Calculate a Reaction Path with NEB
On this page
Use this procedure when reactant and product structures are already prepared and you want to connect them with a nudged elastic band (NEB). For the theory, algorithms, and interpretation of a band, read Nudged Elastic Band Method.
Before opening the calculation
Prepare two structure tabs: one for the initial state and one for the final state. They must contain the same atoms in the same order. For a periodic calculation, verify manually that both structures use the same cell and consistently wrapped coordinates; the endpoint validator does not compare cells.
NEB interpolates atom i in the initial structure to atom i in the final structure, so matching formulas are not sufficient: the atom ordering must correspond. A wrong correspondence creates a mathematically valid but chemically meaningless path.
Use one calculator and one set of calculator parameters for the entire band. If the endpoints were optimized, optimize both with that same model. See Choosing a Method when the calculator has not yet been selected.
Keep the default BFGS reaction optimizer once the band is running. FIRE can converge a climbing-image band to a qualitatively wrong path — a well instead of a barrier — while still reporting neb.converged: true.
Avoid pre-optimizing both endpoints for weakly-bound or degenerate-exchange systems: independent relaxation can collapse the reactant and product into the same minimum. When the mechanism needs a constrained coordinate, apply atomic constraints (see Apply Atomic Constraints) to freeze anchor atoms before building the band.
Open Reaction Path / TS
In the calculation tool rail, select Reaction Path / TS. It is near the bottom of the rail.
The Reaction Path / TS dialog opens without starting a calculation.
Choose the workflow depth
The first selector controls how far Tako carries the workflow:
| Workflow | What Tako does | Use it when |
|---|---|---|
| Interpolate only | Builds the initial band but does not evaluate the calculator | Checking atom correspondence and close contacts |
| NEB (minimum-energy path) | Relaxes an ordinary elastic band | Establishing a minimum-energy path before saddle refinement |
| CI-NEB (barrier + saddle) | Relaxes the band and drives its highest-energy image toward a saddle | Estimating the barrier and preparing a transition-state candidate |
| CI-NEB → Dimer refine (true saddle) | Continues from the band maximum with dimer refinement | Refining the saddle without forming a Hessian |
| Transition state (dimer + vibration check) | Runs the configured band, saddle refinement, and verification stages | A complete transition-state search from two endpoints |
For a new pair of endpoints we suggest starting with Interpolate only. The optimizer fields and Advanced NEB section are hidden in this workflow. Move to CI-NEB (barrier + saddle) only after inspecting the seed band; that selection displays the optimizer and advanced controls.
Select the endpoints
Under Images:
- Set First image to the reactant tab.
- Set Final image to the product tab.
- Set Total images to the number of band images, including both fixed endpoints.
Seven images are a reasonable first band for a single, uncomplicated rearrangement. More images improve spatial resolution but add one force evaluation per interior image at every optimization step. They do not repair incorrect atom mapping or an implausible mechanism.
Tako rejects endpoints with different atom counts or different elements at corresponding indices. That validation cannot detect two atoms of the same element that have been exchanged.
Inspect the initial band
For the first pass, choose:
| Control | Initial value |
|---|---|
| Workflow | Interpolate only |
| Interpolation | IDPP when bonds form or break; otherwise linear is sufficient for inspection |
| Total images | 7 |
Press Start. The dialog closes, an NEB-<N>-calc calculation workspace is created in the Explorer, and its status becomes running. After successful completion, expand the workspace’s output folder and open reaction_path.traj. Step through every image and check that:
- atoms follow the intended correspondence;
- no atom passes through another atom;
- periodic atoms take the short displacement across the cell;
- bonds change in the intended region rather than through global fragmentation;
- fixed atoms occupy the same intended coordinates in both endpoints.
Correct the endpoints when any of these checks fail. Increasing the optimizer step budget does not correct a bad seed path.
Configure the band optimization
Reopen Reaction Path / TS and select CI-NEB (barrier + saddle) for a barrier calculation. The configuration we start from is shown below; IDPP differs from Tako’s registered linear interpolation default:
| Control | Recommended first value | Meaning |
|---|---|---|
| Interpolation | IDPP | Reduces severe pair-distance distortions in the seed band |
| Optimizer | BFGS | Relaxes the interior images using accumulated curvature information |
| NEB method | Improved tangent | Selects the path tangent from neighboring image energies and positions |
| Max NEB steps | 150 | Climbing-stage budget after the separate pre-climb stage |
| Spring constant | 0.1 | Controls image distribution along the path |
| Force convergence | 0.05 eV Å(^{-1}) | Convergence threshold on the NEB residual |
| Max step A | 0.05 Å | Largest geometry displacement in one optimizer step |
For every exposed optimizer, climbing with a positive Preclimb steps value is staged as two optimizer runs. Tako first permits up to that many ordinary-NEB steps, then starts a fresh climbing run with up to Max NEB steps. With the defaults, the total can therefore reach 30 + 150 steps. The result’s neb.steps field counts only the final run. Ordinary NEB uses Max NEB steps directly and does not use the pre-climb budget.
Advanced controls and their exact defaults are documented in Reaction-path settings. Change them to address an observed numerical problem.
Run and monitor the band
Press Start. The dialog closes, an NEB-<N>-calc workspace appears in the Explorer, and the Calc panel reports its running state. Tako shows the broad running, completed, failed, or cancelled state and writes detailed NEB data only when the operation returns successfully. The current interface does not expose a live per-image residual, highest-image trace, or restartable partial band.
Cancelling terminates the calculation worker. Pending NEB artifacts are marked failed, and the partial optimizer state cannot be resumed. Any artifact that had already become available before cancellation remains in the workspace, although a normal NEB run generally publishes its path artifacts only on successful return.
Inspect the result
After successful completion, expand NEB-<N>-calc/output in the Explorer. Open reaction_path.traj to step through the optimized images; frame provenance and how to extract a structure from an image are covered generically in Inspect Trajectories. Open reaction_barrier.json for the image-energy scalars. A trustworthy numerical result requires all of the following:
neb.convergedistrue.- The final residual is below the requested threshold.
- Neighboring images form a continuous mechanism.
- The band does not cross itself or exchange equivalent atoms unexpectedly.
- The energy maximum is spatially resolved rather than hidden between widely separated images.
- The initial and final images still represent the intended states.
reaction_barrier.json reports energies relative to the first endpoint. Therefore
These are electronic-energy differences under the selected calculator. They are not activation free energies unless thermal and entropic corrections have been calculated separately using a consistent protocol.
Continue to a transition state
A climbing image is a saddle-point candidate that still requires verification before it can be assigned as a transition state. Continue with Refine and Verify a Transition State, then require one imaginary vibrational mode whose displacement follows the intended reaction coordinate.
Problems specific to this procedure
| Observation | Diagnosis | Action |
|---|---|---|
| Interpolated atoms cross | Atom correspondence is wrong or the seed is geometrically poor | Repair the endpoint mapping; then compare linear and IDPP seeds |
| A periodic atom traverses the entire cell | Endpoint coordinates use different periodic images | Wrap both endpoints consistently before building the band |
| Both endpoints relax to the same structure | The intended states are not distinct minima under the chosen model or unconstrained relaxation collapses them | Reconsider the model, reaction definition, or physically justified endpoint constraints |
| Images bunch near one endpoint | The band is under-resolved or spacing dynamics are poor | Inspect the tangent and mechanism; then adjust image count or spring constant |
| The band converges to a well | The model or endpoints support a stable intermediate rather than the expected barrier | Inspect the structures and verify the potential-energy surface with another defensible method |
The profile looks smooth but neb.converged is false | A plausible curve has formed before force convergence | Treat it as a diagnostic only; rerun from a saved or reconstructed seed with revised settings |
For scripted execution, result-object fields, and reproducible output handling, use NEB with Tako Script.