[HN Gopher] Numerical Linear Algebra Class in Julia TUM
___________________________________________________________________
Numerical Linear Algebra Class in Julia TUM
Author : darboux
Score : 134 points
Date : 2025-05-03 21:22 UTC (1 days ago)
(HTM) web link (venkovic.github.io)
(TXT) w3m dump (venkovic.github.io)
| staplung wrote:
| Not exactly the same material but U. Michigan has their Robotics
| 101 course up as well: Computational Linear Algebra, also in
| Julia.
|
| https://github.com/michiganrobotics/rob101/tree/main
| ted_dunning wrote:
| This is a nicely comprehensive course, but it looks like it is
| pretty fast paced, especially in the last few lectures (some of
| those later slides definitely aren't finished).
|
| As a reference, it looks very useful.
| stabbles wrote:
| A good resource is Gerard Sleijpen's course:
| https://webspace.science.uu.nl/~sleij101/Opgaven/NumLinAlg/
| me3meme wrote:
| I just selected lecture 07 to take a look: Lecture 07 is about QR
| factorizacion and Householder reflections. The author proves how
| to construct a reflection to make zeros in the first column and
| then he just claims that following this procedure for the other
| columns finish the proof. But he should prove or justify why the
| other reflections do not destroy the zeros of previous
| reflections. Also he proves that a vector v is the vector to
| construct the reflection (but there is a factor of 2 that was not
| correctly simplified, maybe a latex error), but I think that it
| should be more general and easier to prove that for any w the
| vector from w to its image f(w) is the orthogonal vector to the
| plane of the reflection.
|
| I thank the author for the slides, but this little proof need
| some more care, I don't know about the quality of other sections
| or the overall quality of the slides. Anyway I like how he tries
| to make things easy but good work is hard.
|
| Edited: I was wondering whether a LLM reading Lecture 7 would
| detect what was missing in the proof. I tried with deepseek but
| its first feedback on the Lecture 7 was positive, then when
| prompted about the incomplete proof it recognized it as a common
| error and explained how to complete the proof. Also I have to
| prompt it about the bad factor 2 for it to detect it. So it seems
| that deepseek is not a useful tool to judge quality of math
| content without very expert guidance, deepseek suggested to ask
| the LLM to compare this proof with another proof to detect
| important or vital differences.
| Certhas wrote:
| That's an absolutely obvious step though? As in, detailed
| lecture notes should maybe elaborate with a sentence, but in a
| lecture I would not put this on the slides but mention the core
| point and expect students at this level (who should have seen
| some amount of more theoretical LinAlg courses by then) to
| understand how to do the 1 line calculation.
|
| There aren't even any real details to fill in, you iterate on
| the lower right block so anything you do is orthogonal to the
| upper left block. Do a 2x2 block matrix multiplication to
| convince yourself that this preserves the form achieved so far.
| me3meme wrote:
| -- Do a 2x2 block matrix multiplication to convince yourself
| that this preserves the form achieved so far.
|
| I don't consider this a proof. Perhaps you have in mind two
| simple but key properties of reflections about the hyperplane
| orthogonal to a vector v: (a) The hyperplane of a reflection
| is the fixed point of the reflection (b) the hyperplane is
| the orthogonal vector space to the vector space spanned by v.
| From this two properties it follows that each step of making
| zeroes does not change previous zeroes.
|
| Your claim that for advanced students there is no need to
| comment about details it is not falsifiable. Citing Mac Lane:
| A monad is just a monoid in the category of endofunctors.
|
| But from a practical point of view one can see the very basic
| level and simplicity of the definitions and calculations
| prior to the proof. So at this level of detail I consider
| that noticing that one must be careful to not destroy
| previous zeros is matching the level of discourse at the
| proper level.
___________________________________________________________________
(page generated 2025-05-04 23:02 UTC)