Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.

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.

How to delete unwanted stored solutions?

Please login with a confirmed email address before reporting spam

Hello,
I am simulating large 3D RF problem in time domain in version3.5a. And I found my whole stored file is as large as 40GB, since I stored around 600 time steps. And I want to delete the last 100 time steps to save some memory of the server without recalculate the whole process.

I would be very happy if someone knows how to do it.

Best wishes,

20 Replies Last Post 27 août 2013, 00:39 UTC−4
Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 1 oct. 2010, 03:27 UTC−4
Hi

unfortunateely I do not know how to, I often run long simualtions in steps to avoid too large files, ie the first 100 sec, then the next 100<t<200 etc, by using the last of the previous run as starting point to the next run

--
Good luck
Ivar
Hi unfortunateely I do not know how to, I often run long simualtions in steps to avoid too large files, ie the first 100 sec, then the next 100

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 1 oct. 2010, 12:34 UTC−4
Hi,

Thanks a lot for your answer! But can you explain how to set the last of the previous run as starting point to the next run?

Best wishes,
Hi, Thanks a lot for your answer! But can you explain how to set the last of the previous run as starting point to the next run? Best wishes,

Andrew Prudil Nuclear Materials

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 1 oct. 2010, 17:28 UTC−4
To do this:
Run your first simulation for your first block of time. Then go to solver manager and change initial value to "current solution". Then go to solver parameters and change the time range as desired.

I have also found that setting "Times to store in output" to "specified times" and "time steps taken by solver" to "intermediate" only can greatly reduce the file size by only saving at the times from the times input.
To do this: Run your first simulation for your first block of time. Then go to solver manager and change initial value to "current solution". Then go to solver parameters and change the time range as desired. I have also found that setting "Times to store in output" to "specified times" and "time steps taken by solver" to "intermediate" only can greatly reduce the file size by only saving at the times from the times input.

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 2 oct. 2010, 09:06 UTC−4

To do this:
Run your first simulation for your first block of time. Then go to solver manager and change initial value to "current solution". Then go to solver parameters and change the time range as desired.

I have also found that setting "Times to store in output" to "specified times" and "time steps taken by solver" to "intermediate" only can greatly reduce the file size by only saving at the times from the times input.


Many thanks! But I am confused some settings. In solver manager, what is the difference if I set initial value to "current solution" and "stored solution", since COMSOL will store the previous time block results include current solution?

Another question is that, in the window of "solver parameter", go to the tab of "time stepping", there is a setting for "time step tuning". Normally the time step should satisfy the criterion of getting stable results.

dt<dx/c c: the velocity of light dx: the mesh size dt: the time step

Of course the mesh inside the domain is not homogeneous, so is it correct, if I set "initial time step" corresponds to the minimum mesh size and "Maximum time step" corresponds to the maximum mesh size?

However, I do not know what does COMSOL do internally if I choose "Auto" there?
[QUOTE] To do this: Run your first simulation for your first block of time. Then go to solver manager and change initial value to "current solution". Then go to solver parameters and change the time range as desired. I have also found that setting "Times to store in output" to "specified times" and "time steps taken by solver" to "intermediate" only can greatly reduce the file size by only saving at the times from the times input. [/QUOTE] Many thanks! But I am confused some settings. In solver manager, what is the difference if I set initial value to "current solution" and "stored solution", since COMSOL will store the previous time block results include current solution? Another question is that, in the window of "solver parameter", go to the tab of "time stepping", there is a setting for "time step tuning". Normally the time step should satisfy the criterion of getting stable results. dt

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 3 oct. 2010, 03:09 UTC−4
Hi

two partial answers:

current and stored solution are not necesarily the same thing, often you must define a "store solution" to have it saved on file (in 3.5 as well as 4). The current solution is mostly the last in memory, stored or not, and will depend on the segregated / parametric sequences defined

time stepping, normally the solver is rather clever and check the slope and the previous values to define a strategy to set it's own time steps, and hence interpolates the values desired in between. This works very well for transient phenomena with some transient effect at start and then dying out effects towards higehr time values. But for periodic excitation (normally one uses harmonic developments in those cases) the "automatic" time sequencer get fooled and will end up undersampling the solution and miss oscillations, or sudden phenomena appearing after some time.

For these cases you should use either "intermediate" or "strict" time steps: intermedaite ensure that you solve at least one case in between the ones YOU define, "strict" takes strictly the time steps YOU define.

Then in "automatic" mode you still have some freedom to set further limits such as step variation domain, but often the strict is just as good.

The best is to make up some simple cases i.e. cantelever beam oscillating and play with the transient time step values

Hope this helps

--
Good luck
Ivar
Hi two partial answers: current and stored solution are not necesarily the same thing, often you must define a "store solution" to have it saved on file (in 3.5 as well as 4). The current solution is mostly the last in memory, stored or not, and will depend on the segregated / parametric sequences defined time stepping, normally the solver is rather clever and check the slope and the previous values to define a strategy to set it's own time steps, and hence interpolates the values desired in between. This works very well for transient phenomena with some transient effect at start and then dying out effects towards higehr time values. But for periodic excitation (normally one uses harmonic developments in those cases) the "automatic" time sequencer get fooled and will end up undersampling the solution and miss oscillations, or sudden phenomena appearing after some time. For these cases you should use either "intermediate" or "strict" time steps: intermedaite ensure that you solve at least one case in between the ones YOU define, "strict" takes strictly the time steps YOU define. Then in "automatic" mode you still have some freedom to set further limits such as step variation domain, but often the strict is just as good. The best is to make up some simple cases i.e. cantelever beam oscillating and play with the transient time step values Hope this helps -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 7 oct. 2010, 04:56 UTC−4

Hi

two partial answers:

current and stored solution are not necesarily the same thing, often you must define a "store solution" to have it saved on file (in 3.5 as well as 4). The current solution is mostly the last in memory, stored or not, and will depend on the segregated / parametric sequences defined

time stepping, normally the solver is rather clever and check the slope and the previous values to define a strategy to set it's own time steps, and hence interpolates the values desired in between. This works very well for transient phenomena with some transient effect at start and then dying out effects towards higehr time values. But for periodic excitation (normally one uses harmonic developments in those cases) the "automatic" time sequencer get fooled and will end up undersampling the solution and miss oscillations, or sudden phenomena appearing after some time.

For these cases you should use either "intermediate" or "strict" time steps: intermedaite ensure that you solve at least one case in between the ones YOU define, "strict" takes strictly the time steps YOU define.

Then in "automatic" mode you still have some freedom to set further limits such as step variation domain, but often the strict is just as good.

The best is to make up some simple cases i.e. cantelever beam oscillating and play with the transient time step values

Hope this helps

--
Good luck
Ivar


Thanks a lot! By using initial value with current solution, I have decreased my file from 50GB to 10GB.
[QUOTE] Hi two partial answers: current and stored solution are not necesarily the same thing, often you must define a "store solution" to have it saved on file (in 3.5 as well as 4). The current solution is mostly the last in memory, stored or not, and will depend on the segregated / parametric sequences defined time stepping, normally the solver is rather clever and check the slope and the previous values to define a strategy to set it's own time steps, and hence interpolates the values desired in between. This works very well for transient phenomena with some transient effect at start and then dying out effects towards higehr time values. But for periodic excitation (normally one uses harmonic developments in those cases) the "automatic" time sequencer get fooled and will end up undersampling the solution and miss oscillations, or sudden phenomena appearing after some time. For these cases you should use either "intermediate" or "strict" time steps: intermedaite ensure that you solve at least one case in between the ones YOU define, "strict" takes strictly the time steps YOU define. Then in "automatic" mode you still have some freedom to set further limits such as step variation domain, but often the strict is just as good. The best is to make up some simple cases i.e. cantelever beam oscillating and play with the transient time step values Hope this helps -- Good luck Ivar [/QUOTE] Thanks a lot! By using initial value with current solution, I have decreased my file from 50GB to 10GB.

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 20 janv. 2011, 05:36 UTC−5
Hi,
Now I try with different time steps as you mentioned "I often run long simualtions in steps to avoid too large files, ie the first 100 sec, then the next 100<t<200 etc, by using the last of the previous run as starting point to the next run "

But the thing is that when I run the next time step 100<t<200 in the same file, it overwrite the previous result of the first 100 sec. Is this true? What can I do if I want to save the result of the first 100sec, and run the next time step 100<t<200 by using "initial value> current solution"?
Or do I have to save different time steps in different file?
Hi, Now I try with different time steps as you mentioned "I often run long simualtions in steps to avoid too large files, ie the first 100 sec, then the next 100

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 20 janv. 2011, 05:38 UTC−5
sorry,posted by mistake
[i]sorry,posted by mistake[/i]

Muhammad Mohsin Rehman

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 22 nov. 2011, 07:52 UTC−5
Hi Ivar,

Can you please let me know how can I run the simulation in steps i.e. how can I use the result of one step of simulation for the next step?

As far as I have understood, to run a simulation for one year I should proceed as follows:

Firstly, run the simulation for 1 month (86400*30), get the output and then write (86400*30,10000,86400*60) in the Times section and so on. In this way it will give me results for first month and then for next month and then so on.

Am I right? If not, please rectify me...


Any help will be highly appreciated!
Hi Ivar, Can you please let me know how can I run the simulation in steps i.e. how can I use the result of one step of simulation for the next step? As far as I have understood, to run a simulation for one year I should proceed as follows: Firstly, run the simulation for 1 month (86400*30), get the output and then write (86400*30,10000,86400*60) in the Times section and so on. In this way it will give me results for first month and then for next month and then so on. Am I right? If not, please rectify me... Any help will be highly appreciated!

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 22 nov. 2011, 08:06 UTC−5
Hi

normally there is the "continue" for restarting where leftover, but I haven't really tried it out. You must ensure that you have a default solver sequence, and then change the time sequence in the main solver steps. But is this really "smaller" in storage size than solving in one long series ? Not sure. If you clear the result you have lost your starting point

--
Good luck
Ivar
Hi normally there is the "continue" for restarting where leftover, but I haven't really tried it out. You must ensure that you have a default solver sequence, and then change the time sequence in the main solver steps. But is this really "smaller" in storage size than solving in one long series ? Not sure. If you clear the result you have lost your starting point -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 12 avr. 2012, 03:35 UTC−4
I had the same problem,

running a time dependent solver with on off heating (very long calculations >7 days)
but I was only interested in the end situation to start a new study with the cooling.

Because my first study was extensivly large I created a second study, with a very small time (0.1s) so physically nothing changed significantly to the results.

Study 2 was calculated and I cleared the results from study 1. It worked.
I saved the model and started a new study3 to do the new evaluations.

Does anybody knows that what I'm doing is correct, or do I make a major mistake?
My probes are having the same temperatures when doing the next calculations.

Regards
Hans
I had the same problem, running a time dependent solver with on off heating (very long calculations >7 days) but I was only interested in the end situation to start a new study with the cooling. Because my first study was extensivly large I created a second study, with a very small time (0.1s) so physically nothing changed significantly to the results. Study 2 was calculated and I cleared the results from study 1. It worked. I saved the model and started a new study3 to do the new evaluations. Does anybody knows that what I'm doing is correct, or do I make a major mistake? My probes are having the same temperatures when doing the next calculations. Regards Hans

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 12 avr. 2012, 16:13 UTC−4
Hi

I believe , if it's "only" the temperature field you are after, that you can save the T result into a mapped data set (export), and then import it touse it as initial condition called explicitely. But I must admit I havnt tried something like this, as I'm mostly not running such long sequences, for most periodic cases you have the harmonic solver, for linear cases the time - frequency solve rcan do, in certan cases, it calculates one period in the time series, wth a rather high resoliution, and then uses the rms values for a time solving case, you get good results for most diffuxion prolems tha react slowly to the initial oscillation frequency (but its not valid for all model cases)

--
Good luck
Ivar
Hi I believe , if it's "only" the temperature field you are after, that you can save the T result into a mapped data set (export), and then import it touse it as initial condition called explicitely. But I must admit I havnt tried something like this, as I'm mostly not running such long sequences, for most periodic cases you have the harmonic solver, for linear cases the time - frequency solve rcan do, in certan cases, it calculates one period in the time series, wth a rather high resoliution, and then uses the rms values for a time solving case, you get good results for most diffuxion prolems tha react slowly to the initial oscillation frequency (but its not valid for all model cases) -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 25 avr. 2012, 10:37 UTC−4

Hi

I believe , if it's "only" the temperature field you are after, that you can save the T result into a mapped data set (export), and then import it touse it as initial condition called explicitely. But I must admit I havnt tried something like this, as I'm mostly not running such long sequences, for most periodic cases you have the harmonic solver, for linear cases the time - frequency solve rcan do, in certan cases, it calculates one period in the time series, wth a rather high resoliution, and then uses the rms values for a time solving case, you get good results for most diffuxion prolems tha react slowly to the initial oscillation frequency (but its not valid for all model cases)

--
Good luck
Ivar


Ivar, thanks a lot for the intersting-useful information.
With the use of an interpolation function I could make a reference to the temperatures.
Dealing with some other problems during fluid flow (not using boundary layers resulting in no exit speed of the air) and some time between further comsolling (due to other priorities) I came to a sufficient good result.
Now we can fine tune the system and look if it is needed to have the air cooling or not.

Regards
Hans
[QUOTE] Hi I believe , if it's "only" the temperature field you are after, that you can save the T result into a mapped data set (export), and then import it touse it as initial condition called explicitely. But I must admit I havnt tried something like this, as I'm mostly not running such long sequences, for most periodic cases you have the harmonic solver, for linear cases the time - frequency solve rcan do, in certan cases, it calculates one period in the time series, wth a rather high resoliution, and then uses the rms values for a time solving case, you get good results for most diffuxion prolems tha react slowly to the initial oscillation frequency (but its not valid for all model cases) -- Good luck Ivar [/QUOTE] Ivar, thanks a lot for the intersting-useful information. With the use of an interpolation function I could make a reference to the temperatures. Dealing with some other problems during fluid flow (not using boundary layers resulting in no exit speed of the air) and some time between further comsolling (due to other priorities) I came to a sufficient good result. Now we can fine tune the system and look if it is needed to have the air cooling or not. Regards Hans

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 3 mai 2012, 08:38 UTC−4
Now after some intersting studies are done I'm wondering if I could do half of the simulation (heating has two symmetry axes) while cooling has only 1 symmetry axes.

How I would do that is simulate 1/4th of the mould.
And mirror 1/4th in Z-direction leading to a half mould.
Question is can I easily mirror the initial temperatures? Now they are determined by the coordinates and an interpolation function.

Is this possible?
Now after some intersting studies are done I'm wondering if I could do half of the simulation (heating has two symmetry axes) while cooling has only 1 symmetry axes. How I would do that is simulate 1/4th of the mould. And mirror 1/4th in Z-direction leading to a half mould. Question is can I easily mirror the initial temperatures? Now they are determined by the coordinates and an interpolation function. Is this possible?

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 17 mai 2012, 05:50 UTC−4
Hi,

I am stuck at the same thing. I have a bunch of simulations which have large files and unwanted parametric steps. I do quasi-static simulations with plasticity. Is there any way discovered to delete unwanted solutions (other than connecting to MATLAB, because I could not load the file in matlab due to its java heap limitations)?

Would be of great help!

Best regards,
Kodanda
Hi, I am stuck at the same thing. I have a bunch of simulations which have large files and unwanted parametric steps. I do quasi-static simulations with plasticity. Is there any way discovered to delete unwanted solutions (other than connecting to MATLAB, because I could not load the file in matlab due to its java heap limitations)? Would be of great help! Best regards, Kodanda

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 17 mai 2012, 07:36 UTC−4
Hi

indeed you have rotate and mirror solution in the Data set operations, then you can export the data or refer to it directly. By the way referring to a data with a symmetry axis is "just" some equation writing, i.e. to get the symmetry w.r.t. the x=0 Y-Z plane, you use an abs(x) as reference to acess your data map

--
Good luck
Ivar
Hi indeed you have rotate and mirror solution in the Data set operations, then you can export the data or refer to it directly. By the way referring to a data with a symmetry axis is "just" some equation writing, i.e. to get the symmetry w.r.t. the x=0 Y-Z plane, you use an abs(x) as reference to acess your data map -- Good luck Ivar

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 17 mai 2012, 07:37 UTC−4
Hi

deleting solutions or clear them can be done at certain levels, probably too at the data set level. but there should be some simpler ways to export only some results, and then import them again

--
Good luck
Ivar
Hi deleting solutions or clear them can be done at certain levels, probably too at the data set level. but there should be some simpler ways to export only some results, and then import them again -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 22 mai 2012, 04:48 UTC−4

Hi

deleting solutions or clear them can be done at certain elvels, probably too thte data set level. but there should be some simpler ways to export only some results, and then import them again

--
Good luck
Ivar


Hi Ivar,

thanks for your reply. You have mentioned that some solutions can be deleted at certain levels. could you please give some examples?

thanks
Kodanda
[QUOTE] Hi deleting solutions or clear them can be done at certain elvels, probably too thte data set level. but there should be some simpler ways to export only some results, and then import them again -- Good luck Ivar [/QUOTE] Hi Ivar, thanks for your reply. You have mentioned that some solutions can be deleted at certain levels. could you please give some examples? thanks Kodanda

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 22 mai 2012, 05:31 UTC−4
Hi

if you have run a complex model with many studies, you can clear "ALL solutions" by Edit Clear All solutions" but you can also, go down to the solver configuration node and "clear solution" of a given study. or you go directly into the Data set and Delete a Data set (this also Deletes all your linked postprocessing nodes, be aware !, as will the Solver Configuration "Delete Solvers"

--
Good luck
Ivar
Hi if you have run a complex model with many studies, you can clear "ALL solutions" by Edit Clear All solutions" but you can also, go down to the solver configuration node and "clear solution" of a given study. or you go directly into the Data set and Delete a Data set (this also Deletes all your linked postprocessing nodes, be aware !, as will the Solver Configuration "Delete Solvers" -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 27 août 2013, 00:39 UTC−4
Hi Ivar,
when you say using current solution instead of stored solution, does that mean I dont need to store the solution of my previous study (enabling store solution) which I want to use the results in the current study? or it simply means that run a study,store solution and in the new study define the previous study as initial values but use the current solution right?

thanks for your help
Hi Ivar, when you say using current solution instead of stored solution, does that mean I dont need to store the solution of my previous study (enabling store solution) which I want to use the results in the current study? or it simply means that run a study,store solution and in the new study define the previous study as initial values but use the current solution right? thanks for your help

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.