Proof of the inequality of a reduced basis

In summary, the LLL-reduced basis satisfies the following property (Reference): For ##1 \leq i \leq n## concider ##H = span(b_1,...,b_{i-1},b_{i+1},...,b_n)##. Show that ##2^{-n(n-1)/4}||b_i|| \leq dist(H,b_i) \leq || b_i ||##. Hint use ##\prod ||b_i|| \leq 2^{n(n-1)/4} det(\Lambda)##.
  • #1
Peter_Newman
155
11
TL;DR Summary
For ##1 \leq i \leq n## concider ##H = span(b_1,...,b_{i-1},b_{i+1},...,b_n)##. Show that ##2^{-n(n-1)/4}||b_i|| \leq dist(H,b_i) \leq || b_i ||##. Hint use ##\prod ||b_i|| \leq 2^{n(n-1)/4} det(\Lambda)##. Own reasoning below.

I would like to show that a LLL-reduced basis satisfies the following property (Reference):

For ##1 \leq i \leq n## concider ##H = span(b_1,...,b_{i-1},b_{i+1},...,b_n)##. Show that ##2^{-n(n-1)/4}||b_i|| \leq dist(H,b_i) \leq || b_i ||##. Hint use ##\prod ||b_i|| \leq 2^{n(n-1)/4} det(\Lambda)##.



My Idea:
I also have a first approach for the part ##dist(H,b_i) \leq || b_i ||## of the inequality, which I want to present here based on a picture, which is used to explain my thought:

pic.png


So based on the picture we can see that the norm of the orthogonal part (##\tilde{b_i}##) of the vector ##b_i## gives the distance. Therefor I would argue that ##dist(H,b_i) = ||\tilde{b_i}||## and further ##||\tilde{b_i}|| \leq ||b_i||## since the norm of a Gram-Schmidt vector is less than the norm of the regular vector. If I put this together it makes for me a valid argument to say ##dist(H,b_i) \leq || b_i ||##. This would prove one part of the inequality.



Problem:
However I do not manage to show ##2^{-n(n-1)/4}||b_i|| \leq dist(H,b_i)## with the specific hint. There are other useful properties of an reduced basis (see e.g. the reference) with which it is easier to show, but with the hint there I have no idea. This brings me to the question, if someone here might have an idea.



Remark:
This is not a homework exercise, it's just a question I'm thinking about for myself...
 
Physics news on Phys.org
  • #2
I wanted to ask again if anyone would consider my reasoning to be correct? Or can someone contradict that?
 
  • #3
Do you know how ##\det(\Lambda)## is built up by computing ##||\tilde{b_i}||##s?
 
  • Like
Likes Peter_Newman
  • #4
Hey @Office_Shredder, If I remember correctly, it was this: ##\det(\Lambda) = \prod_{i=1}^n ||\tilde{b_i}||##
 
  • #5
That's really the only clever piece. I'll put the rest in spoilers but it's just doing the right algebra, so you should give it a shot yourself.

plugging this into
##\prod ||| b_j|| \leq 2^{n(n-1)/4} \det(\Lambda)##

##\prod ||b_j|| \leq 2^{n(n-1)/4} \prod ||\tilde{b_j}||##

For each ## j\neq i##, ##||\tilde{b_j}|| \leq ||b_j||##. Apply this to the right hand side to get

##\prod ||b_j|| \leq 2^{n(n-1)/4} ||\tilde{b_i}|||\prod_{j\neq i} ||b_j||##

Cancel all the norms left on both sides
 
  • Informative
Likes Peter_Newman
  • #6
Hey @Office_Shredder, then in the end remains: ##2^{-n(n-1)/4}||b_i|| \leq ||\tilde{b_i}||##.

But with my thought above that ##dist(H,b_i) = ||\tilde{b_i}||##, we can derive the required inequality (left side). But this assumes that ##dist(H,b_i) = ||\tilde{b_i}||## is right, and I'am not sure, can you confim that this is right?
 
  • #7
You are correct.
 
  • Like
Likes Peter_Newman
  • #8
Nice! Then thank you very much!

I must say, unconsciously I was able to show the inequality also over another property of the reduced basis (The hint has complicated this however rather :) ).
 

FAQ: Proof of the inequality of a reduced basis

What is "Proof of the inequality of a reduced basis"?

"Proof of the inequality of a reduced basis" is a mathematical concept used in the field of numerical analysis. It is a method of proving that a reduced basis, which is a subset of a larger set of vectors, has a lower bound on its norm that is dependent on the size of the original set of vectors.

Why is proving the inequality of a reduced basis important?

Proving the inequality of a reduced basis is important because it allows us to determine the accuracy and efficiency of using a reduced basis in numerical computations. It also helps us understand the trade-off between accuracy and efficiency, and can guide us in choosing the optimal reduced basis for a specific problem.

How is the inequality of a reduced basis proven?

The inequality of a reduced basis is typically proven using mathematical techniques such as the Cauchy-Schwarz inequality or the Gram-Schmidt process. These techniques involve manipulating the vectors in the reduced basis to show that their norm is bounded by a certain value.

What are the limitations of using a reduced basis?

One limitation of using a reduced basis is that it can only be applied to problems with a certain structure, such as linear or affine problems. Additionally, the accuracy of the reduced basis approximation may decrease as the size of the original set of vectors increases.

What are the practical applications of the inequality of a reduced basis?

The inequality of a reduced basis has practical applications in various fields, including computational fluid dynamics, structural mechanics, and data analysis. It can be used to improve the efficiency of numerical simulations and reduce the computational cost of solving large-scale problems.

Similar threads

Replies
4
Views
1K
Replies
23
Views
1K
Replies
20
Views
2K
Replies
3
Views
1K
Replies
12
Views
1K
Replies
3
Views
1K
Replies
52
Views
3K
Replies
4
Views
1K
Back
Top