-field free energies, Arch. Rational Mech.
Anal., 196, 2010, 143–189.
• J.W. Gibbs, Elementary Principles in Statistical Me-
chanics, Charles Scribners’s Sons, New York, 1902. Digitally
reprinted by Cambridge University Press, Cambridge, 2010.
• H. Liu, H. Zhang & P. Zhang, Axial symmetry and classi-
fication of stationary solutions of Doi-Onsager equation on the
sphere with Maier-Saupe potential, Comm. Math. Sci., 3,
2005, 201–208.
• F. London, The general theory of molecular forces, Trans.
Faraday Soc., 33, 1937, 8–26.
43
• W. Maier & A. Saupe, Eine einfache molekulare Theorie des
nematischen kristallinflüssigen Zustandes, Z. Naturforsch.,
13a, 1958, 564–566.
• P. Palffy-Muhoray, M.Y. Pevnyi, E.G. Virga & X.
Zheng, The effects of particle shape in orientationally ordered
soft materials, in Mathematics and Materials, vol. 23,
PCMS Series, AMS, 2017, in press.
• M. Kleman & O.D. Lavrentovich, Soft Matter Physics:
An Introduction, Springer-Verlag, New York, 2003.
• J.V. Selinger, Introduction to the Theory of Soft Mat-
ter: From Ideal Gases to Liquid Crystals, Springer, Cham,
2016.
• A.M. Sonnet & E.G. Virga, Dissipative Ordered Fluids:
Theories for Liquid Crystals, Springer Science, New York,
2012.

44
• A. Unsöld, Quantentheorie des Wasserstoffmolekülions und
der Born- Landéschen Abstoßungskräfte, Z. Phys., 43, 1927,
563–574.

45