🔬 materials science

A DFT and Machine Learning-Assisted Study on the Lattice Thermal Conductivity of LiCdSb for Thermoelectric Applications

This study employs density functional theory and machine-learning interatomic potentials to investigate the thermoelectric properties of LiCdSb, revealing a low lattice thermal conductivity and a figure of merit (ZT) exceeding 1 at temperatures above 600 K, which positions it as a promising candidate for high-temperature energy conversion.

Original authors: R. Zosiamliana, Lalhriat Zuala, N. T. Tien, Vo Khuong Dien, A. Laref, D. P. Rai

Published 2026-07-28

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: R. Zosiamliana, Lalhriat Zuala, N. T. Tien, Vo Khuong Dien, A. Laref, D. P. Rai

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the world as a giant, humming engine. Every time we drive a car, charge a phone, or even just boil water, we are burning fuel and creating a massive amount of waste heat. This heat usually just floats away into the air, a silent, invisible loss of energy that we can't get back. For decades, scientists have been on a treasure hunt for a special kind of material that acts like a magical heat-to-electricity converter. These materials, called thermoelectrics, can take that wasted warmth and turn it directly into electricity without any moving parts, smoke, or noise. Think of them as the "heat catchers" of the material world.

To make these heat catchers work well, they need to be very good at two things at once: letting electricity flow through them easily (like a superhighway for electrons) while blocking heat from flowing through them (like a thermos keeping your coffee hot). The challenge is that most materials are good at one but bad at the other. If heat flows too fast, the temperature difference disappears, and the electricity stops. So, the holy grail of this field is finding a material that is a "thermal insulator" but an "electrical conductor." Scientists use powerful computer simulations to predict which materials might have these superpowers before they even build them in a lab, saving time and money.

In this study, a team of researchers decided to investigate a specific material called Lithium Cadmium Antimonide (LiCdSb). They wanted to see if this material could be a champion for capturing waste heat, especially at high temperatures. To do this, they didn't just use one tool; they combined two different approaches. First, they used a very precise, heavy-duty computer method called Density Functional Theory (DFT) to map out how electrons move inside the material. It's like using a high-resolution microscope to see the tiny roads electrons travel on. However, calculating how heat moves through the crystal lattice (the atomic grid) is incredibly difficult and slow, like trying to predict the path of every single person in a crowded stadium at once.

To solve this tricky heat problem, the researchers tried a clever shortcut: they used Machine Learning (ML). Imagine training a smart robot to learn the rules of the game by watching a few examples, so it can predict the rest instantly without having to calculate every single step. The team trained this AI on the material's atomic behavior and then let it predict how well the material blocks heat. They compared this "AI guess" against a traditional, more complex calculation method to see which was better.

The results were quite promising. The team found that LiCdSb is indeed a semiconductor, meaning it can conduct electricity, and it has a very low ability to conduct heat. Specifically, at room temperature, the material's lattice thermal conductivity was calculated to be just 0.24 Wm⁻¹K⁻¹. This is a very low number, which is exactly what you want for a thermoelectric material because it means the heat gets stuck in place, ready to be converted into power. When they combined this low heat conductivity with their electronic calculations, they estimated a "figure of merit" (a score that tells you how good the material is) called ZT. At room temperature, the score was around 0.17 to 0.18, which is decent, but the real magic happened at higher temperatures.

As the temperature rose above 600 Kelvin, the ZT score for LiCdSb climbed well above 1. In the world of thermoelectrics, a score above 1 is a major milestone, suggesting the material could be efficient enough for real-world use. The researchers noted that their machine learning predictions matched up very well with existing experimental data, even better than the traditional calculation methods. This suggests that LiCdSb is a strong candidate for high-temperature energy conversion, potentially helping us harvest waste heat from factories or car engines more efficiently. While the paper confirms these results through simulation and comparison with existing data, it highlights that the machine learning approach is a fast and reliable way to find these hidden gems in the future, offering a new path to solving the global energy puzzle.

Technical Summary: A DFT and Machine Learning-Assisted Study on the Lattice Thermal Conductivity of LiCdSb for Thermoelectric Applications

Problem Statement
Thermoelectric (TE) materials offer a sustainable solution for converting waste heat into electricity, addressing global energy demands and reducing reliance on fossil fuels. However, the efficiency of these materials, quantified by the dimensionless figure of merit ( $ZT = S^2\sigma T / (\kappa_e + \kappa_l)$ ), is often limited by high lattice thermal conductivity ( $\kappa_l$ ). Accurately predicting $\kappa_l$ is a significant computational bottleneck; traditional first-principles methods based on Density Functional Theory (DFT) and the Boltzmann transport equation require calculating high-order anharmonic interatomic force constants (IFCs) via massive supercell calculations, which is computationally prohibitive for high-throughput screening. Furthermore, standard DFT functionals (like GGA) often underestimate band gaps, leading to inaccurate electronic transport predictions. This study addresses the dual challenge of obtaining accurate electronic structures and efficient, reliable $\kappa_l$ predictions for the Half-Heusler (HH) compound LiCdSb (LCS).

Methodology
The authors employed a multi-faceted computational approach combining standard DFT, hybrid functionals, and Machine Learning Interatomic Potentials (MLIPs):

Electronic Structure Calculations:
- DFT Framework: Calculations were performed using the Vienna Ab initio Simulation Package (VASP) with Projector Augmented Wave (PAW) potentials.
- Functionals: To address the band gap underestimation inherent in Generalized Gradient Approximation (GGA-PBE), the study utilized the hybrid functional HSE06. Spin-orbit coupling (SOC) was also included to assess its impact on band splitting.
- Transport Properties: Electronic transport coefficients (Seebeck coefficient $S$ , electrical conductivity $\sigma$ , electronic thermal conductivity $\kappa_e$ ) were calculated using the semi-classical Boltzmann transport theory via the Boltztrap2 code. Carrier relaxation times ( $\tau$ ) were estimated using Deformation Potential (DP) theory.
Lattice Thermal Conductivity ( $\kappa_l$ ) Prediction:
- Traditional Approach: $\kappa_l$ was estimated using the standard Slack model and a temperature-dependent extension (Slack+TDEC) incorporating elastic properties derived from Quasi-Static Approximation (QSA).
- Machine Learning Approach: To bypass the computational cost of direct anharmonic phonon calculations, the authors trained an on-the-fly Machine Learning Potential (FMLP) using data from ab initio molecular dynamics (AIMD) simulations at 300 K.
- MLIP Workflow: The trained FMLP was used to calculate second- and third-order IFCs via Phonopy and Phono3py codes. $\kappa_l$ was then computed by solving the phonon Boltzmann transport equation with a $30 \times 30 \times 30$ q-point mesh.
Structural and Mechanical Analysis:
- Structural stability was assessed via Birch-Murnaghan equation of state fitting and phonon dispersion analysis (checking for imaginary frequencies).
- Temperature-dependent elastic properties (TDEC) were calculated using both QSA and Quasi-Harmonic Approximation (QHA) to determine the most reliable model for input into the Slack equation.

Key Results

Structural and Electronic Properties:
- LiCdSb crystallizes in the cubic C1b structure (space group $F\bar{4}3m$ ). The $\beta$ -phase was identified as the ground state, consistent with previous theoretical studies.
- The compound is a non-magnetic semiconductor with a total valence electron count (VEC) of 18, satisfying the Slater-Pauling rule for semiconducting behavior ( $M_t = Z_t - 18 = 0$ ).
- Band Gap: The HSE06 functional yielded a significantly larger and more accurate band gap compared to GGA and GGA+SOC. The band structure indicates a direct gap at the $\Gamma$ point.
- Carrier Effective Masses: The conduction band exhibits high dispersion, resulting in a low electron effective mass ( $|m^*_e| \approx 0.11 m_0$ ), suggesting high electron mobility. The valence band shows anisotropic dispersion with coexisting light and heavy holes.
Mechanical and Thermodynamic Stability:
- The compound satisfies the Born-Huang stability criteria.
- Elastic constants ( $C_{ij}$ ) calculated via QSA showed a smooth, monotonic softening with temperature, maintaining mechanical stability up to 900 K. In contrast, QHA results predicted anomalous behavior (negative $C_{44}$ at ~600 K), leading the authors to rely on QSA-derived TDEC for further analysis.
- The material exhibits a moderate Debye temperature ( $\Theta_D \approx 228$ K) and a melting temperature ( $T_m$ ) of $\sim 881$ K, indicating suitability for moderate-temperature TE applications.
Lattice Thermal Conductivity ( $\kappa_l$ ):
- All methods predicted ultralow $\kappa_l$ values. The MLIP method yielded $\kappa_l = 0.24$ W m $^{-1}$ K $^{-1}$ at room temperature.
- The ML-predicted $\kappa_l$ showed qualitative agreement with the Slack+TDEC model and was closer to experimental trends than the standard Slack model.
Thermoelectric Figure of Merit ($ZT$):
- The study evaluated $ZT$ for both n-type and p-type doping.
- Comparison with Experiment: The $ZT$ values calculated using the ML-based $\kappa_l$ combined with HSE06 electronic properties showed the best agreement with available experimental data ( $ZT_{exp}$ ). Specifically, at 300 K and 500 K, the ML+HSE06 model predicted $ZT \approx 0.17$ and $0.37$, respectively, closely matching experimental values of $0.10$ and $0.32$.
- High-Temperature Performance: The study reports that the $ZT$ value exceeds the benchmark of 1.0 at temperatures above 600 K, suggesting potential for high-temperature energy conversion.

Significance and Claims
The paper claims that the integration of Machine Learning Interatomic Potentials (MLIPs) with hybrid DFT functionals (HSE06) provides a computationally efficient yet highly accurate framework for predicting thermoelectric performance.

Methodological Validation: The study demonstrates that MLIPs can reliably reproduce DFT-level accuracy for lattice dynamical properties (forces and energies) at a fraction of the computational cost, effectively solving the "mammoth task" of calculating anharmonic IFCs for $\kappa_l$ .
Material Potential: LiCdSb is identified as a promising thermoelectric material due to its intrinsically low lattice thermal conductivity and favorable electronic transport properties, particularly when modeled with accurate band structures.
Predictive Accuracy: The authors assert that the combined approach (HSE06 + MLIP) yields $ZT$ predictions that align well with experimental reports, outperforming standard Slack models which tend to overestimate performance due to simplified phonon scattering treatments.

The work concludes that accurate thermoelectric design requires the simultaneous optimization of electronic structure description and lattice thermal transport modeling, with ML-assisted methods offering a viable path forward for screening complex materials.

More like this