Multi-Scale Turing Patterns 2

March 15, 2019

The time has come for another batch of multiscale turing patterns (see part one here). These are all generated with my multiscale turing pattern implementation.

New in recent history is persisting configuration in the URL. You can generate the first pattern with this link. I generally experiment at low resolution, then constrain it to a single tile (inDiv = yes), then set the pixel ratio to 4 (pixelRat = 4) and the resolution to 2122^{12} (res = 12) and . The configuration setup leaves a lot to be desired, but such is the way of things when trying to crank out interesting things in your spare moments.

With this batch, I feel like I’ve explored a fair amount of the parameter space of my current implementation—which is to say I sat there watching a movie hitting the “randomize” button over and over—so that it’d be fun to try some other features. If I were to modify my implementation, I’d do the following:

  • brute-force convolutions with localized kernels to avoid unnecessary FFT passes.
  • try kernels that are neither circular nor gaussian
  • these patterns work by computing the local variation for a number of different scales (as determined by convolution with two different sized kernels). I might try convolving the computed local variation with another kernel to explore smoother patterns.
  • add rotational symmetry in the style of Jonathan McCabe’s paper, Cyclic Symmetric Multi-Scale Turing Patterns.

But until then, enjoy! And feel free to drop me a line via twitter