GSL’s QAG routine’s limit and key should be configurable through the module

[marcpaterno]

The current use of QAG has both the limit and key values hard-wired. This does not allow users to tune these parameters when needed for dealing with harder integrals.

It would be best for these parameters to be settable by the user, through the module configuration. The current values can be left as defaults.

[marcpaterno]

I’ve investigated this module further, and it looks like it does not check whether the calls it makes to the GSL integration routine gsl_integration_qag succeed. It also does report the error estimate on the integral.

[joezuntz]

@marcpaterno which module are you looking at here?

[marcpaterno]

Oops; apologies for the poor issue writing!

The module in question is boltzmann/sigma_cpp.

[zhouconghao]

I tried something more with sigma_cpp. I made a copy of the module files and change the subdivision limit and epsilon. Under this cosmology, 1E8 subdivision limit, and 1e-6 epsilon. gsl still complains this

gsl: qag.c:257: ERROR: maximum number of subdivisions reached

Cosmology:
[cosmological_parameters]
h0 = 7.71358152e-01
omega_m = 0.1 1.87518978e-01 0.3
omega_b = 0.03684521668
omega_k = 0.0
tau = 0.078
n_s = 1.06611261e+00
log1e10As = 3.90460746e+00
w = -1.0
wa = 0.0
omega_nu=0.00624525169

[zhouconghao]

I changed the integration from 15 points Gauss-Kronod to 21 points, and it works fine!