xTB

Official Resources

• Homepage: https://github.com/grimme-lab/xtb
• Documentation: https://xtb-docs.readthedocs.io/
• Source Repository: https://github.com/grimme-lab/xtb
• License: GNU Lesser General Public License v3.0

Overview

xTB (extended tight-binding) is a semiempirical quantum chemistry package implementing various tight-binding methods with parametrizations ranging from GFN0-xTB to GFN2-xTB. It is extremely fast, robust, and covers the entire periodic table, making it ideal for large-scale screening, conformer searches, and preliminary geometry optimizations.

Scientific domain: Computational chemistry, conformational analysis, screening, high-throughput calculations
Target user community: Chemists needing fast quantum calculations for large molecules or extensive screening

Theoretical Methods

• GFN0-xTB: Minimal basis tight-binding
• GFN1-xTB: Generalized Force Field for xTB version 1
• GFN2-xTB: Improved parametrization with H-bonding
• GFN-FF: Force field approximation (ultra-fast)
• Implicit solvation models (GBSA, ALPB)
• D3 and D4 dispersion corrections
• Geometry, Frequency, Noncovalent interactions (GFN methods)
• Periodic boundary conditions
• External electric fields

Capabilities (CRITICAL)

• Geometry optimization (molecules and periodic systems)
• Single-point energy calculations
• Molecular dynamics (NVE, NVT, NVT ensembles)
• Metadynamics for conformational sampling
• Vibrational frequencies and IR/Raman spectra
• Thermochemistry (enthalpies, entropies, Gibbs energies)
• Transition state searches
• Reaction path calculations
• Conformational searches and ensemble generation
• Solvation free energies
• pKa value estimation
• Protonation site prediction
• Noncovalent interaction analysis
• Molecular properties (dipole, polarizability)
• Band structure and DOS for periodic systems
• Systems up to 10,000+ atoms routinely
• Extremely fast (seconds to minutes for typical molecules)

Sources: Official xTB documentation (https://xtb-docs.readthedocs.io/), cited in 6/7 source lists

Key Advantages

Speed:

• Orders of magnitude faster than DFT
• Covers entire periodic table (H-Rn)
• Typical small molecule: <1 second
• Protein (1000 atoms): minutes
• Enables extensive conformer generation

Robustness:

• Stable SCF convergence
• Minimal failures compared to other semi-empirical methods
• Handles challenging systems (metal complexes, radicals)
• Reasonable geometries across chemical space

Accuracy:

• Good geometries (RMSD ~0.1 Å for organic molecules)
• Reasonable energetics for conformers
• Reliable thermochemistry
• Better than most force fields, faster than DFT

Coverage:

• All elements H through Rn
• Organic, inorganic, organometallic
• Main group and transition metals
• No parameterization gaps

GFN Method Comparison

GFN0-xTB:

• Fastest variant
• Minimal basis
• Good for initial screening
• Less accurate than GFN1/GFN2

GFN1-xTB:

• Balanced speed and accuracy
• Good for general applications
• Reliable energies and geometries

GFN2-xTB:

• Most accurate variant
• Improved H-bonding description
• Better thermochemistry
• Recommended for most applications

GFN-FF:

• Ultra-fast force field mode
• For very large systems or long MD
• Quantum-informed force field
• Smooth potential energy surfaces

Inputs & Outputs

• Input formats:

  • XYZ coordinates (primary)
  • SDF molecular files
  • Turbomole coord format
  • POSCAR for periodic systems
  • Command-line driven

• Output data types:

  • Energies and gradients
  • Optimized geometries (XYZ, Turbomole)
  • Vibrational frequencies
  • Thermochemical data
  • Charges (Mulliken, CM5, EEQ)
  • Molecular orbitals
  • Property files

Interfaces & Ecosystem

• CREST integration:

  • Conformer-Rotamer Ensemble Sampling Tool
  • Automated conformer searches
  • Extensive sampling capabilities

• Python interfaces:

  • ASE calculator for xTB
  • Direct Python bindings (xtb-python)
  • Workflow automation

• Standalone utilities:

  • crest for conformer generation
  • xtb4stda for excited states preparation

• Workflow tools:

  • Compatible with standard quantum chemistry workflows
  • Pre-optimizer for DFT calculations

Workflow and Usage

Typical Workflows:

1. Conformer Search:

# Generate conformers with CREST
crest molecule.xyz --gfn2 --alpb water

# xTB optimization
xtb molecule.xyz --opt --gfn2 --alpb water

2. Geometry Optimization:

# Single optimization
xtb molecule.xyz --opt tight --gfn2

# With solvation
xtb molecule.xyz --opt --gbsa water

3. Thermochemistry:

# Frequency calculation
xtb molecule.xyz --hess --gfn2

# Extract thermochemical data from output

4. High-Throughput Screening:

# Loop over many structures
for file in *.xyz; do
  xtb $file --gfn2 --chrg 0 >> results.txt
done

Advanced Features

Conformational Sampling (CREST):

• Metadynamics-based sampling
• iMTD-GC algorithm
• Automated conformer generation
• Degenerate conformer identification
• Boltzmann weighting

Solvation Models:

• GBSA: Generalized Born with surface area
• ALPB: Analytical Linearized Poisson-Boltzmann
• Wide range of solvents parameterized
• Solvation free energies

Property Calculations:

• Fukui functions
• Electron localization function (ELF)
• Partial charges (multiple schemes)
• Bond orders
• Noncovalent interaction index

Periodic Systems:

• Crystal structure optimization
• Band structures
• Density of states
• Bulk modulus calculations

Performance Characteristics

• Speed: 10-1000x faster than DFT
• Scaling: Near-linear for many operations
• Memory: Very low; minimal RAM requirements
• Parallelization: OpenMP support
• Typical times:
  • Small molecule (20 atoms): <1 second
  • Medium molecule (100 atoms): 1-10 seconds
  • Protein fragment (500 atoms): 1-5 minutes

Limitations & Known Constraints

• Semiempirical accuracy: Not as accurate as high-level ab initio
• Energetics: Barriers and reaction energies less reliable
• Excited states: Not directly calculated (use sTDA separately)
• Strong correlation: Not designed for multireference systems
• Absolute energies: Relative energies more reliable
• Metal complexes: Good but not always quantitative
• Learning curve: Low; command-line interface straightforward
• Documentation: Good and accessible
• Platform: Linux, macOS, Windows (via WSL)

Comparison with Other Methods

• vs DFT: 100-1000x faster, lower accuracy
• vs DFTB+: xTB no parameter files needed, better coverage
• vs PM6/PM7: xTB more robust, broader coverage
• vs Force Fields: More accurate, still very fast
• Sweet spot: Preliminary optimizations, conformer searches, screening

Application Areas

Drug Discovery:

• Conformer generation for docking
• Ligand preparation
• pKa prediction
• Tautomer enumeration

Chemical Reaction Screening:

• Reaction pathway exploration
• Barrier estimation
• Product prediction

Materials Science:

• Crystal structure prediction
• MOF screening
• Supramolecular assemblies

Methodology:

• Pre-optimization for DFT
• Initial guess generation
• Filtering for expensive calculations

Integration in Computational Workflows

As Pre-optimizer:

1. xTB geometry optimization (fast)
2. DFT single-point or optimization (accurate)
3. Post-processing

For Ensemble Generation:

1. CREST conformer search (extensive)
2. xTB re-optimization and ranking
3. DFT refinement of top conformers

High-Throughput Screening:

1. Generate large library
2. xTB property calculation (fast)
3. ML model training or direct selection
4. DFT validation

Verification & Sources

Primary sources:

1. GitHub repository: https://github.com/grimme-lab/xtb
2. Documentation: https://xtb-docs.readthedocs.io/
3. C. Bannwarth et al., J. Chem. Theory Comput. 15, 1652 (2019) - GFN2-xTB
4. S. Grimme et al., J. Chem. Theory Comput. 13, 1989 (2017) - GFN1-xTB
5. P. Pracht et al., Phys. Chem. Chem. Phys. 22, 7169 (2020) - CREST

Secondary sources:

1. xTB tutorials and examples
2. CREST documentation
3. Published applications in conformer generation
4. Confirmed in 6/7 source lists (claude, g, gr, k, m, q)

Confidence: CONFIRMED - Appears in 6 of 7 independent source lists

Verification status: ✅ VERIFIED

• Official homepage: ACCESSIBLE (GitHub)
• Documentation: COMPREHENSIVE and ACCESSIBLE
• Source code: OPEN (GitHub, LGPL v3)
• Community support: Very active (GitHub issues, discussions)
• Academic citations: >500 (GFN papers)
• Active development: Regular releases, continuous improvements
• Benchmark validation: Extensive benchmarks published
• Wide adoption: Standard tool for conformer generation and pre-optimization