🔬 materials science

Atomic-level study of dose-dependent lattice thermal conductivity degradation in irradiated tungsten

This study combines Non-Equilibrium Molecular Dynamics simulations with Machine Learning to reveal that vacancy density is the dominant factor driving a nonlinear, dose-dependent degradation of tungsten's lattice thermal conductivity, characterized by a step-like transition at a critical vacancy threshold of 0.05 at.%.

Original authors: Mary K Chessey, José Julio Gutiérrez Moreno, Sajjan Singh Mehta, Paolo Settembri, Arda Erbasan, Dominik Freinberger, Marina García Romero, Mervi J Mantsinen

Published 2026-10-20

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Mary K Chessey, José Julio Gutiérrez Moreno, Sajjan Singh Mehta, Paolo Settembri, Arda Erbasan, Dominik Freinberger, Marina García Romero, Mervi J Mantsinen

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

The quest for clean, limitless energy often leads scientists to the heart of a star, where hydrogen atoms fuse together to release immense power. To harness this fusion on Earth, engineers must build reactors capable of withstanding conditions more extreme than the surface of the sun. The walls of these reactors, known as plasma-facing components, are bombarded by a relentless stream of high-energy neutrons. These neutrons act like microscopic bullets, striking the atoms of the wall material and knocking them out of place. Over time, this bombardment creates a chaotic landscape of missing atoms and misplaced clusters within the solid structure. For a reactor to function safely and efficiently, these walls must not only survive the physical impact but also manage the intense heat they absorb. If the material cannot conduct this heat away quickly to cooling systems, it risks melting or failing, ending the reactor's life.

Tungsten, a dense metal with an incredibly high melting point, has emerged as the leading candidate for these critical walls. It is chosen for its ability to resist heat and its low tendency to absorb the radioactive fuel used in fusion. However, a hidden vulnerability exists: as neutrons damage the metal's internal structure, its ability to conduct heat begins to fade. While metals are generally known for moving heat through their free-moving electrons, tungsten is unique because its rigid atomic lattice also plays a major role in transporting thermal energy. When the lattice is disrupted by radiation damage, this lattice-based heat flow is severely hindered. Understanding exactly how this degradation happens, and at what point the material becomes ineffective, is essential for designing reactors that can operate for decades.

In a recent study, researchers set out to map this degradation process with atomic precision. They focused on tungsten exposed to a wide range of radiation doses, simulating the cumulative damage a reactor wall might endure over years of operation. To do this, they used a powerful combination of computer simulations and artificial intelligence. First, they created digital models of tungsten crystals and introduced defects to mimic the damage caused by neutron strikes. They used a specific method to generate these defects that allowed them to reach high levels of damage without needing to simulate every single collision event, which would be too computationally expensive. They then subjected these damaged models to a flow of heat, watching how the temperature changed across the material to measure its thermal conductivity.

The team ran these simulations on two different scales. They built smaller models containing a quarter of a million atoms to test a vast array of damage levels, pushing the simulations to extreme doses. They also constructed much larger models with two million atoms to ensure their results were not skewed by the size of the simulation box itself. By analyzing the temperature gradients in these models, they calculated how well the heat moved through the damaged metal. The results showed a clear, non-linear relationship between the amount of damage and the loss of heat-conducting ability. As the density of defects increased, the thermal conductivity dropped sharply, but not in a simple, straight-line fashion.

To make sense of the complex data, the researchers employed machine learning, a form of artificial intelligence that can find patterns in vast datasets that humans might miss. They trained these models to predict the thermal conductivity based on the specific types and numbers of defects present, such as empty spots where atoms were missing (vacancies) and clusters of atoms squeezed into spaces they shouldn't occupy. The analysis revealed that the concentration of vacancies was the single most important factor controlling how much heat the material could carry. While other defects played a role, the empty spaces in the atomic grid acted as the primary roadblocks for the heat-carrying vibrations moving through the metal.

Perhaps the most striking discovery was the existence of a critical threshold. The simulations showed that as the concentration of vacancies rose, the thermal conductivity plummeted rapidly until it reached a specific point: a vacancy concentration of 0.05 percent. Beyond this point, the material entered a state of saturation. Even if the researchers added more defects, the thermal conductivity did not drop much further; it hit a floor where the material retained about 70 percent of its original heat-conducting ability. This suggests that once the atomic lattice is damaged enough, the pathways for heat flow are so thoroughly disrupted that adding more damage has a diminishing return. The material has effectively reached its limit of degradation.

This finding provides a crucial piece of the puzzle for fusion reactor design. It indicates that there is a predictable limit to how much radiation damage tungsten can sustain before its thermal performance is permanently compromised. The study confirms that while the material does not fail completely, its ability to dissipate heat is significantly reduced and then stabilizes at a lower level. By identifying the specific defect density that triggers this saturation, engineers can better estimate the lifespan of reactor components and plan for maintenance or replacement before the material becomes a safety risk. The research demonstrates that by combining detailed atomic simulations with advanced data analysis, scientists can uncover the hidden rules governing how materials behave under the extreme conditions of future fusion power plants.

Technical Summary: Atomic-level study of dose-dependent lattice thermal conductivity degradation in irradiated tungsten

Problem Statement
Plasma-facing components (PFCs) in future fusion reactors, particularly those utilizing tungsten (W), must withstand extreme environments. While tungsten is selected for its high melting temperature and low sputtering yield, neutron irradiation introduces complex nanostructural defects that degrade critical properties, specifically thermal conductivity ( $\kappa$ ). Although metallic thermal transport is predominantly electronic, tungsten exhibits an anomalously high lattice (phononic) contribution to $\kappa$ (estimated at 10–30% of the total). This lattice component is highly sensitive to defect-induced scattering. Understanding the degradation pathways of lattice $\kappa$ is essential, as it serves as a proxy for the broader degradation of the material system. However, quantifying the relationship between specific defect structures (vacancies, self-interstitial atoms, and dislocations) and thermal transport across a wide range of irradiation doses remains a challenge, particularly given the computational cost of simulating large-scale defect evolution.

Methodology
The authors employed an integrated computational framework combining three primary components:

Defect Generation: The Creation Relaxation Algorithm (CRA) was used to generate high-dose defect structures in body-centered cubic (BCC) tungsten. This method involves the iterative insertion of Frenkel pairs (vacancies and self-interstitial atoms, SIAs) followed by stress-driven relaxation under isotropic zero pressure. This approach approximates realistic high-dose damage structures while reducing computational costs compared to full collision cascade simulations.
Thermal Transport Simulation: Non-Equilibrium Molecular Dynamics (NEMD) simulations were performed using the LAMMPS code with an Embedded Atom Model (EAM) potential optimized for vacancy clusters. The "direct" method was applied, establishing a temperature gradient across a target region via heat source and sink regions. Simulations were conducted on two system scales:
- Small systems: 250k atoms in the target region (total 2.5M atoms), tested up to 1.8 canonical displacements per atom (cdpa).
- Large systems: 2M atoms in the target region (total 40M atoms), tested up to 0.4 cdpa.
  Local lattice thermal conductivity ( $\kappa$ ) was calculated using Fourier's law based on spatially resolved temperature gradients.
Machine Learning (ML) Analysis: To establish a quantitative relationship between defect features and thermal conductivity, the authors applied multiple ML regression models (Random Forest, Gradient Boosting, XGBoost, LightGBM). Input features included applied dose (cdpa), temperature ( $T$ ), and local concentrations of vacancies ( $N_{Vac}$ ) and SIAs ( $N_{SIA}$ ). The models were trained and validated using 5-fold cross-validation. SHAP (SHapley Additive exPlanations) analysis was utilized to interpret feature importance and non-linear interactions.

Key Results

Defect Evolution and Saturation: Defect concentrations exhibited non-linear behavior with increasing dose. In both system sizes, defect concentrations increased to a maximum and then decreased or leveled off due to self-healing mechanisms (recombination of mobile SIAs with vacancies) at higher doses. Large SIA clusters formed, while vacancies remained largely isolated or located at the edges of these clusters.
Thermal Conductivity Degradation: The study confirmed a significant, non-linear suppression of lattice $\kappa$ with increasing defect density. The pristine large system yielded a bulk thermal conductivity of approximately 22.90 W/m·K.
Machine Learning Performance: ML models trained on the large system data demonstrated superior robustness ( $R^2 > 0.94$ ) compared to those trained on small systems. XGBoost was identified as the most generalizable model.
Dominant Factors: SHAP analysis identified vacancy concentration ( $N_{Vac}$ ) as the dominant structural factor controlling thermal transport, followed by the applied dose (cdpa). While SIAs also contributed to scattering, their influence was secondary to vacancies in the predictive model.
Critical Thresholds and Saturation: The ML-driven mapping revealed a step-like degradation transition at a critical vacancy concentration of 0.05 at.%. Below this threshold, $\kappa$ drops rapidly. Beyond this point, the system enters a saturation regime where further defect accumulation yields diminishing returns on conductivity loss. The maximum relative lattice thermal conductivity degradation was capped at approximately 30%, meaning the material retains roughly 70% of its pristine conductivity.

Significance and Claims
The paper claims to provide a powerful, predictive framework for quantifying the complex, non-linear relationship between irradiation-induced defect structures and lattice thermal conductivity in tungsten. By integrating CRA, large-scale NEMD, and explainable ML, the authors successfully resolved a step-like degradation transition and identified a saturation floor for thermal conductivity loss.

The authors emphasize that while the study focuses on the lattice component, the established structure-property relationships serve as a critical foundation for understanding overall material degradation in fusion environments. The work offers explicit, dose-dependent scaling maps that move beyond qualitative descriptions, potentially serving as input for multi-scale modeling frameworks. The study acknowledges that while it omits direct electronic transport modeling, the high-fidelity structural relationships established here are vital for future efforts to couple lattice-driven models with electronic transport to develop digital twins of fusion reactor components. The findings are presented as a step toward optimizing the performance and lifetime of plasma-facing components in next-generation fusion reactors.

More like this