Thank you for your descriptive reply Ray.

Can you please provide an explanation (or links to an explanation) of the 
connection between the implicit nature of the equation (phi_s on both sides of 
the PDE) and the well-posedness / existence and uniqueness of the solution?

For clarity for the rest of the users on the list, the RHS of the PDE in 
question is strongly non-linear (the j term expands to the Butler-Volmer 
equation).

Since the right hand side of the PDE is non-linear, theoretically there is no 
guarantee of a unique solution. In this context, we would like to better 
understand your concluding statement that the solution gets automatically 
pinned uniquely for this PDE. The implementation strategy – i.e. how to 
represent these boundary conditions in FiPy – still isn’t clear to us – can you 
kindly elaborate?

With best regards,


-          Ian & Krishna

From: [email protected] [mailto:[email protected]] On Behalf Of Raymond 
Smith
Sent: 26 May 2016 19:42
To: [email protected]
Subject: Re: Pure Neumann Boundary Conditions; Elliptic PDE

At least for now, why don't we keep this on the list.
A couple of thoughts. First, I have actually already implemented a Newman-like 
model in Python. I didn't use Fipy, but I did use the finite volume method. 
It's missing a few features compared to dualfoil, but has several extra things. 
If you are interested in using it, please be in touch off the list. I plan to 
publish it, but I haven't quite gotten around to that yet. Hopefully quite soon.
To your questions/comments about the PDE, you raise a couple of critical 
differences. First, your source term, j, is a function of phi (which you didn't 
actually state but is true) and second, the specified flux boundary condition 
at the positive current collector changes the well-posed-ness entirely. In this 
case, the integral of the source term has to match the outlet flux into the 
current collector. That allows for a steady state. Then, the fact that j 
depends on phi is what actually "pins" the solution rather than allowing any 
vertical shift.
Best,
Ray

On Thu, May 26, 2016 at 2:25 PM, Gopalakrishnan, Krishnakumar 
<[email protected]<mailto:[email protected]>> 
wrote:
Hi Ray,

Many thanks for your quick reply.  I am Krishna, Ian’s colleague here at 
Imperial.  Hope you remember me from our earlier conversations about access to 
Martin’s 10.626 materials.

Ian & I are currently working on implementing the basic pseudo-2D porous 
electrode Newman model using FiPy.

The issue is that all of the PDE Boundary conditions are cast as Neumann. 
However, when we try to apply your recommendation in FiPy  (i.e. omit the BC 
definitions/revert to default), it still does not converge.


•        The PDE that we are currently solving is the solid-potential in the 
positive electrode. Sorry, there was an typo in the last email. What we have is:

o   at the positive/separator boundary, there is no potential flux. (a no-flux 
BC)

o     at the positive current collector, we have a fixed flux  ( 
[cid:[email protected]] )  BC


•        Meaning of the source term components:

o   [cid:[email protected]] specific interfacial surface area

o    [cid:[email protected]] faraday’s constant

o    [cid:[email protected]]   = Li-molar flux density 
([cid:[email protected]] )


•    The integral of the source term is  [cid:[email protected]] , 
where [cid:[email protected]] =Solid Phase Porosity, 
[cid:[email protected]] = particle radius and 
[cid:[email protected]] = the thickness of the positive electrode, 
respectively.

o   This above integral is clearly non-zero for a non-zero applied current.

o   However, the Newman model potential equation is considered to be well-posed 
and solution plots are found in papers. All other PDEs of the model also have 
pure Neumann BCs.  Hence, we don’t understand the inconsistency that arises 
from having no  steady-state solution.


Any help will be much appreciated.


Best Regards

Krishna & Ian







From: Campbell, Ian
Sent: 26 May 2016 18:07
To: Gopalakrishnan, Krishnakumar 
<[email protected]<mailto:[email protected]>>
Subject: Fwd: Re: Pure Neumann Boundary Conditions; Elliptic PDE

---------- Forwarded message ----------
From: Raymond Smith <[email protected]<mailto:[email protected]>>
Date: 26 May 2016 3:49 p.m.
Subject: Re: Pure Neumann Boundary Conditions; Elliptic PDE
To: [email protected]<mailto:[email protected]>
Cc:
Hi, Ian.
Some thoughts. Are you able to give us any more detail about your PDE? For 
example, the form/meaning of the things in the source term on the right hand 
side? Presumably the flux is defined by
\sigma \nabla \phi
If so, then not specifying any boundary conditions would lead to no-flux 
conditions because FiPy defaults to assuming zero-gradients for field variables 
on the boundary. However, you could do it explicitly with something like
phi.faceGrad.constrain([0.], mesh.exteriorFaces)
However, the form of the source is important here because if the integral of 
the source is non-zero, then there is no steady solution. If the integral _is_ 
zero and the source is not a function of \phi, then the system admits an 
infinite family of solutions, all shifted by a constant. There's some 
discussion of this issue in the 1D diffusion example:
http://www.ctcms.nist.gov/fipy/examples/diffusion/generated/examples.diffusion.mesh1D.html#module-examples.diffusion.mesh1D
Search for "Fully implicit solutions are not without their pitfalls"
Best,
Ray

On Thu, May 26, 2016 at 6:30 AM, Campbell, Ian 
<[email protected]<mailto:[email protected]>> wrote:
Dear All,

Thanks for the great work on FiPy so far.

We have a 2nd order elliptic PDE* defined in the domain (0, 1). At each 
boundary, we have a no-flux boundary condition.

We have tried to represent this in FiPy, but specifying such conditions results 
only in incorrect results with large error. We have searched this page (below) 
in depth but can’t see which function call is appropriate, if any.

http://www.ctcms.nist.gov/fipy/fipy/generated/fipy.boundaryConditions.html

Do you have a recommendation for how we may implement two no-flux boundary 
conditions on our PDE?

Best regards,


-          Ian

* $$\nabla.(\sigma_{eff} \nabla \bar{\phi}_s) = a_s F \bar{j}$$

Ian Campbell | PhD Candidate
Electrochemical Science & Engineering Group
Room 506, City & Guilds Building, Imperial College London, SW7 2AZ, United 
Kingdom
Phone: +44 (0)7449 815 520<tel:%2B44%20%280%297449%20815%20520> | E-mail: 
[email protected]<mailto:[email protected]>


_______________________________________________
fipy mailing list
[email protected]<mailto:[email protected]>
http://www.ctcms.nist.gov/fipy
  [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]


_______________________________________________
fipy mailing list
[email protected]
http://www.ctcms.nist.gov/fipy
  [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]

Reply via email to