Hermitian symmetric space

Intuition [Beginner]

A Hermitian symmetric space is a symmetric space that is also a complex manifold --- you can multiply tangent vectors by $i$ in a way that respects both the symmetry and the distance structure. The complex upper half-plane (the set of complex numbers with positive imaginary part) is the most familiar example.

The combination of symmetry and complex structure is extremely restrictive. A symmetric space already has the maximum amount of geometric symmetry, and requiring an invariant complex structure on top of that picks out a distinguished subclass. The result is a space where complex analysis, differential geometry, and Lie group theory all interact.

Examples include the complex upper half-plane $\mathfrak{H}=\{x+iy:y>0\}$, the Siegel upper half-space of symmetric complex matrices with positive-definite imaginary part, and the Grassmannian of $k$-planes in complex $n$-space. Why does this concept exist? Hermitian symmetric spaces are the geometric setting for automorphic forms and much of modern number theory.

Visual [Beginner]

A drawing of the complex upper half-plane $\mathfrak{H}$, shaded above the real axis, with a point $z=x+iy$ marked. Arrows show the two directions of motion: horizontal (real translation, changing $x$) and vertical (scaling, changing $y$). The imaginary unit $i$ acts as a 90-degree rotation in the tangent plane at every point.

The complex structure is compatible with the hyperbolic metric: rotating by $i$ preserves distances, making $\mathfrak{H}$ both symmetric and Kahler.

Worked example [Beginner]

Consider the complex upper half-plane $\mathfrak{H}=\{z=x+iy\in \mathbb{C}:y>0\}$ with hyperbolic metric $d{s}^2=(d{x}^2+d{y}^2)/y^2$. We verify that $\mathfrak{H}$ is a Hermitian symmetric space.

Step 1. As a symmetric space, $\mathfrak{H}$ has geodesic symmetries at every point 07.04.07. The isometry group is $PSL(2,\mathbb{R})$ acting by Mobius transformations.

Step 2. The complex structure is multiplication by $i$ in the tangent plane: at the point $z$, the tangent directions "rightward" and "upward" are related by $J(\text{rightward})=\text{upward}$ and $J(\text{upward})=-\text{rightward}$. This is a rotation by 90 degrees.

Step 3. The complex structure is invariant under $PSL(2,\mathbb{R})$: Mobius transformations are holomorphic, so they preserve the multiplication-by-$i$ operation on tangent vectors. The metric is also $J$-invariant since $d{s}^2$ treats $dx$ and $dy$ symmetrically.

What this tells us: the upper half-plane carries both a symmetric space structure and an invariant complex structure, making it the simplest non-flat Hermitian symmetric space.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $M=G/K$ be a Riemannian symmetric space with Cartan decomposition $\mathfrak{g}=\mathfrak{k}\oplus \mathfrak{p}$ at the base point.

Definition (Invariant complex structure). An invariant complex structure on $M=G/K$ is a $G$-invariant endomorphism $J$ of the tangent bundle $TM$ satisfying $J^2=-\mathrm{id}$ and $\nabla J=0$ (parallel with respect to the Levi-Civita connection).

Definition (Hermitian symmetric space). A Hermitian symmetric space is a Riemannian symmetric space $M$ equipped with an invariant complex structure $J$ that is compatible with the metric: $g(JX,JY)=g(X,Y)$ for all tangent vectors $X,Y$.

The conditions $J^2=-\mathrm{id}$, $\nabla J=0$, and $g(JX,JY)=g(X,Y)$ make $M$ a Kahler manifold. So a Hermitian symmetric space is a symmetric space that is also Kahler.

Proposition (Algebraic characterisation). A symmetric space $M=G/K$ of the non-compact type is Hermitian if and only if the centre $Z(\mathfrak{k})$ of the isotropy Lie algebra $\mathfrak{k}$ is non-zero. In that case, $Z(\mathfrak{k})$ is one-dimensional, generated by an element $J_0\in Z(\mathfrak{k})$ with $J_0^2=-\mathrm{id}$ on $\mathfrak{p}$, and the invariant complex structure is $J=\mathrm{ad}(J_0){|}_{\mathfrak{p}}$.

Counterexamples to common slips

• Not every Kahler manifold is symmetric. Complex projective space ${\mathbb{CP}}^n$ is both Kahler and symmetric. But a generic Kahler metric on a complex manifold is not symmetric --- the parallel curvature condition $\nabla R=0$ is a severe additional constraint.

• Not every symmetric space is Hermitian. The real Grassmannian $\mathrm{Gr}(k,{\mathbb{R}}^n)=SO(n)/(SO(k)\times SO(n-k))$ is symmetric but not Hermitian when both $k$ and $n-k$ are odd, because the isotropy algebra has no central element acting as a complex structure on $\mathfrak{p}$.

• The centre of $\mathfrak{k}$ being one-dimensional is a property, not a definition. The definition requires an invariant complex structure; the centre condition is the algebraic characterisation that detects when such a structure exists.

Key theorem with proof [Intermediate+]

Theorem (Cartan's classification of Hermitian symmetric spaces). The irreducible Hermitian symmetric spaces of non-compact type are classified into four infinite families and one exceptional case. The classification is:

Type	Space	Dimension (real)	Rank
AIII	$SU(p,q)/S(U(p)\times U(q))$	$2pq$	$\min (p,q)$
BDI	$S{O}_0(2,n)/SO(2)\times SO(n)$	$2n$	$2$ (for $n\ge 3$)
DIII	$S{O}^{\ast }(2n)/U(n)$	$n(n-1)$	$[n/2]$
CI	$Sp(2n,\mathbb{R})/U(n)$	$n(n+1)$	$n$
EIII	$E_{6(-14)}/(SO(10)\times U(1))$	$32$	$2$
EVII	$E_{7(-25)}/(E_6\times U(1))$	$54$	$3$

Each non-compact Hermitian symmetric space $G/K$ has a compact dual $U/K$ (obtained by replacing the non-compact group $G$ with its compact dual $U$), and vice versa.

Proof. The proof has three parts: (1) establishing the algebraic criterion, (2) classifying irreducible OSLAs with the criterion, and (3) matching each OSLA to its symmetric space.

Part 1: The centre criterion. Let $(\mathfrak{g},\theta )$ be an effective orthogonal symmetric Lie algebra with decomposition $\mathfrak{g}=\mathfrak{k}\oplus \mathfrak{p}$. The symmetric space $G/K$ is Hermitian if and only if $Z(\mathfrak{k})\ne 0$.

For the forward direction, if $G/K$ carries an invariant complex structure $J$, then $J$ corresponds to an element $J_0\in \mathfrak{k}$ with $\mathrm{ad}(J_0){|}_{\mathfrak{p}}=J$. The $G$-invariance of $J$ means $J_0$ commutes with all of $\mathfrak{k}$, so $J_0\in Z(\mathfrak{k})$. Since $J^2=-\mathrm{id}$ on $\mathfrak{p}$, the element $J_0$ is non-zero.

For the reverse direction, if $J_0\in Z(\mathfrak{k})$ is non-zero, then $\mathrm{ad}(J_0)$ preserves $\mathfrak{p}$ (by $[\mathfrak{k},\mathfrak{p}]\subset \mathfrak{p}$) and $J_0\in Z(\mathfrak{k})$ means $\mathrm{ad}(J_0)$ commutes with $\mathrm{ad}(\mathfrak{k})$. The bracket $[J_0,[J_0,X]]=-[[J_0,X],J_0]-[[X,J_0],J_0]$. By the Jacobi identity, $[J_0,[J_0,X]]=-[[J_0,X],J_0]+[[J_0,X],J_0]+[X,[J_0,J_0]]$. Since $J_0\in \mathfrak{k}$, $[J_0,J_0]=0$. So $J=\mathrm{ad}(J_0){|}_{\mathfrak{p}}$ satisfies $J^{2}X=[J_0,[J_0,X]]$. The condition $J^2=-\mathrm{id}$ on $\mathfrak{p}$ follows from the representation theory of $\mathfrak{k}$ on $\mathfrak{p}$: the adjoint action of $J_0$ on $\mathfrak{p}$ has eigenvalues $\pm i$, and by effectiveness these must both appear, forcing $J^2=-\mathrm{id}$.

Part 2: Classification. The irreducible symmetric spaces of non-compact type correspond to simple real Lie algebras $\mathfrak{g}$ with Cartan involution $\theta$. The Hermitian condition adds: the fixed-point algebra $\mathfrak{k}$ has non-zero centre. For each simple $\mathfrak{g}$, the structure of $\mathfrak{k}$ is determined by the Satake diagram. The centre $Z(\mathfrak{k})$ is non-zero if and only if the Satake diagram has exactly one white node connected to the rest of the diagram at a single point (or equivalently, the restricted root system has a one-dimensional centre). Going through the classification:

• Type $A_n$: $SU(p,q)$ with $p,q\ge 1$. Here $\mathfrak{k}=\mathfrak{s}(\mathfrak{u}(p)\oplus \mathfrak{u}(q))$, whose centre is one-dimensional (generated by $\mathrm{diag}(q\cdot I_p,-p\cdot I_q)$). This gives the AIII series.
• Type $C_n$: $Sp(2n,\mathbb{R})$. Here $\mathfrak{k}=\mathfrak{u}(n)$, whose centre is one-dimensional (scalar multiples of $i{I}_n$). This gives the CI series.
• Type $D_n$: $S{O}^{\ast }(2n)$. Here $\mathfrak{k}=\mathfrak{u}(n)$, centre one-dimensional. This gives the DIII series.
• Type $B_n$: $S{O}_0(2,n)$. Here $\mathfrak{k}=\mathfrak{so}(2)\oplus \mathfrak{so}(n)$, centre generated by $\mathfrak{so}(2)$. This gives BDI with $p=1$ (in the notation of $S{O}_0(p,q)$).
• Exceptional types: only $E_6$ and $E_7$ admit Hermitian symmetric spaces, giving EIII and EVII.

Part 3: Matching. Each entry in the table above is obtained by writing the symmetric space $G/K$ for the corresponding real form and computing dimensions. The compact duals are obtained by replacing $G$ with its compact real form and keeping $K$. $\square$

Bridge. This classification builds toward the Harish-Chandra embedding theorem that realises each non-compact Hermitian symmetric space as a bounded domain in ${\mathbb{C}}^n$, and appears again in the theory of automorphic forms on Hermitian symmetric domains. The foundational reason is that the centre of $\mathfrak{k}$ provides a canonical complex structure on $\mathfrak{p}$, and this is exactly the bridge between the Lie-algebraic classification and the complex-geometric reality of bounded symmetric domains. Putting these together with the duality from 07.04.06, the compact Hermitian symmetric spaces (generalised flag varieties) are the Borel-Serre compactifications of their non-compact duals.

Exercises [Intermediate+]

Advanced results [Master]

Theorem 1 (Harish-Chandra embedding). Every non-compact Hermitian symmetric space $M=G/K$ of rank $r$ admits a realisation as a bounded symmetric domain $D\subset {\mathbb{C}}^n$ via the Harish-Chandra embedding. Explicitly, $D$ is the image of the map ${\mathfrak{p}}^{-}\to {\mathfrak{p}}^{+}$ defined by the Cayley transform, where ${\mathfrak{p}}^{\pm }$ are the $\pm i$-eigenspaces of $J_0$ on ${\mathfrak{p}}_{\mathbb{C}}$. The domain $D$ is a circular bounded domain centred at the origin, and the action of $G$ on $D$ extends to a holomorphic action on all of ${\mathbb{C}}^n$.

Theorem 2 (Borel embedding). Let $M^{\ast }=U/K$ be the compact dual of the non-compact Hermitian symmetric space $M=G/K$. The Borel embedding identifies $M$ as an open $G$-orbit in $M^{\ast }$: $M=G\cdot o\hookrightarrow M^{\ast }$, and the complement $M^{\ast }\setminus M$ is a complex-analytic subvariety of complex codimension one. The compact dual $M^{\ast }$ is a generalised flag variety $U/(U\cap P)$ for a suitable parabolic subgroup $P$.

Theorem 3 (Tube domain realisation). Every Hermitian symmetric space of tube type admits a realisation as a tube domain $\mathfrak{V}+i\Omega$ where $\mathfrak{V}$ is a real vector space and $\Omega \subset \mathfrak{V}$ is a self-dual open homogeneous cone. The Hermitian symmetric spaces of tube type are exactly those in which the Shilov boundary has real dimension equal to the rank. The non-tube-type domains require the general Harish-Chandra embedding.

Theorem 4 (Shilov boundary). The Shilov boundary $S$ of a bounded symmetric domain $D$ is a compact homogeneous space $S=K/L$ where $L$ is a subgroup of $K$. It carries a natural CR structure (Cauchy-Riemann structure) and is the maximal boundary for holomorphic functions. The dimension of $S$ equals the rank of $D$ in the tube-type case and is larger otherwise.

Theorem 5 (Connection to automorphic forms). Let $D=G/K$ be a bounded symmetric domain and $\Gamma \subset G$ a torsion-free arithmetic subgroup. The quotient $\Gamma \backslash D$ is a quasi-projective algebraic variety (by the Baily-Borel theorem), and automorphic forms for $\Gamma$ are sections of line bundles on $\Gamma \backslash D$. The Hermitian structure of $D$ is what makes $\Gamma \backslash D$ an algebraic variety --- this fails for general symmetric spaces.

Synthesis. The Hermitian symmetric space is the foundational reason that symmetric spaces interact with complex analysis and number theory. The central insight is that the centre of $\mathfrak{k}$ provides the invariant complex structure, and this is exactly the bridge to the Harish-Chandra embedding. Putting these together with the Borel embedding, the non-compact Hermitian symmetric space sits inside its compact dual as a dense open subset, and the bridge is the tube domain realisation that generalises the upper half-plane. The pattern generalises: every Hermitian symmetric domain of tube type is a Siegel domain of the first kind, and this is exactly the structure that underlies the theory of Siegel modular forms and automorphic representations.

Full proof set [Master]

Proposition 1 (Centre of $\mathfrak{k}$ is at most one-dimensional for irreducible spaces). Let $(\mathfrak{g},\theta )$ be an irreducible orthogonal symmetric Lie algebra. If $\mathfrak{g}$ is of non-compact type and $Z(\mathfrak{k})\ne 0$, then $Z(\mathfrak{k})$ is one-dimensional.

Proof. The adjoint action of $\mathfrak{k}$ on $\mathfrak{p}$ is faithful (by effectiveness of the OSLA). Since $\mathfrak{g}$ is simple and $\mathfrak{k}$ is a maximal compactly embedded subalgebra, the representation $\mathrm{ad}{|}_{\mathfrak{p}}:\mathfrak{k}\to \mathrm{End}(\mathfrak{p})$ is irreducible. The centre $Z(\mathfrak{k})$ acts by scalars on each irreducible $\mathfrak{k}$-submodule of $\mathfrak{p}$ (by Schur's lemma). Since $\mathfrak{p}$ is irreducible, $Z(\mathfrak{k})$ acts by scalars on all of $\mathfrak{p}$. The constraint $J^2=-\mathrm{id}$ forces the scalar to be $\pm i$ (pure imaginary), so $Z(\mathfrak{k})$ is at most one-dimensional. $\square$

Proposition 2 (Hermitian implies irreducible root system of type $B{C}_r$, $C_r$, or standard). Let $M=G/K$ be an irreducible Hermitian symmetric space of non-compact type with rank $r$. The restricted root system of $M$ has the property that the root multiplicities $m_{\alpha }$ are constant on each orbit of the Weyl group, and the restricted root system is of type $B{C}_r$ when the space is not of tube type.

Proof. The Hermitian condition means ${\mathfrak{p}}_{\mathbb{C}}={\mathfrak{p}}^{+}\oplus {\mathfrak{p}}^{-}$ where ${\mathfrak{p}}^{\pm }$ are the $\pm i$-eigenspaces of $J_0$. The root space decomposition of ${\mathfrak{g}}_{\mathbb{C}}$ with respect to a maximal abelian subalgebra $\mathfrak{a}\subset \mathfrak{p}$ decomposes ${\mathfrak{p}}^{+}$ and ${\mathfrak{p}}^{-}$ into root spaces. The restricted roots $\alpha \in {\mathfrak{a}}^{\ast }$ come in pairs $\pm \alpha$, and the root spaces in ${\mathfrak{p}}^{+}$ and ${\mathfrak{p}}^{-}$ have equal dimensions. If there exists a root $\alpha$ such that $2\alpha$ is also a root, the root system is of type $B{C}_r$ and the space is not of tube type. If no such root exists and the root system is of type $C_r$, the space is of tube type. $\square$

Connections [Master]

• Riemannian symmetric space 07.04.07. Hermitian symmetric spaces are the subclass of Riemannian symmetric spaces admitting an invariant complex structure. The algebraic criterion (centre of $\mathfrak{k}$) is detectable from the OSLA data of 07.04.07 without additional geometric input.

• Orthogonal symmetric Lie algebra 07.04.06. The Hermitian condition at the Lie algebra level is the requirement that $Z(\mathfrak{k})$ be non-zero, which constrains the OSLA classification to the four infinite families and two exceptional cases. The duality of 07.04.06 exchanges compact and non-compact Hermitian symmetric spaces while preserving the complex structure.

• Restricted root system 07.04.08. The restricted root system of a Hermitian symmetric space carries additional structure: the root spaces split between ${\mathfrak{p}}^{+}$ and ${\mathfrak{p}}^{-}$, and the Hermitian type is detectable from whether the restricted root system is of type $B{C}_r$ or not.

Historical & philosophical context [Master]

Elie Cartan identified the Hermitian symmetric spaces in his 1935 paper on bounded homogeneous domains (Abh. Math. Sem. Univ. Hamburg 11) ^[Cartan1935], proving that every bounded homogeneous domain in ${\mathbb{C}}^n$ is symmetric and classifying the irreducible ones. Cartan's insight was that the holomorphic automorphism group of a bounded domain forces enough symmetry to make the domain a symmetric space, and the invariant complex structure comes for free from the ambient ${\mathbb{C}}^n$.

Harish-Chandra developed the embedding theory in his series on representations of semisimple Lie groups (Amer. J. Math. 78, 1956) ^{[HarishChandra1956]}, constructing the bounded realisation of every non-compact Hermitian symmetric space. The Borel embedding appears in Borel's work on linear algebraic groups and homogeneous spaces. Satake's 1980 monograph (Algebraic Structures of Symmetric Domains) ^[Satake1980] unified the theory, connecting the Harish-Chandra embedding, tube domains, and the Jordan-algebraic structure underlying tube-type domains.

Bibliography [Master]

@article{Cartan1935,
  author = {Cartan, Elie},
  title = {Sur les domaines bornes homogenes de l'espace de $n$ variables complexes},
  journal = {Abh. Math. Sem. Univ. Hamburg},
  volume = {11},
  year = {1935},
  pages = {116--162},
}

@article{HarishChandra1956,
  author = {Harish-Chandra},
  title = {Representations of semisimple Lie groups {IV}},
  journal = {Amer. J. Math.},
  volume = {78},
  year = {1956},
  pages = {1--41},
}

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

@book{Satake1980,
  author = {Satake, Ichiro},
  title = {Algebraic Structures of Symmetric Domains},
  publisher = {Iwanami Shoten / Princeton Univ. Press},
  year = {1980},
}