Newton Solver

[jamesm6162]

Hi

I'm interested to know some more about the solver used in Newton.
What approaches are followed to get the solver accuracy so high, even in the non-exact solver?

Did you base it on existing research with papers I can read up on?

Regards

James

[Julio Jerez]

Well Newton does not really come from any paper, Newton is my own work.

I am a big fan of David Baraff and his body of work, in my opinion every single paper written after his first publication "Analytical Methods for Dynamic Simulation of Non-penetrating Rigid Bodies"
either plagiarize his work or it is some morons trying to invent a brand new Laws of Physics to go circumvent the complexity of calculating internal reaction forces.

A long time ago, I decided that the calculation are what they are and all I can hope for is that the hardware gets better, so I will focus in doing the best I could while abiding by the laws of Newtonain Mechanic. I would not implement impulsive solver because yes it true they are easy to get up and running and make for impressive fast demos, but in the end, the user is left with implausible physics, lots black art and magic numbers with not physical meaning and terrible unpredictable, unstable and inaccurate simulation.

So yes Newton will be slower but that was my choice, even when I decide to experiment with iterative solvers, this is not a limitation.

When I went to college, I learned Analytical Mechanic as part of my major in Mechanical Engineering,
(Lagragian, Hamiton, Legendre, and Poison mechanic).
My master was on the D'Lambert Method of the virtual works, which only apply to systems in equilibrium, and it the method that is used to design complex structures like building bridges, etc.
I also took classes on advanced mathematics like calculus of variations, differential Geometry and numerical analysis. In the end that math is a much bigger help than the physics.
Knowing the math you can derive almost anything if you can express it as an analytical expression.
The procedure will be consistent, but it will only be correct if the expression represents an accurate model of what you try to simulate.

Whoever while in college and even after graduation, to me the math and the physics only had academic value. I never learned how to apply the knowledge systematically solve problems, like rigid bodies until I saw the game Trespasser. Then I started to search in CompuServe for that kind of work. that's when I found Baraff Thesis and more of his work.
It took me month to really understand it, but then when it finally clicked, I realized it is actually what I learned in College, but he showed how to apply it systematically.

So not with the exception of Baraff work, which I take as inspiration, I do not use any of the MomboJombo garbage we see every year from new graduates students.

The only credit I give to those papers is the Insight into a new way of looking at a problem.
But most the time you can get a much better solution than the one in the paper by using Mathematica or even with a pen on paper to derive the method.