Computational Chemistry GuideSAFE

🔬 A curated collection of 23,000+ agent skills for empirical research across 8 social science disciplines. | 精选 23,000+ AI Agent 技能库,覆盖8大社会科学学科的实证研究。CoPaper.AI 20分钟完成一篇可复现的规范实证论文,并支持用户上传 Skills。-- Maintained by CoPaper.AI from Stanford REAP.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: computational-chemistry-guide
description: "DFT, molecular simulation, and reaction prediction tools for chemists"
metadata:
  openclaw:
    emoji: "⚗"
    category: "domains"
    subcategory: "chemistry"
    keywords: ["computational chemistry", "DFT", "quantum chemistry", "reaction prediction"]
    source: "wentor-research-plugins"
---

# Computational Chemistry Guide

## Overview

Computational chemistry bridges quantum mechanics and practical chemistry, enabling researchers to predict molecular properties, reaction mechanisms, and material behaviors without stepping into a wet lab. From drug design to catalyst optimization, computational methods accelerate discovery by screening thousands of candidates before committing to synthesis.

This guide covers the major computational chemistry paradigms: Density Functional Theory (DFT) for electronic structure calculations, molecular dynamics (MD) for simulating atomic motion, machine learning potentials for scaling up simulations, and reaction prediction tools for retrosynthesis and mechanism elucidation. Each section includes tool recommendations, typical workflows, and code examples.

Whether you are a chemistry PhD student running your first Gaussian calculations, a materials scientist exploring new alloys with VASP, or a medicinal chemist using ML-based property prediction, this skill provides the conceptual framework and practical recipes to get productive quickly.

## Density Functional Theory (DFT)

### When to Use DFT

DFT is the workhorse of quantum chemistry. It provides a good balance of accuracy and computational cost for systems of up to a few hundred atoms.

| Property | DFT Suitability | Typical Error |
|----------|----------------|---------------|
| Molecular geometry | Excellent | < 0.02 Angstrom |
| Vibrational frequencies | Good | 3-5% |
| Reaction barriers | Good with correction | 2-5 kcal/mol |
| Band gaps | Fair (tends to underestimate) | 0.5-1.0 eV |
| Van der Waals interactions | Requires dispersion correction | Varies |
| Excited states | Fair (TD-DFT) | 0.2-0.5 eV |

### Software Comparison

| Software | License | Strengths | Basis Sets |
|----------|---------|-----------|-----------|
| Gaussian | Commercial | Broad functionality, well-documented | Gaussian-type |
| ORCA | Free (academic) | DFT + wavefunction methods, excellent support | Gaussian-type |
| VASP | Commercial | Periodic systems, materials science | Plane-wave |
| Quantum ESPRESSO | Open source | Periodic DFT, phonons | Plane-wave |
| Psi4 | Open source | Reference implementations, Python API | Gaussian-type |
| CP2K | Open source | Mixed Gaussian/plane-wave, large systems | Mixed |

### ORCA DFT Workflow Example

```
# geometry_optimization.inp
! B3LYP def2-TZVP D3BJ OPT FREQ
# B3LYP functional, triple-zeta basis, D3 dispersion, optimize + frequencies

%pal
  nprocs 8
end

%maxcore 4000

* xyz 0 1
C  0.000  0.000  0.000
O  1.200  0.000  0.000
H -0.500  0.866  0.000
H -0.500 -0.866  0.000
*
```

Run with:
```bash
orca geometry_optimization.inp > geometry_optimization.out
```

### Analyzing DFT Results with Python

```python
from ase.io import read
from ase.visualize import view

# Read optimized geometry from ORCA output
atoms = read('geometry_optimization.xyz')

# Extract energies from output file
import re

with open('geometry_optimization.out') as f:
    text = f.read()

# Total energy
energy = float(re.search(r'FINAL SINGLE POINT ENERGY\s+([-\d.]+)', text).group(1))
print(f"Total energy: {energy:.6f} Hartree")
print(f"Total energy: {energy * 627.509:.2f} kcal/mol")

# Thermochemistry
gibbs_match = re.search(r'Final Gibbs free energy\s+\.\.\.\s+([-\d.]+)', text)
if gibbs_match:
    gibbs = float(gibbs_match.group(1))
    print(f"Gibbs free energy: {gibbs:.6f} Hartree")
```

## Molecular Dynamics Simulations

### MD Pipeline

```
Initial Structure (.pdb/.mol2)
    |
    v
[Parameterization] --> Force field assignment (AMBER, CHARMM, OPLS)
    |
    v
[Solvation] --> Add solvent box, ions
    |
    v
[Minimization] --> Energy minimization (steepest descent)
    |
    v
[Equilibration] --> NVT then NPT ensemble (100 ps - 1 ns)
    |
    v
[Production] --> NPT ensemble (10 ns - microseconds)
    |
    v
[Analysis] --> RMSD, RMSF, hydrogen bonds, free energy
```

### OpenMM Quick Start

```python
from openmm.app import *
from openmm import *
from openmm.unit import *

# Load structure
pdb = PDBFile('protein.pdb')
forcefield = ForceField('amber14-all.xml', 'amber14/tip3pfb.xml')

# Create system
modeller = Modeller(pdb.topology, pdb.positions)
modeller.addSolvent(forcefield, model='tip3p', padding=1.0*nanometers)

system = forcefield.createSystem(
    modeller.topology,
    nonbondedMethod=PME,
    nonbondedCutoff=1.0*nanometers,
    constraints=HBonds
)

# Set up simulation
integrator = LangevinMiddleIntegrator(300*kelvin, 1/picosecond, 0.004*picoseconds)
simulation = Simulation(modeller.topology, system, integrator)
simulation.context.setPositions(modeller.positions)

# Minimize
simulation.minimizeEnergy()

# Run production (10 ns)
simulation.reporters.append(DCDReporter('trajectory.dcd', 1000))
simulation.reporters.append(
    StateDataReporter('log.csv', 1000, step=True,
                      potentialEnergy=True, temperature=True)
)
simulation.step(2500000)  # 10 ns at 4 fs timestep
```

## Machine Learning in Computational Chemistry

### ML Potential Energy Surfaces

Machine learning potentials achieve near-DFT accuracy at a fraction of the cost:

| Method | Speed vs DFT | Accuracy | Training Data |
|--------|-------------|----------|---------------|
| ANI | 1000x faster | ~1 kcal/mol | Pre-trained |
| SchNet | 100-1000x | ~1 kcal/mol | 1K-100K configs |
| MACE | 100-1000x | < 1 kcal/mol | 1K-100K configs |
| GemNet | 100-1000x | < 1 kcal/mol | 1K-100K configs |

### Property Prediction with RDKit

```python
from rdkit import Chem
from rdkit.Chem import Descriptors, AllChem
import numpy as np

def compute_molecular_features(smiles):
    ""

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.