WorksPhysics and Astronomy
High-Temperature Effects for Transition State Calculations in Solids
Abstract
Transition state calculation is a critical technique to understand and predict versatile dynamical phenomena in solids. However, the transition state results obtained at zero Kelvin are often utilized for prediction or interpretation of dynamical processes at high temperatures, and the error bars of such approximation are largely unknown. In this benchmark study, all the major temperature effects including lattice expansion, lattice vibration, electron excitation, and band-edge shift are evaluated with first-principles calculations for defects diffusion in solids. With inclusion of these temperature effects, the notable discrepancies between theoretical predictions at zero Kelvin and the experimental diffusivities at high temperatures are dramatically reduced. Particularly, we find that the lattice expansion and lattice vibration are dominant factors lowering the defect formation energies and hopping barriers at high temperatures, but the electron excitation exhibits minor effects. In sharp contrast to typical assumption, the attempt frequency with lattice expansion and vibration varies significantly with materials: several THz for aluminum bulk but surprisingly over 500 THz for 4H-SiC. For defects in semiconductors, the band-edge shift is also significant at high temperatures and plays a vital role in the defect diffusion. We expect this study would help accurately predict the dynamical processes at high temperatures.
Introduction
Transition state calculation of dynamical processes at high temperatures are essential to a range of fields, such as turbine engines, metallurgical processing, oxide film growth, and solid oxide fuel cells. As summarized in Figure 1, high temperatures can induce lattice expansion, lattice vibration, electron excitation, and band-edge shift. These effects are coupled and impact several aspects of a dynamical process, including hopping distance, defect concentration, attempt frequency, and activation barrier. Because of the complexity, most previous studies utilized transition state results at zero Kelvin to predict the processes occurring at high temperatures.[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]

Figure 1. Four types of temperature effects, including lattice expansion, lattice vibration, electron excitation, and band-edge shift, in solids.
Although limited efforts have made to include certain temperature effects in transition state calculations,[12, 13, 14, 15, 16] none of them included all the temperature effects and the differences between metals and semiconductors remain largely unknown. For instance, only the contribution of lattice vibration was considered for the hopping barrier of metallic solute diffusion in silicon.[17] The contributions of lattice vibration to defect concentration and hopping barrier were considered for vacancy intermediated self-diffusion in lithium, but the lattice constants and attempt frequency were fixed to values at zero Kelvin.[18] Neugebauer et al. carefully examined the anharmonicity and vacancies near the melting point of aluminum, while the impacts on dynamics was unexplored.19 Therefore, a comprehensive evaluation of all the temperature effects consistently in representative systems are valuable to clarify their roles in dynamical processes at high temperatures.
In this benchmark study, we employ density functional theory (DFT) calculations to systematically examine all the four types of temperature effects on the hopping distance, defect concentration, attempt frequency, and activation barrier in a consistent manner. By closely examining several distinct diffusion processes, we clarify the impacts of those temperature effects and obtain results with much improved accuracies with respect to the experimental results at high temperatures.
Methods
Our ab initio calculations are carried out based on the spin-polarized DFT as implemented in the Vienna Ab initio Simulation Package.20 The electron excitation is included by adopting the Fermi-Dirac distribution with an energy smearing consistent with the examined temperature, the lattice expansion is included by using the experimental lattice constants (see § 10 of the supplementary material), and the lattice vibration are calculated with finite displacements and analyzed using the package phonopy.21 We also obtain the lattice lengths using first-principles calculations and the corresponding diffusivity resembles that based on the experimental lattice length (see § 11 of the supplementary material). For the aluminum system, the exchange-correlation functional is chosen in the form of Perdew-Burke-Ernzerhof (PBE).22 A self-consistent effective Hubbard U term,23 , is added for the electrons in Fe. Our tests show that the Hubbard U term is critical to obtain reasonable hopping barriers (see § 12 of the supplementary material). To accurately describe the electronic properties of 4H-SiC, we employ the Heyd–Scuseria–Ernzerhof hybrid functional (HSE06)24 for all calculations except the lattice vibration, which is too time-consuming using the hybrid functional and thus is calculated self-consistently with the PBE functional for PBE-relaxed structures. Tests of perfect 4H-SiC bulk show that PBE and HSE06 functionals give very close results (see § 13 of the supplementary material). An energy cutoff of 360, 481, and 450 eV is set to the plane-wave basis sets for the , , and systems, respectively, together with the following projector augmented-wave potentials:25 Al_GW() for Al, Fe_GW() for Fe, Si_GW() for Si, and C_GW() for C. The -point spacing of and Å is used for the Al and the 4H-SiC supercells, which are about 12 and 10 Å in each direction, respectively. Convergence tests of energy cutoff, -point sampling, and supercell size can be found in § 12 of the supplemental material. The hopping barriers are calculated using the climbing nudged elastic band method.26
In the ab initio molecular dynamics (AIMD) simulation of diffusion, the canonical ensemble is carried out with the experimental lattice constants, and the simulation time varies from 560 to 3640 ps with a time step of 2 fs. The diffusion coefficient is determined by the mean squared displacement (MSD) as defined in Eq. (1),27
Here, and is the defect concentration in the supercell and that based on Figure 2(b). The MSD trajectories are analyzed using the Molecular Dynamics Analysis of Neutron Scattering Experiments28 and the MSD results and diffusion coefficients are provided in § 14 of the supplemental material. The AIMD simulations of 4H-SiC bulk are carried out in the canonical ensemble using the PBE functional with a time step of 1.5 fs for 15 ps; the HSE06 functional is adopted to calculate the temperature dependence of band edges for 11-15 sampled AIMD structures.

Figure 2. Temperature dependence of the (a) hopping distance, (b) vacancy concentration, (c) attempt frequency, (d) activation barrier, and (e) diffusion coefficient for diffusion in aluminum bulk. In panel (e), an attempt frequency of 5 THz is adopted for the calculations without lattice vibration.
Results and Discussions
In this study, we examine three representative dynamical processes at high temperatures, including aluminum vacancy, , diffusion in aluminum bulk through vacancy hopping in the range of 643–900 K, diffusion of substitutional defect, , in aluminum bulk though vacancy-assisted hopping in the range of 500–900 K, and carbon vacancy, , diffusion in 4H-SiC bulk through vacancy hopping in the range of 2173–2473 K. These systems are chosen because they represent dynamical processes in non-magnetic metal systems, dilute magnetic systems, and semiconductor systems, and also because they possess sufficient experimental data for an all-aspect verification.
Diffusion in Aluminum Bulk
Our first examined process is diffusion through vacancy hopping in aluminum bulk, which is an ideal system for accurate comparison and fast enough to be verified independently using AIMD. In contrast to the widely-used diffusivity formula shown in Eqs. 2(a-b), 29 we impose the temperature effects to every aspect in the diffusion coefficient of , including hopping distance, defect concentration, attempt frequency, and activation barrier, and update the formulas to Eqs. 3(a-b).
where and are geometric factor and correlation factor, which equals 2 and 0.7815 for FCC structure, 30 respectively; is hopping distance; is fractional concentration of aluminum vacancy; is attempt frequency; and is the internal energy and Gibbs free energy for a dilute aluminum vacancy, respectively; and is the activation barrier of the hopping process.
The attempt frequency was typically treated in one of two ways: a constant in the range of – THz31, 32 or a statistical expression initially for molecular systems (Eq. 4a).33 Here, we extend the formula to solids by averaging over all the -points in the Brillouin zone and include the temperature dependence, as shown in Eq. 4b.
where is the number of sampled -points in the Brillouin zone; and are, respectively, the positive vibration frequency of the initial and transition state for the phonon branch at point in reciprocal space. Given the definition of transition state, the number of is one less than that of at each -point. Note that since Eqs. 4(a-b) already incorporate the entropy of activation barrier, the term in Eq. 3a should solely represent internal energy. However, if is associated with free energy with entropy contribution, the attempt frequency must be expressed as .34, 35
In this work, we consider the thermal expansion by adoption of the experimental lattice constants, 36 which increases by ~2.0% at 900 K relative to that at 300 K [Figure 2(a)]. Figure 2(b) compares the vacancy concentration predicted at different levels of thermal corrections with the experiments in the range of 643–900 K. 37–39 The electron excitation is found to play a minor role, with the results being 0.85–0.88 times of the 0 K results. Here, “0 K” means none of the lattice expansion, lattice vibration, and electron excitation is included. However, further inclusion of the lattice expansion decreases the vacancy concentration to 0.29–0.21 times of that with electron excitation, and lattice vibration dramatically increases it to 15.25–15.50 times of that with lattice expansion. In total, all the temperature effects induce an increasing vacancy concentration 3.78–2.88 times of that at 0 K.
Figure 2(c) shows that the attempt frequency based on our extended formula of Eq. 4b predicts a slow decrease trend from 5.6 to 5.2 THz in the range of 643–900 K. Relative to the values based on the molecular formula of Eq. 4a, they are around 0.2 THz higher. By contrast, the classical mechanics formula of Eq. 4a incorrectly predicts an increase of attempt frequency from 13.40 to 18.75 THz in the same temperature range.
Figure 2(d) compares the temperature effects on activation barrier. The electron excitation is found to induce a negligible decrease of around 1 meV relative to that at 0 K. The lattice expansion decreases the hopping barrier by 0.05–0.10 eV in the range of 643–900 K, and the lattice vibration further decreases it by 0.05–0.08 eV. In total, the lattice expansion and vibration decrease the hopping barrier almost linearly from 0.11 eV at 643 K to 0.17 eV at 900 K, about twice of the thermal energy . Relative to previous studies without considering the full temperature effects and/or with potentially insufficient computational accuracy, [13] [19, 40, 41] noticeable differences are observed in the formation energy, migration barrier, and attempt frequency, as shown in § 15 of the supplementary material.
Figure 2(e) further compares the temperature dependence of diffusivity among results with different thermal corrections and experimental data. 42–44 To independently examine the accuracy and computational cost of the thermal corrections, we also carried out AIMD simulations in the temperature range of 643–900 K up to 0.5–3.6 ns. Our results show that the electron excitation and lattice expansion induce slight changes relative to that at 0 K, and they are about 1/15–1/19 times of the experimental values. By contrast, the diffusivities with all the thermal corrections are significantly improved and agree reasonably well with both experiments and the AIMD results. However, the AIMD simulations in this example are about times more time-consuming than calculations with all the thermal corrections. Therefore, this example clearly indicates that including the thermal corrections and utilizing the Eqs. 3a-3b and 4b, we can largely address the underestimated diffusivity at 0 K and predict reliable diffusivity at high temperatures.
FeAl Diffusion in Aluminum Bulk
To evaluate the thermal corrections for more complex systems, we examine the vacancy-assisted diffusion of FeAl in Al bulk from 500 to 900 K. The diffusion of such dynamical process is often described by the five-frequency model,45 which involves five frequencies – related to the hoppings of vacancy far away or near a dilute solute. Because is close to zero in this specific case, the five-frequency model can be simplified to three frequencies of – (see section S7 of the supplementary material), the thermal effects of which we will focus on. Here, both the hopping distance and vacancy concentration are same as Figure 2(a-Figure 2b). Figure 3(a) shows that the attempt frequency varies with processes and exhibits a decreasing trend with temperature, similar to that of aluminum vacancy diffusion in Figure 2(c). Specifically, the attempt frequency of – decreases from 3.3 to 2.2 THz, from 6.0 to 3.5 THz, and from 9.0 to 8.1 THz, respectively.

Figure 3. Temperature dependence of (a) attempt frequency and (b–d) activation barriers for processes –, and (e) comparison among theoretical diffusion coefficients with different temperature effects and experimental results46-48 for FeAl diffusion in aluminum bulk.
For the temperature dependence of hopping barriers, the electron excitation is found to be trivial, with changes less than 10 meV Figure 3(b–Figure 3d). By contrast, the lattice expansion decreases the of – from 0.05 to 0.20 eV, from 0.03 to 0.15 eV, and from 0.12 to 0.20 eV in the range of 500–900 K, respectively. The lattice vibration decreases further and the whole temperature effects decrease of – from 0.09 to 0.28 eV, from 0.08 to 0.22 eV, and from 0.16 to 0.27 eV in the range of 500–900 K, respectively. These energy changes are about 1.8–3.6 times of the thermal energy . Compared to previous work with partial temperature effects and without Hubbard U interactions[2], noticeable differences in migration barriers and attempt frequencies are observed (see § 15 of the supplementary material).
Figure 3(e) shows that the total temperature effects increase the diffusion coefficients to 213.6–312.2 times of those at 0 K in the range of 500–900 K. Compared with the experimental results,46-48 the theoretical predictions with all temperature effects agrees well with the experiments around 900 K, although noticeably higher predictions than the experimental values are observed around 700 K. Two potential reasons could contribute to such discrepancy. First, the error bar of experimental data may be higher at low temperatures because of the significantly slow dynamical processes. Second, the transition metal iron may need an improved description beyond the current Hubbard U method. We compare an effective of 0, an empirical value of 3 eV,49, 50 and a self-consistent value of 2.7 eV for at 0 K, and the respective activation barrier is calculated to be 1.33, 0.74, and 0.79 eV (see § 12 of the supplementary material). Although the self-consistent is possibly the optimal choice at hand, we are open to the possibility that further improvement could be made in the future. With improved experimental data, more accurate description of transition metal element, and the high temperature corrections proposed in this work, a better agreement should be reached in the future.
Diffusion in 4H-SiC Bulk
Different from metallic systems, defects in semiconductors or insulators could be charged and experience a transition-state-redox effect during dynamical processes,[1] namely that the transition state possesses a different charge state from that of the initial/final state in certain Fermi level range. Moreover, the band edges of semiconductors and insulators can change dramatically with temperature and consequently impact the Fermi energy and defect concentration. To evaluate the temperature effects in semiconductors, we examine the diffusion through vacancy hopping in 4H-SiC bulk in the same temperature range of 2373–2573 K as previous experiments.51 The diffusion coefficient of follows Eq. 3a-b, except that the geometric factor is 0.75,[12] [52] and the correlation factor is 0.7815 in the direction [1] of a hexagonal system.53 The formation energy of a defect with charge state is updated to Eq. (8) based on the original formula,54
where and are the free energies of the system with defect and perfect bulk at temperature , respectively; is the FNV correction with an experimental dielectric constant of to reduce image charge interactions under periodic boundary conditions;1 In the correction, the experimental dielectric constant is used; is the chemical potential of element to balance the composition between the perfect bulk and defective system, and of graphite is used in this case; and are the valence band maximum (VBM) energy and Fermi energy relative to the VBM energy at temperature , respectively.
We first examine the temperature dependence of VBM, . AIMD simulations of 4H-SiC bulk are carried out using the PBE functional with experimental lattice constants for 15 ps and sampled AIMD structures are further calculated using HSE06 functional to obtain the band edges and Si-2p energy level [Fig. 4(a-b)]. Figure 4(c) indicates that the VBM increases slightly with temperature by 0.04 eV at 2473 K relative to that at 0 K, but the CBM decreases dramatically by 1.26 eV. The predicted band gap of 1.9 eV at 2473 K is reasonably consistent with the experimental value of 2.2 eV.56 The energy level of Si-2p decreases by 0.9 eV from 0 to 2473 K and a linear interpolation in the range of 873-1623 K corresponds to an energy decrease of 0.27 eV, consistent with the value of 0.21 eV according to X-ray photoelectron spectroscopy.57, 58 Further charge density analyses reveal that the VBM of 4H-SiC is tightly bound to the C atoms, but the CBM is distributed inside the hollow space of the crystal structure (see § 17 of the supplementary material). Such feature is expected to induce easier changes of the CBM by the structural distortion at high temperatures than the VBM.

Figure 4. Band edges and Si-2p level for 4H-SiC bulk at sampled time steps for AIMD simulations at (a) 1373 and (b) 2473 K; (c) temperature dependence of band edges and Si-2p for 4H-SiC bulk.
Then we examine the charge state and formation energy of in the initial and transition states, with consideration of the lattice expansion, lattice vibration, electron excitation, and band-edge shift. Figure 5(a) shows that the initial ground state of exhibits 2+, 1+, and 0 charge state in the Fermi level range of 0-0.66, 0.66-0.95, and 0.95-1.86 eV, respectively, at 2373 K. By contrast, the transition state of hopping exhibits 2+, 1+, and 0 charge state in the range of 0-0.20, 0.20-0.68, and 0.68-1.86 eV, respectively. Therefore, the initial and transition state possess different charge states in the Fermi level range of 0.20-0.66 and 0.68-0.95 eV, a phenomenon named as transition state redox 1. Similar features are also observed at 2473 and 2573 K [Fig. 5(b-c)]. For 4H-SiC bulk, the intrinsic Fermi level is calculated to be 1.04, 1.03, and 1.01 eV at 2373, 2473, and 2573 K, respectively, according to the charge neutrality relationship in Eq. (9).

Figure 5. Defect formation energy of the initial state (IS) and transition state (TS) during the hopping in 4H-SiC at (a) 2373, (b) 2473, and (c) 2573 K with all temperature effects and (d-f) corresponding plots without lattice vibration. The blue number beside each segment labels its charge state and the vertical violet dashed line indicates the Fermi level in intrinsic 4H-SiC.
where and are the VBM and CBM energy at , respectively; is the density of state at 0 K. Here, a rigid shift of the valence bands and conduction bands obtained at 0 K according to the and in Figure 4(c) is utilized. More rigorous calculations using the average DOS of AIMD structures gives similar result (see § 18 of the supplementary material). According to the Fermi levels and Figure 5(a-Figure 5c), the charge state of in intrinsic 4H-SiC are 0 in both the initial and transition states under the three temperatures. The corresponding defect formation energies and hopping barriers of can also be easily obtained. For comparison, the calculations with only the electron excitation, lattice expansion, and band-edge shift predict to be 1+ in its initial state [Figure 5(d-Figure 5f)], and those without any temperature effects predict it to be 2+ (see section S10 of the supplementary material). For transition state, its charge state is 0 at all levels of thermal corrections. Therefore, the temperature effects are critical in determining the charge state and formation energy of defects in semiconductors and insulators.
Figure 6 summarizes the temperature dependences of hopping distance, vacancy concentration, attempt frequency, hopping barrier, and diffusion coefficient. Experimental result in Figure 6(a) shows that the lattice expansion of 4H-SiC in the range of 2373–2573 K is about 1.2% relative to that at 300 K.59 As for the vacancy concentration, the electron excitation slightly decreases it to 0.66–0.68 times of that at 0 K. The lattice expansion increases the vacancy concentration to 2.09–2.35 times of that with only the electron excitation, and the lattice vibration further increases it by 111.52–113.42 times. In total, all the temperature effects increase the vacancy concentration to 153.25–181.30 times of that at 0 K, significantly more than that in aluminum bulk [Figure 2(b)].

Figure 6. FIG. 6. Temperature dependence of (a) hopping distance, (b) vacancy concentration, (c) attempt frequency of diffusion along the [1] direction of 4H-SiC bulk. Comparison of phonon DOS between (d) 4H-SiC bulk at 2473 K and aluminum bulk at 726 K, and phonon DOS of the initial and transition states for (e) hopping in 4H-SiC bulk at 2473 K and (f) hopping aluminum bulk at 726 K; (g) hopping barrier and (h) diffusion coefficient of in 4H-SiC bulk. The LO-TO splitting is included in the phonon DOS of 4H-SiC bulk (see the effects of LO-TO splitting in section S11 of the supplementary material).
Surprisingly, we find that the attempt frequency in 4H-SiC reaches 530–641 THz in the range of 2373–2573 K [Figure 6(c)], about 100 times of the typical values in aluminum bulk [Figure 2(c) and 3(a)] or about 20 times of the Debye frequency in 4H-SiC. Such high attempt frequency can be ascribed to two reasons. First, the large Young's modulus of 4H-SiC ( GPa) relative to that of aluminum bulk ( Pa) induces much higher phonon frequencies in the former [Figure 6(d)]. Consequently, the ratio of positive phonon frequencies between the initial and transition state is greater in 4H-SiC than that in aluminum bulk according to Eq. 4b. Second, the phonon DOS of transition state is slightly red-shifted by 0.33–0.43 THz in 4H-SiC [Figure 6(e)], while in aluminum bulk, the transition state phonon DOS is blue-shifted by about 0.10 THz [Figure 6(f)]. These noticeable redshifts also contribute to the high attempt frequency in 4H-SiC.
The hopping barrier of along the [1] direction of 4H-SiC is found to be 4.07 eV without any temperature effects, and the electron excitation decreases it slightly by 0.02–0.03 eV in the examined temperature range [Fig. 6(g)]. When the lattice expansion is included, the barrier is increased by 0.13–0.16 eV. With the lattice vibration, however, it is decreased by 0.71–0.74 eV. Such reduction of hopping barrier is also observed by comparing with earlier predictions with partial or without any temperature effects (see section S6 of the supplementary material). The overall diffusion coefficient of in [Fig. 6(h)] shows that the result at 0 K with a 5 THz attempt frequency is 5 orders lower than the experiment.51 The electron excitation and lattice expansion slightly decrease the diffusion coefficient relative to that at 0 K. By contrast, the lattice vibration increases it by – times. With all the temperature effects, the diffusion coefficient agrees satisfactorily with the experiments.
Conclusion
In summary, we propose a scheme for accurate transition state calculation under high temperatures and apply it to three diffusion processes: diffusion in Al bulk, substitutional defect diffusion in Al bulk, and diffusion in 4H-SiC bulk. Our first-principles calculations reveal that electron excitation has a minor influence on diffusion coefficient, and lattice expansion induces a limited improvement on the prediction of diffusion coefficient. The lattice vibration, however, exhibits a remarkable impact on diffusion coefficient. For semiconductors and insulators, band-edge shift can be significant at high temperatures and is critical to determine the charge state, defect formation energy, and activation barrier. Contrary to the conventional assumption, the attempt frequency is found to vary with temperature and can differ significantly between materials. Overall, inclusion of all the temperature effects reduce the defect formation energies and activation barriers; the diffusion coefficients are dramatically increased relative to those at 0 K and significantly improved agreements with experiments are obtained.
Supplementary Material
See the supplementary material for the experimental lattice constants of aluminum bulk and 4H-SiC bulk; lattice constants of aluminum bulk and 4H-SiC bulk predicted by first-principles calculations and impacts on dynamics; tests of -point mesh, Hubbard U value, energy cutoff, and supercell size; tests of vibrational free energy of 4H-SiC bulk; AIMD results of diffusion in aluminum bulk; five-frequency model for diffusion in aluminum bulk; comparison with theoretical literature; charge density of VBM and CBM in 4H-SiC bulk; DOS and intrinsic Fermi level of 4H-SiC bulk at 2473 K; defect formation energy of in 4H-SiC bulk at 0 K; effects of LO-TO splitting on the phonon DOS, attempt frequency, and diffusivity of in 4H-SiC bulk at 2473 K.
Acknowledgement
This work was financially supported by the Guangdong Provincial Key Laboratory of Computational Science and Material Design (Grant No. 2019B030301001), the Introduced Innovative R&D Team of Guangdong (Grant No. 2017ZT07C062), and the Shenzhen Science and Technology Innovation Committee (No. JCYJ20200109141412308). The first-principles calculations were carried out on the Taiyi cluster supported by the Center for Computational Science and Engineering of Southern University of Science and Technology and also on The Major Science and Technology Infrastructure Project of Material Genome Big-science Facilities Platform supported by Municipal Development and Reform Commission of Shenzhen.
AUTHOR DECLARATIONS
Conflict of Interest
The authors have no conflicts to disclose.
Author Contributions
Chengxuan Ke: Data curation (lead); Formal analysis (equal); Methodology (supporting); Writing – original draft (lead); Writing – review & editing (supporting). Chenxi Nie: Formal analysis (supporting). Guangfu Luo: Conceptualization (lead); Formal analysis (equal); Funding acquisition (lead); Methodology (lead); Resources (lead); Supervision (lead); Writing – review & editing (lead).
DATA AVAILABILITY
The data that support the findings of this study are available from the corresponding author upon reasonable request.
Supplemental Material for "High-Temperature Effects for Transition State Calculations in Solids"
Chengxuan Ke; Chenxi Nie; Guangfu Luo
Experimental lattice constants of aluminum bulk and 4H-SiC bulk
Tables S1 and S2, respectively, list the lattice constants of aluminum bulk and 4H-SiC bulk at different temperatures based on experimental results.[1, 2]
| (°C) | 0 | 100 | 200 | 300 | 400 | 500 | 600 | 650 |
| (Å) | 4.0391 | 4.0486 | 4.0592 | 4.0701 | 4.0820 | 4.0947 | 4.1087 | 4.1162 |
Table S1. Lattice constants of aluminum bulk measured at different temperatures.[1]
| (°C) | 1900 | 2100 | 2200 | 2300 |
| (Å) | 9.339 | 9.350 | 9.356 | 9.361 |
| (Å) | 10.189 | 10.201 | 10.208 | 10.214 |
Table S2. Lattice constants of 4H-SiC bulk calculated accordingly to the formula obtained experimentally.[2]
Lattice constants of aluminum bulk and 4H-SiC bulk predicted by first-principles calculations and impacts on dynamics
The lattice parameters can also be predicted using first-principles calculations according to the free energy diagram as a function of the lattice constants and temperature. Here, the free energy consists of the internal energy and the vibrational free energy. The PBE and HSE06 functional is used for the Al bulk and 4H-SiC bulk, respectively. As shown in Figure S1-Figure S1), the predicted lattice length of Al bulk is about 0.09 Å or 0.8% greater than the experiments in the range of 643-900 K, and the predicted lattice length of 4H-SiC is about 0.07 Å or 0.8% greater than the experiments in the range of 2173-2573 K. Tests of the self-diffusion in Al bulk at 643 K show that the predicted lattice length leads to the hopping distance, attempt frequency, and being 1.01, 0.94, and 0.82 times of the respective value obtained using the experimental lattice length. Overall, the self-diffusion coefficient at 643 K is 0.78 times of the value based on the experimental lattice length, as shown in Fig. S1(c).

Figure S1. Free energy diagram as a function of lattice length and temperature for (a) aluminum bulk and (b) 4H-SiC bulk; the red dashed line is the predicted lattice length based on the minimum of free energy diagram and the blue line corresponds to the experimental values;[1, 2] (c) comparison of self-diffusion coefficient in aluminum bulk among experiments,[3, 4, 5] prediction based on experimental lattice constants, and prediction based on the DFT lattice length in (a).
Tests of -point mesh, Hubbard U value, energy cutoff, and supercell size
Figure S2 shows the convergence test of -point mesh for the formation energy and hopping barrier of in aluminum bulk at 0 K. According to these tests, a -point mesh of is finally used, and the corresponding formation energy and hopping barrier are 0.63 and 0.61 eV, and the corresponding error bars are less than 18 and 3 meV, respectively.
Figure S2. (a) -point test of formation energy and hopping barrier in aluminum bulk at 0 K. (b) Effect of on of – in the diffusion of in aluminum bulk at 0 K. (c) Energy cutoff test for the formation energy and hopping barrier of in 4H-SiC with HSE06 functional at 0 K. (d) Supercell size tests of formation energy in 4H-SiC bulk at the PBE functional level. The Fermi level is set to the middle of band gap and is defined as , where is the supercell volume. With the FNV correction, the error of a 72-atom supercell is around 0.09 eV relative to the value at the dilute limit (), which is also extrapolated from the raw data using polynomial.2
Figure S2 compares the hopping barriers of – at 0 K for diffusion in Al bulk using different Hubbard U values. We find that the process is particularly sensitive to : the activation barrier is 1.33 and 0.74 eV with of 0 and an empirical value of 3 eV,[7, 8] respectively. Such large difference highlights the importance of accurate description of 3 electrons and probably explains the significant underestimation of the diffusion in aluminum bulk in literature.3 A self-consistent method based on the linear response approach4 predicts an of 2.7 eV for in Al bulk. This approach was designed to comply with a property of the exact density functional: the total energy as a function of electron occupation exhibits a piecewise linearity.5 Therefore, we choose eV for our final calculations, and the corresponding activation barriers are 0.79, 0.61, and 0.71 eV for the – processes, respectively.
Figure S2 shows the convergence test of energy cutoff for the formation energy and hopping barrier of in 4H-SiC bulk at 0 K. It turns out that an energy cutoff of 450 eV converges the formation energy and hopping barrier at the level of 0.08 and 0.01 eV, respectively, which are supposed to be low.
Figure S2 shows the supercell size tests for the in 4H-SiC bulk. It is found that, with the FNV correction, the formation energy of a 72-atom supercell is 0.09 eV or 2.0% greater than that of the 576-atom supercell. Since the image charge interaction decreases as the charge decrease, the error bars for and are expected to be even less.
Tests of vibrational free energy of 4H-SiC bulk
Figure S3 compares the vibrational free energies per formula of 4H-SiC bulk calculated self-consistently with the PBE and HSE06 functional. The PBE and HSE06 results agree well in the temperature range of 0–2500 K and follows a linear relationship. Such agreement supports the reliability of the PBE vibrational results.

Figure S3. (a) Vibrational free energies per formula of 4H-SiC bulk versus temperature for the PBE and HSE06 results. (b) Linear relationship between the PBE and HSE06 results.
AIMD results of diffusion in aluminum bulk
Figure S4(a) shows the mean squared displacements (MSD) of diffusion in aluminum bulk obtained at different temperatures. The first time-steps have been discarded to ensure that the system reaches thermal equilibrium, and the nonlinear part of the MSD data at the end of AIMD simulation is not considered for linear fitting because of the insufficient data points. The calculated diffusion coefficient exhibits a perfect linear relationship as shown in Figure S4.

Figure S4. (a) MSD results from 643 to 900 K and (b) temperature dependence of diffusion coefficient for diffusion in aluminum bulk
To clarify if there is any diffusion path different from the vacancy-hopping mechanism utilized in our transition state calculations, we examine the atomic coordinates before and after each hopping, as exemplified in Figure S5-Figure S5). It is found that all the hoppings exhibit a hopping distance of Å (Figure S5), which is consistent with the neighboring Al-Al distance of Å. Further direct visualization of the hopping events also confirms the vacancy-hopping mechanism for the self-diffusion in Al bulk.

Figure S5. Atomic coordinates of (a) , (b) , and (c) direction of one hopping atom in the AIMD simulation of in Al bulk at 900 K. Dashed line represents the mean value before and after a hopping. (d) Statistics of hopping distance in the 900 K AIMD simulation with 55 hoppings in total.
Comparison with theoretical literature
| Dynamics | (eV) | (eV) | (THz) | (m/s) | Supercell size, energy cutoff, functional/method, temperature effects | Ref. |
| hopping in Al bulk (643 K 900 K) | 0.63–0.60 | — | — | — | 64 atoms, 204 eV, PW91 Lattice vibration | [12] |
| 0.57–0.57 0.60–0.56 | — | — | — | 108 atoms, 190 eV/462 eV, PBE Lattice expansion Lattice vibration Electron excitation | [13] | |
| 0.62–0.60 | 0.61 | 8.4 (Debye frequency) | – | 80 atoms, 130 eV, PW91 Lattice vibration | [14] | |
| 0.63–0.61 | 0.57 | 19.3 | – | 32 atoms, 300 eV, PBW Lattice expansion Lattice vibration | [15] | |
| 0.56–0.55 | 0.50–0.43 | 5.63–5.22 | – | 108 atoms, 360 eV, PBE Lattice expansion Lattice vibration Electron excitation | This work | |
| FeAl hopping in Al bulk (500 K 900 K) | 0.48 () | 1.43 () 0.62 () 0.65 () | 3.90 () 4.27 () 5.30 () | – | 108 atoms, 350 eV, PBE Lattice vibration for | [9] |
| 0.56–0.55 () | 0.70–0.51 () 0.54–0.39 () 0.55–0.44 () | 3.34–2.16 () 6.01–3.51 () 9.02–8.12 () | – | 108 atoms, 481 eV, PBE+U Lattice expansion Lattice vibration Electron excitation | This work | |
| hopping in 4H-SiC bulk (2373 K 2573 K) | — | 4.70 | 16 (Debye frequency) | — | 240 atoms, DFTB Lattice vibration | [16] |
| 5.03 | 3.78 | — | — | 72 atoms, 450 eV, HSE06 | [17] | |
| 3.69–3.57 | 3.47–3.46 | 517.75–628.58 | – | 72 atoms, 450 eV, HSE06 Lattice expansion Lattice vibration Electron excitation Band-edge shift | This work |
Table S3. Comparison of defect formation energy, activation enthalpy, attempt frequency, and diffusion coefficient of hopping in Al bulk, vacancy-assisted FeAl diffusion in Al bulk, and hopping in 4H-SiC between this work and theoretical literature.
Five-frequency model for FeAl diffusion in aluminum bulk
Vacancy-assisted solute diffusion in FCC bulk is typically described by the five-vacancy-jump-frequency model,[18] which involves five processes as indicated in Figure S6 and quantified by Eqs. (10)–(13). Specifically, is the jump frequency of a vacancy in pure bulk, the jump frequency of a vacancy between nearest-neighbor sites of a solute, the jump frequency of the solute-vacancy exchange process, the jump frequency of a vacancy from nearest-neighbor sites of solute to non-nearest-neighbor sites, and the jump frequency of the reverse jump of process 3. For the diffusion of FeAl in aluminum bulk, because the hopping barrier of process and is around 0.38 and 0.74 eV, respectively, is close to zero (–) in our examined temperature of 500–900 K. Therefore, in Eq. (12) can be approximated as 1, and the diffusion coefficient in Eq. (10) can be solely determined by , , , and , as expressed in Eq. (8).

Figure S6. Five-vacancy-jump-frequency model for FCC lattice.
Here, is the formation energy of and is the process’ attempt frequency calculated using Eq. 4b in the main text.
Charge density of VBM and CBM in 4H-SiC bulk
Figure S7-Figure S7) shows the charge density of CBM and VBM at 0 K. The CBM majorly locates in the hollow region in the crystal, but VBM binds locally around C atoms. Such spatial distribution does not change radically in the MD simulations at high temperatures, as shown in Figure S7-Figure S7). Therefore, the loose (tight) spatial distribution of CBM (VBM) qualitatively explains the fact that the CBM is much more sensitive to the lattice change than VBM as temperature increases.

Figure S7. Charge density of (a) CBM and (b) VBM of 4H-SiC bulk at 0 K; examples of charge density of (c) CBM and (d) VBM for 4H-SiC at 2473 K.
DOS and intrinsic Fermi level of 4H-SiC bulk at 2473 K
Figure S8 compares between the average DOS of 4H-SiC bulk obtained from AIMD structures at 2473 K and the DOS of 0 K with a rigid shift of the conduction bands by -1.31 eV to recover the band gap reduction. Because of the lattice vibration, the average DOS at 2473 K is smoother than that of 0 K. The intrinsic Fermi level calculated based on the average DOS of 2473 K and the shifted DOS of 0 K is 1.029 and 1.026 eV above the VBM, respectively. Given such negligible difference and the high computational cost of average DOS from AIMD structures, the shifted DOS of 0 K is more favorable here.

Figure S8. DOS of (a) the 7000{sup}th, 7400{sup}th, 7800{sup}th, 8200{sup}th, 8600{sup}th, and 9000{sup}th step in AIMD simulation at 2473 K, (b) average DOS of the sampled AIMD structures and DOS of 0 K with the conduction bands shifted by -1.31 eV according to Figure 4c.
Defect formation energy of in 4H-SiC bulk at 0 K
Figure S9 shows the defect formation energy of in 4H-SiC at 0 K without any temperature effects. The Fermi level is set to be at the mid-gap, where the charge state of is in the initial state and 0 in the transition state in intrinsic 4H-SiC.

Figure S9. Defect formation energy of the initial state (IS) and transition state (TS) during the hopping in 4H-SiC at 0 K without any temperature effects.
Effects of LO-TO splitting on the phonon DOS, attempt frequency, and diffusivity of in 4H-SiC bulk at 2473 K
Figure S10-Figure S10) compares the phonon DOS of in 4H-SiC bulk in the initial and transition states at 2473 K with and without the LO-TO splitting. The LO-TO splitting brings in an increase of attempt frequency by THz or (Figure S10) and an increase of diffusivity by (Figure S10).

Figure S10. Comparison of phonon DOS of in 4H-SiC at 2473 K for the (a) initial state and (b) transition state with and without LO-TO splitting; comparison of the (c) attempt frequency and (d) diffusivity of with and without LO-TO splitting.
References
- [1]A. J. C. Wilson, "The thermal expansion of aluminium from 0° to 650 °C," Proc. Phys. Sci. 53, 235-244 (1941).DOI
- [2]M. Stockmeier, R. Muller, S. A. Sakwe, P. J. Wellmann and A. Magerl, "On the lattice parameters of silicon carbide," J. Appl. Phys. 105, 033511 (2009).DOI
- [3]M. Beyeler and Y. Adda, "Détermination des volumes d'activation pour la diffusion des atomes dans l'or, le cuivre et l'aluminium," J. Phys. 29, 345-352 (1968).
- [4]S. Dais, R. Messer and A. Seeger, "Nuclear-magnetic-resonance study of self-diffusion in aluminium," Mater. Sci. Forum 15-18, 419-424 (1987).DOI
- [5]T. S. Lundy and J. F. Murdock, "Diffusion of Al²⁶ and Mn⁵⁴ in aluminum," J. Appl. Phys. 33, 1671-1673 (1962).DOI
- [6]G. Makov and M. C. Payne, "Periodic Boundary-Conditions in Ab-Initio Calculations," Phys. Rev. B 51, 4014-4022 (1995).pubmed.ncbi.nlm.nih.gov/9979237
- [7]P. Mohn, C. Persson, P. Blaha, K. Schwarz, P. Novak and H. Eschrig, "Correlation induced paramagnetic ground state in FeAl," Phys. Rev. Lett. 87, 196401 (2001).
- [8]L. Wang, T. Maxisch and G. Ceder, "Oxidation energies of transition metal oxides within the GGA+U framework," Phys. Rev. B 73, 195107 (2006).
- [9]H. Wu, T. Mayeshiba and D. Morgan, "High-throughput ab-initio dilute solute diffusion database," Sci Data 3, 160054 (2016).DOI
- [10]M. Cococcioni and S. de Gironcoli, "Linear response approach to the calculation of the effective interaction parameters in the LDA+U method," Phys. Rev. B 71, 035105 (2005).DOI
- [11]S. Falletta and A. Pasquarello, "Hubbard U through polaronic defect states," npj Comput. Mater. 8, 263 (2022).arxiv.org/abs/2209.11341
- [12]K. Carling, G. Wahnstrom, T. R. Mattsson, A. E. Mattsson, N. Sandberg and G. Grimvall, "Vacancies in metals: from first-principles calculations to experimental data," Phys. Rev. Lett. 85, 3862-3865 (2000).pubmed.ncbi.nlm.nih.gov/11041946
- [13]B. Grabowski, L. Ismer, T. Hickel and J. Neugebauer, "Ab initio up to the melting point: Anharmonicity and vacancies in aluminum," Phys. Rev. B 79, 134106 (2009).DOI
- [14]N. Sandberg, B. Magyari-Kope and T. R. Mattsson, "Self-diffusion rates in Al from combined first-principles and model-potential calculations," Phys. Rev. Lett. 89, 065901 (2002).pubmed.ncbi.nlm.nih.gov/12190597
- [15]M. Mantina, Y. Wang, R. Arroyave, L. Q. Chen, Z. K. Liu and C. Wolverton, "First-Principles Calculation of Self-Diffusion Coefficients," Phys. Rev. Lett. 100, 215901 (2008).
- [16]E. Rauls, T. Frauenheim, A. Gali and P. Deak, "Theoretical study of vacancy diffusion and vacancy-assisted clustering of antisites in SiC," Phys. Rev. B 68, 155208 (2003).DOI
- [17]G. Luo, T. F. Kuech and D. Morgan, "Transition state redox during dynamical processes in semiconductors and insulators," NPG Asia Mater. 10, 45-51 (2018).DOI
- [18]R. E. Howard and J. R. Manning, "Kinetics of solute-enhanced diffusion in dilute face-centered-cubic alloys," Phys. Rev. 154, 561 (1967).
