Nudged elastic band

Find minimum-energy transition paths and energy barriers between two structures using the nudged elastic band (NEB) method.

MLIP Arena provides two NEB tasks:

• NEB — takes a pre-built list of images (initial, intermediate, and final structures).
• NEB_FROM_ENDPOINTS — takes only the start and end structures, automatically relaxes them, interpolates intermediate images, then runs NEB.

Both tasks use ASE’s NEB implementation and support the climbing-image NEB (CI-NEB) algorithm.

The implementation is adapted from MatCalc.

NEB (from images)

Use this task when you already have a full image list, for example from a prior interpolation or a previous NEB run.

Function signature

from mlip_arena.tasks import NEB

result = NEB(
    images=images,
    calculator=calc,
    optimizer="MDMin",
    optimizer_kwargs=None,
    criterion=None,
    interpolation="idpp",
    climb=True,
    traj_file=None,
)

Parameters

body.images · list[ase.Atoms] · required

Ordered list of Atoms objects representing the NEB path. Must include at least the start and end images. Intermediate images will be interpolated in-place.

body.calculator · ase.calculators.calculator.BaseCalculator · required

Calculator assigned to every image. A shared calculator is used (allow_shared_calculator=True).

body.optimizer · Optimizer | str · default: "MDMin"

Optimizer for the NEB path. Accepts a class or one of: "MDMin", "FIRE", "FIRE2", "LBFGS", "LBFGSLineSearch", "BFGS", "QuasiNewton", "GPMin", "CellAwareBFGS", "ODE12r".

"BFGSLineSearch" is not supported for NEB calculations.

body.optimizer_kwargs · dict | None · default: "None"

Extra keyword arguments forwarded to the optimizer constructor.

body.criterion · dict | None · default: "{}"

Convergence criterion dict forwarded to optimizer.run() (e.g. {"fmax": 0.05}).

body.interpolation · string · default: "idpp"

Interpolation method for intermediate images. One of:

• "idpp" — Image Dependent Pair Potential (recommended, produces smoother paths)
• "linear" — simple linear interpolation of atomic positions

body.climb · boolean · default: "true"

Enable the climbing-image NEB algorithm. The image with the highest energy climbs toward the true saddle point.

body.traj_file · str | Path | None · default: "None"

Path to save the NEB trajectory. Passed directly to the optimizer.

Return value

Key	Type	Description
barrier	tuple	(forward_barrier, reverse_barrier) in eV from NEBTools.get_barrier()
images	list[ase.Atoms]	Final NEB images
forcefit	object	Spline fit object from ase.utils.forcecurve.fit_images, useful for plotting

---

NEB from endpoints

Use this task when you only have the start and end structures. It handles endpoint relaxation and image interpolation automatically.

Function signature

from mlip_arena.tasks import NEB_FROM_ENDPOINTS

result = NEB_FROM_ENDPOINTS(
    start=start_atoms,
    end=end_atoms,
    n_images=7,
    calculator=calc,
    optimizer="BFGS",
    optimizer_kwargs=None,
    criterion=None,
    relax_end_points=True,
    interpolation="idpp",
    climb=True,
    traj_file=None,
    cache_subtasks=False,
)

Parameters

body.start · ase.Atoms · required

Initial endpoint structure.

body.end · ase.Atoms · required

Final endpoint structure.

body.n_images · integer · required

Total number of images in the NEB path, including the start and end images.

body.calculator · ase.calculators.calculator.BaseCalculator · required

Calculator for energy and force evaluations.

body.optimizer · Optimizer | str · default: "BFGS"

Optimizer for endpoint relaxation and the NEB run. See NEB for valid values.

body.optimizer_kwargs · dict | None · default: "None"

Extra keyword arguments forwarded to the optimizer.

body.criterion · dict | None · default: "None"

Convergence criterion for both endpoint relaxations and the NEB run.

body.relax_end_points · boolean · default: "true"

If True, runs OPT on both start and end before interpolating.

body.interpolation · string · default: "idpp"

Interpolation method. One of "linear" or "idpp".

body.climb · boolean · default: "true"

Enable climbing-image NEB.

body.traj_file · str | Path | None · default: "None"

Path to save the NEB trajectory.

body.cache_subtasks · boolean · default: "false"

If True, results from internal OPT and NEB sub-tasks are cached in Prefect’s result store.

Return value

Same as NEB above.

---

Example

1. Define start and end structures

   from ase.build import fcc111, add_adsorbate
   from ase import Atoms
   
   # Example: H atom hopping between two hollow sites on Cu(111)
   from mlip_arena.tasks import NEB_FROM_ENDPOINTS
   from mlip_arena.tasks.utils import get_calculator
   from mlip_arena.models import MLIPEnum
   
   calc = get_calculator(MLIPEnum.MACE_MP)
   
2. Run NEB from endpoints

   result = NEB_FROM_ENDPOINTS(
       start=start_atoms,
       end=end_atoms,
       n_images=7,
       calculator=calc,
       relax_end_points=True,
       interpolation="idpp",
       climb=True,
       criterion={"fmax": 0.05},
       traj_file="neb.traj",
   )
   
3. Inspect barrier

   forward_barrier, reverse_barrier = result["barrier"]
   print(f"Forward barrier:  {forward_barrier:.3f} eV")
   print(f"Reverse barrier: {reverse_barrier:.3f} eV")
   

Interpolation methods

Method	Description
"linear"	Linearly interpolates Cartesian positions between start and end. Fast but may produce high-energy initial guesses.
"idpp"	Image Dependent Pair Potential interpolation. Minimizes changes in interatomic distances and typically yields smoother, more physical paths. Recommended for most cases.

For complex migration events (e.g. involving significant bond breaking), start with "idpp" and verify the path makes physical sense before converging the NEB.