ALIGNN-FF Dynamics

Universal ML force field for atomistic simulations. Three tabs: (1) Geometry Optimization — relax structures with FIRE/BFGS/LBFGS, (2) Molecular Dynamics — NVE simulations with Maxwell-Boltzmann initialization, (3) Phonon Bandstructure — finite-displacement phonon calculations via phonopy. All powered by ALIGNN-FF (AlignnAtomwiseCalculator).

Open App

---

Overview

The ALIGNN-FF app provides a complete atomistic simulation toolkit in the browser. The force field is a pretrained ALIGNN model that predicts energy, forces, and stress for any crystal structure. The app supports geometry optimization with trajectory playback, molecular dynamics with energy/temperature monitoring, and phonon bandstructure calculation via phonopy with interactive Plotly plots.

Data Source

ALIGNN-FF — AlignnAtomwiseCalculator from alignn.ff.ff (default or wt01 model). phonopy — for phonon dispersion via finite displacement method.

Endpoints

GET /alignn_ff/query — Single-point energy (API key)

Compute energy, forces, and stress for a structure without relaxation.

curl "https://atomgpt.org/alignn_ff/query?poscar=Si%0A1.0%0A0%202.734%202.734%0A2.734%200%202.734%0A2.734%202.734%200%0ASi%0A2%0Adirect%0A0%200%200%0A0.25%200.25%200.25&APIKEY=sk-XYZ" \
  -H "accept: application/json"

Returns energy_eV, forces_eV_per_A (Nx3 array), stress (6-component Voigt), natoms.

---

GET /alignn_ff/relax — Quick relaxation (plain text POSCAR)

Relax a structure and return the optimized POSCAR as plain text.

curl "https://atomgpt.org/alignn_ff/relax?poscar=Si%0A1.0%0A5.0%200%200%0A0%205.0%200%0A0%200%205.0%0ASi%0A8%0Adirect%0A0.25%200.75%200.25%0A0%200%200.5%0A0.25%200.25%200.75%0A0%200.5%200%0A0.75%200.75%200.75%0A0.5%200%200%0A0.75%200.25%200.25%0A0.5%200.5%200.5&fmax=0.05&steps=50&APIKEY=sk-XYZ"

Param	Default	Description
poscar	required	URL-encoded POSCAR
fmax	0.05	Force convergence tolerance (eV/Å)
steps	50	Max relaxation steps

Returns plain text relaxed POSCAR.

---

POST /alignn_ff/optimize — Full optimization with trajectory

Geometry optimization with full trajectory data for visualization and playback.

curl -X POST "https://atomgpt.org/alignn_ff/optimize" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "Content-Type: application/json" \
  -H "accept: application/json" \
  -d '{
    "poscar": "Si\n1.0\n5.0 0 0\n0 5.0 0\n0 0 5.0\nSi\n8\ndirect\n0.25 0.75 0.25\n0 0 0.5\n0.25 0.25 0.75\n0 0.5 0\n0.75 0.75 0.75\n0.5 0 0\n0.75 0.25 0.25\n0.5 0.5 0.5",
    "fmax": 0.05,
    "steps": 100,
    "optimizer": "FIRE",
    "relax_cell": true
  }'

Field	Type	Default	Description
poscar	string	required	VASP POSCAR structure (max 100 atoms)
fmax	float	0.05	Force convergence (eV/Å)
steps	int	200	Max optimization steps
optimizer	string	"FIRE"	"FIRE", "BFGS", or "LBFGS"
relax_cell	bool	true	Also optimize cell parameters (via ExpCellFilter)

Response includes converged, final_poscar, initial_energy, final_energy, energy_change, steps_taken, energies (per-step), forces_max (per-step), trajectory (positions/cell every 5 steps for playback), computation_time.

---

POST /alignn_ff/md — Molecular dynamics

NVE molecular dynamics with Maxwell-Boltzmann velocity initialization and VelocityVerlet integrator.

curl -X POST "https://atomgpt.org/alignn_ff/md" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "Content-Type: application/json" \
  -H "accept: application/json" \
  -d '{
    "poscar": "Si\n1.0\n5.494 0 0\n0 5.494 0\n0 0 5.494\nSi\n8\ndirect\n0.25 0.75 0.25\n0 0 0.5\n0.25 0.25 0.75\n0 0.5 0\n0.75 0.75 0.75\n0.5 0 0\n0.75 0.25 0.25\n0.5 0.5 0.5",
    "temperature": 300.0,
    "timestep": 0.5,
    "steps": 50,
    "interval": 5
  }'

Field	Type	Default	Description
poscar	string	required	POSCAR structure (max 50 atoms)
temperature	float	300.0	Initial temperature (K)
timestep	float	0.5	Integration timestep (fs)
steps	int	50	Number of MD steps
interval	int	5	Save trajectory every N steps

Response includes trajectory (positions, velocities, T, KE, PE per saved frame), energies (total/kinetic/potential arrays), temperatures, average_temperature, final_temperature, computation_time.

---

POST /alignn_ff/phonons — Phonon bandstructure

Relax the structure, then compute phonon dispersion via phonopy finite displacements.

curl -X POST "https://atomgpt.org/alignn_ff/phonons" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "Content-Type: application/json" \
  -H "accept: application/json" \
  -d '{
    "poscar": "Si\n1.0\n5.494 0 0\n0 5.494 0\n0 0 5.494\nSi\n8\ndirect\n0.25 0.75 0.25\n0 0 0.5\n0.25 0.25 0.75\n0 0.5 0\n0.75 0.75 0.75\n0.5 0 0\n0.75 0.25 0.25\n0.5 0.5 0.5",
    "fmax": 0.05,
    "relax_steps": 50,
    "force_multiplier": 1.9,
    "dim": [2, 2, 2],
    "relax_cell": true
  }'

Field	Type	Default	Description
poscar	string	required	POSCAR structure (max 50 atoms)
fmax	float	0.05	Force convergence for pre-relaxation (eV/Å)
relax_steps	int	50	Max relaxation steps
force_multiplier	float	1.9	Force scaling for phonopy
dim	list[int]	[2,2,2]	Supercell dimensions for phonon calculation
relax_cell	bool	true	Relax cell before phonon calc

Response includes relaxed_poscar, primitive_poscar, band_data (distances + frequencies for interactive Plotly), plot_base64 (PNG fallback), supercell_dim, computation_time.

---

Python Examples

Optimize structureMolecular dynamicsPhonon bandstructureSingle-point energy

import requests

response = requests.post(
    "https://atomgpt.org/alignn_ff/optimize",
    headers={
        "Authorization": "Bearer sk-XYZ",
        "accept": "application/json",
        "Content-Type": "application/json",
    },
    json={
        "poscar": open("POSCAR").read(),
        "fmax": 0.05,
        "steps": 100,
        "optimizer": "FIRE",
        "relax_cell": True,
    },
)
data = response.json()
print(f"Converged: {data['converged']}")
print(f"E: {data['initial_energy']:.4f} → {data['final_energy']:.4f} eV")
print(f"Steps: {data['steps_taken']}, Time: {data['computation_time']}s")
with open("CONTCAR", "w") as f:
    f.write(data["final_poscar"])

import requests

response = requests.post(
    "https://atomgpt.org/alignn_ff/md",
    headers={
        "Authorization": "Bearer sk-XYZ",
        "accept": "application/json",
        "Content-Type": "application/json",
    },
    json={
        "poscar": open("POSCAR").read(),
        "temperature": 300.0,
        "timestep": 0.5,
        "steps": 100,
        "interval": 10,
    },
)
data = response.json()
print(f"Avg T: {data['average_temperature']:.1f} K")
print(f"Frames: {len(data['trajectory'])}")

import requests

response = requests.post(
    "https://atomgpt.org/alignn_ff/phonons",
    headers={
        "Authorization": "Bearer sk-XYZ",
        "accept": "application/json",
        "Content-Type": "application/json",
    },
    json={
        "poscar": open("POSCAR").read(),
        "dim": [2, 2, 2],
        "force_multiplier": 1.9,
    },
)
data = response.json()
print(f"Primitive atoms: {data['num_atoms_prim']}")
print(f"Supercell: {data['supercell_dim']}")
print(f"Time: {data['computation_time']}s")

# Save phonon plot
if data.get("plot_base64"):
    import base64
    with open("phonon_bands.png", "wb") as f:
        f.write(base64.b64decode(data["plot_base64"]))

import requests

response = requests.get(
    "https://atomgpt.org/alignn_ff/query",
    headers={"accept": "application/json"},
    params={
        "poscar": open("POSCAR").read(),
        "APIKEY": "sk-XYZ",
    },
)
data = response.json()
print(f"Energy: {data['energy_eV']:.4f} eV")
print(f"Max force: {max(max(abs(f) for f in row) for row in data['forces_eV_per_A']):.4f} eV/Å")

AGAPI Agent

from agapi.agents import AGAPIAgent
import os

agent = AGAPIAgent(api_key=os.environ.get("AGAPI_KEY"))

# Relax structure
response = agent.query_sync("Relax the structure of JVASP-1002 using ALIGNN-FF")
print(response)

# Run MD
response = agent.query_sync("Run molecular dynamics on Si at 500K for 100 steps")
print(response)

# Phonons
response = agent.query_sync("Compute the phonon bandstructure of Si using ALIGNN-FF")
print(response)

References

• K. Choudhary, Digital Discovery 2(2), 346 (2023) — ALIGNN-FF DOI
• K. Choudhary, B. DeCost, npj Comp. Mat. 7, 185 (2021) — ALIGNN DOI
• atomgptlab/alignn