Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Let $G$ be a compact, connected Lie group and $\rho$ any of its irreducible, unitary representations. If $\gamma:S^1\to G$ is an injective homomorphism (a periodic geodesic passing through the identity), what can be said about the matrix valued integral $$ I(\rho,\gamma)=\int_{S^1}\rho(\gamma(t))dt? $$ More explicitly, I am interested in the following questions:

  • Given $\rho$, can we find a $\gamma$ so that $I(\rho,\gamma)$ is invertible?
  • Given $\rho$, can we find geodesics $\gamma_i$ and complex numbers $a_i$ so that $\sum_ia_iI(\rho,\gamma_i)$ is invertible?
  • If a square matrix $A$ (of same dimension as $\rho$) satisfies $AI(\rho,\gamma)=0$, what can be said about $A$? This condition holds if $A=B\left.\frac{d}{dt}\rho(\gamma(t))\right|_{t=0}$ for some constant matrix $B$. Are all matrices $A$ satisfying this condition of this form?

The case of the torus $(\mathbb R/\mathbb Z)^n$ is quite simple. Then $\rho$ is parametrized by $\mathbb Z^n$ and $\gamma$ by $\mathbb Z^n\setminus\{0\}$, and $I(\rho_m,\gamma_k)=1$ if $m\cdot k=0$ and $I(\rho_m,\gamma_k)=0$ otherwise. (I have normalized the measure of $S^1$ to $1$.) Also, the irreducibles have dimension one, so we do not have a true matrix problem. Is there something as neat for nonabelian groups?

Although I am looking for a general answer, any examples for any groups are welcome.

share|improve this question
If the representation is faithful, then there is no such geodesic. This follows from Figuera-O'Farrel's answer below, since the integral is zero unless the element $\rho (\gamma) $ is identity. –  Venkataramana Jul 20 '14 at 17:03
@Venkataramana True, but the second question can have an affirmative answer even for a faithful representation. If $G=SU(3)$ and $\rho$ is the tautological representation, the geodesic $\gamma(w)=\text{diag}(w,\bar w,1)$ gives $I(\rho,\gamma)=\text{diag}(0,0,1)$. Summing three matrices like this gives an invertible one. I don't know how general this phenomenon is. –  Joonas Ilmavirta Jul 20 '14 at 17:32
@JoonasIlmavirta I am sorry if my question is trivial. Why the image of $\gamma$ is geodesic? –  Ali Taghavi Oct 8 '14 at 19:22
@JoonasIlmavirta In particular, is every torus subgroup, totally geodesic? –  Ali Taghavi Oct 8 '14 at 19:38
@AliTaghavi, if you take any bi-invariant Riemannian metric on $G$, then the periodic geodesics passing through the identity are precisely the nontrivial homomorphisms $S^1\to G$. Every Lie subgroup is totally geodesic because of this algebraic structure. –  Joonas Ilmavirta Oct 9 '14 at 7:02

2 Answers 2

up vote 3 down vote accepted

I had made a silly error in the first version of this answer.

To try to answer your first question, notice that the image of $\gamma$ is a circle subgroup $T$, say, of $G$ and that the integral $I(\rho,\gamma)$ is (with your normalisation of the measure on the circle) the projector onto the subspace $V^T$ of $T$-invariants in the restriction to $T$ of the representation $\rho: G \to GL(V)$. Hence $I(\rho,\gamma)$ is invertible if and only if $V^T=V$. (Previously I had written $V^T=0$ and hence reached the wrong conclusion.) As Venkataramana points out in the comment, this cannot happen if $\rho$ is a faithful representation, hence for simple $G$.

You could take $G = G_1 \times G_2$ and consider $\rho$ such that the $G_2$ subgroup acts trivially and have $\gamma(S^1)$ be a circle subgroup of the form $(e,\gamma_2(t)) \in G_1 \times G_2$, with $e \in G_1$ the identity element.

share|improve this answer
Do you mean $V^T=V$? –  Venkataramana Jul 20 '14 at 17:02
Oops, yes. I'll edit. –  José Figueroa-O'Farrill Jul 20 '14 at 17:08

José Figueroa-O'Farrill already answered the first question. Let me answer the two other ones:

2) If $G$ is not $S^1$ or $S^3$, the answer is yes. If this did not hold for some $\rho$, there would be a nonzero vector $a\in\mathbb C^{\dim\rho}$ that is in the kernel of $I(\rho,\gamma)$ for all $\gamma$. Let $A$ be the square matrix whose every row is $a^*$. Then $AI(\rho,\gamma)=0$ for all $\gamma$.

Now let $f:G\to\mathbb C$ be the unique smooth function that satisfies $\int_Gf(x)\rho(x)dx=A$ and $\int_Gf(x)\sigma(x)dx=0$ for all irreducible representations $\sigma$ that are not equivalent to $\rho$. The existence and uniqueness of $f$ follows from the Peter-Weyl theorem.

For any $x\in G$ and any nontrivial homomorphism $\gamma:S^1\to G$ let $$ F(x,\gamma)=\int_{S^1}f(x\gamma(t))dt $$ be the integral of $f$ over the geodesic $t\mapsto x\gamma(t)$. For any (unitary, irreducible) representation $\sigma$, a calculation gives $$ \int_G F(x,\gamma)\sigma(x)dx = \int_G f(x)\sigma(x)dx I(\sigma,\gamma). $$ By the definition of $f$, this is always zero (for both $\sigma=\rho$ and $\sigma\neq\rho$). Thus $F$ is orthogonal to all matrix elements of all irreducible representations, so $F=0$.

This means that $f$ is not identically zero but its integral over every geodesic is zero. This is impossible by theorem 1.1 of this paper.

If $G=S^1$, the answer is no for all nonconstant $\rho$. If $G=S^3$, I don't know the exact answer. On this group a function is known to integrate to zero over all geodesics if and only if it is antipodally antisymmetric ($f(-x)=-f(x)$ if we embed $S^3\subset\mathbb R^4$). This with the previous argument might tell for which $\rho$s the answer is yes; it is not yes for all of them.

3) The answer to the third question is positive.

Let $D(\rho,\gamma)=\left.\frac{d}{dt}\rho(\gamma(t))\right|_{t=0}$. Then $\frac{d}{dt}\rho(\gamma(t))=\rho(\gamma(t))D(\rho,\gamma)$. Let $1$ denote the identity matrix, since $I$ is reserved for the integral. If $A$ satisfies $AI(\rho,\gamma)=0$, then \begin{eqnarray} A &=& A(1-I(\rho,\gamma)) \\&=& A\left(1-\int_{S^1}\rho(\gamma(t))dt\right) \\&=& A\int_{S^1}(\rho(\gamma(0))-\rho(\gamma(t)))dt \\&=& A\int_{S^1}\int_t^0\frac{d}{ds}\rho(\gamma(s))dsdt \\&=& A\int_{S^1}\int_t^0\rho(\gamma(s))dsdtD(\rho,\gamma). \end{eqnarray} If we let $$ B=A\int_{S^1}\int_t^0\rho(\gamma(s))dsdt, $$ then we have $A=BD(\rho,\gamma)$. Moreover, the matrix $B$ can be explicitly expressed in terms of $A$.

Since $I(\rho,\gamma)$ is hermitean and $D(\rho,\gamma)$ skew-hermitean, the result states that the image of the derivative of a representation is exactly the kernel of the integral of the representation and vice versa. (Here, of course, the derivative and the integral mean $D$ and $I$.)

share|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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