Anda di halaman 1dari 52

10

Glass Formation and Glassy


Behavior
GREGORY B. McKENNA
National Bureau of Standards, Gaithersburg, MO, USA

10.1 INTRODUCTION 311


10.2 THERMODYNAMICS OF THE GLASS TRANSITION 314
10.2.1 The Kauzmann Paradox 314
10.2.2 Thermodynamic Transitions and Glasses 315
10.2.2.1 Definitions offirst- and second-order transitions 315
10.2.2.2 Pressure dependence of the transition temperature 315
10.2.2.3 Ordering parameter description of glassy thermodynamics 322
10.2.3 Models of the Glass Transition 322
10.2.3.1 Gibbs-DiMarzio theory 323
10.2.3.2 Free-volume concepts 329
10.3 KINETICS OF GLASS-FORMING SYSTEMS 340
10.3.1 Introduction 340
10.3.2 Phenomenology 341
10.3.2.1 Kinetics of glass formation 341
10.3.2.2 Kinetics in the glassy state 342
10.3.3 Phenomenological Equations 344
10.3.3.1 Requirements of the models 344
10.3.3.2 The Kovacs-Aklonis-Hutchinson-Ramos (KAHR) model 344
10.3.4 Free Volume and the Kinetics of Glasses 352
10.3.4.1 Introduction 352
10.3.4.2 The Robertson-Simha-Curro (RSC) model 353
10.3.5 Computer Simulations 356
10.3.6 Physical Aging 357
10.4 REFERENCES 358

10.1 INTRODUCTION
The scientific study of glass-forming substances has been important for well over half a century.
The continuing interest in the behavior of these materials was confirmed by a simple computer
search of the literature from 1968 to 1987 on 'glasses', which resulted in over 125000 citations, and
'polymer glasses', which resulted in over 10000 citations.' A further glimpse of the importance of
glasses as an area of scientific inquiry can be obtained by browsing through the books, special
journal issues and symposium proceedings which have appeared over the past decade and half.2 - 10
In spite of the large body of literature which has accumulated on glasses, there is still much room
for increased understanding of the physics of glass formation and the nature of the glassy state.
There are underlying thermodynamic questions which are still unanswered, such as: 'Is the glass
transition a true thermodynamic transition (second-order transition) or is it a kinetic phenomenon
which saves the thermodynamic "catastrophe"?'. Such questions will be addressed in the ap-
propriate sections of this chapter. First, however, we present some of the phenomenology of the glass
transition itself.
The basic event of the transition from liquid-like to glass-like behavior in glass-forming sub-
stances, including polymers, is observed to be a kinetic phenomenon under ordinary experimental
conditions.v"" For example, as depicted in Figure 1, when isochronal volume measurements are

311
312 Glass Formation and Glassy Behavior

--48°,/7-
I
6


85
J 4

~
'0-
rt>

~
:=.
E I
ci.
fiJ.d/~
~v.1
2

)(
<:t
a

N
)(
/ dl Q

0.//,~
Q 84 0
~ I
,~ II I
, I I
rlr/ /
g/100h I
/1I II -2

g/
/ I T; Tg
/
83
-25
T (OC)

Figure 1 Specific volume v vs. temperature T for a poly(vinyl acetate) polymer. Plot shows effect of duration of experiment
on glass temperature, i.e. T~ = Tg (100 h) < Tg (0.02 h). Also depiected is ex vs. T showing the discontinuity in thermal
expansion coefficient at Tg (after ref. 11, with permission)

taken at different temperatures (at constant pressure), the volume-temperature line for the material
begins to depart from the equilibrium line at a temperature at which the characteristic time! for the
molecular motions leading to volume recovery is longer than the time scale of the experiment.
Therefore, as the time scale of measurement increases, the position of the departure from equilibrium
moves towards lower temperatures. As shown in Figure 1, the intersection of the extrapolated glassy
and liquid lines defines the glass transition temperature Tg or T~ at the relevant annealing time.
Although the above description is a kinetic one phenomenologically, there is a change in the
thermodynamics associated with the change from glassy behavior to liquid behavior. In Figure 2 it is
shown how the heat capacity of a polymer glass changes upon going through the glass transition

200

'.- •
180
T
0
e
T
~
J

~ 160

320

T(K)

Figure 2 C p vs. T for polystyrene showing the specific heat discontinuity at Tg • The points represent data for different
samples and thermal histories (after ref. 12, with permission)
Glass Formation and Glassy Behavior 313

temperature. Interestingly, although T g is a function of the experimental time-scale, the value of the
change of the heat capacity f,.Cp is not. 1 3 Similar behavior is found for f,.a, the change in the
coefficient of thermal expansion in going from the liquid to glassy states (see Figure 1). The quasi-
discontinuous behavior of C p and a at T g suggests that, although the glass transition itself is a kinetic
event under ordinary experimental conditions, there is an underlying thermodynamic transition, as
discussed subsequently.
Returning to the kinetics, when a liquid is cooled to below the glass transition temperature and
the cooling is arrested at some temperature T, the volume departure from equilibrium <5 [= (v
- v<xJ/v oo ] of the non-equilibrium glass evolves spontaneously towards equilibrium, as depicted in
Figure 3 (where v is the specific volume of the glass and V oo the value in equilibrium). Associated with
these changes in glass 'structure', one observes changes in the mechanical properties of the glass.!"
As shown in Figure 4, the creep compliance of a glassy polymer shifts along the time axis with aging
time t e after a quench from above to below T g • This process, whereby the mechanical properties
change as the volume of the glass recovers, is known as physical aging!" and it affects both the linear
and non-linear viscoelastic behavior as well as the ultimate properties of the glass.l": 15
The above paragraphs cover what the following sections of this chapter will describe in some
detail. We will examine first the thermodynamics of the glass transition, because an understanding
of the equilibrium state is essential to any understanding of the non-equilibrium state.l" This
discussion will include both the configurational-entropy model, which is a true thermodynamic
model, as well as the free-volume models which allow prediction of quasi-static thermodynamic

..
5
-.~

-.-.-.-.-.......'... ..-...----.
- ~
4 ,~

- .--.---.
....... """..
'-'9.e oc
3 ....

_.__
-.
....'....
..... ..... 24.9
.......
.. .. 22.4
A_
-~..
., "-,

.,e.
2 . -•.•-.. '••-. ,.. • ..........

..... e.. •.•


..... . . . . ..
-.
......... 27.5 <,

--.--._..... 30....
....... ... ~ ........

35
C -.-.-.-
.-.-.-.
_ _ _ 32.5

......-..
..,

.•••.
, ~..~ ~
-
..
~
--

.. -.
ti.

0.001 0.1 I 10 100

Figure 3 Isothermal contraction of glucose after quenching from To = 40 DC to different temperatures T, as indicated (after
ref. 13, with permission)

4.5

Aging time, te(doys)

0.03 0.1 3 10 30 100 300



Q)
o
c
4.0 /
/

.~
Q.
./
E
o ./
U
Q. 3.5
Q)
Q)
~ ~2%
u
.!
fI)
c
~

Creep time (s)

Figure 4 Creep compliance ofpoly(vinyl chloride) quenched from 90 DC (~ + 10 DC) to 20 DC at various aging times after the
quench (after ref. 14, with permission)
314 Glass Formation and Glassy Behavior

properties. The discussion of free volume will include a description of the models used to describe the
equilibrium behavior of a liquid near to the glass transition. We will then examine the kinetics of
glasses. The discussion of the kinetics will emphasize the phenomenological equations l 7 - 2 1 devel-
oped recently, which have been highly successful in describing the kinetics of the glass in the non-
equilibrium state. Finally, we will discuss briefly the phenomenon of physical aging.!" which has
proved to be of much interest recently and which has important technological implications.

10.2 THERMODYNAMICS OF THE GLASS TRANSITION


10.2.1 The Kauzmann Paradox
Much of what has been written about the thermodynamics of glasses revolves around the question
of how to resolve the Kauzmann P paradox. Kauzmanrr'? examined the thermodynamic behavior
of supercooled glass-forming liquids by extrapolating the equilibrium properties, e.g. volume
(cf Figure 1),enthalpy, entropy, etc., to low temperatures. He found that 'a very startling result' was
obtained. 'Not very far below the glass-formation temperature, but still far above 0 K, the
extrapolated entropy (or other property) of the liquid becomes less than that of the crystalline solid.'
Figure 5 shows the entropy differences between the supercooled liquid and crystalline phases for
several low molecular weight glass-formers after Kauzmann's original work.V (It is interesting to
note that for boron trioxide (B2 0 3 ) the entropy difference does not extrapolate to negative values.)

1.0

0.8

~
0.6
Ethanol ... ~.....

,
(f)E
<J ~ ··7 ...
(f)
<J
.....•......
···········G·I~cerol / >.
/Propanol..··

0.4 ~., Gi~~~~~· ....


/ ....>.. I
/ Lactic acid ....
/ ····························~····1
0.2
/ ;
/ i
/ i
/ i
o 0.2 0.4 0.6

Figure S Differences in entropy between supercooled liquid and crystalline phases for different glass-forming materials, as
indicated. ~S / ~Sm = difference in entropy expressed as fraction of entropy of fusion; T/ Tm = temperature expressed as
fraction of melting temperature (after ref. 22)

Kauzmann argued that the appropriate comparison to be made is between the properties of a
thermodynamically stable crystalline phase and a metastable glassy phase. His arguments imply that
there is no 'thermodynamic' glass transition. He extended a liquid model by Mott and Gurney P to
explain glass formation in terms of the entropy due to the orientational configurations of 'tiny'
crystallites which make up the liquid state, which contributes to the appearance of a 'pseudo-critical'
temperature T K' at which the rate of crystallization of the supercooled liquid becomes slow enough
that glass formation occurs over ordinary experimental time-scales. The model was crude and
quantitatively incorrect; however, it did correctly portray the trend in the thermodynamic properties
of the supercooled liquid discussed previously, i.e. the liquid properties become smaller than the
crystal-state properties above 0 K.
Glass Formation and Glassy Behavior 315

10.2.2 Thermodynamic Transitions and Glasses


10.2.2.1 Definitions offirst- and second-order transitions
Kauzmann P pointed out the difficulties involved in making a thermodynamic interpretation of
the glass transition phenomenon and opted for a kinetic solution to the apparent paradox discussed
above. However, the question of whether or not the glass transition is a true second-order transition
is still one which is widely discussed in the literature. Before describing the two most prominent
models of the glass transition, i.e. the Gibbs-DiMarzi0 24 configurational-entropy model and the
free-volume models,25-29 we will attempt to clarify a point of confusion which arises from attempts
to experimentally prove or disprove the glass transition event as a true thermodynamic transition. We
remind the reader that, regardless of the underlying thermodynamics, the experimentally observed
glass transition is a kinetic phenomenon.
According to Ehrenfest 31, a first-order transition is one for which the free energy as a function of
any given state variable (V, P, T) is continuous, but the first partial derivatives of the free energy with
respect to the relevant state variables are discontinuous. Thus, if the Gibbs free energy G at the
transition temperature is continuous, but (iJG/iJT)p and (iJG/iJP)T are discontinuous, we have a first-
order transition. At a typical first-order transition point, such as fusion or vaporization, there is a
discontinuity in entropy S, volume V and enthalpy H, since 32

[:~l = -s (1)

[:~l v (2)

[a(G/T)] = H (3)
a(I/T) p

(4)
T

- KV (5)

a [[a(G/T)] ] (6)
aT a(I/T) p p

= aV (7)

Some typical phenomena which have been described as second-order transitions are order-
disorder transitions in metal alloys, onset of ferromagnetism, onset of ferroelectricity and onset of
superconductivity.32,34,35
In Figure 6 are depicted schematically the behaviors of the Gibbs free energy function and its first
and second derivatives for first-order, second-order and glass transitions.32,34,35 The important
thing to note from this figure is that, formally, the glass transition event appears to be a second-order
transition. However, the nature of the glass transition is somewhat different from what is normally
observed and classified as a second-order transition. Interestingly, the kinetics involved in the
transitions differ, e.g. for glasses, an increase in the time scale of the experiment reduces the observed
transition temperature, while for other second-order transitions this is not the case."

10.2.2.2 Pressure dependence of the transition temperature


F or a first-order transition from a phase (X to a phase p, the continuity of the free energy function
and the equilibrium condition result in the well-known Clausius-Clapeyrorr'" relation (equation 8)

PS 2-K·
316 Glass Formation and Glassy Behavior

(a) (b) (e)

G G G

Melt
~r r., J;

(I)
i
/
I
I
V) V)
:t:. ---I
~ I :t:" II
~ .. -: /1
~ / 1
I

r; Trr r:

I
~ I ~ ~

~~
r-- <.J~ <.J~
I 0 0
e I

7;r r T.,r T To r

Figure 6 Schematic representation of changes with temperature of the free energy and its first and second derivatives for:
(a) first order, (b) classical second-order and (c) glass transitions (after ref. 35, with permission)

~H
(8)
T~V

where the t, refers to the transition line.


For a second-order transition, we will show the development of two similar relationships, one
which results from the volume continuity at the transition and the other from the entropy continuity.
In the first instance, we can write that the total differential for the volume in the liquid I and the glass
g are the same.
(9)

Regrouping, and because V; = Vg at the transition, we can write

[ddP'r
TJ ~K
= ~(X
(10)

Similarly, for the entropy, we write

C C
dS. = ~dT - (X.~dP = dSg = --!!dT - (Xg VgdP (11)
T T

Regrouping terms, and noting that Sg = S. and Vg = V; at the transition

[::1 =
(12)
Glass Formation and Glassy Behavior 317

Expressions (10) and (12) will always be true for any material which undergoes an apparent
second-order transition. Furthermore, if the volume and entropy surfaces for the transition are the
same, as they must be thermodynamically, (10) and (12) can be equated and we arrive at the
Prigogine-Defay'" ratio, R
~.KACp
R = = 1 (13)
TV(Aex)2

Together, equations (10), (12) and (13) have led to much discussion over the thermodynamics of
the glass transition event. This is because, depending upon how the experiments are carried out, one
can attain agreement, i.e. R = 1, or not with the prediction for a second-order thermodynamic
transition. In this author's view, much of the discussion is unnecessary because the experiments
measure a kinetic phenomenon (perhaps a manifestation of an underlying thermodynamic event)
and, therefore, cannot test whether or not there is a true thermodynamic glass transition.
In the discussion which follows, we examine the thermodynamic data which have been used to
'test' the validity of relations (10), (12) and (13) and describe how the lack of agreement arises only
when 'mixed' data are used. By mixed, we refer to data obtained on glasses with different formation
histories. This effect has been recognized by other workers,37-39 but we emphasize it here because
we feel that experiments other than PVT or SPT types should be developed in order to define the
thermodynamics of glasses. Such experiments will be described when we discuss theoretical models
of glasses.
Typical PVT data for a glass-forming polymer are shown in Figure 7. These data were obtained
by isobarically cooling the polymer liquid (melt) from above to below the glass transition tempera-
ture at a constant rate. Numerical fitting of the data allowed computation of the appropriate
thermodynamic parameters, i.e. ~Kl' ~(Xl and dT; /dP for these!" 'variable-formation glasses'. In
Figure 8 is depicted a similar plot for glasses formed at atmospheric pressure (1 atm, 101 325 Pal,
'constant-formation glasses'. In this instance, all pressure changes occurred in the glassy state after
cooling to the appropriate temperatures. Again ~K2' ~(X2 and dT; /dP were calculated from these
data. There are two points which can be made. First, the P VT data of Figures 7 and 8 are not the
same, indicating a path dependence of the P VT surface. Second, within each set of data, i.e. Figure 7
or Figure 8, dTg/dP = ~K/~(x, because of the tautological development of this relationship (equation
10). However, dT;/dP =1= dT;/dP. This is shown in Figure 9, which also includes data for a glass
formed at 800 bar (8 x 107 Pal (constant-formation glass). Interestingly, dTg/dP for the latter is

0.89

0.88

0.87

0.86

T 0.85
0-
ro
E
~ 0.84
:::.

0.83

0.82

0.81

0.80

-30 -20 -10 0 20 30 40 50 60 70 80 90 100

Figure 7 Specific volume v of poly(vinyl acetate) vs. temperature T at different pressures, as indicated, for variable-formation
glass; note definition of Tg(P) for this type of glass (after ref. 37, with permission)
318 Glass Formation and Glassy Behavior

0.89

0.88

0.87

0.86

'-0-
0.85

0.84
e
f'()

2
~
0.83

0.82

0.81

0.80

-30 -20 -10 0 10 20

Figure 8 Specific volume v of poly(vinyl acetate) vs. temperature T at different pressures, as indicated, for isobaric glass
formed at 1 atm; note definition of T;(P, 0) for this type of glass (after ref. 37, with permission)

70

10

Figure 9 Transition map showing Tg and T; vs. pressure P for different formation histories: (a) variable formation (P' = P),
as in Figure 7; (b) isobaric glass formed at 1 atm (P' = 0), as in Figure 8; and (c) isobaric glass formed at 800 bar
(P' = 0.8 kbar). Note that dTg(P)/dP < dT;(P, O)/dP ~ dT;(P, 0.8) (after ref. 37, with permission)

approximately equal to that of the glass formed at atmospheric pressure. McKinney and
Goldstein's?" data for the three glasses are shown in Table 1.
The fact that dTg/dP depends upon the path of glass formation shows that if one mixes
measurements on glasses formed differently then one can expect an apparent violation of the
tautology represented by equation (10). Thus it is no surprise that there are conflicting reports in the
literature concerning its validity. For example, Zoller 4 0 finds that dTg/dP = AK/A~ for four different
polymers. His data were obtained for variable-formation glasses. On the other hand, Oels and
Rehage 4 1, 4 2 show that dTg/dP =F AK/A~. They would have found a contradiction within their own
data had they carried out an analysis of their PVT data, obtained for variable-formation glasses.
However, they obtained their values for AK by direct measurement of K on glasses cooled isobarically
to appropriate temperatures followed by incremental changes in pressure. Although this is not the
Table 1 Tabulation of Thermodynamic Parameters for Poly(vinyl acetate) Glasses Formed under Different Conditions

Variable-formation glass
p r, v a l x 104 a g 'X 104 J1.(X X 104 1'1 X 105 1'; X 10 5 J1.K* X 105 dTg/dP T;J1.(X
(bar) eC) (cm' g-l) eC- l ) eC- l ) eC- l ) (bar-I) (bar-I) (bar-I) (OC bar-I)

0 31.50 0.84358 7.118 3.031 4.088 4.990 3.902 1.088 0.0266 0.1245
100 34.08 0.84098 6.997 3.000 3.996 4.838 3.834 1.003 0~0251 0.1228
200 36.53 0.83840 6.850 2.969 3.881 4.690 3.766 0.924 0.0238 0.1202
300 38.85 0.83585 6.685 2.937 3.748 4.543 3.696 0.848 0.0226 0.1169
400 41.05 0.83334 6.509 2.904 3.606 4.397 3.624 0.773 0.0214 0.1133
500 43.13 0.83085 6.328 2.869 3.459 4.249 3.552 0.697 0.0202 0.1094
600 45.08 0.82839 6.148 2.834 3.314 4.098 3.478 0.620 0.0187 0.1055
700 46.86 0.82596 5.973 2.797 3.176 3.942 3.403 0.539 0.0170 0.1016
800 48.46 0.82355 5.807 2.759 3.048 3.780 3.326 0.454 0.0149 0.0980 C)
~
~
~
1 atm glass
p 4 4 ~
T*g v a l x 10 (Xg X 10 J1.a x 104 1'1 X 10 5
Kg X 10 5 J1.K X 10 5
dT;/dpa T;J1.a c
(bar) eC) (cm' g-l) (OC- l ) (OC- l ) CC- l ) (bar-I) (bar-I) (bar-I) eC bar-I) ~

3
$::::l

c'
~

0 30.68 0.84309 7.117 2.795 4.322 4.981 2.896 2.085 0.0482 0.1313 ;:s
100 35.42 0.84177 6.996 2.771 4.225 4.856 2.880 1.976 0.0468 0.1304 $::::l
200 40.05 0.84042 6.840 2.740 4.100 4.743 2.869 1.874 0.0457 0.1284 ;:s
~
300 44.57 0.83905 6.664 2.704 3.960 4.637 2.864 1.773 0.0448 0.1258 C)
400 49.00 0.83765 6.480 2.662 3.818 4.532 2.864 1.668 0.0437 0.1230 ~
500 53.29 0.83620 6.300 2.614 3.686 4.420 2.868 1.552 0.0421 0.1203 ~
~
600 57.39 0.83468 6.136 2.559 3.576 4.298 2.877 1.421 0.0397 0.1182 ~

700 61.20 0.83308 5.998 2.499 3.499 4.158 2.889 1.270 0.0363 0.1170 t;x:,
('\:)
800 64.61 0.83137 5.894 2.433 3.461 3.999 2.903 1.096 0.0317 0.1169 ;::-
$::::l
~

800 bar glass c'


~

P T*g v a l x 104 a g x 104 J1.(X X 104 Kl X 10 5 Kg X 10 5 J1.K X 10 5 dT;/dpa T;J1.rx


(bar) CC) (cm' g-l) (OC- l ) CC- l ) (OC- l ) (bar-I) (bar-I) (bar-I) CC bar "")

0 14.75 0.83361 7.085 2.944 4.141 4.849 2.833 2.016 0.0487 0.1192
100 19.44 0.83240 7.010 2.918 4.092 4.668 2.809 1.860 0.0454 0.1197
200 23.87 0.83115 6.885 2.890 3.996 4.511 2.783 1.728 0.0433 0.1187
300 28.12 0.82986 6.726 2.861 3.865 4.371 2.756 1.615 0.0418 0.1164
400 32.25 0.82856 6.542 2.832 3.711 4.244 2.728 1.516 0.0409 0.1133
500 36.30 0.82726 6.348 2.802 3.546 4.126 2.700 1.425 0.0402 0.1097
600 40.29 0.82596 6.153 2.771 3.382 4.010 2.672 1.338 0.0396 0.1060
700 44.21 0.82466 5.968 2.741 3.228 3.894 2.643 1.251 0.0388 0.1024
800 48.02 0.82335 5.804 2.709 3.095 3.773 2.614 1.159 0.0374 0.0994
VJ
......
a dT: /dP = ~.K/ A(X. \0
320 Glass Formation and Glassy Behavior

exact procedure used by McKinney and Goldsteirr'? for the glasses formed at atmospheric or 800
bar pressure, it is similar enough that one could predict Oels and Rehage's result. In particular, they
found."? that AK/Arx was considerably greater than dTg/dP. This, of course, does point up an
interesting problem in defining the properties of a glass. Although the glasses formed by Oels and
Rehage 41 were individually isobarically formed glasses, the PVT surface corresponds to the
variable-formation glasses of McKinney and Golsteirr'? or Zoller.40 However, upon incrementally
changing pressure isothermally once the glass is formed, Oels and Rehage reproduced the com-
pressibility surface (K) corresponding to the constant-formation (isobaric) glasses.
In addition to the path-dependence problem of glass formation, it is also possible that both the
McKinney and Goldstein."? PVTsurfaces for constant-pressure glasses and the Oels and Rehage"!
incremental compressibility measurements reflect the fact that the volume obtained upon changing
the pressure shows kinetic effects similar to those that occur after a temperature change (recall
Figure 3).
One final point concerning the path dependence of the PVT behavior of polymer glasses is the
difference between isobaric glass formation (the variable-formation glass) and isothermal glass
formation. In the latter, the melt is compressed until a glass is obtained. If there were no path
dependence of behavior, a plot of the isobaric data as V vs. P at different temperatures would give
the same result as isothermal measurements. This is not observed, as shown in Figure 10, where
isobaric PVT data of Oels and Rehage"! are replotted and compared with their isothermal PVT
data.

1.00

0.95
T
0'
ro
E
u
~

0.90

0.851..o--~-------_---I_--~---~--~-

Figure 10 Specific volume v vs. pressure P for polystyrene through the glass transition range. Note that the isothermal data
and the isobaric data do not show the same behavior (data from ref. 41)

The above discussion puts into perspective the question of the validity of the 'volume'
equation (10), i.e. dTg/dP = AK/Arx for glass-forming systems. However, the use of the Ehrenfest
classification for second-order thermodynamic transitions also leads to the 'entropy' equation (12),
i.e. dTg/dP = T gVgArx/AC p. We note that it has frequently been reported that this equation is usually
valid,43-ss although discrepancies have been noted. 41,43,44 In a way it is surprising that equation
(12) is so often found to be valid for the glass transition event which, as we have noted previously, is a
kinetic phenomenon over normal experimental time-scales. Equation (12) is derived using the
Maxwell relation (see equations 7 and 11).

(14)

We have already seen that an isothermally formed glass is different from one which is isobarically
formed, and that the transition temperature from isobarically obtained data differs from that for
isothermally obtained data. One may then question the meaning of equation (14) under such
conditions.
Glass Formation and Glassy Behavior 321

However, for nearly all the reported results, equation (12) is a reasonable representation of the
data. This can be seen in Figure 11, taken from a compilation by Angell and Sichina.t? in which
T g vs. P was calculated using equation (12) and compared with experimental data. With only the
exceptions of poly(vinyl chloride) and ZnCl 2 the agreement is quite good. The reasons for this
agreement may stem from the fact that AtX does not vary much with formation history"? (seeTable 1,
for example) and that the dTg/dP values were more than likely obtained either from differential
scanning calorimetry (DSC) scans under the pressure used to determine ACp (P )43,44, 54 or from data
on variable-formation glasses,"? In fact, McKinney and Goldsteirr'? report that equation (13) is
valid, i.e. R = 1, for their variable-formation glasses. Obviously, the same heat capacity measure-
ments resulted in disagreement for the constant-pressure glasses."? We also note that ZnCl 2
glasses43,44 show a larger than usual kinetic effect on Tg , i.e. dTJd log t ~ -13.5 K per decade,
compared with the normally found value of about - 3 K per decade,56-51 where r is the time scale of
the experiment.

60

40
0
e,
......
01
20

-~l 0
cd) 0 -0== 0 J)

-60
-90

-100
0.49

o
-110
-0.60

-120

Figure 11 Tg vs. pressure P from different glass-forming systems, as indicated. Calculations are based on equation (12) (after
ref. 43, with permission)

In summary, dTg/dP depends upon the thermal and pressure history of the glass. With few
exceptions, it is found that, when experiments are run in a self-consistent manner, equation (10) is
valid. Equation (12) is also generally found to be valid, although this may arise from self-consistency
in the experiments. Therefore, although the glass transition event which is observed experimentally
is a kinetic phenomenon, the Prigogine-Defay ratio (equation 13) is equal to unity. In fact, such a
finding, given the nature of the thermodynamic constructions used to derive equations (10), (12) and
(13), is not surprising. If the glass transition event is a kinetic manifestation of an underlying second-
order transition, then theoretical models of the underlying transition must be tested against
predictions other than relations (10), (12) and (13). We will describe these models and the molecular
parameters which most strongly influence the glass transition. As we will see, some of these are
unique to polymers, although the glassy state is essentially ubiquitous. First, however, we make
some comments on ordering-parameter descriptions of glassy thermodynamics.
322 Glass Formation and Glassy Behavior

10.2.2.3 Ordering-parameter description ofglassy thermodynamics


As indicated in the previous section, there has been considerable discussion of the value of the
Prigogine-Defay''" ratio (equation 13), and the validity of the Ehrenfest" relations (equations 10
and 12). We also indicated that much of the problem arises from the fact that the results for dTg/dP
have been compared with ratios of Ii,,/Ii~ or T gVIi~/IiCp determined from different paths. The use of
such discrepancies in discussing the nature of the glass transition (for example, entropy- or volume-
controlled) appears to this author to be inappropriate. However, such discrepancies have been
treated and interpreted within the context of an ordering-parameter description of the glass
transition as a 'freezing-in' process. Here we will describe this approach and how it relates to the
non-equilibrium glassy state
The most important contribution to the extension of ordering parameters to describe the glass
transition is that of Davies and Jones.P The basic idea is that the non-equilibrium thermodynamic
state of the glass is described, for example, by a free energy function F( V, T, Zi)' where the Z, are
internal variables which determine the non-equilibrium state of the glass in addition to the normal
thermodynamic variables.
For a material defined by a single ordering parameter Z, Davies and Jones " showed that the
glassy state was defined at constant Z, while the equilibrium liquid state was defined at (oF / oZ)v, T
= 0, e.g. ~g = (1/ V) (oV/ oT)p,z and ~l = (l/V) (oV/ OT)P,oF/oZ = o- From this point of view, the glass
is defined by a line of constant Z, and in the glassy state Z is 'frozen-in'.
Importantly, Davies and Jones'" found that, for a single ordering-parameter glass, the Ehrenfest
relations (equations 11 and 12 are valid, and the Prigogine-Defay'" ratio is unity. However, upon
defining a material with multiple ordering-parameters, they53 found the following inequality for the
Prigogine-Defay'" ratio

R (15)

As discussed previously, it has often been reported that for real glasses R ~ 1 but that dTg/dP
= TgV Ii~/ IiC p (equation 12). While we have recognized this as being due to the path dependence of
the glass transition temperature, the analysis of Davies and Jones " does not address this problem
specifically, although they did subsequently examine the kinetics of glasses within this context.
Furthermore, the entire approach of Davies and Jones " has recently been debated and discussed in
a series of works by DiMarzio and co-workers,58-62 Goldstein.P-?" and Oels and Rehage."! Of
particular interest is the point of view of DiMarzi0 60- 62 that the Davies and Jonesj ' analysis leads
to a contradiction, and, additionally, the Prigogine-Defay'" ratio is unity even for systems having
more than one ordering parameter. Goldstein'P-?" does not entirely agree, but notes'" that, if
DiMarzio's conclusions are correct, the ordering-parameter approaches taken previously to
describe the thermodynamic aspect of the glass transition 'will be invalidated'.
Without going into this argument further, we note that the true importance of ordering
parameters in describing glassy behavior rests in their use in the development of models of kinetics in
the glassy state, which will be discussed in Section 10.3. Here we add that Davies and Jones."
DiMarzi0 60- 62 and Oels and Rehage'" have commented upon the importance of dealing with the
time-dependent behavior of the ordering parameters in order to correctly account for the glass
transition behavior.

10.2.3 Models of the Glass Transition


As noted previously, the experimentally observed glass transition is dominated by kinetics.
However, it is legitimate to ask whether, independent of the time scale, there exists a purely
thermodynamic glass transition. It is in this context that the Gibbs-Dilvlarzio r'<P" theory was
developed and eventually provided a theoretical basis for the resolution of the Kauzmann P
paradox. Using a lattice model, Gibbs and DiMarzi0 24- 26 found that below a temperature T2 the
configurational entropy was zero. Thus, it was argued that a true second-order thermodynamic
transition at T2 underlies the kinetically observed glass transition at Tg • In the following sections we
will describe the Gibbs-DiMarzio theory and, in the process of describing tests of this theory, survey
the influence of various molecular parameters on the glass transition temperature. Subsequent
sections will deal with free-volume models of the glass transition in the same manner.
Glass Formation and Glassy Behavior 323

10.2.3.1 Gibbs-DiMarzio theory


The Gibbs-Dilvlarziov'<P" theory of the glass transition results from an application of the
Flory-Huggins'v'-?" lattice model of a polymer system. DiMarzio 1 6 has argued that the use of the
lattice- model to study polymeric, as opposed to simple, liquids is more promising, because in
polymers it is possible to form glasses from systems which have no underlying crystalline phase,
which makes arguments about the existence of metastable glasses (which are strongly dependent on
kinetic questions) irrelevant to the equilibrium thermodynamic state described by the lattice model.
Also, the study of polymer glasses must include such parameters as the effects of chain length, chain
stiffness and chain connectivity, with the result that the theory can be compared to a richer class of
experimental phenomena than is possible for simple liquids. These latter become important here
because, as we noted in the previous discussion of equations (10), (12) and (13), experiments other
than the PVT type are necessary to test models of the thermodynamics of glasses.
The parameters relevant to the lattice model are readily described by reference to Figure 12. The
polymer chains of degree of polymerization X (DP = X) have many configurations which fit onto
the lattice of coordination number Z. Each chain has a lowest energy shape, and the more the shape
deviates from it, the greater the internal energy of the molecule. Gibbs and DiMarzio 2 4 - 2 6 assumed
a simple form for this energy, in terms of the fraction! of bonds 'flexed' out of the lowest energy state
and the energy associated with the flexed and unflexed potential wells, B2 and B1• They also allowed
for no vacant sites on the lattice. This latter results in a hole energy, proportional to the number of
intermolecular or van der Waals bonds (bond energy el) broken by introduction of the vacancies into
the lattice. Then there are two energies, dB = B2 - Bland el, and two parameters.j' and no, in the
model. These latter two are controlled by the energies. 16

-.. ,.. ..
'-fo+-r1
+
r-e ~

- ,..h
~ "r+- ~
t"~
~

. .
r1
~ r0-
r-rt
+~

+
~
,..r-e

Figure 12 Representation of the Gibbs-DiMarzio model. A lattice of coordination number Z is used to calculate the
number of complexions for a system of nx molecules each of length X. Energy a is associated with holes and energy ~6 is
associated with flexing of a bond from its low energy shape (after ref. 16, with permission)

The calculation of the partition function can be done by the standard Flory-Huggins lattice
method."? The lattice model predicts the existence of a true second-order transition at a temperature
T2 • This is shown schematically in Figure 13 for the entropy-pressure-temperature equation of state.
As can be seen, the transition occurs at a critical value of the entropy (zero configurational entropy)
and the Kauzmanrr'? paradox is resolved for thermodynamic reasons rather than kinetic ones, i.e.
one is simply not permitted to extrapolate high temperature behavior through the glass transition.
Rather, as the material is cooled, a break in the S-T (or V-T) curves occurs because of a second-
order transition.
The physical reason for the existence of the transition at T2 is then easily understood. As one cools
the system, at a given pressure, the number of allowed arrangements for the molecules decreases for
two reasons. In the first place, the number of holes decreases (a volume decrease) and the
configurational entropy due to permuting holes and chains decreases. Secondly, the configurational
entropy of the molecules decreases, because the chains favor low energy states (shapes) at lower
temperatures. The T(P) transition line defines the point where the total configurational entropy first
becomes zero. It is a direct result of the lattice model that this occurs at a finite non-zero
temperature, i.e. T2 • Gibbs and DiMarzio 2 4 - 2 6 hypothesized that the T(P) line represents the
thermodynamic glass transition in experiments of long time-scale.
324 Glass Formation and Glassy Behavior

Figure 13 Schematic representation of the SPT surface calculated from the Gibbs-DiMarzio lattice model. The mean-field
calculation on the lattice model results in a second-order transition in the Ehrenfest sense (after ref. 16, with permission)

The Gibbs-Dilvlarziov'<F" theory allows the glassy state as a metastable state above the lowest-
energy crystalline state for crystallizable materials, and resolves the Kauzmann paradox even for
crystallizable materials. However, more importantly, it defines the glassy state as a 'fourth state of
matter' in the case of materials, such as atactic polymers, which cannot crystallize. Glass formation is
associated with the condition that the configurational entropy goes to zero. Once one allows that the
glass transition temperature Tg , measured kinetically, is a reflection of the true thermodynamic
transition at T2 , the model makes specific predictions, many of which have been tested experimen-
tally. We now turn 'to the question of the ability of the Gibbs-DiMarzio theory to describe (or
predict) experimental data.

(i) The influence of molecular weight on T g


It is well-known that, for most linear polymers, the glass transition varies with molecular
weight,28,68 -78 although exceptions to this rule have been reported for polymers with ionic 72 or
hydroxyl79,80 end-groups as well as for polyisoprene.81_In fact, it is often reported that, over a
limited range of molecular weight, Tg varies inversely as M n , the number-average molecular weight,
and there are several treatments which give such a reciprocal relationship.69,82,83 However, at low
molecular weights, this relationship breaks down.P?: 78, 84 It is here that one can differentiate
between the Gibbs-DiMarzio theory and other treatments.
The molecular weight dependence of Tg obtained by Gibbs and DiMarzio can be written 69,80,84
as

+ 2exp-- -L\e) (16)


ua,

where x is twice the degree of polymerization, Ae is the flex energy, V o is the volume fraction of holes,
Tg is the glass temperature (associating T2 with Tg ) and k is the Boltzmann constant. Figures 14 and
15 show fits of equation (16) to experimental data70 for poly(vinyl chloride) over a range of
molecular weights from 540 to 45000, using values of V o = 0.025 and Ae = 6.34 kJ mol- 1. The
expression (15) fits the data quite well, even down to a molecular weight of 540 (x = 9). Because Ae is
fixed by the value of Tg at high (infinite) molecular weight this is really a one-parameter fit, i.e. Vo
must be determined. V o has been related 69,70 to the 'free volume' for an infinite molecular weight
polymer at Tg (free-volume concepts will be discussed in a later section). In reality, V o is related to the
hole energy (X in the lattice model.
As mentioned previously, the lattice model used by Gibbs and DiMarzio is readily generalized to
other systems. An interesting prediction is made by the theory for chains which have been formed
into uncatenated rings, Guttman and Dilvlarzio'" have calculated that, unlike for linear chains, the
glass temperature for rings should increase with decreasing molecular weight. There is little data for
the glass temperatures of small rings; however, the data of Clarson et al.2 4 6 for poly(dimethyl
siloxane) (PDMS) rings do show the predicted increase (see Figure 16). The agreement is only
Glass Formation and Glassy Behavior 325

350
: . _---.-- _---.-- , ._! - -- --e----! -- - -- --.-----
",......

...,.1 ·
325
,.•
I

.
:•T

2500------'---------~---~---~~---~---~

Figure 14 Change in Tg with number-average molecular weight for poly(vinyl chloride), the dashed line is the
Gibbs-DiMarzio prediction (after ref. 70, with permission)

350
.,'
.
.~.
~~

..
325
~~............
...........
.........

.
..........
..........
...............
..........

..
............
............
............
..........
..... ......
...... ..................
...........
...........
275 ...........

105 X s:
Figure IS Change in Tg with reciprocal number-average molecular weight. Solid line, linear dependence on 1\1;; 1; dashed
line, Gibbs-DiMarzio prediction (after ref. 70, with permission)

180

170

160

o~
150 Qt,:-:8~rf~Q:::-Q,.;~~~-'::O-D~-..$)~O---~OOO:O-_---o----o----""""o::tF~---
tiP",--
f/
140 01
I

x
Figure 16 Glass transition temperature vs. degree of polymerization X for (0) cyclicand (0) linear poly(dimethyl siloxane).
The lines represent the predictions of Gibbs-DiMarzio theory (after ref. 85, with permission)
326 Glass Formation and Glassy Behavior

qualitative for PDMS, but this is perhaps not surprising as the model does not represent the data for
linear PDMS very well. The reason for this is unknown.

( ii) Prediction of the heat capacity change at T g


As a thermodynamic model, the lattice model of Gibbs and DiMarzio allows the calculation of the
change in specific heat capacities at constant volume, ~Cv, or constant pressure, ~CP' once the
parameters ~e and (X (or ~e and vo) are known. Havlicek et al.8 6 have compiled comparisons
between the predicted and experimental heat capacities for various polymers and find excellent
agreement. Interestingly, they found that values of ~e and (X increase linearly with Tg of the polymer
and are not greatly different one from the other. Their data also show that the fractional free
volumes at T2 and Tg , i.e. vo( T2 ) and vo(Tg ), increase as Tg of the polymer increases. Their
interpretation was that the lattice model parameters are physically meaningful. Berry,"? on the other
hand, argues that the parameters are not truly meaningful, but finds 'surprisingly' that V o ( T)
appears to have a physical sense.
DiMarzio and Dowell'" have added a term for the contribution from lattice vibrations to the
original model. A comparison is made of their predictions with experimental results in Table 2. The
agreement for these polymers is excellent.

Table 2 Comparison of Gibbs-DiMarzio Predictions and Experimentally Observed Values of the


Specific Heat Discontinuity !1Cp at Tg for Different Polymers (After Ref. 88)

Polymers !1Cp theoreticar (J g-1 K -1) !1Cp experimental (J g-1 K -1)

Polyethylene 0.59 0.60


Polypropylene 0.51 0.48
Poly(isobutylene) 0.43 0.40
Poly(vinyl chloride) 036 030
Poly(vinyl acetate) 0.434>.47 0.41
Poly(methyl methacrylate) 0.404>.45 0.30
Polystyrene 031 Q34
Polyla-methylstyrene] 031 0.32
Polycarbonate 0.284>.33 0.24
Poly(ethylene terephthalate) 0.294>.35 0.33

a When two numbers are given, it is because the number offtexes per monomer unit is uncertain for these polymers.

(iii) Change in T g due to crosslinking


An important characteristic of high polymers is their ability to exhibit rubbery behavior above the
glass transition temperature. When crosslinks are introduced into the polymer to form a permanent
network rubber, the glass transition temperature can change significantly. 74,89 - 96 DiMarzi0 5 9
showed that the effect of crosslinking could be calculated quantitatively for real systems. Setting the
configurational entropy to zero at T2 he found that?"
T!1~ 3X 3 3
In (1 + 2e- A£/kT) + f!1e/kT ~ 1 + - + -Xlnf + -Xln (A'X) = 0 (17)
1 + 2T!1o. 2 4 4

where ~e, k, T and f are as before, ~(X is the change in coefficient of expansion from glass to liquid
(Xg), X is the crosslink density expressed as the number of chains per mole of segments and A' is
«(X, -

a number independent of the material.


DiMarzio was able to greatly simplify equation (17) and obtained
T(X) - T(O) KMX/y
(18)
T(O) - KMx/y

where T(X) is the glass temperature of the polymer with X crosslinks per gram, M is the molecular
weight of a residue and y is the number of flexible bonds per residue. K is a constant, independent of
material, approximately equal to 1.3 x 10- 23.
In Figure 17 we have replotted the data of Wood 9 2 for dicumyl-peroxide-crosslinked natural
rubber for Tg vs. parts-per-hundred peroxide, and compare the predictions from equation (18) using
Glass Formation and Glassy Behavior 327

-40r---~---"-"-"'~-""""""""'-"""-

-4~

-~o

E
-~~

~ -60

Dicumyl peroxide (ports per hundred)

Figure 17 Change in Tg with crosslink density for peroxide-crosslinked natural rubber. The line is calculated from equation
(18) (data from ref. 92)

K = 1.3 X 10- 23 and M]» = 22.7, as suggested by DiMarzio. Values for X were obtained using
Wood's97 calculations for the number of crosslinks per em? for different amounts of dicumyl
peroxide added to the rubber. The agreement is quite good over the range of the data. We must point
out here that these data are quite different from those used by DiMarzio in his original discussion. In
that case he used data for sulfur-crosslinked rubber.i" for which Tg increases much more rapidly
than for the peroxide-crosslinked systems. In fact, while the data he used fit an equation of the form
of equation (18), the parameter K differed by an order of magnitude from that which DiMarzio
reported to be a universal number. This may simply reflect the complex nature of sulfur-crosslinked
systems.

(iv) Effect of deformation on T 9


Another interesting aspect of the Gibbs-DiMarzio 24- 26 theory is that it predicts an effect of
deformation on the glass transition temperature due to changes in the entropy of the system upon
straining the network. For uniaxial extension, the result in terms of the small-strain modulus G,
measured at a temperature To, is 59

T(l)
T(I) =
[G
exp 2AC [1 1 - 3]
] (19)
pT o

where T( A) is the transition temperature at a stretch A, T( 1) is the transition temperature at zero


deformation (A = 1), and I 1 = Ai + A~ + A~ (= A2 + 2/A in extension) is the first invariant of the
deformation tensor. This equation results from the classical network theory for entropy of a
deformed chain.?" and could presumably be improved upon, but the prediction is interesting, i.e.
that the glass transition temperature increases exponentially with the deformation invariant. There
are few results available on the effects of deformation on Tg , but that from the most thorough
investigation by Gee et al.9 9 is in reasonable agreement with the model. As shown in Figure 18, the
transition temperature increases with increasing elongation."? Of significance in this respect is the
prediction that, because the configurational entropy always decreases upon deformation, indepen-
dently of the mode of deformation (i.e. tension, simple compression or shear), the glass temperature
should always increase when a rubber sample is deformed.
We note here that other data have been reported which disagree 100- 102 with those of Gee
et al.,99 as well as data which agree. 100, 102, 103 This is a subject which should be investigated further.

(v) Compositional dependence of T9


The glass temperatures of polymer-diluent systems, polymer-polymer blends and copolymers
vary with composition. 57,105 -120 The configurational-entropy model has proven relatively suc-
cessful in describing compositional effects on Tg • However, it is limited in the sense that, while one
328 Glass Formation and Glassy Behavior

0.03

::
i::
..... 0.02
':::::'
i::
I
::<
£ 0.01

Figure 18 Relative change in glass-transition temperature vs. stretch A for various values of G j2AC p T(l) = Y, as indicated.
Experimental data are from Gee et al.9 9 Circles are for natural rubber, for which Y = 0.0032;squares are for GRS rubber, for
which Y = 0.0012; the triangle is for Hycar, for which Y = 0.005, (after ref. 59, with permission)

can calculate Tg (T2 ) for a given composition by assuming that the configurational entropy for the
system is zero, analytical expressions for dependence of Tg on composition are not available from the
Gibbs-DiMarzio model. lOS, 110
For random mixing of polymer and solvent, the Gibbs-DiMarzio theory predictions for the glass
temperature are quite good, as shown in Figure 19. Gordon et al. 1 0 8 have used the configurational-
entropy model, but with an alternative method of calculation of the point of vanishing
configurational-entropy, to derive expressions for the compositional dependence of Tg of regular
solutions. They find that, in terms of the mole fraction x, Tg of a binary mixture can be expressed as

X ~1 + (1
(20)
x + (1

Monomer (%)

Figure 19 Variation of glass transition temperature with composition for solutions of polystyrene in styrene monomer. The
lines represent calculations from the Gibbs-DiMarzio lattice model: upper line calculation allows for 'holes' while the lower
one does not (after ref. 25, with permission)
Glass Formation and Glassy Behavior 329

which fits the data quite well.':" More recently, Chow 106 has used the lattice model to successfully
describe the change in 1: for non-random mixing of polymer solutions.
DiMarzio and Gibbsl~2 have derived equations which successfully describe the variation of Tg
with copolymer composition
(21)

where B A and BB are the fractions of rotatable bonds of polymer A and B, respectively, in the
copolymer. Equation (21) results from a simplified derivation and is only valid when the stiffness
energies used in the lattice model are not too different. De la Campa et al. 1 1 l have shown that the
Gibbs-DiMarzio model can be used ·to describe the behavior of poly(ethylene terephthalate-co-
diethylene terephthalate).
Other molecular parameters affect the glass temperature of polymers, such as tacticity''!: 121 and
blend composition.l P' 114 These have been successfully treated, at least partially, by the entropy
model. In addition, it is observed that the Tg of crosslinked systems is more dramatically reduced
upon addition of diluent than in uncrosslinked systems. 9S,106 This is also accounted for by
considerations of configurational entropy.
Finally, we note that, over and above the Gibbs-DiMarzio approach described here, the
composition dependence of Tg has been thoroughly treated by Couchman and KaraszllS-117 from a
classical thermodynamic point of view. They claim that simple considerations of the thermo-
dynamics of mixing and the glass temperature results in rules similar to those described above for the
composition dependence of Tg , and that agreement with experiment is not truly support for the
Gibbs-DiMarzio model. This point of view has been strongly disputed by Goldstein.P"

10.2.3.2 Free-volume concepts


( il Free volume and molecular mobility
Historically, the free-volume concept was developed to explain the non-Arrhenius dependence of
the fluidity or viscosity of liquids on temperature.F'<P": 122-138 The most well-known formulation of
this is the so-called Doolittle/": 124-126 equation

In '7 = InA + B(v - vd/uf (22)

where 11 is the viscosity, v is the specific volume and Vf is the free volume. Doolittle 29,124-126
recognized early that the definition of the 'free space' Vf was difficult, as was the 'limiting specific
volume' Vo , where Vf = v- V o ' Without describing his definitions (we will subsequently use modern
notation) we find that equation (22) describes quite well the viscosity-temperature behavior of many
simple liquids.
The free-volume concept was subsequently introduced to describe the melt behavior of polymers
in the work of Fox and Fl ory127-129 and that of Williams, Landel and Ferry.P? These latter found
that, when the free volume was assumed to increase linearly with temperature (equation 23), the
equation which is now known as the WLF equation could be derived from the Doolittle equation
(22).
(23)

The WLF equation was developed empirically to describe the change in viscoelastic properties of
polymers above the glass transition.P? Starting from the principle of thermorheological simplicity,
i.e. that the spectrum of relaxation (or retardation times) changes with temperature by a simple shift
along the time axis, it was found that the shift factor for the viscosity could be written as S7

('70 ToPo)/('7oo Tp) (24)


or
log ('70/'700) + log(ToPo/Tp) (25)

where 110 and p are the viscosity and density at the temperature T of interest and 1100 and Po are the
viscosity and density at the reference temperature To, normally taken in the mid-range of the data.
The correction term ToPo/Tp is small and is often ignored in the reduction of the viscosity data. S7
Williams, Landel and Ferry'?" found that log aT could be well represented by the empirical equation

(26)
330 Glass Formation and Glassy Behavior

where C~ and C~ are constants which, while originally thought to be universal, are now recognized
to depend on the polymer.
As noted previously, the WLF equation (26) and the Doolittle equation are found to be same if
one treats the free volume as varying linearly with temperature. If we adopt Ferry's t? notation that
the fractional free-volume is f = Vf / v and the temperature dependence of f is

f = 10 + ~f(T - To) (27)

then, from equations (22), (24) and (27), and ignoring the ToPo/Tp correction, we find
-(B/2.303fo) (T - To)
log aT = (28)
fo/~f + T - To

and we can then identify the constants in the WLF equation (26) with those in equation (28) and
obtain
B
C? -- (29)
2.30310
c~ fo/~f (30)
B
10 (31)
2.303 C?
B
~f = (32)
2.303C?cg

If we identify the reference temperature as the glass transition temperature, then equation (31)
becomes
(33)

Because B is an arbitrary parameter, the free-volume fraction cannot be determined directly from
these equations. As noted previously, Doolittle/?' 124-126 was concerned with the definition of free
volume and in Figure 20 is shown a possible definition for the free-volume fraction based on the
concept of an occupied volume. When one considers B to be unity, the value of the free-volume
fraction at Tg falls"? between 0.013 and 0.034 for many polymers, and the 'universal' value originally
proposed by Williams, Landel and Ferry 1 3 7 was taken to be 0.025 (see equation 23; also the
discussion about the free-volume fraction derived from the Gibbs-Dilvlarzio-v P ? lattice model).
We note that the Doolittle/Wl.F equations predict a rapid increase of viscosity with decreasing
free volume (or as Tg is approached). Another empirical equation used to describe this is the

--- --- --- .---..---- ---- .----


--
...-- ----;'g Tg (00) r; (T )

o
T(K)

Figure 20 Schematic variation of total specific volume V( T)/ Vg , occupied volume Vo ( T) and free volume at Tg(/g) with
temperature T for a supercooled liquid. a g is the coefficient of expansion (total) of the glass, ~ is the coefficient of expansion of
the liquid, ~ f is the coefficient of expansion of the free volume and ~o is the coefficient of expansion of the occupied volume.
L\~ = ~ - ~g is the difference in slope between liquid and glassy states. Tg(r) is the glass transition temperature at some time
r (or equivalent cooling rate) and Tg ( CX)) is the glass transition temperature at infinite time (after ref. 13, with permission)
Glass Formation and Glassy Behavior 331

Vogel.P? Fulcher.v'? Tamann and Hess 141 (VFTH) equation

log 11 = A + B/(T - Too) (34)

where Too is often found to be approximately 50°C below the conventionally measured glass
temperature and has been associated with the T2 of the Gibbs-DiMarzio 24- 26 theory. The three
equations (22), (28) and (34) are formally equivalent and describe the data quite well over a large
range of viscosities, as shown in Figure 21. We remind the reader that these equations are valid 57
from approximately Tg + 10 "C ~ Tg < T g + 100°C.

10

CI)

o
Q..

b
cr[
e
<,

0'
o
...J

120 140

Figure 21 Logarithm of reduced viscosity 110/ P,., vs. temperature T for melts of cyclic polystyrene molecules showing rapid
increase as Tg (~ 100 °C) is approached. The line represents the VFTH equation with To = 43.5 °C (after ref. 245)

The empirical use of the free volume (in particular the WLF model) to describe the mobility of
molecular fluids has been extended to include pressure effects. Then one can write that the shift
factor is related to temperature and pressure through the free volume with the result142-145

(B/2.303k) [T - To - O(P)]
log aT p = (35)
, k/ar(P) + T - To - O(P)

where
O(P) C~(P)ln [1 1
+
+
C~P ]
C~Po
_ C~(P) In [1
1
+
+
CgpJ
cgPo
(36)

and
C~(P) l/k r ar (P) (37)

C~ krK: (38)
C~(P) l/ktPar(P) (39)

cg ktP/K: (40)

where the superscript on the C's refers to the reference temperature To; Po is the reference pressure
and C~ and C~ depend on the experimental pressure P.1o is the free-volume fraction at the reference
332 Glass Formation and Glassy Behavior

temperature and ctr(P) is the pressure-dependent coefficient of thermal expansion of the free volume.
Ki is the bulk modulus of the rubber at To and zero pressure, k, is the Bridgeman pressure coefficient
of the bulk modulus and K: and kq, are the corresponding analogs for the occupied volume. The
equations (35H40) were shown to successfully describe the temperature- and pressure-dependences
of the viscoelastic shifts for a series of lightly crosslinked rubbers. 142 -145
In spite of this success, there is still some argument about the role of free volume or fractional free-
volume in determining the mobility of liquids. Some fluids can apparently be described by the WLF-
type equations over reasonably large ranges of temperature and pressure, but in other cases the
description results in unreasonable values for the free-volume parameters.146 Also, under conditions
where the free-volume fraction is held constant by increasing pressure when the temperature is
increased, the viscosity of the fluid is not constant.146-150 Thus it is argued that an energy of
activation is required in the free-volume model.
Much of the argument surrounding free-volume models results from the difficulty, recognized by
Doolittle.'?' 124-126 of defining both the free volume and the occupied volume. In a sense, they are
fitting parameters for the data in the absence of a specific model which gives a priori the free volume.
A point of importance here is that equations for molecular mobility, similar in form to those which
can be derived from the free-volume concepts (equations 26, 28, 34 and 35),can be obtained from the
configurational-entropy model of Gibbs and DiMarzio. Adam and Gibbs 202 carried out this
extension of the Gibbs-DiMarzio theory by assuming that the relaxation time in the system is
inversely proportional to the configurational entropy.

( ii ) Free volume and the glass transition


The argument described above, that the free volume controls the molecular mobility, leads to a
kinetic resolution of the Kauzmanrr'? paradox. As the free volume collapses with decreasing
temperature, molecular mobility decreases until the time required for molecular rearrangement is
longer than the time scale of the experiment and equilibrium is not reached. In fact, when one
considers the viscosity (or the shift factor aT, p) as a measure of the time required for molecular
rearrangement, the free-volume models give a critical temperature Too (and pressure) at which the
viscosity goes to infinity (see equations 22, 24-26, 28, 34 and 35, Figures 20 and 21) which
corresponds to zero fractional free-volume. Thus the material never reaches equilibrium and the
Kauzmann F problem is resolved for kinetic reasons. The value of Too is found "? to be approxi-
mately 50°C below the conventionally measured Tg •
The concept of free volume is quite descriptive and, as discussed above, able to describe quite well
the mobility of many simple and polymeric fluids. The quantitative interpretation of the free and
occupied volumes, however, is not obvious and has been worried about since Doolittle. 29,124-126
Major breakthroughs in arriving at a molecular basis for free volume were based on ideas of
cooperativityI31-133,151,152 of motion of neighboring molecules. According to Cohen and
Turnbull,153,154 molecular transport in simple liquids occurs when a molecule moves into voids
having a greater size than some critical value v*. The voids are created by the redistribution of free
volume arising from the cooperative motion of neighboring atoms. According to the Cohen and
Turnbull model, 153,154 there is no thermodynamic change in phase in going from the liquid state to
the glassy state; rather, the glass transition exists because the free volume of the amorphous phase
falls below some characteristic value which is small enough to increase the relaxation times (decrease
molecular mobility) enough that one does not attain the equilibrium state. This underlying
assumption of the nature of the glass temperature was also advanced by Fox and Fl ory2s,6s, 127-129
and entered into the thinking of Williams, Landel and Ferry'P? in their interpretation of the
viscosity of polymers near Tg and their development of equation (26). The Cohen-Turnbull free-
volume model arrives at the empirical Doolittle equation (22) from two simple assumptions:
(1) molecular transport occurs only when voids having a volume greater than some critical value v*
form by the redistribution of the free volume; and (2) no energy is required for free-volume
redistribution.
Because of the success of the empirical free-volume relations in describing the behavior of glass-
forming liquids, there have been many attempts since the Cohen and Turnbull free-volume model to
quantify the concept and make the free-volume physics more than a convenient way to correlate
data. The reader is referred to the literature for a general look 1-11 at the various models and also for
some specific developments.Pv I'" However, due to space limitations, we limit our discussion to the
cell model of Simha and Somcynskyl64,165 and the extensive developments of this model which
have been carried out over the years by Simha and co-workers.166-177
Glass Formation and Glassy Behavior 333

( iii) The hole model of Simha and Somcynsky


Simha and Somcynsky'P": 165 developed a cell model for polymer liquids which has been highly
successsful in describing the PVT behavior of melts and glasses. The partition function is computed
from a model which assumes that the polymer repeat units are confined to cells, the centers of which
are in a regular lattice. The classical cell-model, which generally underestimates the entropy, is
extended to include empty cells (holes), which adds a mixing term to the entropy. A square-well
approximation to the Lennard-Jones and Devonshire cell potential is used and the canonical
partition function can be evaluated to obtain the Helmholtz free energy A.
The model is generalized to include chain molecules (polymers) in addition to simple liquids by
factorizing the free volume of a molecule into 3c elements, where 3c is the number of external degrees
of freedom. Then, applying the equilibrium condition that
(oAjoy)v, T = 0 (41)

for a system of N molecules of degree of polymerization s occupying a fraction y of sites, one obtains

Equation of State PVjT [1 - 2- 1/6y(yV)-1/3]-1 + (2yjT) (yV)-2[1.011(yV)-2 - 1.2045] (42)


Internal energy iJ (yj2) (y V) - 2 [1.011 (y V) - 2 - 2.409] (43)
Entropy S -sjc[(1 - y)jy]ln(l - y) + 3In[(yV)1/3 - 2- 1/6 y] + In V*
+ constant (44)

and, from the relation (41), the equation of constraint

(sj3c)[(s - l)js + y- 1 ln (1 - y)] = (yj6T)(yV)-2[2.409 - 3.033(yV)-2]


+ [2- 1/6y(yV)-1/3 - 1j3] [1 - 2- 1/6 y(yV)-1/3]-1 (45)

where 3c/s is a characteristic flexibility parameter. The reduced variables are given by
V VjNsv* = vjv* (46)
T ckTj(qze*) (47)

P Psv* j(qse*) (48)

where e* and v* are the characteristic energy and volume per segment, qz is the number of nearest
neighbor sites per chain equal to s(z - 2) + 2, and z is the coordination number. Additionally we
define three scaling parameters
p* PjP (49)
V* VjV (50)
T* TjT (51)

The Simha-Somcynsky 1 6 4 , 1 6 5 hole theory develops the free-volume model quantitatively by


identifying the Doolittle free-volume fractionf, described previously, with the hole fraction h = 1- y.
Then the liquid state PVT behavior can be described using the hole theory, as shown in Figures 22
and 23. Figure 22 shows isobaric data as log V vs. log f for several polymer melts and Figure 23
shows reduced isotherms for polystyrene melts as PV/ f vs. 1/ V. Importantly we note that the hole
theory is descriptive of the data rather than predictive. However, fitting to data then allows
extraction of physically meaningful parameters, such as the hole fraction (free volume).
Application of the Simha-Somcynsky'P": 165 model to the glass transition allows some interesting
observations about the hole fraction to be made. Because this model treats the glass transition as
wholly kinetic, the transition event itself cannot be predicted a priori from the model but must arise
from kinetic arguments. However, the glassy state can be described by the model. It is relevant to
note that, in the hole model, unlike other free-volume treatments, there is no assumption that the
free-volume fraction remains constant below Tg , although it is constant along the T; vs. P line,
where T; refers to the glass temperature for constant-formation glasses, described previously.
In Figure 24, agreement between experiment and the hole model is excellent in the melt. The open
circles correspond to the data for the glass (constant formation) and liquid. Below Tg , the solid
curve for each pressure gives the theoretical values of V corresponding to a constant value of
h = 1- y = h( Tg ( P)). This constraint condition replaces the equilibrium condition (equation 44). The
dashed lines represent the equilibrium liquid lines predicted by the theory. The large differences
334 Glass Formation and Glassy Behavior

0.9

0.8

0.7

0.6

0.5
~:::.
0-
0
..J 0.4

0.3

0.2

0.1
-1.5 -1.4 -1.3

Log T
Figure 22 Logarithm of reduced volume vs. logarithm of reduced temperature for representative polymer melts and a
n alkane liquid. The line is calculated from equations (42) and (45) (after ref. 169, with permission)

Figure 23 Reduced isotherms for polymer melts. The lines are calculated from equations (42) and (45) (after ref. 171, with
permission)

between the slopes in the glassy region of the experimental volume-temperature curves and those for
constant h (theoretical) led Simha and co-workers170-173,178 to the conclusion that h = h(Tg ) in the
glass is an oversimplification.
An alternative procedure adopted by Quach and Simha 171 results in Figure 25. Here the glassy
P VT data are used to calculate the fraction of holes as a function of temperature in the glass. The
open circles show values of h calculated in this manner. Below Tg the solid line represents the
assumption that the hole fraction in the glass remains constant at h=h(Tg ) . Thus Figure 25 suggests
that the free-volume fraction varies with temperature below Tg and the structure of the glass is not
completely frozen. Finally, we note that the free-volume fraction at Tg given by the Simha-
Glass Formation and Glassy Behavior 335

1.10

1.04
V
1.02

1.00

0.98

0.96

Figure 24 Reduced volume vs. reduced temperature for poly(vinyl acetate) at two different reduced pressures. Solid lines are
calculated from Simha-Somcynsky'P'v t'" theory assuming that the hole fraction is frozen at Tg • Dashed lines are for the
equilibrium liquids below Tg (after ref. 173, with permission)

12

II

10

7'g c O.0 3 226


8 h-0.0780

P -1"-0.0853

26 28 30 32 34 36 38 40

I03 X T
Figure 25 Temperature dependence of hole fraction above and below the glass transition in poly(vinyl acetate). The points
represent two different calculation schemes to determine h. These results show that h is not constant (i.e. frozen) below Tg
(after ref. 170, with permission)

Somcynsky 16 4 , 1 6 5 model differs markedly from that used by Ferry and co-workers"? (WLF-type),
or that which can be derived from the Gibbs-DiMarzio 2 4 - 2 6 lattice model. The Simha-Somcynsky
model results in values of h = f ~ 0.07, while the WLF empiricism and the Gibbs-DiMarzio lattice
model give values of the free-volume fraction of about 0.025, as mentioned previously. The
discrepancy with the WLF approach may come from the arbitrary choice of a value of unity for
parameter B in the Doolittle equation.
336 Glass Formation and Glassy Behavior

The Simha-Somcynsky model is successful in describing much of the thermodynamics (P VT) of


glass-forming substances. Its one major drawback is that it does not develop the kinetics, and
therefore the transition phenomenon, a priori. However, the model is useful as a quantitative free-
volume description of glassy behavior. As we will see, the cell-model limitations on prediction of the
entropy (or energy) equation of state still hold for the Simha-Somcynsky description of the glass.
In the following sections we will examine the ability of the free-volume models, with emphasis on
the Simha-Somcynsky'P": 165 model where possible, to describe the dependence of Tg on different
experimental parameters.

(iv) The influence of molecular weight on T g


As discussed previously, there is a significant dependence of the glass transition temperature
on polymer molecular weight (see Figures 13 and 14) for linear chains. Fox and Flory 1 2 7 - 1 2 9 first
suggested that an empirical relation in which Tg varied inversely with number-average molecular
weight, M n , could describe the dependence for M n > 3000
(52)

where T; is the glass temperature for infinite molecular weight polymer and B is a constant. This
relationship can be understood if the glassy state is taken to be an 'iso-free-volume' state. If we take a
linear chain, as M n decreases the concentration of free ends increases and so does the total free
volume. Assuming that the free volume (or fractional free-volume) is constant at T g for samples of
different molecular weight, and that each terminal group contributes an excess free-volume fJ, a
relation of the form of equation (52) is readily obtained. 70,82
Let the excess free-volume per linear molecule be 2fJ and the free volume per unit volume be equal
to 2fJp/NAM n , where p is density and N A Avagadro's constant. Assuming constant free-volume, at Tg
and a coefficient of expansion for the free volume (Xr (see Figure 20), one can write
(53)

As the excess free-volume jntroduced by the chain ends must be compensated for by thermal
contraction from T: to T~n; i.e.
(54)

then

2fJpNA
Tg = Tt - ---- (55)
~fMn

which is of the same form as equation (52) with B = 2fJpNA/(Xr.


Equation (52) or (55) describes the experimental data quite well except at low molecular weights
(large values of M n- 1 ), and has been used extensively to describe the Tg-M n relationship for
polymers. We note, however, that Gibbs and DiMarzi0 2 5 have pointed out that the glass is not a true
iso-free-volume state, and the free-volume description of Simha and Somcynsk y 1 6 4 , 1 6 5 also pre-
cludes 173 such a description of the glass transition.
The free-volume model does not explain the lack of molecular weight dependence of the T g of
certain polymers 72,79-81 mentioned previously, and obviously the increase in Tg at low molecular
weights for ring-like molecules are not readily explicable in the context of a simple free-volume
model.

(v) Heat capacity change at T,


There has been little use of the free-volume models to describe or predict thermodynamic
functions other than the PVT equation of state for glasses, because these models are primarily
concerned with molecular mobility. However, the cell model of Simha and Somcynsky'P" 165 has
been used 170,178 to predict the heat capacities of poly(vinyl acetate) in the liquid and glassy states.
Table 3 shows a comparison between theory and experiment for C p and C, in the liquid and glassy
states, as well as for L\Cp , L\Cv and L\S at the glass transition. Although the values of C; and C;
obtained from the model are less than 200/0 of the observed ones, values for Cp-C v are reasonably
close to those actually observed. The values for L\Cp , L\Cv and L\S obtained from the model are less
Glass Formation and Glassy Behavior 337

Table 3 Comparison of Heat Capacity Data at T" = 304 K for Poly(vinyl


acetate) with Predictions from Simha-Somcynsky'P'v V'? Cell Model (After
Ref. 178)a

Liquid Glass

C p (T = Tg ) (experimental) 1.77 1.27


C p (theoreticalj/C, (experimental) 0.19 0.08
~Cp (experimental) 0.500
ACp (theoretical) 0.234
AC v (experimental) 0.310
AC v (theoretical) 0.039
Sg - S (T = 273 K) (experimental) 0.055
Sg - S (T = 273 K) (theoretical) 0.023

a All units in J g - 1 K - 1.

than half of the experimental values. This poor agreement with the experimental values undoubtedly
lies in the previously mentioned problem that cell models predict low entropies. Thus, in spite of
obtaining excellent agreement with the PVT behavior of the glass, the SPT behavior is not well
described by the hole model.

( vi) Change in T g due to crosslinking


The increase in glass transition temperature with increasing crosslink density is readily under-
stood within the framework of the free-volume concept if one considers the fact that the introduction
of crosslinks into the polymer involves the exchange of van der Waals' bonds for shorter covalent
bonds. This results in a decrease in the specific volume (and, presumably, the free volume) of the
polymer. Fox and Loshaek 74 and Loshaek'" have treated the crosslinking effect on Tg from this
point of view. They arrived at an expression for the difference between the glass transition of the
crosslinked polymer Tg(p) and the uncrosslinked polymer ~(O)

meAoce Tg(O) )
Tg(p) - TiO) = ( meP (56)
OC e - Be - meAocp

where p is the number of crosslinks per gram, A(X = (X - (Xe is the difference between the coefficient of
thermal expansion of the monomer, (x, and that of the uncrosslinked high molecular weight polymer
above Tg , (Xe. Be is the coefficient of expansion of the glassy high molecular weight polymer and me is
the monomer molecular weight.
Equation (56) results in a linear dependence of T g vs. p, and has been found to describe the change
in Tg with crosslinking for many polymers. Figure 26 shows the results for poly(styrene-co-
divinylbenzene) obtained by Fox and Loshaek.?" We note here that some workers claim that Tg

Figure 26 Change in glass temperature vs. crosslink density for poly(styrene-co-vinylbenzene) ApTg = Tg(p) - Tg(O) and
p = crosslink density in crosslinks g-l. The straight line is calculated from equation (56) (after ref. 74, with permission)
338 Glass Formation and Glassy Behavior

exhibits an exponential dependence on crosslink density, as predicted by the Gibbs-DiMarzio


theory.?" but data are difficult to obtain over a large range of crosslink densities, and distinguishing
between a straight line and an exponential or other non-linear curve~3 is difficult to do.
The free-volume models have also been used successfully to describe the effects of crosslinking on
molecular mobility. 179, 180 The WLF equation is generally invoked in these cases, in which case the
free volume can be used to describe the mobility above Tg 180 and the effect of changing Tg on cure
kinetics. 179,180

(vii) Effect of deformation on T,


The fact that the volume of a material changes upon deformation suggests that the free volume
also changes. Thus, if one associates the sign of the global volume change with the change in free
volume, the glass transition should shift to lower temperatures when the volume increases and to
higher temperatures when the volume decreases. Thus, in uniaxial compression Tg should increase
because the volume decreases. On the other hand, the volume in uniaxial extension increases and
therefore Tg should decrease. In shearing deformations there should be no change in Tg •
These free-volume arguments give quite different predictions for the effect of deformation upon
the glass transition temperature than does the Gibbs-DiMarzio theory discussed previously.
Unfortunately, there have been few experiments designed to test the effects of deformation on the
glass transition temperature. Furthermore, as noted by Johnston and Shen,100 the change in Tg
expected for tensile deformations is only of the order of 1-2 K for 100% elongation, and determi-
nations of Tg are often imprecise and the elongations relatively small. Perhaps the best measure-
ments are those of Gee et al. 9 9 cited previously (Figure 18), which indicate that Tg increases upon
extension. However, Mason;'"! Witte and Anthony'P? and Stevens and Iveyl03 have observed
decreases in Tg upon stretching. On the other hand, in addition to Gee et al.,99 Oshimo and
Kusumotc 'P" have observed increases in Tg with stretching. This author knows of no measurements
of Tg changes due to uniaxial compression or shearing deformations. Such experiments may
elucidate the free volume vs. entropy and the thermodynamic vs. kinetic bases of the glass transition.

(viii) Compositional dependence of T g


As mentioned previously, the glass temperature of polymer-diluent systems, polymer-polymer
blends and copolymers depends upon the composition. The free-volume theories have been used to
describe this behavior with varying degrees of success. First we examine the question of polymer-
diluent systems.
Following Ferry,"? we first remind the reader that addition of low molecular weight diluents
depress the glass temperature, as depicted in Figure 27 (see also Figure 18). J enckel and Heusch 181
first arrived at an equation in which the Tg depression is linear in the weight fraction, w 1 , of the
diluent
(57)

where Tg(O) is the glass temperature of the undiluted polymer and k is a constant depending on the
solvent. However, as seen in Figure 27, the actual dependence on weight fraction of diluent is
nonlinear. This non-linear Tg depression has been attributed to the introduction of additional free-
volume upon addition of the diluent."? as would be expected if the fractional free-volume of the
diluent (It) exceeds that of the polymer (1;). Normally, the Tg of the polymer is greater than that of
the diluent, i.e. Tg 1 < Tg 2 , which corresponds to It >1;. Equation (57) results when the free volumes
of the diluent and polymer are additive. In the linear range.j' depends upon W l in a linear fashion.
Then
f(T, wd ];(T) + {3w t
(58)
or in terms of the volume fraction Vl

f(T,v t ) = ];(T) + B», (59)

where {3 and {3' are constants, {3' = {3p 1I P2. If Tg represents an iso-free-volume-fraction state, then,
from equation (57), k = {3letc. Alternatively,' 82

(60)

where etC1 and etC2 are the expansion coefficients of the fractional free-volume defined previously
(see Figure 20).
Glass Formation and Glassy Behavior 339

-20

-40

~
....0> -60

"-
""","",,,
-80

......
-100 <, ......
<, - -•
............ '0
-120

-140

t:.
-160

0

Figure 27 Glass transition temperatures of polystyrene solutions with various low molecular weight diluents: . , p-naphthyl
salicylate; +, phenyl salicylate; 0, tricresyl phosphate; (!) methyl salicylate; _, nitrobenzene; 0, chloroform; x, methyl
acetate; +, ethyl acetate; ., carbon disulfide; 0, benzene; p,
toluene; 6, amyl butyrate (after ref. 181, as cited in ref. 57, with
permission)

If one assumes that the fractional free-volumes are not additive.l'" the curvature in Figure 27 can
be accounted for and one writes the total fractional free volume as
(61)

where k~ is a negative-valued interaction parameter of the order of 10- 2 . This fits well with the
observation that the total specific volume depends in a similar way on weight fraction'P"
(62)

which implies that the occupied volumes, not the free volumes, are additive. Equation (61) leads to
the following compositional dependence of T g

(63)

Kelley and Bueche'P" derived similar equations for the composition dependence of polymer-
diluent systems from free-volume considerations and also showed acceptable fits to the data. The
parameters used to fit equation (63) to the data are generally found to be in the range expected from
free-volume arguments. Of particular interest is the fact that at small diluent-contents, one finds that
equation (61) corresponds to 1 3 8
P' = it - h + k~ (64)

and, since h and k~ are small, P' ~ It. Then, It is in the range 0.1 to 0.3, which is the correct magnitude
PS 2-L
340 Glass Formation and Glassy Behavior

of the fractional free-volume of small-molecule liquids far above Tg • The above development shows
the power of free-volume models to describe the observed experiments.
The dependence of the glass transition temperature on copolymer composition has been treated
by Gordon and Taylor P" using a volume approach. In their development for 'ideal' copolymers,
Gordon and Taylor P" assume additivity of volumes of the two components in both rubbery and
glassy states. Then two equations result
(65)
(66)

where ~ and VG are the specific volumes of the copolymer in the rubbery and glassy states,
respectively, w, is the weight fraction of the ith component and ~i and VG i are the specificvolumes of
the rubbery and glassy phases of the ith component. Assuming linear expansion coefficients in the
liquid and glassy states for each component (Xli and (Xgi' one arrives at an equation for the
concentration dependence of Tg for the copolymer'": 185

WI ~lk + W2~2
(67)
WI + kW2
where k=~(X2/~(X1 and ~(Xi = (Xli - (Xgi·
In Figure 28 is shown the excellent agreement obtained between equation (67) and the exper-
imental data for poly(butadiene-co-styrene). Loshaek'" reports equally good results for poly(methyl
methacrylate-eo-glycol dimethacrylate)s. Non-additivity of volumes would, of course, alter these
results, as equations (65) and (66) would require additional terms. We note that equation (67) is
identical to equation (63) with k~ = O.

115

i
65

U
e,
15
.....0'

-35

Weight fraction of styrene

Figure 28 Glass transition temperature vs. weight fraction styrene for poly(butadiene-co-styrene) The line is calculated from
equation (67) (after ref. 185, with permission)

Similar equations based on free-volume arguments have been proposed for block copolymers.P"
Again the agreement is quite satisfactory. For polymer blends, equations of the same sort have been
successfully used to describe the change in Tg with blend composition. 1 1 3 , 1 8 7 For these latter
systems, however, the experimental measurement of Tg may dramatically affect the agreement
obtained. Figure 29 shows Tg vs. composition for mixtures ofpoly(2,6-dimethyl-l,4-phenylene oxide)
(PPO) with polystyrene (PS) obtained by different experimental methods.I'" As can be seen, the
shape of the curve depends dramatically on the method of measurement.

10.3 KINETICS OF GLASS-FORMING SYSTEMS


10.3.1 Introduction
In the previous section the thermodynamics of the glass transition and the glassy state were
described. From the discussion there, it is obvious that a complete understanding of the nature of
Glass Formation and Glassy Behavior 341

220

180

E
......0>

140

Figure 29 Glass transition temperature, measured by different techniques (e, DSC; ~,mechanical; _, dielectric), of blends
of poly(phenylene oxide) and polystyrene vs. weight fraction poly(phenylene oxide) w. (after ref. 188, with permission)

glasses remains elusive for the major reason that, regardless of whether or not underlying the glass
transition there is a true second-order thermodynamic transition, the overwhelming features of
observed glassy behavior are kinetic.
The phenomenology of the kinetics of glasses has been studied and known for many years.
However, it is only recently that the mathematical formalism has become available to describe the
major features of the known behavior, which shows a rich array of phenomena. In the sections which
follow, we describe the phenomenology of the kinetics of glassy systems and the equations which
have been developed to explain their kinetic behavior. We subsequently show how the statistical
mechanical models of glasses have been adapted to the mathematical formalism which describes the
kinetics. We will also examine some recent developments in the use of computer simulations to look
at the kinetics of glasses and, finally, we will briefly describe the effects of volume recovery on the
mechanical behavior of glasses, i.e. physical aging.

10.3.2 Phenomenology
10.3.2.1 Kinetics ofglass formations
The glass transition event manifests itself as a change in the temperature dependence of some
property (e.g. volume V, enthalpy H) of a liquid upon cooling. As shown in Figure 1, below some
critical temperature range, the volume (enthalpy, entropy) of the undercooled liquid departs from
equilibrium. The glass transition temperature Tg is determined by the time scale of the experiment t.,
and corresponds to the temperature at which the characteristic time L for the molecular rearrange-
ments leading to structural (volume, enthalpy, entropy) recovery is longer than the time scale of the
experiment, i.e. L > i.. Therefore, as t, increases, the position of the departure from equilibrium
moves towards lower temperatures. The intersection of the extrapolated glassy and liquid lines
defines Tg(te ). This definition was used by Too11 8 9 , 19 0 to define the fictive temperature, TF , which
has come to be used to quantify the structure of the glass at an arbitrary temperature in terms of the
structure of the equilibrium liquid at TF (or Tg ) . This concept, while not entirely valid, is useful and
will be referred to subsequently.
The variation of T g with the experimental time scale t., or rate of cooling q, for a large number of
glass-forming systems is s7
- dTg ~ dTg ~ 3K (68)
dlog i, dlog q

although, as indicated in Section 10.2, there are systems which show a rate sensitivity which is much
higher.
342 Glass Formation and Glassy Behavior

10.3.2.2 Kinetics in the glassy state


( i) Intrinsic isotherms
Upon cooling the equilibrium liquid below the glass transition range, the thermodynamic
properties of the glass (V, H, etc.) continue to evolve toward equilibrium as shown in Figure 3.
Kovacs 11 has referred to the isothermal contraction curves resulting upon interruption of cooling at
different temperatures T 1 as intrinsic isotherms. There are two interesting features of the curves in
Figure 3. First, the departure from equilibrium, b = (v - voo)/v oo does not vary linearly with T l .
Second, the times required to reach equilibrium increases dramatically as b l ' the initial departure
from equilibrium, increases (or T l decreases).

(ii) Asymmetry of isothermal recovery


If the glass is allowed to equilibrate (reach true thermodynamic equilibrium) at a temperature
To = T - J1.T or To = T + J1.T, then the volume recovery upon changing the temperature to Twill
differ depending upon whether or not one is expanding or contracting towards equilibrium. This
asymmetry of approach to equilibrium (depicted in Figure 30) is well understood qualitatively.
When the approach towards equilibrium is from above, the material has a structure which
corresponds to the liquid having higher mobility than the equilibrium mobility. Thus, the contrac-
tion is autoretarded.Conversely, the expansion towards equilibrium from below is autocatalytic,
because mobility increases as the departure from equilibrium decreases towards an equilibrium state
of higher mobility.

-I

-2

0.001 0.01

Figure 30 Expansion and contraction isotherms for poly(vinyl acetate) glass at T = 35°C after heating or cooling from
To = T ± 5°C. This plot shows the asymmetry of the expansion and contraction isotherms (after ref. 13, with permission)

( iii) Memory effects


One of the (apparently) most complicated features of the kinetic behavior of glasses arises when the
material is allowed to recover isothermally for a period of time t 1 insufficient to reach equilibrium,
and then heated to a higher temperature and allowed to recover. As shown in Figure 31, the volume
departure from equilibrium can cross over the actual equilibrium and exhibits a maximum which
depends upon the actual thermal history applied to the sample. These have been referred to as
crossover or memory effects. As will be seen, they arise from the fact that the response function
(e.g. for volume recovery) exhibits behavior equivalent to a multiplicity of retardation mechanisms.

( iv) Uniform heating through the transition range


A common experiment is to study the glass transition behavior by DSe. In such experiments,
a sample is brought into the glassy state either by an unknown route, i.e. 'as received', or by a
controlled thermal history. During the experiment, the heat flow into the sample required to
Glass Formation and Glassy Behavior 343

2.0

1.5
T -30°C

1.0
~
)(
r<>
Q
0.5

~
0.01 0.1 10 100
f - f (h)
j

Figure 31 Isothermal evolution at T = 30°C for poly(vinyl acetate) showing 'memory' effect:(1) quench from 40°C to 30 °C;
(2) quench from 40°C to 10 °C for 160 h followed by rapid heating to 30°C; (3) quench from 40 °C to 15°C for 140 h followed
by rapid heating to 30°C; and (4) quench from 40 °C to 25°C for 90 h followed by rapid heating to 30°C (after ref. 13, with
permission)

1.0
350K/17h f . ,l·-
.\!)
4··-·-



/
~

.
o

1.0 ,.. .............- .

?
360K/1 h ,. 370 K/I h


/
~.
o

360 370 380 390 400 360 370 380 390 400

T(K)

Figure 32 Heat capacity C; from DSC measurements vs. temperature for poly(methyl methacrylate) (PMMA) annealed at
T, for a time i, as indicated (i.e. Te/te). The glasses were cooled at 40 K min -1 and reheated at 10 K min -1 after annealing.
Note the 'features' in these DSC curves (after ref. 204, with permission)
344 Glass Formation and Glassy Behavior

maintain the desired temperature is measured. In an ideal case, the change in heat capacity can be
measured on samples which have been cooled slowly and reheated immediately at approximately
the same rate. However, upon choosing 'poor' experimental conditions, the DSC trace can exhibit
complicated features, as shown in Figure 32. The peaks in Figure 32 have been variously interpreted
as having specific molecular significance. However, as we shall see, they can arise simply from the
kinetics of response to a complicated thermal history. Analogous features would be observed in
volume measurements if a were followed as a function of temperature upon heating."!' 19

10.3.3 Phenomenological Equations


10.3.3.1 Requirements of the models
From the above discussion one can see that the kinetics of glasses exhibit a rich variety of
behaviors which offer a severe test of any model. Until the development by Narayanaswamy"? in
1971 of a mathematical formalism analogous to that of reduced variables used in viscoelasticity
theories, there was no model which incorporated all of the 'ingredients' necessary to explain the
major features of the phenomenology described in the previous section.
The pieces necessary to describe the behavior, in addition to the viscoelasticity formalism, are:
(a) a non-exponential response function which can be represented in terms of multiple retardation
mechanisms characterized by a distribution of retardation times Li; and (b) a departure from
equilibrium, e.g. b, related to T and t by a linear superposition principal but with each retardation
time Li dependent upon the temperature and on the instantaneous state (structure) of the glass.
Several developments based upon these ideas were made by Narayanaswamy,' 7 Kovacs and co-
workers 18,19,191-194 and Moynihan and co_workers. 20,19s,196 The result has been a tremendous
improvement in our ability to understand and describe the kinetic behavior described above. In fact,
once the model is developed, a single non-linear equation describes all of the major features of the
phenomenology. In the next section, we will develop this equation following the approach of Kovacs
and co-workers.l" the so-called KAHR model, because of this author's familiarity with the model.
We note that the resulting equations are formally identical with those of Narayanaswamy'? and
Moynihan and co_workers,2o,19s,196 the differences being not in the underlying assumptions of the
models, i.e. (a) and (b) above, but rather in the specific routes and approximations used to obtain the
equations.!? An excellent treatment of the Narayanaswamy model and applications is given by
Scherer."

10.3.3.2 The Kovacs-Aklonis-Hutchinson-Ramos (KAHR) model


(i) Single-parameter approach
In this section and the next will arrive at a set of equations which describe the volume recovery
under isobaric conditions of model glasses. The KAHR model of Kovacs, Aklonis, Hutchinson and
Ramos '" formally uses a multiple ordering-parameter model to define the 'viscoelastic' response
function for the glass. Here we take a somewhat different approach, summarized by Kovacs, 19which
avoids the specific formalism of ordering parameters and results in the same set of equations. It is
important to remark that the ordering-parameter approach to describing the kinetics of glasses
is interesting and has been increasingly formalized, not only in the KAHR model but also by
Moynihan and co-workers. 19S,196 Here we wish to focus on understanding the important par-
ameters which yield the rich kinetic phenomenology described in the previous sections.
In early work, Too1 189, 190 and Davies and Jones" examined the structural recovery of glasses
assuming that the departure from equilibrium depends upon a single ordering-parameter. Accord-
ingly, the recovery was described by a single retardation time Li. Here we define as the ordering
parameter the volume departure from equilibrium expressed as a dimensionless variable b

(69)

where D is the instantaneous specific volume and Doc> the equilibrium value at the same T and (P).
Then at any temperature the rate of recovery at constant P is 11,S3,191,197

d£5
-- = qJ1rx + £5/,r: (70)
dt
Glass Formation and Glassy Behavior 345

where q = dT/dt, t is the time and L\o: = (0:. - O:g) is the excess expansion coefficient of the liquid with
respect to the glass. The enthalpy recovery can also be described by equation (70) if we refer to the
enthalpy departure as <5 H = H - H 00 and substitute the excess heat capacity L\Cp = (Cp 1 - CPa) for
L\o:. In addition, equation (2) can be cast in terms of the fictive temperature T F to give'": 189, 190

(71)

where T F = T + <5/ L\o: and T F = T + <5 H / L\Cp for volume and enthalpy recoveries, respectively.
(Tool used the fictive temperature to define the structure of the glass. In simple cooling experiments
T F is the intersection of the extrapolated glassy and the extrapolated liquid lines, and is equivalent to
T g (see Figure 1, for example). For more complicated histories TF can be redefined and the reader is
referred to ref. 5 for an excellent discussion of T F .)
The solution of equation (70) depends upon the functionality of the retardation time 'to This
consequently determines how well actual glassy behavior is described by the one-parameter model.
If one assumes that T depends upon temperature alone (in isobaric experiments), then glass
formation and the uniform heating through the glass transition range can be qualitatively
described. 18, 197 However, this model predicts that isothermal recovery proceeds via a simple
exponential decay of <5 in contradiction with observation. Also a dependence of r on T alone cannot
account for the asymmetry of approach depicted in Figure 31.
Tool, 189,190 who recognized this shortcoming, was able to account for the non-linearity and
asymmetry of the isothermal approach to equilibrium by assuming that T depends, in addition to T
and (P), on the instantaneous state of the glass, i.e. on T F or <5. Introduction of a <5-dependent 't into
equation (70) provides qualitative agreement with the isothermal recovery of glasses. In particular,
the intrinsic isotherms 1 1 are reasonably well described in such a model, as is the asymmetry of
isothermal approach towards equilibrium. 5
However, one-parameter models cannot even qualitatively describe the memory effects depicted
in Figure 31. These effects arise precisely because of the contributions of at least two independent
retardation mechanisms involving two or more retardation times. In this context, two or more
ordering parameters are required to describe the kinetics. (We note that this is true within the
context of a discrete spectrum of retardation times. Moynihan et al. 2 0 have pointed out the
usefulness of a non-exponential decay function rather than the multiplicity of mechanisms inherent
in the ordering-parameter approach.)

(ii) Multiparameter approach


The multiparameter treatment of the KAHR 18 model can be viewed as a simple extension of the
single-parameter model. The development is formalized by defining the thermodynamic state of the
glass system at any T and P by a set of N ordering parameters, Zi. Both <5 (the departure) and L\o: are
partitioned among the Zi' with each of the latter being associated with a unique retardation time 'tie
It is furthermore assumed that the contribution of each Z, to <5 is independent of the other ordering
parameters. Then, analogous to equation (70), we have
d£5 i
--
dt = qAC(.I + ~'/t'.
I I
(1 ~ i ~ n) (72)

where L\O:i = giL\o: is the weighted contribution of the ith retardation mechanism to L\o: with
PI

I e. = 1 (73)
i= 1

The KAHR model assumes that each 'ti depends on the total departure from equilibrium (i.e. on <5
not on <5J which couples the set of equations (72). Furthermore, the total departure is the sum of the
individual departures, i.e.

(74)
i= 1

The solution to equation (72) now depends upon the specific temperature (pressure) and structure
dependence of the 'tie These dependencies are put into the KAHR model in a manner equivalent to
the time-temperature superposition principles of viscoelasticity theory. 57, 198 Then KAHR assume
that, by a change in temperature or <5, each retardation time is shifted by the same amount and that
346 Glass Formation and Glassy Behavior

the amount of shift due to a change in temperature is independent from that due to the departure (b)
from equilibrium (i.e. structure). As a result of this, the spectrum of retardation times simply shifts
along the time axis but does not change shape, i.e.
(75)

where aT and alJ are the appropriate shift factors for the spectrum at any T and b relative to 'ti, r in the
reference state at equilibrium. Then simply
ti(T, O)/ti(Tn 0) (76)

ti(T, lJ)/ti(T, 0) (77)

where T, is a reference temperature and b = 0 denotes equilibrium. The set of paired values (gi' 'ti,r)
defines the retardation spectrum, G(ri,r) of the system in a reference equilibrium state at 'I';
Invariance of the shape of the spectrum is assured by assuming the gi are independent of T and b.
Then one can write the thermal history dependence of b in a compact form by using a reduced-time
variable z
t dt'
z = f0 a Ta6
(78)

which reduces the instantaneous rates of change of structure (bi ) at any T and b to those obtained in
equilibrium at T. The relevant expression for b(t) is then given by the convolution integral

lJ(z) = -Aa I o
R(z - z') -dT dz'
dz
(79)

where R(z) is the normalized retardation (recovery) function for the sytem
N
R(z) = L giexP( -z/ti,r) (80)
i= 1

the value of which changes from unity to zero as z (or t) increases from zero to infinity.
Importantly, we now have a formalism which allows one to describe the behavior of a hypo-
thetical glass once one has determined aT' all' Acx and R(z). All of these material functions must be
determined from experiment. This implies further assumptions about the model which we will deal
with in the next section, where we show how well the model agrees with the phenomenology
presented previously. Before leaving this section, we note that two assumptions which have been
made in the model are unnecessary. First is the invariance of the shape of the retardation function
with changes in T and b. The other is that the parameters gi are independent of the magnitude of Tor
b (or bH ), which results in Acx (or AC p ) being a constant. For isobaric experiments this is perhaps
unimportant. The extension of this formalism to include pressure histories as well may require a
pressure dependence in the gi (which weights Acxil as well as another set of weighting functions for the
pressure equivalent of Acx, i.e. AK.1 8 In the formalism of equation (79) AK would no longer be a
constant. Then equation (79) would exhibit non-linearities due to a magnitude dependence of Acx (or
AK) in addition to those due to the structure-dependent retardation spectrum. This will not be dealt
with further.

(iii) A comparison of the phenomenological equations with experiment


Equation (79)is a general equation in which the recovery function R(z) and the shift factors aT and
alJ are unknown. Acx is unknown but easily measured. In the KAHR18 model, R(z) is a function which
can be expressed by a spectrum of retardation times (equation 80). A more easily visualized function
is the so-called Kolrausch199-Williams-Watts200 (KWW) function introduced to describe glassy
kinetics by Moynihan et al.2 0 , 195, 196
R(z) = exp( -z/trf (81)

where 'tr denotes a characteristic retardation time at T, and f3 is a parameter describing the width of
the 'spectrum', 0 ~ f3 ~ 1. In the discussion which follows we will use the KWW function for R(z),
i.e. equation (81), rather than the sum of exponentials, equation (80).
Glass Formation and Glassy Behavior 347

The forms of aT and a~ used in the KAHR model are

exp[ - 8(T - ~)] (82)

a~ exp[ - (1 - x)8~/A(X] (83)

where e and x are material parameters, and x is a partitioning parameter (0:::; x:::; 1) which
determines the relative contributions of temperature and structure to Li. When x = 1, a~ = 1, i.e. there
is no structural dependence of the retardation times and they depend only on temperature. This
parameter has also been referred to as a non-linearity parameter.i'" since the non-linearity is
introduced into the model due to the structure dependence of Li. The forms for aT and a~ introduced
into the KAHR model are those originally put forward by Tool,189,190 but expressed in terms of b
rather than TF • Similar expressions can be obtained from free volume.l "?' configurational
v

entropy 'P" or activation enthalpies. 17,2o, 195 These expressions reduce to equations (82) and (83)
when the temperature interval chosen is not too wide.18, 19
The problem of determining the adequacy of equation (79) to describe the data is one of
determining Aa, (or ACp ) , x and e, and R(z). The difficulty is that the non-linear nature of
equation (79) makes unique determination of these parameters and functions difficult. However,
upon assuming reasonable values for Aa, e and x and a two-box representation of the distribution
of retardation times, KAHR18 were able to solve equation (79) numerically for various thermal
histories. The results obtained reproduce 18, 19all of the features shown in Figures 1,3 and 30-32, i.e.
the specific non-linearity and asymmetry of the isothermal approach curves, memory effects and the
peaks obtained on constant heating through the transition range, although in the latter a was used
rather than Cpo
It is interesting to note here that the combination of a broad distribution of retardation times,
combined with a structure-dependent shift factor for these times explains some extremely compli-
cated features of glassy behavior. A set of experiments which are of particular interest are depicted in
Figure 32, i.e. uniform heating through the transition range. The peaks exhibited in such DSC traces
of heat capacity vs. temperature have been related to specific relaxation mechanisms in polymers.
However, great care must be used in interpretation of such peaks, because in non-isothermal
experiments the glassy kinetics themselves can give rise to such peaks. As an example, Hodge
and Berens 203 have carried out calculations using the Narayanaswamy'YMoynihan''? form of
equation (79) for different conditions. As seen in Figures 33(a) and (b) the annealing time and
temperature of the glass can affect considerably the position and magnitudes of the peaks. A full
analysis and discussion of this is presented by Hodge and Berens,201,203-205 by Kovacs and co-
workers 18,19, 191-194, 197 and Tribone et al.2 0 6 One question remains: 'How well can the models
describe the experimental data?'. The answer is very well- but! The 'but' implies that there are still
some things which have not been completely explained. Kovacs and co-workers 'P' 19, 191-194, 197
treat the problem as one of developing a rigorous test, not only of the phenomenology predicted by
the model, but also of the underlying physical assumption of the model. In this case, it is necessary to
determine by independent measurements the values of Aa, x, e and R(z), and to then compare actual
data with the model predictions. The non-linear nature of equation (79) makes such independent
determinations difficult. However, KAHR 18 were able to give methods for determining all of the
parameters in the model except x, the structure-partitioning or non-linearity term. Although
Lagasse et al.2 0 7 claim to have found a method to obtain x, this has been disputed by Kovacs.F'"
Without x, one is left to examine the model by curve-fitting' procedures. Several groups have adopted
this approach with exceptional success. 201,203-206,209-212
Finally, we note that the extensive testing of the phenomenological equations, while indicating
satisfactory' agreement with the experimental results for most 'practical' applications, has shown
less than satisfactory agreement from a scientific point of view, for two reasons. The first is that the
values of the parameters required for optimal fit to the data change from one set of experiments to
the next. In particular the shape of the distribution of retardation times, as measured by f3 in
equation (81) and the structure dependence of the retardation times as measured by the non-linearity
parameter x, change with thermal history.5,201,206,209-212 In Figure 34, for example, a DSC curve
calculated using parameters determined from one thermal history is compared with the data for a
different history. Although the calculated behavior is similar to that observed, the agreement is not
perfect. Better fits are obtained by allowing both x and f3 to vary with history, which then violates
some underlying assumptions about the model, i.e. that the shape of the spectrum is invariant and,
more importantly, the current retardation times depend only upon the instantaneous departure
from equilibrium. The second difficult problem in the description of the data arises in the fits to the
asymmetric-approach curves (see Figure 30). Although one can fit these curves quite well, a close

PS 2-L*
348 Glass Formation and Glassy Behavior

1.2
(a)

1.0

0.8

c, 0.6

0.4

0.2

-.---
~o:Vo~-~-:;-;!:;:::-----='~7---~~--_..l-_-_..L
400
T(K)

T(K)

Figure 33 DSC curves calculated using a Narayanaswamy17-Moynihan2°-type phenomenological model for glasses
subjected to different annealing times teo (0, 1 h; . , 1 day; ~, 1 week; . , 1 month; D, 1 year) after quenching to different
temperatures: (a) Te = 40 °C; (b) T; = 60°C. This demonstrates how complex features in DSC curves can be reproduced by
phenomenological equations (after ref. 201, with permission)

scrutiny of the fit in the light of the so-called 'r-effective paradox' (see next section) implies that there
is an 'ingredient' missing in the underlying physics of the phenomenological models. We turn now to
this paradox.

(iv) The x-effective paradox


Upon bringing a glass from equilibrium at a temperature To = T1 ±~Tto the new temperature T1 ,
the volume (enthalpy) recovery towards equilibrium exhibits either autocatalytic or autoretarded
behavior depending upon whether the approach towards equilibrium is from below or above. We
previously referred to this behavior as the asymmetry of approach in T-jump experiments
(Figure 30). As described above, the phenomenological equations of the Narayanaswamy-
Glass Formation and Glassy Behavior 349

4.7

""
3.5 (0)

2.3

1.1

315 325 335 345 355

3.1

2.3 (b)

0.7

365

3.1

2.3 (c)

1.5

0.7

390 400 410 420 430


T(K)

Figure 34 Comparison of experimental (solid) and calculated (dashed) enthalpy recovery curves: (a) isotactic PMMA,
T, = 311 K; (b) atactic PMMA, T, = 362 K; (c) syndiotactic PMMA, T; = 380 K. Annealing temperatures are approximately
15 K below Tg. The curves represent t; = 0,6000 and 60000 s, with peak height increasing with annealing time (after ref. 206,
with permission)

KAHR-Moynihan-type model this behavior quite well. However, in earlier work, Kovacs''"
examined this behavior through an effective retardation time teff defined as

_ ~(d<5) (84)
<5 dt
where t is the time and b is the departure from equilibrium. teff is a derivative parameter which
makes fitting its behavior an extremely demanding test of any model.P: 19,21,213 In Figure 35 are
depicted teff plots for the expansion and contraction isotherms reported by Kovacs'" for poly(vinyl
acetate). The data on the right-hand side of the figure (b > 0) come from contraction experiments in
which the sample was brought from equilibrium at To = T1 + AT to T= T1 • The left-hand side of the
figure (b < 0) shows the response in the expansion experiments in which the sample was brought from
equilibrium at To = T1 - AT to T= T1 • The curves are labeled in families by the given T1 value, and
350 Glass Formation and Glassy Behavior

the values for T1 ± tiT are given near each individual curve. These data, of course, reflect the
asymmetric approach to equilibrium. In discussing the families of curves, we want to keep in mind
that each set of contraction and expansion isotherms corresponds to the same value of T1 , e.g. 35 "C.
Of considerable interest is that the effective retardation times in the approaches from b > 0
(contraction) merge near b = 0, while from b < 0 (expansion) there is no corresponding merging of the
r eff values. For example, at T 1 = 35 "C the responses to the temperature jumps are markedly
different: as equilibrium is approached, at b ~ 10- 4 , [reff ] 30 DC/ [ reff ] 32.5 DC ~ 5, i.e. there is a 'gap'
between the curves. This behavior has been commented on previously by Kovacs et al., 18,191 Ng and
Aklonis P ? and others,"?' 214 and is surprising because, near equilibrium, one would not expect such
a difference in an effective retardation time, since the effects of the previous history should have faded
away. In the original KAHR 18 paper, plots of log reff vs. b were calculated for reasonable values of
the material parameters and, as seen in Figure 36, for all of the curves, contraction and expansion,
the values of reff converge at b=O. Furthermore, Ng and Aklonis/!" have analyzed many sets of
parameters, and they were unable to reproduce the effect (even qualitatively) except with extremely
unrealistic values of the material parameters. To this author's knowledge, there is only one model
which accounts for the 'r-effective paradox' -the coupling model of Ngai.

( v) The coupling model of N,gai and the t-effective paradox


An interesting model of relaxation phenomena has appeared recently.": 215-223 The appeal of the
model from the point of view of glassy relaxations is that it can describe the r-effective paradox

-2
37.5

-I

J
o
o
...J
I

3'---_.....I.- .....I.- ....L.... ...L- ....L...._ _

-4 -2 o 2 4

Figure 3S Log "Ceff vs. departure from equilibrium b for expansion (b ::; 0) and contraction behaviors (b ~ 0) of a poly(vinyl
acetate) glass. Experiments were conducted by equilibrating the samples at the temperatures indicated on each curve and
jumping the temperature to that indicated with each family of curves. Thus, volume contraction experiments were carried out
at T = 30°C after equilibration at temperatures of 32.5, 35, 37.5, 40 and 60°C. Importantly, note that while contraction
curves merge at b = 0, the expansion curves do not. The latter is referred to as the 'expansion gap' (after ref. 13, with
permission. Note that the figure is corrected courtesy ~f A. J. Kovacs)
Glass Formation and Glassy Behavior 351

Q)
t-
o
o
...J
I

Figure 36 Calculations oflog !eff vs. departure from equilibrium b. Calculations are based on the KAHR model. This figure is
a simulation using the KAHR model of the experimentally observed behavior depicted in Figure 35. T, is the temperature of
the volume contraction experiment and A T indicates the equilibration temperature. Note that the expansion gap is not
reproduced (after ref. 18, with permission)

without recourse to additional assumptions beyond those used in the Narayanaswamy-KAHR-


Moynihan models, other than that relaxation in, 'complex' systems proceeds as described by the
coupling model of Ngai": 215-223 Thus, the basic physics of glassy recovery remains the same, i.e. it
is necessary to have a retardation function (spectrum) which exhibits non-exponential decay and
retardation times which depend upon the instantaneous state (structure) of the glass. As mentioned
previously, the shape of the retardation spectrum may change during recovery (i.e. as a function of
the structure of the glass), but this is not the missing 'ingredient' required to explain the r-effective
paradox. The piece that is missing is the correct description of the relaxation process in 'complex'
systems as given by the coupling model.
There is insufficient space here to derive the coupling model in detail, therefore we present the
salient features of the model and show how it describes the isothermal recovery behavior of glasses in
response to aT-jump experiment. In particular, we will pay attention to the expansion experiment,
i.e. ~ < 0 where a 'gap' exists between the curves for log 'teff vs. ~ as ~ approaches zero.
According to the coupling model, relaxation parameters (e.g. ordering parameters or mechanisms
in the KAHR model) do not exhibit simple exponential decay. Rather, the physics of complex
systems requires cooperativity among the relaxing species which acts to slow the simple constant
relaxation rate into a time-dependent rate. Then one finds that the relevant variable initially decays
at a constant rate, Wo = 'to 1 , but at times larger than a characteristic time i, == w; 1 the rate becomes
explicitly time-dependent, i.e. of the form W = 'to 1 (wet)-n where 0 < n < 1. This is the basic result of
the coupling model which can be summarized as
W(t) !o 1, wet ~ 1 (85)
W(t) !ol(w et)-n, wet ~ 1 (86)

The fact that the relaxing species exhibit a time-dependent rate in the complex environment has
some important consequences. The relaxation rate of the macroscopic variable l/J becomes

d4>
- W(t)4> (87)
dt

and integration of equations (85) and (86) gives

4> (t) 4> (0)exp ( - tl!o), wet ~ 1 (88)

4> (t) 4> (O)exp[ - L 'to


dt' I (wct'}-n J. ov » 1 (89)

The slower-than-exponential decay of l/J in equation (89) reflects the coupling of the primitive
relaxing-species to its complex environment. Ngai and co-workers 2 1 7 , 2 1 8 have been able to
determine We for several polymers and find that it is typically in the range 109 -10 1 0 S - 1.
352 Glass Formation and Glassy Behavior

The form of equation (89) for equilibrium conditions, when -ro, We and n are constant, becomes

(90)
where 1
r* = [(1 - n)())~ro]t=n (91)

is the effective relaxation time of the complex system (i.e. that observed experimentally), and is
related to -ro, the relaxation time of the primitive species (uncoupled). We remark that equations (90)
and (91) contain the same parameter n, thus the measured relaxation time -r* is related to the shape
of the decay function (spectrum), which leads to testable predictions because n, which describes
material behavior, must be the same in both equations (90) and (91). The model has proven
successful in many instances.217-223
Importantly for the structural recovery of glasses, the model predicts an equilibrium decay
function which is of K WW form (/3 = 1- n), see equation (81), for even a single primitive species.
Thus the requirement of a non-exponential decay function is fulfilled by the model. Although the
other models use a broad relaxation function to describe behavior, they neither make the prediction
of equation (91), nor can the general equation (89) result from them. In general, nand -r* can be
functions of Tand <5; however, we treat only the case where -r* is a function of <5, i.e. -r* =-r*(T, <5(t')).
Then, rewriting equation (86) in terms of -r* and identifying the macroscopic variable <p with the
departure from equilibrium <5, we find that, for isothermal volume recovery, equation (89) becomes

(92)

There are then three important facets to equation (92). First, at equilibrium, the decay function is
non-exponential (equation 90). Second, the observed retardation time depends upon <5 and T. Third,
both nand -r* appear inside the integral, which implies that the history-dependent change of both
these parameters is inherent in the model and so they affect the volume-recovery response differently
than in the Narayanaswamy-KAHR-Moynihan-type models. For the same isothermal recovery
after a T-jump, we can compare equation (92) with the response'" for the KAHR 18 model

o(t) = oo .f. e. exp(- It


1=1 0
'1- 1dt)' 'I = 'I( T, 0) (93)

and the Narayanaswamy'Yvlvloynihan''" form

(94)

As can be seen, equation (92) is different from (93) and (94), which is a direct result of the coupling
model.i! Rendell et al. 2 l have solved equation (92) for the T-jump experiment. As can be seen from
Figure 37, the expansion curve is reproduced, as is the 'gap' as <5 approaches zero. The lower portion
of Figure 37 shows the shift factor a(T, <5) which was used to fit the data compared with independent
measurements obtained by Kovacs, Stratton and Ferry224 (KSF).
An important result of the coupling model is that the expansion gap is directly related to n, the
coupling parameter in the model. In fact, this is a severe test of the coupling model, as the value of n
which fits the shape of the equilibrium retardation spectrum, must be that which determines the
magnitude of the 'gap'. This is, in fact, what Rendell et al. 2 1 found.
We remark that the coupling-model predictions for contraction, i.e. <5 > 0, while reasonable at
small AT, fit less well at large AT. This was attributed." to finite cooling-rate effects, not accounted
for in equation (92). Furthermore, the coupling-model predictions have not been compared with
other experiments, e.g. memory effect, uniform heating through the transition, etc., and ultimate
confirmation of the model awaits such quantitative comparison. The results presented above as an
explanation of the r-effective paradox are encouraging.

10.3.4 Free Volume and the Kinetics of Glasses


10.3.4.1 Introduction
The phenomenological models described in the previous section contain the essential ingredients
needed to describe most of the kinetic behavior of glasses, i.e. a retardation function exhibiting non-
Figure 37 (a) Calculation of expansion side of the r-effective plot using the Ngai coupling model. The expansion gap is
reproduced. (b) shows the shift factor a( T, (5) used in the model vs. the experimental data of Kovacs, Stratton and Ferry2 3 0
(after ref. 21, with permission)

exponential decay and a dependence of retardation times on the instantaneous state (structure) of
the glass. The Narayanaswamy-KAHR-Moynihan models are virtually identical, while the coup-
ling model of Ngai results in a somewhat different phenomenological equation, not by changing the
'essential ingredients' above, but by treating the relaxation process itself in a new way. In the next
section we examine the use of a detailed model of free volume to describe glassy kinetics.

10.3.4.2 The Robertson-Simha-Curro (RSC) model


Robertson, Simha and Curro (RSC)225-227 have developed a molecular theory of the kinetics of
structural recovery of glasses. It is based upon a stochastic model originally proposed by
Robertson,22s-231 and makes use of the previously discussed Simha-Somcynsky'P" 165 cell model
to calculate the free-volume function and its fluctuations. As we will show, the theory develops
expressions for a spectrum of retardation times and the time-dependent distribution of the free
volume.
RSC assume that the changes in measured volume associated with structural recovery are changes
in free volume. The free volume is defined using the Simha-Somcynsky164,165 equation of state
(recall Section lO.2.3.2.iii)
(42)

where the P, V and T are reduced variables and y is the fraction of occupied cells. Then fractional
free-volume f is related to y by
f = 1 - y (95)

At equilibrium, the free volume can be determined by minimization of the free energy
(s/3c)[(s - 1)/s + y- 1 ln (1 - y)] = (y/6T)(yV)-2[2.409 - 3.003(yV)-2]
+ [2- 1/6y(yV)-1/3 - 1/3][1 - 2- 1/6 y( y V )- 1/3] - 1 (45)

As RSC note, equation (45) is not valid when the polymer is not in equilibrium, and a different
condition is needed to reduce the number of variables in equation (42).
354 Glass Formation and Glassy Behavior

To compute changes in free volume during recovery, the theory requires that one examine changes
throughout the entire sample. This is because local changes occur at rates that depend upon the local
free-volume, which can vary from point to point because of thermal fluctuations. (We note here
that this assumption, due originally to Robertson,228-231 is at variance with the assumption in
the phenomenological models where the 'local' rate depends on the global departure b =1:b i ) . At
equilibrium, one can write an appropriate expression for the mean-squared fluctuation in free
volume 225,227 (bf2) = (b y 2 ) :

+ (1/3 yT) [6.066(yV)-4 - 2.409(yV)-2] (96)

where N; represents the size of the regions of interest and is the nominal number of chain units in
those regions. Equation (96) represents the fact that the mean-square fluctuation in free volume is a
constant independent of volume size.225 At equilibrium, equation (96) is valid. In the 'frozen' glass,
away from equilibrium, one must use the measured volume with equations (42) and (96)to determine
the free volume. Also, although the value of N; is the number of chain units required for local
segmental rearrangements, and should be approximately 26 for vinyl-type polymers.P? RSC treat it
as an adjustable parameter in their computations.
In the RSC model there is a discrete free-volume distribution function, which allows one to write a
set of stochastic equations in terms of the probability wi(tO ) of finding a local region out of
equilibrium in the ith state at time to, i.e. it contains a fractional free-volumej.. RSC go on to define a
probability matrix Pij(t - to) which gives the probability that a region known to be in the ith state at
time to will be in the jth state after some infinitesimal time interval At. Assuming that a region
undergoing a change from state i to state j will pass through all intervening states, the transition
probabilities become
Pi,i-l (At) AtAi- + O(At) (97)

Pi,i+1 (At) so; + O(At) (98)

Pii(At) 1 - At (Ai- + At) + O(At) (99)

where O(At) denotes a quantity which approaches zero faster than does At. At is the upward
transition rate from state i and Ai- is the downward rate. At equilibrium, microscopic reversibility
requires that
(100)

where the ~i are the state occupancies given by the binomial distribution.P?
In the model, the upward and downward transition rates are expected to depend on both the local
free-volume and on the free volume of the neighbors. This is because any increase or decrease in local
free-volume must be obtained from or passed on to a neighboring region. RSC 227 note that in a
strict cell or lattice model the free volume would be required to diffuse to or from the surface for any
net change to occur in the global free-volume of the system. This is one aspect of the free-volume
models which is disturbing. Measurements by Braun and Kovacs.P? using bulk and powdered
samples, showed no effect of specimen geometry on the observed kinetics. RSC feel that, for real
systems, net changes in free volume are 'probably liberated and absorbed by changes in various local
strains of the system'. This problem is also discussed by Curro et al.,233 who used a different model
to describe the kinetics of recovery in glasses. In any event, there is no correlation between free-
volume content of neighboring regions in equilibrium, and RSC further assume that the correiation
remains weak away from equilibrium. Then the combined free-volume of the neighboring regions
can be approximated by the average global free-volume 1. The free volume which affects the
transition rate in the ith region of interest is

]; = [h + (Z - l)f]/z (101)

where Z is the number of regions needed to liberate or absorb the net free-volume changes in the
region of interest.
The rates of transition, At and Ai- ,are assumed to have the same dependence on local free-volume
as their global dependence. Therefore, RSC are able to use the Doolittle or WLF forms, described
Glass Formation and Glassy Behavior 355
previously, to represent the structure dependence of the transition rates

(102)

(103)

where R is an adjustable parameter, (~i+ 1/~Jl/2 enforces the miscroscopic reversibility condition
(equation 100) and f3 is the step between adjacent free-volume states.
With the above assumptions, RSC were able to compute volume-recovery curves which mimic
the major features of the glassy phenomenology. In Table 4 are shown the parameters used in the
calculations. Figure 38 shows the asymmetry of the calculated approach curves compared with the
experimental data of Kovacs"! for poly(vinyl acetate). In Figure 39 is shown a plot of log 'teff vs. b
calculated from the model compared with the data for poly(vinyl acetate). As was the case for the
phenomenological models, the r-effective paradox is not resolved.
An interesting aspect of the RSC model, however, is the prediction that the retardation spectrum
changes in shape as recovery proceeds. As shown in Figure 40, it appears to narrow as equilibrium is

-I

-2

Log time (s )

Figure 38 The departure from equilibrium ~ vs. log time following steps in temperature to 35°C from 40 °C (D) and 30°C
(0) for poly(vinyl acetate) The curves are computed from the RSC model (after ref. 225, with permission)

37. 5°C 40°C

. .Q)
4
0-
0
...J

7
-5 -4

103 X 8

Figure 39 Computed curves (from RSC model) and experimental data for poly(vinyl acetate) for r ef f vs. departure from
equilibrium ~. Values of T, and To are indicated (after ref. 225, with permission)
356 Glass Formation and Glassy Behavior

Table 4 Parameters Used to Calculate the Kinetics of Volume


Recovery in the Robertson-Simha-Curro Model (After Ref. 225)

N, = 26 Number of segments in free-volume region


C t = 15.6 }
Time-temperature shift parameters (ref. 57)
C 2 = 46.8 K
r, = 308 K Glass transition temperature at 100 kPa
/g = 0.08134 Free volume at t; and 100kPa
L g = 3600 s Retardation time at Tg
Z = 13 Size-ratio controlling free volume

0.6


•/
CD 0.4
'0
Z
Q..
E
0
CD
.~
0
'i 0.2
0::

Log T

Figure 40 Retardation time spectra at stages during the volume recovery of poly(vinyl acetate) following a temperature step
from 40°C to 30 °C. Progression from the initial spectrum (D) immediately after the step to the final spectrum (.) shows
narrowing as volume recovery proceeds (after ref. 225, with permission)

approached in the contraction experiment. This is attributed to a change in the distribution of the
free volume as equilibrium is approached. Such behavior may mean that a more complicated free-
volume treatment is required than the one presented here, as the observed experimental behavior
appears to support a broadening rather than a narrowing of the spectrum.209-212 (Recall that both
f3 and x changed in the phenomenological Narayanaswamy! 7 -Moyniharr'" model.)
The free-volume model of Robertson, Simha and Curro allows one to examine the role of free-
volume fluctuations on the kinetics of glassy recovery and provides a detailed model of the physics of
such processes. An interesting extension of this type of calculation might be a non-equilibrium
Flory-Huggins lattice calculation, i.e. an extension of the Gibbs-DiMarzio theory to the non-
equilibrium properties of glasses. This is particularly interesting since the Simha-Somcynsky cell
model, as mentioned previously, does not correctly describe the energetics of glasses but only the
P VT surface.

10.3.5 Computer Simulations


There has been considerable recent interest234-244 in carrying out computer simulations of the
behavior of glassy systems with the intention of understanding the kinetics by varying certain
parameters. It is interesting to note that, as implied in the phenomenological and free-volume
models, the computer simulations also imply that an important component in glassy kinetics is
cooperativity of motion. A detailed treatment of the models is beyond the range of this article;
however, an interesting finding was made recently by Fredricksoni:" when he compared the kinetics
from the computer simulation with those given by the Narayanaswamy'Pvlvloyniharr'" phenom-
Glass Formation and Glassy Behavior 357

enological model. He found that, for certain cases in which the phenomenological equations did not
describe the simulation, the deviations were similar to those observed in real experiments. However,
Fredericksorr'P" does note that the simulations are versatile enough that one may be able to
simulate almost anything. Apparently more work is required before simulations reach the stage of
sophistication where new physical insights will be obtained.

10.3.6 Physical Aging


In the above sections we have described the kinetics of the structural recovery of glasses from both
phenomenological and microscopic viewpoints. As the structure (free volume, configurational
entropy, etc.) of the non-equilibrium glass evolves in response to thermal history, the mechanical,
dielectric, etc. properties also change. The changes in properties accompanying the structural
changes of glasses have been known for some time,224 but were intensively studied by Struik!" and
labeled 'physical aging' in his extensive treatise on the subject. A lengthy review of physical aging is
outside the scope of this article and the reader is referred to Struik's14.24 7 work and the more recent
work of Plazek and Berry248 for reviews of the subject. However, a few comments about aging are
worth making.
The picture of aging formulated by Struik!" is based on the postulate that, as the volume of the
glass decreases with increasing aging-time after a quench (cf Figure 3), the change in the glassy
structure changes the viscoelastic spectrum governing the material response (e.g. mechanical,
dielectric) by shifting it along the time axis by an amount ate where t, denotes the time after the
quench. Such behavior can be seen in the creep-compliance data presented in Figure 4 for poly(vinyl
chloride) quenched from above T g to 20 "C. Here we see that, shortly after the quench, the
compliance is greater than at longer aging times teo The creep curves can be superimposed by a shift
along the time axis, i.e. the retardation spectrum shifts but does not change shape. The amount of
shift per decade of time, u, was found to be approximately equal to unity, i.e.
d log a.,
~ = ------ ~ 1 (104)
d log t,

for many systems in the linear viscoelastic range.


Importantly, for this type of analysis to work, the duration of the mechanical measurement must
be short relative to the total aging time, so that the tests are carried out at fixed glassy structure.!"
i.e. t 1/te ~ 0.10. Under such conditions, one can consider the measurement of the mechanical or
dielectric response to be a probe of the glassy structure. In fact, Kovacs 19 has proposed associating
ate from the mechanical or dielectric measurements with the aa required for the phenomenological
theories as a means of independently determining the non-linearity parameter x discussed previously
(see equation 83).
Struik14also investigated the change with aging of the large stress (or deformation) properties of
glasses. Along with other workers he reports two additional phenomena: (i)the viscoelastic spectrum

Q) High stress Small stress


o
c: q' a
.2 ,
Q.
E f.'e
o
o
Q.
Shifts due
Q) ~ to aging
Q)
~
o
.!
en Shifts due
c:
f to stress
~
o
~
e
Q)
J::.
(J)

Log creep ti me

Figure 41 Schematic representation of the relative effects of aging on the small-stress and high-stress creep compliance of a
glass (after ref. 14, with permission)
358 Glass Formation and Glassy Behavior

1.0

o'-
CD

...s:...
0.5

C/)

o 40

Figure 42 Variation of the rate of aging, Jl. (i.e. double logarithmic rate of change of the shift factor) with maximum shear
stress for PVC quenched from 90°C to various temperatures: 0, 60; ~,50; \1,40; D, 20; 0, -20°C (after ref. 14, with
permission)

is shifted 14, 15,248-250 to shorter times by large stresses or deformations (see Figure 41); and (ii) the
rate of aging J..l decreases dramatically as the level of stress (or deformation) increases 14,251,252
(see Figure 42).
Struik!" attributed the above effects to 'erasure' of the aging due to the mechanical deformation,
possibly due to change in the structure of the glass created by the mechanical energy dissipated by
the system. This contention has been supported by work of Smith and co-workers P ? and Vee and
co-workers. 254,255 However, Myers et al. 2 5 6 claim the opposite result, i.e. an enhancement of the
aging response due to the applied deformation, and McKenna and Zapas 257,258 have argued
against the Struik interpretation on the grounds that non-linear memory effects explain, from simple
viscoelasticity considerations, many of the observed 'rejuvenation' or 'erasure' phenomena.
Although the basic phenomenon of physical aging, i.e. the impact of structural changes in a glass
on the viscoelastic or dielectric response, are well understood, particularly in the linear
range, 14,247-248,259-261 there are details of behavior, such as aging effects near subglass
relaxations262-264 and changes in spectral shape,264-267 which need to be clarified. Finally, the
coupling between aging effects, large mechanical deformations, the structure of the glass and non-
linear viscoelastic properties has become an area of great interest recently,251-263,268-275 and merits
considerable research effort in the future.

ACKNOWLEDGEMENTS
The author appreciates many fruitful discussions during the writing of this article with E. A.
DiMarzio. Thanks are also due to the Polymers Division and the Institute for Materials Science and
Engineering of the National Bureau of Standards, which made the time available to produce this
work.

10.4 REFERENCES
1. A computer search of 'Chemical Abstracts' (American Chemical Society, Washington, DC) using DIALOG (DIALOG
Information Services) for 1967 to May 1987 resulted in the numbers quoted.
2. R. N.Haward, 'The Physics of Glassy Polymers', Wiley, New York, 1973.
3. 'Physical Aging Processes in Molecular and Atomic Glasses', NATAS Conference, 1983, Special Issue of Polym. Eng.
Sci., 1984, 24 (14).
4. M. R. Tant and G. L. Wilkes, Polym. Eng. Sci., 1981, 21, 874.
5. G. W. Scherer, 'Relaxation in Glass and Composites', Wiley, New York, 1986.
6. M. Goldstein and R. Simha (eds.), Ann. N. Y. Acad. Sci., 1976, 279.
7. J. M. O'Reilly and M. Goldstein (eds.), Ann. N. Y. Acad. Sci., 1981,371.
8. C. A. Angell and M. Goldstein (eds.), Ann. N. Y. Acad. Sci., 1986, 484.
9. K. L. Ngai and G. B. Wright (eds.),'Relaxations in Complex Systems', U. S. Government Printing Office, Washington,
DC, 1984(available from National Technical Information Service, 5285 Port Royal Road, Springfield, VA 22161, USA).
10. P. H. Gaskell (ed.), 'The Structure of Non-Crystalline Materials', Taylor and Francis, London, 1977.
11. A. J. Kovacs, J. Polym. Sci., 1958, 30, 131.
12. S. S. Chang, J. Polym. Sci., Po/ym. Symp., 1984,71,59.
13. A. J. Kovacs, Fortschr. Hochpolym.-Forsch., 1964, 3, 394.
14. L. C. E. Struik, 'Physical Aging in Amorphous Polymers and Other Materials', Elsevier, Amsterdam, 1978.
Glass Formation and Glassy Behavior 359

15. J. M. Crissman and G. B. McKenna, J. Polym. Sci., Po/ym. Phys. Ed., 1987, 25, 1667.
16. E. A. DiMarzio, Ann. N. Y. Acad. Sci., 1981, 371, 1.
17. O. S. Narayanaswamy, J. Am. Ceram. Soc., 1971, 54, 491.
18. A. J. Kovacs, J. J. Aklonis, J. M. Hutchinson and A. R. Ramos, J. Polym. Sci., Po/ym. Phys. Ed., 1979, 17, 1097.
19. A. J. Kovacs, Ann. N. Y. Acad. Sci., 1981, 371, 38.
20. C. T. Moynihan, P. B. Macedo, C. 1. Montrose, P. K. Gupta, M. A. DeBolt, J. F. Dill, B. E. Dom, P. W. Drake, A. J.
Easteal, P. B. Elterman, R. P. Moeller, H. Sasabe and J. A. Wilder, Ann. N. Y. Acad. Sci., 1976,279, 15.
21. R. W. Rendell, J. J. Aklonis, K. L. Nagi and G. R. Fong, Macromolecules, 1987, 20, 1070.
22. W. Kauzmann, Chem. Rev., 1948, 43, 219.
23. N. F. Mott and R. W. Gurney, Trans. Faraday Soc., 1939, 35, 364.
24. J. H. Gibbs, J. Chem. Phys., 1956, 25, 185.
25. 1. H. Gibbs and E. A. DiMarzio, J. Chem. Phys., 1958, 28, 373.
26. E. A. DiMarzio and J. H. Gibbs, J. Chem. Phys., 1958, 28, 807.
27. A. J. Batschinski, Z. Phys. Chem. (Leipzig), 1913,84, 643.
28. T. G. Fox, Jr. and P. J. Flory, J. Appl. Phys., 1950, 21, 581.
29. A. K. Doolittle, J. Appl. Phys., 1951, 22, 1031.
30. I. Prigogine and R. Defay, 'Thermodynamique Chimique', Edition Desoer, Liege, Belgium, 1950.
31. P. Ehrenfest, Proc. K.Ned. Akad. Wet., 1933,36, 153.
32. R. A. Swalin, 'Thermodynamics of Solids', 2nd edn., Wiley, New York, 1972.
33. J. J. Aklonis and W. J. MacKnight, 'Introduction to Polymer Viscoelasticity', Wiley, New York, 1983.
34. P. G. Shewmon, 'Transformations in Metals', McGraw-Hill, New York, 1969.
35. G. Rehage and W. Borchard, in 'The Physics of Glassy Polymers', ed. R. N. Haward, Wiley, New York, 1973, p. 54.
36. F. T. Gucker and R. L. Seifert, 'Physical Chemistry', Norton, New York, 1966, p. 386.
37. J. E. McKinney and M. Goldstein, J. Res. Natl. Bur. Stand., Sect. A, 1974, 78A, 331.
38. G. Gee, Polymer, 1966,7, 177.
39. E. Passaglia and G. M. Martin, J. Res. Natl. Bur. Stand., Sect. A, 1964, 68A, 273.
40. P. Zoller, J. Polym. Sci., Po/ym. Phys. Ed., 1982, 20, 1453.
41. H.-J. Oels and G. Rehage, Macromolecules, 1977,10, 1036.
42. G. Rehage and H.-J. Oels, High Temp. - High Pressures, 1977, 9, 545.
43. C. A. Angell and W. Sichina, Ann. N. Y. Acad. Sci., 1976,279,53.
44. C. A. Angell, E. Williams, K. J. Rao and J. C. Tucker, J. Phys. Chem., 1977,81,238.
45. M. Goldstein, J. Phys. Chem., 1973, 77, 667.
46. J. M. O'Reilly, J. Polym. Sci., 1962, 57, 429.
47. J. E. McKinney, Ann. N. Y. Acad. Sci., 1976, 279, 88.
48. I. Havlicek, Macromolecules, 1981, 14, 1595.
49. A. 1. Staverman, Rheol. Acta, 1966, 5, 283.
50. R.-J. Roe and A. E. Tonelli, Macromolecules, 1978, 11, 114.
51. A. Quach and R. Simha, J. Phys. Chem., 1972, 76, 416.
52. M. Goldstein, J. Chem. Phys., 1963, 39, 3369.
53. R. o. Davies and G. o. Jones, Adv. Phys., 1953, 2, 370.
54. E. Williams and C. A. Angell, J. Phys. Chem., 1977,81,232.
55. K. D. Pae, C.-L. Tang and E.-S. Shin, J. Appl. Phys., 1984,56,2426.
56. C. T. Moynihan, A. J. Easteal, M. A. DeBolt and J. Tucker, J. Am. Ceram. Soc., 1976, 59, 12.
57. 1. D. Ferry, 'Viscoelastic Properties of Polymers', 3rd edn., Wiley, New York, 1980.
58. E. A. DiMarzio, J. H. Gibbs, P. D. Fleming III and I. C. Sanchez, Macromolecules, 1976, 9, 763.
59. E. A. DiMarzio, J. Res. Natl. Bur. Stand., Sect. A, 1964, 68A, 611.
60. E. A. DiMarzio, J. Appl. Phys., 1974, 45, 4143.
61. E. A. DiMarzio, Macromolecules, 1977, 10, 1407.
62. E. A. DiMarzio, in 'Relaxations in Complex Systems', ed. K. L. Ngai and G. B. Wright, U. S. Government Printing
Office, Washington, DC, 1985, p. 43 (available from National Technical Information Service, 5285 Port Royal Road,
Springfield, VA 22161, USA).
63. M. Goldstein, J. Appl. Phys., 1975, 46, 4153.
64. M. Goldstein, Macromolecules, 1977, 10, 1407.
65. P. J. Flory, 'Principles of Polymer Chemistry', Cornell University Press, Ithaca, NY, 1953.
66. M. L. Huggins, Ann. N. Y. Acad. Sci., 1942, 43, 1.
67. A. R. Miller, 'The Theory of Solutions of High Polymers', Oxford University Press, London, 1948.
68. T. G. Fox and P. 1. Flory, J. Polym. Sci., 1954, 14, 315.
69. R. B. Beevers and E. F. T. White, Trans. Faraday Soc., 1960,56,744.
70. G. Pezzin, F. Zilio-Grandi and P. Sammartin, Eur. Polym. J., 1970, 6, 1053.
71. R. B. Beevers, J. Polym. Sci., Part A, 1964, 2, 5257.
72. A. Eisenberg and T. Sasada, in 'Proceedings of the Conference on Physics of Non-Crystalline Solids', ed. 1. A. Prins,
North-Holland, Amsterdam, 1965, p. 99.
73. R. Larrain, L. H. Tagle and F. R. Draz, Polym. Bull. (Berlin), 1981, 4, 487.
74. T. G. Fox and S. Loshaek, J. Polym. Sci., 1955, 15, 371.
75. E. V. Thompson, J. Polym. Sci., Part A-2, 1966, 4, 199.
76. T. Hatakeyama and M. Serizawa, Polym. J., 1982, 14, 51.
77. F. Bastard and B. Jasse, Polym. Bull. (Berlin), (1982), 7, 331.
78. K. Ueberreiter and G. Kanig, J. Colloid Sci., 1952, 7, 569.
79. B. E. Read, Polymer, 1962, 3, 529.
80. J. A. Faucher, J. Polym. Sci., Part B, 1965, 3, 143.
81. J. M. Widmaier and G. C. Meyer, Macromolecules, 1981,14,450.
82. F. 1. Bueche, 'Physical Properties of Polymers', Interscience, New York, 1962.
83. T. Somcynsky and D. Patterson, J. Polym. Sci., 1962, 62; 151.
360 Glass Formation and Glassy Behavior

84. J. M. G. Cowie and P. M. Toporowski, Eur. Polym. J., 1968, 4, 621.


85. C. M. Guttmann and E. A. DiMarzio, Macromolecules, 1988, in press.
86. I. Havlicek, V. Vojta, M. Ilavsky and J. Hrouz, Macromolecules, 1980,13, 357.
87. G. C. Berry, Macromolecules, 1980,13, 550.
88. E. A. DiMarzio and F. Dowell, J. Appl. Phys., 1979, SO, 6061.
89. G. M. Martin and L. Mandelkern, J. Res. Natl. Bur. Stand. (U. S.), 1959,62, 141.
90. K. Uebrreiter and G. Kanig, J. Chem. Phys., 1950,18,399.
91. S. Loshaek, J. Polym. Sci., 1955, 15, 391.
92. L. A. Wood, J. Res. Natl. Bur. Stand., Sect. A, 1972, 76A, 51.
93. J. M. Charlesworth, J. Macromol. Sci., Phys., 1987, 826, 105.
94. A. Schiraldi, P. Baldini and E. Pezzati, J. Therm. Anal., 1985, 30, 1343.
95. T. S. Ellis, F. E. Karasz and G. ten Brinke, J. Appl. Polym. Sci., 1983, 28, 23.
96. X. Jin, T. S. Ellis and F. E. Karasz, J. Polym. Sci., Polym. Phys. Ed., 1984, 22, 1701.
97. L. A. Wood, J. Res. Natl. Bur. Stand. (U. S.), 1979,84,353.
98. L. R. G. Treloar, 'The Physics of Rubber Elasticity', 3rd edn., Oxford University Press, Oxford, 1975.
99. G. Gee, P. N. Hartley, 1. B. M. Herbert and H. A. Lanceley, Polymer, 1960, 1, 365.
100. W. V. Johnston and M. Shen, J. Polym. Sci., Part A-2, 1969, 7, 1983.
101. P. Mason, Trans. Faraday Soc., 1959, 55, 1461.
102. R. S. Witte and R. L. Anthony, J. Appl. Phys., 1951, 22, 689.
103. J. R. Stevens and D. G. Ivey, J. Appl. Phys., 1958, 29, 1390.
104. K. Oshima and H. Kusumoto, J. Chem. Phys., 1956, 24, 913.
105. R. K. Chan, K. Pathmanathan and G. P. Johari, J. Phys. Chem., 1986, 90, 6538.
106. T. S. Chow, Ferroelectrics, 1980, 30, 139.
107. G. ten Brinke, F. E. Karasz and T. S. Ellis, Macromolecules, 1983, 16, 244.
108. J. M. Gordon, G. B. Rouse, J. H. Gibbs and W. M. Risen, Jr., J. Chem. Phys., 1977, 66, 4971.
109. C. A. Angell, J. M. Sare and E. 1. Sare, J. Phys. Chem., 1978,82,2622.
110. E. A. DiMarzio and J. H. Gibbs, J. Polym. Sci., Part A, 1963, 1, 1417.
111. J. G. de la Campa, J. Guzman, J. de Abajo and E. Riande, Makromol. Chem., 1981,182,3163.
112. E. A. DiMarzio and J. H. Gibbs, J. Polym. Sci., 1959, 40, 121.
113. H. A. Schneider and M.-J. Brekner, Polym. Bull. (Berlin), 1985, 14, 173.
114. P. Zoller and H. H. Hoehn, J. Polym. Sci., Polym. Phys. Ed. 1982, 20, 1385.
115. P. R. Couchman and F. E. Karasz, Macromolecules, 1978,11,117.
116. P. R. Couchman, Polym. Eng. Sci., 1984,24, 135.
117. P. R. Couchman, Macromolecules, 1978,11,1156.
118. M. Goldstein, Macromolecules, 1985,18,177.
119. A. E. Tonelli, Macromolecules, 1977,10,633.
120. A. E. Tonelli, Macromolecules, 1977, 10, 716.
121. J. C. Wittmann and A. J. Kovacs, J. Polym. Sci., Part C, 1969, 16, 4443.
122. J. H. Hildebrand, 'Viscosity and Diffusivity: A Predictive Treatment', Wiley, New York, 1977.
123. J. H. Hildebrand, Science (Washington, D. C.), 1971,174,490.
124. A. K. Doolittle, J. Appl. Phys., 1951, 22, 1471.
125. A. K. Doolittle, J. Appl. Phys., 1952, 23, 236.
126. A. K. Doolittle and D. B. Doolittle, J. Appl. Phys., 1957, 28, 901.
127. T. G. Fox, Jr. and P. 1. Flory, J. Am. Chem. Soc., 1948, 70, 2384.
128. T. G. Fox, Jr. and P. J. Flory, J. Polym. Sci., 1954, 14, 315.
129. T. G. Fox, Jr. and P. J. Flory, J. Phys. Chem., 1951,55, 221.
130. A. Bondi, J. Phys. Chem., 1954, 16, 929.
131. F. Bueche, J. Chem. Phys., 1953, 21, 1850.
132. F. Bueche, J. Chem. Phys., 1956, 24, 418.
133. F. Bueche, J. Chem. Phys., 1959, 30, 738.
134. F. Bueche, J. Chem. Phys., 1962, 36,2940.
135. J. I. Frenkel, 'Kinetic Theory of Liquids', Clarendon Press, Oxford, 1946, p. 93.
13~. H. Eyring, J. Chem. Phys., 1936, 4, 283.
137. M. L. Williams, R. F. Landel and 1. D. Ferry, J. Am. Chem. Soc., 1955, 77, 3701.
138. J. D. Ferry and R. A. Stratton, Kolloid-Z., 1960, 171, 107.
139. H. Vogel, Phys. Z., 1921,22,645.
140. G. S. Fulcher, J. Am. Ceram. Soc., 1925, 8, 339.
141. G. Tammann and W. Hesse, Z. Anorg. AUg. Chem., 1926, 156, 245.
142. R. W. Fillers and N. W. Tschoegl, Trans. Soc. Rheol., 1977, 21, 51.
143. W. K. Moonan and N. W. Tschoegl, Macromolecules, 1983, 16, 55.
144. W. K. Moonan and N. W. Tschoegl, Int. J. Polym. Mater., 1984, 10, 199.
145. W. K. Moonan and N. W. Tschoegl, J. Polym. Sci., Polym. Phys. Ed., 1985, 23, 623.
146. J. Koppelmann, in 'Proceedings of the International Congress on Rheology, 4th', ed. E. H. Lee and A. L. Copley, Wiley,
New York, 1965, vol. 3, p. 361.
147. P. B. Macedo and T. A. Litovitz, J. Chem. Phys., 1965, 42, 245.
148. N. H. Nachtrieb and J. Petit, J. Chem. Phys., 1956,24,746.
149. R. E. Hoffman, J. Chem. Phys., 1952,20, 1567.
150. J. D. McKenzie, J. Chem. Phys., 1958, 28, 1037.
151. R. M. Barrer, Trans. Faraday Soc., 1942,38,322.
152. R. M. Barrer, Trans. Faraday Soc., 1943, 39, 48.
153. M. H. Cohen and D. Turnbull, J. Chem. Phys., 1959,31,1164.
154. D. Turnbull and M. H. Cohen, J. Chem. Phys., 1961, 34, 120.
155. N. Hirai and H. Eyring, J. Appl. Phys. 1958, 29, 810.
156. N. Hirai and H. Eyring, J. Polym. Sci., 1959,37,51.
Glass Formation and Glassy Behavior 361

157. R. P. Smith, J. Polym. Sci., Part A-2, 1970, 8, 1337.


158. M. H. Cohen and G. S. Grest, Phys. Rev. B: Condens. Matter, 1979, 20, 1077.
159. G. S. Grest and M. H. Cohen, Phys. Rev. B: Condens. Matter, 1980, 21, 4113.
160. G. S. Grest and M. H. Cohen, Adv. Chem. Phys., 1981,48,455.
161. M. H. Cohen and G. S. Grest, Ann. N. Y. Acad. Sci., 1981,371, 199.
162. Y. Lipatov, Adv. Polym. Sci., 1978, 26, 63.
163. R. N. Haward, J. Macromol. Sci., Rev. Macromol. Chem., 1970, C4, 191.
164. R. Simha and T. Somcynsky, Macromolecules, 1969,2,342.
165. T. Somcynsky and R. Simha, J. Appl. Phys., 1971,42,4545.
166. V. S. Nanda, R. Simha and T. Somcynsky, J. Polym. Sci., Part C, 1966,12,277.
167. R. Simha, Polym. Eng. Sci., 1980, 20, 82.
168. P. S. Wilson and R. Simha, Macromolecules, 1973, 6, 903.
169. R. Simha and P. S. Wilson, Macromolecules, 1973, 6, 908.
170. J. E. McKinney and R. Simha, Macromolecules, 1976,9,430.
171. A. Quach and R. Simha, J. Appl. Phys., 1971, 42, 4592.
172. J. E. McKinney and R. Simha, J. Res. Natl. Bur. Stand., Sect. A, 1977, 81A, 283.
173. 1. E. McKinney and R. Simha, Macromolecules, 1974~ 7, 894.
174. R. Simha, Macromolecules, 1977, 10, 1025.
175. R. Simha and C. E. Weil, J. Macromol. Sci., Phys., 1970,84,215.
176. S. T. J. Peng, R. F. Landel, J. Moacanin, R. Simha and E. Papazoglou, J. Rheol., 1987,31,125.
177. E. Papazoglou and R. Simha, J. Rheol. (N. Y.), 1987, 31, 135.
178. R. Simha, Ann. N. Y. Acad. Sci., 1976,279,2.
179. C. Feger and W. J. MacKnight, Macromolecules, 1985,18,280.
180. M. Gordon and W. Simpson, Polymer, 1961,2,383.
181. E. Jenckel and R. Heusch, Kolloid-Z., 1953, 130, 89.
182. L. J. Garfield and S. E. Petrie, J. Phys. Chem., 1964,68, 1750.
183. G. Braun and A. J. Kovacs, in 'Proceedings of the Conference on Physics of Non-Crystalline Solids', ed. J. A. Prins,
North-Holland, Amsterdam, 1965, p. 303.
184. F. N. Kelley and F. Bueche, J. Polym. Sci., 1961, SO, 549.
185. M. Gordon and J. S. Taylor, J. Appl. Chem., 1952, 2, 493.
186. I. J. McEwen and A. F. Johnson, in 'Alternating Copolymers', ed. J. M. G. Cowie, Plenum Press, New York, 1985, p. 239.
187. J. A. Manson and L. H. Sperling, 'Polymer Blends and Composites', Plenum Press, New York, 1976.
188. W. J. MacKnight, J. Stoelting and F. E. Karasz, Adv. Chem. Ser., 1971, 99, 29.
189. A. Q. Tool, J. Am. Ceram. Soc., 1946, 29, 240.
190. A. Q. Tool, J. Res. Natl. Bur. Stand. (U. S.), 1946, 37, 73.
191. A. J. Kovacs, J. M. Hutchinson and J. J. Aklonis, in 'The Structure of Non-Crystalline Materials', ed. P. H. Gaskell,
Taylor and Francis, London, 1977, p. 153.
192. J. M. Hutchinson and A. J. Kovacs, in 'The Structure of Non-Crystalline Materials', ed. P. H. Gaskell, Taylor and
Francis, London, 1977, p. 167.
193. A. 1. Kovacs and J. M. Hutchinson, J. Polym. Sci., Polym. Phys. Ed., 1979, 17, 2031.
194. 1. M. Hutchinson, J. J. Aklonis and A. J. Kovacs, Polym. Prepr., Am. Chem. Soc., Div. Polym. Chem., 1975, 16 (2),94.
195. M. A. DeBolt, A. J. Est~al, P. B. Macedo and C. T. Moynihan, J. Am. Ceram. Soc., 1976, S9, 16.
196. C. T. Moynihan and A. V. Lesikar, Ann. N. Y. Acad. Sci., 1981, 371, 151.
197. J. M. Hutchinson and A. J. Kovacs, J. Polym. Sci., Polym. Phys., Ed., 1976, 14, 1575.
198. H. Leaderman, 'Elastic and Creep Properties of Filamentous Materials and other High Polymers', Textile Foundation,
Washington, DC, 1943.
199. F. Kolrausch, Pogg. Ann. Phys., 1847, 12, 393.
200. G. Williams and D. C. Watts, Trans. Faraday Soc., 1970,66,80.
201. I. M. Hodge and A. R. Berens, Macromolecules, 1982, IS, 762.
202. G. Adam and J. H. Gibbs, J. Chem. Phys., 1965;43, 139.
203. I. M. Hodge and A. R. Berens, Macromolecules, 1981,14, 1598.
204. I. M. Hodge, Macromolecules, 1983, 16, 898.
205. I. M. Hodge and A. R. Berens, Macromolecules, 1985, 18, 1980.
206. J. J. Tribone, J. M. O'Reilly and J. Greener, Macromolecules, 1986, 19, 1732.
207. R. R. Lagasse, R. E. Cohen and A. Letton, J. Polym. Sci., Polym. Phys. Ed., 1982, 20, 375.
208. J. M. Hutchinson and A. J. Kovacs, J. Polym. Sci., Polym. Phys. Ed., 1983, 21, 2419.
209. I. M. Hodge and G. S. Huvard, Macromolecules, 1983, 16, 371.
210. W. M. Prest, Jr., F. 1. Roberts, Jr. and I. M. Hodge, in 'Proceedings of the North American Thermal Analysis Society
Conference, 12th, Williamsburg, VA', ed. J. C. Buck, 1983, p. 119 (available from NATAS, A-I Business Service, 219
Park Avenue, Scotch Plains, NJ, USA).
211. W. M. Prest, Jr., F. 1. Roberts, Jr., D. M. Teegarten and K. M. Sheriden, in 'Proceedings of the North American Thermal
Analysis Society Conference, 12th, Williamsburg, VA', ed. J. C. Buck, 1983, p. 251 (available from NATAS, A-I Business
Service, 219 Park Avenue, Scotch Plains, NJ, USA).
212. R. W. Rendell, T. K. Lee and K. L. Ngai, Polym. Eng. Sci., 1984, 24, 1104.
213. D. Ng and J. J. Aklonis, in 'Relaxations in Complex Systems', ed. K. L. Ngai and G. B. Wright, U. S. Government
Printing Office, Washington, DC, 1985, p. 53 (available from National Technical Information Service, 5285 Port Royal
Road, Springfield, VA 22161, USA).
214. M. Goldstein and M. Nakonecznyj, Phys. Chem. Glasses, 1965, 6, 126.
215. K. L. Ngai, Comments Solid State Phys., 1979, 9, 127.
216. K. L. Ngai, Comments Solid State Phys., 1980, 9, 141.
217. R. W. Rendell and K. L. Ngai, in 'Relaxations in Complex Systems', ed. K. L. Ngai and G. B. Wright, U. S. Government
Printing Office, Washington, DC, 1985, p. 309 (available from National Technical Information Service, 5285 Port Royal
Road, Springfield, VA 22161, USA).
218. K. L. Ngai and G. Fytas, J. Polym. Sci., Polym. Phys. Ed., 1986, 24, 1683.
362 Glass Formation and Glassy Behavior

219. K. L. Ngai and D. J. Plazek, J. Polym. Sci., Po/ym. Phys. Ed., 1985, 23, 2159.
220. G. B. McKenna, K. L. Ngai and D. J. Plazek, Polymer, 1985, 26, 1651.
221. R. W. Rendell, K. L. Ngai and G. B. McKenna, Macromolecules, 1987, 20, 2250.
222. K. L. Ngai and D. J. Plazek, J. Polym. Sci., Polym. Phys. Ed., 1986, 24, 619.
223. K. L. Ngai, R. W. Rendell, A. K. Rajagopal and S. Teitler, Ann. N. Y. Acad. Sci., 1986,484,150.
224. A. J. Kovacs, R. A. Stratton and J. D. Ferry, J. Phys. Chem., 1963, 67, 152.
225. R. E. Robertson, R. Simha and J. G. Curro, Macromolecules, 1984, 17, 911.
226. R. Simha, J. G. Curro and R. E. Robertson, Polym. Eng. Sci., 1984, 24, 1071.
227. R. E. Robertson, R. Simha and J. G. Curro, Macromolecules, 1985,18, 2239.
228. R. E. Roberston, J. Polym. Sci., Polym. Symp., 1978, 63, 173.
229. R. E. Roberston, J. Polym. Sci., Polym. Phys. Ed., 1979, 17, 597.
230. R. E. Robertson, J. Appl. Phys., 1978, 49, 5048.
231. R. E. Robertson, Ann. N. Y. Acad. Sci., 1981, 371, 21.
232. G. Braun and A. J. Kovacs, Phys. Chem. Glasses, 1963, 4, 152.
233. J. G. Curro, R. R. Lagasse and R. Simha, Macromolecules, 1982, IS, 1621.
234. G. H. Fredrickson and H. C. Andersen, J. Chem. Phys., 1985,83,5822.
235. G. H. Fredrickson and S. A. Brawer, J. Chem. Phys., 1986, 84, 3351.
236. G. H. Fredrickson, Ann. N. Y. Acad. Sci., 1986,484, 185.
237. J. J. Fernandez, C. Z. Anderico and T. S. 1. Streit, J. Appl. Phys., 1982, 53, 7991.
238. C. Z. Anderico, J. F. Fernandez and T. S. J. Streit, Phys. Rev. B: Condens. Matter, 1982, 26, 3824.
239. A. Baumgartner, J. Chem. Phys., 1980, 73, 2489.
240. R. D. de la Batie, J.-L. Viovy and L. Monnerie, J. Chem. Phys. 1984, 81, 567.
241. S. M. Rekhson, D. M. Heyes, C. J. Montrose and T. A. Litovitz, J. Non-Cryst. Solids, 1980, 38/39, 403.
242. C. A. Angell, Ann. N. Y. Acad. Sci., 1981, 371, 136.
243. L. V. Woodcock, Ann. N. Y. Acad. Sci., 1981, 371, 274.
244. H. J. Raveche, Ann. N. Y. Acad. Sci., 1976, 279, 36.
245. G. B. McKenna, G. Hadziioannou, P. Lutz, G. Hild, C. Strazielle, C. Straupe, P. Rempp and A. J. Kovacs,
Macromolecules, 1987,20, 498.
246. S. J. Clarson, K. Dodgson and 1. A. Semlyen, Polymer, 1985, 26, 930.
247. L. C. E. Struik, Polymer, 1980, 21, 962.
248. D. J. Plazek and G. C. Berry, in 'Glass: Science and Technology, Volume 3, Viscosity and Relaxation', ed. D. R.
Uhlmann and N. J. Kriedl, Academic Press, Orlando, FL, 1986, p. 363.
249. S. Matsuoka, S. J. Aloisio and H. E. Bair, J. Appl. Phys., 1973, 44, 4265.
250. G. C. Berry, J. Polym. Sci., Polym. Phys. Ed. 1976, 14, 451.
251. G. B. McKenna and A. J. Kovacs, Polym. Eng. Sci., 1984, 24, 1138.
252. G. B. McKenna, in 'Relaxations in Complex Systems', ed K. L. Ngai and G. B. Wright, U. S. Government Printing
Office, Washington, DC, 1985, p. 129 (available from National Technical Information Service, 5285 Port Royal Road,
Springfield, VA 22161, USA).
253. T. L. Smith, T. Ricco, G. Levita and W. K. Moonan, Plast. Rubber Process. Appl., 1986,6,81.
254. A. F. Vee, R. J. Bankert, K. L. Ngai and R. W. Rendell, in 'Advances in Rheology 3. Polymers', ed. B. Mena, A. Garcia-
Rejon and C. Rangel-Nafaile, Universidad Nacional Autonoma de Mexico, 1984, p. 231.
255. A. F. Vee, R. J. Bankert, K. L. Ngai and R. W. Rendell, J. Polym. sa.. Polym. Phys. Ed., 1988, in press.
256. F. A. Myers, F. C. Cama and S. S. Sternstein, Ann. N. Y. Acad. Sci., 1976,279,94.
257. G. B. McKenna and L. J. Zapas, J. Polym. Sci., Polym. Phys. Ed., 1985, 23, 1647.
258. G. B. McKenna and L. J. Zapas, Polym. Eng. Sci., 1986, 26, 725.
259. D. L. Huntson, W. T. Carter and J. L. Rushford, in 'Developments in Adhesives-2', ed. A. J. Kinloch, Applied Science,
Barking, 1981, p. 125.
260. L. Guerdoux, R. A. Duckett and D. Froelich, Polymer, 1984,25,1392.
261. S. Matsuoka, J. Rheol. (N. Y.), 1986, 30, 869.
262. J. Heijboer, Ann. N. Y. Acad. Sci., 1976, 279, 104.
263. G. P. Johari, in 'Relaxations in Complex Systems', ed. K. L. Ngai and G. B. Wright, U. S. Government Printing Office,
Washington, DC, 1985, p. 17 (available from National Technical Information Service, 5285 Port Royal Road,
Springfield, VA 22161, USA).
264. B. E. Read and G. D. Dean, Polymer, 1984, 25, 1679.
265. N. G. McCrum, Polymer, 1984, 25, 309.
266. C. K. Chai and N. G. McCrum, Polymer, 1984, 25, 291.
267. N. G. McCrum, Polymer, 1984, 25, 299.
268. G. B. McKenna and L. J. Zapas, Polymer, 1985, 26, 543.
269. T. S. Chow, J. Rheol. (N. Y.), 1986, 30, 729.
270. R. M. Shay, Jr. and J. M. Caruthers, J. Rheol. (N. Y.), 1986,30, 781.
271. B. D. Snow, P. Potnis and D. C. Bogue, .1.. Rheol. (N. Y.), 1984, 28, 517.
272. B. D. Snow and D. C. Bogue, J. Rheol. (N. Y.), 1984,28, 533.
273. R. F. Landel and S. T. J. Peng, J. Rheol. (N. Y.), 1986, 30, 741.
274. J. T. Bendler, AlP Conf. Proc., 1985, 137, 227.
275. J. T. Bendler, D. G. LeGrand and W. V. Olszewski, in 'Transport and Relaxation in Random Materials', ed. J. Klafter,
R. J. Rubin and M. F. Shlesinger, World Scientific, Singapore, 1986, p. 240.

Anda mungkin juga menyukai