Conservative transfers

[vincentlaboure]

Rationale

Some customers are increasingly requesting conservative transfers.

Description

A (scalable and transfer-agnostic) option could perhaps be to specify - in the Transfers block - the postprocessors on both the master and sub app that should match. The target fields could then be scaled using the ratio of these postprocessors.

Impact

Key quantities of interest could be preserved during transfers.

[YaqiWang]

My understanding: what @vincentlaboure wants are two more parameters in the transfer base class

postprocessor_to_be_preserved_on_master_side = 'pp_on_master'
postprocessor_to_be_preserved_on_subapp_side = 'pp_on_subapp'

When a transfer is done, on the to transfer side, do a postprocessor evaluation, get one scalar value v2, then use the ratio between the from transfer side postprocessor value v1 and v2 as a scaling factor to redo the transfer (assuming v2 is linear). This sounds simple but can be quite complicated.

[fdkong]

Looks like we need to execute postprocessor twice. But we only trigger this when users speficy to use the conservative transfer.

[vincentlaboure]

@fdkong Thanks for looking at this! @YaqiWang's summary seems accurate.

I think a good test problem could be as follows:

Master app defined for 0 < x < 1, 0 < y < 1; block 0 for 0 < x < 0.5, block 1 for 0.5 < x < 1
Sub app defined for 0 < x < 0.5, 0 < y < 1; block 10 for 0.2 < x < 0.3 and 0.2 < y < 0.8, block 11 otherwise
Two sub abbs: positions = '0 0 0 0.5 0 0'

The goal would be to:

• Transfer power density field p from master app block 0 to sub app 0, block 10 such that \int_{master, block 0} p dx = \int_{sub0, block 10} p dx

• Transfer power density field p from master app block 1 to sub app 1, block 10 such that \int_{master, block 1} p dx = \int_{sub1, block 10} p dx

• Transfer power density field q from sub app 0, block 10 to master app block 0 such that \int_{master, block 0} q dx = \int_{sub0, block 10} q dx

• Transfer power density field p from sub app 1, block 10 to master app block 1 such that \int_{master, block 1} q dx = \int_{sub1, block 10} q dx

A first step might be to do that same problem without block restriction on the sub apps.

Does this make more sense? If you want, I can create some input files for that proposed test problem.

[fdkong]

@fdkong Thanks for looking at this! @YaqiWang's summary seems accurate.

I think a good test problem could be as follows:

Master app defined for 0 < x < 1, 0 < y < 1; block 0 for 0 < x < 0.5, block 1 for 0.5 < x < 1
Sub app defined for 0 < x < 0.5, 0 < y < 1; block 10 for 0.2 < x < 0.3 and 0.2 < y < 0.8, block 11 otherwise
Two sub abbs: positions = '0 0 0 0.5 0 0'

The goal would be to:

• Transfer power density field p from master app block 0 to sub app 0, block 10 such that \int_{master, block 0} p dx = \int_{sub0, block 10} p dx
• Transfer power density field p from master app block 1 to sub app 1, block 10 such that \int_{master, block 1} p dx = \int_{sub1, block 10} p dx
• Transfer power density field q from sub app 0, block 10 to master app block 0 such that \int_{master, block 0} q dx = \int_{sub0, block 10} q dx
• Transfer power density field p from sub app 1, block 10 to master app block 1 such that \int_{master, block 1} q dx = \int_{sub1, block 10} q dx

A first step might be to do that same problem without block restriction on the sub apps.

Does this make more sense? If you want, I can create some input files for that proposed test problem.

Sounds good for me.

@vincentlaboure, please create some input files, and added to my PR #13060. I will take care of them.

[vincentlaboure]

@fdkong I created a branch showing input files that should make a really good test. I am aware that this might not be a trivial task but it would extremely useful for us and our customers.

Please let me know if you have any questions about it.

[vincentlaboure]

The input files are here