5

The question is quite long since I give some background but really I am interested in some very concrete fact about matrix representations. Scroll all the way to the bottom for a self contained question, feel free to ignore the rest.

Let $L/K$ be a Galois field extension with Galois group $\Gamma$. Let $G$ be a finite group and $V$ a representation of $G$ over $L$ (aka, a module over $L[G])$.

Suppose there is a semi linear action of $\Gamma$ on $V$ with respect to $L(G)$. That is, for all $a \in L[G]$,$\gamma \in \Gamma$ and $v \in V$, we have $(av)^\gamma = a^\gamma v^\gamma$.

Does there necessarily exist a $k[G]$ submodule $W \subset L$ such that the natural map $W\otimes_{k[G]} L[G] \to V$ is an isomorphism?

I believe I can prove this using general ideas from faithfully flat descent (extend to a separably closed base field...) but I am looking for a more concrete way of proving this along the following lines:

Let us decompose $V = \bigoplus_k V_k$ as a sum of irreducible representations. Now, note that we can define conjugate representations $g^\gamma$ on $V$ since $g^\gamma (v) = (g(v^{\gamma^{-1}}))^\gamma \in V.$ Moreover, this conjugate action preserves irreducibility and so the representation $(G^\gamma,V_k)$ corresponds to some $(G,V_{\gamma(k)})$.

In concrete terms, if we fix a basis of vectors fixed under the semilinear action of $\Gamma$ and $g$ is given by a matrix $[a_{ij}]$ on $V_k$, then $g^\gamma$ is given by the matrix $[\gamma(a_{ij})]$.

Therefore, we can group the conjugate representations together and now suppose $W_k = \bigoplus_{\gamma \in \Gamma}V_k^\gamma$. (In general, there might be repeated terms in this sum but let me ignore that for now). This is closed under both the $G$ action and the $\Gamma$ action. Concretely, for any $g$ with matrix $[a_{ij}]$ for $W_k$, the matrix for $W$ is given by a block matrix with $|\Gamma|$ blocks and the block corresponding to $\gamma$ is equal to $[a_{ij}^\gamma]$.

In this case, I want to show that the matrix for each $g$ is similar to a matrix over $k$ and moreover that this similarity is realized by the same matrix (over $L$) for all $g \in G$.

The second condition is what is tripping me up. Since the characteristic polynomial for any $g$ is a polynomial over $k$, the rational canonical form realizes an element in the conjugacy class over the base field. But how do I make sure that I can do just one base change to get all the matrices corresponding to all the $g \in G$ into rational form?

Concrete question:

Let $L/K$ be a Galois extension with Galois group $\Gamma$. Let $G$ be a finite group. Let $\rho: G \in M_n(L)$ be a representation and suppose that for each $g, \rho(g)$ is a block matrix with the blocks indexed by $\gamma \in \Gamma$ such that if the block corresponding to the identity in $\Gamma$ is $[a_{ij}]$, then the block corresponding to $\gamma$ is $[\gamma(a_{ij})]$.

Can I find an invertible matrix $P$ over $L$ such that for all $g \in G$, $P\rho(g)P^{-1}$ is a matrix over $K$? Note that $\rho(g)$ always has characteristic polynomial defined over $K$ so that the rational canonical form gives us some $P_g$ for each $g \in G$ with the above property.

The question is eliminate the dependence of $P_g$ on $g$.

vidyarthi
  • 7,315
Asvin
  • 8,489
  • Your notation confuses me. Could it be that the line "Does there necessarily exist a $k[G]$ submodule $W \subset L$ such that the natural map $W\otimes_{k[G]} L[G] \to W$ is an isomorphism?" should actually read "Does there necessarily exist a $\color{red}{K}[G]$ submodule $W \subset \color{red}{V}$ such that the natural map $W\otimes_{k[G]} L[G] \to \color{red}{V}$ is an isomorphism?" Then later, I would assume you want to decompose $V$ not $W$, but I stopped reading there for now. – Torsten Schoeneberg Jul 19 '19 at 21:18
  • Reading the question after a year, I am a little confused too. I think you must be right. I am using $k$ and $K$ to denote the same field and I want to decompose $V$, not $W$. I think I was basically asking for descent for modules over the Galois extension $L[G]/K[G]$ that is more explicit/elementary than fpqc descent by working with matrices. – Asvin Jul 20 '19 at 04:11
  • I hope the question makes sense now. – Asvin Jul 20 '19 at 04:18
  • First, I have a feeling that ultimately this will be answered by Hilbert 90, which is the basic case of descent anyway, cf. https://mathoverflow.net/q/21110/27465. But to get there, I'm still troubled by technical issues. So are the two $\color{red}{\text{red } V}$'s in my first comment to be inserted (you did not address them in your edit)? Further, the $\Gamma$-action on $L[G]$ is the natural on $L$ and trivial on $G$? – Torsten Schoeneberg Jul 21 '19 at 01:22
  • I think yes and yes. I fixed the V (hopefully). – Asvin Jul 21 '19 at 01:40
  • Hmm. If the $\Gamma$-action on $G \subset L[G]$ is trivial, that means that the actions of $\Gamma$ and $G$ on $V$ commute. But then you can just find a $\Gamma$-fixed basis of $V$ (with Hilbert 90 or without: https://kconrad.math.uconn.edu/blurbs/galoistheory/galoisdescent.pdf), and then w.r.t. basis, the $[a_{ij}]$ for all $g$ will have coefficients in $K$. – Torsten Schoeneberg Jul 22 '19 at 17:19
  • I think I came to the same conclusion yesterday after thinking for a while! The G action is a red herring! – Asvin Jul 22 '19 at 17:20
  • What is a non-trivial example for this situation to begin with? Note that for most actual $\Gamma$-actions from representation theory (where $\Gamma$ and $G$ don't have commuting actions), the statement is just wrong. Example $G= \mathbb Z/3$, $L\vert K = \mathbb C\vert \mathbb R$, $V= \mathbb C$ with a generator of $G$ acting as $\zeta_3$. Does not have a $K[G]$-form. – Torsten Schoeneberg Jul 22 '19 at 17:23
  • I don't remember why I was thinking about this anymore. I might have just been trying to understand fpqc descent in specific cases explicitly. Thanks for thinking about the question! – Asvin Jul 22 '19 at 17:25

0 Answers0