It seems to me that this discretization directly ties the timestep to the spatial resolution since a single time-step would have to exactly match up such that the neighboring node is reached with the given discrete speed vector. Am I correct in this assumption? This is really the main thing that has been confusing me about LBM since many papers and texts seem to treat both of these as variables and don't even mention this, although I have found some that do exactly what I describe. Is there a way to do substeps in time with this method? Or is there simply no need, numerically speaking, i.e. is this exact matchup between spatial resolution and timestep simply also the numerically most accurate way to do it?
great presentation: it helped me make sense of LBM. My focus is a deep learning model based on LBM (Bedrunka et al) and I would like to recommend the 2
Hi, I try to ask 2 questions here, don't worry if you can't answer it :) I'm strugling to code a LBM simulation and I always get some strange ripples (region where f becomes alternatively big and small in the direction of propagation). It seems that the method doesn't like big f gradient : they get bigger and bigger (with any tau value). I suspect that I'm not doing the operations in the right order : I'm doing every node's advection (parallel calculations on GHPU), THEN every collision. I see at 42:03 that it's not the way you recommand it, but I don't really understand your description... This iterative method will change a node after the other. So the one I just change will then be read with it's new value for the next one ? I'm confused. How could this be parallelized ? Also, what is the 4th step: "Post-collision rearrangment" ?? Sorry for my approximative english and of course, thanks for this nice video :D
that was a brilliant presentation, thank you, it brighten the path. I also have a question, what application do you suggest for coding LBM ?? python or Matlab or any other one?
33:20 I made it to this point. For this presentation to make the Lattice Bolzman method understandable, it needs to spend some more time on this concept of phase space. Instead this presentation suddenly made a deep-dive into hardcore math saying "I will not explain about this in detail because it is very complex".... ok... First part was good but here I stopped watching
This is a very good video to get into the lattice boltzmann method. Thank you!!! Please never take this video offline! :D
Very nice to know about the physics behind the equations !
It seems to me that this discretization directly ties the timestep to the spatial resolution since a single time-step would have to exactly match up such that the neighboring node is reached with the given discrete speed vector. Am I correct in this assumption? This is really the main thing that has been confusing me about LBM since many papers and texts seem to treat both of these as variables and don't even mention this, although I have found some that do exactly what I describe.
Is there a way to do substeps in time with this method? Or is there simply no need, numerically speaking, i.e. is this exact matchup between spatial resolution and timestep simply also the numerically most accurate way to do it?
I'm always watching it well. Thank you. Please share it with us more.
Complex yet, explain quite interesting ❤
great presentation: it helped me make sense of LBM. My focus is a deep learning model based on LBM (Bedrunka et al) and I would like to recommend the 2
Thanks so much. Great job. Enjoyed your lecture.
Hi, I try to ask 2 questions here, don't worry if you can't answer it :)
I'm strugling to code a LBM simulation and I always get some strange ripples (region where f becomes alternatively big and small in the direction of propagation). It seems that the method doesn't like big f gradient : they get bigger and bigger (with any tau value). I suspect that I'm not doing the operations in the right order : I'm doing every node's advection (parallel calculations on GHPU), THEN every collision. I see at 42:03 that it's not the way you recommand it, but I don't really understand your description... This iterative method will change a node after the other. So the one I just change will then be read with it's new value for the next one ? I'm confused. How could this be parallelized ?
Also, what is the 4th step: "Post-collision rearrangment" ??
Sorry for my approximative english and of course, thanks for this nice video :D
Thankyou for uploading this video. I found it very interesting.
Great stuff dude, keep it up!
Thanks, it is really clear and interesting 👍
am i supposed to mesh my geometry while using LBM ?? Kindly reply
thank you for good presentation!
What is the exact form of f(x,ei,t) and feq(x,ei,t)?
Does LBM treat the particles as elastic or rigid?
awesome video! Thank you!
that was a brilliant presentation, thank you, it brighten the path. I also have a question, what application do you suggest for coding LBM ?? python or Matlab or any other one?
C++ or C
To code a solver you need speed
If you just want to code for fun, then you can use any code, but using python for example will lead to very slow calculations
Can you please make a video on Turbulent plane channel flow with smooth walls using OpenFOAM?
33:20 I made it to this point. For this presentation to make the Lattice Bolzman method understandable, it needs to spend some more time on this concept of phase space. Instead this presentation suddenly made a deep-dive into hardcore math saying "I will not explain about this in detail because it is very complex".... ok... First part was good but here I stopped watching