Are the p-norm and and the p'-norm equivalent?

  • Thread starter dirk_mec1
  • Start date
  • Tags
    Equivalent
In summary, we can prove that the p-norm and p'-norm are equivalent for 1 ≤ p < ∞ and 1 ≤ p' < ∞ by showing that the norms are bounded above and below by constants M and m, respectively. For the upper bound, we can use the fact that if the p-norm is 1, then the q-norm must be less than or equal to 1. For the lower bound, we can use the fact that if the p-norm is 1, then at least one element of the vector must be larger than (1/k)^(1/p), which implies that the q-norm must also be greater than this value. Therefore, we can conclude that the
  • #1
dirk_mec1
761
13

Homework Statement


Are the p-norm and and the p'-norm equivalent?

With
[tex]1 \leq p < \infty[/tex] and [tex]1 \leq p' < \infty[/tex]


Homework Equations



Two norms p and q are equivalent if there is a M>0 and m>0 such that [tex] ||x||_p \leq M ||x||_q [/tex] and

[tex] ||x||_q \leq m ||x||_p [/tex]

The Attempt at a Solution


I do not seem to get started. How can I find the constants?
 
Physics news on Phys.org
  • #2
Before expecting help you need to
1. Tell us what vector space ##x## is in.
2. Show us what you have tried.
 
  • #3
[tex] x \in \mathbb{R} ^k [/tex]

Suppose I take k = 2 and I presume p>q (I do not like the accent so I choose to use q) then I need to prove that:[tex] (|x_1|^p + |x_2|^p)^{1/p} \leq M (|x_1|^q + |x_2|^q)^{1/q} [/tex]This is not correct but if I assume the vector elements to be larger than one then:

I know that [itex] |x_i|^p \geq |x_i|^q [/itex] with i=1,2.

I also know that [itex] |x_i|^{1/p} \leq |x_i|^{1/q} [/itex] with i=1,2.

Then I'm stuck.
 
Last edited:
  • #4
Suppose ##1 \leq p < q < \infty##.

If ##\|x\|_p = 1##, can you find an upper bound for ##|x_i|^p##? What does this imply about ##\|x\|_q## in this case?
 
  • #5
if [itex] ||x||_p =1 [/itex] then [itex] |x_i|^p \leq 1 [/itex] which implies then that [itex] |x_i|^q \leq |x_i|^p [/itex] for i = 1,...,k but then:

[tex] |x_1|^q +... + |x_k|^q \leq |x_1|^p +... + |x_k|^p [/tex]

taking root at both sides

[tex] (|x_1|^q +... + |x_k|^q)^{1/p} \leq (|x_1|^p +... + |x_k|^p)^{1/p} [/tex]

but the RHS is 1

[tex] (|x_1|^q +... + |x_k|^q)^{1/p} \leq 1 [/tex]

taking root at both sides

[tex] ||x||_q^{1/p} \leq 1 [/tex]

This has an lower bound

[tex] ||x||_q \leq ||x||_q^{1/p} [/tex]So the conclusion is that:

[itex] ||x||_q \leq ||x||_p [/itex] with [itex] ||x||_p=1[/itex]Suppose the vector x has length A (in the p-norm) then we can adapt this to:

[itex] ||x||_q \leq ||x||_p \cdot A^{1/q} [/itex]Is this correct?
 
Last edited:
  • #6
dirk_mec1 said:
if [itex] ||x||_p =1 [/itex] then [itex] |x_i|^p \leq 1 [/itex] which implies then that [itex] |x_i|^q \leq |x_i|^p [/itex] for i = 1,...,k but then:

[tex] |x_1|^q +... + |x_k|^q \leq |x_1|^p +... + |x_k|^p [/tex]

taking root at both sides

[tex] (|x_1|^q +... + |x_k|^q)^{1/p} \leq (|x_1|^p +... + |x_k|^p)^{1/p} [/tex]

but the RHS is 1

[tex] (|x_1|^q +... + |x_k|^q)^{1/p} \leq 1 [/tex]

taking root at both sides

[tex] ||x||_q^{1/p} \leq 1 [/tex]
OK so far.

This has an lower bound

[tex] ||x||_q \leq ||x||_q^{1/p} [/tex]
OK, here you're using the fact that if you take the right hand side to the ##p##'th power, then the result is smaller since ##\|x\|_q^{1/p} \leq 1##.

You can save a few steps if you avoid taking ##p##'th roots. If ##\|x\|_p = 1##, then ##|x_1|^p +... + |x_k|^p = 1##, so
$$|x_1|^q +... + |x_k|^q \leq |x_1|^p +... + |x_k|^p = 1$$
and therefore
$$\|x\|_q = (|x_1|^q +... + |x_k|^q)^{1/q} \leq 1 = \|x\|_p$$

Suppose the vector x has length A (in the p-norm) then we can adapt this to:

[itex] ||x||_q \leq ||x||_p \cdot A^{1/q} [/itex]

Is this correct?
No. If ##\|x\|_p = A##, then we can write ##x = Au## where ##\|u\|_p = 1##. Then:
$$\|x\|_q = \|Au\|_q = A\|u\|_q \leq ?$$
Now apply the result obtained above.
For the other inequality ##||x||_p \leq M ||x||_q##, I recommend that you once again start with the case ##\|x\|_p = 1##. In that case, you need to prove that there is some ##M## for which
$$\|x\|_q \geq \frac{1}{M}$$
The key idea here is that if ##\|x\|_p = 1##, it's not possible for ALL of the elements of ##x## to be small. Can you find a lower bound for ##\max |x_i|##?
 
  • #7
[tex]\|x\|_q = \|Au\|_q = A\|u\|_q \leq A\|u\|_p = \|Au\|_p= ||x||_p [/tex]

So m=1 in this case?

I think that the lower bound for [itex] \mbox{max}|x_i| = (\frac{1}{k})^{1/p}[/itex]

But I don't understand how to proceed.
 
Last edited:
  • #8
dirk_mec1 said:
[tex]\|x\|_q = \|Au\|_q = A\|u\|_q \leq A\|u\|_p = \|Au\|_p= ||x||_p [/tex]

So m=1 in this case?
Yes.

I think that the lower bound for [itex] \mbox{max}|x_i| = (\frac{1}{k})^{1/p}[/itex]
Yes, this is correct. Sketch of proof: if ##|x_i| < (1/k)^{1/p}## for all ##i##, then ##|x_i|^p < 1/k## for all ##i##, so ##|x_1|^p + \cdots + |x_k|^p < 1##, contradicting the fact that ##\|x\|_p = 1##. So for at least one ##i## we must have ##|x_i| > (1/k)^{1/p}##.

For the next step, note that the above implies (for the same ##i##) that ##|x_i|^q > (1/k)^{q/p}##. What does this imply about ##\|x\|_q##?
 
  • #9
I think it implies that [itex] ||x||_q \geq \left(\frac{1}{k} \right)^{q/p} [/itex]Now suppose [itex]||x||_p=A [/itex] We can write [itex]x=Au [/itex] with [itex]||u||_p= 1[/itex] then we get[itex]||x||_q =||Au||_q =A ||u||_q \geq A \left(\frac{1}{k} \right)^{q/p} = M\cdot A||u||_p = M\cdot ||Au||_p = M ||x||_p[/itex] with [itex]M= \left(\frac{1}{k} \right)^{q/p} [/itex]

Is this correct?
 
Last edited:
  • #10
dirk_mec1 said:
I think it implies that [itex] ||x||_q \geq \left(\frac{1}{k} \right)^{q/p} [/itex]
Close but not quite right:
$$\|x\|_q^q = |x_1|^q + \cdots + |x_k|^q \geq |x_i|^q > (1/k)^{q/p}$$
where ##i## is as in my previous post. Taking ##q##'th roots of both sides, we get
$$\|x\|_q > (1/k)^{1/p}$$
The rest of your post is fine if you make the corresponding adjustment.
 
  • #11
Ok got it thanks.
 

FAQ: Are the p-norm and and the p'-norm equivalent?

1. What is the difference between the p-norm and the p'-norm?

The p-norm and the p'-norm are both ways of measuring the size of a vector in a vector space. The p-norm is calculated by taking the p-th root of the sum of the absolute values of the vector's components. The p'-norm, also known as the dual norm, is calculated by taking the p-th root of the sum of the absolute values of the vector's components, multiplied by the p-th power of the vector's length.

2. How are the p-norm and the p'-norm related?

The p-norm and the p'-norm are related through a mathematical concept known as duality. This means that for any given vector, the p-norm multiplied by the p'-norm will always be equal to the p-th power of the vector's length. In other words, the two norms are inversely proportional to each other.

3. Can the p-norm and the p'-norm have different values for the same vector?

Yes, the p-norm and the p'-norm can have different values for the same vector. This is because the two norms measure different aspects of the vector. The p-norm measures the absolute size of the vector, while the p'-norm measures the relative size of the vector in relation to its length.

4. In what situations are the p-norm and the p'-norm equivalent?

The p-norm and the p'-norm are equivalent when the p-th power of the vector's length is equal to the sum of the absolute values of the vector's components. This is known as the unit ball condition, and it is satisfied when the vector space is finite-dimensional and has a finite basis.

5. What are the applications of the p-norm and the p'-norm in science?

The p-norm and the p'-norm have many applications in various fields of science, including physics, engineering, and computer science. They are used to measure the size of vectors in different contexts, such as in optimization problems, signal processing, and data analysis. The choice of which norm to use depends on the specific problem at hand and the desired properties of the solution.

Similar threads

Replies
1
Views
1K
Replies
2
Views
1K
Replies
43
Views
3K
Replies
12
Views
2K
Replies
2
Views
895
Back
Top