Question regarding fundamental region of a lattice

In summary: It's not difficult to check that a given point is in a fundamental region. All you have to do is find a point in the region, and see if it's in the set of points that you're looking for. It's not difficult to check that a given point is in a fundamental region. All you have to do is find a point in the region, and see if it's in the set of points that you're looking for.
  • #1
Peter_Newman
155
11
Hello,

I want to prove that for any lattice ##\Lambda = \Lambda(B)## (the ##B## is a basis) the orthogonalized parallelepiped ##P(B^*) = B^*\left[-0.5,0.5\right)^n## is a fundamental region of a lattice.

If I wanted to show this, I would try to establish a volume argument here. After all, the determinant of a lattice is no more than the volume of the fundamental region, by definition. Now I already know that ##det(\Lambda) = |det(B)| = |det(B^*U)| = \prod ||b_i^*||##. The ##B = B^{*}U## is just the Gram-Schmidt composition as matrix, ##B^*## is an orthogonal matrix and ##U## is an upper triangular matrix containing the Gram-Schmidt coefficients. So now we go on, by asking what is the volume of ##P(B^*)##, that is the determinant ##det(B^*)## and that is just the product again ##\prod ||b_i^*||##. So here we have a direct relationship between ##vol(P(B))## and ##vol(P(B^*))##. If we already know that ##P(B)## is a fundamental region, then we can conclude from the volume argument that ##P(B^*)## is also a fundamental region.

This raises two questions for me. First, is this meant as a "proof" in the right way? Second, can we formalize this a little better, make it a little more concrete?
 
Last edited:
Physics news on Phys.org
  • #2
This is largely unnecessary.

It is straightforward that if [itex]B[/itex] is a basis, then [itex]B^{*}[/itex] is a basis. Thus if [itex]P(B)[/itex] is a fundamental region for [itex]\Lambda (B)[/itex] then [itex]P(B^{*})[/itex] is a fundamental region for [itex]\Lambda(B^{*})[/itex], because anything you can prove for an arbitrary lattice [itex]\Lambda(B)[/itex] also holds for the lattice [itex]\Lambda(B^{*})[/itex].
 
  • Like
Likes Peter_Newman
  • #3
Hello @pasmith, thank you for your reply! I can understand what you write. I would then like to ask whether the argument about the volume would be correct, or how one leads the proof about the volume concretely.
 
  • #4
I think reference to volume is not necessary. The result follows from the 1D case: for every [itex]x \in \mathbb{R}[/itex] there exists a unique [itex]N \in \mathbb{Z}[/itex] and a unique [itex]r \in [a, a + b)[/itex] such that [itex]x = Nb + r[/itex].
 
  • Like
Likes Peter_Newman
  • #5
pasmith said:
I think reference to volume is not necessary. The result follows from the 1D case: for every [itex]x \in \mathbb{R}[/itex] there exists a unique [itex]N \in \mathbb{Z}[/itex] and a unique [itex]r \in [a, a + b)[/itex] such that [itex]x = Nb + r[/itex].
Yes, I agree on that. But the hint for the proof is to consider the volume of ##P(B^*)## :smile:
 
  • #6
Well, I'm still interested in a solution. The hint for the solution is given: "Notice that the volume of ##P(B^*)## is ##\prod_i ||b_i^*||##, as expected". I am interested if my suggestion from above is correct so far :)
 
  • #7
pasmith said:
[...] anything you can prove for an arbitrary lattice [itex]\Lambda(B)[/itex] also holds for the lattice [itex]\Lambda(B^{*})[/itex].
@pasmith I thought about this sentence again and wondered how you could justify it. I would refer this to the "arbitrary", then ##\Lambda(B^*)## is "only" a special case of ##\Lambda(B)##. Is that what you meant by that?
 
  • #8
Yes.
 
  • Like
Likes Peter_Newman
  • #9
The volume argument isn't enough. A fundamental region has to contain a single lattice point, and tile the whole space (I think there are other similar definitions). A 100x0.01 rectangle in 2 dimensions is not a fundamental region of the integer lattice, but does have a volume of 1. In this case if you center it at the origin it contains many more than 1 lattice point (which means when you shift it, it overlaps with itself in the x direction, and misses space in the y direction)
 
  • Like
Likes Peter_Newman
  • #10
Thank you @Office_Shredder for your answer! So if you continue the thought you would have to check if there is an overlap. If there is no overlap, then it is a fundamental region.

Then one would have to show that e.g. ##x = \Lambda(B) + P(B^*)## is unique? or?
 
  • #11
What does x= mean there? The right hand side is a set of approximately the whole space (depending on whether it is a fundamental region or not)
 
  • #12
Sorry, I had forgotten to mention that here. The ##x## stands for ##x \in span(B)##.
 
  • #13
Office_Shredder said:
A fundamental region has to contain a single lattice point, and tile the whole space
Good. Then the question would be how to show that ##P(B^*)## is a fundamental region of ##\Lambda(B)##.

The other question: how to show that ##P(B^*)## is a fundamental region of ##\Lambda(B^*)## (note the change here) is not difficult.

My problems are with the first question: how to show that ##P(B^*)## is a fundamental region of ##\Lambda(B)##.

For this, I would define ##x = P(B^*) + \Lambda(B)## and check that every ##x \in R^n## can be uniquely represented by ##P(B^*) + \Lambda(B)##. However, this is not so simple...
 
  • #14
Given that you have the right volume, if the lattice shifts don't intersect they have to cover the whole space. Consider an enormous sphere, so that effects around the surface don't matter, then you have an equal number of non intersecting volumes using this region vs a region you know is a fundamental unit (all the lattice shifts in the sphere) which both have the same volume, and the latter leaves no empty space, so the former can't leave empty space either.

Suppose ##z\in x+P\cap y+P## for ##x,y## lattice points and z any arbitrary point. Then ##z=x+p_1=y+p_2## for ##p_1,p_2\in P##. Hence ##x-y=p_2-p_1##.

##x-y## is another lattice point, so a contradiction can be generated by showing you can't get a lattice point by adding two elements of the region (I've been a little sloppy here, you can add two elements from the surface of the region and get a lattice point, the real issue is if you can add two interior points. I'll let you think it over)
 
  • #15
Office_Shredder said:
Suppose ##z\in x+P\cap y+P## for ##x,y## lattice points and z any arbitrary point. Then ##z=x+p_1=y+p_2## for ##p_1,p_2\in P##. Hence ##x-y=p_2-p_1##.
Hey @Office_Shredder, thanks for your answer! I would also proceed similarly. See below:

Peter_Newman said:
For this, I would define ##x = P(B^*) + \Lambda(B)## and check that every ##x \in R^n## can be uniquely represented by ##P(B^*) + \Lambda(B)##. However, this is not so simple...
We would have to show uniqueness and assume that there is an overlap, then let ##x_1 = P(B^*) + \Lambda(B)## and ##x_2 = P(B^*) + \Lambda(B)##. If there is an overlap, then it means that ##x_1 = x_2##, this is noting else then
$$x_1 = P(B^*) + \Lambda(B) = a_1 b_1^* + ... + a_n b_n^* + a_1'b_1 + ... + a_n'b_n$$
and
$$x_2 = P(B^*) + \Lambda(B) = c_1 b_1^* + ... + c_n b_n^* + c_1'b_1 + ... + c_n'b_n$$
with ##a_i,c_i \in## [0,1) and ##a_i', c_i' \in \mathbf{Z}##.

The problem now is that once here we have the Gram Schmidt vectors and the basis vectors. You could try to rewrite the basis vectors, but I think that makes things even more difficult. Even if I am sure that this is the way for the proof. But what happens next?

Peter_Newman said:
The other question: how to show that ##P(B^∗)## is a fundamental region of ##Λ(B^∗)## (note the change here) is not difficult.
This can be shown relatively easy with the idea from above, this is easier since we only have one "sort" of vectors. The calculation of uniqueness works out excellently here!
 
  • #16
The ##a_i## and ##c_i## are actually in ##[-0.5,0.5)##.

This is actually important (though obviously shifting a fundamental region doesn't change its fundamentalness) because you can rearrange to
##\sum m_i b_i = \sum x_k b^*_m## where ##m_i## are integers, and ##x_k\in [-1,1)##. I would recommend assuming ##x_k\neq -1## for your first pass at the proof, as it just adds technical annoyance. The goal is to prove all the coefficients are zero. Remember that ##b^*_k## is orthogonal to many of the ##b_i##s, which should help you go through step by step and prove that the coefficients must all be zero.
 
  • Like
Likes Peter_Newman
  • #17
Hey! Yes the coefficients should be in ##[-0.5,0.5)##. according to the original definition, that is correct. But as you also say, the interval is not necessarily crucial, so I thought I'd just try it for [0,1) first. You probably mean [0,1) instead of [-1,1], right?

Office_Shredder said:
The goal is to prove all the coefficients are zero.
Yes that's what I wanted to show. My next step would be:

$$x_1 = P(B^*) + \Lambda(B) = a_1 b_1^* + ... + a_n b_n^* + a_1'b_1 + ... + a_n'b_n$$
$$x_2 = P(B^*) + \Lambda(B) = c_1 b_1^* + ... + c_n b_n^* + c_1'b_1 + ... + c_n'b_n$$

$$x_1 = x_2 $$
$$a_1 b_1^* + ... + a_n b_n^* + a_1'b_1 + ... + a_n'b_n = c_1 b_1^* + ... + c_n b_n^* + c_1'b_1 + ... + c_n'b_n$$
$$a_1 b_1^* + ... + a_n b_n^* -(c_1 b_1^* + ... + c_n b_n^*) = c_1'b_1 + ... + c_n'b_n - (a_1'b_1 + ... + a_n'b_n)$$
$$(a_1 - c_1)b_1^* + ... + (a_n - c_n)b_n^* = (c_1' - a_1')b_1 + ... + (c_n' - a_n')b_n$$

At this point, however, we have the problem that two kinds of vectors appear and we have to decide what to transform. One idea would be to replace the ##b_i##'s by Gram Schmidt vectors, but this would introduce new coefficients. Here I don't see a solution to move forward right now....
 
Last edited:
  • #18
Peter_Newman said:
Hey! Yes the coefficients should be in ##[-0.5,0.5)##. according to the original definition, that is correct. But as you also say, the interval is not necessarily crucial, so I thought I'd just try it for [0,1) first. You probably mean [0,1) instead of [-1,1], right?

If you add two arbitrary numbers in ##[-.5,.5)## the sum could be anywhere in ##[-1,1)##
 
  • #19
Regarding my post #17, I would continue by writing ##b_i##'s in the Gram Schmidt version. This results on the right hand side in:

$$(a_1 - c_1)b_1^* + ... + (a_n - c_n)b_n^* = (c_1' - a_1')b_1 + ... + (c_n' - a_n')b_n$$
$$(a_1 - c_1)b_1^* + ... + (a_n - c_n)b_n^* = (c_1' - a_1')b_1^* + ... + (c_n' - a_n')(b_n^* + \sum_{i=1}^{n-1} db_i^*)$$

where ##d## is arbitrary Gram Schmidt constant.
If one compares the coefficients e.g. for ##b_n^*## this is
$$(a_n - c_n) = (c_n' - a_n')$$
and since the right hand side is an integer number we can say ##(c_n' - a_n') = 0## and then we have ## a_n = c_n##.

We can proceed in that manner and obtain that for all other coefficients we have ## a_i = c_i ##. This shows then that ##x_1 = x_2## are unique since the coefficients are all equal.

Is this right so far, or did I miss something?
 
  • #20
That looks right to me.
 
  • Like
Likes Peter_Newman
  • #21
Excellent. Then I would say the proof is complete. Thank you @Office_Shredder!
 

1. What is the fundamental region of a lattice?

The fundamental region of a lattice is the smallest region in which all points of the lattice can be generated by translations. It is usually defined as a parallelogram or a rectangle.

2. How is the fundamental region of a lattice related to the lattice points?

The fundamental region of a lattice contains exactly one lattice point at each of its corners. This means that all other lattice points can be generated by translating the fundamental region.

3. Can the fundamental region of a lattice be any shape?

No, the fundamental region of a lattice must have a specific shape in order to contain all lattice points. It must be a parallelogram or a rectangle, and the angles and sides of the shape are determined by the lattice basis vectors.

4. How is the fundamental region of a lattice used in crystallography?

In crystallography, the fundamental region of a lattice is used to represent the repeating pattern of atoms in a crystal structure. It is also used to determine the symmetry and symmetry operations of the crystal lattice.

5. Is the fundamental region of a lattice unique?

Yes, the fundamental region of a lattice is unique for a given lattice. However, different lattices can have the same fundamental region if they have the same lattice basis vectors.

Similar threads

  • Linear and Abstract Algebra
Replies
21
Views
1K
  • Linear and Abstract Algebra
Replies
7
Views
805
Replies
3
Views
2K
  • Electrical Engineering
Replies
10
Views
2K
Replies
5
Views
1K
  • Calculus and Beyond Homework Help
Replies
6
Views
3K
  • Linear and Abstract Algebra
Replies
9
Views
3K
Replies
3
Views
1K
  • Engineering and Comp Sci Homework Help
Replies
8
Views
1K
Replies
19
Views
1K
Back
Top