RE: C and Reynolds sensitivity
Kerstein, Alan
More important than this small change in the molecular viscosity will be to see what effect the eddy viscosity has. The eddy viscosity is implemented by commenting out the code bracketed by !SM and !!SM in SMSetVis.f and changing airvisc to the value applicable to water. The latter will then matter only in the stable region, where its main effect will be on the scalars that control buoyancy (if their molecular transport coefficients are expressed as multiples of the viscosity).
The switch from underresolved to resolved is a tricky issue because it's hard to get an empirical handle on Z. Z=600 seems to be a good value for boundary layers, in which it controls the wall-normal location of the transition from the viscous sub-layer to the logarithmic region. The physics associated with that determination of Z probably doesn't apply far from walls. Chances are that you would find little sensitivity to Z because it controls the cutoff of small eddies, which are less important than large eddies. So far it appears that even the frequency of the large eddies doesn't much matter. Based on this I don't think you would see a detectable effect of the choice of Z unless it were set so large that it killed off most of the turbulence, which is not entirely implausible in this moderate-Re flow but a similar effect could be achieved by reducing C.
If we knew what a good value of Z was, then it could be included whether running underresolved with eddy viscosity, or resolved (but in either case the molecular viscosity would be used in the viscous penalty term. In principle a good value of Z could be determined by comparing simulations to measurements of a property sensitive to Z, but far from walls in nonreacting flows, this would typically require high-order statistical information such as spectra.
So I suggest proceeding for now with underresolved simulations using eddy viscosity and Z=0 and doing resolution tests to verify the irrelevance of unresolved scales (except that the eddy viscosity represents their transport contribution). Hopefully this will provide more insight that will enable us to judge whether/how to perform resolved simulations.
I will copy this to the google group because some of this might be of interest to others.
Alan
-----Original Message-----
From: Heiko Schmidt [mailto:heis...@mi.fu-berlin.de]
Sent: Friday, May 08, 2009 3:49 AM
To: Kerstein, Alan
Cc: renaud....@centrale-marseille.fr
Subject: Re: C and Reynolds sensitivity
Hi,
I thought the same and this tiny difference perhaps brings us also closer to the observed E values.
since it will Kerstein, Alan wrote:
>Hi,
>
>I think you should run with Sayler's value. At the time we started using 1e-6, we didn't know Sayler's value so we just rounded off a standard value.
>
>An important point regarding Re sensitivity is that sufficient resolution must be maintained. I'll mention my recollection about this (unfortunately I don't have notes on this to remind me, but hopefully Heiko recalls). I think we determined that the results are insensitive to resolution of the smallest scales of the convective region provided that the smoother profiles of the stable region are resolved.
>
Yes
> Therefore we decided to run with under-resolution of the convective region, and accordingly set the viscous penalty equal to zero because the viscous scales are not resolved in the region where eddies occur.
>
Yes
> I also recall that we bypassed the eddy-viscosity model in SMSetVis.f (see !SM through !!SM in that file), which in any case would need to be revised for the present application because in its original form it sets the lowest allowable viscosity equal to the viscosity of air. Instead it should be the viscosity of water.
>
>
Just to double check my understanding,
when we are underresolved viscous penalty does not make sense and Z is
therefore set to zero. At the same time we are in an LES type setting
where we need closure.
This can be achieved in the code using, e.g. an eddy viscosity model.
>So my question is, are you running under-resolved in the convective region with zero viscous penalty? If so, I suggest using the eddy-viscosity model with the above change. We bypassed this model because the original intention was to run resolved, but I think we then switched to under-resolved simulations with zero viscous penalty. Heiko, is my recollection accurate?
>
>
>
Yes we bypassed it and we have to check, but it should be switched off.
Renaud has already started calculating the Kolmogorov scales for all
the runs. So we will add tables including the latter as well as the
resolution number (how is this called?) dz/\eta. In the latest
calculations I have done it should be order 1, but with increasing Re in
sensitivity studies we will get higher values.
Where do you see the limit of the Z=0, Z=something switch?
Should it be a Heavyside type switch or a smoother one?
>If you adopt this approach, the
>
We have this in mind.
Renaud will provide you with more details concerning the parameter
ranges before he starts the various runs.
Cheers
Heiko
>n it will be important to do tests of the adequacy of resolution at various Re values.
>
>Alan
>
>-----Original Message-----
>From: renaud....@centrale-marseille.fr [mailto:renaud....@centrale-marseille.fr]
>Sent: Thursday, May 07, 2009 8:30 AM
>To: Kerstein, Alan
>Cc: heis...@mi.fu-berlin.de
>Subject: C and Reynolds sensitivity
>
>Hello,
>
>Thank you for your very quick answer to my last e-mail. We begin now to study the sensitivity of entrainement to C parameter and Reynolds number. We will do it as you suggested in your mail. We will vary C*Re by varying C and/or Re and will check the effect on the entrainement rate.
>As we want to see the effect of Re I have recalculated it using both Saylers data and those which are used in the code. The only difference is the viscosity. But I only get the same value as Sayler (he got 932) with his precise value for viscosity.
>
>
>case H01 Data Sayler Data code
>q0(W.m-2) 251 idem
>alpha(K-1) 2,33E-004 idem
>g(m.s-2) 9,81E+000 idem
>Rho_0(kg.m-3) 1,00E+003 idem
>Cp(J.kg-1,K-1) 4,18E+003 idem
>h(m) 2,66E-001 idem
>nu(m2.s-1) 9,43E-007 1,00E-006
>
>Q 1,37E-007 idem
>w1 3,32E-003 idem
>Re_w1 933,43 880,22
>
>
>Do you think that we should run the code with the Sayler's value for viscosity?
>
>Best regards,
>Renaud Nedelec
>
>
>
>
>