3

Can a group have two faithful irreducible unitary representations with distinct degrees? For example, if you were to find a faithful irreducible unitary representation of a group $G$ of dimension $\aleph_0,$ could you immediately deduce that any other faithful irreducible unitary representation must also have this dimension?

Miles Gould
  • 1,284
  • 4
    Every nontrivial representation of a simple group is faithful, so you can look at e.g. the character table of $A_5$; it has faithful irreducible unitary representations of degrees $3, 4, 5$. – Qiaochu Yuan Oct 16 '24 at 17:55
  • @QiaochuYuan Thank you! Can you think of any with infinite dimension? I would be shocked to find that there can be faithful irred. unitary rep's of different infinite cardinal dimensions. – Miles Gould Oct 16 '24 at 17:58
  • 1
    That seems more plausible to me but I'm not very familiar with infinite-dimensional unitary representations. – Qiaochu Yuan Oct 16 '24 at 21:11
  • 2
    This is a partial answer, as I don't know in general. Assume that the field $k$ is "small", i.e., $|k|\leq |G|$. If $V$ is a vector space of dimension greater than $|k|$ then $|V|=\dim(V)$. Also, $|V|$ is the same as the cardinality of linear transformations from $V$ to $V$, so if $G$ embeds in $End(V)$ then $|V|\geq |G|$. On the other hand, the $G$-orbit of a vector in $V$ has cardinality at most $|V|$, so any irreducible representation has cardinality at most $|G|$. – David A. Craven Oct 16 '24 at 21:49
  • 1
    @DavidA.Craven: do you mean, in your last sentence, "the $G$-orbit of a vector in $V$ has cardinality at most $|G|$, so any irreducible representation has dimension (and hence cardinality) at most $|G|$"? – krm2233 Oct 17 '24 at 22:01
  • @krm2233 Yes, I skipped a step. It was only a comment, so I thought that bit was fine, since the same trick (which definitely requires $k$ to not be too big) was used earlier. – David A. Craven Oct 17 '24 at 22:07
  • @DavidA.Craven It wasn't the skipped step I meant to draw attention to -- I agree with you that that should be fine to elide given the trick was already in play. I just meant to flag the stupid point that your original comment has a typo in that it says the G-orbit has cardinality at most $|V|$ rather than $|G|$. – krm2233 Oct 17 '24 at 22:12
  • 1
    @krm2233 Ah, right. Yes, you are correct, I didn't notice that, even after you mentioned it! Thanks. – David A. Craven Oct 17 '24 at 23:08

1 Answers1

3

Qiaochu already provided a counterexample for finite-dimensional cases in the comments. Here is an infinite-dimensional counterexample, constructed using $C^\ast$-algebra and von Neumann algebra machinery. There should be much more simpler constructions, but this is the only one I came up with.

Fix a separable II$_1$ factor $M$. Let $U(M)$ be its unitary group. Then its center is $U(1)$, the group of scalar operators. Let,

$$G = (U(M) \times U(M))/\{(z, z): z \in U(1)\}$$

Now, consider the following two representations of the group:

  1. Let $H = L^2(M)$, the standard representation of $M$, which comes equipped with both left and right actions of $M$. Let $\pi: G \to U(H)$ be given by,

$$\pi([(g, h)])\xi = g\xi h^{-1}, \forall \xi \in H$$

Note that, for any $z \in U(1)$, $z\xi z^{-1} = \xi$, so this is indeed a well-defined homomorphism. Furthermore, we note that it is faithful. Indeed, if $g\xi h^{-1} = \xi$ for all $\xi \in H$, then in particular, $ghh^{-1} = h \Rightarrow g = h$. But then, when $g = h$, $g\xi g^{-1} = \xi$ for all $\xi \in H$ implies $g \in Z(M)$, so $g \in U(1)$, i.e., $(g, h) \in \{(z, z): z \in U(1)\}$.

We also note that it is irreducible. Indeed, $\pi([(g, 1)])$ gives all unitaries in the left action of $M$ while $\pi([(1, h)])$ gives all unitaries in the right action of $M$. The left and right actions together generate the entirety of $\mathbb{B}(L^2(M))$, so the representation is irreducible. Since $M$ is separable, $H = L^2(M)$ is a separable Hilbert space. Thus, $G$ has a faithful irreducible representation of dimension $\aleph_0$.

  1. By Krein-Milman, $M$ has a pure state and thus an irreducible representation $\phi: M \to \mathbb{B}(H)$ for some Hilbert space $H$. It is known that, if $H$ is separable, then $\phi$ must be normal. (See Theorem V.5.1 in Takesaki’s Theory of Operator Algebras I.) But if $\phi$ is normal, then as $M$ is of type II$_1$, $\phi(M)$ and thus $\phi(M)’$ must be of type II. However, the representation is irreducible, so $\phi(M)’ = \mathbb{C}$, which is a contradiction, so $H$ must be non-separable. Note also that, as II$_1$ factors are simple as $C^\ast$-algebras, $\phi$ must be faithful.

Now, let $\pi: G \to U(H \otimes \overline{H})$ be defined by,

$$\pi([(g, h)]) = \phi(g) \otimes \overline{\phi(h)}$$

Since $z \otimes \overline{z} = 1$ for any scalar $z \in U(1)$, we again have $\pi$ is a well-defined homomorphism. $\pi$ is also faithful, as $\phi(g) \otimes \overline{\phi(h)} = 1$ implies both $\phi(g)$ and $\phi(h)$ are scalars, in which case it is easy to check that $\phi(g) \in U(1)$ and $\phi(h) = \phi(g)$. But $\phi$ is faithful, so,

$$(g, h) = (\phi(g), \phi(g)) \in \{(z, z): z \in U(1)\}$$

This verifies that $\pi$ is faithful. Finally, it is irreducible as $\pi([(g, 1)])$ generates $\phi(M) \otimes 1$ and thus generates $\mathbb{B}(H) \otimes 1$, as $\phi$ is irreducible. Similarly, $\pi([(1, h)])$ generates $1 \otimes \mathbb{B}(\overline{H})$, so $\pi(G)$ generates the entirety of $\mathbb{B}(H \otimes \overline{H})$, i.e., $\pi$ is irreducible. Hence, $G$ also has a faithful irreducible representation on a non-separable Hilbert space.

David Gao
  • 22,850
  • 9
  • 28
  • 48
  • The reason I asked this question was because I was interested in the representations of $S_\mathbb{N}$ (the automorphisms of the set $\mathbb{N}$) with a particular topology. I suppose I got lucky since I was able to show that my conjecture held in this case, i.e. every faithful irreducible has degree $\aleph_0.$ – Miles Gould Oct 22 '24 at 23:50
  • @MilesGould You meant that all faithful irreducible representations of $S_\mathbb{N}$ continuous under your particular topology are of degree $\aleph_0$? I suppose that’s not that surprising. In general, if you have an uncountable group, it’s always a good idea to consider some topology on it. Even in the example in my answer, if you equip $G$ with the SOT topology and only consider strongly continuous representations, you also get the only faithful irreps have dimension $\aleph_0$. It’s only when topology is ignored that these bizarre counterexamples begin to show up. – David Gao Oct 23 '24 at 05:35