Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.
Stationary and Time Dependent giving different solutions?
Posted Jul 9, 2020, 6:13 a.m. EDT Electromagnetics, Fluid & Heat, Chemical Version 5.4 7 Replies
Please login with a confirmed email address before reporting spam
I've successfully modelled a transport of coupled dilute species, electrostatics and deformed geometry problem in 1D and I'm now trying increase it to 2D. I'm having an issue with concentrations going negative in a particular regions of my model, depsite using the same boundary conditions that were successful in my 1D model. At the moment I am ignoring the deformed geometry and trying to solve the stationary case to use as initial values for the deforming time resolved case. The stationary model converges, but with the negative concentration region.
What's interesting is that if I use this erroneous solution as the initial condition for a time resolved study of some arbitrary time I get a reasonable looking result for all time steps (i.e. no negative concentrations and good agreement with the 1D model). So the time-resolved solution for what should be a static system does indeed look static but the stationary solver on the same system will not give this result.
Does anyone have any suggestions as to why this would be?
Please login with a confirmed email address before reporting spam
Hello Tom,
Have you checked this knowledge base entry already? https://www.comsol.com/support/knowledgebase/952
It has many useful tips.
Please login with a confirmed email address before reporting spam
Hi Alexis,
Yes I did check that. The problem with the stationary solver doesn't change with mesh refinement so I don't think its that. And as I said the same mesh will give me a sensible answer using the time dependent solver (right from the first time step!). I have no reactions in the domain, only at the boundaries. The concentration is dropping negative only adjacent to straight boundaries of my domain, and look fine next to the curved boundaries. (domain is rounded rectangle)
Please login with a confirmed email address before reporting spam
Could you share a screenshot with a map of concentration and your mesh ?
Please login with a confirmed email address before reporting spam
Sure thing. Here's the region in question. I've highlighted the region of negative concentration as you can see that the overall solutions are very similar.
Attachments:
Please login with a confirmed email address before reporting spam
So the negative concentrations are only located in regions where the concentration is nearly 0, right?
Could you change the data and/or color range so that the color bar's maximum is 0 ? I'm wondering if the magnitude of the negative concentrations might be neglected.
Also sorry to be inquisitive but your channel is only 2nm wide. Is this correct? It seems extremely narrow to me.
Please login with a confirmed email address before reporting spam
The domain is very small, I'm attempting to model the diffusion and migration of vacancies and interstitials in an passive layer. (Some further information here if you're interested http://dx.doi.org/10.1016/j.electacta.2012.02.008 ). After looking at the data to try and produce the plots requested, I've established that none of the mesh elements appear to be negative concentration. However if I plot the concentration along the outer boundary, or integrate along it I am given a negative concentration. See plots attached. I'm now thinking that this is a discretization issue as the concentration gradient is large at this boundary and the concentration should tend to zero. Perhaps the gradient across the mesh element is overshooting in the stationary case. I don't understand why I get different results between the solvers though.
Attachments:
Please login with a confirmed email address before reporting spam
Very interesting system, thanks for the read!
I would not be very worried about these concentrations. After all ±10⁻¹¹ mol/m³ is essentially 0 for any practical purpose, when the typical scale of concentration is 100 mol/m³ The ratio between the two is in fact not too far from the representation limit of double precision floats anyway.
Another way to look at it is that over a volume of 10 × 10 × 10 nm³, this concentration multiplied by the Avogadro number is many orders of magnitude below 1 ion.