Silver-Antimony-Telluride: From First-Principles Calculations to Thermoelectric Applications Silver-Antimony-Telluride: From First-Principles Calculations to Thermoelectric Applications

Silver-antimony-telluride (AgSbTe 2 ) based compounds have emerged as a promis- ing class of materials for thermoelectric (TE) power generation at the mid-temperature range. This Chapter demonstrates utilization of first-principles calculations for predict- ing TE properties of AgSbTe 2 -based compounds and experimental validations. Predictive calculations of the effects of La-doping on vibrational and electronic properties of AgSbTe 2 compounds are performed applying the density functional theory (DFT), and temperature-dependent TE transport coefficients are evaluated applying the Boltzmann transport theory (BTE). Experimentally, model ternary (AgSbTe 2 ) and quaternary (3 at. % La-AgSbTe 2 ) compounds were synthesized, for which TE transport coefficients were measured, indicating that thermal conductivity decreases due to La-alloying. The latter also reduces electrical conductivity and increases Seebeck coefficients. All trends corre- spond with those predicted from first-principles. Thermal stability issues are essential for TE device operation at service conditions, e.g. changes of matrix composition and second-phase precipitation, and are also addressed in this study on both computational and experimental aspects. It is shown that La-alloying affects TE figure-of-merit posi- tively, e.g., improving from 0.35 up to 0.50 at 260 °C. We highlight the universal aspects of this approach that can be applied for other TE compounds. This enables us screening their performance prior to synthesis in laboratory. A significant step in this direction is introduced by our previous investigations of AgSbTe 2 based phase, involving both computational and experimental aspects [57, 58]. Vibrational properties of both AgSbTe 2 -based and La-doped-AgSbTe 2 alloys, including frequency-depen-dent vibrational density of states functions (v-DOS), temperature-dependent heat capacity, sound velocities, and Debye temperatures were evaluated employing lattice dynamics first-principles calculations. It was reported that La-doping reduces average sound velocity and varies v-DOS of AgSbTe 2 -based phase [57, 58]. Quantitatively, lattice thermal conductivity of La 0.125 Ag 0.875 SbTe 2 alloy was calculated to be lower by ca. 14%, than that of AgSbTe 2 -based phase at 300 K [58]. Experimental validations of these effects of La-alloying on reducing lattice thermal conductivity were made, as well [58]. Experimentally, model ternary (AgSbTe 2 ) and quaternary (3 at.% La-AgSbTe 2 ) alloys were synthesized by vacuum melting followed by quenching and hot-pressing. The appropriate conditions enabling formation of AgSbTe 2 -matrix that dissolves La-atoms with no La-rich precipitates were established. DSC tests enable observation of Ag 2 Te+Sb 2 Te 3 → δ-AgSbTe 2 phase transition at 360°C for La-free alloys only, indicating improvement of alloy’s thermal stability due to La-additions. Temperature-dependent thermal conductivity of both alloys indicate reduction in thermal conductivity as a result of La-alloying from 0.92 to 0.71 W m −1 K −1 at 573 K, which corresponds with the trend predicted from first-principles. Measurements of tem-perature-dependent electrical conductivity and Seebeck coefficients indicate that La-doping reduces electrical conductivity and increases Seebeck coefficients, as predicted from first-principles. Eventually, it is shown that La-doping has positive effects on TE figure-of-merit ZT , which is improved, e.g., from 0.35 up to 0.50 at 260°C.


Introduction
It is of utmost technological importance to develop predictive tools that will provide us with information about design of materials' functional properties. In this context, density functional theory (DFT) first-principle calculations offer us such possibilities [1][2][3][4], allowing us this context, the molar formation energies of both phases and the free energies of their interfaces with AgSbTe 2 -matrix are simulated to predict their thermal stability and nucleation sequence.

5.
Additionally, to address the influence of deviations from AgSbTe 2 -stoichiometry on electron transport properties, the latter is simulated for two off-stoichiometric model alloys Ag 3 SbTe 4 and AgSb 3 Te 4 .
Experimental procedures are divided into the following steps: 1. Model ternary (AgSbTe 2 ) and quaternary (3 at.% La-AgSbTe 2 ) alloys are synthesized by vacuum melting followed by quenching and hot-pressing. The appropriate conditions enabling formation of AgSbTe 2 -matrix that dissolves La-atoms with no La-rich precipitates are found.
tetragonal (P4/mmm, No. 123), and rhombohedral (R-3m, No. 166) have been simulated from first-principles [57]. Calculations of temperature-dependent Helmholtz free energy for these three polymorphs indicate that P4/mmm polymorph is the most stable one at temperatures above 400 K, whose energy exhibits close proximity to that of Pm-3m polymorph. Based on calculated v-DOS function for P4/mmm model alloy consisting of 4 atoms per simulation cell, Figure 1a, it was decided to test effects of doping with lanthanum atoms (to be discussed further below). To represent effects of La-doping with effective concentration of La atoms, that is close to realistic doping levels, a model compound of Ag 7 LaSb 8 Te 16 stoichiometry was constructed having the same P4/mmm space group symmetry as of the original AgSbTe 2 lattice. In this compound, consisting of 32 atoms per simulation cell, Figure 1b, La-atom substitutes for 1 ⁄ 8 of Ag-atoms, so that, the resulting concentration is 3.125 at.% La. Computational parameters concerning structural relaxation and vibrational properties are provided in detail [57]. The effects of La-doping on vibrational and thermal properties will be discussed further below. Silver-Antimony-Telluride: From First-Principles Calculations to Thermoelectric Applications http://dx.doi.org/10.5772/66086

The base AgSbTe 2 phase-electronic calculations
To simulate the effects of La-doping on electrical conductivity and Seebeck coefficient, electronic band structures are calculated for both lattices from first principles. A plane-wave basis set is implemented in Vienna ab-initio simulation package (VASP) [60][61][62] and MedeA ® software environment [63]. The exchange-correlation electronic energy is expressed by means of generalized gradient approximation (GGA) using PBEsol energy functional [64] and projector augmented wave (PAW) potentials, which are utilized to represent core electron density [65]. Sampling of Brillouin zone is carried out using a set of uniform Monkhorst-Pack k-point mesh with density ranges between 0.14 and 0.17 Å −1 and smearing method of linear-tetrahedron with Blöchl corrections [66]. To represent Kohn-Sham electronic wave functions, the plane waves are spanned with 400 eV energy cutoff for the structural relaxation or electronic calculations, respectively. Electronic optimization procedures are performed applying 10 −5 eV energy convergence threshold.
The calculated 0 K band structures are used for evaluation of temperature-dependent electrical conductivity, electronic component of thermal conductivity, and Seebeck coefficient, applying near-equilibrium Boltzmann transport theory with constant relaxation time approximation, as implemented in BoltzTrap code [67].
The partial electrical conductivity tensor, σ ′ αβ (i, k ) , represented for ith energy band and a given k-point, is obtained from Cartesian component of electron group velocity by derivation of ith energy band, ε i,k , with respect to α -and β -components of electron's wave vector [68]. σ ′ αβ (i, k ) is then given by: where e is electron unit charge, ℏ is reduced Planck constant, and τ i,k is electron relaxation time, which is assumed to be constant. This yields temperature and chemical potential, μ , dependent electrical conductivity tensor with respect to α -and β -components, summed over N-energy bands: where Ω is characteristic unit cell volume and f 0 (T, ε, μ ) is equilibrium Fermi-Dirac distribution function [69]. The electronic component of thermal conductivity tensor, κ e , is, accordingly, expressed by: Finally, the explicit expression for Seebeck coefficient tensor, S ij , is given by [70,71]: 2 Te 3 and Sb 8 Te 3 phases   Nonmagnetic DFT calculations are performed for Sb 2 Te 3 and Sb 8 Te 3 crystal structures having (R-3m) space group symmetry, which incorporate 15 and 33 atoms per simulation cell, respectively. Both lattice structures are rendered in Figure 1c and d, respectively. A computational routine similar to aforementioned one is implemented with several differences.

The Sb
GGA approximation is applied for a set of uniform 9 × 9 × 9 Monkhorst-Pack k-point mesh, and plane waves are spanned with either 400 or 350 eV energy cutoff for structural relaxation or electronic calculations, respectively. Electronic optimization procedures are performed applying 10 −6 eV energy convergence threshold.
Structural relaxation procedures are first performed, allowing variation of cell volume and atom positions at all degrees of freedom, setting a convergence threshold of 10 −4 eV Å −1 for Hellman-Feynman forces. The resulting lattice parameters obtained for relaxed crystal struc- To address bulk and interfacial energetic aspects related with nucleation of Sb 2 Te 3 and Sb 8 Te 3 phases in AgSbTe 2 phase, we have simulated formation energies of Sb 2 Te 3 and Sb 8 Te 3 phases and their interfaces with AgSbTe 2 phase. Molar formation energy of model Sb p Te q cell, E _ S b p T e q tot , is calculated using the following expression [57]: where E _ S b p T e q tot is cell's molar total energy and μ Sb o and μ Te o are chemical potentials of Sb-and Te-atoms in their standard states, which are evaluated to be −397.72 and −303.12 kJ mol −1 , respectively. The free energy of silver-antimony-telluride (AST)/antimony-telluride (SBT) interface is calculated constructing a slab model having AST/SBT generic form, and using the following expression [74]: where A is AST/SBT interface cross-sectional area, E AST/SBT f is calculated formation energy of slab model, E _ AST f and E _ SBT f are calculated molar formation energies of AST and SBT sub-cells, and n AST and n SBT are their number of moles in the entire model slab, respectively.

Ag 3 SbTe 4 and AgSb 3 Te 4 model compounds
To simulate the effects of deviations from stoichiometric AgSbTe 2 composition, we construct three model alloys based on P4/mmm space group symmetry, which is reduced to cubic P1 symmetry, by setting equal lattice parameter of a = 6.113 Å for all. The resulting structures simulated are: (AgSbTe 2 ) 2 , Ag 3 SbTe 4 , and AgSb 3 Te 4 , which appear in Figure 1e, f, and g, respectively. All three structures contain 8 atoms per unit cell and Sb/Ag ratios of 1, 1/3, and 3, respectively. To calculate band structures of these three model alloys, spin-orbit (SO) magnetic calculations were performed utilizing a similar GGA/PAW routine as described above for uniform 7 × 7 × 7 Monkhorst-Pack k-point mesh and 400 eV energy cutoff to represent Kohn-Sham electronic wave functions, applying 10 −6 eV energy convergence threshold. SO coupling is often being considered in band structure calculations [32,70,75]. TE transport properties were calculated according to the procedure detailed by Eqs. (1)-(4).

Materials synthesis
Experimental procedures implemented in this study are intended to validate the effects of La-alloying on TE performance, as predicted from first principles. They include synthesis of two model alloys, La-free and La-alloyed, having molar ratios (Ag:Sb:Te:La) of 18:29:53:0 and 15.75:29:53:2.25, respectively. Generally, synthesis procedures comprise vacuum melting and iced-water quenching, followed by uniaxial hot-pressing at two distinct temperatures, 540 and 500°C, yielding two series of 12.7 mm dia. pellets referenced below as Series A and Series B, respectively. The difference between these two series of alloys is manifested by their phase contents and average composition in matrix. These factors significantly affect TE performance, as will be discussed further below. A detailed description of the experimental procedures appears elsewhere [58].

Materials and thermoelectric property characterization
Materials characterization procedures include microstructure, phase identification, and composition analysis employing scanning electron microscopy and X-ray diffraction [58].
Assessment of alloys' thermal stability is investigated using SETARAM 1600 DSC with a scanning rate of 25 K min −1 at temperatures ranging from room temperature through 973 K.
MicroFlash ® LFA-457 laser flash analyzer (LFA; Netzsch GmbH) is utilized to measure directly of thermal diffusivity, α(T), of pellets in the same temperature range applying pulse-corrected Cowan approximation to consider heat loss of the samples [79], yielding instrumental accuracy of 2%. Material's density, ρ , is measured at room temperature, and density's dependence on temperature is neglected. Temperature-dependent heat capacity, C p (T), is simultaneously measured in LFA by comparative method using pure Al 2 O 3 -reference sample having similar geometry [76]. The resulting accuracy of evaluation of thermal conductivity values is equal to 10%. Pellets' thermal conductivity values, κ, are then determined by measuring their temperature-dependent thermal diffusivity and heat capacity, as well as, density; κ is then expressed by [80]:

Effects of La-alloying on thermoelectric performance
In this section, we introduce the concept resting behind La-alloying: its origin and implications, predictions from first principles, and experimental validations. Comparative discussion of the results in view of TE performance is provided.

. Structural and vibrational properties
P4/mmm form of AgSbTe 2 phase is found to be the most stable one compared to all three polymorphs at temperatures larger than 400 K and exhibits Helmholtz free energy values with close proximity to those of cubic polymorph [57]. Frequency-dependent v-DOS, g p ( ω ) , calculated for this compound applying Debye approximation exhibits two major peaks at ca. 2.0 and 2.7 THz, and discloses interesting feature. Whereas, 2.7 THz peak comprises equal contributions from lattice vibrations of all sublattice sites, 2.0 THz one is primarily ascribed to vibrations of Ag-sublattice site atoms [57,58]. This opens up the option of tuning v-DOS pattern by introducing point defects, a discipline for which the term phonon engineering has been coined [35]. Particularly, substitutions for Ag-sublattice sites by elements of different mass or atomic radius are expected to modify v-DOS with respect to that of pure AgSbTe 2 phase by suppressing its major v-DOS peak. This, consequently, will reduce lattice thermal conductivity. La has been suggested as optional substitution atom due to its relatively large mass and atomic radius compared to average values of AgSbTe 2 , that is, 138.91 a.m.u. and 187 pm vs. 121.21 a.m.u. and 143.98 pm, respectively, giving rise to enhanced phonon scattering by point defects [81][82][83][84]. Furthermore, La-alloying has commercial outcomes, since La is the most inexpensive element compared to constituents of AgSbTe 2 alloy and is one of the less inexpensive ones among energy-critical elements [85].
Three substitutional options were tested, in which La substitutes for Ag, Sb, or Te, and it was found that substitution at Ag-sublattice sites is the most energetically preferred state for P4/mmm symmetry [57]. Accordingly, La-doped structure was constructed, in which one La-atom substitutes for 1/8. of Ag-atoms, and is shown in Figure 1b. First, v-DOS was calculated for La-doped structure and 2.0 THz peak was suppressed, as expected. Second, phonon dispersion curves were calculated for both AgSbTe 2 and LaAg 7 Sb 8 Te 16 alloys close to Γ-point along c-crystallographic direction, indicating, that the slopes of the one longitudinal and two transverse acoustic modes of AgSbTe 2 -lattice are greater, than those of La-alloyed one [57]. Quantitatively, average sound velocities derived for pure and La-alloyed materials are 1727 and 1046 m s −1 , respectively. Moreover, temperature-dependent heat capacity functions were determined for both structures, yielding slightly lower values for La-alloyed material. Both values of sound velocity and heat capacity that are found to decrease due to La-alloying imply, that La-alloying should reduce lattice thermal conductivity [57]. Additional calculations employing Debye approximation for low-temperature range of heat capacity yield Debye temperatures and sound velocities for both pure and La-alloyed materials, which are 112 K and 1684 m s −1 vs. 104 K and 1563 m s −1 , respectively [58]. It is noteworthy that evaluation of sound velocity in this manner is considered to be more physically reliable, since it represents the entire space of lattice directions, rather than individual one. It is, therefore, expected that this way of calculation should yield thermal conductivity values, that fit experimental data better than the former way does.

Effects of La-doping on thermal conductivity
Average sound velocity, v s , and Debye temperature, θ D , evaluated from first-principles serve as input, that is required to evaluate lattice thermal conductivity, κ p . To this end, one possibility is to employ Callaway model for lattice thermal conductivity [86,87], which has become conventional, particularly in the field of TE materials [36][37][38][88][89][90][91][92][93]. In present case, however, there is no need to employ Callaway model for several reasons. First, Callaway model is specified for low temperatures, where contributions of either Normal (N)-or Umklapp (U)-processes are at the same order of magnitude. For temperatures adequately higher than Debye temperature (e.g., θ D ≈ 112 K for AgSbTe 2 alloy) [30,57], only U-processes dominate. Second, Callaway model considers g p ( ω ) and C p ( T ) functions that are simplistically approximated based on Debye model [94]. In present case, however, the explicit g p ( ω ) and C p ( T ) functions have already been calculated for both pure and La-alloyed materials. Alternatively, the following expression for lattice thermal conductivity is employed [94][95][96]: where τ is phonon relaxation time. To first approximation, it has been assumed that La-doping influences mostly sound velocity and heat capacity and has negligible effect on τ . The ratio of C v v s 2 -products obtained for LaAg 7 Sb 8 Te 16 andAgSbTe 2 alloys, therefore, reflects the lower limit of relative reduction in thermal conductivity due to La-alloying. Applying dispersion curves close to Γ-point along c-crystallographic direction, it is predicted, that κ p should decrease by factor of ca. 2.7 due to La-doping. Alternatively, applying sound velocity values derived from Debye approximation, κ p is expected to decrease by ca. 14% at room temperature due to La-doping [58].
A more thorough and accurate treatment of expression (8) considers the effects of La-alloying on τ , as well. To evaluate τ, contributions of two major scattering mechanisms are taken into account. The first one is phonon-phonon inelastic interactions, i.e., U-processes, that prevail for these alloys above room temperature. Relaxation time for U-processes, τ U , is represented by [96,97]: where γ is Grüneisen parameter that reflects the degree of lattice anharmonicity [69], and M is average atomic mass of alloy. Second, to account for internal composition inhomogeneity or compositional modulations at unit-cell length scales, that are typical for such materials [5,8,16,91], the boundary scattering mechanism is employed for characteristic period l , represented by relaxation time τ B , so that [96]: To consider dependence of v-DOS on phonon frequency, frequency-averaged expression for τ U is introduced, so that g p ( ω ) serves as weighting function: where ω D is Debye frequency. Equivalent relaxation time is then expressed as: The resulting values of lattice thermal conductivity for LaAg 7 Sb 8 Te 16 and AgSbTe 2 alloys are obtained from Eq. (8) by substituting the respective physical magnitudes for both alloys in Eqs. (9)-(11) [19,20,30,57,58,92] with l ≈ 1 nm [5,8,16,91]. Lattice thermal conductivity for LaAg 7 Sb 8 Te 16 and AgSbTe 2 alloys calculated as function of temperature appear in It is shown that thermal conductivity exhibits realistic values, that correspond with data documented in the literature [19,20,25,98] with marked decrease due to La-doping, ranging between relative values of 11 and 19%, depending on temperature.

Effects of La-doping on electrical properties
It was shown that La-alloying reduces lattice thermal conductivity values, which affects TE performance positively. To address, however, the total effects of La-alloying on TE performance, evaluation of electrical conductivity and Seebeck coefficient is essential. This goal was achieved from first-principles applying Boltzmann transport theory as described above for LaAg 7 Sb 8 Te 16 and AgSbTe 2 alloys. The results are plotted in Figure 3 in temperature range 50-1000 K.
It is found that La-doping results in reduction in electrical conductivity (e.g., from ca. 1800 down to 250 S cm −1 at room temperature) and, at the same time, increase in Seebeck coefficient, e.g., from ca. 4 up to 40 µV K −1 at room temperature. For the sake of comparison, Jovovic and Heremans reported on experimental measurements of electrical conductivity and Seebeck coefficients of stoichiometric and doped AgSbTe 2 alloys at temperatures up to 400 K [19,98].
For example, they report on electrical resistivity value of 5 × 10 −5 Ohm m at 100 K for stoichiometric AgSbTe 2 alloy, which is equivalent to 200 S cm −1 .
They report also on electrical resistivity that increases with temperature, indicating charge carriers scattering. Additionally, electrical resistivity may either increase or decrease with  It is shown that La-doping has considerably positive effect on PF. This also corresponds with the data reported by Min et al. [27], specifically for low La-concentration regime. We note that, moreover, La-doping reduces lattice thermal conductivity, as shown in Figure 2. We conclude that La-alloying should improve energy conversion efficiency of AgSbTe 2 (P4/mmm), as reflected by increased TE figure-of-merit.

Formation of Sb 2 Te 3 and Sb 8 Te 3 (R-3m) phases
The single δ-phase is Sb-rich phase based on AgSbTe 2 alloy. Since it has limited solubility to Sb with relatively moderate slope of Sb-solvus, it is likely to decompose to δ+Sb 2 Te 3 phase mixture [28,29,[100][101][102][103], whereas Sb 2 Te 3 is equilibrium phase and may appear as different homologous forms [72,73,[104][105][106][107]. Precipitation of antimony-telluride second phase in δ-matrix is expected to affect TE performance due to contributions from both matrix and precipitate phases or variation of the average matrix composition. In the following sections, we address both aspects.   and 〈10 ¯ 1 〉 AgSbTe 2 ∥ 〈 ¯ 2 110〉 Sb p Te q orientation relationship, which was observed experimentally [100]. Both structures are displayed in Figure 5. Figure 5a and b consist of different Sb-and Te-terminating planes, so that interfacial free energies calculated according to Eq. (6) represent an average value for both terminations. Correction factor is, therefore, applied to represent interfacial free energy of low-energy Sb-termination. The resulting values for Sb 2 Te 3 / AgSbTe 2 and Sb 8 Te 3 /AgSbTe 2 interfaces are γ = 208 and 175 mJ m −2 , respectively. These values are considered to be relatively low compared to those of intermetallic compounds and are comparable with those of pure metals [108]. This is, however, not surprising, considering the extremely small atomic misfit between the ( 111 ) AgSbTe 2 and ( 0001 ) Sb p Te q crystallographic planes [100], which encourages formation of Sb 2 Te 3 or Sb 8 Te 3 precipitates in the form of long lamellae along these planes [28,29,57,58,[100][101][102][103]. These low values of interfacial free energy also initiate fast nucleation, thanks to low activation energy for nucleation, which is proportional to γ 3 [109]. Interestingly, Sb 2 Te 3 phase exhibits lower value of formation energy and higher value of interfacial free energy compared to those of Sb 8 Te 3 phase. This implies that Sb 8 Te 3 is metastable phase that may form prior to the nucleation of Sb 2 Te 3 equilibrium phase [28,29,[100][101][102][103]. Suggested nucleation sequence is, therefore, supersaturated-δ → supersaturated-δ + Sb 8 Te 3 → equilibrium-δ + Sb 2 Te 3 .

Effects of Sb 2 Te 3 and Sb 8 Te 3 formation on electronic properties
In view of aforementioned prospect for the presence of either of Sb 2 Te 3 -or Sb 8 Te 3 -phases in AgSbTe 2 -matrix, calculations of transport coefficients of these phases provide us with predictions of the effects such phase mixture on TE performance. Figure 6 displays electrical conductivity and Seebeck coefficients calculated in temperature range 50-1000 K.
It is shown that both electrical conductivity and Seebeck coefficient values of Sb 2 Te 3 phase are larger than those of Sb 8 Te 3 phase in wide temperature range, e.g., ca. 2100 S cm −1 and 85 µV K −1 for Sb 2 Te 3 compared to 1380 S cm −1 and 29 µV K −1 for Sb 8 Te 3 at 300 K, respectively. Moreover, comparison of these results with the data shown in Figure 3 for AgSbTe 2 -matrix implies that precipitation of Sb 2 Te 3 phase yields positive influence on TE performance: both electrical conductivity and Seebeck coefficient increase. It is strikingly indicated that the effects of Sb 2 Te 3 precipitation are even greater, than those of La-doping. It is, therefore, concluded that the desirable material from TE viewpoint is La-doped, Sb-supersaturated δ-AgSbTe 2 -matrix that is aged for a certain duration to form considerable amount of Sb 2 Te 3 phase. The effects of Sb 8 Te 3 phase on TE performance are, conversely, inferior to those of Sb 2 Te 3 phase. Sb 8 Te 3 is, however, metastable phase and is not expected to prevail for long durations at elevated temperatures (e.g., under service conditions of TE generator) due to low thermal stability.

Effects of off-stoichiometry on electronic properties of the AgSbTe 2 phase
An additional effect taking place during precipitation of any Sb p Te q phase from Sb-supersaturated δ-matrix is enrichment of δ-matrix with Ag-atoms and depletion of Sb. To simulate these compositional variations, two off-stoichiometric model alloys are constructed, namely Ag 3 SbTe 4 and AgSb 3 Te 4 , in addition to stoichiometric AgSbTe 2 phase. These model compounds, appearing in Figure 1e, g, and f, exhibit Sb/Ag ratios of 1/3, 3, and 1, respectively. It is shown, that increase in Sb/Ag ratio results in decrease in electrical conductivity simultaneously with increase in Seebeck coefficient. This trend corresponds well with study of Jovovic and Heremans [19], who reported on decrease in both Seebeck coefficient and electrical resistivity due to additions of 2% AgTe to stoichiometric AgSbTe 2 -phase, i.e., reducing Sb/Ag ratio. It should be noted that comparison of this trend with data reported in the literature is not straightforward, since compositional changes involve in practice not only Sb/Ag ratio, but also ratio of Te to any of the other species. Additionally, deviations from given stoichiometry often involve formation of second phases, which is not directly simulated here. For instance, Zhang et al. reported on dependence of TE properties on composition for Ag 2−y Sb y Te 1+y -based alloys and found that electrical conductivity increases, while Seebeck coefficient decreases with y-values increasing from 1.26 up to 1.38 [101].
Most importantly, this predicted effect of Sb/Ag ratio on electrical properties has major implications on the temporal evolution of TE performance of the material during aging heat treatments (below Sb-solvus), or of TE generator during service. Since Sb p Te q phases nucleate from Sb-supersaturated δ-matrix during heat treatments, Sb/Ag ratio in δ-matrix decreases. This should be accompanied by increase in electrical conductivity concurrently with decrease in Seebeck coefficient.

Experimental results
It was shown above how first-principles calculations provide us with information about all aspects concerning TE transport behavior, including thermal and electrical conductivity and Seebeck coefficient. Most importantly, this assists us in tailoring the material by introducing lattice defects to enhance its TE performance. In the following section, we introduce experimental procedures taken for validating the above predictions. Comparing between both aspects is, moreover, very instructive not only on engineering aspects, but also on universal aspect, by realizing how to implement computational tools to predict properties of other materials.

Microstructure and implications on thermoelectric behavior
As mentioned above, two classes of La-alloyed AgSbTe 2 -based materials were prepared by uniaxial hot-pressing at 540 or 500°C, and are classified as Series A and Series B, respectively. The ideal case for testing the effects of La-doping is single δ-phase dissolving La homogeneously. This, however, is difficult to achieve. Hot-pressing at 540°C, that is, above Sb-solvus, expected to yield the desirable single δ-phase that does not contain Sb 2 Te 3 precipitates [101][102][103]. These Sb 2 Te 3 precipitates, indeed, were not observed in Series A samples; however, La-rich precipitates having stoichiometry close to LaTe 2 were observed [110]. As a result, δ-matrix was found to be depleted of La [58], which does not allow us comparison between La-free and La-doped samples. Conversely, samples of Series B, that were hot-pressed at 500°C, exhibit considerable amount of Sb 2 Te 3 phase, which is expected; however, LaTe 2 precipitates are dissolved, so that δ-matrix contains adequately large amount of La, close to its nominal concentration. Series B is, therefore, more suitable to exemplify the effects of La-doping. Moreover, as predicted from first-principles, the presence of Sb 2 Te 3 precipitates, in addition to La solute atoms, has positive effects on electronic transport.

Thermal analysis
Thermal conductivity measurements performed for both Series A and Series B indicate the expected trend. First, all thermal conductivity values lie in the range 0.6-0.8 W m −1 K −1 [58]. Second, samples of Series A did not exhibit any considerable difference between La-free and La-doped materials [58]. This is associated to depletion of δ-matrix from La solute atoms, so that, matrix composition of La-doped and La-free materials is practically the same. Third, and most importantly, it was found that thermal conductivity of La-doped materials is significantly lower than those of La-free materials of Series B, e.g., 0.8 W m −1 K −1 for La-free and 0.6 W m −1 K −1 for La-doped samples at 500 K. This is strikingly corroborated by predictions from first-principles, both qualitatively and quantitatively, as shown in Figure 2. It is also noteworthy that both values coincide at temperatures larger than 650 K, which can be associated with phase transition [58,[101][102][103]. Thorough discussion of thermal conductivity values measured for Series A and Series B materials and their relationship with microstructure appears elsewhere [58]. To address this issue of phase transition, DSC measurements were implemented for both La-free and La-doped samples, Figure 8.
Endothermic peak at around 630-650 K is observed for La-free material, which is associated with Ag 2 Te + Sb 2 Te 3 → δ-AgSbTe 2 phase transition at 360°C [101][102][103]. La-doped material, however, does not exhibit this transition. This corresponds well with thermal conductivity behavior reported by us earlier [58], in which temperature-dependent thermal conductivity of La-doped materials show up continuous trend, whereas La-free materials exhibit sharp drop of thermal conductivity around this temperature. This implies that La-additions help in stabilizing δ-phase against decomposition, which is expected to contribute to stability of TE device operation at service conditions. Sharp endothermic peak at ca. 860 K, which is common for both La-free and La-doped materials, is associated to melting.

Electrical property measurements
It was predicted from first-principles that La-doping reduces electrical conductivity and increases Seebeck coefficient, Figure 3. Measurements of electrical conductivity and Seebeck coefficients were carried out for both Series A and Series B materials. The samples of Series B are of our interest, since they dissolve La-atoms in δ-matrix; we will, therefore, introduce these results first. Figure 9 displays experimentally collected electrical conductivity and Seebeck coefficient values of La-free and La-doped materials of Series B.
It is shown that electrical conductivity values decrease, e.g., from ca. 1400 down to 900 S cm −1 at room temperature, and Seebeck coefficient increase, e.g., from ca. 30 up to 70 µV K −1 at room temperature, due to La-doping. This behavior is, qualitatively, the same as that observed for calculated values shown in Figure 3. Moreover, temperature dependence, that is, electrical conductivity decreasing and Seebeck coefficient increasing with temperature for both La-free and La-doped materials, is identical to that indicated by calculated values shown in Figure 3.
There are two major differences between experimental and calculated values appearing in Figures 3 and 9, respectively. First, the absolute values of measured Seebeck coefficient values are greater than calculated ones. Also, difference of electrical conductivity between La-doped and La-free materials is smaller for measured dataset than for calculated ones. This is probably due to difficulty to simulate low dopant concentrations in DFT [70]. Second, it is noteworthy that both values of electrical conductivity and Seebeck coefficients measured for La-free and La-doped materials converge at temperatures >650 K, Figure 9. Interestingly, these convergences occur due to sharp deviations of the values featured by La-free material, whereas the values of La-doped materials preserve their continuous trendline. This observation corresponds well with the behavior shown by DSC curves in Figure 8, where La-free compound decomposes at around 650 K, whereas La-doped compound seem to preserve its thermal stability. This also corresponds with converging thermal conductivities of the samples of Series B as discussed above [58]. Following our comparative discussion in Section 4.1.1.3, experimental values of electrical properties are found to be closer to experimental values reported in the literature than to calculated values [19,25,27,98,99].
To complement our understanding of the effects of La-doping on electronic properties, we measured temperature-dependent electrical conductivity and Seebeck coefficient values for the samples of Series A, as well. The results are plotted against temperature in Figure 10.
Comparison between the results attained for alloys of Series A and Series B is very instructive. As noted, the samples of Series A exhibited formation of LaTe 2 -like precipitates, which "drain out" La atoms from δ-matrix, resulting in matrix compositions, that are nearly identical to each other for La-free and La-doped materials. For this reason, thermal conductivity values measured for La-free and La-doped materials seem to be practically identical in wide temperature range [58]. It is, therefore, not surprising to observe the same behavior for electrical properties, Figure 10.
It is indicated, that both electrical conductivity and Seebeck coefficient values measured for La-free and La-doped materials seem to be very close to each other in the entire temperature range, probably due to nearly identical matrix compositions for La-free and La-doped materials. Additionally, electrical conductivity and Seebeck coefficients featured by La-doped alloys exhibit relatively continuous temperature-dependent behavior, whereas values, measured for La-free alloys exhibit curled behavior. This, again, can be explained in terms of poor thermal stability of La-free materials, as discussed above.

Implications for thermoelectric power conversion
It has been shown that La-doping has unequivocally positive effect on reducing lattice thermal conductivity, both computationally and experimentally. The effects on electrical properties, particularly electrical conductivity and Seebeck coefficient, are opposing each other. To assess the effects of La-doping on device's power capacity, TE PFs of La-free and La-doped materials of Series B are evaluated based on the data displayed in Figure 9. The results are shown in Figure 11.
It is clearly shown that La-doping affects positively PF for temperatures lower than 500 K, e.g., PF determined for room temperature increases from ca. 200 to 400 µW m −1 K −2 due to La-doping. At higher temperatures, PFs of La-free and La-doped materials are practically identical. The maximum PF values observed are around 1000 µW m −1 K −2 . This trend is similar to that reported by Min et al. [27], that is, PF increasing from 300 up to 1500 µW m −1 K −2 in respective temperature range from room temperature to 400°C for AgSbTe 2 alloy. Finally, determination of TE figure-of-merit for both La-free and La-doped materials will provide us with the ultimate indication whether La-doping enhances TE power conversion efficiency. Based on thermal and electrical properties measured for Series A and Series B, temperature-dependent ZT values were determined and appear in Figure 12.  Most importantly, it is shown that La-doping increases ZT values markedly, Figure 12b, e.g., from ca. 0.3 to 0.45 at 473 K. This improvement is due to decrease in thermal conductivity in almost the entire temperature range and increase in PF at low-temperature regime due to La-doping. Above 600 K, again, both values of La-free and La-doped alloys converge due to poor thermal stability of La-free materials. ZT values of La-free and La-doped materials shown in Figure 12b correspond with those reported by Zhang et al. [101], where the effects of La-doping are comparable to those of stoichiometric variations about AgSbTe 2 composition. Similar values are reported by Mohanraman et al. [43] and Jovovic and Heremans [19] for Bi-doping, as well as, for Pb-doping [19]. Chen et al. obtain similar ZT values for Ge-doping [111] and for Sn-doping [112], depending on concentration. ZT values reported in the present study are, however, lower than those reported by Du et al. [25,99], probably owing to different processing conditions yielding higher electrical conductivity values [113].
The picture, revealed for alloys of Series A, is, however, different; it is shown in Figure 12a that La-doping has little or no effect on ZT. This is, again, not surprising and follows the trends featured by Series A alloys for electrical conductivity and Seebeck coefficient, Figure 9, and thermal conductivity [58] associated to depletion of La-atoms from δ-matrix in La-alloyed materials hot-presses at 540°C.

Summary and concluding remarks
This chapter introduced the following findings. Computationally, total energy calculations for different polymorphs of AgSbTe 2 phase yield their Helmholtz free energies, implying, that P4/mmm space group symmetry is the most stable one at temperatures adequately higher than room temperatures. Predictions of the effects of doping on thermal conduc-  Attempts to infer conclusions with practical implications from DFT calculations must consider engineering aspects that extend further beyond single phase state having high symmetry unit cell, that maintains its physical properties with time. For example, considerations, such as long-term device operation under elevated service temperatures, should be taken into account. Such case requires original solution for simplified (or, sometimes, over-simplified) approach offered by DFT. Particularly, for the case of thermal stability, exposure of Sb-rich δ-phase to elevated temperatures results in precipitation of Sb p Te q -based phases at the same time with decrease in Sb/Ag ratio in δ-matrix. In this manner, thermal stability issues can be addressed by dividing the realistic conditions into a set of simplified problems, each can be handled by DFT. To this end, we first consider the case in which Sb 2 Te 3 and Sb 8 Te 3 phases precipitate inside AgSbTe 2 -matrix. It is found that both electrical conductivity and Seebeck coefficient values of Sb 2 Te 3 -phase are larger, than those of Sb 8 Te 3 -phase in wide temperature range, e.g., ca. 2100 S cm −1 and 85 µV K −1 for Sb 2 Te 3 compared to1380 S cm −1 and 29 µV K −1 for Sb 8 Te 3 at 300 K, respectively. Moreover, it is estimated that precipitation of Sb 2 Te 3 -phase in AgSbTe 2 -matrix is expected to improve the total values of both electrical conductivity and Seebeck coefficient. Concerning nucleation sequence of Sb 2 Te 3 and Sb 8 Te 3 phases in AgSbTe 2 , their molar formation energies and interfacial free energies were calculated, suggesting that Sb 8 Te 3 nucleates first as metastable phase, prior to the formation of equilibrium Sb 2 Te 3 phase. Second, to address the influence of deviations from AgSbTe 2 stoichiometry on electron transport properties, off-stoichiometric model alloys Ag 3 SbTe 4 and AgSb 3 Te 4 were simulated. It is found that increase in Sb/Ag ratio results in decrease in electrical conductivity simultaneously with increase in Seebeck coefficient. Considering both effects of Sb 2 Te 3 precipitation accompanied by simultaneous decrease in Sb/Ag ratio in δ-matrix taking place with aging time at temperatures below δ-solvus, it is expected that electrical conductivity of two-phase δ+Sb 2 Te 3 alloy should increase with aging time, disregarding effects, such as electron boundary scattering. Interestingly, these two effects have opposite consequences regarding Seebeck coefficient, so that, it is difficult to assess resulting Seebeck coefficient.
Experimentally, model ternary (AgSbTe 2 ) and quaternary (3 at.% La-AgSbTe 2 ) alloys were synthesized by vacuum melting followed by quenching and hot-pressing. The appropriate conditions enabling formation of AgSbTe 2 -matrix that dissolves La-atoms with no La-rich precipitates were established. DSC tests enable observation of Ag 2 Te+Sb 2 Te 3 →δ-AgSbTe 2 phase transition at 360°C for La-free alloys only, indicating improvement of alloy's thermal stability due to La-additions. Temperature-dependent thermal conductivity of both alloys indicate reduction in thermal conductivity as a result of La-alloying from 0.92 to 0.71 W m −1 K −1 at 573 K, which corresponds with the trend predicted from first-principles. Measurements of temperature-dependent electrical conductivity and Seebeck coefficients indicate that La-doping reduces electrical conductivity and increases Seebeck coefficients, as predicted from firstprinciples. Eventually, it is shown that La-doping has positive effects on TE figure-of-merit ZT, which is improved, e.g., from 0.35 up to 0.50 at 260°C.
We demonstrate how first-principles calculations serve as trustworthy tool for predicting TE performance of materials, screening the best candidates for application in TE devices. It is noteworthy that such DFT routines prove to be very efficient by prediction of TE properties in a way saving expensive and time-consuming experiments. The resulting materials that seem to possess improved performance are, eventually, processed in laboratory. We show how simple physical considerations can be implemented in DFT calculations and lead to improvement of power conversion efficiency. La-doping improves the alloys' thermal stability and reduces their thermal conductivity, as well as enhances TE power factor in certain temperature range. As a result, the total TE figure-of-merit improves significantly. We, finally, emphasize the universal aspects of this approach that can be applied for other TE materials, as well.