500,000 particles. The Newton's gravity law. Real time simulation

[@AlexNize]

Hi! More than 10 years have passed since the creation of the simulation, to my surprise, the video is still popular. I will try to briefly answer the questions that might arise. First of all, I've also made N-Body simulation, you can find it on my GitHub: github.com/DrA1ex/JS_ParticleSystem/blob/main/README.md#n-body-simulation About Simulation in the video: The simulation simulation is simplified, it's not an N-Body (each particle interacts with each other). All calculations been made for one point, which is controlled by the mouse. In order to implement precise calculation of physics, it will be necessary to do optimizations (e.g. clustering of points), otherwise the performance will be so much lower (the complexity of calculations instead of O(N) will be O(N^2)), but even with optimizations it will not be possible to achieve a significant increase in performance, since for a precise gravity calculation will need to calculate the interaction of each point with each other. For 1 thousand points, this will be up to 1 million operations. In my simplified version, 1 million points requires 1 million operations. You can try GPU (best for Desktop) in-browser simulation via WebGL: - 500,000 particles: dra1ex.github.io/WebGL_ParticleSystem/?particle_count=500000 - 1,000,000 particles: dra1ex.github.io/WebGL_ParticleSystem/?particle_count=1000000 - 2,000,000 particles: dra1ex.github.io/WebGL_ParticleSystem/?particle_count=2000000 - 4,000,000 particles: dra1ex.github.io/WebGL_ParticleSystem/?particle_count=4000000 My MacBook Pro 16 can simulate ~2,500,000 in 60fps Or you can try CPU (best for mobile) simulation via HTML5 Canvas: - 100,000 particles: dra1ex.github.io/JS_ParticleSystem/?particle_count=300000 - 500,000 particles: dra1ex.github.io/JS_ParticleSystem/?particle_count=500000 - 1,00,000 particles: dra1ex.github.io/JS_ParticleSystem/?particle_count=1000000 My IPhone 11 Pro can simulate ~600,000 in 60 fps At this moment I've made 3 main versions: - DirectX 9 (you see it in the video) version uses CPU for physics calculation, uses AVX vectorization for speedup. Points are drawn through the billboard technique (for texture mapping). Maximum number of points in real-time: about 1 million. - DirectX 10 version uses GPU for physics calculation. It's rendered through geometry shaders (a polygon is created from a point). The calculations located in shader, and results of iteration are dumpred into memory via the StreamOut technique. Maximum number of points in real-time: about 10 million. - WebGL 2.0 version uses GPU for physics calculation. Used Transform Feedback technique for GPGPU physics calculation. Number of points in real-time: about 5 million (maybe more). So what about source? - DirectX 9 version: github.com/DrA1ex/ParticleSystem - DirectX 10 version: github.com/DrA1ex/D3D10_ParticleSystem - Windows Forms version: github.com/DrA1ex/GDI_ParticlesSystem - JavaScript version: github.com/DrA1ex/JS_ParticleSystem - WebGL version: github.com/DrA1ex/WebGL_ParticleSystem I used DirectX for rendering, but the implementation can be done on any engine: OpenGL, Vulkan, Metal, even HTML5. I choose DirectX because 10 years ago there were more manuals about DirectX on the Internet. I'm also thinking about implementing a gravity simulation in WebGL (JavaScript or WASM), but a precice N-Body simulation with calculation optimization and of course with source code. So tell me, do you interested in such a simulation?

[@9551Dev]

wow playing with this is really mesmorizing. looks very nice !

[@water_is_wet]

great work

[@JanPBtest]

What happened at 0:48 ?

[@buttafan4010]

Like a description of star birth, star death by super nova, black hole and void formation.

[@Hibbyhubby]

yes for Precise N-Body Simulation!!! good work

[@m.sierra5258]

The youtube compression algorithm REALLY doesn't like this

[@roygalaasen]

Oh man, not even in 720p could The compression algorithm handle, but it is not really weird. “Noise” like this require bandwidth to be rendered properly, and the compression algorithm is trying the hardest to keep the bandwidth down. Conflicting interests to say the least.

[@m.sierra5258]

@@roygalaasen Upscaled to 4k could probably handle it :D

[@vincentschumann937]

@@m.sierra5258 double/triple the bitrate at 720 and it would look fine

[@jaredf6205]

Need to upload it to Vimeo. They have much higher bitrates if your connection can handle it.

[@vitorstreetboys]

TRUE

[@AlexNize]

I was changing the simulation parameters while recording video

[@cicik57]

you should not, instead you d better create many simulations with different params and compare but still amazing work, dud

[@KK-pq6lu]

Edge artifacts? Looks like the particles can wrap?

[@soliderarmatang5664]

This is why I love KZbin, because of videos and people like this! Thanks for all the links!

[@mikaelpaavola4781]

This is so rad man. Cool of you to share the code!!

[@yougeo]

Nice bit code appears to be 2d forces only. Also would be nice to see without infinite walls bounce of particles.

[@AvZNaV]

What CPU are you running? Intel i81 @ 50GHz?

[@samb7291]

He is using his GPU. its magnitudes faster than that

[@ShredST]

@@samb7291 The description says it's using an AVX particle system, so presumably it's running on the CPU.

[@ylliimeri]

Nice looking simulation. But it doesn't seem to conserve angular momentum. At one point all particles were rotating in the same direction.

[@monkeyman5513]

You still conserve angular momentum when you have backwards rotation from the total momentum if other particles have extra angular momentum in the opposite direction to cancel it out

[@ylliimeri]

@@monkeyman5513 well, true, but it doesn’t seem to be the case here. If you look at 3:10 all the particles are rotating counterclockwise, and unless a few of them rotate clockwise at very high speeds the angular momentum isn’t conserved. Btw, the initial angular momentum seems to be zero.

[@srijanraghavula]

Beautiful simulation

[@iinRez]

*This is what the Universe is doing. To us it's billions of years, at this scale it's 6 minutes.* *We are currently in the expansion period of this undulating phenomenon.*

[@shivas3003]

the nightmare for compression algorithm ^^

[@vaughnalexander31]

this video has been blessed by the algorithm

[@AlexNize]

GPU version can. ~ 5m (2m pixels at FullHD) particles at 60 fps on GTX670

[@grecinos2]

I downloaded the CUDA SDK and the samples. After installing it, I was able to play with some of the demos and they are great! I have VS 12 and it is having a tough time loading the demo programs. The CUDA docs say that the samples are compatible with VS 8 and 10. Needless to say, I was unsuccessful at compiling them. My goal is to be able to animate an n-body simulation in 2-d w/o collision detection. Say, 100000 little circles and perhaps a planetoid for kicks.

[@nullbeyondo]

This is beautiful.. I just wish KZbin didn't hate this video :D

[@caseyford3368]

It depends on what particles you're using and many other variables.

[@shinshia]

After 10 years, algorithm finally bring me here

[@marquizzo]

Thank you for sharing the source code!

[@jyxtheberzerking4824]

new xbox screensaver just dropped, looks awesome. edit: OLD xbox screensaver rediscovered, looks awesome.

[@marianotrebino6027]

Good job man!!!! Its beautiful, the only physics is the newtons gravity law? Do you apply any external force? Can you explain it to me or share a resource? I want to do something similar. Thank you!!!

[@Azakadune]

Well that’s the only force you’d need. I’m guessing there ain’t any collisions based on how it behaves?

[@jsstt]

"What are you watching son?" Chaos.

[@Ryrzard]

We have just witnessed a star's life from beggining to the end.

[@Ryrzard]

What do you mean?

[@ImperativeGames]

I would say it represents Universe, with Big Bang and stuff like that

[@Ryrzard]

Where was this big bang? I don't really see it here.

[@ImperativeGames]

TheFarciarz777 When it all goes into one point, then everything explodes and new Universe created.

[@Ryrzard]

Well, you can't really simulate a Big Bang in an euclidian space. Besides, plenty of stuff in the universe goes into a very close space and explodes after some time.

[@nazardidkovskyi]

Hi, again) Do you use 3D space in this program? And do you use such parameters as mass of particles? Do this program compute force for all particles?(I mean the force with which each particle interacts with all 499 999 other) Because I can't get such performance in my program, though I'm using multithreading.

[@AlexNize]

+Nazar Didkovskyy, привет. Это не N-Body симуляция, каждая частица взаимодействует только с одной. К тому же z-координата фиксирована и не рассчитывается.

[@NimerionTech]

Hello Nazar. I just saw this video and your comment, and once encountered a similar problem, but with Flocking simulation. What I've learned was that you don't really need to go over all the other entities in your world to get them collide successfully. You can simply either make a grid model on which you can detect if the neighbor positions (sells) are empty, or not for every object. OR you can do a QuadTree which will increase the performance drastically. I've never tried a QuadTree, tho.

[@patrickthepure]

See multilevel fast multipole method. It reduces the complexity from N^2 to somewhat like NlogN in these kind of situations.

[@mastershooter64]

@@patrickthepure woahhh bro really???? that's insane!!

[@c.howard9413]

Just stumbled upon this video and it’s pretty cool. I understand the GPU aspect of improving runtimes but what other approximations did you make? Did you use the fast multipole method to speed up the approximation?

[@cmilkau]

0:48 What's going on here? Looks weird, like a sudden change in mean impulse.

[@StanislavMarchevsky]

Это просто космос!

[@timoakeley6085]

truly amazing

[@kazimierzmarkiel5400]

What this animation is modelling? Gravitation itself causes the mass focusing in the the gravity center. What is causing the rotation of mass, repeated many times asymetric explosions, the changes of the center of gravity in the space ? Stary

[@anchorcult]

BEAUTIFUL AN AMAZING MOMENT! GLAD I COULD BE A PART OF IT! OOIIOO

[@emanueol]

great resources :) simple question, imagining you just have 3 particles.. does calculation use same location/vel/accl of each particle on beginning of each frame to calc (using newton f=m1*m2/D^2) or something more exotic ?

[@aadiduggal1860]

you can just use Newton's law as that's complex enough for a numerical solver. For the case of 3 particles, each particle would have a force vector in 2 directions. From that, you numerically integrate the summed forces which would result in position. For say, particle 1, we can simulate its acceleration (ẍ) as a function of its position to the other 2 particles (x1, x2, x3) as per Newton's law, f_i = m_i * ẍ_i = m_i * m_j / x_i ^ 2 ( where f_i is the force acted on the ith particle to the jth particle). The math would be something like this: f1 = m1 * ẍ1 = m1 * m2 / x2^2 f2 = m1 * ẍ1 = m1 * m3/ x3^2 (f1 + f2)/m1 = ẍ1 = m2/x2^2 + m3/x3^2 From this, you can probably tell a pattern that the other 2 particles would follow a similar procedure. In total we would have (assuming G = 1) : ẍ1 = m2/x2^2 + m3/x3^2 ẍ2 = m1/x1^2 + m3/x3^2 ẍ3 = m1/x1^2 + m2/x2^2 This is a fully defined system of differential equations which can now be numerically integrated with respect to time by a computer, giving us the positions of x1, x2, x3 which can be plotted with time and animated. You would, of course need the initial conditions for x and ẋ. Extending this to 500,000 particles would result in a system of 500,000 differential equations, which can really only be solved if you do some further optimization (and probably will need a supercomputer) which I'm not too familiar with. UPDATE: So it seems that the way this video works is that the particles are following the user's mouse. That's the same thing as here, with the difference being that your mouse would define the initial position coordinates of a very, very big mass. This video is NOT an N-body, so each particle is not being attracted to the other. Each particle is very small (pretty much 0 mass) and ONLY being attracted to the mouse. This greatly simplifies the calculations. Still, it's kinda cool to see everything coalesce into a sphere given enough time (which is why planets are round).

[@emanueol]

@@aadiduggal1860 thanks for confirming :) I think you got typo on first x1 equation: I think instead of: ẍ1 = m2/x2^2 + m1/x1^2 should be ? ẍ1 = m2/x2^2 + m3/x3^2

[@aadiduggal1860]

@@emanueol yep, thanks for catching that.

[@Broxzier]

Nicely done!

[@riki_XD]

A 10 years old comment !

[@Broxzier]

@@riki_XD A one-day-old reply!

[@NAANsoft]

Do you include the time propagation of gravitaion - i.e. that gravity is propagating with the speed of light, not with infinite speed?

[@Khepramancer]

0:48 Was that spontaneous?

[@matthewsupra]

Mesmerising

[@Xonatron]

Are they attracting each other, or just the center mass? I'm guessing the latter.

[@professionalgoob]

This is amazing!

[@mitchellkeenan4356]

Best film i've ever seen

[@szilardorosz371]

do the particles collide or just gravitate?

[@christopherrogers532]

lol nothing like seeing a random video like this bob to the surface of youtube and getting sucked in. :)

[@ShredST]

Was inspired to write my own version after watching this video.

[@SolarMusicZ]

How are they all together and then all of a sudden shoot apart again?

[@korek3787]

Very stable formation :D

[@NeoVamp]

Ran it at 15 million, holy shit NICE CODE!

[@buttafan4010]

Like a description of star birth, star death by super nova, black hole and void formation.

[@jdahllquist6823]

Fascinating. I kept waiting for it to form a galaxy. I wonder why it can collapse to a near sphere then burst out again? Simple rules can lead to complex behavior.

[@עמריטפלא]

J Dahllquist It bursts out because of momentum reservation: One of the partical gets so close to the overall origin of gravitation, that it gets flung at a huge velocity, resulting in the movement of the other particals.

[@MaxAltair]

J Dahllquist Looks like particles doesn't have collision.

[@Scelenaarwen]

it doesn't. This is from a website where when you hold left click you attract the particels when you hold right click you repell them. The bursting out is just the guy pressing the rght mouse button. You won't get that type of behaviour just out of a purelly attractional force. Only if a particle gets acceleated really fast to the center of mass and gets sling-shot outwards, but not a bulk of the cloud bursting.

[@AltumNovo]

@@Scelenaarwen Not true. Every head of a supernova?

[@ShredST]

I'm reading the source code and I just got to the AVX part. Man, it's going to be a long night. Edit: fuck it, I checking out an older commit.

[@AlexNize]

It's CPU only. GPU version can simulate 5 million particles in real time.

[@grecinos2]

I've been searching on and off for some years for someone to publish a simple gravity simulation that uses DirectX. I think your's is the best I've found yet! I've been tinkering around with your code. I was trying to see if I can convert it into a particle simulation whereby each particle is attracted to all others. I've yet to figure it out. Can you please give me some insight on how I might go about doing this?

[@JoeCenatiempo]

Look up multi body physics and the three body problem perhaps

[@dejanpopovic7903]

Why do all the bodies coalesce towards the end of the simulation? I thought they would continue their respective eccentric behavior?

[@ItsNotAllRainbows_and_Unicorns]

There might be a dissipative term involved

[@marceliwanicki2843]

From the comments i assume that the program involves some user input so the guy is probably changing stuff

[@cicik57]

is it attraction to center or fair calculation for every pair of particles?

[@Astromath]

Attraction to a point (the mouse)

[@cicik57]

@@Astromath not between particles?

[@Mahishasur]

Music?

[@nenotu8400]

Did you used Blender or something like After Effects to make this

[@bitterlemonboy]

no he used opengl

[@TH0M2610]

What are the colors corresponding to ?

[@matthewbarnes7029]

...such beauty...

[@potatocrusher1922]

Why did it bounce back ?🤔

[@arc8216]

Can you see any Lagrange points?

[@nothsim]

Closed loop universe?

[@Reuleaux99]

Cool simulation! Why did this happen?? 5:19

[@grecinos2]

I sent a reply earlier this week. Not sure what happened. Anyways, lol, yes, I’ve been searching for some years, but only every 3 to 4 months. In particular, I’m searching for an n-body simulation that takes advantage of today’s GPU’s. I would like to find one that I can modify and recompile. I’ve yet to find one, even the original source from this video. I tried your recommendations. I used the compiler of their choice, but no go. I’m real frustrated.

[@nielsdaemen]

KZbin compression raped the quality!

[@evanwonisch788]

RK4? Or Euler? Or something else?

[@ishaanlohani]

"And then there was Light" -God

[@ger10izq92]

wonderful

[@megadalton3006]

At 2:02, did we just witness a virtual big bang?

[@WiocorEnergy]

Great simulation ;-)

[@Markus9705]

Wonderful.

[@gountzas]

*"when the center of gravity is literally the center of your screen"*

[@eduardositedotcom]

Wonderful!!!

[@Razorcarl]

So cool

[@solcarzemog5232]

Are particles elastic and bounce agaist each other?

[@FreekHoekstra]

there seems to be no collision, so they pass through the centre of gravity and their inertia caries them forward. untill they slow down due to the gravity again and go towards the centre

[@ps4games164]

law?

[@karlsoriso2304]

pov: isagi is about to do somethin lit

[@NTC]

holy... amazing!!! and inspiring

[@LordBackuro]

Kinda looks like what i imagine the Universe to be like

[@d.........]

War between gravity and entropy.

[@grecinos2]

I'm not sure what the deal is, but my replies keep getting deleted. Anyways... I installed the CUDA SDK, and was able to run the demos. They are great, and close to what I want to accomplish. My goal is to make a 2d Nbody sim with 50-100k particles. Pretty much, just for the eye-candy. I have VS 12, and the CUDA samples were made in VS 8 and 10. Needless to say, I was unable to compile the samples (yet). I think I need to set the Path, referencing the CUDA's libraries, includes, etc.

[@prostytroll]

I don't see any consistent circular motion...

[@bsgsujgd3406]

this is great to watch once youve studied astrophysics

[@Knowledge.o.Everything]

Name of simulator or this software

[@rejeenaldflores5039]

how did you do that?

[@1.4142]

It went supernova

[@sleeppy66]

this is from an app for the apple ipad ( the original)

[@MisterClassified]

Is that a supernovae at 5:20.

[@mm-ov2gg]

Just need some otherworldly music and it's perfect

[@SimSoupYT]

computers this good existed in 2012?

[@YoungDegenerate]

Darkness cannot exist without light and light Cannot exist without darkness.

[@h3nnn4n]

Real time? How? gpgpu?

[@StefanoBusnelliGuru]

Gravitation on a toroidal manifold :D

[@lol532-t2g]

Now we can create our own virtual universes using OpenGL, can't we?

[@karolakkolo123]

No, we would have to know all the physics rules and equations, but for now we don't know very much about our universe. Including dark energy, antimatter, and other elementary particles that we yet have to discover

[@bitterlemonboy]

Have fun trying to gather the computing resources for that

[@swarog3]

I am just trying to implement the Einstein's gravity. Later I dream to make a little atom. At the end I want to create a world with animals having artificial intelligence brains.

[@SeanSkyhawk]

This is basically gravity on a toroidal plane

[@darchojandreoski2451]

Der Weltraum Raumschiff sagte 3 Milliarden Pixel orteten wir Jahr 1988 /1989 auf Planet Erde

[@zloykushnir]

Thats fucking awesome!

[@DigitalAlligator]

2d or 3d

[@xirustam]

Does it really work in real time?... 500.000 x 500.000 = 250.000.000.000 calculations per frame.

[@C0braGameplays]

It looks like there is black hole in there

[@sirahmad]

the animation is beautiful but i still didn't understand how its relate to newton law of gravity

[@openleft4214]

Got really confuse when I thought this was a GR sim.

[@zackbarkley7593]

Do the same simulation using MOND theory.

[@JohnDoe_1237]

damn, you need some really cool music for this video!