You are on page 1of 8

THE JOURNAL OF CHEMICAL PHYSICS 130, 234503 共2009兲

Composition dependence of glass transition temperature and fragility.


II. A topological model of alkali borate liquids
John C. Mauro,1,a兲 Prabhat K. Gupta,2 and Roger J. Loucks3
1
Science and Technology Division, Corning Incorporated, Corning, New York 14831, USA
2
Department of Materials Science and Engineering, Ohio State University, Columbus, Ohio 43210, USA
3
Department of Physics and Astronomy, Alfred University, Alfred, New York 14802, USA
共Received 23 March 2009; accepted 19 May 2009; published online 15 June 2009兲

Glass transition temperature and fragility are two important properties derived from the temperature
dependence of the shear viscosity of glass-forming melts. While direct calculation of these
properties from atomistic simulations is currently infeasible, we have developed a new topological
modeling approach that enables accurate prediction of the scaling of both glass transition
temperature and fragility with composition. A key feature of our approach is the incorporation of
temperature-dependent constraints that become rigid as a liquid is cooled. Using this approach, we
derive analytical expressions for the composition 共x兲 dependence of glass transition temperature,
Tg共x兲, and fragility, m共x兲, in binary alkali borate systems. Results for sodium borate and lithium
borate systems are in agreement with published values of Tg共x兲 and m共x兲. Our modeling approach
reveals a natural explanation for the presence of the constant Tg regime observed in alkali borate
systems. © 2009 American Institute of Physics. 关DOI: 10.1063/1.3152432兴

I. INTRODUCTION ing the number of constraints to the atomic degrees of free-


dom, it is possible to predict compositions with optimum
Accurate knowledge of the temperature 共T兲 and compo- glass-forming ability; glass network topology is also respon-
sition 共x兲 dependence of shear viscosity, ␩共T , x兲, is of critical sible for critical behavior in a number of mechanical and
importance for all stages of industrial glass production.1,2 It thermal properties.25–38 While previous topological models
is also vital for our scientific understanding of glass transi- are formulated for zero temperature conditions, incorporation
tion and relaxation phenomena. Much effort—both of temperature-dependent constraints13 allows us to extend
experimental and theoretical—has been invested over several the approach to enable calculation of ␩共T , x兲.
decades in the measurement, characterization, and under- The glass transition temperature, Tg共x兲, and the fragility,
standing of the viscosity of glass-forming liquids. Extensive m共x兲, are two important parameters describing the tempera-
sets of measured viscosity data are available for a wide va- ture dependence of viscosity at the glass transition. Although
riety of liquids over large temperature ranges.1,3,4 While the glass transition temperature can be defined in many
there has been much effort to systematize this large amount ways, here we follow the convention of Angell39–42 by defin-
of data in the hope of uncovering universal trends,5,6 our ing Tg as the temperature at which the shear viscosity is
understanding of the mechanisms of viscous flow remains equal to a fixed value, generally taken as 1012 Pa s:1,3
rather poor. There has recently been growing attention and
research activity in understanding viscosity and various as- ␩关Tg共x兲,x兴 = 1012 Pa s. 共1兲
sociated phenomena from the theoretical physics and chem-
istry communities.7–10 While the recent enthalpy landscape With this definition, the viscosities of all glass-forming com-
modeling of Mauro and Loucks11 has enabled the calculation positions can be mapped onto a single universal plot in terms
of both glass transition temperature and fragility for simple of Tg / T.39–42 This definition of Tg is also equivalent to the
systems such as selenium, where the interatomic potentials annealing point,1 where the structural relaxation time is on
can be computed accurately from ab initio physics, conven- the order of 100 s. Following Angell,39–42 the fragility de-
tional atomistic modeling techniques such as molecular dy- scribes the slope of the viscosity curve with respect to in-
namics cannot access the long times required to compute verse temperature at the glass transition temperature and is
viscosity in the supercooled regime.12 defined as39–44
In the first paper of this series,13 we presented a new
topological modeling approach that formulates shear viscos-
ity in terms of temperature-dependent constraints that be-
come frozen as the liquid is cooled from high temperature.
m共x兲 ⬅ 冏 ⳵ log10 ␩共T,x兲
⳵ 关Tg共x兲/T兴
冏 T=Tg共x兲
. 共2兲

Our approach builds on the pioneering work of Phillips and In our previous paper,13 we presented a new modeling
Thorpe14–19 and that of Gupta and Cooper,20–24 who first pro- approach for the calculation of Tg共x兲 and m共x兲, starting with
posed treating glass as a network of constraints. By compar- the Adam–Gibbs relation.45–50 The Adam–Gibbs relation de-
scribes viscosity in terms of the configurational entropy,
a兲
Electronic mail: mauroj@corning.com. Sc共T , x兲, of the melt:

0021-9606/2009/130共23兲/234503/8/$25.00 130, 234503-1 © 2009 American Institute of Physics

Downloaded 15 Jun 2009 to 199.197.135.217. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
234503-2 Mauro, Gupta, and Loucks J. Chem. Phys. 130, 234503 共2009兲

B共x兲 rigid 共q␣ = 1兲; in the limit of infinite temperature, constraints


log10 ␩共T,x兲 = log10 ␩⬁ + . 共3兲 are easily broken and hence do not contribute to the rigidity
TSc共T,x兲
of the network 共q␣ = 0兲. At a finite temperature, some fraction
Here, ␩⬁ is the extrapolated composition-independent infi- of the constraints are broken while others are rigid. In this
nite temperature viscosity39,51 and B共x兲 is a fitting parameter. paper we employ two forms of q␣共T兲, continuous and dis-
While B共x兲 is commonly associated with the activation en- crete. In terms of the energy landscape approach, q␣共T兲 can
ergy of the Arrhenius model for viscosity, Adam and Gibbs45 be expressed as a continuous function of temperature,55

冋 冉 冊册
showed that B共x兲 is given by ␯tobs
⌬F␣ⴱ
sⴱc 共x兲 q␣共T兲 = 1 − exp − , 共9兲
B共x兲 = n p⌬␮共x兲 , 共4兲 kT
k
where ⌬F␣ⴱ is the activation free energy for breaking the ␣
where n p is the number of particles in the system, ⌬␮ is the constraint, ␯ is the vibrational attempt frequency, and tobs is
average energy barrier for structural rearrangement, sⴱc is the the observation time. Owing to this dependence on observa-
configurational entropy of the smallest rearranging region, tion time, the continuous form of q␣共T兲 contains information
and k is Boltzmann’s constant. While both ⌬␮ and sⴱc are on the cooling rate dependence of the glass transition.55,56 A
functions of composition, Toplis52 showed that B共x兲 often fast cooling rate indicates a lower tobs and an earlier onset of
varies unappreciably within a given class of material 共e.g., the glass transition. Since typically q␣共T兲 is a rather sharp
alkali silicates兲. In our model we assume that the major com- function of temperature, it can be approximated in discrete
position dependence of Tg共x兲 is due to variations in Sc共T , x兲, form with a unit step function:
and hence we take B共x兲 as a constant. With this approxima-
q␣共T兲 = ␪共T␣ − T兲, 共10兲
tion, the glass transition temperature Tg共x兲 can be computed
relative to that of some reference composition xR; combining where T␣ represents the onset temperature below which a
Eqs. 共1兲 and 共3兲, we have particular constraint ␣ becomes rigid. In practice, the dis-
Tg共x兲 Sc关Tg共xR兲,xR兴 crete form of q␣共T兲 in Eq. 共10兲 is convenient for deriving
= . 共5兲 analytical expressions for glass transition temperature, Tg共x兲,
Tg共xR兲 Sc关Tg共x兲,x兴 employing Eq. 共6兲. However, as fragility is computed in
According to the recent energy landscape modeling of terms of a temperature derivative of f共T , x兲,13
Naumis,53,54 the configurational entropy of a system is
largely proportional to the density of atomic degrees of free-
dom, f共x兲, i.e., the number of low-frequency “floppy modes.”
冉 冏
m共x兲 = m0 1 +
⳵ ln f共T,x兲
⳵ ln T
冏 冊
T=Tg共x兲
, 共11兲

With this result, Eq. 共5兲 becomes the continuous form of q␣共T兲 in Eq. 共9兲 is used for calcula-
Tg共x兲 f关Tg共xR兲,xR兴 d − n关Tg共xR兲,xR兴 tion of m共x兲. Here, m0 ⬇ 17 is the fragility of a strong liquid
= = , 共6兲 such as silica.13
Tg共xR兲 f关Tg共x兲,x兴 d − n关Tg共x兲,x兴
Application of our topological modeling approach to any
where d = 3 is the dimensionality of the network and n共T , x兲 new system involves three basic steps:
is the average number of constraints per atom. The con-
共1兲 Identify and count the number of distinct network-
straints per network node can be calculated by averaging
forming species as a function of x. By distinct, we
over all network-forming species i and each type of con-
mean either chemically distinct or chemically similar
straint ␣. By “network-forming species” we mean any atom
but having different short-range orders 共i.e., coordina-
that is bound to at least two other atoms as part of the infi-
tion number兲.
nitely large, topologically disordered glass network. Defining
共2兲 Identify and count the number of constraints associated
Ni共x兲 as the mole fraction of network-forming species i in
with each species. Rank different constraints in terms
composition x, we may write the average number of atomic
of their relative bond strengths 共or onset temperatures兲.
constraints as
共3兲 Apply Eqs. 共6兲 and 共11兲 to calculate Tg共x兲 and m共x兲,
n共T,x兲 = 兺 Ni共x兲 兺 wi,␣q␣共T兲. 共7兲 respectively.
i ␣
In this paper, we apply our topological constraint model
The first sum in Eq. 共7兲 is over the network-forming species for the binary alkali borate system. Equation 共6兲 is used to
i, and the second sum is over the various constraints ␣, provide an analytical form for Tg共x兲, and m共x兲 is computed
which may include two-body linear, three-body angular, or using the straightforward numerical calculation of Eq. 共11兲.
higher-order constraints. Here, wi,␣ is the number of ␣-type We show agreement between calculated and published val-
constraints associated with species i, and the temperature de- ues of Tg共x兲 and m共x兲 for both sodium and lithium borate
pendence of the constraints is accounted by q␣共T兲, a measure systems.
of rigidity for constraint ␣. Each q␣共T兲 satisfies the limits
lim q␣共T兲 = 1 and lim q␣共T兲 = 0. 共8兲
T→0 T→⬁ II. STRUCTURE AND TOPOLOGY

In the limit of zero temperature there is no thermal energy The alkali borate system, xM 2O · 共1 − x兲B2O3, has proved
available to break a constraint, so all constraints are fully a significant challenge for theoretical modeling due to the

Downloaded 15 Jun 2009 to 199.197.135.217. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
234503-3 Topological model of alkali borates J. Chem. Phys. 130, 234503 共2009兲

0.7
so-called boron anomaly.1,57–65 In contrast to alkali silicate

Fraction of Network-Forming Species


systems, where addition of alkali invariably creates non- 0.6
Bridging Oxygen
bridging oxygen by rupturing the linkages between SiO4
0.5
tetrahedra1,66–69 and leading to an increase in floppy modes,
3
when an alkali is added to borate glass there are two possi- 0.4 Q
bilities: 0.3

共1兲 Create a nonbridging oxygen, rupturing the linkage be- 0.2


tween two trigonally coordinated BO3 groups, decreas-
0.1
ing the rigidity of the network. Q
4
Q
2

共2兲 Convert boron from a three-coordinated state to a four- 0


coordinated state, effectively increasing the rigidity of 0 0.1 0.2 0.3 0.4 0.5

the network. In this case, no nonbridging oxygen is (a) x

created.
1

Limiting ourselves to the x ⱕ 1 / 2 regime, there are four 0.9


Q
3

distinct types of network-forming nodes to consider: 共i兲 four- 0.8

Fraction of Boron Species


coordinated boron bonded to four bridging oxygens 共Q4兲, 共ii兲 0.7

three-coordinated boron bonded to three bridging oxygens 0.6

共Q3兲, 共iii兲 three-coordinated boron bonded to two bridging 0.5

oxygens 共Q2兲, and 共iv兲 the bridging oxygen itself 共OB兲. Nei- 0.4

ther nonbridging oxygens nor alkali ions are considered as 0.3

x = 1/3
4
Q 2
network formers. 0.2 Q

Properly accounting for boron speciation is especially 0.1

challenging for molecular dynamics simulations.70–72 Fortu- 0


0 0.1 0.2 0.3 0.4 0.5
nately, the random pair model of Gupta73 has proved suc-
(b) x
cessful at capturing the fraction of each boron species across
a wide range of alkali content. The random pair model is FIG. 1. 共Color online兲 共a兲 Fraction of network-forming species in the
based on the following rules for network formation: xM 2O · 共1 − x兲B2O3 system following the random pair model of Gupta 共Ref.
73兲. 共b兲 Boron speciation in the xM 2O · 共1 − x兲B2O3 system. Initially alkali
• BO4 tetrahedra occur in corner-sharing pairs, where the act to convert boron from the Q3 to the Q4 state. Further addition of alkali in
B–O–B angle within a pair is random. the x ⬎ 1 / 3 regime leads to formation of nonbridging oxygen and Q2 boron.

• Pairs of BO4 tetrahedra cannot be bound to each other.

冦 冧
3 − 2x 1
• Nonbridging oxygens occur in BO3 groups only and not , xⱕ ,
5 − 4x 3
in BO4 groups. N共OB兲 = 共15兲
21 − 28x 1 1
With the chemical formula xM 2O · 共1 − x兲B2O3 and the , ⬍xⱕ ,
31 − 38x 3 2
structural rules above, an alkali-to-boron ratio of less than
one-half 关x / 共1 − x兲 ⱕ 1 / 2, or x ⱕ 1 / 3兴 will lead to all alkali which are plotted in Fig. 1.
converting boron from three to four coordinations. Non- We consider the following types of constraints in the
bridging oxygens begin to form only at higher alkali concen- alkali borate system:
trations, x ⬎ 1 / 3. With the random pair model,73 the fraction
• ␣: B–O linear bonds. There are two ␣ constraints at
of each network-forming species is
each bridging oxygen.

冦 冧
2x 1
, xⱕ , • ␤: O–B–O angular constraints. There are five ␤ con-
5 − 4x 3 straints at each Q4 unit to form a rigid BO4 tetrahedron,
N共Q 兲 =
4
共12兲
6 − 8x 1 1 three ␤ constraints at each Q3 to keep the BO3 unit
, ⬍xⱕ , planar, and a single ␤ constraint at each Q2.
31 − 38x 3 2
• ␥: B–O–B angular constraints. There is one ␥ constraint

冦 冧
2 − 4x 1 at every bridging oxygen.
, xⱕ ,
5 − 4x 3
N共Q3兲 = 共13兲 The bond strengths are ordered such that
10 − 20x 1 1
, ⬍xⱕ ,
31 − 38x 3 2 T␥ ⬍ Tg共0兲 ⬍ T␤ ⬍ T␣ . 共16兲

冦 冧
1 The linear bonds are strongest and freeze in at the highest
0, xⱕ , temperature. Following the structural analysis of Mozzi and
3
N共Q2兲 = 共14兲 Warren,74 the O–B–O angle poses a stronger constraint than
18x − 6 1 1 the B–O–B angle. The glass transition temperature of pure
, ⬍xⱕ ,
31 − 38x 3 2 B2O3, Tg共0兲 = 543 K, must fall between T␥ and T␤ in order to

Downloaded 15 Jun 2009 to 199.197.135.217. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
234503-4 Mauro, Gupta, and Loucks J. Chem. Phys. 130, 234503 共2009兲

satisfy the conditions that 0 ⬍ f关Tg共x兲 , x兴 ⬍ 3 and ⳵Tg共x兲 / ⳵x 3.5


x M2O·(1-x )B2O3
⬎ 0 at x = 0. 3.0
T >T
: B-O
2.5

Atomic Degrees of Freedom


III. GLASS TRANSITION TEMPERATURE : O-B-O
: B-O-B
2.0
Following the analysis of Sec. II, the number of atomic
T <T <T

Cooling
constraints is given by 1.5

冦 冧
12 − 6x 1 1.0
, xⱕ ,
5 − 4x 3 T <T <T
n共Tg共x兲,x兲 = 共17兲 0.5
96 − 138x 1 1
, ⬍xⱕ , 0.0
31 − 38x 3 2
T <T
in the temperature range T␥ ⬍ Tg共x兲 ⬍ T␤. Here we have as- -0.5
0 0.1 0.2 0.3 0.4 0.5
sumed the discrete form of q␣共T兲, as in Eq. 共10兲. The corre-
sponding degrees of freedom are (a) x

冦 冧
1300
3 − 6x 1
, xⱕ , 1200
5 − 4x 3 1100 T  < T g (x ) < T 
f共Tg共x兲,x兲 = 共18兲
24x − 3 1 1 1000
⬍xⱕ ,

Temperature (K)
,
31 − 38x 3 2 900

800
plotted in Fig. 2共a兲. Combining Eqs. 共6兲 and 共18兲 leads to the T  = 740 K
700
following scaling of Tg with composition: T g (x )

冉 冊
600

冦 冧
1 5 − 4x 1 500
Tg共0兲, xⱕ , T g (0) = 543 K
5 1 − 2x 3 400

冉 冊 冉冊
Tg共x兲 = 共19兲 300
T  = T 0 (0) = 328 K
1 31 − 38x 1 1 1
Tg , ⬍xⱕ , 0 0.1 0.2 0.3 0.4 0.5
11 8x − 1 3 3 2 (b) x

where Tg共1 / 3兲 is the glass transition temperature of the dibo-


FIG. 2. 共Color online兲 共a兲 Temperature and composition dependence of the
rate composition. Solving for Tg共1 / 3兲, we obtain average number of atomic degrees of freedom in the binary alkali borate

冉 冊
冦 冧
system, xM 2O · 共1 − x兲B2O3. We consider the following constraints, in order
1 5 − 4x 1 of decreasing strength: B–O linear bonds 共␣兲, O–B–O angular constraints
Tg共0兲, xⱕ ,
5 1 − 2x 3 共␤兲, and B–O–B angular constraints about the bridging oxygen. The dashed

冉 冊
Tg共x兲 = 共20兲 line corresponds to the isostatic situation, where f = 0. 共b兲 Plot of the con-
1 31 − 38x 1 1 straint onset temperatures and Tg共x兲 for the sodium borate system. In order
Tg共0兲, ⬍xⱕ ,
5 8x − 1 3 2 to satisfy the condition that the product f关Tg共x兲 , x兴 · Tg共x兲 is a constant, the
final Tg共x兲 for the alkali borate system is given by Eq. 共23兲.
which is valid for glass transition temperatures falling in the
range T␥ ⬍ Tg共x兲 ⬍ T␤. Equation 共20兲 is plotted as the blue
f关Tg共x兲 , x兴. In Fig. 2共a兲, this is equivalent to jumping from
共gray兲 curve in Fig. 2共b兲 and shows a clear peak at x = 1 / 3.
the blue curve 共T␥ ⬍ T ⬍ T␤兲 to the green curve at higher
However, at some point the glass transition temperature will
temperature 共T␤ ⬍ T ⬍ T␣兲. When this happens, the Tg共x兲
exceed T␤, the onset temperature for the O–B–O angular
must decrease in order to achieve a constant f关Tg共x兲 , x兴 · Tg共x兲
constraints, violating our assumption of T␥ ⬍ Tg共x兲 ⬍ T␤ and
and compensate for this increased floppiness. However,
leading to a very interesting effect.
when Tg共x兲 decreases slightly below T␤ the network sud-
From Eq. 共6兲, it is clear that the product f关Tg共x兲 , x兴 · Tg共x兲
denly becomes more rigid due to a decrease in f关Tg共x兲 , x兴,
is a constant for all x. In order to satisfy this condition, there
and the Tg共x兲 must increase to compensate for this loss of
must be a horizontal jump in Tg共x兲 between the two intersec-
floppiness. The net result is a regime of constant Tg共x兲 = T␤
tion points of the blue 共gray兲 curve with the T␤ isotherm in
until enough nonbridging oxygens form to satisfy the condi-
Fig. 2共b兲, i.e., the two intersection points of Eq. 共20兲 with T␤.
tion of constant f关Tg共x兲 , x兴 · Tg共x兲 with lower values of the
Denoting these two intersection points x1 and x2, we have
glass transition temperature. Hence, the final Tg共x兲 curve for


5 T␤/Tg共0兲 − 1
冊 共21兲
the alkali borate system is given by

冉 冊
x1 =

冦冉 冧
2 5T␤/Tg共0兲 − 2 1 5 − 4x
Tg共0兲, x ⱕ x1 ,
and 5 1 − 2x
31 + 5T␤/Tg共0兲 Tg共x兲 = T ␤, x1 ⬍ x ⬍ x2 , 共23兲
x2 =
38 + 40T␤/Tg共0兲
.

Physically, when Tg共x兲 increases slightly above T␤ the net-


共22兲 1 31 − 38x
5 8x − 1
Tg共0兲, 冊 1
x2 ⱕ x ⱕ .
2
work suddenly becomes more floppy due to an increase in If Tg共x兲 were to fall below T␥, there could be another

Downloaded 15 Jun 2009 to 199.197.135.217. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
234503-5 Topological model of alkali borates J. Chem. Phys. 130, 234503 共2009兲

850 TABLE I. Experimental values of Tg共x兲 and m共x兲 in the xNa2O · 共1


x Na2O·(1-x )B2O3 − x兲B2O3 system. The measurements of Nemilov 共Ref. 79兲 use the penetra-
800

Glass Transition Temperature (K)


tion method. Stolyar et al. 共Ref. 80兲 employed the beam bending technique,
750 and Suzuki et al. 共Ref. 81兲 used application of a compressive force.
700
x Tg 共K兲 m Reference
650
0.0249 543.6 35.1 Nemilova
600 0.0481 568.3 36.9 Nemilova
550 0.0715 593.5 39.2 Nemilova
Model Nemilov 0.0959 616.7 40.1 Nemilova
500 Suzuki Stolyar 0.1215 646.2 42.6 Nemilova
450
Moynihan 0.1379 663.9 43.1 Nemilova
0 0.1 0.2 0.3 0.4 0.5 0.1582 694.6 49.1 Nemilova
(a) x 0.1767 718.0 48.9 Nemilova
850 0.1831 717.5 51.8 Nemilova
x Li2O·(1-x )B2O3 0.1934 728.4 55.7 Nemilova
800
Nemilova
Glass Transition Temperature (K)

0.2111 732.8 57.4


750 0.2214 732.8 57.3 Nemilova
0.2242 735.3 59.2 Nemilova
700
0.2364 734.9 57.1 Nemilova
650 0.2505 738.9 61.7 Nemilova
0.2733 736.2 59.9 Nemilova
600
0.2999 737.7 62.1 Nemilova
550 0.3018 742.6 60.7 Nemilova
500
0.3268 737.5 61.4 Nemilova
Model Kodama 0.3567 732.5 66.3 Nemilova
Chryssikos Moynihan
450 0.3615 727.8 69.3 Nemilova
0 0.1 0.2 0.3 0.4 0.5
0.0490 600.4 38.5 Stolyar et al.b
(b) x
0.0941 643.1 40.6 Stolyar et al.b
0.1297 692.0 47.4 Stolyar et al.b
FIG. 3. 共Color online兲 Composition dependence of glass transition tempera- 0.1343 689.1 47.8 Stolyar et al.b
ture for the 共a兲 sodium borate and 共b兲 lithium borate systems. The solid lines
0.1878 738.6 65.6 Stolyar et al.b
show the predicted Tg共x兲 using the topological model with temperature-
dependent constraints. Experimental data points 共Refs. 78–83兲 are provided 0.2111 747.2 66.3 Stolyar et al.b
in Tables I and II. 0.2242 749.2 70.4 Stolyar et al.b
0.2373 750.5 70.2 Stolyar et al.b
0.2524 749.9 73.0 Stolyar et al.b
horizontal jump in Tg共x兲 at lower temperatures. However, as 0.2581 748.8 74.2 Stolyar et al.b
shown in Fig. 2共b兲, Tg共x兲 remains above T␥ for the entire 0.2619 748.5 73.5 Stolyar et al.b
composition range of interest 共x ⱕ 1 / 2兲. Here, we have esti- 0.2923 746.9 79.8 Stolyar et al.b
mated T␥ based on the Vogel temperature of pure B2O3, T␥ 0.3037 744.6 84.9 Stolyar et al.b
0.3152 740.7 84.3 Stolyar et al.b
⬇ T0共0兲 = 328 K.75 We base this approximation on the violet
0.3200 737.6 84.5 Stolyar et al.b
curve in Fig. 2共a兲, which shows that pure B2O3 loses all
0.0445 586.2 ¯ Suzuki et al.c
configurational degrees of freedom at T ⬍ T␥. In other words, 0.0896 627.9 ¯ Suzuki et al.c
the B2O3 network is isostatic at T = T␥, i.e., f共T␥ , 0兲 = 0. Such 0.1352 684.3 ¯ Suzuki et al.c
vanishing of the topological degrees of freedom can be as- 0.1813 732.8 ¯ Suzuki et al.c
sociated with the Vogel temperature T0共x兲 at which the vis- 0.2279 741.2 ¯ Suzuki et al.c
cosity of a supercooled liquid diverges.1,48,76,77 0.2752 741.4 ¯ Suzuki et al.c
In Fig. 3 we compare our model result of Eq. 共23兲 to a
Reference 79.
published experimental values of Tg共x兲 for the sodium borate b
Reference 80.
c
and lithium borate systems. The published data are also sum- Reference 81.
marized in Tables I and II. The agreement between predicted
and measured values of Tg共x兲 is truly remarkable, especially Tg, as in Eq. 共1兲, the lithium borate data of Kodama and
since there is only one free parameter in Eq. 共23兲, the Kojima82 and Chryssikos et al.83 are, respectively, based on
O–B–O constraint onset temperature T␤. Based on the ex- dilatometry and calorimetry. Also, the critical temperature
perimental data in Tables I and II, we estimate T␤ = 740 K for the immiscibility dome is higher in lithium borates than
for the sodium borate system and T␤ = 760 K for the lithium in the sodium borate system.84,85 With Eqs. 共21兲 and 共22兲 and
borate system. This difference is partly due to the effect of these values of T␤, the iso-Tg regime falls between 共x1 , x2兲
changing alkali and partly because the experiments used for = 共0.19, 0.41兲 for the sodium borate system and 共x1 , x2兲
fitting the sodium borate model employ a different definition = 共0.20, 0.40兲 for the lithium borates.
of Tg compared to those used for fitting the lithium borate The presence of an iso-Tg regime in both of the alkali
model. Whereas all three sets of sodium borate borate systems is testament to the importance of including
measurements79–81 are based on a viscometric definition of the temperature dependence of constraints when constructing

Downloaded 15 Jun 2009 to 199.197.135.217. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
234503-6 Mauro, Gupta, and Loucks J. Chem. Phys. 130, 234503 共2009兲

TABLE II. Experimental values of Tg共x兲 and m共x兲 in the xLi2O · 共1 100
− x兲B2O3 system. Kodama and Kojima 共Ref. 82兲 measured glass transition x Na2O·(1-x )B2O3
90
temperature using differential thermal analysis. Here we estimate values of
Tg共x兲 from their Fig. 5 共Ref. 82兲. The measurements of Chryssikos et al. 80
共Ref. 83兲 were made using differential scanning calorimetry. All of the re-
70
ported compositions are nominal compositions. Values of fragility are esti-

Fragility
mated from their Fig. 4 共Ref. 83兲. 60

x Tg 共K兲 m Reference 50

40
0.00 518 ¯ Kodama and Kojimaa
0.02 528 ¯ Kodama and Kojimaa 30
Model Nemilov
0.04 543 ¯ Kodama and Kojimaa Stolyar Chryssikos
20
0.06 558 ¯ Kodama and Kojimaa 0 0.1 0.2 0.3 0.4 0.5
0.08 598 ¯ Kodama and Kojimaa x
(a)
0.10 618 ¯ Kodama and Kojimaa
90
0.12 638 ¯ Kodama and Kojimaa
x Li2O·(1-x )B2O3
0.14 673 ¯ Kodama and Kojimaa 80
0.16 693 ¯ Kodama and Kojimaa
0.18 723 ¯ Kodama and Kojimaa 70
0.20 748 ¯ Kodama and Kojimaa
60

Fragility
0.22 753 ¯ Kodama and Kojimaa
0.24 758 ¯ Kodama and Kojimaa 50
0.26 763 ¯ Kodama and Kojimaa
0.28 763 ¯ Kodama and Kojimaa 40

0.00 ¯ 33 Chryssikos et al.b


30
0.02 ¯ 35 Chryssikos et al.b
Model Chryssikos
0.06 ¯ 37 Chryssikos et al.b 20
0.08 ¯ 36 Chryssikos et al.b 0 0.1 0.2 0.3 0.4 0.5

Chryssikos et al.b x
0.10 634 37 (b)
0.12 ¯ 37 Chryssikos et al.b
0.15 699 ¯ Chryssikos et al.b FIG. 4. 共Color online兲 Composition dependence of fragility in the 共a兲
0.16 ¯ 44 Chryssikos et al.b sodium borate and 共b兲 lithium borate systems. The solid lines show the
0.18 ¯ 47 Chryssikos et al.b predicted m共x兲 using the topological model with temperature-dependent
constraints. Experimental data points 共Refs. 79, 80, and 83兲 are provided in
0.19 ¯ 51 Chryssikos et al.b
Tables I and II.
0.20 752 60 Chryssikos et al.b
0.21 ¯ 55 Chryssikos et al.b
0.25 770 61 Chryssikos et al.b ⌬F␣ⴱ = − kT␣ ln共1 − 2−1/␯tobs兲. 共24兲
0.30 770 ¯ Chryssikos et al.b
This leaves us with one unknown parameter, ␯tobs, which can
0.35 757 ¯ Chryssikos et al.b
be adjusted to study the cooling rate dependence of Tg共x兲 and
0.48 699 ¯ Chryssikos et al.b
other properties. Since we are concerned with composition
a
Reference 82. rather than cooling rate dependence, we consider a constant
b
Reference 83.
value of ␯tobs = 103 for all calculations in this paper. The
value of ␯tobs = 103 is obtained by fitting to the published
topological models of glass-forming systems. Indeed, this value of fragility for pure B2O3, m共0兲 = 32 共our reference
should be a universal phenomenon since iso-Tg regimes can composition兲.86 Additional investigation is required to deter-
occur in any glass-forming system when Tg共x兲 reaches some mine a more precise form of Eq. 共9兲.
constraint onset temperature. Combining Eq. 共11兲 with the Tg共x兲 in Eq. 共23兲, we can
calculate the fragility numerically. As shown in Fig. 4, the
qualitative trend of the computed m共x兲 is in good agreement
IV. FRAGILITY with published experimental values. Because of the large
values of fragility for alkali-rich compositions, there is con-
As we have just shown, the discrete form of q␣共T兲 in Eq. siderable uncertainty in both the experimental and theoreti-
共10兲 is convenient for obtaining accurate analytical forms of cally calculated values of fragility. It is interesting to note
Tg共x兲. In practice, we find little difference between analytical that m共x兲 continues to change during the flat Tg共x兲 = T␥ re-
calculations of Tg共x兲 using the discrete form of q␣共T兲 and gime owing to the changing slope of f关Tg共x兲 , x兴 as the num-
numerical calculations of Tg共x兲 using the continuous form of ber of each species evolves with x. Fragility peaks at the
q␣共T兲 in Eq. 共9兲. However, calculations of fragility involve diborate composition, x = 1 / 3, which marks the onset of non-
taking the derivative of f共T , x兲 and hence require use of the bridging oxygen.
continuous q␣共T兲. When converting between discrete and
V. DISCUSSION
continuous forms of q␣共T兲, the free energy activation barrier,
⌬F␣ⴱ in Eq. 共9兲, can be computed directly from the onset In this work we have generalized the notion of bond
temperature, T␣ in Eq. 共10兲, by constraints for finite temperature conditions. Using the well-

Downloaded 15 Jun 2009 to 199.197.135.217. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
234503-7 Topological model of alkali borates J. Chem. Phys. 130, 234503 共2009兲

established Adam–Gibbs theory45 and the recent energy land- not limited to these two parameters. In fact, a similar ap-
scape modeling of Naumis,53,54 we relate the temperature- proach could be used to calculate the composition depen-
dependent floppy mode density to the shear viscosity, dence of other viscosity reference points such as the strain
␩共T , x兲, obtaining a simple explanation for the striking uni- point and softening point. With this approach the entire equi-
versal scaled viscosity plot of Angell.42 Our model is clearly librium viscosity curves of these systems can be computed.
a simplification of a complex fluid; to understand its appli- Our approach gives a natural explanation for the con-
cability and limitations it is useful to examine the assump- stant Tg共x兲 regions observed in the alkali borate systems,
tions involved. since the condition of Eq. 共6兲 can only be satisfied with a
First, we have taken simplifying expressions for the tem- “horizontal jump” in Tg共x兲 as in Fig. 4共b兲. If Tg共x兲 were to
perature dependence of the bond constraints, q␣共T兲. We have increase above the constraint onset temperature, the network
employed two distinct forms for q␣共T兲, the continuous form would become more floppy, driving the Tg共x兲 downward.
in Eq. 共9兲 derived from a simplified energy landscape However, when Tg共x兲 falls below the constraint onset tem-
approach55 and the discrete form in Eq. 共10兲 employing a perature, the network becomes more rigid, driving the Tg共x兲
unit step approximation. With the discrete model, a bond upward. The net result is a region of constant Tg共x兲 in com-
constraint becomes rigid suddenly as the temperature is de- position space. This should be a general phenomenon for any
creased below a certain onset temperature T␣. On the other glass-forming liquid when the Tg共x兲 reaches a constraint on-
hand, in the continuous model a bond constraint stiffens set temperature.
gradually over a range of temperatures. While the discrete
form of q␣共T兲 is useful for deriving analytical expressions for
VI. CONCLUSIONS
the composition dependence of glass transition temperature,
Tg共x兲, the continuous form must be used for accurate calcu- Topological modeling of glass networks is useful for
lation of fragility, m共x兲. In practice, we have found that there predicting trends in thermal and mechanical properties. We
is little difference between computed values of Tg共x兲 using have extended the concept of rigidity to finite temperatures
the discrete and continuous forms of q␣共T兲, but the fragility and related it to a temperature-dependent floppy mode den-
calculations are sensitive to the tails of the q␣共T兲 functions. sity. The floppy mode density is used to calculate configura-
Additional work is required to assess the validity of Eq. 共9兲 tional entropy associated with atomic rearrangements and to
and its application to the sub-Tg regime where thermal his- determine viscosity in the vicinity of the glass transition tem-
tory plays a strong role. perature. Our modeling approach allows for derivation of
Second, the counting of constraints from medium- and simple analytical expressions for the scaling of glass transi-
long-range interactions has been approximated by counting tion temperature with composition, Tg共x兲, and provides a
over the nearest neighbors only. This needs further examina- natural explanation for horizontal jumps in Tg共x兲, a possible
tion, particularly in relation to the reported “intermediate cause of phase separation behavior. Fragility, m共x兲, can be
phase” behavior where medium-range ordering is of critical computed using a straightforward temperature derivative of
importance.87,88 Boolchand and co-workers89–92 demon- the floppy mode density.
strated a link between the self-assembled intermediate phase Our approach, accounting for the temperature-dependent
and nanoscale phase separation. The iso-Tg phenomenon re- nature of constraints, enables the calculation of Tg共x兲 and
ported in this paper may also be indicative of such nanoscale m共x兲 for a variety of systems, specifically demonstrated in
phase separation. this paper for the binary alkali borate system. The Tg共x兲 cal-
Third, in the calculation of Tg共x兲, we have assumed that culation is insensitive to the particular form of q␣共T兲 and
the parameter B共x兲 of the Adam–Gibbs relation is constant demonstrates good quantitative agreement with experiment.
with respect to composition. This assumption may not be On the other hand, the m共x兲 calculation depends on the de-
strictly valid, especially for application over a wide range of tails of q␣共T兲 and is 共currently兲 useful for determining quali-
compositions, and hence the dependence of B共x兲 on compo- tative trends. A more accurate model of q␣共T兲 is necessary to
sition should be explored more thoroughly from both theo- obtain a quantitatively accurate model of m共x兲.
retical and experimental perspectives. This work represents an intermediate level of approach
Fourth, we have neglected temperature-dependent between traditional atomistic simulation techniques such as
changes in the short-range atomic level structure of the liq- molecular dynamics and other more empirical macroscopic
uid, such as the temperature dependence of four-coordinated modeling. While the topological modeling approach does not
boron in alkali borate glasses.57 capture the minute level of detail as molecular dynamics, it
Despite these limitations and assumptions, we have focuses on the most relevant aspects of the interactions to
shown that the calculated composition dependence of glass enable accurate prediction of property scaling with composi-
transition temperature and fragility agrees remarkably well tion. We do not, however, consider the modeling approach of
with experimental measurements across a wide range of al- this paper as a replacement for atomistic simulation tech-
kali content. The substantial agreement between our model niques such as molecular dynamics. Rather, we consider both
and the experimental measurements indicates that the above approaches as providing critical pieces of a larger whole.
assumptions are reasonable. While not capturing every detail Only when systems are modeled on multiple scales does the
of these complex liquids, the topological model in this paper complete nature of glass-forming systems become clear. In
apparently captures the most important aspects governing particular, molecular dynamics can be useful for providing
Tg共x兲 and m共x兲. We also note that our modeling approach is key inputs into a topological model, such as identifying the

Downloaded 15 Jun 2009 to 199.197.135.217. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp
234503-8 Mauro, Gupta, and Loucks J. Chem. Phys. 130, 234503 共2009兲

43
basic structural units of the glass and ordering the strengths J. C. Mauro and R. J. Loucks, Phys. Rev. E 78, 021502 共2008兲.
44
P. K. Gupta and J. C. Mauro, Phys. Rev. E 78, 062501 共2008兲.
of the various constraints. Indeed, such knowledge is a pre- 45
G. Adam and J. H. Gibbs, J. Chem. Phys. 43, 139 共1965兲.
requisite for using our topological modeling approach. 46
G. W. Scherer, J. Am. Ceram. Soc. 67, 504 共1984兲.
47
Y. Bottinga and P. Richet, Chem. Geol. 128, 129 共1996兲.
48
ACKNOWLEDGMENTS M. D. Ediger, C. A. Angell, and S. R. Nagel, J. Phys. Chem. 100, 13200
共1996兲.
We gratefully acknowledge valuable discussions with 49
R. Richert and C. A. Angell, J. Chem. Phys. 108, 9016 共1998兲.
50
Doug Allan, Roger Araujo, Adam Ellison, T. J. Kiczenski, J.-P. Bouchaud and G. Biroli, J. Chem. Phys. 121, 7347 共2004兲.
51
and Amy Rovelstad. D. Giordano, J. K. Russell, and D. B. Dingwell, Earth Planet. Sci. Lett.
271, 123 共2008兲.
52
1 M. J. Toplis, Am. Mineral. 83, 480 共1998兲.
A. K. Varshneya, Fundamentals of Inorganic Glasses, 2nd ed. 共Society of 53
G. G. Naumis, Phys. Rev. E 71, 026114 共2005兲.
Glass Technology, Sheffield, 2006兲. 54
2 G. G. Naumis, J. Non-Cryst. Solids 352, 4865 共2006兲.
J. Zarzycki, Glasses and the Vitreous State 共Cambridge University Press, 55
P. K. Gupta and J. C. Mauro, J. Chem. Phys. 126, 224504 共2007兲.
Cambridge, 1990兲. 56
3 J. C. Mauro, P. K. Gupta, and R. J. Loucks, J. Chem. Phys. 126, 184511
D. R. Uhlmann and N. J. Kreidl, Glass: Science and Technology
共Academic, New York, 1983兲. 共2007兲.
57
4
P. B. Macedo and A. Napolitano, J. Res. Natl. Bur. Stand., Sect. A 71, R. J. Araujo, J. Non-Cryst. Solids 42, 209 共1980兲.
58
231 共1967兲. R. J. Araujo, J. Non-Cryst. Solids 46, 217 共1981兲.
59
5
O. V. Mazurin, J. Non-Cryst. Solids 351, 1103 共2005兲. R. J. Araujo, J. Non-Cryst. Solids 58, 201 共1983兲.
60
6
A. Fluegel, J. Am. Ceram. Soc. 90, 2622 共2007兲. P. J. Bray, J. Non-Cryst. Solids 75, 29 共1985兲.
61
7
N. P. Lazarev, A. S. Bakai, and C. Abromeit, J. Non-Cryst. Solids 353, R. Kerner, J. Non-Cryst. Solids 135, 155 共1991兲.
62
3332 共2007兲. O. Majérus, L. Cormier, G. Calas, and B. Beuneu, Phys. Rev. B 67,
8
V. Lubchenko and P. G. Wolynes, Annu. Rev. Phys. Chem. 58, 235 024210 共2003兲.
63
共2007兲. T. Yano, N. Kunimine, S. Shibata, and M. Yamane, J. Non-Cryst. Solids
9
M. C. C. Ribeiro, T. Scopigno, and G. Ruocco, J. Chem. Phys. 128, 321, 157 共2003兲.
64
191104 共2008兲. T. J. Kiczenski, L.-S. Du, and J. Stebbins, J. Non-Cryst. Solids 351, 3571
10
U. R. Pedersen, T. Christensen, T. B. Schrøder, and J. C. Dyre, Phys. Rev. 共2005兲.
65
E 77, 011201 共2008兲. L. Cormier, O. Majérus, D. R. Neuville, and G. Calas, J. Am. Ceram.
11
J. C. Mauro and R. J. Loucks, Phys. Rev. B 76, 174202 共2007兲. Soc. 89, 13 共2006兲.
12
L. Wondraczek and J. C. Mauro, J. Eur. Ceram. Soc. 29, 1227 共2009兲. 66
S. A. Brawer and W. B. White, J. Chem. Phys. 63, 2421 共1975兲.
13
P. K. Gupta and J. C. Mauro, J. Chem. Phys. 130, 094503 共2009兲. 67
T. F. Soules, J. Chem. Phys. 71, 4570 共1979兲.
14
J. C. Phillips, J. Non-Cryst. Solids 34, 153 共1979兲. 68
A. A. Tesar and A. K. Varshneya, J. Chem. Phys. 87, 2986 共1987兲.
15
J. C. Phillips and M. F. Thorpe, Solid State Commun. 53, 699 共1985兲. 69
C. Huang and A. N. Cormack, J. Chem. Phys. 93, 8180 共1990兲.
16
M. F. Thorpe, J. Non-Cryst. Solids 57, 355 共1983兲. 70
T. F. Soules and A. K. Varshneya, J. Am. Ceram. Soc. 64, 145 共1981兲.
17
H. He and M. F. Thorpe, Phys. Rev. Lett. 54, 2107 共1985兲. 71
A. N. Cormack and B. Park, Phys. Chem. Glasses 41, 272 共2000兲.
18
Y. Cai and M. F. Thorpe, Phys. Rev. B 40, 10535 共1989兲. 72
M. A. González, C. Mondelli, G. D’Angelo, C. Crupi, and M. R.
19
M. F. Thorpe, J. Non-Cryst. Solids 182, 135 共1995兲. Johnson, J. Non-Cryst. Solids 354, 203 共2008兲.
20
A. R. Cooper, Phys. Chem. Glasses 19, 60 共1978兲. 73
P. K. Gupta, Proceedings of the International Congress on Glass, New
21
P. K. Gupta and A. R. Cooper, J. Non-Cryst. Solids 123, 14 共1990兲. Delhi, India, 1986 共unpublished兲, p. 1.
22
P. K. Gupta, J. Am. Ceram. Soc. 76, 1088 共1993兲. 74
R. L. Mozzi and B. E. Warren, J. Appl. Crystallogr. 3, 251 共1970兲.
23
P. K. Gupta, in Rigidity Theory and Applications, edited by M. F. Thorpe 75
I. Avramov, J. Non-Cryst. Solids 351, 3163 共2005兲.
and P. M. Duxbury 共Kluwer, New York, 1999兲, p. 173. 76
24 A. Napolitano, J. H. Simmons, D. H. Blackburn, and R. E. Chidester, J.
P. K. Gupta and D. B. Miracle, Acta Mater. 55, 4507 共2007兲.
25 Res. Natl. Bur. Stand., Sect. A 78, 323 共1974兲.
J. Grothaus and P. Boolchand, J. Non-Cryst. Solids 72, 1 共1985兲. 77
26 C. A. Angell and D. L. Smith, J. Phys. Chem. 86, 3845 共1982兲.
B. L. Halfpap and S. M. Lindsay, Phys. Rev. Lett. 57, 847 共1986兲. 78
27 C. T. Moynihan, J. Am. Ceram. Soc. 76, 1081 共1993兲.
W. Bresser, P. Boolchand, and P. Suranyi, Phys. Rev. Lett. 56, 2493 79
S. V. Nemilov, Neorg. Mater. 2, 349 共1966兲.
共1986兲. 80
28 S. V. Stolyar, V. P. Klyuev, and A. V. Bulaeva, Fiz. Khim. Stekla 10, 447
B. Norban, D. Pershing, R. N. Enzweiler, P. Boolchand, J. E. Griffiths,
and J. C. Phillips, Phys. Rev. B 36, 8109 共1987兲. 共1984兲.
81
29
I. Zitkovsky and P. Boolchand, J. Non-Cryst. Solids 114, 70 共1989兲. S. Suzuki, T. Kobayashi, M. Takahashi, and M. Imaoka, J. Ceram. Soc.
30
D. R. Swiler, A. K. Varshneya, and R. M. Callahan, J. Non-Cryst. Solids Jpn. 89, 252 共1981兲.
82
125, 250 共1990兲. M. Kodama and S. Kojima, J. Therm. Anal. Calorim. 69, 961 共2002兲.
83
31
A. N. Sreeram, A. K. Varshneya, and D. R. Swiler, J. Non-Cryst. Solids G. D. Chryssikos, J. A. Duffy, J. M. Hutchinson, M. D. Ingram, E. I.
130, 225 共1991兲. Kamitsos, and A. J. Pappin, J. Non-Cryst. Solids 172–174, 378 共1994兲.
84
32
A. N. Sreeram, D. R. Swiler, and A. K. Varshneya, J. Non-Cryst. Solids R. R. Shaw and D. R. Uhlmann, J. Am. Ceram. Soc. 51, 377 共1968兲.
85
127, 287 共1991兲. I. G. Polyakova, Phys. Chem. Glasses 41, 247 共2000兲.
86
33
A. N. Sreeram, A. K. Varshneya, and D. R. Swiler, J. Non-Cryst. Solids R. Böhmer, K. L. Ngai, C. A. Angell, and D. J. Plazek, J. Chem. Phys.
128, 294 共1991兲. 99, 4201 共1993兲.
34 87
U. Senapati and A. K. Varshneya, J. Non-Cryst. Solids 185, 289 共1995兲. Y. Wang, P. Boolchand, and M. Micoulaut, Europhys. Lett. 52, 633
35
U. Senapati and A. K. Varshneya, J. Non-Cryst. Solids 197, 210 共1996兲. 共2000兲.
88
36
U. Senapati, K. Firstenberg, and A. K. Varshneya, J. Non-Cryst. Solids M. Micoulaut and J. C. Phillips, Phys. Rev. B 67, 104204 共2003兲.
89
222, 153 共1997兲. P. Boolchand, D. G. Georgiev, and B. Goodman, J. Optoelectron. Adv.
37
A. K. Varshneya, J. Non-Cryst. Solids 273, 1 共2000兲. Mater. 3, 703 共2001兲.
38 90
J. C. Mauro and A. K. Varshneya, J. Am. Ceram. Soc. 90, 192 共2007兲. P. Boolchand, D. G. Georgiev, and M. Micoulaut, J. Optoelectron. Adv.
39
C. A. Angell, J. Non-Cryst. Solids 73, 1 共1985兲. Mater. 4, 823 共2002兲.
40 91
C. A. Angell, J. Non-Cryst. Solids 102, 205 共1988兲. D. G. Georgiev, P. Boolchand, H. Eckert, M. Micoulaut, and K. Jackson,
41
C. A. Angell, J. Non-Cryst. Solids 131–133, 13 共1991兲. Europhys. Lett. 62, 49 共2003兲.
42 92
C. A. Angell, K. L. Ngai, G. B. McKenna, P. F. McMillan, and S. W. P. Boolchand, M. Jin, D. I. Novita, and S. Chakravarty, J. Raman Spec-
Martin, J. Appl. Phys. 88, 3113 共2000兲. trosc. 38, 660 共2007兲.

Downloaded 15 Jun 2009 to 199.197.135.217. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/jcp/copyright.jsp

You might also like