Using Clapeyron along JutulDarcy

[golghanddashtih]

Hi
I am trying to use Clapeyron as thermodynamic engine for JutulDarcy. I have tries it but multiple errors just happens. may you share a sample to see how you define the system so it can run without any problem. However I just tried to change the code so I resolved some errors but it finally endded up to an error that is really confusing me. I share the error here as well. However I just did some modification to reach this extent.

ERROR: Cannot determine ordering of Dual tags Cells() and ForwardDiff.Tag{Clapeyron_Mod.var"#f#184"{PR{BasicIdeal, PRAlpha, NoTranslation, vdW1fRule}, Float64, SubArray{ForwardDiff.Dual{Cells(), Float64, 2}, 1, Matrix{ForwardDiff.Dual{Cells(), Float64, 2}}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}}, ForwardDiff.Dual{ForwardDiff.Tag{Clapeyron_Mod.var"#f#189"{PR{BasicIdeal, PRAlpha, NoTranslation, vdW1fRule}, Float64, SubArray{ForwardDiff.Dual{Cells(), Float64, 2}, 1, Matrix{ForwardDiff.Dual{Cells(), Float64, 2}}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}}, Float64}, Float64, 1}}

As you have certainly have ran a case for this please share your sample for to check at my end to see if it works or not.

Thanks

[longemen3000]

hmm, can you share the complete stacktrace?. It is a Tag ordering issue, and it would help looking at where this error is happening.

[golghanddashtih]

@longemen3000 If you can share an example of yourself that works with JutulDarcy I might be able to better understand the flow and source of problem. My email is: hassan.golghanddashti@adelaide.edu.au

[longemen3000]

Hmm, it seems like some things changed from the last time we tried to integrate JutulDarcy to Clapeyron. i'm gonna check the original code to see if it still runs.

The main error seems to be a tag ordering issue. This is a ForwardDiff.jl problem that appears when 2 instances of ForwardDiff.Dual numbers with different signatures cannot know which one has priority at evaluation time. It seems like JutulDarcy defines a Cells tag that interferes with the default ForwardDiff.Tag type (that we use in Clapeyron).

[golghanddashtih]

Jutuldarcy normally has two different tags one is "cell" the other is "face"....but for both the number and order of partials are preserved the same.

[moyner]

I would be happy to help out if you can share an example. I think the Clapeyron/JutulDarcy interaction should be put under regression testing on my end so that it keeps working.

[golghanddashtih]

@moyner The fact is that I have kept everything as JutulDarcy case except the eos and sys that I just defined as a Clapeyron case. You can replace your fluid part in any JutulDarcy case and expect that works.

using Clapeyon, Jutul,JutulDarcy, MultiComponentFlash
eos = PR(["methane", "nitrogen"])
sys = MultiPhaseCompositionalSystemLV(eos, ( JutulDarcy.LiquidPhase(), JutulDarcy.VaporPhase()))

The fact is that based on Extension part it is expected to work but seems partials are not the same order and making conflict.

[longemen3000]

If i remember correctly, the example code we used to integrate Clapeyron with JutulDarcy was this one:
https://github.com/sintefmath/JutulDarcy.jl/blob/a7da60f069ee02ba60ae9a9f756ce7abd63c1da4/examples/co2_brine_2d_vertical.jl#L45
(we replaced eos for our own EoSModel).

I Will take a look at it tomorrow, probably there are some changes

[golghanddashtih]

Hi. Any outcome?

[longemen3000]

i added the necessary patches to allow the problem to almost execute up until the error, but the error seems to be inherent to how JutulDarcy seems to set their Dual numbers. a more minimal reproducer:

using ForwardDiff: ForwardDiff, Dual
using Jutul
z = [Dual{Cells()}(1.0,0.0,1.0), Dual{Cells()}(0.0,-0.0,-1.0)]
f(x,z) = 0.5*sum(z)/x
ForwardDiff.derivative(_x -> f(_x,z),2.0) #fails with the same error

@moyner is it possible to change how Jutul creates its duals, from ForwardDiff.Dual{<: JutulEntity} to ForwardDiff.Dual{::Type{Tag{<:JutulEntity}} ? when calling ForwardDiff.Tag(f,R), ForwardDiff registers an unique tag count that is used for tag ordering:

https://github.com/JuliaDiff/ForwardDiff.jl/blob/463e8304d9d08a3a21b4d19484767cf79532cefb/src/config.jl#L8-L26

[longemen3000]

The compositional examples seem to run when adding the following lines:

using ForwardDiff, Jutul
ForwardDiff.:≺(::Jutul.Cells, ::Type{<:ForwardDiff.Tag}) = true
ForwardDiff.:≺(::Type{<:ForwardDiff.Tag}, ::Jutul.Cells) = false

But i cannot add those lines in Clapeyron.jl, as they are a type of type piracy (i don't own any of the involved packages).

[moyner]

i added the necessary patches to allow the problem to almost execute up until the error, but the error seems to be inherent to how JutulDarcy seems to set their Dual numbers. a more minimal reproducer:

using ForwardDiff: ForwardDiff, Dual
using Jutul
z = [Dual{Cells()}(1.0,0.0,1.0), Dual{Cells()}(0.0,-0.0,-1.0)]
f(x,z) = 0.5*sum(z)/x
ForwardDiff.derivative(_x -> f(_x,z),2.0) #fails with the same error

@moyner is it possible to change how Jutul creates its duals, from ForwardDiff.Dual{<: JutulEntity} to ForwardDiff.Dual{::Type{Tag{<:JutulEntity}} ? when calling ForwardDiff.Tag(f,R), ForwardDiff registers an unique tag count that is used for tag ordering:

https://github.com/JuliaDiff/ForwardDiff.jl/blob/463e8304d9d08a3a21b4d19484767cf79532cefb/src/config.jl#L8-L26

Thanks for digging into this and setting up the reproducer - I think this is a straightforward solution. I will go over the Dual setup code to use the Tag type instead. Until I get this fixed, the type piracy option can be used if @golghanddashtih wants to get the code running.

[moyner]

I have fixed this on main and added an example to demonstrate usage: https://sintefmath.github.io/JutulDarcy.jl/dev/examples/compositional/clapeyron

Once the next JutulDarcy release is out, it should keep working in the future!