xi calculation in SB.py (cpu-float32 backend) does not include h term #32
Description
In mindquantum/mindquantum/algorithm/qaia/SB.py, the calculation of self.xi is inconsistent across different backends.
For "gpu-float32" and "npu-float32", when h is not None, the formula includes the h contribution:
- 2 * ((self.h / 2) ** 2).sum()
However, for "cpu-float32", the current implementation is:
self.xi = 0.5 * np.sqrt(self.N - 1) / np.sqrt(csr_matrix.power(self.J, 2).sum())
This ignores the h term, which seems inconsistent and may lead to incorrect results when h is not None.
Proposed fix:
Update the "cpu-float32" branch to also include the h contribution, similar to other backends, for consistency and correctness.