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.
Failed to find consistent initial values.
Posted Jan 10, 2012, 5:11 a.m. EST Results & Visualization, Studies & Solvers Version 4.3b, Version 5.2a 9 Replies
Please login with a confirmed email address before reporting spam
Hi.
I coose frekvenci transient study and get thise error.
Failed to find consistent initial values.
Last time step is not converged.
THANK
I coose frekvenci transient study and get thise error.
Failed to find consistent initial values.
Last time step is not converged.
THANK
9 Replies Last Post Sep 4, 2017, 12:38 p.m. EDT
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
as it's stated, your initial conditions (default is "0") are too far from the final result so COMSOL cannot mange to get where you want.
This can come from many things :
- poor or too coarse mesh,
- missing BC (boundary conditions) as your system must be uniquely defined, all physics need enough BC's to propose a unique solution,
- initial conditions node too far from result, or unphysical, given the values in the BC's
take a new look at your model, do you believe its fully defined, then load the initial conditions: select the Dependent Variables node of your solver and right click: "compute to selected" and look at the plot of the initial values, can you define them better ?
--
Good luck
Ivar
as it's stated, your initial conditions (default is "0") are too far from the final result so COMSOL cannot mange to get where you want.
This can come from many things :
- poor or too coarse mesh,
- missing BC (boundary conditions) as your system must be uniquely defined, all physics need enough BC's to propose a unique solution,
- initial conditions node too far from result, or unphysical, given the values in the BC's
take a new look at your model, do you believe its fully defined, then load the initial conditions: select the Dependent Variables node of your solver and right click: "compute to selected" and look at the plot of the initial values, can you define them better ?
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi Ivar,
I have this problem also, but I just made the mesh finer, and all other conditions were unchanged. I do not think it belongs to the three reasons you mentioned above. Do you have any suggestions?
Thanks!
Jiangtao
I have this problem also, but I just made the mesh finer, and all other conditions were unchanged. I do not think it belongs to the three reasons you mentioned above. Do you have any suggestions?
Thanks!
Jiangtao
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
I had the same problem,
The magnitudes of my patameters were right, but I had a mistake in the units ([a] instead of [A] in electric current).
The solution was verifying all units in the Parameters' section.
Greetings,
ABM.
The magnitudes of my patameters were right, but I had a mistake in the units ([a] instead of [A] in electric current).
The solution was verifying all units in the Parameters' section.
Greetings,
ABM.
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Hi,
I also have same problem for thermal stress model for a ceramic material. I found the source of problem but I couldn't it.
I selected young modul of material around 200GPa but the model gave error. When I give small values around zero or 1 Pa (not GPa), it works. I think it come from two physics (heat transfer and stress) since these physics include the numbers having huge difference (comsol get the unit of young modulus as Pa so my value is around 200*10^9 Pa but thermal conductivity is 1 W /mK).
Does anybody has an idea?
Thanks
I also have same problem for thermal stress model for a ceramic material. I found the source of problem but I couldn't it.
I selected young modul of material around 200GPa but the model gave error. When I give small values around zero or 1 Pa (not GPa), it works. I think it come from two physics (heat transfer and stress) since these physics include the numbers having huge difference (comsol get the unit of young modulus as Pa so my value is around 200*10^9 Pa but thermal conductivity is 1 W /mK).
Does anybody has an idea?
Thanks
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
I have same problem if you find soloution please guide me
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
Updated:
8 years ago
Hello,
It is worth reiterating Ivar's comments from earlier in this thread. The issue that you're facing is usually caused by:
- Having a physically inconsistent set of boundary and domain conditions. Use your knowledge of the physics to guide yourself here. This can require some deep thought. It is also often instructive to first solve a steady-state problem with the same boundary conditions (if reasonable to do so) to examine the resulting fields. You might also want to perform an adaptive mesh refinement (www.comsol.com/blogs/using-adaptive-meshing-local-solution-improvement/) on the steady state solution, this can tell you if there are any singularities in your model: www.comsol.com/blogs/fillet-away-your-electromagnetic-field-singularities/
- Having initial conditions that are "too far" away from the actual solution. Keep in mind that the default initial values in most physics interfaces are zero. For a discussion about how to address this in the context of fluid flow problems (where we find that a lot of people make this mistake) please see:
www.comsol.com/support/knowledgebase/1172/
Basically, if you do want to introduce a step change in the boundary conditions from the initial conditions, it makes sense to add smoothing, so that the change in boundary conditions happens gradually, over some finite, but small, timespan:
www.comsol.com/support/knowledgebase/905/
On the other hand, if you do want to introduce an instantaneous change in boundary conditions during the simulation, look to using the Events interface:
1) www.comsol.com/blogs/modeling-a-periodic-heat-load/
2) www.comsol.com/blogs/implementing-a-thermostat-with-the-events-interface/
- Having "too coarse" of a mesh. The mesh must be fine enough to resolve the variations in the solution fields in the time domain. The mesh refinement studies that you do on a steady state problem will not necessarily be a good guide here. Again, use your knowledge of the physics.
Is the governing equation a hyperbolic PDE with second time derivatives? That is, will the solution be wave-like? Then think about the element size and CFL number, as described here:
www.comsol.com/support/knowledgebase/1118/
and in this example:
www.comsol.com/model/transient-gaussian-explosion-1374
Or, is the governing equation a parabolic PDE, with first time derivative, such as the transient heat transfer equation? In that case, you might want to use boundary layer meshing or swept meshing when you know ahead of time that the fields will vary rapidly in the directions normal to faces with applied loads.
I'll add one suggestion with regards to meshing: If your are using 2nd order elements (www.comsol.com/blogs/keeping-track-of-element-order-in-multiphysics-models/) you may want to try switching to 1st order elements. Although you will need a finer mesh with 1st order vs 2nd order (www.comsol.com/blogs/meshing-considerations-linear-static-problems/) the disadvantage of 2nd order elements is that they may "under" or "overshoot" the solution, as described here: www.comsol.com/support/knowledgebase/952/. In fact, for reasons related to this last point, our transport formulations (Fluid, Flow, Chemical Species Transport, etc.) by default use 1st order elements.
Of course there may be other issues in your model that are more subtle and usually require looking over the model file in question to diagnose, but the above points cover the vast majority of the issues that we find our users encountering. The first two points are the most common source of issues, and we suggest that you start there.
Best Regards,
It is worth reiterating Ivar's comments from earlier in this thread. The issue that you're facing is usually caused by:
- Having a physically inconsistent set of boundary and domain conditions. Use your knowledge of the physics to guide yourself here. This can require some deep thought. It is also often instructive to first solve a steady-state problem with the same boundary conditions (if reasonable to do so) to examine the resulting fields. You might also want to perform an adaptive mesh refinement (www.comsol.com/blogs/using-adaptive-meshing-local-solution-improvement/) on the steady state solution, this can tell you if there are any singularities in your model: www.comsol.com/blogs/fillet-away-your-electromagnetic-field-singularities/
- Having initial conditions that are "too far" away from the actual solution. Keep in mind that the default initial values in most physics interfaces are zero. For a discussion about how to address this in the context of fluid flow problems (where we find that a lot of people make this mistake) please see:
www.comsol.com/support/knowledgebase/1172/
Basically, if you do want to introduce a step change in the boundary conditions from the initial conditions, it makes sense to add smoothing, so that the change in boundary conditions happens gradually, over some finite, but small, timespan:
www.comsol.com/support/knowledgebase/905/
On the other hand, if you do want to introduce an instantaneous change in boundary conditions during the simulation, look to using the Events interface:
1) www.comsol.com/blogs/modeling-a-periodic-heat-load/
2) www.comsol.com/blogs/implementing-a-thermostat-with-the-events-interface/
- Having "too coarse" of a mesh. The mesh must be fine enough to resolve the variations in the solution fields in the time domain. The mesh refinement studies that you do on a steady state problem will not necessarily be a good guide here. Again, use your knowledge of the physics.
Is the governing equation a hyperbolic PDE with second time derivatives? That is, will the solution be wave-like? Then think about the element size and CFL number, as described here:
www.comsol.com/support/knowledgebase/1118/
and in this example:
www.comsol.com/model/transient-gaussian-explosion-1374
Or, is the governing equation a parabolic PDE, with first time derivative, such as the transient heat transfer equation? In that case, you might want to use boundary layer meshing or swept meshing when you know ahead of time that the fields will vary rapidly in the directions normal to faces with applied loads.
I'll add one suggestion with regards to meshing: If your are using 2nd order elements (www.comsol.com/blogs/keeping-track-of-element-order-in-multiphysics-models/) you may want to try switching to 1st order elements. Although you will need a finer mesh with 1st order vs 2nd order (www.comsol.com/blogs/meshing-considerations-linear-static-problems/) the disadvantage of 2nd order elements is that they may "under" or "overshoot" the solution, as described here: www.comsol.com/support/knowledgebase/952/. In fact, for reasons related to this last point, our transport formulations (Fluid, Flow, Chemical Species Transport, etc.) by default use 1st order elements.
Of course there may be other issues in your model that are more subtle and usually require looking over the model file in question to diagnose, but the above points cover the vast majority of the issues that we find our users encountering. The first two points are the most common source of issues, and we suggest that you start there.
Best Regards,
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
Hello All,
Walter: many thanks for your valuable input. I've been struggling with the issue of starting time-dependent solvers for a while now. I get the "Failed to find consistent initial values" error in a number of different problems, but in all cases the situation does not fit in the common causes that you have very helpfully listed.
Here are some details for an example problem that I have been solving:
- I'm computing a steady-state gas discharge, so my system of equations contains Poisson equation and equations of conservation of charged species. The PDE are elliptic (the non-stationary term is absent).
- I've implemented this problem in COMSOL using either the Plasma Module, or Electrostatics plus Transport of Diluted Species.
- I'm able to compute a steady-state solution as a matter of routine using stationary solvers.
- If I try to recompute the steady-state solution with a time-dependent solver, I get the error "Failed to find consistent initial values", regardless of whether I'm using Plasma Module or Electrostatics plus Transport of Diluted Species. My initial condition is the solution obtained by the stationary solver, and the mesh is the same.
Obviously, the boundary conditions and domain conditions are physically consistent, otherwise I would not be able to find a solution using stationary solvers. The mesh is also fine enough. I've used CFL criterion to setup time-stepping for the mesh used with no success: the error "Failed to find consistent initial values" keeps showing up.
Two more notes: 1) I stress that the above-described case is just an example. The issue appears not only in my models on gas discharges, but also thermal conduction and other problems based on diffusion-convection-reaction equations; in all of them, the initial condition is a previously computed solution with a stationary solver. 2) Sometimes I am able to overcome the issue by playing with solver parameters, time-stepping, etc. Basically, by pressing buttons and engaging in trial and error, which, I think, is not the proper way of working with COMSOL.
All this strikes me as very strange: how can a time-dependent solver complain about an initial condition which is a solution to the problem? It seems that either something is wrong with time-dependent solvers of COMSOL, or I am very wrong in the use I make of them. I would really like to hear your thoughts on this issue, Walter. Thanks!
Kind regards,
Pedro Almeida
Walter: many thanks for your valuable input. I've been struggling with the issue of starting time-dependent solvers for a while now. I get the "Failed to find consistent initial values" error in a number of different problems, but in all cases the situation does not fit in the common causes that you have very helpfully listed.
Here are some details for an example problem that I have been solving:
- I'm computing a steady-state gas discharge, so my system of equations contains Poisson equation and equations of conservation of charged species. The PDE are elliptic (the non-stationary term is absent).
- I've implemented this problem in COMSOL using either the Plasma Module, or Electrostatics plus Transport of Diluted Species.
- I'm able to compute a steady-state solution as a matter of routine using stationary solvers.
- If I try to recompute the steady-state solution with a time-dependent solver, I get the error "Failed to find consistent initial values", regardless of whether I'm using Plasma Module or Electrostatics plus Transport of Diluted Species. My initial condition is the solution obtained by the stationary solver, and the mesh is the same.
Obviously, the boundary conditions and domain conditions are physically consistent, otherwise I would not be able to find a solution using stationary solvers. The mesh is also fine enough. I've used CFL criterion to setup time-stepping for the mesh used with no success: the error "Failed to find consistent initial values" keeps showing up.
Two more notes: 1) I stress that the above-described case is just an example. The issue appears not only in my models on gas discharges, but also thermal conduction and other problems based on diffusion-convection-reaction equations; in all of them, the initial condition is a previously computed solution with a stationary solver. 2) Sometimes I am able to overcome the issue by playing with solver parameters, time-stepping, etc. Basically, by pressing buttons and engaging in trial and error, which, I think, is not the proper way of working with COMSOL.
All this strikes me as very strange: how can a time-dependent solver complain about an initial condition which is a solution to the problem? It seems that either something is wrong with time-dependent solvers of COMSOL, or I am very wrong in the use I make of them. I would really like to hear your thoughts on this issue, Walter. Thanks!
Kind regards,
Pedro Almeida
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
Hello Pedro,
Since these questions do revolve around a particular set of models which you've developed, we would generally not hazard a blind guess. Of course there can certainly be reasons other than those previously discussed, in particular it sounds like you'd had success with using a tighter solver tolerance and/or timestep, which is entirely reasonable, and is a point that I should have included in the previous posting.
The question of timestepping and tolerances is a quite complex topic and there are volumes of textbooks written on the subject, but the (rather egregiously oversimplified) summary is: The tighter you make the tolerances, the more accurate the solution, and less convergence difficulties you will have. It is a subject that we are planning a series of technical blogs about in the future, so we encourage you to keep an eye on this space: www.comsol.com/blogs/tag/technical-content/
Since these questions do revolve around a particular set of models which you've developed, we would generally not hazard a blind guess. Of course there can certainly be reasons other than those previously discussed, in particular it sounds like you'd had success with using a tighter solver tolerance and/or timestep, which is entirely reasonable, and is a point that I should have included in the previous posting.
The question of timestepping and tolerances is a quite complex topic and there are volumes of textbooks written on the subject, but the (rather egregiously oversimplified) summary is: The tighter you make the tolerances, the more accurate the solution, and less convergence difficulties you will have. It is a subject that we are planning a series of technical blogs about in the future, so we encourage you to keep an eye on this space: www.comsol.com/blogs/tag/technical-content/
Please login with a confirmed email address before reporting spam
Posted:
7 years ago
Hiii All,
I am also in my learning phase of COMSOL.
I also got this error message at least 50 times while solving the problem.
I checked my values regularly for inconsistency in BCs. Because I was pretty sure that my meshing is perfect.
Finally I took the debugging as a challenge instead of getting irritated.
And what I found was after changing some geometries in my model, I forgot to select the point for pressure point constraint in heat transfer module due to which it gave me error message.
So do check many times your model.
Regards,
Subha
I am also in my learning phase of COMSOL.
I also got this error message at least 50 times while solving the problem.
I checked my values regularly for inconsistency in BCs. Because I was pretty sure that my meshing is perfect.
Finally I took the debugging as a challenge instead of getting irritated.
And what I found was after changing some geometries in my model, I forgot to select the point for pressure point constraint in heat transfer module due to which it gave me error message.
So do check many times your model.
Regards,
Subha