Removed because even if the parent has the same variables, the child is created in its neighborhood and the purpose of this test, to be the same individual as the parent, is not achieved.

Is the valid range for $\eta$ is $\geq 0$? I want to validate the value here since I find that setting $\eta = -1$ causes ZeroDivisionError. This may also happen in SBXCrossOver.

    beta_1 = np.power(1 / (2 * us), 1 / (eta + 1))
                                    ~~^~~~~~~~~~~
ZeroDivisionError: division by zero

I could not find a description of the range of eta in the references in the document, but the following ones mention that it is a non-negative value.
Add a check for non-negative values, since there is also the issue of zero divisions.

Self-adaptive simulated binary crossover for real-parameter optimization

I will implement the same for the SBX as a another PR.

Although it is a very small probability, there is a chance of becoming the denominator in 1 / (2 * us) can be zero since rng.rand() $\in [0, 1)$ (ref).

The below might be safer. (This pseudocode might require modification since us is vector whereas eps is scalar.)

        beta_1 = np.power(1 / max((2 * us), eps), 1 / (eta + 1))

There is no strong reason for the eps, but I have chosen to use the same values used in the SBX.

Although out of the range of values actually generated(rng.rand()∈[0,1)), 1 is set as us value in the test. Since zero divisions occur even in beta_2, a similar correction has been made.

beta_2 = np.power(1 / (2 * (1 - us)), 1 / (eta + 1))

Question: Where do these values come from? I have no idea about what is tested here.

Based on #6033 (comment), I modified the following so that zero divisions do not occur.

beta_1 = np.power(1 / np.maximum((2 * us), eps), 1 / (eta + 1))

The original test was [-inf, -inf] because of the zero division, but has been corrected as above.

However, beta_1 still be a very large value, so vsbx generate a child on the -inf side. The value tested here is considered to be a result that depends on the eps value.

I followed the original test with a random value of 0 and adjusted the test results, but as you pointed out, this is not a very meaningful test, so it may be more appropriate to specify a different meaningful random value.

I think it's important to keep the test with us = 0 since it covers an edge case. That said, I agree that the test is ambiguous. How about adding a comment to clarify the behavior, as you mentioned here?

I added a comment on the test content.