Lecture 18: Introduction to Physical Simulation (33)

zhangyifei-chelsea

The intuition behind adaptive step size is that numeric error tends to be large when the derivative of underlying curve changes greatly. So when the derivative varies greatly, we need to adaptively decrease the step size.

SainanChen

Tuning hyperparameters, like the threshold and step size reduction every time for repetition, is important, and it will influence the performance and running time of the parameter.

briana-jin-zhang

I am a little bit confused by the drawing of x_{T/2}. It would think that x_{T/2} is the midpoint of vector to x_T but that doesn't seem to be what's in the drawing.

catherinecang

I think here x_{T/2} is referring to 2 timesteps each with size T/2, so that's why it looks like similar length to X_T

The intuition behind adaptive step size is that numeric error tends to be large when the derivative of underlying curve changes greatly. So when the derivative varies greatly, we need to adaptively decrease the step size.

Tuning hyperparameters, like the threshold and step size reduction every time for repetition, is important, and it will influence the performance and running time of the parameter.

I am a little bit confused by the drawing of x_{T/2}. It would think that x_{T/2} is the midpoint of vector to x_T but that doesn't seem to be what's in the drawing.

I think here x_{T/2} is referring to 2 timesteps each with size T/2, so that's why it looks like similar length to X_T