[Action.Kamen]

Hello Foamers ,


I am trying to implement sigmoid function in OpenFOAM-v1912 using

Foam::exp() and Foam::pow().

volScalarField tmpVar = nGAGR*scalar(-1) + averageGAR;

sigmoidGAGR=Foam::pow((1+Foam::exp(tmpVar)),-1);

here nGAGR and averageGAGR both are dimensionless volScalarField. But the code wont work and stops at the second line.


What am I doing wrong here, or is there any other way to implement sigmoid function.

[HPE]

Hi,

- What is the error?
- Sigmoid function is a generic name, which expression you try to code?
- If the value of "tmpVar" becomes very large, the computation of "exp" function may throw floating point overflow error.

[Action.Kamen]

Hi HPE,


- There is no error displayed in the terminal, the simulation just stops after first time-interval. But when I comment out the line the simulation starts running (and does not stops ).


- Sigmoid function : The basic idea to use sigmoid function is that to take in some large number and pump out result that is either 0 or 1(or something in between). Here nGAGR is the order of 10^11 and averageGAR is in the order of 10^6.



- Value should not be a problem since machine I'm using is a 64-bit one, and the macros DBL_MIN and DBL_MAX give the ranges as 2.22507e-308 and 1.79769e+308 respectively. And the range I'm working with is around 10^-8 to 10^11.


The thing I want to do is basically assign 0s and 1s all over my mesh into volScalarField. There are two regions in the mesh one where nGAGR is around 10^11 and other where nGAGR is around 10^-8. Now that could have been easily achieved by looping all over the mesh and using if else. But I have a quite large mesh and deltaT is pretty small 10^-5. And what I learned from scientific computing course is that if/else and loops are lot slower than compared to mathematical statements. So I'm trying to implement sigmoid function


PS : I have attached image of the sigmoid function which I want to implement