First-principles transition-state tensorial cluster expansion of vacancy diffusion in Ta-W beyond the kinetically-resolved activation approximation

arXiv:2605.23612 · cond-mat.mtrl-sci · Submitted 2026-05-22 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.

Kai: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "First-principles transition-state tensorial cluster expansion of vacancy diffusion in Ta-W beyond the kinetically-resolved activation approximation".

Mira: Here's the gist of the paper. Predicting diffusion in chemically complex alloys remains challenging due to the strong dependence of migration barriers on local atomic environments The framework captures nontrivial composition-dependent diffusion behavior arising from a crossover between solute trapping and percolated low-barrier transport pathways, yielding a maximum in the apparent activation energy near intermediate compositions This approach establishes a general and scalable route for integrating first-principles transition-state energetics into mesoscale kinetic simulations, enabling predictive multiscale modeling of diffusion in chemically complex materials and providing a pathway for uncovering emergent transport phenomena.

Kai: Modeling the Energy Landscape The researchers developed a tensorial cluster expansion (TCE) technique to represent the energy as a linear function of cluster counts, including transition states This is expressed as E(X) = 1/2!

Mira: ε(n) αβ N(n) αβ (X) + 1/3!

Kai: ζ(n) αβγ M(n) αβγ(X), which can be cast as a simple linear problem E(X) = jT t(X).

Title and authors: Kai: We started by looking at the title of this paper, "First-principles transition-state tensorial cluster expansion of vacancy diffusion in Ta-W beyond the kinetically-resolved activation approximation," and seeing who wrote it. It sounds like a lot, but essentially they are trying to get a more accurate prediction for how atoms move in tantalum tungsten alloys.

Mira: And the authors are Jacob Jeffries, Brianna Sebastian Olazabal, and Enrique Martinez. They seem to be coming from different areas—materials science and mechanical/automotive engineering—which is interesting because this work bridges the gap between fundamental quantum calculations and applied kinetic modeling >

Lev: I’m curious about how they approach the complexity of composition dependence here. When you have a system that's chemically complex, the barriers should change based on what’s around it, and this paper seems to be tackling that dependency head-on >

Kai: Yeah, they are using a tensorial cluster expansion technique derived from density functional theory and nudged elastic band calculations to represent the migration barriers. It’s not just a simple calculation; it includes transition states in the expansion >

Mira: They are trying to move beyond simpler models, like the KRA relation, because that relation makes assumptions about correlations between energy differences that they suspect aren't accurate for vacancy hopping in Ta-W alloys >

Lev: That’s the key assumption they are challenging; if those correlations don't hold up, then relying on a model built on them isn't very reliable when you need to predict real transport behavior >

Kai: They propose modeling the energy barrier using feature vector differences instead of just absolute energies, which they show is a generalized broken bond model that lets them predict differences directly >

Mira: That approach seems promising because it allows for a more direct prediction of the energy barriers based on what's locally present rather than relying on an approximation about the pathway itself >

The paper's summary: Kai: So, what they summarized is that this framework successfully captures the nontrivial composition-dependent diffusion behavior in Ta-W systems by modeling a crossover between solute trapping and those low-barrier, percolated transport pathways >

Mira: They found that this modeling yields a maximum in the apparent activation energy right near intermediate compositions, which tells us exactly where we should expect to see this transition happening >

Lev: That peak in the activation energy is significant because it’s not just some minor fluctuation; it's a feature that shows how the system switches from one transport mode to another >

Kai: They also detailed four distinct regimes over composition and temperature when they looked at their kinetic Monte Carlo simulations, highlighting that dilute tantalum inhibits vacancy diffusion up to a percolation threshold >

Mira: And they found that dilute tungsten leaves vacancy diffusion largely unaffected, which is interesting because it suggests the effect of one solute can be much more localized than you might expect >

Lev: That difference between how Ta and W affect diffusion is something we need to consider when designing materials for applications where we need specific transport control >

Kai: Overall, they are showing that this AI approach establishes a general and scalable route for integrating those first-principles transition state energetics into mesoscale kinetic simulations >

Mira: That integration is what makes this useful because it creates a pathway for predictive multiscale modeling of diffusion in complex materials, which opens up the door to uncovering emergent transport phenomena >

The paper's improvements: Lev: Now let’s talk about how they improved upon the previous work. They are focusing on replacing that KRA approximation with a more direct way to calculate energy barriers using those feature vector differences >

Kai: They show that they can model the energy barrier using E‡ minus E◦ equals j transpose t‡ minus t◦, which is this generalized broken bond model, which lets them predict the barrier difference directly rather than needing absolute energies from first principles >

Mira: That’s a major step because it means you don't have to perfectly calculate every single state energy from scratch if you can just predict the differences between them using those learned parameters j >

Lev: From an engineering standpoint, that means the computational overhead for running KMC simulations is reduced because we aren't needing to solve a massive set of absolute energies for every possible transition state >

Kai: They also trained their model by fitting those interaction parameters using ridge regression on feature vector differences p one and p two, which constrains the fitting process so that non-physical interactions end up with zero fitted interaction energy >

Mira: That constraint is crucial because it keeps the learned parameters physically sensible, ensuring the AI isn't learning random noise from the data >

Lev: So they are essentially taking complex quantum calculations and turning them into a set of simple, learnable parameters that describe the local chemistry effectively >

Conclusion: Kai: So to wrap up on this paper, this work on "First-principles transition-state tensorial cluster expansion of vacancy diffusion in Ta-W beyond the kinetically-resolved activation approximation" shows how to use AI to create a way to predict composition-dependent diffusion behavior >

Mira: They found that the framework successfully captures the crossover between solute trapping and percolated transport pathways, which helps us understand where that maximum in activation energy is located in the composition space >

Lev: The real impact for materials design is knowing exactly when to use dilute tantalum to minimize void growth versus when you need to increase it above ninety-eight percent to start building those fast diffusion networks >

Kai: This approach gives us a scalable route to integrate first-principles energetics into kinetic simulations, which means we can predict how things move in chemically complex materials and find those emergent transport phenomena >

Mira: It’s a method that allows us to tune the transport by controlling percolation pathways based on composition, which is really useful for materials design applications >

Lev: To finish up, this paper on "First-principles transition-state tensorial cluster expansion of vacancy diffusion in Ta-W beyond the kinetically-resolved activation approximation" shows how to use AI to create a way to predict composition-dependent diffusion behavior >

Kai: It’s a method that allows us to tune the transport by controlling percolation pathways based on composition, and that’s a pretty concrete finding for anyone working in this space >

Mira: And we've got some clear guidelines on when to use specific solute concentrations for optimal outcomes in these kinds of systems >

Lev: That’s the summary of what this paper gives us about using first-principles data to guide kinetic simulations, and it’s a solid foundation for future work >

Jacob Jeffries, Brianna Sebastian-Olazabal, Enrique Martinez

Department of Materials Science and Engineering, Clemson University · School of Mechanical and Automotive Engineering, Clemson University

cond-mat.mtrl-sci

Submitted: 2026-05-22

Updated: 2026-06-05

Comments: 19 pages, 14 figures

DOI: 10.1016/j.commatsci.2026.115142

Code: https://github.com/MUEXLY/TaW-tce-kmc

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 81/100

The gist: Predicting diffusion in chemically complex alloys remains challenging due to the strong dependence of migration barriers on local atomic environments The framework captures nontrivial…

Key concepts

Tensorial Cluster Expansion (TCE)
A mathematical technique used to represent the total energy of a system as a linear function of cluster counts, including transition states. It simplifies complex energy calculations into a solvable linear problem, making it easier to predict energy differences between different atomic configurations.
Feature Vector Differences
Instead of relying on approximations like the KRA relation for barrier prediction, this method uses the difference between feature vectors (like ∆ft and ∆fE) to directly calculate energy differences (E‡ − E◦). This provides a more accurate way to model migration barriers based on local atomic environments.
Percolated Low-Barrier Transport
This refers to the formation of connected pathways within the alloy where diffusion barriers are low. When these pathways form a network, they allow for long-range transport, which is crucial for high diffusion rates in Ta-W alloys at specific compositions.
Composition-Dependent Diffusion Regimes
The study maps how vacancy movement changes based on the amount of Ta in the alloy. It identifies distinct behaviors, such as dilute Ta inhibiting diffusion up to a certain point, and finding critical concentrations where fast transport networks begin to form.

Terminology

Summary

Predicting diffusion in chemically complex alloys remains challenging due to the strong dependence of migration barriers on local atomic environments The framework captures nontrivial composition-dependent diffusion behavior arising from a crossover between solute trapping and percolated low-barrier transport pathways, yielding a maximum in the apparent activation energy near intermediate compositions This approach establishes a general and scalable route for integrating first-principles transition-state energetics into mesoscale kinetic simulations, enabling predictive multiscale modeling of diffusion in chemically complex materials and providing a pathway for uncovering emergent transport phenomena. p. 0

Where the paper says it

Applied to the Ta-W system, the framework captures nontrivial composition-dependent diffusion behavior arising from a crossover between solute trapping and percolated low-barrier transport pathways, yielding a maximum in the apparent activation energy near intermediate compositions. This approach establishes a general and scalable route for integrating first-principles transition-state energetics into mesoscale kinetic simulations, enabling predictive multiscale modeling of diffusion in chemically complex materials and providing a pathway for uncovering emergent transport phenomena.

Modeling the Energy Landscape

The researchers developed a tensorial cluster expansion (TCE) technique to represent the energy as a linear function of cluster counts, including transition states This is expressed as E(X) = 1/2! ε(n) αβ N(n) αβ (X) + 1/3! p. 1

Where the paper says it

2 Model First, following a previously developed tensorial cluster expansion model[31], we write our energy as a linear function of the number of n’th order α-β pairs and the number of n’th order α-β-γ three body clusters: E(X) = 1 2! ε (n) αβ N (n) αβ (X) + 1 3!

ζ(n) αβγ M(n) αβγ(X), which can be cast as a simple linear problem E(X) = jT t(X). p. 1
Where the paper says it

With respect to the lower indices, all of the above tensors are symmetric, necessitating the overcounting factors 1/2! This can be more easily cast as a simple linear problem: E(X) = j ⊺ t(X) (3) where X describes the occupation of each lattice site on a static lattice: Xiα = ( 1 site i is occupied by atom type α 0 else (4) which allows us to calculate the counts of each cluster type as polynomials in X: N (n) αβ (X) = A (n) ij XiαXjβ M (n) αβγ(X) = B (n) ijk XiαXjβXkγ (5) 2

The energy barrier is modeled using the feature vector differences to predict energy differences directly, rather than relying on the KRA relation. p. 2
Where the paper says it

vec(ζ) ⊺ ⊺ (8) For a given reaction pathway X◦ → X‡ → X′ , respectively denoting the initial, transition, and final states, the transition rate can be approximated by harmonic transition state theory (HTST)[32]: r = ν exp − E‡ − E◦ kBT (9) where E‡ and E◦ are respectively the energies of the transition and initial states, and ν is an attempt frequency. Rather than predict absolute energies, we can instead use the feature vector differences to predict energy differences directly, similarly to Musa et al.

Specifically, the energy barrier is modeled using E‡ − E◦ = jT t‡ − t◦. p. 8
Where the paper says it

Specifically, only including sites that are local to the event, i. sites in D, offers a roughly quadratic speedup in energy barriers with respect to system size (Fig.

Data Generation and Model Fitting

The training set was generated by equilibrating structures of varying compositions between 20% and 80% Ta with a step of 5%, followed by random vacancy insertion and nearest-neighbor hops. The model is trained using ridge regression to fit the interaction parameters j based on feature vector differences p1 = (∆fti, ∆fEi) and p2 = (∆bti, ∆bEi). p. 7

Where the paper says it

Lf(j) = X i (∆ft ⊺ i j − ∆fEi) 2 Lb(j) = X i (∆bt ⊺ i j − ∆bEi) 2 L(j) = Lf(j) + Lb(j) + λ∥j∥ 2 2 = ∥∆Tj − ∆e∥ 2 2 + λ∥j∥ 2 2 jopt = arg min j∈Rk L(j) = (∆T ⊺∆T + λI) −1 ∆T ⊺∆e (19) where Lf is the loss for the forward reactions, Lb is the loss for the backwards reactions, L is the total loss, λ > 0 is the regularization strength, ∆T is the design matrix containing the feature vector differences ∆fti and ∆bti defined above, ∆e is a target vector containing the energy barriers ∆fEi and ∆bEi also defined above, j is the learnable matrix containing cluster interaction energies from Eq. (8), k = 8·3 2+6·3 3 = 234 is the feature size, i.

The fitting routine respects physical constraints, such as non-physical interactions having a resulting fitted interaction energy of 0. p. 7
Where the paper says it

We find that our fitted coefficients respect important physical constraints. Firstly, non-physical interactions have a resulting fitted interaction energy of 0, e.

Key Findings on Diffusion Regimes

The KMC simulations reveal four categorically distinct regimes over composition, notably that dilute Ta inhibits vacancy diffusion up to a percolation threshold, and that dilute W leaves vacancy diffusion largely unaffected. The paper finds two main peaks emerging as a function of composition and temperature in the distribution of traversed barriers. p. 9

Where the paper says it

Namely, there seems to be a boundary in composition-temperature space in which the vacancy traverses a bimodal distribution of energy barriers, rather than a unimodal one as one might expect for a system like Ta-W, where the elements present in the solution have a large degree of chemical similarity (Fig. From the distribution of traversed barriers, we see that there are two main peaks emerging as a function of composition and temperature.

These peaks suggest that for Ta-rich environments, the lower-barrier peak becomes more prominent over the high-barrier peak. p. 9
Where the paper says it

From the distribution of traversed barriers, we see that there are two main peaks emerging as a function of composition and temperature. For Ta-rich environments, the lower-barrier peak becomes more prominent over the high-barrier peak.

Implications for Materials Design

The framework enables the combination of first-principles transition-state energetics with datadriven cluster expansions beyond the KRA approximation in a way that is generalizeable to a wide class of diffusive mechanisms. This provides a natural route to tuning transport via percolation, allowing alloy compositions to be designed to either promote or suppress the formation of connected networks of fast diffusion channels. p. 0

Where the paper says it

Applied to the Ta-W system, the framework captures nontrivial composition-dependent diffusion behavior arising from a crossover between solute trapping and percolated low-barrier transport pathways, yielding a maximum in the apparent activation energy near intermediate compositions. This approach establishes a general and scalable route for integrating first-principles transition-state energetics into mesoscale kinetic simulations, enabling predictive multiscale modeling of diffusion in chemically complex materials and providing a pathway for uncovering emergent transport phenomena.

The results suggest that dilute Ta is optimal for minimizing void growth, while Ta concentration larger than xTa ≈ 9.8% is needed to enhance diffusion by forming percolated networks. p. 0
Where the paper says it

Migration barriers computed using density functional theory and nudged elastic band calculations are represented via a tensorial cluster expansion including transition states and deployed in on-lattice kinetic Monte Carlo simulations. Applied to the Ta-W system, the framework captures nontrivial composition-dependent diffusion behavior arising from a crossover between solute trapping and percolated low-barrier transport pathways, yielding a maximum in the apparent activation energy near intermediate compositions.

Limitations and Future Directions

The migration rates are evaluated within the framework of HTST, which is formally valid in the low-temperature limit, but entropic effects can become nonnegligible at elevated temperatures. The study also notes that the KMC simulations are performed on finite supercells (bcc 5 × 5 × 5), which may introduce finite-size effects. Future work could involve incorporating vibrational contributions to extend the formalism beyond the HTST approximation. p. 6

Where the paper says it

As such, we save this testing for a future work. Then, rather than rely on the KRA approximation to compute energy barriers, which will quite poorly approximate energy barriers within our dataset, we will instead compute feature vectors for the initial, transition, and final states within our aforementioned multi-lattice model.

The framework is not inherently restricted to the single-defect limit, and augmenting the training set with multiple vacancies could enable the study of vacancy clustering kinetics.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Automated migration barrier prediction using feature vector differences This system can predict vacancy hop energy barriers directly from local features rather than relying on potentially inaccurate KRA approximations, as shown by Eq 10: E‡ − E◦ = j⊺t‡ − t◦, which is a generalized broken bond model.

  2. Composition-dependent diffusion regime mapping This AI can predict the four categorically distinct regimes of vacancy diffusion over composition and temperature, such as Dilute Ta (0% ≤ xTa ≤ 5%), in which Ta acts as a dragging solute, based on KMC simulations.

  3. Solute drag and percolation pathway engineering This system can identify compositions that inhibit or enhance vacancy motion by modeling the percolation of low-barrier pathways, suggesting that for applications where inhibiting diffusion is desired, dilute Ta is optimal.

  4. Identification of critical concentration thresholds This AI can predict the critical concentration xTa ≈ 9.8% above which Ta forms a percolated network, directly linking this to the observed sharp drop in activation energy between 5% and 11% Ta and supporting long-range diffusion.

  5. Multi-scale kinetic simulation of vacancy transport This improved system can perform KMC simulations over large temperature and composition ranges to calculate diffusivity D(x, T) using the Arrhenius relation D(x, T) = D∞(x) exp − Ea(x)/kBT, providing composition-dependent diffusion coefficients.

Abstract

Predicting diffusion in chemically complex alloys remains challenging due to the strong dependence of migration barriers on local atomic environments. Migration barriers computed using density functional theory and nudged elastic band calculations are represented via a tensorial cluster expansion including transition states and deployed in on-lattice kinetic Monte Carlo simulations. Applied to the Ta-W system, the framework captures nontrivial composition-dependent diffusion behavior arising from a crossover between solute trapping and percolated low-barrier transport pathways, yielding a maximum in the apparent activation energy near intermediate compositions. This approach establishes a general and scalable route for integrating first-principles transition-state energetics into mesoscale kinetic simulations, enabling predictive multiscale modeling of diffusion in chemically complex materials and providing a pathway for uncovering emergent transport phenomena.

Related papers