[doublestrong]

I use open temperature and pressure for outlet boudary condition of my case. The values of temperature and pressure should be given by user routine. I wrote user fortran to predict the two values. However, it seems that only one value can be returned to user routine.

My solution is to assemble the values of pressure and temperature into one value. For instance, if the pressure is 123,456 [Pa] and the temperature is 789[K], then I make my subroutine return 123456.789. This number can be splitted to two numbers using int(123456.789) and 1000*(123456.789 - int(123456.789)) in CFX expressions.

However, the problem is that the default type of return value in user routine is REAL. This type makes 123456.789 be returned as 123456.8, therefore the temperature cannot be accurate anymore. Is there a way to make the default type of return value to be double precision? Or can anyone provide me with other solutions?

Thanks!

[ghorrocks]

That looks like a horrible way to pass variables into CFX. Why can't you split it into two fortran routines and get one to pass the pressure variable and one to pass the temperature variable?

[doublestrong]

Dear Glenn,

Thank you! The problem can be solved if I split it into two fortran routines. But I prefer to use just one fortran routine if two values can be returned in one fortran routine in any way. My fortran routine is a one-dimensional simulation. There are hundreds of iterations in the fortran routine every physical time step. And the temperature and pressure actually can be figured out in one run together. If I split it into two fortran routines, I soppuse the routines are almost the same but return different variables. I am worried about the extra computation time caused by another fortran routine. I will think about using two routines if there is no other way to solve the problem. Thanks again.

[ghorrocks]

By far the easiest way is just to calculate it twice. I would check the computational time of this option before discarding it. Anything else will be significantly more difficult to implement.

[techtuner]

Doublestrong, you can pass your results as array of dimensionless values. Later, in CEL just choose appropriate index of your required value.

Sent from my M040 using CFD Online Forum mobile app

[LorenzoSaintDubois]

Funky little problem for the afternoon...
Every effort to shave off some computational time is noble, in my opinion.
Solution shown below. Do try to test this thoroughly though.
Best of luck!
LSD

program test
implicit none
doubleprecision:: T=789.12,P=123456.1,F,s=1000000,tn,pn
character*(100):: ct,cp,cf


! Encode T and P into doubleprecision F.
ct=''
write(ct,*)nint(T)
write(*,*)ct
ct=adjustl(trim(ct))

cp=''
write(cp,*)nint(P)
write(*,*)cp
cp=adjustl(trim(cp))

cf=''
cf=adjustl(trim(ct))//adjustl(trim(cp))
write(*,*)cf

read(cf,*)F
write(*,*)F

! Retrieve tn and pn from F.
tn=int(F/s)
write(*,*)tn
pn=F-tn*s
write(*,*)pn

end program test

[Opaque]

I am still puzzled by this question.

What is the point of returning multiple output for CFX UserFortran CEL function when the left hand side of the expression only takes one input.

The usage of CEL expressions is as follows

Parameter of Interest = MyCELFunction (P, T, ...)

How can the software split the multiple outputs w/o indication which output goes into which parameter ? Sure, we could add an index to indicate which single output to return, say

Parameter1 = MyCELFunction (1, P, T, ...)
Parameter2 = MyCELFunction (2, P, T, ...)

That will definitely work; however, the software still has to call the same function twice (if that is the number of possible outputs), and the time saving logic can only be embedded into the function.

It seems focus has been lost from solving the engineering problem at hand, which led to a CFD model, and time is being spent on an optimization of a low level (though time consuming) detail of the big picture.

Advice, create a local data area in the ANSYS CFX memory, or equivalent, store the time consuming results and reuse the data whenever it has already been computed. Alternatively, find a much faster algorithm to recompute the data.

My 2 cents,

[LorenzoSaintDubois]

The way I see it, the user has a single computationaly expensive Fortran function that however delivers both needed variable values for the BC definition. So the trick is to call this expensive function once and somehow give CFX the values of both the temperature and the pressure using this one function call. So, I guess, you define a CFX user-variable (say, F) that is calculated using the Fortran function and holds the information of both T and P. Then using a couple of expressions you extract T and P from the values of F and set these expressions to the BC definition. So the solver calls the Fortran function once to populate F and then uses the available values on the grid for calculating T and P.

Your suggestion of using the CFX MMS is the most elegant way to go, but definitely not trivial and I'm not sure whether the documentation is sufficient for figuring out how to do this. I'd have to call support if I'd go that way.

LSD