Seeking Advice: Parameter Settings for Stress-Strain Curve of EPDM/Carbon Black Composite in Materials Studio

Dear MD Community,

I am currently working on molecular dynamics simulations using Materials Studio (Forcite module) to obtain the macroscopic stress-strain curves of a complex rubber composite system. I would greatly appreciate your insights on optimizing my loading script parameters to balance computational efficiency and physical accuracy.

My System Details:

  • Polymer Matrix: 12 EPDM chains (Degree of Polymerization = 100).

  • Filler: 3 layers of graphene (serving as an equivalent model for Carbon Black).

  • Additives: Vulcanizing agents and antioxidants (RD/MB).

  • Note: The initial structure has been thoroughly relaxed (Geometry Optimization followed by a 1 ns NPT equilibration at 298 K and 0 GPa) to reach a stable density and minimum energy state.

My Core Questions:

1. Loading Method: Stress-controlled vs. Strain-controlled?

I am debating whether to use stress-controlled loading (applying a constant force via Barostat in NPT) or strain-controlled loading (applying affine deformation in steps/continuously and sampling stress in NVT). Given the high viscosity and complex entanglement of my filled EPDM system, which method is more stable and reliable for capturing the full 0-60% deformation range without crashing?

2. Equilibration Time per Level?

If using a step-wise loading approach, the system is forced into a high-energy state after each deformation step. How do you practically determine the necessary relaxation/equilibration time required for the polymer chains and filler network to adapt to the new box dimensions before sampling?

3. Production/Sampling Time?

At 298 K, the thermal fluctuations (virial stress noise) are massive (\\\$\\pm\\\$ 100-200 MPa) compared to the actual mechanical response of the rubber (~10-30 MPa). How long should the sampling window be at each strain/stress step to effectively average out this thermal noise?

4. Balancing Efficiency and Accuracy?

Running 500 ps to 1 ns per data point is computationally expensive. Are there any proven strategies (e.g., Continuous Non-Equilibrium MD at specific strain rates) that offer a good trade-off between getting a physically accurate absolute modulus and saving computational resources?

Thank you in advance for any scripts, literature, or personal experience you can share!

Best regards