It seems that B_for_integrand variable can fail to be between the background integration bounds (even though it should be within those), when the bounds are determined by both the user given support and the default limits. This can lead to numerical problems for the integration, and thus GSL can fail to produce a correct integration result and lead to X-PSI setting an arbitrarily low likelihood value for a parameter vector when that is not necessary.
It seems that
B_for_integrand
variable can fail to be between the background integration bounds (even though it should be within those), when the bounds are determined by both the user given support and the default limits. This can lead to numerical problems for the integration, and thus GSL can fail to produce a correct integration result and lead to X-PSI setting an arbitrarily low likelihood value for a parameter vector when that is not necessary.Here the lines related to the issue: https://github.com/xpsi-group/xpsi/blob/1dfb38f7a598f4dc9b2cd5d507c52018cdcbd492/xpsi/likelihoods/default_background_marginalisation.pyx#L532
Solution would be adding these lines after the linked line: