[WS2122] Homework 4 Discussion

[changkun]

This thread is open for the discussion of homework 4.

[JohannesMerkt]

I get this weird result when i try to solve the linear equation with the cholesky solver. Does anyone have an idea why that might be?

[changkun]

That's a good assumption. As we observe from there the solutions are too small. Maybe weights are not properly constructed.

[JohannesMerkt]

I cant use the cholesky solver since the weight values are negative.

L*U' = U

I have no idea how i can make them positive but I can use a different solver and that works

[MarlQ]

L is not even a lower triangular matrix...
Edit: Ok, that is actually irrelevant. I was a bit confused. L needs to be Hermitian positive-definite, which it isn't.

[MarlQ]

I tried using a different solver as well, and my solution is suddenly very nice. With Cholesky it doesn't work at all.

[changkun]

Oh. Sorry I completely missed the messages here... I hope this does not cause too much trouble.

As discussed in the class, the Cholesky solver is not a suitable solver because the matrix does not satisfy symmetric. For this type of linear equation, we need to use other solvers such as a general LU solver. I should update the comments on the provided code too.

[alvaxgln]

When calculating the coordinates for the boundary vertices, they should be evenly distributed on the circle/rectangle, right?
Also can we just pick any radius/side length?

[JohannesMerkt]

The uv display area is from negative 0.5 to positive 0.5 in both x and y axis. so if you want the uvs to be in that area you should pick the radius 0.5 and side length 1

[JohannesMerkt]

For an optimal result the boundary vertices should be evenly distributed i think

[alvaxgln]

Ok, thank you :)

[changkun]

Yep, as Johannes mentioned. Uniformly distributed vertices should be better for looping with a fixed step (for x := vec(0, 0); x.neq(vec(0,0)); x.inc(step) { ... })

[changkun]

Close due to the end of the submission period.