Input Parameters

This page documents the RunInputs TOML schema accepted on the master branch. It is derived from neb_dynamics/inputs.py on this branch, so it reflects the actual keys and defaults the CLI accepts today.

Top-Level Layout

A typical inputs file looks like this:

engine_name = "chemcloud"
program = "xtb"
chemcloud_queue = "celery"
path_min_method = "NEB"

[path_min_inputs]
max_steps = 500
v = false

[chain_inputs]
k = 0.1
delta_k = 0.09
do_parallel = true
use_geodesic_interpolation = true
friction_optimal_gi = true
node_freezing = true
node_rms_thre = 5.0
node_ene_thre = 5.0
frozen_atom_indices = ""

[gi_inputs]
nimages = 10
friction = 0.001
nudge = 0.1
align = true

[gi_inputs.extra_kwds]

[program_kwds]
cmdline_args = []

[program_kwds.keywords]
threads = 1

[program_kwds.extras]

[program_kwds.files]

[program_kwds.model]
method = "GFN2xTB"
basis = "GFN2xTB"

[program_kwds.model.extras]

[optimizer_kwds]
name = "cg"
timestep = 0.5

Top-Level Keys

Key	Type	Default	Notes
engine_name	str	"chemcloud"	Supported: "qcop", "chemcloud", "ase"
program	str	"xtb"	Common values: "xtb", "crest", "terachem"; ase currently only supports "omol25"
chemcloud_queue	str \| null	null	Queue passed to QCOPEngine when using engine_name = "chemcloud"
path_min_method	str	"NEB"	Supported: "NEB", "FNEB", "MLPGI"
path_min_inputs	table	branch-dependent defaults	See below
chain_inputs	table	ChainInputs() defaults	See below
gi_inputs	table	GIInputs() defaults	See below
program_kwds	table	inferred from engine_name/program	Parsed into qcio.ProgramArgs for qcop / chemcloud
optimizer_kwds	table	{ name = "cg", timestep = 0.5 }	See optimizer names below

path_min_inputs for NEB

These keys are accepted when path_min_method = "NEB".

Key	Type	Default
climb	bool	false
en_thre	float	1e-4
rms_grad_thre	float	0.02
max_rms_grad_thre	float	0.05
skip_identical_graphs	bool	true
ts_grad_thre	float	0.05
ts_spring_thre	float	0.02
barrier_thre	float	0.1
early_stop_force_thre	float	0.0
negative_steps_thre	int	5
positive_steps_thre	int	10
use_geodesic_tangent	bool	false
do_elem_step_checks	bool	false
max_steps	float	500
v	bool	false

Example:

[path_min_inputs]
climb = false
en_thre = 0.0001
rms_grad_thre = 0.02
max_rms_grad_thre = 0.05
skip_identical_graphs = true
ts_grad_thre = 0.05
ts_spring_thre = 0.02
barrier_thre = 0.1
early_stop_force_thre = 0.03
negative_steps_thre = 5
positive_steps_thre = 10
use_geodesic_tangent = false
do_elem_step_checks = false
max_steps = 500
v = false

path_min_inputs for FNEB

When path_min_method = "FNEB", the branch provides these defaults:

Key	Default
max_min_iter	100
max_grow_iter	20
verbosity	1
skip_identical_graphs	true
do_elem_step_checks	true
grad_tol	0.05
barrier_thre	5
tangent	"geodesic"
tangent_alpha	1.0
use_xtb_grow	true
distance_metric	"GEODESIC"
min_images	10
todd_way	true
dist_err	0.1

Any provided values override these defaults.

path_min_inputs for MLPGI

MLPGI uses an empty default table on this branch. Keys are effectively backend-specific passthrough.

chain_inputs

Key	Type	Default	Notes
k	float	0.1	NEB spring constant
delta_k	float	0.09	Energy-weighted spring parameter
do_parallel	bool	true	Parallel gradient/energy computation
use_geodesic_interpolation	bool	true	Use geodesic interpolation for initial path generation
friction_optimal_gi	bool	true	Optimize GI friction parameter
node_freezing	bool	true	Freeze converged nodes
node_rms_thre	float	5.0	Node RMS threshold
node_ene_thre	float	5.0	Node energy threshold
frozen_atom_indices	str	""	Space-separated atom indices

Example:

[chain_inputs]
k = 0.1
delta_k = 0.09
do_parallel = true
use_geodesic_interpolation = true
friction_optimal_gi = true
node_freezing = true
node_rms_thre = 5.0
node_ene_thre = 5.0
frozen_atom_indices = ""

gi_inputs

Key	Type	Default
nimages	int	10
friction	float	0.001
nudge	float	0.1
align	bool	true
extra_kwds	table	{}

Example:

[gi_inputs]
nimages = 10
friction = 0.001
nudge = 0.1
align = true

[gi_inputs.extra_kwds]

program_kwds

For engine_name = "qcop" or "chemcloud", this section is parsed into a qcio.ProgramArgs object.

Structure

[program_kwds]
cmdline_args = []

[program_kwds.keywords]
threads = 1

[program_kwds.extras]

[program_kwds.files]

[program_kwds.model]
method = "GFN2xTB"
basis = "GFN2xTB"

[program_kwds.model.extras]

Defaults

If program_kwds is omitted, defaults depend on program:

• program = "xtb"
• if crest is available: method/basis gfn2, threads = 1, and program is internally changed to "crest"
• otherwise: method/basis GFN2xTB
• program containing "terachem"
• method ub3lyp
• basis 3-21g

optimizer_kwds

optimizer_kwds.name selects the optimizer implementation.

Supported names:

• cg
• conjugate_gradient
• vpo
• velocity_projected
• lbfgs
• adam
• amg
• adaptive_momentum
• fire

Defaults:

[optimizer_kwds]
name = "cg"
timestep = 0.5

Minimal Examples

ChemCloud xTB run

engine_name = "chemcloud"
program = "xtb"
chemcloud_queue = "celery"
path_min_method = "NEB"

[path_min_inputs]
max_steps = 250
v = true

[chain_inputs]
use_geodesic_interpolation = true

[gi_inputs]
nimages = 10

[optimizer_kwds]
name = "cg"
timestep = 0.5

Local QCOP TeraChem run

engine_name = "qcop"
program = "terachem"
path_min_method = "NEB"

[path_min_inputs]
max_steps = 100
v = true

[program_kwds]
cmdline_args = []

[program_kwds.keywords]
gpus = 1
sphericalbasis = false

[program_kwds.model]
method = "b3lyp"
basis = "6-31g**"

[optimizer_kwds]
name = "cg"
timestep = 0.5

Loading and Saving

from neb_dynamics.inputs import RunInputs

ri = RunInputs.open("inputs.toml")
ri.save("copied_inputs.toml")

CLI Notes

The CLI reads this file with:

mepd run --start start.xyz --end end.xyz --inputs inputs.toml

A few options are CLI flags, not TOML keys. For example:

• --recursive
• --use-tsopt
• --create-irc
• --minimize-ends

Those belong on the command line, not inside inputs.toml.