Left-Invariant Vector Fields on a Coset Space

211 Views Asked by At

This is intended as a general question regarding whether or not a notion of left-invariant vector fields makes sense on a coset space $G/H$ (where, of course, $G$ is a Lie group and $H$ is some subgroup), and how one can determine them; however, for the sake of clarity (hopefully...), I'll illustrate what I'm getting at with an example.

$\fbox{$\textbf{Example:}$}$

$\quad$ Consider the Lie group $W:=\{w(p,x,\theta) \ | \ p,x,\theta\in\mathbb{R}\}$ with the group operation given by

$$ w(p',x',\theta')w(p,x,\theta)=w(p'+p,x'+x,\theta'+\theta+p'x-x'p), $$

called the Heisenberg-Weyl group. This can also be thought of as the unique Lie group obtained by taking the Heisenberg algebra defined by the single commutator (all others being zero) $[X,P]=2iI$ (Note: The factor of 2 here is to compensate for some factors of $1/2$ I missed below -- it's easier to fix it here. Besides, some authors discussing this algebra prefer to use the "$\hbar=2$" units anyways, so... ;) ), and exponentiating it to get

$$ w(p,x,\theta)=\exp\big[i\big(pX-xP+\theta I\big)\big]. $$

One can check that, by applying the Baker-Campbell-Hausdorff formula to a product of the above exponentials will reproduce the group operation defined above.

$\quad$ To find, for example, a representation $D(X)$ of the Lie algebra generator $X$ as a (left-invariant) vector field on $W$, one can consider a function $f\in C^{\infty}(W)$ and compute (setting $f(w(p,x,\theta))\equiv f(p,x,\theta)$, for simplicity)

\begin{align*} -i\frac{d}{dt}\bigg|_{t=0}f(w(-t,0,0)w(p,x,\theta)) &= -i\frac{d}{dt}\bigg|_{t=0}f(p-t,x,\theta-tx) \\[1em] &= -i\bigg[\frac{d(p-t)}{dt}\partial_p+\frac{d(x)}{dt}\partial_x+\frac{d(\theta-tx)}{dt}\partial_\theta\bigg]f(p-t,x,\theta-tx)\bigg|_{t=0} \\[1em] &= i\bigg[\partial_p+x\partial_\theta\bigg]f(p,x,\theta); \end{align*}

hence, we obtain the vector field $D(X)=i(\partial_p+x\partial_\theta)$. Similarly, one has (the standard expressions) $D(P)=-i(\partial_x-p\partial_\theta)$ (keeping in mind that one uses $id/dt$ in this case, due to the $-$ in the exponential for the coefficient of $P$) and $D(I)=i\partial_\theta$.


$\quad$ Now, consider the coset space $\widetilde{W}:=W/H$, where $H=\{(0,0,\theta) \ | \ \theta\in\mathbb{R}\}$. In a similar vein to the above, we can write the elements of this coset space as

$$ \widetilde{w}(p,x)=\exp\big[i\big(pX-xP\big)\big]. $$

The idea here is that we are setting $\widetilde{w}(p,x)=e^{-i\theta}w(p,x,\theta)$.

$\quad$ Even though this clearly doesn't give one a group structure, due to an extra "phase factor" (rewriting things in terms of the elements of $W$, once again using BCH, and collecting the appropriate factors in the $\theta$-direction to go back to $\widetilde{W}$):

$$ \widetilde{w}(p',x')\widetilde{w}(p,x)=\widetilde{w}(p'+p,x'+x)\exp\big[i\big(x'p-p'x\big)\big], $$

for the purposes of physics, this isn't an issue -- quantum states are only unique up to such a phase factor -- and this simply says that a representation of $W$ induces a projective representation on $\widetilde{W}$. Not only that, but they also talk about representing the generators $X$ and $P$ as (left-invariant) vector fields on $\widetilde{W}$, and this finally leads to the question at hand. Trying to repeat the above process in a naive way, one finds:

\begin{align*} -i\frac{d}{dt}\bigg|_{t=0}f(\widetilde{w}(-t,0)\widetilde{w}(p,x)) &= -i\frac{d}{dt}\bigg|_{t=0}f(\widetilde{w}(p-t,x)e^{itx}) \\[1em] \text{"} &= \text{"} -i\frac{d}{dt}\bigg|_{t=0}\bigg[f(p-t,x)e^{itx}\bigg] \\[1em] &= -i\bigg[\frac{d(p-t)}{dt}\partial_p+\frac{d(x)}{dt}\partial_x\bigg]f(p-t,x)e^{itx}\bigg|_{t=0} \\[1em] &\hspace{1.1in} -i^2xf(p-t,x)e^{itx}\bigg|_{t=0} \\[1em] &= (i\partial_p+x)f(p,x,\theta); \end{align*}

thus we arrive at a representation $\widetilde{D}(X)$ of $X$ as a differential operator on $C^{\infty}(\widetilde{W})$ given by $\widetilde{D}(X)=i\partial_p+x$. Similarly, $\widetilde{D}(P)=-i\partial_x+p$ (still keeping in mind that one uses $id/dt$ for this term).

Question: The above successfully reproduces standard expressions in the literature for both the group $W$ and the coset space $W/H$, but is there actually any sense to be made of the "$=$" above? Is it as simple as saying that what we mean by $C^{\infty}(W/H)$ is the collection of functions $f$ satisfying $f(\widetilde{w}\exp(\text{whatever}))=f(\widetilde{w})\exp(\text{whatever})$?

Or, more generally, on an arbitrary coset space $G/H$ can one make sense of left-invariant vector fields in any way remotely resembling this? (Or, if not arbitrary coset spaces, in what cases can you do something along these lines?)

1

There are 1 best solutions below

0
On BEST ANSWER

After more thought, I'm fairly certain that I've already answered my own question. Instead of thinking of the operators as acting on something like $C^{\infty}(G)$, as I did in the first part of the example, for a coset space $G/H$ one works with

$$ C^{\infty}_e(G/H):=\{f\in C^{\infty}(G) \ | \ f(ge^{ia^jh_j})=f(g)e^{ia^jh_j}\}, $$

(where the $h_j$ are the generators for the Lie algebra of $H$ and the $a^j$ are parameters) and simply apply exactly the above procedure to get a suitable notion of left-invariant vector fields on $G/H$, which correctly reproduces the standard expressions given in the literature for the example considered, but can be applied more generally.