Let $G$ be a locally compact group with unimodular Haar measure $\mu$. We consider the Hilbert space $\mathscr{H}:= L_{\mu}^2(G)$ together with the unitary representation $\pi : G \to U(\mathscr{H})$ given by $\pi(g): \mathscr{H} \to \mathscr{H}, f \mapsto f \circ r_g$, where $r_g$ is right multiplication by $g$. Hence $(\pi(g)f)(x) = f(xg)$.

We studies the following: For a continuous function $\varphi \in C_c(G)$ with compact support we define $$\pi(\varphi):= \int_G \varphi(g)\pi(g) \, d\mu(g).$$ More explicitly, for a measurable function $f: G \to \mathbb{C}$ we define $$(\pi(\varphi)f)(x) :=\int_G \varphi(g)f(xg) \, d\mu(g).$$

My question in now: Does this define a well defined representation of $C_c(G)$ on $\mathscr{H}$, where the problem is, that if $f \in L_{\mu}^2(G)$, then we need to show that $\pi(\varphi)f$ is in $L_{\mu}^2(G)$.

I doubt that this is true for a general $G$. However, I can prove the statement if $G$ is compact as follows: Note that if $G$ is compact, then $L_{\mu}^2(G) \subset L_{\mu}^1(G)$. Hence for $f \in L_{\mu}^2(G)$, \begin{align*} ||\pi(\varphi)f||_2^2 &= \int_G|(\pi(\varphi)f)(x)|^2 \, d\mu(x) \\ &= \int_G \bigg| \int_G \varphi(g)f(xg) \, d\mu (g) \bigg|^2 d\mu(x) \\ & \leq ||\varphi||_{\infty}^2 \int_G \bigg| \int_G f(xg) \, d\mu (g) \bigg|^2 d\mu(x) \\ &\leq ||\varphi||_{\infty}^2 \int_G ||f||_1^2 \,d \mu(x) \\ &= ||\varphi||_{\infty}^2 \cdot ||f||_1^2 \cdot \mu(G) < \infty. \end{align*} Where we used the invariance of the Haar measure in line 4. And the last term is less than infinity, since $\varphi$ is continuous of compact support and hence bounded, $L_{\mu}^2(G) \subset L_{\mu}^1(G)$ and hence for $f \in L_{\mu}^2(G)$ we have $||f||_1 < \infty$ and finally since $G$ is compact, the Haar measure is finite.

So the problem is can, one do a similar calculation if $G$ is not compact, or if one can find a counterexample.


It's true: $\pi(\varphi)f\in L^2(G,\mu)$.

Let $M$ be the support of $\varphi$, so $$(\pi(\varphi)f)(x) :=\int_M \varphi(g)f(xg) \, d\mu(g).$$ By the Cauchy-Schwarz inequality, we have $$|(\pi(\varphi)f)(x)|^2\le \|\varphi\|^2_2\int_M|f(xg)|^2d\mu(g)=\|\varphi\|^2_2\int_{xM}|f(g)|^2d\mu(g),$$ so $$\int_G|(\pi(\varphi)f)(x)|^2d\mu(x)\le \|\varphi\|^2_2\int_G\int_{G}\mathbb{1}_{xM}(g)|f(g)|^2d\mu(g)d\mu(x).$$ Since $g\in xM$ $\Leftrightarrow$ $x\in gM^{-1}$, this yields (I should assume $G$ $\sigma$-compact to apply Fubini, but otherwise since $f$ has $\sigma$-compact support I apply it inside a suitable $\sigma$-compact open subgroup) $$\int_G|(\pi(\varphi)f)(x)|^2d\mu(x)\le \|\varphi\|^2_2\int_G\int_{gM^{-1}}d\mu(x)|f(g)|^2d\mu(g).$$ So $$\int_G|(\pi(\varphi)f)(x)|^2d\mu(x)\le \|\varphi\|^2_2\mu(M^{-1})\int_G|f(g)|^2d\mu(g)=\|\varphi\|^2_2\mu(M^{-1})\|f\|_2^2<\infty.$$

Note that I only used (twice) left-invariance of $\mu$ (in the non-unimodular case, the right-regular representation $\pi$ is not unitary, however).

| cite | improve this answer | |
  • $\begingroup$ Thank you very much. To simplify a little: $$\begin{align*} ||\pi(\varphi)f||_2^2 &= \int_G |(\pi(\varphi)f)(x)|^2 \, d\mu(x) \\ &= \int_G \bigg| \int_{\mathrm{supp}(\varphi)} \varphi(g)f(xg) \, d\mu(g) \bigg|^2 \, d\mu(x) \\ & \leq ||\varphi||_2^2 \int_G\int_{\mathrm{supp}(\varphi)} |f(xg)|^2 \, d\mu(g)d\mu(x) \\ &= ||\varphi||_2^2 \int_{\mathrm{supp}(\varphi)} \int_G |f(xg)|^2 \, d\mu(x)d\mu(g) \\ &= ||\varphi||_2^2 \int_{\mathrm{supp}(\varphi)} \int_G |f(x)|^2 \, d\mu(x)d\mu(g) \\ &= \mu(\mathrm{supp}(\varphi)) \cdot ||\varphi||_2^2 \cdot ||f||_2^2 < \infty. \end{align*} $$ $\endgroup$ – Constantin K Mar 31 '18 at 10:09
  • $\begingroup$ Yes. Actually you're using right-invariance of $\mu$ and not left as I did. In this case the right-regular representation is unitary. Said otherwise: for $G$ possibly non-unimodular and $\mu$ left-invariant, my argument works for the right-regular representation, and your argument works for the left-regular representation (these two representations are of distinct interest when $G$ is non-unimodular; only the left one is unitary). $\endgroup$ – YCor Mar 31 '18 at 10:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.