82 discussed below when calorimetric techniques are discussed.

Unfortunately, there is not complete agreement among practitioners on the preferred method of measurement, but it is of primal importance to specify the experiment as completely as is possible. In the following sections we describe some typical techniques for measuring the glass transition temperature and attempt to indicate the strengths and weaknesses of each. We also indicate the ability of such techniques to measure kinetic properties associated with the glassy structure. We do not look at methods such as dynamic mechanical or dielectric measurements as they have been widely described elsewhere and their interpretation is, in fact, more difficult than the methods presented here that provide a transition in "thermodynamic" properties.
5.2. Dilatometric methods

Dilatometry is a method for measuring volume. Although dilatometry is not the most common method of measurement in glass forming systems, we discuss it first here because the data first presented in the prior sections was obtained in very elegant experiments performed by A. J. Kovacs using fluid confinement dilatometry methods for a poly(vinyl acetate) polymer. Generally, dilatometric measurements can be made either in a confining fluid to obtain the volume change of the sample or through length change measurements. It is common, particularly in fluid confinement dilatometry experiments, to report the first derivative of the free energy state variable, the volume. The data can then, if appropriate, be analyzed in terms of the coefficient of thermal expansion. Hence, in Fig. 2, we see a volume-temperature plot and the Tx is estimated from the intersection of the slopes of lines drawn through the points at high temperature and low temperature (above and below the apparent transition). In addition, dilatometric techniques, including the length change methods, are often continuous or semi-continuous measurements in which volume (or its change relative to a reference) can be measured as functions of time, temperature or pressure. This can lead to very good accuracy in both the change and in the derivative of the change with time, temperature or pressure. (See, for example, references 9, 10, 11, 3 9, 40, 41 ).

5.2.1. Fluid confinement dilatometry

The methods used by Kovacs in his studies were based on the Bekkedahl [73] dilatometer depicted schematically in Fig. 24. The apparatus is essentially a quartz tube in which a sample is placed and then surrounded by an appropriate fluid. The tube is attached to a capillary of known (calibrated) diameter and volume changes are observed by measuring the height change of the fluid column in the capillary. However, despite the simplicity, the method is somewhat cumbersome and few people carry out such dilatometric measurements today because they are

83 not as convenient as other techniques. Perhaps the biggest disadvantages of the method lie in the need to evacuate the system in order to fill it with the confining fluid and the fact that, for polymers, the fluid should not interact with the polymer, which leads, commonly, to the use of mercury, a material whose physical properties are well known, as the confining fluid. Mercury has the advantage of excellent thermal conductivity there minimizes temperature gradients in the sample cell. Also, the temperature range over which mercury can be used is fairly broad (melting point is -39~ and the one atmosphere boiling point is 357~ ) [74]. Toxicity of the vapors, however, reduce the upper limit somewhat. An advantage of mercury confinement dilatometry is that the system can be immersed in a liquid bath, which lends itself to relatively easy temperature control and stability when high precision in the measurements is required. Typical estimates [40,41 ] for the accuracy of the Kovacs measurements, for example, are that the long time volume stability was approximately 2 x 10 .5 cm3/cm 3. This limitation is related to the fact that liquid mercury has a relatively high coefficient of thermal expansion and generally makes up 50 % or more of the total sample cell volume. Finally, the fluid confinement dilatometer has a significant advantage over linear measurements. Because the material is under very small hydrostatic stresses there is little tendency to change dimensions due to creep. Therefore, measurements are readily made through the glass transition where the material might flow. The reader is referred

Fig. 24. Schematic of the Bekkedahl fluid confinement dilatometer. (After Reference 73).

84 to Figures 2-4, 6-8, 10-14 for typical data taken using fluid confinement dilatometers. From the Figures 3, 4, 6 showing volume-temperature plots at different pressures, it is clear that the confinement technique also lends itself well to measurements under hydrostatic pressure. The actual measurement of the glass transition is made by taking the point of intersection of the liquid state v-T line with that of the glassy state v-T line. Typical errors in the estimates of Tg from isobaric measurements are reported to be of the order of+(1 to 2~ ), due primarily to the uncertainty of the estimates of the slopes in the v-T behavior and corresponding extrapolation of the uncertain lines to a point of intersection (See reference 11 for a discussion of the statistical methods used to make the error estimates). An advantage of the fluid confinement measurement is that it can be made on cooling, which reduces some of the difficulties which occur due to structural recovery in heating experiments. It is important, as in all T~ measurements, to fully specify the thermal history applied to the sample. For example, recalling Fig. 2, the Tg depends on the time-scale of the experiment. We also note that special, deformation dilatometers have been built for fundamental studies [75-78] of the effects of deformation on volume changes and could, conceivably, be used to examine the influence of deformation on the glass transition temperature.

5.2.2. Length change dilatometry

The most straight forward means of measuring the change in volume of an isotropic sample is to measure the change in length with temperature. This is evidently a popular method as reflected by three different standard test methods published by the ASTM [79-81 ]. Furthermore, when sample flow (creep) is not a potential problem, length change measurements can be more accurate than the fluid confinement techniques discussed above. [The reason for this is that, in the confinement method, the fluid volume changes make up a significant part of the measured volume change and these need to be taken into account.] The problem of sample flow is normally not a big effect far below the glass transition, but when the T~ is approached it becomes an issue. Most of the standard length change methods all require some sort of mass be applied to the sample. Furthermore, the mass of the sample itself can lead to flow when the viscosity ofthe sample becomes small enough. When flow becomes a problem, the fluid confinement methods are the best techniques because they measure the volume of the sample. The accuracy of the length change methods is determined by the ability to measure the absolute length, L, and the magnitude of the length change. If one is interested in the coefficient of thermal expansion, the interferometric methods lead to an estimated expanded uncertainty of 40 nm/m/K. Comparing this with a typical coefficient of thermal expansion of 3 .5 m/m/K, the relative expanded

85 uncertainty of the measurement is then approximately 1.3x10 3. This would assume, of course, that the sample did not flow and that the sample temperature is known perfectly. Additional errors in the measurement would result from fluctuations or gradients in temperature. A typical uncertainty, which is valid for all of the methods of measurement, can be estimated based on the coefficient of thermal expansion of the material and the sample length. Additionally, errors of measurement can be due to other factors, such as the actual temperature fluctuations, thermal expansion of the apparatus supports and, perhaps most importantly near to the glass transition, flow of the sample under the sample weight and weights of any instrument parts, such as mirrors in the interferometric method. In the case of less sophisticated methods [79,80] in which the length change is measured with a micrometer or a linear variable differential transformer (LVDT) displacement transducer, one needs to be concerned with flow and with the sample geometry. For example, a resolution of the sample length change of 1 gm would correspond to a 3~ temperature change for the above material ifthe sample length were approximately 1 cm. On the other hand, measurements with thin films by such a technique would require temperature accuracy of approximately 3 x 105~ if the film had a thickness of 100 gm. Clearly, such "push rod" methods are not made for measuring the change of coefficient of thermal expansion of thin films in the thickness direction. On the other hand, for research purposes, the expansion of films of thickness of hundreds ofnm can be measured using X-ray [82,83] or neutron [84] reflectivity or ellipsometry [85]. Such techniques are clearly not for routine use. In the case of measurement of the thermal expansion coefficients of films having thicknesses between several gm and 100's ofgm, there are interferometric methods [86] which are reasonably sophisticated. In addition, current research at NIST [87] and elsewhere [88] also suggests that very careful capacitance methods may be possible for measurement of the coefficient of thermal expansion in the thickness direction of thin films. Also, for kinetic studies of structural recovery, dilatometry is the only method for which it is relatively easy to obtain more or less continuous data. When structural recovery is measured in, for example, differential scanning calorimetry (described below), one is required to make a separate measurement for each data point in time. This can be very cumbersome and time consuming.

5.3. Calorimetric techniques Absolute calorimetry to measure the enthalpy is extremely difficult to carry out and it is much more common to measure the constant pressure heat capacity. Since the Cpis a second derivative (of the free energy) parameter, what is observed experimentally is a "jump" in going from the liquid to the glass. Modern scanning calorimetric apparatuses exist that make such measurements easy. In addition,