We can exploit the conjugate gradient formulation. The theory for it is resumed in the following PDF [odd_correction_with_lanczos.pdf](https://github.com/user-attachments/files/26109813/odd_correction_with_lanczos.pdf)