Это видео недоступно.
Сожалеем об этом.
Crystals: raw footage - 3D simulation of crystallization (Molecular Dynamics)
HTML-код
- Опубликовано: 13 дек 2018
- For those who are more curious, this video is the complete (20 minute long) simulation of the 3D crystalization demo. (like I expect anybody to watch it front to back...)
The simulation is based off of the "Energy Verlet" molecular dynamics method. I wrote it in Matlab to run on a 1060 6GB GPU. By the time all the particles are in the simulation, the program makes multiple trips to the graphics card for every timestep or else runs out of graphics memory. massively parallel computation requires a LOT of memory, apparently... There is no sub-binning of the simulation. Every pair of particles has its forces calculated in every timestep. Although far-apart particles don't apply a lot of force on each other, I couldn't make a more efficient system to only solve for adjacent particles that was better than calculating all the inter-pair distances in one line of code... For 16000 particles, that's something like 250 million square roots that need to be calculated every timestep, let alone the actual force vectors after you figure out how far apart everything is...
The particles obey a sum of the Lennard Jones potential (so that particles don't go through each other) and the Coulomb potential (so that red and blue attract and like colors repel). Even with these two perfectly "spherical" potentials, an ordered 3D structure forms, with both a crystal lattice and habit.
Since a couple comments were asking for it, I cleaned up my code a bit and have it posted here. If you have any questions, let me know in the comments so that anybody can read the answers!
github.com/BrianHaidet/AlphaP...
This Video Series:
Crystals (The main video): • Crystals: Building pat...
The entire CsCl crystal simulation, growing a crystalline "nanoparticle" from 2 to 16384 particles: • Crystals: raw footage ...
Extra footage of the water-bath 2D demo: • Crystals: raw footage ...
Epilogue - Why cubic crystals don't always make cubic shapes: coming soon
GIFs in this series:
2D crystal water bath:
3D Simulated nanoparticle:
Music Credits:
I Dunno by grapes is licensed under a Creative Commons Attribution license (creativecommons.org/licenses/...)
ccmixter.org/files/grapes/16626
Operatic 3 by Vibe Mountain is licensed under a Creative Commons license
Brighton Lights by Noir Et Blanc Vie is licensed under a Creative Commons license
Move Ya by Max Surla/Media Right Productions is licensed under a Creative Commons license
Birth Noir by Noir Et Blanc Vie is licensed under a Creative Commons license
Space Walk by Silent Partner is licensed under a Creative Commons license
Flickering by VYEN
New Land by ALBIS is licensed under a Creative Commons license
Arcadia - Wonders by Kevin MacLeod is licensed under a Creative Commons Attribution license (creativecommons.org/licenses/...)
Source: incompetech.com/music/royalty-...
Artist: incompetech.com/
Very nice
hey, can you answer a doubt for me? Is it possible to simulate this exercise with nonspherical potential? Let's say a potential that has an equipotential surface like a cube and then we see the emerging order? thanks in advance. Good work here By the way
Oh yeah absolutely! Real atoms are NOT perfect spheres and they are simulatable. The weirder the shape of the valence orbitals, the more specific and bizarre crystal structures you can generate. Some equilibrium crystal lattices have a ridiculous number of atoms in the unit cell. The repeating pattern can get super complicated. If you did a cube-shaped potential, you’d also have to account for rotational degrees of freedom and the associated torques, angular momenta, etc. of each particle and I decided that was a serious pain when I could do a pretty good representation of simpler materials with near spherical ions
@@AlphaPhoenixChannel thanks for clarifying mate. :)
Love this simulation! If we were to try to mimic a change in ionic radius (a la your changing to larger blue magnets in the 2D model) how might we achieve that in the code? I'd love to develop an array of these and run the simulations in the classroom.
Great channel!! At these experiments try to use gromacs(no quantum mec, only MD), it may be more effective) but you should set custom forcefield and low temperature.
Is there a way that you can put a counter on how many cubes are formed. It would also be nice to see squares of cubes, where ever time a cubed basically cloned itself you would count that and so on. Double cube counter. That would be fire.
I wonder if this can be used to predict the structure of MOFs, by minimize the energy of the complexes at set intervals.
Cool! Emergent order is fascinating. I have a couple questions: how are you handling boundary conditions? It seems as though atoms are reflecting/bouncing off the cell walls. Have you ever considered periodic boundary conditions and a shorter LJ cut-off? The second question is the solvent: are you calculating this trajectory in a vacuum, or is there implicit solvent (like GBIS or some other continuum model)?
Crystallization in MD is really interesting. I once tried to freeze TIP3P water -- both with and without a nucleation site -- but with no luck. Might be an interesting video (at least for me lol). Keep up the great content!
Ahh... "heated walls." Sorry for not reading the text in the vid!
haha yep - added a bunch of random annotations over there. I don't know if it says in the video but if I remember right the wall heating is a 2% increase in speed every bounce. As it turns out, if I don't heat the walls and maintain a thermal gradient from outside to center, I get chunks of crystal precipitating in the vacuum around the nanoparticle in the middle. I didn't want to deal with the vacancies and defects associated with running nanoparticles into each other, so the heated walls keep the "gas" diatomic or monoatomic.
Wow great simulation, would this happen in solution and at a certain PH?
In real life, yes, but I'm not sure how I'd implement that in this simulation. I just assumed vacuum
Great work, May I know the steps using MD software to create this growth model?
Not sure - I wrote this from scratch
Feeling internal pain watching that computation time accelerate....
n^2.....
How can contact u? I need some serious help related to astrophotography
@Hubert Dungen 😢
Believe it or not, but if you roll up to the local games club with this in your pocket they'll tell you it's "not random enough"
Hey, may I know what MD method and algorithm you used and how you modeled the atoms?
I think energy verlet but it was a while back - should be on github
@@AlphaPhoenixChannel really nice, I've started learning MD with Gromacs a few months ago, we mostly use leap-frog verlet for the main production simulation and steepest gradient descent for potential energy minimization. I found this video really cool because this is something I wanted to see in molecular dynamics. But I've only done simulations with constant number of atoms and using periodic boundary conditions. But I'm definitely going to try this some day!
When I was building gold nanoparticles I'd just generate a bunch of atoms within a zone at the center of the unit cell, heat it up, hold it until they'd coalesced, and then cooled it to the final temperature. I was only making particles on the order of 30A at the largest though IIRC.
Also, is there a reason you're using Matlab rather than a dedicated program like LAMMPS?