|
|
|
|
|
|
|
|
10nm.png (30,720 bytes) |
|
|
100nm.png (37,963 bytes) |
|
|
1000nm.png (31,417 bytes) |
|
|
|
|
|
Your changes are for an older release of OpenFOAM and it is hard to integrate your changes into OpenFOAM-dev. Could you upgrade to OpenFOAM-dev? It is not clear why you have changed the method to calculate the random vector, what was wrong with the method Bruno implemented? Your implementation value.Su()[1] = mass*f*G1; value.Su()[2] = mass*f*G2; value.Su()[3] = mass*f*G3; value.Su()[4] = mass*f*G4; is clearly incorrect because it set the elements 1-4 of a vector with elements 0-2, could you check? |
|
|
I have corrected the Boltzmann constant: OpenFOAM-dev commit ac67f6a84bb07ed27897e4cefa724d198965e3e2 and I am running the test-case you supplied but it is not clear how the results should be processed. |
|
|
Dear Prof Weller Thank you very much for your kind reply. - Yes, I can upgrade to openfoam-dev but I need a bit more time to compile openfoam-dev on my system. Please kindly realize that this process takes at least about 24 hr. - The random vector that have been implemented by Bruno didn’t generate zero mean unit variance Gaussian random number efficiently. I’m spent much time to test Brownian force of openfoam and I’m sure that it’s incorrect. For example you can see that scalar “f” twice multiply by a random number “eta” ( in line 194 and 202 ) that is incorrect, because in the reference paper “f” multiply by a random number once. If you test it, you can see that particles highly speed up that is physically unreal. If you are in doubt, I can plot the random numbers generated by Bruno implemented method and Box-muller method to compare them and see their difference. I have just used the method suggested by Prof Goodarz Ahmadi in the attached PDF file. - First, I used the random vector of 0-2 but the results were not validated. Finally, I choose random vector of 1-4 by using trial and error method. It was compiled without any error and the results were validated. Moreover, it is better to use an even random vector, because the Box-Muller transform generates a pair of random number and it is more efficient to use the pair. Process of result: As noted earlier, the RMS displacement of particles in Y direction should be equal to the following theory equation: Y_RMS=(2*D*time)^0.5 D=(Kb*temperature)/(3*pi*dynamic viscosity*diameter) In attached test case, 2000 particles are injected from position of (0.5 0.5) at initial time (t=0). The drag and Brownian forces act on these particles. Accordingly, the particles move only due to Brownian random force and their Y RMS displacement can be used to validate the process. Temperature = 273 K Kb = 1.38e-23 Pi = 3.1415 dynamic viscousity = 0.001 Pa.s diameter = 1e-6 m then: D = 4e-13 When you run the test case, you should compute particles Y-RMS displacement in a specific time from the start of solution by using the following correlation: Y-RMS = ((1/number of particles )*(sum (Y(i)-Y0)^2))^0.5 for i=1,number of particles In the test case n=2000 and Y0=0.5 . I used a simple fortran code. This code is attached here. This code reads the particles positions file and computes Y-RMS. For example at t=1 sec: 1- Run the test case 2- Copy the file “testcase/1/lagrangian/reactingcloud1/positions” and paste it in the fortran project directory. 3- Run the fortran code. This code reads the positions file and shows value “ Y RMS = 8.88e-7” 4- Use the theory equation and calculate Y RMS = (2*4e-13*1)^0.5 = 8.94e-7 5- You can see that the RMS displacement is validated for t=1 sec. 6- You can repeat this step for other times. |
|
|
Y-RMS-fortran.f90 (507 bytes) |
|
|
> For example you can see that scalar “f” twice multiply by a random number “eta” ( in line 194 and 202 ) that is incorrect Agreed, and if you take the first eta out are the results then correct? > Finally, I choose random vector of 1-4 by using trial and error method. It was compiled without any error and the results were validated. This is clearly incorrect; if you compile with full-debug and range checking you will find the code stops with an error accessing the vector elements out of range and the first element is not set. |
|
|
I have removed the spurious additional 'eta': commit 96645225ed6e088d5ab77fc3dd5411cd28994cff |
|
|
Dear Prof Weller Thank you for your fast reply > Agreed, and if you take the first eta out are the results then correct? No. I tried that but the result were incorrect. > This is clearly incorrect; if you compile with full-debug and range checking you will find the code stops with an error accessing the vector elements out of range and the first element is not set. Sorry. I'm not expert in C++. only I know that brownian random forces must act on particles sequential ( each pair for one direction ) and my result validated for Y direction. |
|
|
The force vector has 3 components, you cannot set 4 irrespective of the programming language you use. |
|
|
Sorry. You are right. but I say the truth. the result were validated for Y direction. you can test that. may be just set component 1 and 2 of the force vector cause the result to validated for Y direction. |
|
|
Why are you calculating the RMS of the y-displacement? Why not calculate the RMS of the vector displacement? |
|
|
Because in the file Brownian.PDF ,Y-RMS were calculated for validation process. |
|
|
lagrangian::BrownianMotionForce: Changed from a cubic to a spherical distribution See also: StochasticDispersionRAS Resolved in OpenFOAM-dev by commit 46d69e1deaed6b035b96524a12670ee0602da542 Note that to compare with the RMS displacement in the 1D analysis cited you need to calculate the RMS of the magnitude of the displacement in the spherical distribution. |
- Anonymous
