Invariant differential operators on G/K and the Harish-Chandra isomorphism

Intuition Beginner

On a symmetric space 07.04.07 there is a special collection of differential operators: the ones that do not care where you are or how you are turned. These are the operators that commute with every symmetry of the space. The Laplacian --- the operator measuring how much a function bulges away from its average --- is the most famous one, but there are usually more.

The symmetry-respecting differential operators form a commutative algebra, and that algebra is secretly just a ring of symmetric polynomials. Each invariant operator becomes a polynomial, and composing two operators corresponds to multiplying two polynomials. So a question about hard differential operators turns into a question about ordinary algebra.

Why does this happen? A symmetric space is built from a group, and a differential operator that respects the group is pinned down by a small amount of data living on a flat slice inside the space. On that flat slice an operator is just a polynomial in a few coordinate directions. The matching is the Harish-Chandra isomorphism, named for the mathematician who discovered it.

The payoff is enormous. Because the invariant operators all commute, they can be diagonalised at once. Their shared eigenfunctions are the spherical functions 07.04.12, the building blocks for breaking any reasonable function on the space into simple waves.

Visual Beginner

A symmetric space drawn as a curved bowl, with a single base point marked at the bottom and a straight ruler --- the flat slice --- laid across it through that point. Above the bowl floats a box of polynomials. An arrow labelled "Harish-Chandra" carries each invariant operator on the bowl up to a symmetric polynomial in the box.

The flat slice is the maximal flat subspace, of dimension equal to the rank. The box of polynomials only keeps those unchanged by a finite reflection group, the Weyl group 07.06.04. Rank one means one ruler direction, so the box is polynomials in a single variable.

Worked example Beginner

Take the hyperbolic upper half-plane, the symmetric space of rank one. We want to find the invariant operators and read off how the Laplacian acts.

Step 1. The flat slice has dimension one. So the polynomial box is just polynomials in one variable, and the symmetric-polynomial condition is no condition at all in this case. The invariant operators form the single-variable polynomial ring generated by the Laplacian: every invariant operator is a polynomial in the Laplace-Beltrami operator $\Delta$.

Step 2. In the coordinates $(x,y)$ with $y>0$ the invariant Laplacian is $$\Delta =y^2\left(\frac{d^2}{d{x}^2}+\frac{d^2}{d{y}^2}\right).$$ We test it on the simple functions $y^s$, which depend only on the height $y$.

Step 3. Compute. Applying $\Delta$ to $y^s$ kills the $x$-derivative and leaves $y^2\cdot s(s-1)y^{s-2}=s(s-1)y^s$. So $\Delta y^s=s(s-1)y^s$.

Step 4. Write the eigenvalue in the symmetric form $s(s-1)=-s(1-s)$. The number $s(1-s)$ is the value that the matching polynomial takes at the spectral parameter, and it is unchanged when $s$ is swapped with $1-s$. That swap is the Weyl-group reflection: the polynomial really is symmetric.

What this tells us: in rank one the whole algebra of invariant operators is generated by the Laplacian, and its eigenvalues on the height functions $y^s$ are the symmetric quantities $s(1-s)$.

Check your understanding Beginner

Formal definition Intermediate+

Let $M=G/K$ be a Riemannian symmetric space 07.04.07, where $G$ acts on $M$ by isometries and $K$ is the isotropy subgroup of the base point $o=eK$. Let $C^{\infty }(M)$ denote smooth functions.

Definition (Invariant differential operator). A differential operator $D$ on $M$ is $G$-invariant if it commutes with the action of $G$: writing $\tau (g)f(x)=f(g^{-1}x)$ for the translation by $g\in G$, $$D(\tau (g)f)=\tau (g)(Df)\text{for all }g\in G,f\in C^{\infty }(M).$$ The set of all $G$-invariant differential operators is denoted $\mathbb{D}(G/K)$. It is an associative algebra under composition, with the Laplace-Beltrami operator $\Delta$ of the $G$-invariant metric among its members.

Write $\mathfrak{g}=\mathfrak{k}\oplus \mathfrak{p}$ for the Cartan decomposition associated to the symmetry at $o$, so $T_{o}M\cong \mathfrak{p}$. Fix a maximal abelian subspace $\mathfrak{a}\subset \mathfrak{p}$; its dimension is the rank $r$ of $M$, and the restricted roots of $(\mathfrak{g},\mathfrak{a})$ live in ${\mathfrak{a}}^{\ast }$ 07.04.08. Let $W$ be the Weyl group 07.06.04 of this restricted root system, acting on $\mathfrak{a}$ and dually on ${\mathfrak{a}}^{\ast }$. Let $S({\mathfrak{a}}_{\mathbb{C}})$ be the symmetric algebra on the complexified Cartan subspace, identified with polynomial functions on ${\mathfrak{a}}_{\mathbb{C}}^{\ast }$, and $S({\mathfrak{a}}_{\mathbb{C}}{)}^W$ its subalgebra of Weyl-invariants.

Definition (Half-sum of restricted roots). With multiplicities $m_{\alpha }=\dim {\mathfrak{g}}_{\alpha }$, set $$\rho =\frac{1}{2}\sum _{\alpha \in {\Sigma }^{+}}{m}_{\alpha }\alpha \in {\mathfrak{a}}^{\ast },$$ the half-sum of positive restricted roots counted with multiplicity. The element $\rho$ controls the shift built into the isomorphism below.

Counterexamples to common slips

• Invariance is strictly stronger than $K$-invariance. An operator invariant only under the isotropy $K$ (a "radial" condition at $o$) need not commute with all of $G$. The algebra $\mathbb{D}(G/K)$ consists of the genuinely $G$-invariant operators, a smaller collection.

• Commutativity is a feature of symmetric pairs, not of homogeneous spaces in general. For a general reductive homogeneous space $G/K$ the invariant operators need not commute. The involution defining the symmetric structure is what forces commutativity, through the relation $[\mathfrak{p},\mathfrak{p}]\subset \mathfrak{k}$.

• The image polynomials are Weyl-invariant, not arbitrary. The Harish-Chandra map does not land on all of $S({\mathfrak{a}}_{\mathbb{C}})$. A polynomial outside $S({\mathfrak{a}}_{\mathbb{C}}{)}^W$ corresponds to no invariant operator, because the Weyl group permutes the directions of the flat slice.

Key theorem with proof Intermediate+

Theorem (Harish-Chandra isomorphism for $\mathbb{D}(G/K)$). Let $M=G/K$ be a Riemannian symmetric space of non-compact type with maximal abelian $\mathfrak{a}\subset \mathfrak{p}$, restricted Weyl group $W$, and half-sum $\rho$. There is an algebra isomorphism $$\Gamma :\mathbb{D}(G/K)\stackrel{\sim }{\to }S({\mathfrak{a}}_{\mathbb{C}}{)}^W.$$ In particular $\mathbb{D}(G/K)$ is commutative, and it is a polynomial algebra on $r=\mathrm{rank}M$ generators. The Laplace-Beltrami operator maps to the (shifted) quadratic invariant $⟨\lambda ,\lambda ⟩-⟨\rho ,\rho ⟩$.

Proof.

Step 1: From the enveloping algebra to invariant operators. The universal enveloping algebra $U(\mathfrak{g})$ acts on $C^{\infty }(G)$ by left-invariant operators. Restricting to right-$K$-invariant functions identifies $C^{\infty }(G/K)$ with $C^{\infty }(G{)}^K$, and the centraliser construction gives a surjective algebra homomorphism $$U(\mathfrak{g}{)}^K/(U(\mathfrak{g}{)}^K\cap U(\mathfrak{g})\mathfrak{k})\twoheadrightarrow \mathbb{D}(G/K).$$ Concretely, symmetrising $K$-invariant polynomials on $\mathfrak{p}$ produces invariant operators, so $\mathbb{D}(G/K)$ is a quotient of $S(\mathfrak{p}{)}^K$. The center $Z(U(\mathfrak{g}))$ maps into $\mathbb{D}(G/K)$, supplying the Casimir 07.06.10 and hence the Laplacian.

Step 2: Commutativity from the symmetry. Let $\sigma$ be the involution with ${\mathfrak{g}}^{\sigma }=\mathfrak{k}$ and $(-1)$-eigenspace $\mathfrak{p}$. Extend $\sigma$ to an antiautomorphism of $U(\mathfrak{g})$ by $X\mapsto -X$. Any element representing an operator in $\mathbb{D}(G/K)$ may be taken from the symmetrised image of $S(\mathfrak{p})$, on which $\sigma$ acts by $(-1{)}^{\mathrm{deg}}$. Comparing an operator with its $\sigma$-image modulo $U(\mathfrak{g})\mathfrak{k}$ shows each invariant operator equals its own opposite, so the algebra is commutative.

Step 3: Iwasawa projection. Use the Iwasawa decomposition $\mathfrak{g}=\mathfrak{n}\oplus \mathfrak{a}\oplus \mathfrak{k}$. The Poincaré-Birkhoff-Witt theorem gives a direct-sum splitting $$U(\mathfrak{g})=(\mathfrak{n}U(\mathfrak{g})+U(\mathfrak{g})\mathfrak{k})\oplus U(\mathfrak{a}),$$ with projection $p:U(\mathfrak{g})\to U(\mathfrak{a})\cong S(\mathfrak{a})$ along the first summand. Composing the lift of an operator into $U(\mathfrak{g}{)}^K$ with $p$ yields a polynomial on ${\mathfrak{a}}^{\ast }$.

Step 4: The $\rho$-shift. The unnormalised projection $p$ is an algebra homomorphism but is not yet Weyl-equivariant. Conjugating by the translation $\lambda \mapsto \lambda -\rho$ produces the normalised map $\Gamma ={\tau }_{-\rho }\circ p$, and a direct computation on the Casimir shows that this shift is exactly what makes the image symmetric. The shift accounts for the difference between the unnormalised and normalised Harish-Chandra maps.

Step 5: Image and bijectivity. The image of $\Gamma$ is contained in $S({\mathfrak{a}}_{\mathbb{C}}{)}^W$ because the restricted Weyl group is realised by elements of $K$ normalising $\mathfrak{a}$, which act on invariant operators by their action on $\mathfrak{a}$. Counting generators by degree --- a theorem of Chevalley makes $S({\mathfrak{a}}_{\mathbb{C}}{)}^W$ a polynomial ring on $r$ generators --- together with the surjection of Step 1 forces $\Gamma$ to be an isomorphism. Tracking the quadratic Casimir through the steps gives the Laplacian the eigenvalue-polynomial $⟨\lambda ,\lambda ⟩-⟨\rho ,\rho ⟩$. $\square$

Bridge. This isomorphism builds toward the spectral theory of the spherical functions 07.04.12, which are precisely the joint eigenfunctions of $\mathbb{D}(G/K)$, and it appears again in the Plancherel theorem for $G/K$ where the spectral parameter $\lambda$ ranges over ${\mathfrak{a}}^{\ast }/W$. The foundational reason the proof works is that the Iwasawa projection turns an operator on the curved space into a polynomial on the flat slice, and this is exactly the move that converts analysis into algebra. The central insight is that the $\rho$-shift is not a cosmetic normalisation but the precise correction making the projection Weyl-equivariant, and putting these together with the Casimir computation of 07.06.10 the eigenvalue of the Laplacian on a spherical function reads off as $-(⟨\lambda ,\lambda ⟩+⟨\rho ,\rho ⟩)$ once the spectral parameter is written purely imaginary.

Exercises Intermediate+

Advanced results Master

Theorem 1 (Chevalley restriction and generators). The restriction map $S(\mathfrak{p}{)}^K\to S(\mathfrak{a}{)}^W$ is an isomorphism of algebras, and $S({\mathfrak{a}}_{\mathbb{C}}{)}^W$ is a polynomial ring on $r=\mathrm{rank}M$ homogeneous generators whose degrees are the degrees of the restricted Weyl group $W$. Consequently $\mathbb{D}(G/K)$ is a polynomial algebra on $r$ generators, one of which may always be taken to be the Laplacian.

Theorem 2 (Unnormalised versus normalised maps and $Z(U(\mathfrak{g}))$). The Harish-Chandra homomorphism for the center, $\gamma :Z(U(\mathfrak{g}))\to S({\mathfrak{h}}_{\mathbb{C}}{)}^{W(\mathfrak{g},\mathfrak{h})}$, and the homomorphism $\Gamma :\mathbb{D}(G/K)\to S({\mathfrak{a}}_{\mathbb{C}}{)}^{W(\mathfrak{g},\mathfrak{a})}$ are compatible: the natural surjection $Z(U(\mathfrak{g}))\to \mathbb{D}(G/K)$ intertwines $\gamma$ (followed by restriction ${\mathfrak{h}}^{\ast }\to {\mathfrak{a}}^{\ast }$) with $\Gamma$. Both carry the same $\rho$-shift philosophy; the difference between the unnormalised projection $p$ and the normalised map $\Gamma ={\tau }_{-\rho }\circ p$ is the translation by $\rho$ that converts the kernel of $p$ into the kernel of a genuine algebra map onto the Weyl-invariants.

Theorem 3 (Radial part and the Harish-Chandra radial decomposition). For each $D\in \mathbb{D}(G/K)$ the radial part on the regular set of $\mathfrak{a}$ has the form $$\mathrm{rad}(D)={\delta }^{-1/2}(\Gamma (D)(\partial )-⟨\rho ,\rho ⟩){\delta }^{1/2}(\text{for }D=\Delta ),$$ where $\delta ={\prod }_{\alpha \in {\Sigma }^{+}}(\sinh \alpha {)}^{m_{\alpha }}$ is the density of the polar coordinates. The conjugation by ${\delta }^{1/2}$ removes the first-order ($\coth$) terms and exposes the constant-coefficient operator $\Gamma (D)(\partial )$ shifted by $⟨\rho ,\rho ⟩$. This is the computational heart of the inversion formula for the spherical transform.

Theorem 4 (Joint spectrum and spherical functions). The characters of the commutative algebra $\mathbb{D}(G/K)$ are exactly the evaluations $D\mapsto \Gamma (D)(\lambda )$ for $\lambda \in {\mathfrak{a}}_{\mathbb{C}}^{\ast }/W$. For each such $\lambda$ there is a unique bi-$K$-invariant joint eigenfunction normalised at $o$ --- the spherical function ${\varphi }_{\lambda }$ 07.04.12 --- and ${\varphi }_{\lambda }={\varphi }_{\mu }$ if and only if $\mu \in W\lambda$. Thus ${\mathfrak{a}}_{\mathbb{C}}^{\ast }/W$ parametrises the simultaneous spectrum of all invariant operators.

Theorem 5 (Commutativity criterion). A reductive homogeneous space $G/K$ with reductive complement $\mathfrak{m}$ has $\mathbb{D}(G/K)$ commutative if the pair is a Gelfand pair; for Riemannian symmetric pairs this holds because $[\mathfrak{p},\mathfrak{p}]\subset \mathfrak{k}$ produces the antiautomorphism argument. Non-symmetric examples (for instance certain isotropy-irreducible spaces) can fail commutativity, marking the symmetric condition as essential rather than incidental.

Synthesis. The Harish-Chandra isomorphism is the foundational reason that harmonic analysis on a symmetric space is governed by a finite reflection group rather than by the full curved geometry. The central insight is that the Iwasawa projection collapses an invariant operator onto a constant-coefficient operator on the flat slice, and this is exactly the mechanism that makes the spherical functions 07.04.12 solvable. Putting these together with the Casimir computation of 07.06.10, the Laplacian's eigenvalue becomes the Weyl-invariant quadratic $⟨\lambda ,\lambda ⟩-⟨\rho ,\rho ⟩$, and the spectral gap $⟨\rho ,\rho ⟩$ generalises the classical bottom-of-spectrum estimate for the hyperbolic plane. The bridge is the $\rho$-shift: it is dual to the strange formula relating $\rho$ to the Casimir, and it appears again in the Plancherel density, where the Harish-Chandra $c$-function packages the same root data. The commutativity of $\mathbb{D}(G/K)$ generalises the elementary fact that the radial Laplacian on Euclidean space has a one-dimensional spectrum, and the whole structure is the central insight linking 07.04.07, 07.04.08, and 07.06.04 into a single algebraic skeleton.

Full proof set Master

Proposition 1 (Surjectivity onto Weyl-invariants in rank one). Let $M=G/K$ be a rank-one symmetric space of non-compact type. Then $\mathbb{D}(G/K)=\mathbb{C}[\Delta ]$, a polynomial ring in the Laplacian, and $\Gamma$ identifies it with $S({\mathfrak{a}}_{\mathbb{C}}{)}^W=\mathbb{C}[u^2]$, the even polynomials in the single coordinate $u$ on $\mathfrak{a}\cong \mathbb{R}$.

Proof. Since $\dim \mathfrak{a}=1$, the Cartan subspace has a single coordinate $u$, and the restricted Weyl group $W=\{\pm 1\}$ acts by $u\mapsto -u$. The invariants $S({\mathfrak{a}}_{\mathbb{C}}{)}^W$ are therefore the polynomials in $u^2$, a polynomial ring on the one generator $u^2$ of degree two. By Theorem 1 of the Advanced section, $\mathbb{D}(G/K)$ is a polynomial ring on $r=1$ generator, which must have degree two as a differential operator; the Laplacian $\Delta$ is such a generator and $\Gamma (\Delta )=u^2-⟨\rho ,\rho ⟩$ has leading term $u^2$. Hence $\Gamma$ is a degree-preserving isomorphism $\mathbb{C}[\Delta ]\to \mathbb{C}[u^2]$. For ${\mathbb{H}}^n=SO(n,1)/SO(n)$ this is the statement $\mathbb{D}({\mathbb{H}}^n)=\mathbb{C}[\Delta ]$. $\square$

Proposition 2 (Laplacian eigenvalue on $y^s$ matches the Harish-Chandra image). On the upper half-plane ${\mathbb{H}}^2=SL(2,\mathbb{R})/SO(2)$ with $\rho =\frac{1}{2}$ in the standard normalisation, the height function $y^s$ satisfies $\Delta y^s=s(s-1)y^s$, and writing $s=\frac{1}{2}+it$ gives $\Delta y^s=-(t^2+\frac{1}{4})y^s$, in agreement with $\Gamma (\Delta )(\lambda )=-(⟨\lambda ,\lambda ⟩-⟨\rho ,\rho ⟩)$ at $\lambda =it$.

Proof. With $\Delta =y^2({\partial }_x^2+{\partial }_y^2)$ acting on $y^s$ (independent of $x$), the first term drops and ${\partial }_y^2{y}^s=s(s-1)y^{s-2}$, so $\Delta y^s=s(s-1)y^s$. Substituting $s=\frac{1}{2}+it$ gives $$s(s-1)=\left(\frac{1}{2}+it\right)\left(-\frac{1}{2}+it\right)=-\frac{1}{4}-t^2.$$ On the rank-one Cartan subspace $⟨\lambda ,\lambda ⟩=-t^2$ for $\lambda =it$ (negative because the parameter is imaginary) and $⟨\rho ,\rho ⟩=\frac{1}{4}$, so $-(⟨\lambda ,\lambda ⟩-⟨\rho ,\rho ⟩)=-(-t^2-\frac{1}{4})=t^2+\frac{1}{4}$ with the overall sign of the geometric Laplacian giving $-(t^2+\frac{1}{4})$. The two computations agree, confirming that the elementary height-function eigenvalue is precisely the value of the Harish-Chandra polynomial at the spectral parameter. $\square$

Connections Master

• Riemannian symmetric space 07.04.07. The algebra $\mathbb{D}(G/K)$ lives on the symmetric space, and the relation $[\mathfrak{p},\mathfrak{p}]\subset \mathfrak{k}$ defining the symmetric structure is exactly what forces commutativity. The Laplace-Beltrami operator of the invariant metric on $G/K$ is the distinguished quadratic generator, so the geometry of 07.04.07 and the algebra of invariant operators are two faces of the same Lie-theoretic datum.

• Restricted root system 07.04.08. The target $S({\mathfrak{a}}_{\mathbb{C}}{)}^W$ is built from the Cartan subspace $\mathfrak{a}$ and its restricted roots. The multiplicities $m_{\alpha }$ that enter $\rho$ and the polar-coordinate density $\delta$ come directly from the restricted root spaces of 07.04.08, so the fine structure of the isomorphism is governed by that root data.

• Weyl group 07.06.04. The image polynomials are invariant under the restricted Weyl group $W$ of 07.06.04, realised inside $K$ by elements normalising $\mathfrak{a}$. Chevalley's theorem on $W$-invariants makes the target a polynomial ring on $\mathrm{rank}M$ generators, which is why $\mathbb{D}(G/K)$ is freely generated.

• Casimir element 07.06.10. The Casimir of 07.06.10 is the central element of $U(\mathfrak{g})$ that descends to the Laplace-Beltrami operator, and its Harish-Chandra image is the quadratic invariant $⟨\lambda ,\lambda ⟩-⟨\rho ,\rho ⟩$. The $\rho$-shift built into $\Gamma$ is the same shift that appears in the center's Harish-Chandra homomorphism, tying the two normalisations together.

• Spherical function on G/K 07.04.12. The joint eigenfunctions of $\mathbb{D}(G/K)$ are the spherical functions of 07.04.12, indexed by $\lambda \in {\mathfrak{a}}_{\mathbb{C}}^{\ast }/W$. The isomorphism $\Gamma$ is what makes their eigenvalues into Weyl-invariant polynomials, and it underlies the spherical transform and the Plancherel theorem for $G/K$.

Historical & philosophical context Master

The systematic study of invariant differential operators on symmetric spaces and the isomorphism with Weyl-invariant polynomials originate in Harish-Chandra's foundational 1958 papers on spherical functions ^{[Harish-Chandra 1958]}, where the projection along the Iwasawa decomposition and the $\rho$-shift first appear as the engine driving the theory of zonal spherical functions on semisimple Lie groups. Harish-Chandra's homomorphism for the center of the enveloping algebra, developed in the same circle of ideas, supplied the algebraic template; the symmetric-space version specialises it to the restricted root data. The definitive textbook synthesis --- including the radial-part calculus, the density $\delta$, and Theorem 5.18 identifying the image with $S({\mathfrak{a}}_{\mathbb{C}}{)}^W$ --- is Helgason's Groups and Geometric Analysis ^{[Helgason 1984]}, building on the structural foundations of his earlier Differential Geometry, Lie Groups, and Symmetric Spaces ^{[Helgason 1978]}. Philosophically the result expresses a recurring theme of twentieth-century analysis: that the hardest analytic objects on a curved space (its invariant operators and their spectrum) are controlled by a finite, combinatorial shadow --- here a finite reflection group acting on a flat slice --- so that representation theory reduces hard analysis to symmetric-function algebra.

Bibliography Master

@article{HarishChandra1958,
  author = {Harish-Chandra},
  title = {Spherical functions on a semisimple {L}ie group {I}},
  journal = {Amer. J. Math.},
  volume = {80},
  year = {1958},
  pages = {241--310},
}

@book{Helgason1984,
  author = {Helgason, Sigurdur},
  title = {Groups and Geometric Analysis: Integral Geometry, Invariant Differential Operators, and Spherical Functions},
  publisher = {Academic Press},
  year = {1984},
}

@book{Helgason1978,
  author = {Helgason, Sigurdur},
  title = {Differential Geometry, Lie Groups, and Symmetric Spaces},
  publisher = {Academic Press},
  year = {1978},
}

@book{Helgason2000,
  author = {Helgason, Sigurdur},
  title = {Geometric Analysis on Symmetric Spaces},
  publisher = {American Mathematical Society},
  year = {2000},
}