iiinova — where your digital self keeps evolving

Repeated subdivision of triangles by centroid- infinite zoom

58513

Well maybe there is one at another depth, or maybe we can get arbitrarily close- don’t know!

010

another deep render, from filling the triangles

2312

No it is about 4% off but I sneak in a transformation to get a perfect loop

271

Repeatedly subdividing a triangle using its centroids

1021936

Reposted by Matt Henderson

And here is how to make a parabola with sand

1452

And here is how to make a parabola with sand

1452

Real life demo (me in 2020)

1352

Matt Henderson @matthen.com · 18/09/2026

How to make an ellipse, using sand and physics

1538648

Then I think you are pretty limited if you also say you want linearity and possibly commutativity!

051

Matt Henderson @matthen.com · 18/09/2026

Yeah I would say we prefer v times w to be an object that doesn’t depend on the coordinate system we happened to choose

180

It is an honour to inspire you! I did this same visualisation recently for ball bouncing inside a circle and ball bouncing on a floor of circles. Curious what you think to do with a Galton board.

Good question! Yes in a simulation it is deterministic. There is numerical precision that will make solutions diverge depending on how you compute things, but with infinite precision it’s deterministic. So to visualise the predictability I add small random noise to the velocity and project forward

270

Backwards Steps in Rigid Body Simulation

Their Backward Steps in Rigid Body Simulation (graphics.cs.cmu.edu/projects/bac...) is also similar line of thinking to my two chambers post: bsky.app/profile/matt...

graphics.cs.cmu.edu

160

Oh yeah this seems very similar line of thinking

Trying to forecast the future on a Galton board

1658081

This is just entropy based on the counts in left and counts in right

121

bsky.app/profile/matt...

120

It is event driven elastic collisions. So it solves to find the next collision rather than incrementing timesteps. Energy is conserved up to numerical precision. So is momentum in ball ball collisions, ball wall collisions reflect the balls so don’t conserve momentum

220

Yeah very fair question!

020

I am just plotting the binomial entropy, based on the count in left and count in right

220

Okay yea i suppose it is equivalent to flipping all the velocities. So you put all the balls in one chamber and run the sim, save those frames. Then you go back to start, flip velocities and run sim again. Save those frames, reverse and prepend. That gets you one entropy drop. The others from search

130

No it is not reversing the velocities it is running the sim forwards and backwards

For running time backwards I mean more your second option- the physics work just as well here with t -> -t