Hi Ayla,

I've included a SCIRun v4.7 (latest build) srn file, as well as all the data field files you need to run it:

http://www.ashuc.com/FEMexample.zip (~148.8 MB)

The complete network is far more complicated than the one I've included, but hopefully this will give you enough of an idea of what's going on.

Inside the network I've tried to label all the modules so you can follow it.  In the ChooseField module, Port 0 will use the v4.7 CFDT module which doesn't work, and Port 1 will load the v3.0.2 CFDT module which does work.  You will see in the UI for SolveLinearSystem that one solution converges while the other doesn't.

I hope this helps.  Thanks again!

--Ashu Chaturvedi

ashu.chaturvedi@gmail.com

On Jun 19, 2012, at 7:55 PM, Ayla Khan wrote:

Hi Ashu,

I had a look at the fields you sent, and couldn't see much difference between the field converted by SCIRun 4.7 and SCIRun 3.0.2 other than the data max values are different (the values are essentially the same in the input field and the SCIRun 4.7 converted field) as shown in the attached image. The algorithm code for a converting mesh with a constant basis to a mesh of the same size, such as in this case, is pretty straightforward - it's a copy with a C++ static_cast to the new type.

Can you send a sample network for your FEM simulation - there could be something else going on here.

Thanks,

Ayla

<convertfielddatatype.png>

On Jun 5, 2012, at 12:34 PM, Ashu Chaturvedi wrote:

Hi Ayla,

Sorry to bring this up again, but it appears this module still isn't working for me.  I'm having pretty much the same issues as before.  You can read my original email below to remember the details.

I still have the problem where ConvertFieldDataType doesn't allow my FEM solutions to converge.  I've saved two field files, which are the outputs to the ConvertFieldDataType (CFDT) module in v3.0.2 (works) and v4.7 (broken).  Also included is the input field file to CFDT in v4.7.

Since these are large files, you can download it here:

http://www.ashuc.com/ConvertFieldDataType_2.zip

I'm basically using CFDT to convert the input field from datatype double to int.  Thanks,

--Ashu Chaturvedi

ashu.chaturvedi@gmail.com

On Feb 23, 2012, at 5:20 PM, Ayla Khan wrote:

Hi Ashu,

We found the bug, patched and rebuilt SCIRun v4.5. The 64-bit OS X binary installer is available on the SCIRun download page, and the 32-bit binary installer will be up shortly. Thank you for reporting this bug and let us know if you run into any more problems.

Ayla

On Thu, 23 Feb 2012 17:10:49 -0500

Ashu Chaturvedi <ashu.chaturvedi@gmail.com> wrote:

Hi Ayla,

I was wondering if you had any updates regarding this bug?  Thanks,

--Ashu Chaturvedi

ashu.chaturvedi@gmail.com

On Feb 17, 2012, at 6:30 PM, Ayla Khan wrote:

Hi Ashu,

I was able to reproduce your bug, and I'm looking into the differences between the modules in SCIRun v3.0.2 and the latest version.

Ayla

On Feb 16, 2012, at 2:22 PM, Ashu Chaturvedi wrote:

Hey,

I have a simple SCIRun network which reads in two input files and solves for the FEM voltage solutions in space.  In SCIRun v3.0.2 the solutions converge correctly in the SolveLinearSystem module. However, in SCIRun v4.5 they do not; in fact, the cumulative error goes off to infinity.

I have narrowed down the problem to the ConvertFieldDataType module. When I use the v4.5 module, further downstream I get a warning inside ApplyFEMVoltageSource that says:

"WARNING: (BioPSE_Forward_ApplyFEMVoltageSource_0) The input field doesn't contain Dirichlet boundary conditions."

This warning does not appear in the v3.0.2 ApplyFEMVoltageSource module.

In v4.5, when I read in the saved output from the ConvertFieldDataType module from v3.0.2, that warning does not appear and the FEM solutions converge correctly in v4.5.

You can download (87 mb) the example network and data which depicts this problem:

http://www.ashuc.com/ConvertFieldDataType.zip

Thanks in advance for your help!

--Ashu Chaturvedi

ashu.chaturvedi@gmail.com