WorksPhysics and Astronomy
Simultaneous suppression and temperature-driven crossover of particle-like and wave-like thermal transport in trigonal CsAgS
Abstract
The simultaneous suppression of particle-like ( ) and wave-like () thermal transport channels in crystalline materials remains a challenging problem, as reducing one channel can enhance the other. Here, we investigate the thermal transport in trigonal P\Bar{3}m1 CsAgS using self-consistent phonon theory with cubic and quartic anharmonicity and the unified theory of thermal transport. Our calculations predict an ultralow lattice thermal conductivity of 0.222 W m K at 300 K, featuring simultaneously low and with a weak temperature dependence of T, which deviates significantly from the conventional T behavior. This simultaneous suppression arises from localized Cs modes, which enhance anharmonic scattering, and the softer Ag-S framework, which reduces phonon group velocities, together suppressing , in conjunction with sparse, well-separated high-frequency S-derived optical modes that suppress by limiting the coherent coupling between the phonon modes. Furthermore, the dominant transport channel exhibits a temperature-dependent crossover, with contributing 74 % of total lattice thermal conductivity at 300~K and contributing 63 % at 700~K. Combining the ultralow thermal conductivity with electrical transport calculated within the momentum relaxation time approximation (MRTA), we obtain a of 1.28 for -type doping at 700 K. The present work not only clarifies the microscopic mechanism of ultralow thermal conductivity in trigonal CsAgS but also identifies the covalent M-Ch framework modification within the AMCh family of ternary chalcogenides as an additional tuning parameter for further reducing within the bonding-hierarchy design principle.
Introduction
Materials with ultralow thermal conductivity are of great importance for various technological applications, including thermoelectrics [1, 2], thermal barrier coatings [3], thermal insulation [4, 5], and nanoscale thermal confinement [6]. This has motivated the design and discovery of several classes of materials with intrinsically low lattice thermal conductivity, such as clathrates [7], filled skutterudites [8, 9], chalcogenides [10, 11, 12], Zintl phases [13, 14, 15], argyrodites [16]. The suppressed thermal transport in these materials originates from diverse structural and dynamical features such as complex crystal structures [17], rattling-like sublattice vibrations [18, 19], lone-pair electrons [20, 21], ferroelectric instabilities [22, 23], and bonding hierarchies [24], etc. Therefore, identifying and understanding the microscopic mechanisms of thermal transport in crystalline materials is crucial for designing materials with ultralow lattice thermal conductivity. Phonons in crystals carry heat through two microscopically distinct mechanisms: particle-like propagation between well-separated wave packets, and coherent, wave-like tunnelling between near-degenerate modes whose linewidths exceed their frequency spacing [25]. The unified theory of thermal transport formalizes this split into a diagonal, Peierls-like contribution and an off-diagonal coherence contribution [26, 27]. The two
channels respond to structural perturbations in opposite ways: phonon softening, mass disorder, or structural complexity that suppresses by shortening phonon lifetimes and lowering group velocities typically densify the phonon spectrum and narrow the interbranch spacing, thereby enhancing [28]. This intrinsic antagonism imposes a practical lower bound on in crystalline solids, highlighting the need to decouple the two thermal transport channels to further reduce lattice thermal conductivity.
Two recent studies show that this tradeoff is not fundamental. Xiong et al. demonstrated that the superionic crystals XReSI (X = Rb, Cs) suppress both transport channels simultaneously through a bonding hierarchy [29]. A rigid, covalently bonded [ReSI] covalent network is interpenetrated by weakly, ionically bonded cations. The mechanism was further generalized by Wu et al. into an explicit structural descriptor - a phonon density of states sparsity metric combined with a mass contrast criterion [24]. Together, these studies establish a soft ionically bound rattler embedded in a stiff covalent framework as a general route to simultaneous suppression of particle- and wave-like transport channels.
The AMCh (Where, A = alkali metal, M = Cu, Ag, Au, and Ch = S, Se, Te) family of ternary chalcogenides provides a particularly compelling platform for exploring this concept [30, 31, 32]. These materials comprise strongly bonded [MCh] frameworks weakened by pronounced M-Ch antibonding character near the Fermi level, separated by a weakly bonded alkali metal sublattice. This combination of a stiff, strongly bonded framework and weakly coupled, rattler-like sublattice is expected to promote distinct mechanisms for suppressing the particle-like and coherent component of heat transport. Recent work on CsCuS demonstrated a low lattice thermal conductivity, attributed primarily to Cu-S antibonding hybridization and an avoided crossing between Cs-dominated acoustic modes and low-lying optical branches [33]. Additionally, the sparse high-frequency optical modes corresponding to the strongly bonded lighter S atoms contribute to the suppression of coherence’s thermal conductivity. However, remains substantial in CsCuS with of 1.8 and 1.37 W m K along in-plane and out-of-plane directions, respectively at 300 K. This contrast highlights a key limitation of the bonding hierarchy picture: while the weakly coupled sublattice can efficiently disrupt particle-like transport, the strongly bonded covalent framework may still promote substantial particle-like conduction, posing a key challenge for simultaneous suppression of both thermal transport channels.
This contrasting behavior reveals a fundamental limitation of the bonding-hierarchy strategy proposed for achieving ultralow thermal conductivity by simultaneously suppressing both transport channels. While the weakly bonded sublattice suppresses particle-like transport through rattling-induced scattering and avoided-crossing-induced flattening of phonon branches, the rigid covalent framework may still provide an efficient pathway for particle-like heat conduction, leaving the particle-like transport channel only partially suppressed. Thus, simultaneous suppression of both transport channels requires not only a weakly bonded rattler sublattice but also targeted engineering of the strongly bonded covalent framework itself to reduce its intrinsic capacity for particle-like transport.
The M-site cation therefore provides a natural target for further engineering. For instance, in CsCuS, the S sublattice governs the spectral sparsity that suppresses , while the weakly bonded Cs sublattice induces rattling and avoided crossings that suppress . The Cu–S bonds, which form the connective backbone of the covalent framework, remain comparatively untargeted and are responsible for the residual particle-like transport [33]. To examine this possibility, we replace Cu with its heavier group-11 congener Ag. This is an isovalent substitution that preserves the S-dominated high-frequency modes and the Cs rattler sublattice, while increasing the mass of the framework, forming M–Ch units. The increased mass is expected to soften the acoustic and low-lying optical modes, reducing their frequencies and group velocities and thereby providing a chemically minimal route to further suppress without disrupting the mechanism responsible for the low in CsCuS. Similar mass-induced phonon softening and reduced group velocities have been associated with suppressed thermal transport [34].
In the present work, we systematically investigate the structural stability, chemical bonding, anharmonic lattice dynamics, and thermal transport properties of CsAgS
in the trigonal phase using first-principles simulations. We combine self-consistent phonon theory, including quartic and cubic anharmonic renormalization, with the unified theory of thermal transport. In particular, the anharmonic phonon properties are treated beyond the harmonic approximation by incorporating both cubic and quartic anharmonicities into the phonon self-energy. The unified thermal transport framework explicitly separates the particle-like and coherent contributions to lattice thermal conductivity. Together with the inclusion of higher-order phonon scattering and dual-channel phonon transport, we provide a comprehensive description of thermal transport in trigonal CsAgS. The formation energy and convex hull analysis confirm the structural stability. Our analysis further shows that the trigonal phase is dynamically and mechanically stable and retains its structural integrity up to 700 K. A pronounced bonding hierarchy was observed between the covalently bonded [AgS] framework and weakly coupled independent Cs ions, resulting in an ultralow lattice thermal conductivity of 0.222 W m K at 300 K. Importantly, both particle-like and coherent transport channels are strongly suppressed, while their relative contributions change significantly with temperature. The coherent contribution becomes dominant at 700 K, revealing a crossover from particle-like to wave-like heat transport. To complement the thermal transport analysis, we further investigate the electrical transport properties within the momentum relaxation time approximation (MRTA). Combining the ultralow lattice thermal conductivity with the resulting electrical transport coefficients, we obtain a maximum of 1.28 for -type doping at 700 K. Overall, our analysis demonstrates that targeted substitution on the covalent framework within a bonding hierarchy structural motif can simultaneously suppress the two thermal transport channels, providing a new direction for engineering ultralow thermal conductivity materials.
Computational methods
First-principles calculations
We investigated the lattice dynamics, thermal, and thermoelectric properties of CsAgS using first-principles calculations based on density functional theory (DFT) as implemented in the Vienna ab initio simulation package (VASP) [35, 36]. The projector augmented wave method with valence configurations for Cs, for Ag and for S was used to describe the interactions between the ion cores and the valence electrons. The electronic wave functions were represented using a plane-wave basis set with a kinetic energy cutoff of 500 eV. To describe the electron-electron interactions, we utilized the Perdew–Burke–Ernzerhof (PBE) exchange-correlation functional [37] within the generalized gradient approximation (GGA). The conjugate gradient algorithm was used to optimize the crystal structure by relaxing all degrees of freedom with the energy and forces convergence criteria of eV and eV/Å, respectively. The Brillouin zone was sampled using a k-mesh of , while a denser mesh of was utilized for the self-consistent energy calculations. Spin-orbit coupling (SOC) was included in the electronic-structure calculations to account for relativistic effects arising from the presence of heavy elements. The crystal orbital Hamilton population (COHP) analysis was performed with the LOBSTER code [38] to elucidate the nature of chemical bonds and orbital interactions.
Interatomic force constants
We utilized the ALAMODE package [39, 40] to efficiently extract the harmonic and anharmonic interatomic force constants (IFCs). The 0 K harmonic IFCs were obtained using the finite displacement method. We utilized a supercell to create 12 displaced structures. In each structure, the atoms were displaced by 0.01 Å. Subsequently, a Monkhorst-Pack k-mesh along with an energy cutoff of 500 eV was utilized for the single-point DFT calculations. The resulting displacement-force dataset was subsequently used to extract the harmonic force constants at 0 K. We obtained anharmonic IFCs by employing the compressive sensing lattice dynamics (CSLD) technique [41, 42], which can efficiently collect the physically important terms of the anharmonic IFCs by utilizing the limited displacement-force dataset [39]. We conducted the ab initio molecular dynamics simulations (AIMD) using a supercell with a Monkhorst-Pack k-mesh of and energy convergence criteria better than eV to generate the displaced structures necessary to obtain the displacement and force dataset to compute the anharmonic force constants within the framework of CSLD. The AIMD simulations were carried out within the NVT ensemble using a Nosé-Hoover thermostat at 300 K for 5000 steps with a step size of 0.5 femto seconds (fs). After disregarding 1000 initial steps, 80 snapshots were sampled using the AIMD trajectory at equal intervals of 50 steps to generate the quasirandom configurations. Then a random displacement of 0.1 Å was further applied to all the atoms in the sampled structures to minimize the residual correlations between the structures. Subsequently, high-precision static DFT calculations were performed to generate the training and validation data using the sampled configurations on a k-point mesh with the energy convergence criteria of eV and a plane-wave kinetic energy cutoff of 500 eV. Using the displacement-force dataset and the 0 K harmonic force constants, the cubic and quartic IFCs were obtained using the least absolute shrinkage and selection operator (LASSO) technique [43, 44]. We applied real-space cutoff radii of 7.94 Å and 5.29 Å for cubic and quartic IFCs, respectively. Notably, we observed a low relative fitting error of 4.1% for anharmonic IFCs, which is in excellent agreement with the small values reported in the literature [40, 45, 46, 47]. This highlights the robustness and reliability of the present approach to accurately model the anharmonic force constants.
Anharmonic phonon renormalization
The obtained 0 K harmonic and anharmonic IFCs are then used to compute the temperature-renormalized phonon dispersions within the framework of self-consistent phonon (SCP) theory as implemented in the ALAMODE package [39, 47, 48]. The first-order SCP equation considering only the quartic anharmonicity, i.e., the loop diagram, can be written in the following form [49]
Where is the SCP renormalized phonon frequency corresponding to wave vector . is the unitary polarization matrix, represents the phonon branch index, , is the total number of sampled phonon wave vectors in the first Brillouin zone, is the reciprocal representation of order IFCs and is the Bose-Einstein distribution function. The anharmonic polarization mixing was explicitly considered in the calculations by including the off-diagonal components associated with the loop self-energy. The SCP calculations were performed on a q-mesh of , which corresponds to the same supercell dimensions in real space. For the inner loop of the SCP equation, a denser q-mesh of was utilized to obtain the converged anharmonic phonon frequencies. To accurately capture the effect of cubic anharmonicity on the anharmonic phonon energies, we further incorporated the bubble self-energy corrections to the anharmonic phonon energies (SCPB), which can be evaluated using the following self-consistent equation within the quasiparticle approximation [49]
Where is the phonon-frequency dependent bubble self-energy defined as
Here, represents the third-order IFCs in reciprocal space, is a function that becomes 1 when is an integral multiple of the reciprocal lattice vector and 0 otherwise. The dependent function is given by
Where is the Bose-Einstein distribution function, and are the internal phonon wavevectors, and represents the analytic continuation of the Green’s function to the real axis by a positive infinitesimal . Within the framework of the quasiparticle (QP) approximation, several treatments, including QP[0] (), QP[S] (), and the non-linear QPNL () can be implemented to solve Eq. (2) [49]. We employed the QPNL scheme to obtain fully renormalized anharmonic phonon energies accounting for both cubic and quartic anharmonicities. A -mesh of was employed for the SCPB calculations. Since calculating the bubble self-energy is computationally expensive as compared to the SCP calculations, a relatively coarse -grid of was used for the inner loop of the SCPB equation. Additionally, the non-analytic part of the harmonic dynamical matrix was included in all calculations. The Born effective charges and dielectric tensor necessary to compute the non-analytic correction were obtained using density functional perturbation theory (DFPT) [50].
Unified theory of thermal transport
Utilizing the temperature-renormalized harmonic and anharmonic IFCs, the lattice thermal conductivity was computed by solving the linearized Boltzmann transport equation (BTE) within the framework of the unified theory of thermal transport [26, 27]. Within this formalism, the total lattice thermal conductivity comprises the contribution from (capturing the particle-like phonon transport) and (accounting for the wave-like phonon tunneling) described by the diagonal and off-diagonal elements of the Wigner heat-flux operator, respectively [25]. The lattice thermal conductivity tensor under the single-mode relaxation-time approximation is given by
Where , , , , and represent the reduced Planck’s constant, Boltzmann constant, absolute temperature, unit cell volume, number of sampled -points in the first Brillouin zone, and equilibrium Bose-Einstein distribution function, respectively. denotes the phonon scattering rate, which includes three-phonon (), four-phonon () and isotope () scattering. is the generalized group velocity tensor including both intra- and inter-branch contributions and is given by [25]
with and representing the polarization vector and the dynamical matrix, respectively. When only the diagonal elements of the Wigner heat flux operator are considered, i.e., in the second summation, Eq. (5) reduces to the standard Peierls-Boltzmann contribution to the lattice thermal transport given as
Wherein, and are the modal heat capacity and phonon group velocity, respectively. represents the phonon relaxation time. Finally, the total lattice thermal conductivity was obtained by summing the contributions from and , i.e., . In the present work, a -mesh of along with a scalebroad parameter of 0.5 was used for both 3ph and 4ph scattering processes. Considering the memory-intensive nature of the computation of 4ph scattering probabilities, the maximum likelihood approximation [51] was used to accelerate the convergence of the thermal conductivity. It should be noted that 3ph scattering processes are solved iteratively, while the 4ph scattering processes are treated under the relaxation time approximation. All thermal transport calculations were carried out utilizing the FourPhonon package along with its modified version to account for the coherent conductivity [52, 53, 54].
Electrical transport properties
We calculated the electrical transport parameters by solving the electronic BTE within the momentum-relaxation-time approximation as implemented in the AMSET package [55]. The electron-phonon interactions were modelled by explicitly considering the acoustic deformation potential (ADP), the polar-optical phonon (POP) scattering, and the influence of ionized impurity (IMP) scattering. The partial decay rate for each mechanism was obtained using Fermi’s golden rule [56], and the total scattering rate was calculated by combining the following contributions via Matthiessen’s rule [57],
The input parameters, such as the electronic structure, dielectric tensor, elastic constants, deformation potential, and polar optical phonon frequencies were obtained from first-principles calculations.

Figure 1. Convex hull diagram for CsAg3S2 in trigonal phase. The color scale denotes the formation energy in eV/atom. Black circles indicate phases on the convex hull, while orange diamonds denote phases above the hull.
Results and Discussion
Structural Stability
CsAg3S2 belongs to the AM3Ch2 (Where, A = alkali metal, M = Cu, Ag, Au and Ch = S, Se, Te) family of ternary chalcogenides, whose members crystallize in two closely related structure types: monoclinic C2/m and trigonal Pm1, featuring corrugated [M3Ch2] units separated by alkali cations [30, 32]. CsAg3S2 was first synthesized in 1977 by the high-temperature melt reaction method [32] and has since been reported only in the monoclinic form. Despite sharing the same bonding hierarchy, both structure types differ in the spectral sparsity of phonon modes they produce. The trigonal cell contains single formula units, yielding a phonon spectrum with well-separated optical branches. In contrast, the large monoclinic cells distribute twice as many branches within a similar frequency range. The coherent phonon transport becomes significant when the phonon linewidths become comparable to the interbranch spacing between the phonon modes [58], and therefore the monoclinic structure is prone to enhanced wave-like thermal transport. The trigonal structure already realized in Cu and Au members of the family [30] therefore provides a natural platform to test the hypothesis that targeted engineering of the M-Ch framework can drive particle-like transport into the ultralow regime while leaving the coherent channel undisturbed.
Before examining the lattice dynamics and transport properties, it is essential to establish the structural stability of CsAg3S2 in the trigonal phase and its energetic competitiveness with the experimentally reported monoclinic phase. We start the quest of stability by analyzing the formation energy of both polymorphs. The formation energies were computed using the expression
Where is the total energy per formula unit, whereas , , and are the energies of the respective elements per atom. Our calculated formation energies for CsAg3S2 in the trigonal and monoclinic phases are -2.87 eV and -2.37 eV, respectively. Both phases exhibit negative formation energies, indicating stability against decomposition into constituent elements. The more negative value of formation energy in the trigonal phase indicates that it is more favorable energetically as compared to the experimentally observed monoclinic phase.
Having established the energetic favorability of both polymorphs, we next assess their thermodynamic stability against competing phases in the Cs-Ag-S chemical space through convex hull analysis. The convex hull provides a rigorous measure of the stability of the material with respect to all possible combinations of competing phases [59]. Phases lying on the convex hull are thermodynamically stable, whereas materials lying above the convex hull are prone to decomposition into the competing phases and therefore thermodynamically unstable. We constructed the convex hull using all available binary and ternary phases from the OQMD database [60, 61], including experimentally reported compounds and theoretically predicted structures based on common structure prototypes. A total of 48 competing phases were optimized using identical computational parameters employed for CsAg3S2, followed by static calculations to obtain their respective formation energies and ensure computational consistency. As shown in Fig. S3 [62], the trigonal phase lies on the convex hull, indicating its thermodynamic stability at 0 K. In contrast, the experimentally reported monoclinic phase lies 0.057 eV/atom above the hull. This relatively small hull distance is consistent with the experimental realization of the monoclinic phase [63]. The convex hull analysis therefore establishes the thermodynamic stability of the trigonal phase against decomposition into competing phases at 0 K.
Establishing the thermodynamic stability, we conducted the AIMD simulations at 300 K, 500 K, and 700 K to verify the structural stability at higher temperatures. As shown in Fig. S2 (b) [62] the free energy fluctuated only slightly during the 20 ps AIMD run, and no evidence of structural distortion or phase transitions was observed. This suggests that CsAg3S2 can withstand higher temperatures relevant to the practical operating conditions.
Having confirmed the thermodynamic and thermal stability of trigonal CsAg3S2, it is equally important to examine its mechanical stability, which is necessary for assessing a material’s ability to withstand external stresses and mechanical load without undergoing perma- @page 5/17 starts_mid=0 ends_mid=1 nent structural deformation or collapse. Our computed elastic constants, listed in Table X, obey the Born–Huang elastic stability criteria given by
The above findings underscore the mechanical robustness of CsAg₃S₂. Complementing this, the absence of imaginary frequencies in the phonon dispersions [see Figure 3 (a) and (e)] establishes the dynamical stability, thereby ruling out any possibility of soft-mode-driven structural phase transition. Collectively, the favorable formation energy, thermodynamic stability against competing phases as evidenced by the convex hull analysis, finite-temperature structural integrity under AIMD simulations, and the fulfillment of mechanical and dynamical stability criteria provide compelling evidence for the intrinsic stability of the trigonal phase of CsAg₃S₂. Having firmly established the stability of CsAg₃S₂ in the trigonal symmetry, we now turn our attention to its crystal structure and bonding characteristics.

Figure 3. (a) Phonon dispersions for CsAgS including quartic anharmonicity (loop diagram). (b) Temperature dependence of the lowest and (c) highest optical modes using loop and bubble diagrams. (d) Phonon dispersion for CsAgS including both cubic (bubble diagram) and quartic anharmonicity (loop diagram). (e) Calculated mean-square displacements for Cs, Ag, and S atoms within the SCP and SCPB frameworks, respectively. Note that Ag and S values overlap.
Crystal Structure and Bonding
The crystal structure of trigonal CsAg₃S₂ is shown in Figure 2 (a) and (b). CsAg₃S₂ crystallizes in the space group (No. 164) with 6 atoms in the primitive cell, where the Cs, Ag, and S occupy the 1b, 3e, and 2d Wyckoff positions, respectively. The optimized lattice parameters are listed in Table S1 [62]. The structure is isostructural with CsCu₃S₂ [64] and consists of corrugated [Ag₃S₂]⁻ layers parallel to the (001) plane, separated by Cs⁺ ions. Within this structure, the Ag atoms are in strictly linear coordination, while S is triply coordinated within the Ag-S framework. In contrast, Cs is surrounded by S atoms forming a distorted octahedral environment [Fig. S4 (a) [62]], with a Cs-S distance of 3.81 Å and S-Cs-S angles of 76.6° and 103.4°. The distorted Cs environment, together with the large mass of Cs and its weak interaction with the Ag-S framework, is expected to produce soft, low-frequency Cs-dominated modes and strong lattice anharmonicity.
Figure 2. (a) Top view of CsAg₃S₂ crystal structure in the trigonal phase. (b) Corrugated Ag-S layers running parallel to the (001) direction. (c) Calculated crystal orbital Hamilton population (ICOHP) for CsAg₃S₂. The inset shows the enlarged view near the Fermi level featuring the Ag-S antibonding interactions. (d) Integrated COHP and bond distance analysis for CsAg₃S₂.
To further understand the bonding characteristics, we performed crystal orbital Hamilton population analysis as shown in Figure 2 (c) and (d). The Ag-S interactions exhibit pronounced bonding states between -4 eV to -6 eV, with an integrated COHP (ICOHP) value of -1.85 eV, confirming the strong covalent character of the Ag-S framework. In comparison, the Cs-S and Cs-Ag interactions are considerably weaker, with ICOHP values of -0.21 eV and -0.15 eV, respectively. For the isostructural CsCu₃S₂, when computed within identical input parameters, the corresponding Cu-S and Cs-S values are -2.35 eV and -0.75 eV, respectively [see Fig. S5 in SI [62]]. This indicates that Cu to Ag substitution not only weakens the covalent framework but also diminishes the Cs-S interaction strength arising from the enlargement of the CsS₆ cavity and increased Cs-S bond lengths. Notably, antibonding Ag-S states occur near the Fermi level due to Ag- and S- hybridization. Although these states represent only a small fraction of the occupied manifold, their presence provides an additional source of anharmonicity while the overall Ag-S framework remains strongly bonded [65].
This bonding picture is further supported by Bader charge analysis. Cs, Ag, and S carry charges of +0.84 , +0.19 , and -0.70 , respectively, indicating substantial charge transfer from Cs and appreciable charge sharing between Ag and S. Thus, the lattice exhibits a clear bonding hierarchy consisting of a strongly bonded covalent Ag-S framework with a weakly coupled Cs sublattice. The large contrast in bond strength and atomic mass is expected to produce distinct vibrational characteristics of Cs and Ag-S sublattices, providing the structural basis for the anomalous lattice dynamics and ultralow thermal conductivity as discussed in subsequent sections.
Anharmonic Lattice Dynamics
The temperature-renormalized phonon dispersions of CsAg₃S₂ are presented in Figure 3. Upon inclusion of the quartic anharmonicity through the loop diagram, the phonon spectrum exhibits a strong mode-dependent temperature response. The acoustic and low-frequency optical branches below 130 cm⁻¹ progressively harden with increasing temperature, whereas the intermediate frequency modes between 212 and 280 cm remain nearly unchanged. In contrast, the high-frequency optical modes soften at elevated temperatures. This contrasting behavior reflects the distinct contribution of quartic anharmonicity to different regions of the phonon spectrum. The loop-only contribution to phonon self-energy is known to overestimate the phonon frequencies [49, 66]. We therefore include the cubic anharmonicity through the bubble diagram to obtain fully renormalized phonon frequencies [49]. As shown in Figure 3 (e), the inclusion of the bubble contribution substantially modifies the temperature evolution of the phonon spectrum. The combined effect of the loop and bubble self-energies determines the net temperature renormalization.
The high-frequency optical modes undergo a pronounced softening because the cubic and quartic contributions act cooperatively. The low-frequency optical modes also demonstrate an overall softening behavior; the temperature dependence is, however, weak due to the competing effects of quartic and cubic anharmonicities. Similar mode-dependent temperature renormalization was also observed in CsCuS [33], CsCuSe [67], and CsCuI [68].
The contrasting roles of cubic and quartic anharmonicity are further illustrated by tracing the lowest and highest frequency optical modes, M and M, respectively [See Fig. 3 (b), (c)]. The M mode hardens with temperature under both quartic and cubic anharmonicities, although the bubble contribution reduces its temperature dependence. In contrast, M softens under both treatments, with the bubble contribution substantially enhancing its temperature dependence. These results indicate that the mid- and high-frequency modes are strongly influenced by cubic anharmonicity, whereas the low-frequency optical modes exhibit a significant dependence on quartic anharmonicity. Compared to the isostructural CsCuS, which exhibits the softening of high-frequency optical modes along with the hardening of the acoustic and low-frequency optical modes [33], CsAgS exhibits contrasting temperature dependence. Both the high- and low-frequency optical modes show appreciable softening. Notably, the acoustic modes show a decrease in frequency as the temperature increases, indicating that the Ag-S framework can effectively suppress the group velocities of these modes and is therefore expected to lower the particle-like thermal conductivity.
The atom-resolved phonon density of states provides further insights into the microscopic origin of the unusual lattice dynamics in CsAgS, as shown in Fig. S6 [62]. The distinct mass and bonding hierarchy of Cs, Ag, and S results in a clear separation of their vibrational contributions across the phonon spectrum. The acoustic branches are dominated by Cs vibrations, while the low-lying optical modes in the frequency range 50-150 cm arise predominantly from Ag atoms. In contrast, the high-frequency optical manifold above the phonon gap, extending from 210-350 cm, is dominated by the lighter S atoms. Additionally, an avoided crossing between the acoustic and low-lying optical branches is observed along the -A direction. Such hybridization is characteristic of rattling-like motion of the loosely bound Cs atoms, which is further supported by the larger mean square displacement (MSD) values of Cs atoms compared to Ag and S atoms [see Figure 3 (e)] and weak Cs-S interactions. This hybridization leads to the branch flattening of the interacting modes and consequently reduces phonon group velocities [19].
To elucidate the microscopic nature of the vibrational modes, we computed the atomic participation ratio (APR) and participation ratio (PR) [69], defined as
where is the atom index, denotes the number of atoms, represents the phonon eigenvector with wave number and mode index , and is the atomic mass. The APR quantifies the contribution of individual atoms to a particular phonon mode, whereas PR can be used to distinguish extended propagative modes from localized vibrations.
As shown in Figure 4 (a)-(Figure 4c), the vibrational modes below 50 cm are predominantly associated with Cs, whereas Ag dominates the low-lying optical modes spanning from 50-150 cm. Above the phonon gap, the high-frequency optical modes are dominated by S, consistent with its low mass and strong incorporation into the covalent framework. Notably, the Cs contribution rapidly vanishes above 50 cm, indicating its confined motion to a narrow frequency window rather than being distributed throughout the spectrum. Additionally, the PR shown in Figure 4 (d) further resolves the character of the low-frequency manifold. The acoustic modes exhibit PR values close to unity, characteristic of extended propagative vibration, whereas the PR decreases sharply above 25 cm with pronounced localization in the 25-50 frequency window. Thus, the Cs sublattice contributes to both the acoustic modes and the localized vibrations immediately above them. The avoided crossing between the Cs-derived modes hybridizes the localized optical character with the acoustic branches, leading to branch flattening, reduced group velocities, and enhanced anharmonic scattering. A similar mechanism of mode localization and suppression of particle-like thermal transport has been reported in filled skutterudites [70], type-I clathrates [71], and related chalcogenides such as CsCuS [33] and CsCuSe [67].

Figure 4. Atomic participation ratio for (a) Cs, (b) Ag, and (c) S atoms at 300 K. (d) Total participation ratio for CsAgS at 300 K.
Taken together, the lattice dynamics of trigonal CsAgS reveal complementary mechanisms for suppressing particle-like thermal transport: Cs-derived localized modes enhance anharmonic scattering and shorten phonon lifetimes, while the heavy Ag atoms in the more weakly bonded Ag-S framework reduce the group velocity of heat-carrying phonon modes. Compared with the isostructural CsCuS, the Cu to Ag substitution enhances both mechanisms by weakening the covalent framework and loosening the Cs rattlers. The combined influence of shortened phonon lifetimes and diminished group velocities leads to a substantial reduction in particle-like thermal transport, as quantified below.
Lattice Thermal Conductivity
To elucidate the interplay between anharmonic phonon renormalization and multiphonon scattering in governing lattice thermal transport in CsAgS, we calculated the lattice thermal conductivity using temperature-renormalized phonon frequencies obtained within the HA, SCP, and SCPB frameworks together with three-phonon (3ph) and combined three- and four-phonon (3,4ph) scattering processes. Consequently, we constructed six theoretical models, namely HA+3ph, HA+3,4ph, SCP+3ph, SCP+3,4ph, SCPB+3ph, and SCPB+3,4ph. As shown in Figure 5 (a)-(Figure 5c), all six approaches predict pronounced anisotropy in , with the in-plane component substantially exceeding the out-of-plane component. Within the SCPB+3,4ph framework, reaches 0.22 and 0.05 W m K along the - and -axis at 300 K, as compared to 1.80 and 1.37 W m K for the isostructural CsCuS [33]. The in-plane reduction reflects the heavier and more weakly bonded Ag-S framework, whereas the much stronger suppression along -axis is consistent with the considerably weaker Cs-S interaction, since heat transport across the layers is mediated by the Cs-S contacts.

Figure 5. Temperature dependence of particle-like thermal conductivity () for CsAgS along in-plane and out-of-plane directions within (a) HA, (b) SCP, and (c) SCPB frameworks considering 3ph (circles) and 3,4ph (diamonds) processes. The downward arrows represent the fitted temperature dependence . (d) Spectral (shaded, left axis) and cumulative (line, right axis) at 300 K along in-plane and out-of-plane directions considering 3ph processes within the HA framework. (e) Same as (d) but considering 3,4ph processes. (f) Same as (d) but within the SCP framework. (g) Same as (e) but within the SCP framework. (h) Same as (d) but within the SCPB framework. (i) Same as (e) but within the SCPB framework.
Figure 5 (a)-(Figure 5c) further illustrates the systematic temperature dependence of as predicted by different theoretical models. Within the 3ph approximation, the SCP+3ph model yields the highest (0.49 and 0.13 W/mK along the - and -axis, respectively, at 300 K), followed by HA+3ph (0.33 and 0.07 W/mK) and SCPB+3ph (0.27 and 0.06). The enhancement within SCP+3ph framework arises from the quartic-induced hardening of the acoustic and low-frequency optical modes, which increases the group velocities, whereas the bubble self-energy in SCPB+3ph framework softens these modes and markedly suppresses .
The inclusion of 4ph scattering systematically reduces in both directions throughout the investigated temperature range. The reduction becomes progressively larger with increasing temperature, indicating the growing importance of higher-order anharmonic scattering at elevated temperatures. The in-plane decreases by 24–40%, 16.5–27%, and 17–29% within HA, SCP, and SCPB, respectively, with corresponding out-of-plane reductions of 23–36%, 15–23%, and 16–24%. These values are comparable to those reported for CsAgBiBr (34.6% and 51.8% at 300 and 700 K) [46] and CsCuSe (∼46% in-plane and ∼38% cross-plane at 300 K) [67]. Notably, the influence of 4ph scattering is consistently stronger within the HA approximation, whereas the results are comparatively smaller for the SCP and SCPB frameworks, suggesting that temperature-induced renormalization partially compensates for additional thermal resistance introduced by the 4ph scattering channels by modifying the phonon spectrum and the scattering phase space. Nevertheless, the inclusion of 4ph scattering does not alter the pronounced anisotropy of , but primarily lowers its magnitude by introducing additional resistive phonon scattering processes that shorten the phonon lifetimes.
To quantify the temperature dependence of the lattice thermal conductivity, we fitted to the power law expressed as , where can be used as an effective descriptor of the temperature evolution of underlying phonon scattering mechanism. Within the HA+3ph framework, exhibits a temperature dependence of () along a/b-axis (c-axis), close to the classical dependence of Umklapp-limited heat transport. In contrast, SCP+3ph yields a much weaker temperature dependence of (), significantly diverging from the scaling, reflecting the partial compensation of the temperature-induced increase in scattering by quartic-anharmonic phonon renormalization. Including the bubble correction restores the stronger temperature dependence of () within the SCPB+3,4ph approach, consistent with the phonon softening induced by cubic anharmonicity. With 4ph scattering, the exponents evolve to (), (), and () within HA+3,4ph, SCP+3,4ph, and SCPB+3,4ph, respectively. The obtained in-plane SCPB+3,4ph exponent is comparable to that of CsCuS () [33], indicating that the Cu to Ag substitution mainly lowers the magnitude of rather than changing its temperature scaling. Such deviations from classical behavior are also reported in other strongly anharmonic materials [46, 72, 73, 74].
To gain further insight into the microscopic origin of the ultralow , we analyze the spectral and cumulative lattice thermal conductivities, in Figure 5 (d)-(Figure 5i). Irrespective of the theoretical framework adopted, the cumulative conductivity shows that the c-axis thermal transport is almost entirely contributed by phonon modes below frequency cm, whereas along the -axis more than half of originates from the same frequency range and exceeds 80 % after including modes up to cm. Together with the projected phonon density of states, this shows that the c-axis transport is governed by Cs-dominated acoustic and low-lying optical modes which explains its strong sensitivity to weak Cs-S coupling, while Ag-related modes between 50 and 150 cm provide an additional in-plane contribution. The high-frequency optical modes contribute negligibly. The inclusion of 4ph scattering reduces the magnitude of without appreciably changing the frequency distribution of heat-carrying phonons.
To further elucidate the origin of suppressed , we examine the phonon group velocities () in Figure 6 (a)-(Figure 6c). The group velocities are higher in the in-plane direction than along the out-of-plane directions, which is consistent with the bonding hierarchy and the anisotropy in . The acoustic branches retain the largest and dominate the particle-like transport. Importantly, the evolution of across the low- and intermediate-frequency regimes reveals two complementary mechanisms for suppressing . A pronounced dip in around 40-60 cm arises from the Cs-dominated rattling modes, which flatten the low-frequency optical branches at the avoided crossing. Beyond this region, decreases progressively across the Ag-dominated part of the spectrum, reflecting the heavier and weakly bonded Ag-S framework compared to Cu-S framework in CsCuS, which lowers the phonon frequencies. Thus, while the Cs-rattling primarily impedes the low-frequency phonons, the softer Ag-S framework extends the suppression of to the heat-carrying low-lying optical modes, indicating that the two mechanisms work cooperatively to reduce . The recovery of in the optical branches above the phonon gap is consistent with the small contribution to the cumulative near 200 cm, although its overall magnitude remains low.

Figure 6. (a) Phonon group velocities as a function of frequency at 300 K along in-plane and out-of-plane directions within the HA framework. (b) Same as (a) but within the SCP framework. (c) Same as (a) but within the SCPB framework. (d) Three-phonon (3ph), four-phonon (4ph), and isotope (iso) scattering rates as a function of frequency at 300 K within the HA framework. (e) Same as (d) but within the SCP framework. (f) Same as (d) but within the SCPB framework.
We next examine the phonon scattering rates governing the phonon lifetimes and thermal transport in Figure 6 (d)-(Figure 6f). The scattering rates, in general, increase with frequency, with negligible isotope scattering compared to the anharmonic 3ph and 4ph contributions. 3ph scattering dominates over most of the frequency spectrum; however, near the avoided crossing region, the 3ph rates decrease while the 4ph rates increase sharply and become comparable to the 3ph contributions. The interaction between the acoustic and low-lying optical branches opens a small frequency gap and flattens the dispersions, restricting certain 3ph processes that can simultaneously satisfy energy and momentum conservation. In contrast, the localized, rattling-like Cs vibrations weakly bonded to the Ag-S framework, provide strong fourth-order anharmonicity and enhanced 4ph scattering. A similar behavior was reported for CsAgBiBr, where the 4ph scattering rates are comparable to the 3ph rates at 300 K [46]. The fourth-order processes therefore make a significant contribution to the suppression of particle-like thermal transport in CsAgS.
To gain further insights into the underlying scattering mechanisms, we resolve the 3ph and 4ph scattering rates into their respective sub-processes. The 3ph processes comprise recombination () and splitting (). As shown in Figure 7 (a)-(Figure 7c), recombination dominates at low frequencies, whereas splitting grows rapidly with frequency and dominates for the high-frequency optical modes, consistent with the larger splitting phase space (Fig. S7 (a)-(c)[62]). Notably, only the recombination channel is suppressed in the avoided crossing region, indicating that the localized Cs-derived modes restrict recombination processes and thereby reduce the total 3ph scattering rates. Similarly, the 4ph processes can be decomposed into recombination (), splitting (), and redistribution () sub-processes. As shown in Figure 7 (d)-(Figure 7e), the redistribution channel dominates throughout the spectrum and exhibits a pronounced enhancement near the avoided crossing. The corresponding 4ph scattering phase space also reaches its maximum in this region [see Fig. S7 (d)-(e) [62]], establishing redistribution as the main source of the enhanced fourth-order scattering corresponding to the localized Cs modes. Similar redistribution-dominated 4ph scattering is also reported in CsAgBiBr [46], while flat rattler modes driven strong 4ph scattering was observed in CsCuSe [67], AgCrSe [75] and CuSbS [76]. These results highlight the important role of higher-order anharmonic scattering in suppressing the particle-like thermal transport in CsAgS.

Figure 7. (a) Process-resolved 3ph scattering processes as a function of frequency at 300 K within the HA framework. (b) Same as (a) but within the SCP framework. (c) Same as (a) but within the SCPB framework. (d) Process-resolved 4ph scattering processes as a function of frequency at 300 K within the HA framework. (e) Same as (d) but within the SCP framework. (f) Same as (d) but within the SCPB framework. (g) Umklapp (U) and normal (N) contributions to the 4ph processes as a function of frequency at 300 K within the HA framework. The inset shows the ratio of Umklapp to normal scattering processes (U/N). (h) Same as (g) but within the SCP framework. (i) Same as (g) but within the SCPB framework.
We further resolved the 4ph scattering rates into Umklapp (U) and normal (N) processes [Fig. 7 (g)-(h)]. U processes violate crystal-momentum conservation and directly provide thermal resistance, whereas N processes conserve total crystal momentum and redistribute it among the phonons without directly generating any thermal resistance. Both U and N rates increase with frequency and peak near the avoided crossing region, indicating that the enhanced 4ph scattering in this frequency region arises from both channels. Additionally, the U/N ratio provided in the inset reveals the dominance of the U processes in the 4ph scattering rates, suggesting that the 4ph scattering suppresses primarily by generating thermal resistance.
To elucidate the respective effects of anharmonic phonon renormalization and 4ph scattering on particle-like thermal transport, we examine the thermal conductivity ratios , , and , for renormalization and , , and for 4ph scattering [77] in Fig. S9 of SI [62]. The loop-only renormalization increases by 70% relative to the harmonic-only treatment, whereas the bubble correction mitigates this effect, yielding a 10% decrease relative to the harmonic-only value. Consequently, the bubble correction reduces by relative to the SCP theory, significantly larger than the decrease reported for isostructural CsCuS [33] for the same level of theory. This indicates a considerably stronger cubic anharmonicity in CsAgS, consistent with its weaker bonding. Four-phonon scatter-
ing further suppresses with strongest reduction within HA (24%), followed by SCPB (17%) and SCP (16%) at 300 K. The detailed temperature and axis dependence of these ratios are discussed in the SM [62].
Wave-like Thermal Transport
Having established the strong suppression of particle-like thermal transport in CsAgS, we now turn our attention to the coherence’s contribution () to thermal transport. originates from the off-diagonal component of the Wigner heat-flux operator [25]. When the phonon linewidth becomes comparable to the interbranch spacing, this leads to coherent Zener-like tunneling between phonon modes, enabling wave-like heat transport [26]. This provides an intrinsic challenge for reducing the total thermal conductivity. The structural features that suppress through strong anharmonic scattering and phonon mode softening can simultaneously broaden phonon linewidths and enhance interbranch coupling, thereby increasing [28]. This compensation in CsAgS remains remarkably weak. The calculated at 300 K is 0.057 W m K and increases to 0.098 W m K at 700 K within the SCPB+3,4ph framework and remains below 0.1 W m K for all levels of theoretical frameworks considered [see Fig. 8 (a) and Figs. S10 (a)–(c) and S11 (a)–(c) in SI [62]. The origin of this suppression can be traced to the phonon dispersions. The strongly bonded light S atoms form sparse, spectrally isolated, high-frequency optical modes with large interbranch separations, thereby reducing the coherent coupling between the neighbouring phonon modes. Importantly, although the weaker Ag–S bonding shifts the spectrum to lower frequencies than in CsCuS, the phonon gap and the large interbranch separation persists, so the substitution that suppresses does not compromise the spectral sparsity that limits . Thus, while the weakly bonded Cs sublattice and the softer Ag–S framework suppress , the S-dominated high-frequency manifold independently constrains , so that both transport channels are limited simultaneously by distinct features of hierarchical lattice dynamics
To elucidate the microscopic origin of coherence’s thermal conductivity, we further analyze the spectral and cumulative in Figure 8 (b) and Figs. S10 (d)–(f), S11 (d)–(f) [62]. The coherence’s contribution is concentrated predominantly below cm, with most in-plane arising from this frequency window. Along the out-of-plane direction, exhibits two distinct windows contributing to ; below cm and between –270 cm. The latter coincides with the lower part of the S-dominated optical modes, indicating a near-degenerate mode coupling. Beyond this region, the large interbranch separations suppress coherent coupling, rendering

Figure 8. (a) Temperature-dependent coherence’s thermal conductivity () within the SCPB+3,4ph framework. (b) Spectral (shaded, left axis) and cumulative (line, right axis) within the SCPB+3,4ph framework. (c) Temperature-dependent total lattice thermal conductivity for CsAgS computed within different theoretical frameworks. The downward arrows represent the fitted temperature dependence . (d) Contribution of and to at 300 K and 700 K using different theoretical frameworks considering 3ph and 3,4ph scattering processes. (e) Phonon lifetimes and (f) mean-free-paths for CsAgS as a function of frequency at 300 K. The circle area is proportional to the modal contribution to and the colorbar indicates the origin of the transport, from particle-like (green color) to wave-like (blue color). In (e), the horizontal line denotes the Wigner limit in time , and the curve marks the Iofee–Regel limit in time (). In (f), the horizontal line represents the Iofee–Regel limit in space.
the high-frequency optical modes inactive in heat transfer. Thus, while strong anharmonicity promotes coherent transport through linewidth broadening, substantial additionally requires a dense, closely packed phonon spectrum of energetically accessible neighbouring modes. This spectral contrast is also observed in isostructural CsCuS [33] and CsReS [29], where sparse optical modes contribute weakly to wave-like thermal transport.
Combining both channels, we obtain an ultralow thermal conductivity of 0.222 W m K at 300 K, which further reduces to 0.156 W m K at 700 K within the SCPB+3,4ph [see Figure 8 (c)]. Because the growing contribution from partially offsets the effect of decreasing , follows a weak temperature dependence of , in contrast to the and dependencies of in-plane and out-of-plane , respectively. The room temperature values are comparable to those of materials with simultaneously suppressed and , such as such as CsReS [29] and TlGeS [24]. Furthermore, in CsAgS is nearly an order of magnitude smaller than 1.66 W m K reported for isostructural CsCuS, reflecting the combined weakening of the Ag-S framework and Cs-S coupling upon Cu to Ag substitution.
The SCP+3,4ph description consistently overestimates relative to the SCPB+3,4ph by 60.36 % at 300 K and 62.17 % at 700 K, highlighting the importance of the cubic bubble correction. Interestingly, the HA+3,4ph and SCPB+3,4ph approaches yield nearly identical at 300 K despite their substantially different , owing to the enhanced within the SCPB treatment, which compensates for its lower . This agreement disappears at elevated temperatures as acquires a stronger temperature dependence within the SCPB+3,4ph framework. The inclusion of four-phonon scattering further suppresses for all renormalization schemes, although the magnitude of this reduction decreases at elevated temperatures. This can be attributed to the growing , which partially compensates for the reduction in .
Next, we quantify and analyze the individual contributions of and to in Figure 8 (e) and Table S3 in SM [62]. The relative contributions reveal a pronounced temperature-driven crossover in CsAgS. At 300 K, particle-like transport dominates within the SCPB+3,4ph framework, with accounting for approximately 74% of the total . With increasing temperature, decreases rapidly, whereas increases. At 700 K, the coherence channel contributes 62.82% surpassing . A similar particle-to-wave crossover has been reported in strongly anharmonic materials such as CuBiS [78], TlAgI [79], AgTlI [80]. In contrast, remains dominant in CsCuS up to 900 K [33], indicating that the crossover is enabled by the stronger suppression of upon Cu to Ag substitution. The important distinction in CsAgS is that the crossover results primarily from the continued suppression of by weakly bound Cs atoms and softer Ag-S framework, rather than from the emergence of large absolute coherence conductivity, as remains below 0.1 W m K. Therefore, this crossover reflects the progressive collapse of particle-like channel against a comparatively weak coherence background. Therefore, CsAgS provides a platform in which the dominant transport mechanism changes with temperature while both channels remain strongly suppressed.
To further elucidate the origin of the suppressed coherence channel and ultralow thermal conductivity in CsAgS, we analyze the frequency-dependent lifetime and mean free path in Figure 8 (e) and (f). Here, the area of each circle is proportional to each mode’s contribution to , and the color bar represents the origin of the contribution. The green color represents the particle-like contribution, whereas the blue color corresponds to the wave-like tunneling channel. The intermediate colors represent a regime where both transport channels coexist and contribute to the heat conduction. Modes with lifetime above the Wigner limit in time, defined as [27], retain predominantly particle-like character, whereas modes with progressively acquire the wave-like character. As shown in Figure 8(e), the acoustic and nearby low-lying optical modes largely remain above and dominate the thermal transport. The high-frequency modes increasingly fall below this limit, but contribute only weakly to overall thermal conductivity, consistent with the spectral in Figure 8 (b). An analogous picture emerges in the frequency-mean-free-path domain Figure 8(f), where the high-frequency modes fall below the Ioffe-Regel limit in space [25, 81], at which the mean-free-path becomes comparable to the average interatomic spacing, indicating a predominantly wave-like character of these modes. The contribution to , however is small owing to the large separation of the high-frequency branches. Thus, the loss of quasi-particle character does not ensure a large coherent contribution.
Figure 9 (a)–(c) and (d)–(f) resolve in terms of the frequencies and of the two coupled modes within the SCP+3,4ph and SCPB+3,4ph frameworks, respectively. At 300 K, originates mainly from coupling between the quasi-degenerate phonon modes below 100 cm. This can be attributed to the strong 4ph scattering rates, and hence broader phonon linewidths of highly localized phonon modes associated with the rattling-like vibrations of weakly bound Cs atoms. In contrast, the high-frequency modes above 100 cm have minimal contribution, consistent with the cumulative analysis related to Figure 8 (b). At higher temperatures, increased scattering broadens the linewidths and promotes coupling between the phonon modes of increasingly dissimilar frequencies. A comparison of the two frameworks shows that the enhancement is stronger for the SCPB+3,4ph framework than for the SCP+3,4ph framework. Accordingly, at 700 K, the contribution from the high-frequency phonons with larger frequency differences has a finite contribution to coherence’s thermal transport. This behavior differs from CsAgBiBr [46], where the high-frequency modes contribute significantly to . This phenomenon resembles the behavior observed in CsCuS [33], CsReS [29], and TlSiS [24], where the high-frequency modes do not actively participate in the wave-like thermal transport. Together, these results show that the weakly bound Cs atoms and the softer Ag-S framework drive low , while the sparse S-derived optical branches, preserved upon substitution, keep small, leading to the simultaneous suppression of both transport channels.

Figure 9. (a) Three-dimensional visualization of modal contributions to coherence’s thermal conductivity resolved in terms of frequencies and at 300 K within the SCP+3,4ph framework. (b) Same as (a) but at 500 K. (c) Same as (a) but at 700 K. (d) Same as (a) but within the SCPB+3,4ph framework. (e) Same as (d) but at 500 K. (f) Same as (d) but at 700 K. Contributions along the diagonal () arise from the coupling between the quasi-degenerate phonon modes.
Electronic Structure and Thermoelectric Transport
The obtained ultralow thermal conductivity makes CsAgS an interesting prospect for thermoelectric applications. Therefore, to complement the analysis of weak thermal transport, we further examined the electronic structure and electrical transport properties to assess the overall thermoelectric performance. The calculated electronic band structure Fig. S14 [62] reveals a semi-conducting nature with an indirect band gap of 2.2 eV. Additionally, the multi-valley nature in the conduction band region, in conjunction with the multi-valley convergence in the valence band region around point A, further provides a suitable electronic environment for favorable electrical transport properties [82]. Furthermore, we note a distinct orbital character in the electronic band structure. The valence band maximum (VBM) is predominantly derived from the Ag and S states, whereas the conduction band minimum (CBM) is mainly composed of Ag and S states. This indicates that the electronic states near the band edges are primarily governed by the Ag-S framework, with dominant - hybridization in the CBM region and - hybridization in the VBM region which is consistent with the antibonding Ag-S states near the Fermi level identified in the COHP analysis.
The carrier-dependent transport coefficients (Fig. S16 [62]), obtained within the MRTA, exhibit the expected competition between the Seebeck coefficient () and electrical conductivity (). The Seebeck coefficient increases with temperature but decreases with the carrier concentration. In contrast, electrical conductivity increases with carrier concentration and decreases with temperature. The electronic thermal conductivity follows the trend of electrical conductivity closely, consistent with the Wiedemann-Franz law. The relatively large band gap further suppresses bipolar transport throughout the investigated temperature range. Despite the favorable , the comparatively low limits the power factor. Combining and , we obtained the highest power factor values of 8.07 W cm K and 3.17 W cm K for - and -type doping, respectively, at 700 K.
The resulting thermoelectric figure of merit () [see Figure 10], determined by the and the ultralow lattice thermal conductivity, displays a characteristic nonmonotonic dependence on carrier concentration. increases initially with carrier concentration for either type of doping to reach a maximum value at an optimum carrier concentration, and subsequently decreases at higher carrier concentrations. Furthermore, the calculated increases with temperature, reaching maximum values of approximately 1.28 and 0.88 at 700 K for -type and -type doping, respectively, corresponding to a carrier concentration of cm. Altogether, these results indicate that, although the electrical transport properties contribute to the thermoelectric response, the ultralow lattice thermal conductivity plays a key role in enabling respectable values at elevated temperatures.

Figure 10. Thermoelectric figure of merit as a function of carrier concentration at various temperatures for - and -type doping.
Conclusion
In summary, we investigated the structural, vibrational, and lattice thermal transport properties of trigonal CsAgS using first-principles calculations in conjunction with self-consistent phonon theory, higher-order anharmonic scattering, and unified theory of thermal transport. Although CsAgS has so far been reported in the monoclinic symmetry, the trigonal phase is energetically accessible, dynamically, mechanically, and structurally stable up to 700 K. The bonding analysis suggests the presence of a unique bonding hierarchy consisting of covalently bonded [AgS] units and a loosely bonded Cs sublattice. Compared to isostructural CsCuS, the Cu to Ag substitution weakens both the covalent framework and the Cs-S interactions, giving rise to an exceptionally low of 0.222 W m K at 300 K and 0.16 W m K at 700 K with a weak temperature dependence of . The localized Cs modes enhance anharmonic scattering and shorten phonon lifetimes, and the heavier, softer Ag-S framework reduces phonon group velocities, together suppressing particle-like transport, while the sparsely sparse and well-separated S-derived optical branches restrict coherent phonon coupling, leading to the simultaneous suppression of both transport channels. With increasing temperature, the coherent contribution rises from about 26 % to 63 %, leading to a crossover from particle-like thermal transport to wave-like tunneling. This crossover is not observed in CsCuS and arises primarily from the decrease of rather than from a large increase in . Combined with the computed electrical transport parameters, this yields a of approximately 1.28 at 700 K for -type doping.
Beyond establishing ultralow thermal conductivity and clarifying the distinct roles of particle- and wave-like transport channels in the thermal transport of trigonal CsAgS, our results show that modifying the covalent framework in low thermal conductivity materials featuring bonding hierarchy structural motif can provide an additional route to suppress without disturbing the mechanisms limiting . In CsAgS, the heavier and softer Ag-S framework suppresses the phonon group velocities, while the accompanying weakening of Cs-S coupling reduces the phonon lifetimes by promoting the rattler-driven anharmonic scattering. Together with the sparse S-derived optical modes limiting the coherent coupling, these mechanisms provide a route for simultaneous suppression of both heat transport channels. More broadly, systematic modification of the covalent framework across the AMCh family and related bonding hierarchy materials provide a natural platform to examine its generality. Since the framework modification also tunes the coupling of the ionic sublattice, it may offer a route to modulate the relative contribution of and . We believe this study will motivate further investigations to test this strategy for design and discovery of ultralow thermal conductivity materials.
Acknowledgments
We would like to thank Md. Zeeshan for the useful discussions. V.G. is thankful to BUKRP00034052 for the financial assistance. B.K.M. acknowledges the funding support from the SERB, DST (CRG/2022/000178). The calculations were performed using TEJAS cluster at the Department of Physics, IIT Delhi and the PARAM Rudra, a national supercomputing facility at the Inter-University Accelerator Centre (IUAC), New Delhi.
References
- [1]G. J. Snyder and E. S. Toberer, Nature materials 7, 105 (2008).
- [2]M. Mukherjee, A. Srivastava, and A. K. Singh, Journal of Materials Chemistry C 10, 12524 (2022).
- [3]Y. Wang, W. Zhang, J. Du, D. Li, and R. Liu, Journal of the European Ceramic Society 45, 117038 (2025).
- [4]J. Liu, Y. Li, S. Li, and P. Chen, Ceramics International 48, 10480 (2022).
- [5]T. Zhou, Y. Xu, Y. Zhen, K. Wu, H. Ding, L. Wang, X. Tai, X. Cai, X. Zhang, T. Xia, et al., Advanced Materials 35, 2306135 (2023).
- [6]F. Du, J. Zhou, and Q. Li, Advanced Materials Technologies , e71253 (2026).
- [7]T. Takabatake, K. Suekuni, T. Nakayama, and E. Kaneshita, Reviews of Modern Physics 86, 669 (2014).
- [8]Y. Tang, Z. M. Gibbs, L. A. Agapito, G. Li, H.-S. Kim, M. B. Nardelli, S. Curtarolo, and G. J. Snyder, Nature materials 14, 1223 (2015).
- [9]D. Li, X.-L. Shi, J. Zhu, T. Cao, X. Ma, M. Li, Z. Han, Z. Feng, Y. Chen, J. Wang, et al., Nature Communications 15, 4242 (2024).
- [10]L.-D. Zhao, S.-H. Lo, Y. Zhang, H. Sun, G. Tan, C. Uher, C. Wolverton, V. P. Dravid, and M. G. Kanatzidis, nature 508, 373 (2014).
- [11]J. Pei, B. Cai, H.-L. Zhuang, and J.-F. Li, National science review 7, 1856 (2020).
- [12]B. Jia, D. Wu, L. Xie, W. Wang, T. Yu, S. Li, Y. Wang, Y. Xu, B. Jiang, Z. Chen, et al., Science 384, 81 (2024).
- [13]C. Chen, W. Xue, S. Li, Z. Zhang, X. Li, X. Wang, Y. Liu, J. Sui, X. Liu, F. Cao, et al., Proceedings of the National Academy of Sciences 116, 2831 (2019).
- [14]C. Wei, Z. Feng, Y. Yang, G. Zhao, Y. Fu, and D. J. Singh, Physical Review B 108, 235203 (2023).
- [15]M. Zeeshan, C. K. Vishwakarma, and B. Mani, Physical Review Materials 6, 085404 (2022).
- [16]Q. Ren, M. K. Gupta, M. Jin, J. Ding, J. Wu, Z. Chen, S. Lin, O. Fabelo, J. A. Rodríguez-Velamazán, M. Kofu, et al., Nature materials 22, 999 (2023).
- [17]W. Li, S. Lin, M. Weiss, Z. Chen, J. Li, Y. Xu, W. G. Zeier, and Y. Pei, Advanced Energy Materials 8, 1800030 (2018).
- [18]D. Voneshen, K. Refson, E. Borissenko, M. Krisch, A. Bosak, A. Piovano, E. Cemal, M. Enderle, M. Gutmann, M. Hoesch, et al., Nature materials 12, 1028 (2013).
- [19]H. Lin, G. Tan, J.-N. Shen, S. Hao, L.-M. Wu, N. Calta, C. Malliakas, S. Wang, C. Uher, C. Wolverton, et al., Angewandte Chemie International Edition 55, 11431 (2016).
- [20]Z. Wang, Y. Zhao, J. Ni, and Z. Dai, Phys. Rev. Appl. 25, 014061 (2026).
- [21]M. D. Nielsen, V. Ozolins, and J. P. Heremans, Energy & Environmental Science 6, 570 (2013).
- [22]V. Taneja, S. Mishra, U. V. Waghmare, and K. Biswas, ACS Energy Letters 10, 3866 (2025).
- [23]D. Sarkar, T. Ghosh, S. Roychowdhury, R. Arora, S. Sajan, G. Sheet, U. V. Waghmare, and K. Biswas, Journal of the American Chemical Society 142, 12237 (2020).
- [24]Y. Wu, Y. Chen, S. Zeng, G. Li, H. Zhang, L. Zhou, S.-H. Wei, and C. Liu, Nano Letters 26, 1632 (2026).
- [25]P. B. Allen and J. L. Feldman, Phys. Rev. B 48, 12581 (1993).
- [26]M. Simoncelli, N. Marzari, and F. Mauri, Nat. Phys. 15, 809 (2019).
- [27]M. Simoncelli, N. Marzari, and F. Mauri, Phys. Rev. X 12, 041011 (2022).
- [28]A. Rodriguez, R. Rurali, C. Lin, J. Ojha, M. Al-Fahdi, G. J. Snyder, and M. Hu, npj Computational Materials 12, 137 (2026).
- [29]W. Xiong, H. Huang, Y. Wu, X. Xu, G. Li, Z. Gu, and S. Zeng, Advanced Science 12, e06807 (2025).
- [30]K. Klepp and C. Weithaler, Journal of Alloys and Compounds 243, 1 (1996).
- [31]A. J. Rettie, C. D. Malliakas, A. S. Botana, J. M. Hodges, F. Han, R. Huang, D. Y. Chung, and M. G. Kanatzidis, Journal of the American Chemical Society 140, 9193 (2018).
- [32]C. Burschka and W. Bronger, Zeitschrift für anorganische und allgemeine Chemie 430, 61 (1977).
- [33]J. Yue, J. Zheng, J. Li, X. Shen, W. Ren, Y. Liu, and T. Cui, Materials Today Physics 46, 101517 (2024).
- [34]V. Gusain, M. Zeeshan, and B. Mani, Materials Today Physics , 102055 (2026).
- [35]G. Kresse and J. Furthmüller, Comput. Mater. Sci. 6, 15 (1996).
- [36]G. Kresse and J. Furthmüller, Phys. Rev. B 54, 11169 (1996).
- [37]J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996).
- [38]S. Maintz, V. L. Deringer, A. L. Tchougréeff, and B. Dronskowski, J. Comput. Chem. 37, 1030 (2016).
- [39]T. Tadano and S. Tsuneyuki, Phys. Rev. B 92, 054301 (2015).
- [40]T. Tadano, Y. Gohda, and S. Tsuneyuki, J. Phys.: Condens. Matter 26, 225402 (2014).
- [41]F. Zhou, W. Nielson, Y. Xia, and V. Ozoliņš, Phys. Rev. B 100, 184308 (2019).
- [42]F. Zhou, W. Nielson, Y. Xia, and V. Ozoliņš, Phys. Rev. Lett. 113, 185501 (2014).
- [43]L. J. Nelson, G. L. W. Hart, F. Zhou, and V. Ozoliņš, Phys. Rev. B 87, 035125 (2013).
- [44]E. J. Candès and M. B. Wakin, IEEE Signal Process. Mag. 25, 21 (2008).
- [45]Y. Xia, K. Pal, J. V. Ozoliņš, and C. Wolverton, Phys. Rev. Lett. 124, 065901 (2020).
- [46]J. Zheng, C. Lin, C. Lin, G. Hautier, R. Guo, and B. Huang, npj Comput. Mater. 10, 30 (2024).
- [47]T. Tadano and S. Tsuneyuki, Phys. Rev. Lett. 120, 105901 (2018).
- [48]I. Errea, B. Rousseau, and A. Bergara, Phys. Rev. Lett. 106, 165501 (2011).
- [49]T. Tadano and W. A. Saidi, Phys. Rev. B 129, 185901 (2022).
- [50]S. Baroni, S. de Gironcoli, A. Dal Corso, and P. Giannozzi, Rev. Mod. Phys. 73, 515 (2001).
- [51]Z. Guo, Z. Han, D. Feng, G. Lin, and X. Ruan, npj Comput. Mater. 10, 31 (2024).
- [52]W. Li, J. Carrete, N. A. Katcho, and N. Mingo, Comput. Phys. Commun. 185, 1747 (2014).
- [53]Z. Han, X. Yang, W. Li, T. Feng, and X. Ruan, Comput. Phys. Commun. 270, 108179 (2022).
- [54]L. Ji, A. Huang, Y. Huo, Y.-m. Ding, S. Zeng, Y. Wu, and L. Zhou, Phys. Rev. B 109, 214307 (2024).
- [55]A. M. Ganose, J. Park, A. Faghaninia, R. Woods-Robinson, K. A. Persson, and A. Jain, Nature Commun. 12, 2222 (2021).
- [56]P. A. M. Dirac, Proc. R. Soc. A 114, 243 (1927).
- [57]A. Matthiessen and C. Vogt, Ann. Phys. 198, 19 (1864).
- [58]Q.-Y. Xie, P.-F. Liu, J.-J. Ma, L.-M. Wu, K.-W. Zhang, and B.-T. Wang, Physical Review B 108, 014302 (2023).
- [59]C. J. Bartel, Journal of Materials Science 57, 10475 (2022).
- [60]J. E. Saal, S. Kirklin, M. Aykol, B. Meredig, and C. Wolverton, Jom 65, 1501 (2013).
- [61]S. Kirklin, J. E. Saal, B. Meredig, A. Thompson, J. W. Doak, M. Aykol, S. Rühl, and C. Wolverton, npj Computational Materials 1, 15010 (2015).
- [62]See Supplemental Material at [URL] for Convergence tests, tabulated data, convex hull analysis for mono-clinic phase, thermal stability analysis, ICOHP comparison of ison, phonon DoS, scattering phase space, comparison of thermal transport coefficients, thermal conductivity ratios, wave-like and total thermal conductivity, electronic structure and electrical transport for CSAg3S2..
- [63]J. Ma, V. I. Hedge, K. Munira, Y. Xie, S. Keshavarz, D. T. Mildebrath, C. Wolverton, A. W. Ghosh, and W. Butler, Physical Review B 95, 024411 (2017).
- [64]H. Sato and N. Kojima, Phosphorus, Sulfur, and Silicon and the Related Elements 67, 361 (1992).
- [65]J. Yuan, Y. Chen, and B. Liao, Journal of the American Chemical Society 145, 18506 (2023).
- [66]J. Zheng, D. Shi, Y. Yang, C. Lin, H. Huang, R. Guo, and B. Huang, Physical Review B 105, 224303 (2022).
- [67]J. Yue, Y. Liu, and J. Zheng, Physical Review B 111, 024313 (2025).
- [68]S. Tao, X. Ding, K. Yang, W. Liu, N. Li, X. Ding, X. Zhou, M. He, Z. Yang, and X. Yang, Applied Physics Letters 127 (2025).
- [69]J. Hafner and M. Krajci, Journal of Physics: Condensed Matter 5, 2489 (1993).
- [70]H.-J. J. Pang, L.-C. Chen, H. Yu, P.-F. Qiu, G.-H. Zhong, Q. Peng, and X.-J. Chen, Physical Review B 105, 094115 (2022).
- [71]S. Pailhès, H. Euchner, V. M. Giordano, R. Debord, A. Assy, S. Gomès, A. Bosak, D. Machon, S. Paschen, and M. De Boissieu, Physical review letters 113, 025506 (2014).
- [72]F. Xiao, Q.-Y. Xie, R. Yu, H. Li, J. Zhang, and B.-T. Wang, Physical Review B 111, 184304 (2025).
- [73]X. Song, K. Li, D. Liu, and J. Guan, Physical Review B 111, 195209 (2025).
- [74]N. Virakante, R. Kumar, and A. Jain, Phys. Rev. B 114, 065202 (2026).
- [75]L. Xie, J. Feng, R. Li, and J. He, Physical Review Letters 125, 245901 (2020).
- [76]Y. Xia, V. Ozoliņš, and C. Wolverton, Physical review letters 125, 085901 (2020).
- [77]Y. Zhao, S. Zeng, G. Li, C. Lian, Z. Dai, S. Meng, and J. Ni, Physical Review B 104, 224304 (2021).
- [78]J. Yue, J. Zheng, X. Shen, K. Maji, C.-Y. Yang, S. Lin, P. Lemoine, E. Guilmeau, Y. Liu, and T. Cui, Small 21, e06386 (2025).
- [79]R. Pathak, S. Paul, S. Biswas, S. K. Pati, and K. Biswas, Proceedings of the National Academy of Sciences 123, e2521353123 (2026).
- [80]Z. Ju, D. Ma, and Z. Chang, Physical Review B 111, 134302 (2025).
- [81]P. B. Allen, J. L. Feldman, J. Fabian, and F. Wooten, Philosophical Magazine B 79, 1715 (1999).
- [82]Y. Pei, X. Shi, A. LaLonde, H. Wang, L. Chen, and G. J. Snyder, Nature 473, 66 (2011).
