A number of recent experiments report on a correlation between the low-temperature slope of the thermopower, ␣ / T, and the specific heat coefficient ␥ = C V / T for heavy fermions and valence fluctuating compounds with Ce, Eu, Yb, and U ions. Assuming that charge and heat currents at low temperatures are transported by quasiparticles, we first derive the universal value for the ratio q = ␣ / ␥T using macroscopic transport equations. We then calculate the thermal response of the Fermi liquid

more »

... ͒ regime of the periodic Anderson model and of the Falicov-Kimball model by dynamical mean field theory and find the q ratio. Eventually, we calculate the temperature dependence of ␣͑T͒ above the FL regime using the "poor man's" approach, which describes the scattering of conduction electrons on the lattice of f ions by the single impurity Anderson model with crystal field ͑CF͒ splitting. The overall temperature dependence is obtained by interpolating between the FL and the poor man's solution, and is explained in simple terms. The shape of ␣͑T͒ is determined by the relative magnitude of the Kondo scale T K and the CF splitting. Pressure or doping ͑chemical pressure͒ affects ␣͑T͒ by transforming the narrow Kondo resonances into a broad spectral function typical of valence fluctuators. This changes the effective degeneracy of the f state and results in a drastic modification of ␣͑T͒. Temperature also changes the degeneracy of the f state by populating the excited CF states. Since T K is strongly pressure dependent, while the CF splitting is not, the shape of ␣͑T͒ is a sensitive function of pressure or doping. These results explain the near universality of the q ratio and the overall behavior of ␣͑T͒ in EuCu 2 ͑Ge 1−x Si x ͒ 2 , CePt 1−x Ni x , YbIn 1−x Ag x Cu 4 , and similar systems. PACS number͑s͒: 72.15.Jf, 75.30.Mb, 62.50.ϩp, 75.30.Kz of a magnetic field and consider only a single Cartesian component of ĵ and ĵ E ͑this is appropriate for cubic systems͒. Thermoelectric effects are usually described in terms of the heat current rather than the energy current. Hence, we transform J and J E to J and J Q = J E − ͑ / e͒J to yield 12,14 where = L 11 , ␣T = ͑L 12 / L 11 − / e͒, and T = ͑L 22 − L 12 2 / L 11 ͒. A simple analysis shows that ͑T͒, ␣͑T͒, and ͑T͒ are the ZLATIĆ et al. PHYSICAL REVIEW B 76, 085122 ͑2007͒ 085122-2