search

xzackli / Bolt.jl

differentiable boltzmann code
MIT License
43 stars 5 forks source link

Minor conformal time integration issue #79

Closed jmsull closed 1 year ago

jmsull commented 1 year ago

At least on the ie branch, when I run the conformal time ODE solve, I find ┌ Warning: dt(1.8189894035458565e-12) <= dtmin(1.8189894035458565e-12) at t=0.0006687152338204658. Aborting. There is either an error in your model specification or the true solution is unstable. └ @ SciMLBase ~/.julia/packages/SciMLBase/MXZxL/src/integrator_interface.jl:529

But only the first time - on a second (or any subequent) solve, the solver runs fine. This seems like something we want to avoid?

I saw Chris Rackauckas mention something about this in Julia diff-eq slack a while ago, so I am tempted to think this is not our fault - will dig it up here.

jmsull commented 1 year ago

On the topic of conformal time, based on something else I am seeing for the neutrinos we may want to change the way we are doing the background integral. Right now it uses the quadrature points but we can probably just do a (faster) cumulative trapezoidal rule...

jmsull commented 1 year ago

This issue is apparently not so minor - for the conformal time solves I am now finding actual SingularMatrix errors - will deal with this carefully after I get a working iterative prototype up first with the x solves

jmsull commented 1 year ago

This appears to have been resolved by dropping unused elements of the u vector in the truncated hierarchy, so I will close this