SMALL, GREAT and ROOTVSMALL - what is that?

Dommy · July 18, 2013, 18:18

Dear Foamers!

I'm working on Euler/Lagrange validation cases at the moment.
While looking into Disoersion Models I discoverd something in the Code - SMALL, GREAT and ROOTVSMALL.

Some examples (from /openfoam211/src/lagrangian/intermediate/submodels/Kinematic/DispersionModel/GradientDispersionRAS/GradientDispersionRAS.C)

Code:

 
scalar sigma = sqrt(2.0*k/3.0);
vector dir = -gradk/(mag(gradk) + SMALL);

Code:

 else
    {
        tTurb = GREAT;
        UTurb = vector::zero;
    }

Code:

const scalar epsilon =
        this->epsilonPtr_->internalField()[cellI] + ROOTVSMALL;
    const vector& gradk = this->gradkPtr_->internalField()[cellI];

In order to completly understand what I'm doing I need to know what this model is doing. Thanks a lot in advance!

Best,
Thomas

Lieven · July 18, 2013, 18:31

Hi Thomas,

SMALL, VSMALL, LARGE, ... are simply constants defined in src/OpenFOAM/primitives/Scalar/scalar/scalar.H and src/OpenFOAM/primitives/Scalar/floatScalar/floatScalar.H

In summary:

Code:

GREAT = 1.0e+6;
VGREAT = 1.0e+37;
ROOTVGREAT = 1.0e+18;
SMALL = 1.0e-6;
VSMALL = 1.0e-37;
ROOTVSMALL = 1.0e-18;

When you have something like 1/(x+SMALL) in the code, the SMALL simply prevents an floating point error from occurring when x = 0.

Cheers,

Lieven

l_r_mcglashan · July 18, 2013, 18:47

Slight correction; the values depend on whether openfoam is compiled in single or double precision.

Lieven · July 18, 2013, 18:55

How indeed, my apologies for this laxity. The values mentioned above are for a double precision compilation.

Cheers

L

Dommy · July 19, 2013, 05:47

Thanks for the quick reply - it all makes sense now!

l_r_mcglashan · July 19, 2013, 05:48

Correction; the values above are for single precision.

ndev · March 22, 2017, 20:53

I know this is old, but sometimes these posts get a lot of traffic. Just to eliminate any confusion, the double precision values are:

static const doubleScalar doubleScalarGREAT = 1.0e+15;
static const doubleScalar doubleScalarVGREAT = 1.0e+300;
static const doubleScalar doubleScalarROOTVGREAT = 1.0e+150;
static const doubleScalar doubleScalarSMALL = 1.0e-15;
static const doubleScalar doubleScalarVSMALL = 1.0e-300;
static const doubleScalar doubleScalarROOTVSMALL = 1.0e-150;

evamaria · February 20, 2020, 10:12

Be aware that the hard-coded scalar limits have been replaced: https://github.com/OpenFOAM/OpenFOAM...9d7f1ee5912713

July 18, 2013, 18:18	SMALL, GREAT and ROOTVSMALL - what is that?	#1
Dommy New Member Thomas F. Join Date: Jun 2013 Location: Graz, Austria Posts: 14 Rep Power: 13	Dear Foamers! I'm working on Euler/Lagrange validation cases at the moment. While looking into Disoersion Models I discoverd something in the Code - SMALL, GREAT and ROOTVSMALL. Some examples (from /openfoam211/src/lagrangian/intermediate/submodels/Kinematic/DispersionModel/GradientDispersionRAS/GradientDispersionRAS.C) Code: scalar sigma = sqrt(2.0*k/3.0); vector dir = -gradk/(mag(gradk) + SMALL); Code: else { tTurb = GREAT; UTurb = vector::zero; } Code: const scalar epsilon = this->epsilonPtr_->internalField()[cellI] + ROOTVSMALL; const vector& gradk = this->gradkPtr_->internalField()[cellI]; In order to completly understand what I'm doing I need to know what this model is doing. Thanks a lot in advance! Best, Thomas

July 18, 2013, 18:31		#2
Lieven Senior Member Lieven Join Date: Dec 2011 Location: Leuven, Belgium Posts: 299 Rep Power: 23	Hi Thomas, SMALL, VSMALL, LARGE, ... are simply constants defined in src/OpenFOAM/primitives/Scalar/scalar/scalar.H and src/OpenFOAM/primitives/Scalar/floatScalar/floatScalar.H In summary: Code: GREAT = 1.0e+6; VGREAT = 1.0e+37; ROOTVGREAT = 1.0e+18; SMALL = 1.0e-6; VSMALL = 1.0e-37; ROOTVSMALL = 1.0e-18; When you have something like 1/(x+SMALL) in the code, the SMALL simply prevents an floating point error from occurring when x = 0. Cheers, Lieven louisgag, gautami, apple-tree and 30 others like this.

July 18, 2013, 18:47		#3
l_r_mcglashan Senior Member Laurence R. McGlashan Join Date: Mar 2009 Posts: 370 Rep Power: 23	Slight correction; the values depend on whether openfoam is compiled in single or double precision. vivek05 likes this. __________________ Laurence R. McGlashan :: Website

July 18, 2013, 18:55		#4
Lieven Senior Member Lieven Join Date: Dec 2011 Location: Leuven, Belgium Posts: 299 Rep Power: 23	How indeed, my apologies for this laxity. The values mentioned above are for a double precision compilation. Cheers L vivek05 likes this.

July 19, 2013, 05:48		#6
l_r_mcglashan Senior Member Laurence R. McGlashan Join Date: Mar 2009 Posts: 370 Rep Power: 23	Correction; the values above are for single precision. __________________ Laurence R. McGlashan :: Website

July 19, 2013, 05:47		#5
Dommy New Member Thomas F. Join Date: Jun 2013 Location: Graz, Austria Posts: 14 Rep Power: 13	Thanks for the quick reply - it all makes sense now!

March 22, 2017, 20:53	Double Precision Values	#7
ndev New Member Nate Join Date: Oct 2013 Location: Amherst, MA Posts: 13 Rep Power: 13	I know this is old, but sometimes these posts get a lot of traffic. Just to eliminate any confusion, the double precision values are: static const doubleScalar doubleScalarGREAT = 1.0e+15; static const doubleScalar doubleScalarVGREAT = 1.0e+300; static const doubleScalar doubleScalarROOTVGREAT = 1.0e+150; static const doubleScalar doubleScalarSMALL = 1.0e-15; static const doubleScalar doubleScalarVSMALL = 1.0e-300; static const doubleScalar doubleScalarROOTVSMALL = 1.0e-150; xpqiu, serles, mrishi and 7 others like this.

February 20, 2020, 10:12		#8
evamaria New Member Join Date: May 2018 Location: Vienna, Austria Posts: 14 Rep Power: 8	Be aware that the hard-coded scalar limits have been replaced: https://github.com/OpenFOAM/OpenFOAM...9d7f1ee5912713 GerhardHolzinger, mrishi, chathuranga and 1 others like this.