search

pybamm-team / PyBaMM

Fast and flexible physics-based battery models in Python
https://www.pybamm.org/
BSD 3-Clause "New" or "Revised" License
1.09k stars 541 forks source link

Particle surface concentrations extrapolated below zero / above one #375

Closed valentinsulzer closed 5 years ago

valentinsulzer commented 5 years ago

With the Finite Volumes, we get a RuntimeWarning in Power when we try to take the square root of a negative number. This is because the particle surface concentration in the exchange-current density is extrapolated from the bulk concentration, so can go below zero or above one even if the bulk concentration is within the limits. (I think @jamiemfoster has encountered this problem before).

Options are:

  1. Leave as is and accept the warning
  2. Use more points for extrapolation (not obvious that this would work)
  3. Cut off if extrapolated c hits the limits (from what I have seen, this only happens near the cut-off time anyway. But this might not be the case at higher C-rates.)
  4. Accept as an issue for Finite Volumes and rely on other spatial methods to fix this
  5. ... others?

My preference is for a combination of 3 & 4.

Scottmar93 commented 5 years ago

Cut off if extrapolated c hits the limits (from what I have seen, this only happens near the cut-off time anyway. But this might not be the case at higher C-rates.)

I'm for this. Should be quite simple since we already have the surface values in the output dictionary.

jamiemfoster commented 5 years ago

This is one, but not the only, reason to switch to finite elements/control volumes. One reason to not be tempted by the cheap and dirty solution of "cut it off" is that this introduces discontinuities in the system of ODEs and you may find that it causes further issues down the line when time stepping.

On Fri, 10 May 2019, 18:59 Scott Marquis, notifications@github.com wrote:

Cut off if extrapolated c hits the limits (from what I have seen, this only happens near the cut-off time anyway. But this might not be the case at higher C-rates.)

I'm for this. Should be quite simple since we already have the surface values in the output dictionary.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/pybamm-team/PyBaMM/issues/375#issuecomment-491378489, or mute the thread https://github.com/notifications/unsubscribe-auth/AKQDSJI3MIMQDLHFPXSD3HTPUWZWNANCNFSM4HMESE6A .

Scottmar93 commented 5 years ago

One reason to not be tempted by the cheap and dirty solution of "cut it off" is that this introduces discontinuities in the system of ODEs

Ah, interesting. Does this also apply to using other "Events" like having a voltage cutoff etc? Or if we cut off when the internal concentrations reach zero? Should we also try to avoid cutting things off in general?

valentinsulzer commented 5 years ago

Not sure if we mean the same thing by "cut off", what I mean is terminate the time-stepping if this occurs (i.e. cut off in time rather than space). However, I guess it might be possible for this extrapolation problem to occur even if the bulk concentration is not close to zero/one (which is when we cut off already)?

jamiemfoster commented 5 years ago

Not in the very same way, but they might share some similarities. The thing which I believe is particularly obnoxious about the "cut it off" (by which I mean "round it up to zero" or something similar) is that you might find that when trying to take a time step close to an instance where we "round it up to zero" the root finding step gets confused because because the Jacobian appears singular; slightly perturbing the state vector \bf{u} does not affect the RHS of the ODEs, \bf{f(u)}, and so you gets zeros in the Jacobian where there shouldn't be. I am sure that this is soler dependent, but I have found it to be a stumbling block with ode15s and so the issue might be somewhat generic.

By contrast, things like approaching a cut-off voltage tend to be reasonably smooth and so you might have to take smaller time steps to stop exactly where you want to, but that's not normally too much of an issue.

Jamie M. Foster, School of Mathematics and Physics, University of Portsmouth, Portsmouth, Hampshire, UK (PO1 2UP). Phone: +44 02392 846 356 (internal: 6356) Email: jamie.michael.foster@gmail.com URL: http://ms.mcmaster.ca/~jmfoster/

On Fri, May 10, 2019 at 7:21 PM Valentin Sulzer notifications@github.com wrote:

Not sure if we mean the same thing by "cut off", what I mean is terminate the time-stepping if this occurs (i.e. cut off in time rather than space). However, I guess it might be possible for this extrapolation problem to occur even if the bulk concentration is not close to zero/one (which is when we cut off already)?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/pybamm-team/PyBaMM/issues/375#issuecomment-491385502, or mute the thread https://github.com/notifications/unsubscribe-auth/AKQDSJP2ZB75SOV5R24LQETPUW4LRANCNFSM4HMESE6A .