Orthogonal projection formula: Difference between revisions

From Groupprops
 
(7 intermediate revisions by the same user not shown)
Line 1: Line 1:
{{quotation|This page needs to be cleaned up to accommodate improved terminology and also to tackle the non-splitting field version of the formula.}}
==Statement==
 
===Splitting field case using bilinear product===
 
Let <math>G</math> be a [[finite group]] and <math>k</math> a [[field]] whose characteristic does not divide the order of <math>G</math>. Suppose <math>\varphi:G \to GL(V)</math> is a finite-dimensional [[linear representation]] of <math>G</math> over <math>k</math>. Suppose further that <math>k</math> is a [[splitting field]] for <math>G</math>.
 
By Maschke's lemma, <math>\varphi</math> must be completely reducible i.e. it is the direct sum of irreducible representations. Suppose the irreducible representations are <math>\varphi_1, \varphi_2, \ldots, \varphi_s</math> and their multiplicities are <math>a_1, a_2, \ldots, a_s</math> respectively. Then if <math>\chi_i</math> is the character of <math>\phi_i</math> and <math>\chi</math> of <math>\phi</math> we have:
 
<math>\langle \chi,\chi_i \rangle_G = a_i</math>
 
where  <math>\langle \ , \ \rangle_G</math> is the [[fact about::inner product of functions]]:
 
<math>\langle f_1,f_2 \rangle = \frac{1}{|G|} \sum_{g \in G} f_1(g)f_2(g^{-1})</math>
 
Also <math>\chi</math> is orthogonal to the character of any irreducible linear representation ''not'' among the <math>\varphi_i</math>s, i.e., <math>\langle \chi, \alpha \rangle = 0</math> for any such <math>\alpha</math>.
 
===Non-splitting field case===
 
Let <math>G</math> be a [[finite group]] and <math>k</math> a [[field]] whose characteristic does not divide the order of <math>G</math>. Suppose <math>\varphi:G \to GL(V)</math> is a finite-dimensional [[linear representation]] of <math>G</math> over <math>k</math>. Suppose further that <math>k</math> is a [[splitting field]] for <math>G</math>.
 
By Maschke's lemma, <math>\varphi</math> must be completely reducible i.e. it is the direct sum of irreducible representations. Suppose the irreducible representations are <math>\varphi_1, \varphi_2, \ldots, \varphi_s</math> and their multiplicities are <math>a_1, a_2, \ldots, a_s</math> respectively. Then if <math>\chi_i</math> is the character of <math>\phi_i</math> and <math>\chi</math> of <math>\phi</math> we have:
 
<math>\langle \chi,\chi_i \rangle_G = a_id</math>
 
where <math>d</math> is a positive integer, which is actually the sum of squares of degrees of irreducible constituents of <math>\varphi_i</math> when it is decomposed over a splitting field.
 
As before  <math>\langle \ , \ \rangle_G</math> is the [[inner product of functions]]:


==Statement==
<math>\langle f_1,f_2 \rangle = \frac{1}{|G|} \sum_{g \in G} f_1(g)f_2(g^{-1})</math>


Let <math>G</math> be a [[finite group]] and <math>k</math> a [[field]] whose characteristic does not divide the order of <math>G</math>. Suppose <math>\phi:G \to GL(V)</math> is a finite-dimensional [[linear representation]] of <math>G</math> over <math>k</math>. Suppose further that <math>k</math> is a [[sufficiently large field]] for <math>G</math>, viz <math>k</math> contains all the <math>m^{th}</math> roots of unity where <math>m</math> is the [[exponent of a group|exponent]] of <math>G</math>.
===Alternate version for characteristic zero using a Hermitian inner product===


By Maschke's lemma, <math>\phi</math> must be completely reducible i.e. it is the direct sum of irreducible representations. Suppose the irreducible representations are <math>\phi_1, \phi_2, \ldots, \phi_s</math> and their multiplicities are <math>a_1, a_2, \ldots, a_s</math> respectively. Then if <math>\chi_i</math> is the character of <math>\phi_i</math> and <math>\chi</math> of <math>\phi</math> we have:
If <math>k</math> is a subfield of the complex numbers closed under complex conjugation, we can, instead of using the inner product specified here, use the Hermitian inner product (see [[Inner product of functions#Hermitian inner product]]):


<math><\chi,\chi_i> = a_i</math>
<math>\langle f_1, f_2 \rangle_G = \frac{1}{|G|} f_1(g)\overline{f_2(g)}</math>


where
It so turns out that when restricted to characters of representations, the Hermitian inner product coincides with the inner product used in general. See [[Inner product of functions#Relation between the definitions]].


<math><f_1,f_2> = \frac{1}{|G|} \sum_{g \in G} f_1(g)f_2(g^{-1})</math>
===Key distinction between characteristic zero and prime characteristic===


Also <math>\chi</math> is orthogonal to the character of any irreducible linear representation ''not'' among the <math>\phi_i</math>s.
Although the statement above is valid in characteristic zero and in prime characteristics not dividing the order of the group, there is a key distinction:


Note that when <math>k</math> is not sufficiently large, it is no longer true that <math><\chi,\chi_i></math> is the multiplicity of <math>\phi_i</math> in <math>\phi</math>. Actually we get:
* In characteristic zero, the inner product value is an element living in a field of characteristic zero, and so the equality <math>\langle \chi,\chi_i \rangle_G = a_i</math> allows us to retrieve either side from the other.
* In characteristic <math>p</matH>, the inner product value is in a field of characteristic <math>p</math>, whereas the ''multiplicity'' <math>a_i</math> is an actual nonnegative integer. So the equality <math>\langle \chi,\chi_i \rangle_G = a_i</math> for splitting fields needs to be interpreted more carefully: given <math>a_i</math>, we know <math>\langle \chi, \chi_i \rangle_G</math> uniquely as <math>a_i \pmod p</math>. However, the value <math>\langle \chi,\chi_i \rangle_G = a_i</math> only tells us <math>a_i \pmod p</math> and does ''not'' allow us to disambiguate between the various possibilities for <math>a_i</math>. The same caveat applies to non-splitting fields.


<math><\chi,\chi_i> = a_id</math>
==Facts used==


where <math>d</math> is a positive integer, which is actually the number of irreducible constituents of <math>\phi_i</math> in a sufficiently large field containing <math>k</math>.
# [[uses::Maschke's averaging lemma]]: This just guarantees complete reducibility.
# [[uses::Character orthogonality theorem]]


It still remains true, though, that the inner product is nonzero if and only if <math>\phi_i</math> is a part of <math>\phi</math>.
==Applications==
==Results used==


* [[Maschke's averaging lemma]]
* [[Character determines representation in characteristic zero]]
* [[First orthogonality theorem]]


==Proof==
==Proof==


The orthogonal projection formula follows from the [[first orthogonality theorem]], which states that the characters of irreducible linear representations form an orthonormal set. The idea is that if:
===Splitting field case===
 
<math>\phi = \sum_i a_i\phi_i</math>


Then by linearity of trace, we get:
'''Given''': <math>\varphi</math> is a representation of a finite group <math>G</math> over a splitting field <math>k</math>, and it reduces completely as a sum of <math>a_1</math> copies of <math>\varphi_1</math>, <math>a_2</math> copies of <matH>\varphi_2</math>, and so on till <math>a_s</math> copies of <math>\varphi_s</math>, All the <math>\varphi_i</math> are distinct irreducible representations.


<math>\chi = \sum_i a_i \chi_i</math>
<math>\chi_i</math> is the character of <math>\varphi_i</math>.


Taking the inner product with any particular <math>\chi_i</math>, all the terms other than that <math>\chi_i</math> give zero, and <math>\chi_i</math> gives <math>a_i</math> (this is where we use the first orthogonality theorem).
'''To prove''': <math>\langle \chi, \chi_i \rangle_G = a_i</math>. Further, if <math>\psi</math> is an irreducible representation not among the <math>\varphi_i</math>s, and <math>\alpha</math> is its character, then <math>\langle \chi, \alpha \rangle_G = 0</math>.


Further, if we take the character of any representation other than the <math>\phi_i</math>s, its inner product with all the <math>\chi_i</math>s is zero, and hence its inner product with <math>\chi</math> is zero.
{| class="sortable" border="1"
! Step no. !! Assertion/construction !! Facts used !! Given data used !! Previous steps used !! Explanation
|-
| 1 || <math>\chi = \sum_{i=1}^s a_i\chi_i</math> || Trace is linear || <math>\varphi</matH> is a sum of <math>\varphi_i</math>s ||  ||
|-
| 2 || <math>\langle \chi_i, \chi_i \rangle_G = 1</math> and <math>\langle \chi_i,\chi_j \rangle_G = 0</math> for <math>i \ne j</math> || Fact (2) || <math>k</math> is a splitting field, <math>\varphi_i</math> are irreducible. || ||
|-
| 3 || We get <math>\langle \chi, \chi_i \rangle = a_i</math> by additivity of the inner product || || || Step (2) || Combine Step (2) and additivity of the inner product.
|-
| 4 || <math>\langle \chi_i, \alpha \rangle = 0</math> for all <math>i</math>. ||  Fact (2) || <math>k</math> is a splitting field, <math>\varphi_i</math> are irreducible and so is <math>\psi</math>. || ||
|-
| 5 || <math>\langle \chi, \alpha \rangle = 0</math> || || || Step (4) || Combine Step (4) and additivity of the inner product.
|}


==Consequences==
==Consequences==

Latest revision as of 15:06, 13 July 2011

Statement

Splitting field case using bilinear product

Let be a finite group and a field whose characteristic does not divide the order of . Suppose is a finite-dimensional linear representation of over . Suppose further that is a splitting field for .

By Maschke's lemma, must be completely reducible i.e. it is the direct sum of irreducible representations. Suppose the irreducible representations are and their multiplicities are respectively. Then if is the character of and of we have:

where is the Inner product of functions (?):

Also is orthogonal to the character of any irreducible linear representation not among the s, i.e., for any such .

Non-splitting field case

Let be a finite group and a field whose characteristic does not divide the order of . Suppose is a finite-dimensional linear representation of over . Suppose further that is a splitting field for .

By Maschke's lemma, must be completely reducible i.e. it is the direct sum of irreducible representations. Suppose the irreducible representations are and their multiplicities are respectively. Then if is the character of and of we have:

where is a positive integer, which is actually the sum of squares of degrees of irreducible constituents of when it is decomposed over a splitting field.

As before is the inner product of functions:

Alternate version for characteristic zero using a Hermitian inner product

If is a subfield of the complex numbers closed under complex conjugation, we can, instead of using the inner product specified here, use the Hermitian inner product (see Inner product of functions#Hermitian inner product):

It so turns out that when restricted to characters of representations, the Hermitian inner product coincides with the inner product used in general. See Inner product of functions#Relation between the definitions.

Key distinction between characteristic zero and prime characteristic

Although the statement above is valid in characteristic zero and in prime characteristics not dividing the order of the group, there is a key distinction:

  • In characteristic zero, the inner product value is an element living in a field of characteristic zero, and so the equality allows us to retrieve either side from the other.
  • In characteristic , the inner product value is in a field of characteristic , whereas the multiplicity is an actual nonnegative integer. So the equality for splitting fields needs to be interpreted more carefully: given , we know uniquely as . However, the value only tells us and does not allow us to disambiguate between the various possibilities for . The same caveat applies to non-splitting fields.

Facts used

  1. Maschke's averaging lemma: This just guarantees complete reducibility.
  2. Character orthogonality theorem

Applications

Proof

Splitting field case

Given: is a representation of a finite group over a splitting field , and it reduces completely as a sum of copies of , copies of , and so on till copies of , All the are distinct irreducible representations.

is the character of .

To prove: . Further, if is an irreducible representation not among the s, and is its character, then .

Step no. Assertion/construction Facts used Given data used Previous steps used Explanation
1 Trace is linear is a sum of s
2 and for Fact (2) is a splitting field, are irreducible.
3 We get by additivity of the inner product Step (2) Combine Step (2) and additivity of the inner product.
4 for all . Fact (2) is a splitting field, are irreducible and so is .
5 Step (4) Combine Step (4) and additivity of the inner product.

Consequences

Uniqueness of decomposition as a sum of irreducible representations

The orthogonal projection formula tells us that given a representation, we can determine the multiplicities of irreducible representations in it. Thus, a representation cannot be expressed as a sum of irreducible representations in more than one way.

Character determines the representation

In characteristic zero, a representation is determined upto equivalence, by its character. This is essentially because the character determines the multiplicities of the irreducible constituents, which in turn determines the representation uniquely. We can rephrase this as: any field of characteristic not dividing the order of a finite group, is a character-determining field for the group. That is, every representation over the field is uniquely determined by its character.

Further information: Character determines representation in characteristic zero

Regular representation as a sum of irreducible representations

The orthogonal projection formula can be used to show that the regular representation is:

where are the characters of irreducible linear representations, and is the degree of .