Research data supporting "Mechanical properties of graphene oxide from machine-learning-driven simulations"

This repository contains code and data supporting the following work:

Zakariya El-Machachi, Bowen Cheng, Volker L. Deringer, Chem. Commun., 2025, 61, 11405-11408

The purpose is to enable readers to access the structural models and simulation code for reproducing the work.

Contents

The repository is structured as follows:

• Code: The straining code used in this work. This code has been modified from the original production code only in documentation to remove typos and to remove absolute paths. The Young's modulus is calculated as follows:

  def get_young_modulus(stress, strain):
      # Fit a linear model to the stress-strain data in elastic region 
      coefficients = np.polyfit(strain[1:101], stress[1:101], 1)
      young_modulus = coefficients[0]
  return young_modulus
  

• Structures: 10 input structures for GO and 10 for rGO.

The MACE model used in this work is available at zakmachachi/GO-MACE-23.

Due to large file sizes, trajectory data can be found on Zenodo:

Figures

The following figures were constructed as follows:

• Figure 1: Structural model can be found at zakmachachi/GO-MACE-23. Trajectory data can be found using the Zenodo link above.

• Figure 2: Input structures can be found here. For panel a), b) and c), the structures are in p1-p2-GO.xyz. The table below maps the structure configurations for a "batch":

  • Top row: OH/O ratio
  • Left column: O content (%)
  • Cells: Index of each configuration

  OH/O → O content (%) ↓	0.00	0.25	0.50	0.75	1.00
  0.10	0	1	2	3	4
  0.20	5	6	7	8	9
  0.30	10	11	12	13	14
  0.40	15	16	17	18	19
  0.50	20	21	22	23	24

  In total there are 10 batches, meaning 10 structures per config type, and having 25 config types means a total of 250 structures total. Since we are straining in x and in y, we plot the average of the combined x and y data meaning we have 20 configurations per line in the plot (shaded with ± 1 standard deviation).

• Figure 3: Input structures can be found here again. For panel a), the structures are in p1-p2-GO.xyz, for panel b) they are in p1-p2-rGO.xyz. We are now separating x and y values of $E$, which are slightly different. Each data point is the average $E$ of the 10 config types for x strained (full symbol) and the same 10 strained in y (hollow symbol). rGO will lose some species under thermal annealing and since these are removed, it results in a lower O% content in some cases. We also see a transformation in functional groups and so this is also in reference to the initial structure.