Skip to content

fort.99 File

The fort.99 file provides a detailed breakdown of how well ReaxFF reproduces the reference data defined in trainset.in. It is the most informative output file for diagnosing why a force field performs well or poorly during optimization.

While fort.13 reports only the total error, fort.99 shows the individual contributions that make up that total.

Purpose

  • Compare ReaxFF-computed values against QM or literature references
  • Compute weighted squared errors for each training-set entry
  • Provide transparency into the force-field cost function

This file is generated during force-field optimization runs.

File Structure

Each line corresponds to one entry in trainset.in and reports:

Column Meaning
Description Type of data (charge, bond, angle, energy, etc.)
ReaxFF value Value computed by ReaxFF
QM/Lit value Reference value
Weight Weight from trainset.in
Error Squared weighted error
Total error Cumulative error up to this line

Example: fort.99 Output

FField value   QM/Lit value   Weight   Error   Total error

methane Heat of formation:
-17.8000       -17.8000       2.0000   0.0000  0.0000

chexane Charge atom: 1
-0.1604        -0.1500        0.1000   0.0109  0.0109

Heat of formation:
-29.4900       -29.4900       2.0000   0.0000  0.0109

Bond distance: 1 2
1.5586         1.5400         0.0100   3.4571  3.4679

Bond distance: 1 7
1.1696         1.1000         0.0200   12.1227 15.5906

Bond distance: 1 8
1.1713         1.1000         0.0200   12.7203 28.3109

Valence angle: 1 2 3
110.8117       111.0000       1.0000   0.0354  28.3463

Valence angle: 7 1 8
104.3207       107.0000       1.0000   7.1788  35.5251

chex_cryst a:
11.8448        11.2000        0.4000   2.5987  38.1238

Energy +butbenz/1 -butbenz_a/1
-96.6941       -90.0000       1.5000   19.9158 58.0396

Energy +butbenz/1 -butbenz_b/1
-63.4751       -71.0000       1.5000   25.1663 83.2060

Energy +butbenz/1 -butbenz_c/1
-77.1805       -78.0000       1.5000   0.2985  83.5045

Energy +chex_cryst/16 -chexane/1
-6.2139        -11.8300       2.0000   7.8851  91.3896

Error Definition

The force-field error is computed as:

[ \text{Error}^{\text{ReaxFF}} = \left( \frac{v^{\text{ReaxFF}} - v^{\text{QC/Lit}}}{\text{weight}} \right)^2 ]

The sum of all individual errors (final value in the last column) is the total error used in:

  • fort.13 (summary error per parameter trial)
  • fort.79 (parabolic extrapolation)

Relation to Other Files

File Role
trainset.in Defines target data and weights
fort.99 Per-datapoint error breakdown
fort.13 Total error per parameter trial
fort.79 Parameter optimization logic

Practical Notes

  • Large contributions in fort.99 immediately highlight:
  • Poorly fit bonds or angles
  • Overweighted training data
  • Conflicting targets in trainset.in
  • Energy terms often dominate the total error
  • fort.99 is the best starting point for manual force-field debugging

ReaxKit Context

In ReaxKit, fort.99 is ideal for:

  • Visualizing error contributions
  • Ranking training-set importance
  • Building diagnostic plots for force-field optimization