Applying Monte-Carlo integration to solving almost-generic PDE without using any kind of meshing of the domain.
Like with most stuff Keenan Crane and his lab produces and has produced) ... super high-quality research, easy to understand, almost immediately applicable to engineering problems, and super-well explained.
Maybe this is the standard for the computer graphics community, but I love how the paper is accompanied by this great webpage and videos. Very inspiring, this is how publications should be done in this day and age: Alongside with material making it easier to understand it.
Does anyone know how he writes his papers? His illustrations will forever be beyond my skills, but I’d like to up my paper mojo. Since his references are in bibtex, I assume he’s using some flavor of TeX, but this looks more sophisticated than LaTeX. In particular, he has small sub-illustrations next to his section headings that are really hard to get right using LaTeX. Is he using ConTeXt? Which fonts does he use? The ligatures are really nice, also hard to do using standard LaTeX available fonts.
Wow, you're correct, he uses ACM TOG for the paper. As a scientist, I mostly just use the REVTeX packages or tufts-latex. Evidently there's a brave new world of other templates out there that I need to learn about. Thanks.
The inset figures are probably placed using the “wrapfigure” LaTeX package. It’s janky and frustrating to use, but I agree the results look great!
EDIT: Incidentally the template changed recently and people within the graphics community complained about some aspects of it, e.g. the equation typesetting and how subsections and lists are formatted. It's good to hear that people outside the community like it.
I'm not sure this is the standard. I have a feeling Keenan Crane is trying to set a (fantastic) example for others to follow. His presentations are really top notch and something everyone should strive for, a really inspiring smart guy.
Noob Q could this be used to solve other problems that uses PdE for example fluid dynamics? My naive mind has a made up example like simulating how rocket fuel might behave when subjected to certain pressure/temperature/injection method
Currently the method only works for elliptic, linear PDEs like the heat equation. Keenan and his students are trying to extend the method to more general PDEs (they recently had a paper showing how to do elliptic PDEs with non-constant coefficients) but currently the Navier-Stokes equation from fluid dynamics is out of scope.
That's my view of the method in a nutshell: using walk-on-spheres as a practical method to solve PDEs arising in simulations is a departure from what most people have tried in recent years. Whether it's a total game-changer or a curious novelty will depend on whether it generalizes beyond diffusion-like PDEs.
> babe wake up, a new keenan crane lecture just dropped.
First time I hear about Keenan Crane, but gosh, from now on, I'll keep an eye for videos produced by him. Like I keep an eye out for 3Blue1Brown, Mathologer and Veritasium. Now Keenan Crane is on that list. Except, he's a bit higher.
18 comments
[ 19.6 ms ] story [ 72.2 ms ] threadLike with most stuff Keenan Crane and his lab produces and has produced) ... super high-quality research, easy to understand, almost immediately applicable to engineering problems, and super-well explained.
What a treat!
Well worth the watch.
Maybe this is the standard for the computer graphics community, but I love how the paper is accompanied by this great webpage and videos. Very inspiring, this is how publications should be done in this day and age: Alongside with material making it easier to understand it.
https://www.cs.cmu.edu/~kmcrane/Projects/Other/IllustratingG...
For example, a recent paper of his team "Repulsive Surfaces" is on arXiv, where the source files of the preprint PDF are available for download
https://arxiv.org/format/2107.01664
The inset figures are probably placed using the “wrapfigure” LaTeX package. It’s janky and frustrating to use, but I agree the results look great!
EDIT: Incidentally the template changed recently and people within the graphics community complained about some aspects of it, e.g. the equation typesetting and how subsections and lists are formatted. It's good to hear that people outside the community like it.
https://www.lyx.org/
That's my view of the method in a nutshell: using walk-on-spheres as a practical method to solve PDEs arising in simulations is a departure from what most people have tried in recent years. Whether it's a total game-changer or a curious novelty will depend on whether it generalizes beyond diffusion-like PDEs.
engineers in life: "OMG how do you draw like that, amazing" (guy draws a ball)
"OMG how do u explain things so well" (uses basic essay structure)
After much futzing it seems to be the only sane way. Fast too.
> babe wake up, a new keenan crane lecture just dropped.
First time I hear about Keenan Crane, but gosh, from now on, I'll keep an eye for videos produced by him. Like I keep an eye out for 3Blue1Brown, Mathologer and Veritasium. Now Keenan Crane is on that list. Except, he's a bit higher.
You might want to start with this most excellent series from 2021 on discrete differential geometry.
https://www.youtube.com/watch?v=mas-PUA3OvA
This guy really has a gift to explain complex math problems in a very intuitive fashion.
There's also the fact that - unlike much advanced math stuff - most of the things he presents is very new and can actually be implemented.