Luc Grosheintz
Luc Grosheintz
Since we anyway have access to the Jacobian, we can (somewhat) inexpensively do Newton error propagation: https://en.wikipedia.org/wiki/Propagation_of_uncertainty#Simplification The idea is that it should give us an estimate of how big...
It requires two fixes: 1. Set `dt = 1e9` before computing the steady state. 2. Use Newton error propagation to estimate how large roundoff error in `F` are. If we...
I don't see a segfault, only a very long error message leading to an assert followed by a core dump: ``` special: /home/hines/neuron/temp/share/examples/nrniv/nmodl/x86_64/capmp.cpp:824: void neuron::nrn_init_capr(const {anonymous}::_nrn_model_sorted_token&, NrnThread*, Memb_list*, int): Assertion...
That's plausibly explained by something I didn't mention in the summary. One more difference is how `F` scales in `dt`: ``` (X - X_old)/dt = dX # NOCMODL X =...
The changes have been implemented here: #1403.
One could also have a flag in NMODL to produce the files. Which could be use by `nrnivmodl` to dump them in a location the compiler can pick them up...
I'm sorry, I confused myself. We're getting rid of `vectorize` by renaming it `thread-safe` and sometimes genuinely removing it. The problem is that `vectorize` refers to both putting data into...
As discussed offline, it doesn't reproduce on BB5 or a mac.
Thank you, nice idea. We implemented it in #1025 but since `boost::span` is a very recent addition, the Boost version installed in the CI image (and I believe commonly on...
Should be available on the main branch now.