Brief description: The bug was unfortunately introduced during extensive optimizations to the CalC numerical engine in versions x.8.0.
Therefore, this bug is present in CalC versions x.8.0 through x.8.9, and affects scripts that define more than a single Ca
2+ channel, in the case
of non-zero
background [Ca2+].
The bug leads to a reduction of the Ca
2+ current through the 2
nd and each consecutive Ca
2+ channel.
This reduction factor equals
1/(1+2×Ca.bgr×V), where
V is the
diffusion volume (in units of μm
3),
and
Ca.bgr is the
background [Ca2+] (in units of μM).
For example, if Ca.bgr=0.05 (in μM) and the
volume is 1x1x1 μm
3, the reduction is about 9% through all but the first channel
(see diagnostic output below for this case).
If on the other hand the volume is smaller, (0.5 μm)
3, this reduction is around 1.2%.
The bug is fixed in versions x.9.1 posted
on the main CalC homepage and also below.
Great thanks to
Christian Stricker for alerting me to large charge loss in one of his
scripts!
Technical explanation: Ironically, the bug was a result of a second layer of current amplitude validation checks added in CalC versions 6.8.0 and 7.8.0
to remove the possibility of tiny Ca
2+ current leaks due to the partial overlap of current source profile with domain boundaries.
The validation check had a mis-placed initialization statement that caused complex behavior described above when more than one channel is present, therefore it only affects
some of the scripts, thereby slipping by version deployment checks.
Which simulations may be affected: If you used
CalC versions 6.7.4/7.7.4 or ealier,
or if your script defines only a single Ca
2+ channel,
or if your
background [Ca2+] is zero, your results were
not affected by the bug. If you think you may have used
versions of
CalC x.8.0-x.8.9, and your
model script had multiple
Ca.source specifications
and non-zero
background [Ca2+], I recommend redoing your
simulations with the most recent version of
CalC (x.9.1 and above), to compare with the results you obtained previously.
How to detect the problem: To spot the problem with your simulation results when running buggy versions of
CalC in question,
run your script with the version you had used, or if you didn't save that executable, use versions x.8.9 posted below.
You should look for the following diagnostic output indicating that your model script run results were affected:
#### Source #1 boundary spill-over adjustment factor = 2
#### Source #2 boundary spill-over adjustment factor = 1.82
#### Source #3 boundary spill-over adjustment factor = 1.82
For Cartesian coordinates and point currents on boundary surfaces, and for sufficiently high resolution, all adjustment factors should be close to an integer.
The diagnostic output above indicates that the bug indeed affects this script, reducing
the current through the second and third channels by 9%. You can confirm this further by monitoring the "
Charge loss" diagnostic output,
after removing any
uptake commands from your script, and switching to
Noflux (i.e. Neumann) boundary conditions.
Charge loss should remain small (< 1%).
Epilogue: I started this software project to facilitate reproducibility of simulation
results (in fact,
the first CalC publication
was a correction to a previously published flawed simulation), so
this bug is quite upsetting to me. I hope none of the published results were affected (the ones I checked
so far were not). I would be happy to help with any requests to provide old executables,
run and verify your simulation scripts, or post your new or old simulation scripts on my
script homepage.
Although I run extensive checks on each version of
CalC before release, my battery of tests turns out to be insufficient.
I will improve my release test procedures to catch potential problems such as this one.
