Why sometimes momentum predictor step is not performed

liu · July 2, 2005, 00:50

In some of the solvers, such as interFoam, the momentum equation is not solved (i.e., no solve(UEqn == fvc::...)). Some solver (such as rhoTurbFoam) has a switch to choose whether this step is performed or not.

The original paper of PISO algorithm has this as a essential step.

1. Why the difference?
2. If the gravity is included as in interFoam and I want to do the momentum predictor step, should I include this gravity effect in the UEqn? Like:
solve(UEqn == -fvc::grad(p) - fvc::grad(rho)*gh)?

henry · July 2, 2005, 09:01

1) Doing a momentum predictor is not an essential step for the convergence of the PISO corrector loop although it is sometimed benefitial but not always. For example in very low-Re flows the momentum predictor step can be severely detrimental to the convergence. For interFoam I found that the momentum predictor step did not improve the convergence behaviour and is a bit complicated to include so I removed it for simplicity. I have reinstated it for the 1.2 release just so people can find out for themselves if it is helpful or not.

2) No that formulation of the momentum equation sources is not consistent with the pressure-equation and momentum corrector.

liu · July 2, 2005, 15:45

Is solve(UEqn == -fvc::grad(pd) - fvc::grad(rho)*gh)? Here gh = g & mesh.C().
I forgot to change the total pressure to dynamic pressure.

henry · July 2, 2005, 19:27

That is still inconsistent with the pressure-equation and momentum corrector, take a look at pEqn.H.

henry · July 2, 2005, 19:30

I ran the dam-break case with and without a momentum predictor and found that with the predictor the pressure solution is slightly faster but it does not offset the cost of the momentum solution and overall it ran ~5% slower.

liu · July 2, 2005, 20:03

Ok, I derived the equation with pd again and find it seems to be solve(UEqn == -fvc::grad(pd)).
The gravity is totally absorbed into the pressure term and density difference only take effect in the pEqn.

henry · July 2, 2005, 20:59

That is not correct.

liu · July 2, 2005, 21:08

Well, what is correct one?

henry · July 2, 2005, 21:15

I have implemented it in the 1.2 version of interFoam which will be released soon but due to other improvements it is not compatible with 1.1 so there is no point me posting it here. I could write you a momentum predictor for the 1.1 version of interFoam but because it does not improve the performance of the code in any way I don't think it's a good use of my time. However, if this is very importantto you, enough for you to purchase a support contract I would do this work as part of that contract.

July 3, 2005, 00:02

I think it that "- fvc::grad(rho)*gh" should move into UEqn.

henry · July 3, 2005, 07:56

I disagree.

July 2, 2005, 00:50	In some of the solvers, such a	#1
liu Senior Member Xiaofeng Liu Join Date: Mar 2009 Location: State College, PA, USA Posts: 118 Rep Power: 17	In some of the solvers, such as interFoam, the momentum equation is not solved (i.e., no solve(UEqn == fvc::...)). Some solver (such as rhoTurbFoam) has a switch to choose whether this step is performed or not. The original paper of PISO algorithm has this as a essential step. 1. Why the difference? 2. If the gravity is included as in interFoam and I want to do the momentum predictor step, should I include this gravity effect in the UEqn? Like: solve(UEqn == -fvc::grad(p) - fvc::grad(rho)*gh)? kmou likes this. __________________ Xiaofeng Liu, Ph.D., P.E., Assistant Professor Department of Civil and Environmental Engineering Penn State University 223B Sackett Building University Park, PA 16802 Web: http://water.engr.psu.edu/liu/

July 2, 2005, 09:01	1) Doing a momentum predictor	#2
henry Senior Member Join Date: Mar 2009 Posts: 854 Rep Power: 22	1) Doing a momentum predictor is not an essential step for the convergence of the PISO corrector loop although it is sometimed benefitial but not always. For example in very low-Re flows the momentum predictor step can be severely detrimental to the convergence. For interFoam I found that the momentum predictor step did not improve the convergence behaviour and is a bit complicated to include so I removed it for simplicity. I have reinstated it for the 1.2 release just so people can find out for themselves if it is helpful or not. 2) No that formulation of the momentum equation sources is not consistent with the pressure-equation and momentum corrector. sharonyue, wc34071209, minh khang and 7 others like this.

July 2, 2005, 15:45	Is solve(UEqn == -fvc::grad(pd	#3
liu Senior Member Xiaofeng Liu Join Date: Mar 2009 Location: State College, PA, USA Posts: 118 Rep Power: 17	Is solve(UEqn == -fvc::grad(pd) - fvc::grad(rho)*gh)? Here gh = g & mesh.C(). I forgot to change the total pressure to dynamic pressure. __________________ Xiaofeng Liu, Ph.D., P.E., Assistant Professor Department of Civil and Environmental Engineering Penn State University 223B Sackett Building University Park, PA 16802 Web: http://water.engr.psu.edu/liu/

July 2, 2005, 19:27	That is still inconsistent wit	#4
henry Senior Member Join Date: Mar 2009 Posts: 854 Rep Power: 22	That is still inconsistent with the pressure-equation and momentum corrector, take a look at pEqn.H.

July 2, 2005, 19:30	I ran the dam-break case with	#5
henry Senior Member Join Date: Mar 2009 Posts: 854 Rep Power: 22	I ran the dam-break case with and without a momentum predictor and found that with the predictor the pressure solution is slightly faster but it does not offset the cost of the momentum solution and overall it ran ~5% slower. randolph likes this.

July 2, 2005, 20:03	Ok, I derived the equation wit	#6
liu Senior Member Xiaofeng Liu Join Date: Mar 2009 Location: State College, PA, USA Posts: 118 Rep Power: 17	Ok, I derived the equation with pd again and find it seems to be solve(UEqn == -fvc::grad(pd)). The gravity is totally absorbed into the pressure term and density difference only take effect in the pEqn. __________________ Xiaofeng Liu, Ph.D., P.E., Assistant Professor Department of Civil and Environmental Engineering Penn State University 223B Sackett Building University Park, PA 16802 Web: http://water.engr.psu.edu/liu/

July 2, 2005, 20:59	That is not correct.	#7
henry Senior Member Join Date: Mar 2009 Posts: 854 Rep Power: 22	That is not correct.

July 2, 2005, 21:08	Well, what is correct one?	#8
liu Senior Member Xiaofeng Liu Join Date: Mar 2009 Location: State College, PA, USA Posts: 118 Rep Power: 17	Well, what is correct one? __________________ Xiaofeng Liu, Ph.D., P.E., Assistant Professor Department of Civil and Environmental Engineering Penn State University 223B Sackett Building University Park, PA 16802 Web: http://water.engr.psu.edu/liu/

July 2, 2005, 21:15	I have implemented it in the 1	#9
henry Senior Member Join Date: Mar 2009 Posts: 854 Rep Power: 22	I have implemented it in the 1.2 version of interFoam which will be released soon but due to other improvements it is not compatible with 1.1 so there is no point me posting it here. I could write you a momentum predictor for the 1.1 version of interFoam but because it does not improve the performance of the code in any way I don't think it's a good use of my time. However, if this is very importantto you, enough for you to purchase a support contract I would do this work as part of that contract.

July 3, 2005, 00:02	I think it that "- fvc::grad(r	#10
li Guest Posts: n/a	I think it that "- fvc::grad(rho)*gh" should move into UEqn.

July 3, 2005, 07:56	I disagree.	#11
henry Senior Member Join Date: Mar 2009 Posts: 854 Rep Power: 22	I disagree.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
running gambit journal step by step	giogio	FLUENT	1	March 10, 2008 15:03
How are a transient simulations performed??	Luk	Main CFD Forum	0	October 19, 2007 11:09
Can microgravity simulations be performed by FOAM	zou_mo	OpenFOAM Running, Solving & CFD	1	February 20, 2006 23:35
define moving mesh for engine step by step	homan	FLUENT	0	October 26, 2005 04:27
time step for predictor-corrector method	jenn	Main CFD Forum	0	February 21, 2005 16:43