Questions on the Kolmogorov flows code

[cjunho]

I am Junho, a researcher who is studying about PINN at SungKun Korea. At my paper, I am going to write and refer your results about Kolmogorov flows implemented by Physics-Informed Neural Operator. To do that, I would like to ask you questions about your code and the numerical results.

1. Could you confirm if the result I reproduced is correct? I ran your code “python3 train_operator.py --config_path configs/pretrain/Re500-pretrain-05s-4C1.yaml” with data NS_fft_Re500_T4000.npy when setting ic_loss=1, f_loss=1, xy_loss=0. As a result, I got 0.35570 data L2 error when 40000 epochs as shown in ns_err.txt. Is it correct and can I refer the error at my paper?

2. I am wondering what Equation error is?

3. What is instance wise learning fine-tuning and how can I run it at your code? At your paper, “Physics-Informed Neural Operator for Learning Partial Differential Equations” you said instance wise learning fine-tuning produced more accurate results. The instance wise learning fine-tuning looks like an additional operator loss to PDE loss made from PDE residual, but I could not find where the operator loss is at your code.

Thank you for reading my questions.
ns_error.txt
Re500-pretrain-05s-4C1.txt