The lattice specific heat of minerals at T 1000 K is close to the classic limit 3Rn = 3RM/m (the molar value, where R is the gas constant and n is the number of atoms in chemical formula). From calorimetry, we have information on the isobaric heat capacity CP, which exceeds CV by 1-3% at 300 K and 10-15% at T > Q (Table 2 and Pankov et al., 1997]. Since m 20-22 g/mole for mantle minerals, the classic value of CV for them is 1.13-1.25 J/g K. Depending on mineral, the high-temperature anharmonic corrections to CV become singnificant either near the melting point or even at room temperature (sometimes, at T equal 1/6 of the melting point) [Mulargia and Boschi, 1980; Quareni and Mulargia, 1988; Reynard et al., 1992; Fiquet et al., 1992].
From the identities 2E/ V T = 2E/ T V and 2S/ P T = 2S/ T P, using the Maxwell relations, we find
Note that (57) and (58) are the alternate forms of (33) and (34), repectively. The logarithmic volume derivatives at T = constant can be expressed as follows:
These identities were used to compute the derivative values given in Tables 3 and in Pankov et al. . O. Anderson et al.  noted that, for the Debye model, ( CV/ P)T 0 and therefore a 2dT - K.
The difference CP - CV satisfies the identity
i.e., it decreases with pressure for usual values of K and dT (see Table 3 and Pankov et al. ).
Birch  estimated the decrease in CP with pressure in the lower mantle, setting a 4 and agT 0.1. By the power law for CP, this gives a 13-16% decrease in CP along an isotherm, for x descending from 1.0 to 0.7. According to Birch, the maximum decrease in CP in the mantle does not appear to exceed 20%.
The adiabatic volume derivative of CP is
Substituting ( ln CP/ ln T)P 0.15 (the typical value for minerals for T 1000 K), g 1-1.5, agT 0.1, and ( ln CP/ ln V)T 0.5 (for a 4 ), we find ( ln CP/ ln V)S 0.3-0.4. Consequently, the power law for CP yields a 8-13% decrease of this value along the mantle adiabat (to x 0.7 ).
By using (29), the temperature derivative of CP can be represented in the form
The second term arising from the extrinsic anharmonicity can be estimated by making use of (58), so that (63) in conjuction with data for ( CP/ T)P allows the first term coming from the intrinsic anharmonicity to be evaluated. At room temperature, ( ln CP/ ln V)T is on the order of ( ln CP/ ln T)P, but nevertheless, the contribution of the second term to the sum (63) is small because of the small factor agT. At high temperatures, this contribution generally increases to 15-30% (perhaps, 60-70% for ilmenite and perovskite, according to our estimates) [Pankov et al., 1997]. In the classical limit, CV = constant, assuming that aa(V) and gg(V), we find ( CP/ T)V agCV.
Further, from (17) and (18), it is easy to obtain the identity
which we used to estimate the values of this derivative presented in Table 3 and Pankov et al. . Then, with the help of the identity
it is possible to compute (( ln CV)/( ln T))V, provided that the second term in (65) is given by (59).
An explicit dependence CV(V, T) can be derived from models and measurements of the vibrational spectra of solids (e.g., Pitzer and Brewer, 1961]):
where k is the Boltzman constant, y = hn/kT, g(n) is the spectrum density, and n are the lattice frequencies including optic and acoustic modes [Kieffer, 1979a, 1979b, 1979c, 1980; Hofmeister, 1991a, 1991b; Richet et al., 1992]. This method provides information on the inadequacy of the Debye theory and the related approximation g = -d lnQ/d ln V. The characteristic temperature Q in the Debye theory is usually estimated from acoustic data, but Q found from data for CV at T 300 K (labelled Qth ), on average, exceeds the acoustic Q (labelled Qa ) by about 20% (larger deviations are common to quartz and coesite, see Table 2 and Pankov et al.  and Watanabe ). Chopelas [1990b] found, however, good agreement of the Debye model with the spectrum data for MgO at pressures to 200 kbar, provided that Q Vg, i.e., q = 0.
Spectroscopic measurements at high pressures allow us to estimate the derivative ( CV/ P)T. For example, the data of Chopelas [1990a, 1990b] to 200 kbar show that CV linearly decreases with P, so that the gradient -( CV/ P)T is 17.6 10-3 ( T = 300 K) and 0.91 10-3 ( T = 1800 K) for MgO and 49.8 10-3 ( T = 300 K) and 3.4 10-3 ( T = 1800 K) J/(mole K kbar) for forsterite. With these values, CV being extrapolated (by the power volume dependence) to the maximum pressure P = 1357 kbar in the mantle will decrease 40-60, 8-12, and 2-3% on the 300, 1000, and 1700 K isotherms, respectively. Comparing these results with the decrease in CP estimated above, we verify that the difference CP-CV in the lower mantle must be exceedingly small. From the same Chopelas' data, using also the KT and CV values from Table 2, we find ( ln CV/ ln V)T = 0.77 ( T = 300 K) and 0.022 ( T = 1800 K) for MgO and 0.54 ( T = 300 K) and 0.022 ( T = 1700 K) for forsterite. These results are comparable to our estimates of this derivative from thermodynamic data (Table 3).
Calorimetric data for CP versus temperature at P = 0 are commonly fitted to various empirical expressions [e.g., Fei and Saxena, 1987; Berman, 1988; Saxena, 1989; Richet and Fiquet, 1991]:
Richet and Fiquet  showed that the last of the above formalas are favored but no one of them provides an accurate description of CP over a wide temperature range.
In addition, Figure 6 compares CP(T) found by the simpler formula used by Watanabe  to fit the measurement in the temperature interval of 350-700 K. An example of MgO shows that the extrapolation by this formula can lead to series errors.
As well as in the analysis of thermal expansivity in section 6.3, we calculated CP(T) (Figure 6) from the same Mie-Grüneisen EOS as was used to compute a. One can see that the theoretical curves can be reconciled with the data shown by varying parameter q in the limits 1-2. In so doing, we find q 0 for MgO, q = 1-2 for Al 2 O 3, and q 0.5 for forsterite. However, these values of q are not always consistent to data on a (see section 6.3), and this fact also suggests a certain inaccuracy of the Mie-Grüneisen EOS model.
The thermal pressure coefficient defined by (9) or (12) is the basic characteristic of the thermal pressure and can also be defined as t = ( Pth/ T)V. Note that t has also the meaning of the latent heat of expansion per 1 K. O. Anderson and his co-workers [O. Anderson, 1982, 1984, 1988; O. Anderson and Sumino, 1980; O. Anderson and Goto, 1989; O. Anderson et al., 1982, 1991, 1992a] paid special attention to this parameter, in particular, in relation to their development of the rectangular parallelepiped resonance technique for measuring elastic properties of minerals at high temperatures.
The basic identities for the derivatives of t can easily be derived from those given in sections 6 and 7. The following identities are especially suitable [Brennan and Stacey, 1979; Birch, 1978; O. Anderson and Yamamoto, 1987]:
Formula (68) is obtained by expanding the derivative at the left side and then by using (30); (69) is a consequence of (57), and (70) is easily derived by equating the second derivatives of P with respect to V and T taken in one order or another; finally, (71) follows from (31) and (70).
Combining (68) and (69), we have
According to (71), for the common inequality dT > K, parameter t decreases with pressure along an isotherm. However, at high pressure, the decrease can change to an increase, as, e.g., in the PIB model for MgO [O. Anderson et al., 1993]. It is clear from (68) that t increases with temperature at constant pressure, at least for T < Q.
As follows from (29), the logarithmic temperature derivative of t at P = constant is represented as
where the first, intrinsic anharmonic term is positive due to (69) and can be written as the sum of two terms
and the second term in (73), in view of (70) and (71), is
Here, the symbol dVT is introduced for convenience (see a further analysis in section 9). Although due to large values of a, the intrinsic anharmonic term in (73) is dominant in value at T < Q, the sign of (73) at high temperatures can be either positive or negative.
The temperature behavior of t resembles that of CV, so that at T > Q, t tends to be independent of temperature [O. Anderson, 1984]. Accordingly, the thermal pressure Pth tends to a linear dependence on T. Our estimates of t (Table 2 and Pankov et al. ) show that the nonlinear terms in Pth versus T makes a contribution not greater than 1-3% at the highest temperatures indicated in Table 3 and Pankov et al. . The linear temperature behavior of Pth is considered to be the universal property of solids at high temperatures [O. Anderson et al., 1992a]. Unlike the temperature dependence, the extent to which t depends on volume at T = constant varies from one type of solid to another. According to O. Anderson et al. [1992a], the earth minerals fall into an intermediate group between materials with significant (e.g., gold) and relatively weak (e.g., sodium chloride, alkali metals, noble elements) volume dependences of t. It is important that these inferences are based on both P-V-T data (analysis of Pth(V, T) ) and high-temperature data for KT and aKT at P = 0 (analysis of (70)).
At first glance, the observed regularity in t(T) at T > Q is explained by the fact that CV constant in this temperature range, where, in view of (8) and (69), g is therefore independent of temperature. In other words, the quasiharmonic Mie-Grüneisen EOS is seemingly justified at high temperatures. However, this is not quite true to be the general case when we start with the condition t = constant or t = t(V) which are compatible with the case of CV(T) (see (69)) and therefore with a dependence of g on both volume and temperature [O. Anderson and Yamamoto, 1987]. Theoretically, the departure of CV from the Dulong and Petit law [e,g., Mulargia and Broccio, 1983; O. Anderson and Suzuki, 1983; Gillet et al., 1991; Reynard et al., 1992] is in part related to the intrinsic anharmonicity described by the third and higher order terms in the lattice Hamiltonian expansion. In describing experimental data, the thermal EOS generally requires smaller number of terms in this expansion than the caloric EOS [Leibfried and Ludwig, 1961; Wallace, 1972; Davies, 1973]. Thus, when anharmonicity in Pth versus T and the temperature dependence of g are not observed, this may suggest that either the quasiharmonic limit for the vabrational g has not yet been achieved or the higher order terms in the thermal part of the EOS are mutually cancelled [O. Anderson et al., 1982].
In conclusion to the above analysis of t, we formulate the following important assumptions and their consequences that can easily be verified: (1) Let CV be independent of V, i.e., either CV = CV(T) or CV = constant. (2) Assume that KT = KT(V) that is equivalent to dT = K. From (1), it follows that either t = t(V) or t = constant, and in addition, KT is either a linear function of T or KT(V), which leads to q = q(V) (or q = 1 ) and either g = Vt(V)/CV(T) or g = g(V). The statement (2) is equivalent to either t = t(T) or t = constant. If both (1) and (2) statements are valid (but CV constant), then g = const V/CV(T) and q = 1. The conditions CV = constant and KT = KT(V) yield g = const V.
The thermal EOS resulting directly from integrating (12) is of the form
where f(V) is the static lattice pressure plus the zero oscillation pressure. The second term in (76) is the total thermal pressure Pth accounting for all anharmonic contributions. This EOS can be rewritten in the form
where, for example, T0 = 300 K. In accordance with the behavior of t described above, the thermal pressure can be approximated as [O. Anderson, 1984, 1988]
where T T1 Q. As already noted, the variation with volume in (78) is insignificant for some solids.
In the Mie-Grüneisen EOS, the thermal pressure is however defined as
where the quasiharmonic approximation for CV is used. Thus, here, at high temperatures T T2 Q, when CV constant,
Even when b(V) = b(V) in some temperature range, the distinction between (78) and (80) is retained since, in the general case, g = g(V, T) in (78) and a a. In practice, for certain minerals and for the present accuracy of measurements, Pth from (78) and the Mie-Grüneisen theory can be indistinguishable [Fei et al., 1992a, 1992b; Mao et al., 1991], especially for minerals with low Debye Temperature.
The term DPth in (77) can be approximated in various ways. For example, with given volume dependences of K and dT, by integrating (75), we can find t(V) at T = constant [O. Anderson et al., 1992a, 1993]. The temperature dependence of t is derived from data on a(T) and KT(T) at P = 0. Another possibility to explicitly approximate the thermal pressure is given by the power law for t(V) on an isotherm.
Fei et al. [1992a, 1992b] and Mao et al.  used a number of models for Pth in order to describe P-V-T X-ray data. Specifically, they assumed ( KT/ T)V = constant. Then, (70) yields
where t(0, T) is determined from data on a provided that (dKT/dT)P = constant, the condition assumed over the entire P-V-T range of measurements. From the assumption that both temperature derivatives of KT are constant, it follows that tK = const. However, in such a case, K will increase with pressure along an isotherm and decrease with temperature along an isobar--the behavior that disagrees with the usual properties of this parameters (see section 6). Note that the estimated dT values can be very sensitive to the adopted model of Pth [e.g., Mao et al., 1991], although the Pth values themselves from various models can be close.
Finally, in Figure 7, we illustrate the temperature dependences of t at P = 0, calculated from the Mie-Grüneisen EOS for periclase, corundum, and forsterite, described in section 6, with various values of q in the interval 0-2. Although this type EOS gives correct orders of magnitude and the correct regularities in the P-T variations of t, it is difficult to achieve the complete consistency for all of the data given, as well as in the cases of specific heat (section 7) and thermal expansivity (section 6). A better accuracy of the EOS is undoubtedly required than that of the Mie-Grüneisen EOS in order to describe experimental data, to reliably predict unmeasured properties, and in particular, to calculate the phase diagrams at high pressures (when a 10% error in Pth can substantially affect the estimated phase boundary slopes and positions).