[Bug]: Batch Relaxer is missing gradient for converged check

[luisbro]

Contact Details

No response

Bug Description

Using the MatterGen evaluation script call MatterSim's BatchRelaxer. This leads to the error TypeError: Optimizer.converged() missing 1 required positional argument: 'gradient'. The issue lies with ase, which changed their API and now require the gradient argument. An issue has been raised to ase's gitlab, where it seems like it'll be discussed if this chage should be reverted.

MatterSim Version

v1.1.2

Python Version

3.10.14

Reproduction Steps

1. Install MatterSim as one of MatterGen's dependencies
   • This installs an older version v1.1.2, but the issue likely affects newer ones as well
2. Run MatterGen's evaluation script

Expected Behavior

MatterSim relaxes the given structures

Actual Behavior

Early termination of script with error TypeError: Optimizer.converged() missing 1 required positional argument: 'gradient', due to an argument missing here:

mattersim/src/mattersim/applications/batch_relax.py

Line 126 in 5b1ee33

if opt.converged() or opt.nsteps >= self.max_n_steps:

.
Workaround / fix

gradient = opt.optimizable.get_gradient()
if opt.converged(gradient) or opt.nsteps >= self.max_n_steps:

Error Logs

Code of Conduct

• I agree to follow the project's Code of Conduct

[ShenshenYan]

good! By the way, the ase==3.25.0 is fine without workaround / fix.