betaFeedback

WorksPhysics and Astronomy

Combined magnetic and structural transition state in first-principles calculations

Abstract

We present a generalized nudged elastic band (NEB) implementation within potential-based constrained density-functional theory (cDFT) for calculating transition pathways that simultaneously optimize atomic, lattice, and magnetic degrees of freedom. Unlike standard fixed-cell NEB calculations restricted to atomic displacements, our framework consistently couples atomic forces, stresses, and spin torques. The method supports fixed- and variable-cell approaches for collinear and non- collinear magnetic configurations, by incorporating magnetic torques and generalized cell coordinates within the NEB optimization. To validate the methodology, we present benchmark calculations for representative magnetostructural systems. Specifically, we investigate the bcc-to-hcp transformation in Fe and the antiferromagnetic-to-ferromagnetic transition in FeRh using the variable-cell formalism. Furthermore, we examine spin-reversal processes in two-dimensional CrI3 to analyze the influence of spin-orbit coupling on transition paths and their associated energy barriers.

PACS numbers:

Introduction

Many fascinating phenomena in condensed-matter physics are governed by activated transformations between competing metastable states, where both the atomic structure and magnetic order evolve cooperatively. Representative examples include magnetostructural phase transitions, pressure-induced magnetic transformations, magnetovolume effects, and magnetic switching[1, 2, 3, 4, 5]. In these systems, the strong coupling between magnetic and structural degrees of freedom gives rise to significant lattice distortions and anisotropic cell variations during magnetic transitions[6]. These processes occur on complex multidimensional potential-energy landscapes, making the identification of the minimum-energy path (MEP) connecting the initial and final states essential for understanding transition mechanisms, activation barriers, and switching kinetics.

The nudged elastic band (NEB) method and its climbing-image extension are standard approaches for locating transition pathways in atomistic simulations[7, 8]. For solid-state transformations involving lattice deformations, the solid-state NEB (SS-NEB) method extends the configuration space to include both atomic positions and lattice vectors within a unified Euclidean framework[9, 10, 11, 12]. Magnetic transitions introduce an additional geometric challenge because orientations of fixed-magnitude local magnetic moments evolve on the curved Riemannian manifold ∏iS2\prod_i S^2 rather than in Euclidean space. To address this, Bessarab et al. developed the geodesic NEB (GNEB) method[13, 14], later complemented by geodesic minimum-mode following (GMMF) for direct saddle-point searches[15]. However, these works rely on spin Hamiltonians and neglect the explicit self-consistent evolution of the electronic structure, limiting their applicability to itinerant magnets and systems with significant charge redistribution.

Constrained-spin density functional theory (DFT) provides a first-principles route to magnetic transition states[16, 17, 18, 19]. Recent developments, including the self-adaptive DeltaSpin framework[20] and the potential-based constrained DFT (cDFT) formulation of Gonze et al.[21], have substantially improved the robustness of constrained magnetic calculations. In the latter approach, magnetic constraints are incorporated directly into the potential-based self-consistency problem. The target magnetic moments are thereby treated as external parameters on the same footing as the atomic positions and lattice parameters, allowing spin torques, atomic forces, and stresses to be evaluated consistently. This formulation is therefore particularly well suited for coupled magnetostructural transition-state calculations.

The importance of coupling magnetic and structural relaxations has recently been emphasized. Lawrence et al. showed that structural relaxation near magnetic transition states can reduce activation barriers by up to 20% in antiferromagnetic systems[22]. Tang et al. subsequently introduced a magnetic string method based on DeltaSpin-constrained DFT calculations[23]. Nevertheless, their approach neglects lattice degrees of freedom, restricting the calculations to fixed simulation cells and preventing a fully coupled magnetostructural description.

To our knowledge, no first-principles framework simultaneously optimizes atomic positions, lattice vectors, and magnetic moments along a transition pathway while maintaining full electronic self-consistency. Existing approaches either decouple structural and magnetic optimizations or neglect macroscopic strain, which can lead to inaccurate transition mechanisms and energy barriers when lattice-spin coupling is strong.

In this work, we address this challenge by combining potential-based cDFT with a generalized transition-path methodology implemented in the ABINIT package. The resulting cDFT-NEB framework enables the simultaneous optimization of atomic coordinates, lattice vectors, and non-collinear magnetic moments under full electronic self-consistency.

We demonstrate the capabilities of the method using three representative applications: the AFM-to-FM transition in FeRh, the coupled structural and magnetic transformation from bcc-FM to hcp-NM iron, and noncollinear spin reversal in monolayer CrI3_3. Together, these applications demonstrate that the framework consistently captures the coupled evolution of atomic, lattice, and magnetic degrees of freedom and allows the effects of cell relaxation and pressure on transition pathways and activation barriers to be assessed. Our work therefore provides a unified first-principles methodology for investigating transition mechanisms and activation barriers in strongly coupled magnetostructural materials.

The remainder of this paper is organized as follows. Section II briefly reviews the potential-based constrained DFT formalism underlying the present implementation. Section III presents the generalized cDFT-NEB methodology, including the treatment of atomic, lattice, and magnetic degrees of freedom as well as the optimization algorithms. Section IV reports the computational details and benchmark applications to FeRh, Fe, and monolayer CrI3_3, demonstrating the capabilities of the proposed approach. Finally, Section V summarizes the main conclusions and perspectives of this work.

Method

Potential-based constrained density functional theory

The present implementation is based on the potential-based constrained density-functional theory (cDFT) formalism introduced in Ref. 21. In this framework, the screened Kohn-Sham potential is treated as the fundamental variational quantity, allowing charge and magnetic constraints to be incorporated directly within the self-consistent procedure. In contrast to conventional density-based approaches, the potential-based formulation places constrained variables, atomic coordinates, and lattice degrees of freedom on the same footing within the variational formalism.

For a trial screened potential uu, the associated Kohn-Sham equations are solved according to

(T^+u^)∣ϕi,u⟩=εi,u∣ϕi,u⟩(1)(\hat{T}+\hat{u})|\phi_{i,u}\rangle=\varepsilon_{i,u}|\phi_{i,u}\rangle \tag*{(1)}

where T^\hat{T} is the kinetic-energy operator and ∣ϕi,u⟩|\phi_{i,u}\rangle are the Kohn-Sham orbitals associated with the trial potential uu, and εi,u\varepsilon_{i,u} denote the corresponding Kohn-Sham eigenvalues. The corresponding electronic density is obtained as

ρv[u](r)=∑ifiϕi,u∗(r)ϕi,u(r)(2)\rho^{v}[u](\mathbf{r})=\sum_i f_i\phi_{i,u}^{*}(\mathbf{r})\phi_{i,u}(\mathbf{r}) \tag*{(2)}

with fif_i denoting the occupation numbers.

Within the potential-based formalism, the self-consistent electronic energy is recast as a minimization problem in the space of screened potentials,

EvextSC=min⁡uEvextv[u](3)E_{\mathrm{vext}}^{\mathrm{SC}}=\min_u E_{\mathrm{vext}}^{v}[u] \tag*{(3)}

with

Evextv[u]=T[{ϕi,u}]+∫ρv[u](r)vext(r) dr+EHxc[ρv[u]](4)E_{\mathrm{vext}}^{v}[u]=T[\{\phi_{i,u}\}]+\int\rho^{v}[u](\mathbf{r})v_{\mathrm{ext}}(\mathbf{r})\,d\mathbf{r}+E_{\mathrm{Hxc}}[\rho^{v}[u]] \tag*{(4)}

where vextv_{\mathrm{ext}} is the external ionic potential and EHxcE_{\mathrm{Hxc}} contains the Hartree and exchange-correlation contributions. The gradient of the functional with respect to the screened potential is given by[21]

δEvextv[u]δu(r)=∫χ0(r,r′)Rv[u](r′) dr′(5)\frac{\delta E_{\mathrm{vext}}^{v}[u]}{\delta u(\mathbf{r})}=\int\chi_0(\mathbf{r},\mathbf{r}')R^{v}[u](\mathbf{r}')\,d\mathbf{r}' \tag*{(5)}

where χ0(r,r′)\chi_0(\mathbf{r},\mathbf{r}') is the independent-particle susceptibility and

Rv[u]=Kv[u]−u(6)R^{v}[u]=K^{v}[u]-u \tag*{(6)}

is the potential residual. Here, Kv[u]K^{v}[u] represents the output screened potential obtained after solving the Kohn-Sham equations for a trial input potential uu. Self-consistency is reached when the input and output screened potentials become identical, corresponding to a vanishing residual Rv[u]=0R^{v}[u]=0.

Within cDFT, constraints are introduced through Lagrange multipliers. For multiple constraints labelled by II, the augmented functional is written as

Evext,{NI}+v[u,{ΛI}]=Evextv[u]+∑IΛI(ρIv[u]−NI)(7)E_{\mathrm{vext},\{N_I\}}^{+\mathrm{v}}[u,\{\Lambda_I\}]=E_{\mathrm{vext}}^{v}[u]+\sum_I\Lambda_I\left(\rho_I^{v}[u]-N_I\right) \tag*{(7)}

where ΛI\Lambda_I denotes the Lagrange multiplier associated with constraint II, while NIN_I is the target value of the corresponding constrained quantity. When constraining the local magnetization, NIN_I represents the target value of the IIth local magnetic moment, denoted by MI,targetM_{I,\mathrm{target}}.

For noncollinear magnetism, the electronic state is described through the spin-density matrix,

ρss′(r)=∑ifiϕs,i∗(r)ϕs′,i(r)(8)\rho_{ss'}(\mathbf{r})=\sum_i f_i\phi_{s,i}^{*}(\mathbf{r})\phi_{s',i}(\mathbf{r}) \tag*{(8)}

with spin indices s,s′∈{↑,↓}s,s' \in\{\uparrow,\downarrow\}.

Local magnetic moments are obtained from weighted integrations of the spin-density matrix within atom-centered regions. For atom κ\kappa, the local magnetic moment is defined as

Mκ=∑ss′∫d3r wκ(r)σss′ρss′(r),(9)\mathbf{M}_{\kappa}=\sum_{ss'}\int d^{3}r\,w_{\kappa}(\mathbf{r})\boldsymbol{\sigma}_{ss'}\rho_{ss'}(\mathbf{r}), \tag*{(9)}

where wκ(r)w_{\kappa}(\mathbf{r}) denotes the atom-centered weight function associated with atom κ\kappa and σ\boldsymbol{\sigma} are the Pauli matrices.

In Ref. 21, it is shown how the Lagrange parameters ΛI\Lambda_I can be obtained. For simplicity, here we show the line of thought in the case of a single constraint, labeled AA, with weight function wA(r)w_A(\mathbf{r}). The projection of the potential residual onto the corresponding constraint subspace is defined as

RAv[u]=∫dr Rv[u](r)wA(r),(10)R_A^v[u]=\int d\mathbf{r}\,R^v[u](\mathbf{r})w_A(\mathbf{r}), \tag*{(10)}

while the overlap of the weight function with itself is

WAA=∫dr wA(r)wA(r).(11)W_{AA}=\int d\mathbf{r}\,w_A(\mathbf{r})w_A(\mathbf{r}). \tag*{(11)}

The corresponding Lagrange multiplier is ΛA[u]=−RAv[u](WAA)−1\Lambda_A[u]=-R_A^v[u](W_{AA})^{-1}. Substituting this expression into the augmented functional yields the potential-based cDFT functional

Evext,NAcDFT[u]=Evextv[u]−RAv[u](WAA)−1(ρAv[u]−NA).(12)E^{\mathrm{cDFT}}_{v_{\mathrm{ext}},N_A}[u]=E^v_{v_{\mathrm{ext}}}[u]-R_A^v[u](W_{AA})^{-1}\left(\rho_A^v[u]-N_A\right). \tag*{(12)}

This functional depends only on the trial screened potential uu and does not require the Lagrange multiplier to be determined through a separate optimization procedure. Consequently, the target constraint NAN_A is treated as an external parameter on the same footing as the atomic coordinates and lattice parameters.

The different gradients with respect to the external parameters, as well as to the constraints, can now be obtained. The derivative of the total energy with respect to the constrained magnetic direction defines the spin torque,

TJ=−∂Evext,{MJ,target}v+[v∗]∂MJ,target=ΛJ[v∗].(13)T_J=-\frac{\partial E^{v+}_{v_{\mathrm{ext}},\{\mathbf{M}_{J,\mathrm{target}}\}}[v^*]}{\partial\mathbf{M}_{J,\mathrm{target}}}=\Lambda_J[v^*]. \tag*{(13)}

Owing to the stationary character of the cDFT functional at the self-consistent screened potential v∗v^*, the Hellmann-Feynman theorem can be applied directly to constrained calculations. The force acting on atom κ\kappa along Cartesian direction α\alpha is therefore written as

Fκα=−∂Evextv∂τκα∣v∗+∑JTJ∂RJv[u]∂τκα∣v∗,(14)F_{\kappa\alpha}=-\left.\frac{\partial E^v_{v_{\mathrm{ext}}}}{\partial\tau_{\kappa\alpha}}\right|_{v^*}+\left.\sum_J T_J\frac{\partial R_J^v[u]}{\partial\tau_{\kappa\alpha}}\right|_{v^*}, \tag*{(14)}

where τκα\tau_{\kappa\alpha} denotes the atomic coordinate.

Similarly, the stress tensor is obtained from the derivative of the total energy with respect to the strain tensor ηαβ\eta_{\alpha\beta},

σαβ=1Ω∂Evextv∂ηαβ∣v∗−∑JTJ1Ω∂RJv[u]∂ηαβ∣v∗(15)\sigma_{\alpha\beta}=\left.\frac{1}{\Omega}\frac{\partial E^v_{v_{\mathrm{ext}}}}{\partial\eta_{\alpha\beta}}\right|_{v^*}-\left.\sum_J T_J\frac{1}{\Omega}\frac{\partial R_J^v[u]}{\partial\eta_{\alpha\beta}}\right|_{v^*} \tag*{(15)}

where Ω\Omega is the unit-cell volume. The deformation tensor is defined through

rα′=rα+∑βηαβrβ.(16)r'_{\alpha}=r_{\alpha}+\sum_{\beta}\eta_{\alpha\beta}r_{\beta}. \tag*{(16)}

Within the present implementation, atomic coordinates, lattice vectors, and local magnetic moments are optimized simultaneously during the transition-path calculations described below.

Finding the Transition Path

The present implementation is based on a generalized nudged elastic band (NEB) formalism in which atomic coordinates, lattice degrees of freedom, and magnetic variables are optimized simultaneously along the transition pathway. The method unifies standard NEB, climbing-image NEB (CI-NEB), generalized solid-state NEB (G-SSNEB), and variable-cell NEB formulations within a single framework.

Path Initialization

For each image ii, the generalized coordinate vector is defined as

Xi={Ri,hi,Mi},(17)\mathbf{X}_i=\{\mathbf{R}_i,\mathbf{h}_i,\mathbf{M}_i\}, \tag*{(17)}

where Ri\mathbf{R}_i denotes the atomic coordinates, hi\mathbf{h}_i the lattice degrees of freedom, and Mi\mathbf{M}_i the magnetic degrees of freedom.

The initial pathway is constructed by interpolation between the initial and final images. Atomic coordinates, lattice vectors, and collinear magnetic moments are interpolated linearly. For noncollinear magnetic configurations, a geometrically consistent interpolation of the spin orientations is required. Accordingly, the magnetic moment magnitudes are interpolated linearly, whereas their orientations are initialized using spherical linear interpolation (Slerp)^{24},

Slerp⁡(M1,M2;u)=sin⁡[(1−u)θ]sin⁡θM^1+sin⁡(uθ)sin⁡θM^2,(18)\operatorname{Slerp}(\mathbf{M}_1,\mathbf{M}_2;u)=\frac{\sin[(1-u)\theta]}{\sin\theta}\widehat{\mathbf{M}}_1+\frac{\sin(u\theta)}{\sin\theta}\widehat{\mathbf{M}}_2, \tag*{(18)}

where M^1⋅M^2=cos⁡θ\widehat{\mathbf{M}}_1\cdot\widehat{\mathbf{M}}_2=\cos\theta and u∈[0,1]u\in[0,1]. The spin orientations are thus interpolated along the shortest geodesic on the unit sphere, ensuring a smooth and geometrically consistent initialization of the magnetic pathway. The distinction between linear and spherical interpolation is illustrated schematically in Figure 1. Whereas linear interpolation is appropriate for collinear magnetic configurations, noncollinear spin orientations must be interpolated on the spin manifold to preserve their rotational character.

Schematic illustration of the interpolation of magnetic moments between adjacent images

Figure 1. Schematic illustration of the interpolation of magnetic moments between adjacent images. Collinear moments are interpolated linearly (left), while noncollinear spin orientations are interpolated using spherical linear interpolation (Slerp) along the shortest geodesic on the unit sphere (right).

Generalized Energy Gradients

For each image, the energy derivatives are evaluated with respect to the atomic, lattice, and magnetic degrees of freedom. At finite external pressure, the relevant thermodynamic potential is the enthalpy

Hi=Ei+PextΩi,(19)\mathcal{H}_i = E_i + P_{\mathrm{ext}}\Omega_i, \tag*{(19)}

which reduces to the total energy EiE_i when Pext=0P_{\mathrm{ext}}=0.

Atomic Degrees of Freedom

The force acting on atom κ\kappa in image ii is defined as

Fiκat=−∂Hi∂Riκ,(20)\mathbf{F}^{\mathrm{at}}_{i\kappa} = -\frac{\partial\mathcal{H}_i}{\partial\mathbf{R}_{i\kappa}}, \tag*{(20)}

where Riκ\mathbf{R}_{i\kappa} denotes the position of atom κ\kappa in image ii.

Variable-Cell Degrees of Freedom

To describe structural transformations involving strain and lattice deformation, the generalized configurational space includes cell degrees of freedom following the G-SSNEB formalism of Sheppard et al.[9].

The lattice deformation relative to a reference cell is written as

Δhi=hi−h0,(21)\Delta\mathbf{h}_i = \mathbf{h}_i - \mathbf{h}^{0}, \tag*{(21)}

and the corresponding generalized lattice coordinate is defined as

Xicell=ηi(h0)−TΔhi,(22)\mathbf{X}^{\mathrm{cell}}_i = \eta_i \left(\mathbf{h}^{0}\right)^{-T}\Delta\mathbf{h}_i, \tag*{(22)}

where

ηi=Ωi1/3N1/6,(23)\eta_i = \Omega_i^{1/3}N^{1/6}, \tag*{(23)}

with Ωi\Omega_i denoting the cell volume and NN the number of atoms. To keep the Jacobian constant throughout the NEB optimization, the image-dependent volume Ωi\Omega_i is replaced by the average volume of the fixed initial and final images[9],

Ω‾=Ωinitial+Ωfinal2.(24)\overline{\Omega} = \frac{\Omega_{\mathrm{initial}}+\Omega_{\mathrm{final}}}{2}. \tag*{(24)}

The scaling factor used for all images is therefore

η=Ω‾1/3N1/6.(25)\eta= \overline{\Omega}^{1/3}N^{1/6}. \tag*{(25)}

The generalized lattice force is obtained from the stress tensor as

Ficell=−Ωiησi,(26)\mathbf{F}^{\mathrm{cell}}_i = -\frac{\Omega_i}{\eta}\boldsymbol{\sigma}_i, \tag*{(26)}

where

σi=σiint+PextI.(27)\boldsymbol{\sigma}_i = \boldsymbol{\sigma}^{\mathrm{int}}_i + P_{\mathrm{ext}}\mathbf{I}. \tag*{(27)}

Here, σiint\boldsymbol{\sigma}^{\mathrm{int}}_i is the internal stress tensor, PextP_{\mathrm{ext}} is the external hydrostatic pressure, and I\mathbf{I} is the identity tensor.

Magnetic Degrees of Freedom

Magnetic degrees of freedom are incorporated into the generalized configurational space whenever the magnetic configuration changes along the transition pathway. For each image ii, the magnetic contribution to the generalized force is given by the cDFT magnetic torque Ti\mathbf{T}_i defined in Eq. (13). For noncollinear configurations, the local magnetic moments are updated according to the magnetic torque Ti\mathbf{T}_i, allowing both their orientations and magnitudes to evolve during the pathway optimization. The atomic, lattice, and magnetic contributions are collected into the generalized physical-force vector

Fi={Fiat,Ficell,Ti}.(28)\mathbf{F}_i = \left\{\mathbf{F}^{\mathrm{at}}_i,\mathbf{F}^{\mathrm{cell}}_i,\mathbf{T}_i\right\}. \tag*{(28)}

Generalized Nudged Elastic Band Method

Starting from the initialized pathway, the generalized NEB method relaxes the intermediate images toward a minimum-energy path in the combined atomic, lattice, and magnetic configurational space. The generalized NEB force acting on image ii is decomposed into a perpendicular physical force and a parallel spring contribution,

FiNEB=Fi⊥+Fispring,(29)\mathbf{F}^{\mathrm{NEB}}_i = \mathbf{F}^{\perp}_i + \mathbf{F}^{\mathrm{spring}}_i, \tag*{(29)}

The perpendicular component is obtained by projecting out the component parallel to the tangent direction,

Fi⊥=Fi−(Fi⋅τ^i)τ^i.(30)\mathcal{F}_i^{\perp}=\mathcal{F}_i-(\mathcal{F}_i\cdot\hat{\tau}_i)\hat{\tau}_i. \tag*{(30)}

where τ^i\hat{\tau}_i denotes the normalized tangent vector.

The tangent construction follows the improved prescription of Henkelman and Jónsson.[8] Defining

ΔXi=Xi−Xi−1,(31)\Delta\mathbf{X}_i=\mathbf{X}_i-\mathbf{X}_{i-1}, \tag*{(31)}

the tangent vector is evaluated according to the local energy ordering,

τi={ΔXi+1,Ei+1>Ei>Ei−1,ΔXi,Ei+1<Ei<Ei−1.(32)\tau_i= \begin{cases} \Delta\mathbf{X}_{i+1}, & E_{i+1}>E_i>E_{i-1},\\ \Delta\mathbf{X}_i, & E_{i+1}<E_i<E_{i-1}. \end{cases} \tag*{(32)}

At extrema along the pathway, an energy-weighted interpolation is used,

τi={ΔXi+1ΔEmax⁡+ΔXiΔEmin⁡,Ei+1>Ei−1,ΔXi+1ΔEmin⁡+ΔXiΔEmax⁡,Ei+1<Ei−1.(33)\tau_i= \begin{cases} \Delta\mathbf{X}_{i+1}\Delta E_{\max}+\Delta\mathbf{X}_i\Delta E_{\min}, & E_{i+1}>E_{i-1},\\ \Delta\mathbf{X}_{i+1}\Delta E_{\min}+\Delta\mathbf{X}_i\Delta E_{\max}, & E_{i+1}<E_{i-1}. \end{cases} \tag*{(33)}

where

ΔEmax⁡=max⁡(∣Ei+1−Ei∣,∣Ei−1−Ei∣),(34)\Delta E_{\max}=\max\left(\lvert E_{i+1}-E_i\rvert,\lvert E_{i-1}-E_i\rvert\right), \tag*{(34)}

and

ΔEmin⁡=min⁡(∣Ei+1−Ei∣,∣Ei−1−Ei∣).(35)\Delta E_{\min}=\min\left(\lvert E_{i+1}-E_i\rvert,\lvert E_{i-1}-E_i\rvert\right). \tag*{(35)}

The normalized tangent vector is then

τ^i=τi∣τi∣.(36)\hat{\tau}_i=\frac{\tau_i}{\lvert\tau_i\rvert}. \tag*{(36)}

The spring force is evaluated using the distance-based formulation,

Fispring=[ki+1∣Xi+1−Xi∣−ki∣Xi−Xi−1∣]τ^i,(37)\mathcal{F}_i^{\mathrm{spring}}=\left[k_{i+1}\lvert\mathbf{X}_{i+1}-\mathbf{X}_i\rvert-k_i\lvert\mathbf{X}_i-\mathbf{X}_{i-1}\rvert\right]\hat{\tau}_i, \tag*{(37)}

where kik_i denotes the spring constant associated with image ii.

Both constant and adaptive spring schemes are implemented. In the adaptive formulation, the spring constants increase near the saddle-point region in order to improve the image distribution around the transition state.

Climbing-Image NEB

The implementation includes the climbing-image extension of the NEB method. After an initial relaxation stage, the image with the highest energy is converted into a climbing image by removing the spring-force contribution and inverting the parallel component of the physical force,

FiCI=Fi−2(Fi⋅τ^i)τ^i.(38)\mathcal{F}_i^{\mathrm{CI}}=\mathcal{F}_i-2(\mathcal{F}_i\cdot\hat{\tau}_i)\hat{\tau}_i. \tag*{(38)}

This procedure drives the climbing image toward the first-order saddle point.

Numerical Optimization

The generalized NEB forces are relaxed using several optimization algorithms, including steepest descent (SD)[25], local limited-memory Broyden-Fletcher-Goldfarb-Shanno (LL-BFGS), and global limited-memory BFGS (GL-BFGS)[26]. Both BFGS schemes retain only a limited history of coordinate and force differences. The distinction between them concerns the configuration space over which the approximate inverse Hessian is constructed.

In the steepest-descent method, the generalized coordinates of each image are updated along the corresponding NEB force,

Xi(n+1)=Xi(n)+αFiNEB,(n),(39)\mathbf{X}_i^{(n+1)}=\mathbf{X}_i^{(n)}+\alpha\mathcal{F}_i^{\mathrm{NEB},(n)}, \tag*{(39)}

where α\alpha is the optimization step size. In practice, separate step-size parameters are used for the structural and magnetic degrees of freedom to account for their different energy scales and convergence behavior. The atomic and lattice variables are updated using αstr\alpha_{\mathrm{str}}, whereas the magnetic variables are updated using αmag\alpha_{\mathrm{mag}}.

In the LL-BFGS scheme, each intermediate image has an independent optimization history and an independent approximate inverse Hessian. The search direction for image ii is therefore evaluated as

di(n)=(Hi,nLL)−1FiNEB,(n),(40)\mathbf{d}_i^{(n)}=\left(\mathbf{H}_{i,n}^{\mathrm{LL}}\right)^{-1}\mathcal{F}_i^{\mathrm{NEB},(n)}, \tag*{(40)}

and the image is updated according to

Xi(n+1)=Xi(n)+λi,ndi(n).(41)\mathbf{X}_i^{(n+1)}=\mathbf{X}_i^{(n)}+\lambda_{i,n}\mathbf{d}_i^{(n)}. \tag*{(41)}

Although all images are updated during the same NEB iteration, the inverse-Hessian approximations are constructed image by image. Consequently, the LL-BFGS approximation does not explicitly contain the off-diagonal Hessian blocks describing couplings between neighboring images.

In the GL-BFGS scheme, all PP intermediate images are treated collectively. The band-wide coordinate vector Xband(n)\mathbf{X}_{\mathrm{band}}^{(n)} is formed by concatenating the generalized coordinates Xi(n)\mathbf{X}_{i}^{(n)} of all images, while the corresponding force vector Fband(n)\mathbf{F}_{\mathrm{band}}^{(n)} is formed by concatenating their generalized NEB forces FiNEB,(n)\mathbf{F}_{i}^{\mathrm{NEB},(n)}. The entire band is then updated using a single approximate inverse Hessian,

Xband(n+1)=Xband(n)+λn(HnGL)−1Fband(n).(42)\mathbf{X}_{\mathrm{band}}^{(n+1)} = \mathbf{X}_{\mathrm{band}}^{(n)} + \lambda_n \left(\mathbf{H}_{n}^{\mathrm{GL}}\right)^{-1}\mathbf{F}_{\mathrm{band}}^{(n)}. \tag*{(42)}

Unlike LL-BFGS, which constructs an independent inverse-Hessian approximation for each image, GL-BFGS includes inter-image couplings by treating the complete band within a single optimization space.

Results and discussion

To validate the proposed generalized cDFT-NEB framework, we consider a series of representative magnetic and magnetostructural transitions. These examples are designed to assess the capability of the method to simultaneously treat atomic, lattice, and magnetic degrees of freedom within the potential-based constrained density-functional theory (cDFT) formalism.

We first investigate the first-order metamagnetic transition in FeRh between the antiferromagnetic (AFM) and ferromagnetic (FM) phases by comparing fixed-cell and variable-cell transition pathways. We then examine the more challenging direct magnetostructural transformation from bcc-FM to hcp-NM iron, involving substantial lattice deformation, atomic shuffle, volume reduction, and collapse of the local magnetic moments. Finally, we study the collective spin reversal in monolayer CrI3_3, in which the transition is governed predominantly by the evolution of the magnetic moments while the accompanying atomic and lattice relaxations remain negligible. This system also provides a convenient platform for assessing the influence of spin-orbit coupling (SOC) on the minimum-energy path and the associated magnetic anisotropy barrier.

Together, these applications demonstrate the broad applicability of the proposed framework, spanning strongly coupled magnetostructural transformations, predominantly magnetic transitions, and systems involving noncollinear spin rotations.

Computational Details

All calculations were performed within the framework of density-functional theory (DFT) as implemented in the ABINIT software (version 10.8)[27, 28, 29]. The electron-ion interactions for Fe and FeRh were described using the projector augmented-wave (PAW) method[30] with datasets obtained from the Pseudo-Dojo project[31]. To accurately represent the rapidly varying density and potential terms within the PAW augmentation spheres, an auxiliary double-grid energy cutoff of 80 Ha was employed.

FeRh: AFM-FM Metamagnetic Transition

Conversely, optimized norm-conserving pseudopotentials were employed for CrI3_3. Exchange-correlation effects were treated within the Perdew-Burke-Ernzerhof (PBE) generalized-gradient approximation (GGA)[32] for Fe and FeRh, while the local-density approximation (LDA)[33] was adopted for CrI3_3. Spin-orbit coupling was included in all calculations.

For Fe and FeRh, the Kohn-Sham wave functions were expanded in a plane-wave basis set with a kinetic-energy cutoff of 25 Ha. Brillouin-zone integrations employed Monkhorst-Pack kk-point meshes of 16×16×1016 \times16 \times10 for the two-atom Fe cells and 8×8×88 \times8 \times8 for the four-atom FeRh cells. Electronic occupations were treated using a Gaussian smearing scheme with a broadening width of 10−210^{-2} Ha to facilitate self-consistent field convergence. For the monolayer CrI3_3 system, a higher plane-wave kinetic-energy cutoff of 50 Ha and an 8×8×18 \times8 \times1 Monkhorst-Pack kk-point mesh were required. The CrI3_3 simulation cell contained eight atoms and included a vacuum spacing of approximately 18.5 Å along the out-of-plane direction to eliminate spurious interactions between periodic images.

Ground-state structures were fully relaxed until the residual atomic forces fell below 10−410^{-4} Ha/Bohr. Non-collinear magnetic configurations were treated explicitly using spinor wave functions within the spin-density-matrix formalism. Unless otherwise specified, crystal symmetry operations were disabled throughout the nudged elastic band (NEB) calculations to permit the unrestricted evolution of the atomic, lattice, and magnetic degrees of freedom.

Minimum-energy pathways (MEPs) were determined using the generalized cDFT-NEB implementation described in Sec. III, employing the improved-tangent formulation of Henkelman and Jónsson[8]. For fixed-cell calculations, steepest-descent, LL-BFGS, and GL-BFGS optimizers were benchmarked and found to converge to identical MEPs and activation barriers within numerical accuracy. Variable-cell calculations were executed via the SD optimizer, enabling the simultaneous relaxation of atomic coordinates, lattice vectors, and magnetic degrees of freedom within the potential-based constrained DFT framework. Unless stated otherwise, the NEB calculations were considered converged when the mean, over all dynamically optimized images, of the absolute total-energy difference between two successive image iterations fell below 10−510^{-5} Ha.

Equiatomic FeRh is a prototypical metamagnetic alloy exhibiting a first-order transition from an antiferromagnetic (AFM) state to a ferromagnetic (FM) state at an experimental transition temperature of approximately 340–370 K[34, 35, 36]. The material crystallizes in the chemically ordered B2 (CsCl-type) structure with cubic space group Pm3ˉmPm\bar{3}m (No. 221), where Fe atoms occupy the cube corners and Rh atoms occupy the body-centered positions. The AFM-FM transition is isostructural and is accompanied by a small isotropic lattice expansion, making FeRh a benchmark system for studying coupled magnetic and lattice degrees of freedom.

Because the transition occurs close to room temperature and can be manipulated by temperature, strain, magnetic field, and ultrafast excitation, FeRh has attracted considerable attention for spintronic and magnetocaloric applications[36, 37]. Owing to its well-established magnetic transition pathway and the availability of previous string-method and DeltaSpin calculations[23], FeRh provides an ideal test system for validating the present generalized NEB implementation.

In the low-temperature AFM phase, Fe magnetic moments are aligned antiferromagnetically while the Rh moments vanish by symmetry. In the FM phase, the Fe moments align parallel and finite induced magnetic moments develop on the Rh atoms. In the present calculations, the AFM configuration was taken as the initial state and the FM configuration as the final state. A four-atom cell corresponding to the primitive AFM FeRh cell was employed.

To facilitate comparison with previous magnetic string-method calculations[23], the transition pathway was discretized using 11 images. The initial magnetic path was generated by prescribing one Fe moment to remain aligned along the positive spin direction while the second Fe moment gradually rotates from antiparallel to parallel alignment. A middle image was constructed with the two Fe moments perpendicular to each other while preserving approximately constant magnetic-moment amplitudes. The remaining images were initialized using spherical interpolation (Slerp) between neighboring magnetic configurations, ensuring smooth rotational evolution of the local moments during the NEB optimization.

Both fixed-cell and variable-cell calculations were performed. For the fixed-cell calculations, the lattice parameters were constrained to the equilibrium AFM structure, whereas in the variable-cell calculations both lattice and magnetic degrees of freedom were optimized simultaneously along the pathway. The generalized NEB calculations employed the improved-tangent formalism with 11 images. Three optimization schemes were tested for the fixed-cell calculations, namely steepest descent (SD), LL-BFGS, and GL-BFGS optimizations. All three methods converged to identical minimum-energy pathways and barrier heights, demonstrating the robustness of the present implementation. In the current implementation, variable-cell calculations were performed using the steepest-descent optimizer.

The resulting minimum-energy pathways are presented in Figure 2. The magnetic transition proceeds continuously through noncollinear intermediate states connecting the AFM and FM configurations. During the transition, the Fe magnetic moments undergo a gradual reorientation accompanied by a small increase in their magnitudes, while finite induced magnetic moments develop on the

Figure 2. Antiferromagnetic (AFM) to ferromagnetic (FM) transition in FeRh obtained using the generalized cDFT-NEB method. (a) Minimum-energy pathways obtained from fixed-cell (blue circles) and variable-cell (red squares) calculations. Representative magnetic configurations are illustrated for selected images along the pathways. Fe and Rh atoms are shown in gold and gray, respectively, while the arrows indicate the local magnetic moments. Fe1 denotes the Fe site whose magnetic moment remains approximately aligned along the positive zz direction, whereas Fe2 denotes the Fe site whose magnetic moment rotates from antiparallel to parallel alignment with Fe1. The transition proceeds through intermediate non-collinear magnetic configurations. Allowing lattice relaxation slightly reduces the energy barrier, highlighting the role of magnetovolume coupling. (b) Evolution of the magnetic moment magnitudes ∣M1∣|\mathbf{M}_1| and ∣M2∣|\mathbf{M}_2| of Fe1 and Fe2, respectively, along the fixed-cell and variable-cell pathways. (c) Evolution of the polar angles θ\theta of Fe1 and Fe2 magnetic moments with respect to the positive zz axis along the fixed-cell and variable-cell pathways. The angle for Fe1 remains close to 0∘0^\circ, with small deviations at intermediate images, whereas that for Fe2 decreases from 180∘180^\circ to 0∘0^\circ.

Rh atoms.

As shown in Figure 2(b), the magnitudes of the Fe magnetic moments increase from approximately 3.124 μB\mu_B in the AFM state to 3.178 μB\mu_B and 3.191 μB\mu_B in the FM state for the fixed-cell and variable-cell calculations, respectively. Small differences between the magnitudes of the Fe1 and Fe2 moments are observed at intermediate images in the calculations including spin–orbit coupling (SOC). The induced Rh magnetic moments increase from zero in the AFM state to 1.056 μB\mu_B and 1.054 μB\mu_B in the FM state for the fixed-cell and variable-cell calculations, respectively.

Figure 2(c) shows the evolution of the polar angles of the Fe magnetic moments with respect to the positive zz axis. The Fe1 moment remains nearly aligned with the positive zz direction, exhibiting small angular deviations at intermediate images, whereas the polar angle of Fe2 decreases continuously from 180° to 0°, corresponding to its rotation from antiparallel to parallel alignment with Fe1. The magnetic moments rotate predominantly within the xzxz plane, with negligible components along the yy direction.

For the fixed-cell and variable-cell calculations, the energy barrier is found to be approximately 29 meV/atom and 28 meV/atom, respectively, in good agreement with previous DeltaSpin and magnetic string-method studies[23] reporting barriers of ∼\sim 33–35 meV/atom. Using the relation E≈kBTE \approx k_B T, this barrier corresponds to a characteristic temperature close to the experimentally observed AFM-FM transition temperature range.

Allowing simultaneous lattice relaxation lowers the energy barrier further, highlighting the importance of magnetovolume coupling in FeRh. The variable-cell pathway exhibits a smoother evolution of the total energy and a reduced saddle-point energy relative to the fixed-cell calculations, consistent with the experimentally observed lattice expansion accompanying the metamagnetic transition. These results demonstrate that the present generalized cDFT-NEB framework can naturally describe coupled spin and lattice transitions within a unified formalism.

bcc-to-hcp structural and magnetic transition in iron

As a more demanding benchmark of the generalized NEB implementation, we consider elemental iron, a prototypical system in which structural stability and magnetism are strongly coupled. In contrast to FeRh, where the crystal structure remains essentially unchanged during the magnetic transition, the bcc-to-hcp transformation in iron involves simultaneous changes in the internal atomic coordinates, lattice degrees of freedom, and local magnetic moments.

At ambient conditions, ferromagnetic body-centered cubic iron (bcc-FM, α\alpha-Fe) is the ground-state phase. Under pressure, iron undergoes a martensitic transformation

to the hexagonal close-packed phase, accompanied by a reduction in volume and magnetization. Previous first-principles work described the pressure-dependent transition pathway in terms of strongly coupled shear and shuffle modes[38], while in situ x-ray tomography and diffraction measurements confirmed that the transformation follows the Burgers orientation relationship[39]. Within the present computational setup, the nonmagnetic hcp phase lies approximately 54 meV/atom above the bcc-FM ground state at zero pressure.

For a continuous mapping between the bcc and hcp endpoints, the conventional cubic representation of bcc Fe is replaced by the reoriented two-atom cell illustrated in Figure 3. The cell is chosen such that

Construction of the reoriented bcc cell used for the bcc-to-hcp NEB calculations

Figure 3. Construction of the reoriented bcc cell used for the bcc-to-hcp NEB calculations. The conventional cubic bcc cell is outlined in red, while the reoriented two-atom cell is outlined in black. The cell is selected such that (001)bcc′_{\mathrm{bcc}}^{\prime} corresponds to (110)bcc_{\mathrm{bcc}} and evolves continuously into the (0001)hcp_{\mathrm{hcp}} basal plane.

(001)bcc′∥(110)bcc∥(0001)hcp,(43)(001)_{\mathrm{bcc}}^{\prime} \parallel(110)_{\mathrm{bcc}} \parallel(0001)_{\mathrm{hcp}}, \tag*{(43)}

where bcc′^{\prime} denotes the reoriented representation used in the NEB calculations. This change does not modify the physical bcc structure, but provides a common two-atom periodicity for the two endpoints and establishes a one-to-one correspondence between their atomic positions. The transformation follows a Burgers-type mechanism in which a lattice deformation converts the (110)bcc(110)_{\mathrm{bcc}} planes into hexagonal nets, while alternating planes undergo a shuffle to generate the hcp stacking sequence[38, 39].

At zero pressure, the volume of the two-atom cell decreases from 22.707 Å3^3 in bcc-FM to 20.399 Å3^3 in hcp-NM, corresponding to a volume reduction of approximately 10.2%. The magnitude of the Fe local magnetic moment decreases from approximately 2.26 μB\mu_B in bcc-FM to zero in hcp-NM.

The transition pathway was represented by 15 images, including the two fixed endpoints. To examine the influence of the magnetic initialization, two different initial pathways were considered. In the first, the two Fe moments were kept parallel in the initial band, while their magnitudes were interpolated linearly from the bcc-FM value to zero at the hcp-NM endpoint. This FM-like initialization resulted in a higher activation barrier and did not reveal an intermediate metastable configuration. In the second, AFM-assisted initialization, the relative orientation of the two Fe moments was varied smoothly from parallel toward antiparallel alignment. A perpendicular intermediate direction was specified to define the rotation plane, and the intermediate orientations were generated using spherical linear interpolation. Concurrently, the magnitudes of both moments were progressively reduced to zero toward the hcp endpoint. Starting from this band, the optimization converged to the lower-barrier pathway.

Under an imposed hydrostatic pressure, the relevant thermodynamic potential is the enthalpy,

H=E+PΩ.(44)\mathcal{H} = E + P\Omega. \tag*{(44)}

The lower-barrier pathways obtained at 0, 20, and 30 GPa are shown in Figure 4, where the enthalpy of each image is reported relative to the bcc-FM endpoint at the same pressure. At zero pressure, the path exhibits a single maximum with an activation barrier of approximately 154 meV/atom. Increasing pressure stabilizes the hcp phase relative to bcc and lowers the activation barrier to approximately 117 meV/atom at 20 GPa and 103 meV/atom at 30 GPa. No intermediate metastable minimum is observed along these direct bcc-to-hcp pathways.

Top and side views of the bcc-FM structure, the zero-pressure transition-state configuration, the hcp-NM structure, and enthalpy pathways

Figure 4. Top and side views of (a) the bcc-FM structure, (b) the zero-pressure transition-state configuration, and (c) the hcp-NM structure. The gold and gray spheres both represent Fe atoms, with the different colors used to distinguish the two alternating atomic planes. During the transformation, the gray Fe atom moves from the center of the (001)bcc_{\mathrm{bcc}} plane, with reduced coordinates (1/2, 1/2, 1/2), toward the hcp position (1/3, 2/3, 1/2), or its symmetry-equivalent position, through the atomic shuffle. The arrows indicate the local Fe magnetic moments. The in-plane lattice angles are indicated in panels (a)-(c), while the corresponding c/ac/a ratios are 1.633, 1.645, and 1.583, respectively. Panel (d) shows the enthalpy relative to the bcc-FM endpoint along the 15-image generalized NEB pathways at 0, 20, and 30 GPa. The corresponding activation barriers are 154, 117, and 103 meV/atom, respectively.

The evolution of the magnetic degrees of freedom along the converged zero-pressure lower-barrier pathway is resolved in Figure 5. No symmetry constraint was imposed to enforce equal magnitudes of the two Fe magnetic moments. Nevertheless, ∣M1∣\lvert M_1\rvert and ∣M2∣\lvert M_2\rvert remain nearly equal along the optimized pathway and decrease from approximately 2.25 μB2.25\,\mu\mathrm{B} at the bcc endpoint to zero at the hcp endpoint. At the same time, their relative angle increases continuously from 0∘0^\circ, passes close to 90∘90^\circ at the seventh image, reaches approximately 103∘103^\circ at the transition state, and approaches 180∘180^\circ before the moments collapse. Thus, the magnetic part of the pathway involves both a progressive rotation and a reduction of the local-moment magnitudes, rather than a purely longitudinal collapse.

Figure 5. Evolution of the magnetic degrees of freedom along the zero-pressure lower-barrier bcc-to-hcp pathway. The upper panel shows the magnitudes ∣M1∣|\mathbf{M}_1| and ∣M2∣|\mathbf{M}_2| of the two Fe magnetic-moment vectors, and the lower panel shows their relative angle θ\theta. The horizontal dashed and dotted lines indicate 90∘90^\circ and 180∘180^\circ, respectively, and the vertical dashed line marks the transition state at image 8. The angle is not reported for the final hcp-NM image because both moments vanish and their relative orientation is therefore undefined. Lines connecting the images are guides to the eye.

The structural evolution along the zero-pressure pathway is quantified in Figure 6. In addition to the monotonic volume reduction, the in-plane lattice angle γ\gamma increases smoothly from 109.44∘109.44^\circ in bcc to 120∘120^\circ in hcp. The axial ratio c/ac/a shows a weak nonmonotonic variation: it increases from 1.633 at the bcc endpoint to a maximum of approximately 1.656 near the sixth image and then decreases to 1.583 in hcp. At the transition state (image 8), the volume is 10.85110.851 Å3^3/atom, γ=114.38∘\gamma= 114.38^\circ, and c/a=1.645c/a = 1.645.

Figure 6. Structural evolution along the zero-pressure lower-barrier bcc-to-hcp pathway. (a) Volume per atom, (b) in-plane lattice angle γ\gamma, (c) axial ratio c/ac/a, and (d) normalized atomic-shuffle coordinate ss. For the slightly distorted intermediate cells, aa in panel (c) is defined as the mean length of the first two in-plane lattice vectors. The shuffle coordinate is the projection of the second Fe reduced coordinate onto the segment connecting (1/2,1/2,1/2)(1/2,1/2,1/2) in bcc to (1/3,2/3,1/2)(1/3,2/3,1/2) in hcp, so that s=0s=0 for bcc and s=1s=1 for hcp. The vertical dashed line marks the zero-pressure transition state at image 8. Lines connecting the images are guides to the eye.

To separate the internal atomic displacement from the homogeneous cell deformation, we define a normalized shuffle coordinate by projecting the reduced coordinate ui_i of the second Fe atom onto the line joining its bcc and

hcp endpoint positions,

si=(ui−ubcc)⋅(uhcp−ubcc)∣uhcp−ubcc∣2,(45)s_i = \frac{(\mathbf{u}_i-\mathbf{u}_{\mathrm{bcc}})\cdot(\mathbf{u}_{\mathrm{hcp}}-\mathbf{u}_{\mathrm{bcc}})}{|\mathbf{u}_{\mathrm{hcp}}-\mathbf{u}_{\mathrm{bcc}}|^2}, \tag*{(45)}

where ubcc=(1/2,1/2,1/2)\mathbf{u}_{\mathrm{bcc}}=(1/2,1/2,1/2) and uhcp=(1/3,2/3,1/2)\mathbf{u}_{\mathrm{hcp}}=(1/3,2/3,1/2). With this definition, s=0s=0 and s=1s=1 correspond to the bcc and hcp endpoints, respectively. The shuffle remains small in the first part of the pathway, increases rapidly through the central images, and reaches s=0.727s=0.727 at the transition state. The correlated changes in γ\gamma, c/ac/a, and ss demonstrate that the shear, axial distortion, and atomic shuffle are strongly coupled along the transformation pathway.

The bcc-to-hcp transformation therefore cannot be described solely as a lattice deformation. It combines a substantial volume reduction, coupled cell deformation and atomic shuffle, and the progressive rotation and collapse of the local magnetic moments. The lower barrier reached from the AFM-assisted initialization further highlights the importance of treating magnetic degrees of freedom explicitly when determining transition pathways in iron.

Spin reversal in monolayer CrI3

As a third application, we consider the collective magnetization reversal in a ferromagnetic monolayer of CrI3_3. In the initial state, the Cr magnetic moments are aligned along the out-of-plane +z+z direction, while the small induced moments on the I atoms are oriented antiparallel to them. In the final state, all these moments are reversed. Spin-orbit coupling is included in the calculations, giving rise to a finite magnetic anisotropy barrier between the two magnetic states. The transition pathway is represented by 13 images, including the initial and final states. In contrast to the FeRh and Fe transformations discussed above, the spin-reversal process in CrI3_3 induces only negligible atomic and lattice relaxations. Nevertheless, the coupled magnetostructural framework is retained throughout the pathway optimization, allowing structural and magnetic degrees of freedom to evolve self-consistently. This example therefore probes the limit in which the transition is governed predominantly by magnetic degrees of freedom. The pathway optimization was performed using an energy-convergence threshold of 2.0×10−62.0 \times10^{-6} Ha.

The calculated minimum-energy path is shown in Figure 7. The energy profile exhibits the expected symmetric behavior, with a maximum at the midpoint of the pathway. At this midpoint, corresponding to the seventh image, the Cr moments are oriented along the in-plane +y+y direction, while the induced I moments remain antiparallel and point along −y-y. The magnitudes of the local moments remain nearly constant during the reversal, with values of approximately 2.6822.682 μB\mu_{\mathrm{B}} on Cr and 0.03050.0305 μB\mu_{\mathrm{B}} on I. The resulting magnetic anisotropy barrier is approximately 0.20.2 meV/atom, representing the energy required for coherent rotation of the magnetization from the out-of-plane easy axis toward the in-plane direction.

Collective spin reversal in monolayer CrI3

Figure 7. Collective spin reversal in monolayer CrI3_3. Top: crystal structure viewed along the out-of-plane direction. Bottom: minimum-energy path obtained from cDFT-NEB calculations using 13 images for the transition between ferromagnetic states with the Cr magnetization oriented along +z+z and −z-z. At the midpoint of the pathway, the magnetic moments are oriented in plane. Representative magnetic configurations are shown for selected images.

This example demonstrates that the cDFT-NEB formalism is capable of describing not only strongly coupled magnetostructural transformations, such as those encountered in FeRh and Fe, but also transitions in which magnetic degrees of freedom dominate the energetics. The method therefore provides a unified first-principles framework for investigating both magnetostructural and magnetic transition pathways in a wide range of materials.

Conclusion

We have developed a generalized first-principles transition-path framework that combines potential-based constrained density functional theory with a magnetostructural extension of the nudged elastic band method. The resulting cDFT-NEB formalism enables the simultaneous optimization of atomic coordinates, lattice degrees of freedom, and noncollinear magnetic moments within a unified variational framework. By consistently incorporating atomic forces, stresses, and spin torques, the method provides access to minimum-energy pathways and transition states in systems with coupled structural and magnetic degrees of freedom.

The implementation has been integrated into the ABINIT package and supports fixed- and variable-cell calculations for both collinear and noncollinear magnetic configurations while maintaining full electronic self-consistency throughout the pathway optimization. Applications to FeRh, elemental iron, and monolayer CrI3_3 demonstrate the versatility of the method. In FeRh, fixed- and variable-cell calculations yield similar activation barriers, indicating a modest effect of cell relaxation on the transition pathway. In iron, the framework captures the simultaneous lattice deformation, atomic shuffle, volume reduction, and collapse of the magnetic moments during the bcc-FM to hcp-NM transformation, as well as the pressure-induced reduction of the activation enthalpy. In CrI3_3, it enables the direct determination of the noncollinear spin-reversal pathway and the associated magnetic anisotropy barrier in the presence of spin-orbit coupling.

The present framework provides a general first-principles platform for investigating activated magnetostructural processes. It can be applied to pressure-induced magnetic transitions, polymorphous transformations, noncollinear magnetic materials, and other systems governed by coupled structural and magnetic energy landscapes.

Acknowledgments

This work is part of the SHAPEme project (EOS ID 400077525) that has received funding from the FWO and F.R.S.-FNRS under the Excellence of Science (EOS) programme funded by the Federal government of Belgium. This work was supported by the Fonds de la Recherche Scientifique - F.R.S.-FNRS under Grant number J.0091.26 (CDR - ABINIT on GPUs : Equal footing approach). We thank Denis Sabani for suggesting the extension of the NEB method to include magnetic degrees of freedom.

References

References

  1. [1]Q. Liu, W. Su, Y. Gu, X. Zhang, X. Xia, L. Wang, K. Xiao, N. Zhang, X. Cui, M. Huang, et al., Nature Communications 16, 2025), ISSN 2041-1723, URL http://dx.doi.org/10.1038/s41467-025-59561-0.
  2. [2]T.-H. Yang, S. Okamoto, D. A. Tennant, M. A. McGuire, and Q. Zhang, Nature Communications 17 (2026), ISSN 2041-1723, URL http://dx.doi.org/10.1038/s41467-026-71683-7.
  3. [3]Z. He, S. P. Poudel, S. Stolz, T. Wang, A. Rossi, F. Wang, S.-K. Mo, A. Weber-Bargioni, Z. Q. Qiu, S. Barraza-Lopez, et al., Nano Letters 24, 8535–8541 (2024), ISSN 1530-6992, URL http://dx.doi.org/10.1021/acs.nanolett.4c01286.
  4. [4]T. Li, S. Jiang, N. Sivadas, Z. Wang, Y. Xu, D. Weber, J. E. Goldberger, K. Watanabe, T. Taniguchi, C. J. Fennie, et al., Nature Materials 18, 1303–1308 (2019), ISSN 1476-4660, URL http://dx.doi.org/10.1038/s41563-019-0506-1.
  5. [5]H. Han, H. Lin, W. Gan, R. Xiao, Y. Liu, J. Ye, L. Chen, W. Wang, L. Zhang, C. Zhang, et al., Physical Review B 107, 075423 (2023), URL https://link.aps.org/doi/10.1103/PhysRevB.107.075423.
  6. [6]L. Lewis, R. Barua, and B. Lejeune, Journal of Alloys and Compounds 650, 482–488 (2015), ISSN 0925-8388, URL http://dx.doi.org/10.1016/j.jallcom.2015.07.255.
  7. [7]G. Henkelman, B. P. Uberuaga, and H. Jónsson, The Journal of Chemical Physics 113, 9901 (2000).
  8. [8]G. Henkelman and H. Jónsson, The Journal of Chemical Physics 113, 9978 (2000).
  9. [9]D. Sheppard, P. Xiao, W. Chemelewski, D. D. Johnson, and G. Henkelman, The Journal of Chemical Physics 136, 074103 (2012).
  10. [10]N. A. Zarkevich and D. D. Johnson, The Journal of Chemical Physics 143 (2015), ISSN 1089-7690, URL http://dx.doi.org/10.1063/1.4927778.
  11. [11]L. Gao, X. Ding, T. Lookman, J. Sun, and E. K. H. Salje, Applied Physics Letters 109, 031912 (2016).
  12. [12]A. Fantasia, F. Rovaris, O. Abou El Kheir, A. Marzegalli, D. Lanzoni, L. Pessina, P. Xiao, C. Zhou, L. Li, G. Henkelman, et al., The Journal of Chemical Physics 161, 014110 (2024).
  13. [13]P. F. Bessarab, V. M. Uzdin, and H. Jónsson, Computer Physics Communications 196, 335–347 (2015), ISSN 0010-4655, URL http://dx.doi.org/10.1016/j.cpc.2015.07.001.
  14. [14]G. P. Müller, M. Hoffmann, D. Dießelkamp, D. Schürhoff, S. Mavros, M. Sallermann, N. S. Kiselev, H. Jónsson, and S. Blügel, Physical Review B 99, 224414 (2019).
  15. [15]H. Schrautzer, M. Sallermann, P. F. Bessarab, and H. Jónsson, Physical Review B 112, 104433 (2025), URL https://link.aps.org/doi/10.1103/z673-hhnp.
  16. [16]P. H. Dederichs, S. Blügel, R. Zeller, and H. Akai, Physical Review Letters 53, 2512 (1984).
  17. [17]P. Kurz, F. Förster, L. Nordström, G. Bihlmayer, and S. Blügel, Physical Review B 69, 024415 (2004).
  18. [18]Q. Wu and T. Van Voorhis, Physical Review A 72, 024502 (2005).
  19. [19]P.-W. Ma and S. L. Dudarev, Physical Review B 91, 054420 (2015).
  20. [20]Z. Cai, K. Wang, Y. Xu, S.-H. Wei, and B. Xu, Quantum Frontiers 2 (2023), ISSN 2731-6106, URL http://dx.doi.org/10.1007/s44214-023-00041-1.
  21. [21]X. Gonze, B. Seddon, J. A. Elliott, C. Tantardini, and A. V. Shapeev, Journal of Chemical Theory and Computation 18, 6099–6110 (2022), ISSN 1549-9626, URL http://dx.doi.org/10.1021/acs.jctc.2c00673.
  22. [22]R. A. Lawrence, S. J. Donaldson, and M. I. J. Probert, Magnetochemistry 9, 42 (2023), ISSN 2312-7481, URL http://dx.doi.org/10.3390/magnetochemistry9020042.
  23. [23]W. Tang, Y. Li, and B. Xu, npj Computational Materials 11 (2025), ISSN 2057-3960, URL http://dx.doi.org/10.1038/s41524-025-01603-8.
  24. [25]J. Nocedal and S. J. Wright, Numerical Optimization (Springer, New York, 2006), 2nd ed.
  25. [26]D. Sheppard, R. Terrell, and G. Henkelman, The Journal of Chemical Physics 128, 134106 (2008).
  26. [27]X. Gonze, B. Amadon, G. Antonius, F. Arnardi, L. Baguet, J.-M. Beuken, J. Bieder, F. Bottin, J. Bouchet, E. Bousquet, et al., Computer Physics Communications 248, 107042 (2020).
  27. [28]A. H. Romero, D. C. Allan, B. Amadon, G. Antonius, T. Appencourt, L. Baguet, J. Bieder, F. Bottin, J. Bouchet, E. Bousquet, et al., Journal of Chemical Physics 152, 124102 (2020).
  28. [29]M. J. Verstraete, J. Abreu, G. E. Allemand, B. Amadon, G. Antonius, M. Azizi, L. Baguet, C. Barat, L. Bastogne, R. Béjaud, et al., The Journal of Chemical Physics 163 (2025), ISSN 1089-7690, URL http://dx.doi.org/10.1063/5.0288278.
  29. [30]M. Torrent, F. Jollet, F. Bottin, G. Zérah, and X. Gonze, Comput. Materials Science 42, 337 (2008).
  30. [31]M. van Setten, M. Giantomassi, E. Bousquet, M. Verstraete, D. Hamann, X. Gonze, and G.-M. Rignanese, Computer Physics Communications 226, 39 (2018).
  31. [32]J. P. Perdew, K. Burke, and M. Ernzerhof, Physical Review Letters 77, 3865 (1996).
  32. [33]J. P. Perdew and Y. Wang, Physical Review B 45, 13244 (1992).
  33. [34]J. S. Kouvel and C. C. Hartelius, Journal of Applied Physics 33, 1343 (1962).
  34. [35]G. Shirane, R. Nathans, and C. W. Chen, Physical Review 134, A1547 (1964).
  35. [36]M. P. Annaorazov, K. A. Asatryan, G. Myalikgulyev, S. A. Nikitin, A. M. Tishin, and A. L. Tyurin, Cryogenics 32, 867 (1992).
  36. [37]L. H. Lewis, C. H. Marrows, and S. Langridge, Journal of Physics D: Applied Physics 49, 323002 (2016).
  37. [38]J. B. Liu and D. D. Johnson, Phys. Rev. B 79, 134113 (2009).
  38. [39]E. Boulard, C. Denoual, A. Dewaele, A. King, Y. L. Godec, and N. Guignot, Acta Mater. 192, 30 (2020).

Paper details

vlnc:W_z2x4h6gv
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY-SA 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents