Mutation (Jordan algebra)

In mathematics, a mutation, also called a homotope, of a unital Jordan algebra is a new Jordan algebra defined by a given element of the Jordan algebra. The mutation has a unit if and only if the given element is invertible, in which case the mutation is called a proper mutation or an isotope. Mutations were first introduced by Max Koecher in his Jordan algebraic approach to Hermitian symmetric spaces and bounded symmetric domains of tube type. Their functorial properties allow an explicit construction of the corresponding Hermitian symmetric space of compact type as a compactification of a finite-dimensional complex semisimple Jordan algebra. The automorphism group of the compactification becomes a complex subgroup, the complexification of its maximal compact subgroup. Both groups act transitively on the compactification. The theory has been extended to cover all Hermitian symmetric spaces using the theory of Jordan pairs or Jordan triple systems. Koecher obtained the results in the more general case directly from the Jordan algebra case using the fact that only Jordan pairs associated with period two automorphisms of Jordan algebras are required.

Definitions

Let A be a unital Jordan algebra over a field k of characteristic ≠ 2.[1] For a in A define the Jordan multiplication operator on A by

\displaystyle{L(a)b=ab}

and the quadratic representation Q(a) by

 Q(a)=2L(a)^2 -L(a^2). \,

It satisfies

Q(1)=I. \,

the fundamental identity

\displaystyle{Q(Q(a)b)=Q(a)Q(b)Q(a)}

the commutation or homotopy identity

\displaystyle{Q(a)R(b,a)=R(a,b)Q(a) = 2Q(Q(a)b,a),}

where

 R(a,b)c=2Q(a,c)b,\,\,\, Q(x,y)=\frac{1}{2} (Q(x+y)-Q(x)-Q(y)).

In particular if a or b is invertible then

\displaystyle{R(a,b)=2Q(a)Q(a^{-1},b)=2Q(a,b^{-1})Q(b).}

It follows that A with the operations Q and R and the identity element defines a quadratic Jordan algebra, where a quadratic Jordan algebra consists of a vector space A with a distinguished element 1 and a quadratic map of A into endomorphisms of A, aQ(a), satisfying the conditions:

The Jordan triple product is defined by

 \{a,b,c\}=(ab)c+(cb)a -(ac)b, \,

so that

 Q(a)b=\{a,b,a\},\,\,\, Q(a,c)b=\{a,b,c\},\,\,\, R(a,b)c=\{a,b,c\}. \,

There are also the formulas

 Q(a,b)=L(a)L(b)+L(b)L(a) - L(ab),\,\,\, R(a,b)= [L(a),L(b)] + L(ab). \,

For y in A the mutation Ay is defined to the vector space A with multiplication

 a\circ b= \{a,y,b\}. \,

If Q(y) is invertible, the mutual is called a proper mutation or isotope.

Quadratic Jordan algebras

Let A be a quadratic Jordan algebra over a field k of characteristic ≠ 2. Following Jacobson (1969), a linear Jordan algebra structure can be associated with A such that, if L(a) is Jordan multiplication, then the quadratic structure is given by Q(a) = 2L(a)2L(a2).

Firstly the axiom Q(a)R(b,a) = R(a,b)Q(a) can be strengthened to

\displaystyle{Q(a)R(b,a)=R(a,b)Q(a)=2Q(Q(a)b,a).}

Indeed applied to c, the first two terms give

\displaystyle{2Q(a)Q(b,c)a=2Q(Q(a)c,a)b.}

Switching b and c then gives

\displaystyle{Q(a)R(b,a)c=2Q(Q(a)b,a)c.}

Now let

\displaystyle{L(a)=\frac{1}{2}R(a,1).}

Replacing b by a and a by 1 in the identity above gives

\displaystyle{R(a,1)=R(1,a)=2Q(a,1).}

In particular

\displaystyle{L(a)=Q(a,1),\,\,\,L(1)=Q(1,1)=I.}

The Jordan product is given by

\displaystyle{a\circ b = L(a)b=\frac{1}{2}R(a,1)b=Q(a,b)1,}

so that

\displaystyle{a\circ b = b \circ a.}

The formula above shows that 1 is an identity. Defining a2 by aa = Q(a)1, the only remaining condition to be verified is the Jordan identity

\displaystyle{[L(a),L(a^2)]=0.}

In the fundamental identity

\displaystyle{Q(Q(a)b)= Q(a)Q(b)Q(a),}

Replace a by a + t1, set b = 1 and compare the coefficients of t2 on both sides:

\displaystyle{Q(a)=2Q(a,1)^2 - Q(a^2,1)= 2L(a)^2 - L(a^2).}

Setting b = 1 in the second axiom gives

\displaystyle{Q(a)L(a)=L(a)Q(a),}

and therefore L(a) must commute with L(a2).

Inverses

Let A be a unital Jordan algebra over a field k of characteristic ≠ 2. An element a in a unital Jordan algebra A is said to be invertible if there is an element b such that ab = 1 and a2b = a.[2]

Properties.[3]

  • a is invertible if and only if there is an element b such that Q(a)b = a and Q(a)b2 =1. In this case ab = 1 and a2b = a.

If ab = 1 and a2b = a, then Q(a)b = 2a(ab) − (a2)b = a. The Jordan identity [L(x),L(x2)] = 0 can be polarized by replacing x by x + ty and taking the coefficient of t. This gives

\displaystyle{[L(x^2),L(y)]+2[L(xy),L(x)]=0.}

Taking x = a or b and y = b or a shows that L(a2) commutes with L(b) and L(b2) commutes with L(a). Hence (b2)(a2) = 1. Applying L(b) gives b2a = b. Hence Q(a)b2 = 1. Conversely if Q(a)b = a and Q(a)b2 = 1, then the second relation gives Q(a)Q(b)2 Q(a) = I. So both Q(a) and Q(b) are invertible. The first gives Q(a)Q(b)Q(a) = Q(a) so that Q(a) and Q(b) are each other's inverses. Since L(b) commutes with Q(b) it commutes with its inverse Q(a). Similarly L(a) commutes with Q(b). So (a2)b = L(b)a2 = Q(a)b = a and ab = L(b)Q(a)b= Q(a)Q(b)1= 1.

  • a is invertible if and if only Q(a) defines a bijection on A. In that case a−1 = Q(a)−1a. In this case Q(a)−1 = Q(a−1).

Indeed if a is invertible then the above implies Q(a) is invertible with inverse Q(b). Any inverse b satisfies Q(a)b = a, so b = Q(a)−1a. Conversely if Q(a) is invertible let b = Q(a)−1a. Then Q(a)b = a. The fundamental identity then implies that Q(b) and Q(a) are each other's inverses so that Q(a)b2 = Q(a)Q(b)1=1.

  • If an inverse exists it is unique. If a is invertible, its inverse is denoted by a−1.

This follows from the formula a−1 = Q(a)−1a.

  • a is invertible if and only if 1 lies in the image of Q(a).

Suppose that Q(a)c = 1. Then by the fundamental identity Q(a) is invertible, so a is invertible.

  • Q(a)b is invertible if and only if a and b are invertible, in which case (Q(a)b)−1 = Q(a−1)b−1.

This is an immediate consequence of the fundamental identity and the fact that STS is invertible if and only S and T are invertible.

  • If a is invertible, then Q(a)L(a−1) = L(a).

In the commutation identity Q(a)R(b,a) = Q(Q(a)b,a), set b = c2 with c = a−1. Then Q(a)b = 1 and Q(1,a) = L(a). Since L(a) commutes with L(c2), R(b,a) = L(c) = L(a−1).

  • a is invertible if and only if there is an element b such that ab = 1 and [L(a),L(b)] = 0 (a and b "commute"). In this case b = a−1.

If L(a) and L(b) commute, then ba = 1 implies b(a2) = a. Conversely suppose that a is invertible with inverse b. Then ab = 1. Morevoer L(b) commutes with Q(b) and hence its inverse Q(a). So it commutes with L(a) = Q(a)L(b).

  • When A is finite-dimensional over k, an element a is invertible if and only if it is invertible in k[a], in which case a−1 lies in k[a].

The algebra k[a] is commutative and associative, so if b is an inverse there ab =1 and a2b = a. Conversely Q(a) leaves k[a] invariant. So if it is bijective on A it is bijective there. Thus a−1 = Q(a)−1a lies in k[a].

Elementary properties of proper mutations

  • The mutation Ay is unital if and only if y is invertible in which case the unit is given by y−1.
  • The mutation Ay is a unital Jordan algebra if y is invertible
  • The quadratic representation of Ay is given by Qy(x) = Q(x)Q(y).

In fact [4] multiplication in the algebra Ay is given by

\displaystyle{a\circ b = \{a,y,b\},}

so by definition is commutative. It follows that

\displaystyle{a\circ b = L_y(a)b,}

with

\displaystyle{L_y(a)=[L(a),L(y)] + L(ay).}

If e satisfies ae = a, then taking a = 1 gives

\displaystyle{ye=1.}

Taking a = e gives

\displaystyle{e(ya)=y(ea)}

so that L(y) and L(e) commute. Hence y is invertible and e = y−1.

Now for y invertible set

\displaystyle{Q_y(a)=Q(a)Q(y),\,\,\, R_y(a,b)=R(a,Q(y)b).}

Then

\displaystyle{Q_y(e)=Q_y(y^{-1})= Q(y^{-1})Q(y)=I.}

Moreover

\displaystyle{Q_y(a)Q_y(b)Q_y(a)=Q(a)Q(y)Q(b)Q(y)Q(a)Q(y)=Q(a) Q(Q(y)b) Q(a)Q(y) = Q(Q(a)Q(y)b)Q(y)=Q_y(Q_y(a)b).}

Finally

\displaystyle{Q(y)R(c,Q(y)d)Q(y)^{-1}=R(Q(y)c,d),}

since

\displaystyle{Q(y)R(c,Q(y)d)x=2Q(y)Q(c,x)Q(y)d=2Q(Q(y)c,Q(y)x)d=R(Q(y)c,d)Q(y)x.}

Hence

\displaystyle{Q_y(a)R_y(b,a)=Q(a)Q(y)R(b,Q(y)a)=Q(y^{-1}) Q(Q(y)a)R(b,Q(y)a)= Q(y)^{-1} R(Q(y)a,b)Q(Q(y)a)=R_y(a,b) Q_y(a).}

Thus (A,Qy,y−1) is a unital quadratic Jordan algebra. It therefore corresponds to a linear Jordan algebra with the associated Jordan multiplication operator M(a) given by

\displaystyle{M(a)b=\frac{1}{2} R_y(a,e)b= \frac{1}{2} R(a,Q(y)e)b= \frac{1}{2}R(a,y)b=\{a,y,b\}=L_y(a)b.}

This shows that the operators Ly(a) satisfy the Jordan identity so that the proper mutation or isotope Ay is a unital Jordan algebra. The correspondence with quadratic Jordan algebras shows that its quadratic representation is given by Qy.

Nonunital mutations

The definition of mutation also applies to non-invertible elements y. If A is finite-dimensional over R or C, invertible elements a in A are dense, since invertibility is equivalent to the condition that det Q(a) ≠ 0. So by continuity the Jordan identity for proper mutations implies the Jordan identity for arbitrary mutations. In general the Jordan identity can be deduced from Macdonald's theorem for Jordan algebras because it involves only two elements of the Jordan algebra. Alternatively, the Jordan identity can be deduced by realizing the mutation inside a unital quadratic algebra.[5]

For a in A define a quadratic structure on A1 = Ak by

\displaystyle{Q_1(a\oplus \alpha 1)(b\oplus \beta 1)= \alpha^2\beta 1\oplus [\alpha^2 a + \alpha^2 b + 2\alpha\beta a + \alpha 
\{a,y, b\} + \beta Q(a)y + Q(a)Q(y)b].}

It can then be verified that (A1, Q1, 1) is a unital quadratic Jordan algebra. The unital Jordan algebra to which it corresponds has Ay as an ideal, so that in particular Ay satisfies the Jordan identity. The identities for a unital quadratic Jordan algebra follow from the following compatibility properties of the quadratic map Qy(a) = Q(a)Q(y) and the squaring map Sy(a) = Q(a)y:

Hua's identity

Let A be a unital Jordan algebra. If a, b and ab are invertible, then Hua's identity holds:[6]

\displaystyle{a^{-1} = (a-b)^{-1} + (a - Q(a)b^{-1})^{-1} = (a-b)^{-1} + Q(a)^{-1}(a^{-1} - b^{-1})^{-1}.}

In particular if x and 1 – x are invertible, then so too is 1 – x−1 with

\displaystyle{(1-x)^{-1} + (1-x^{-1})^{-1} = 1.}

To prove the identity for x, set y = (1 – x)−1. Then L(y) = Q(1 – x)−1L(1 – x). Thus L(y) commutes with L(x) and Q(x). Since Q(y) = Q(1 – x)−1, it also commutes with L(x) and Q(x). Since L(x−1) = Q(x)−1L(x), L(y) also commutes with L(x−1) and Q(x−1).

It follows that (x−1 – 1)xy =(1 – x) y = 1. Moreover y – 1 = xy since (1 – x)y = 1. So L(xy) commutes with L(x) and hence L(x−1 – 1). Thus 1 – x−1 has inverse 1 – y.

Now let Aa be the mutation of A defined by a. The identity element of Aa is a−1. Moreover an invertible element c in A is also invertible in Aa with inverse Q(a)−1 c−1.

Let x = Q(a)−1b in Aa. It is invertible in A, as is a−1Q(a)−1b = Q(a)−1(ab). So by the special case of Hua's identity for x in Aa

\displaystyle{a^{-1}= Q(a)^{-1}(a^{-1} - Q(a)^{-1}b)^{-1} + Q(a)^{-1}(a^{-1} - b^{-1})^{-1}= (a -b)^{-1} + (a-Q(a)b^{-1})^{-1}.}

Bergman operator

If A is a unital Jordan algebra, the Bergman operator is defined for a, b in A by[7]

\displaystyle{B(a,b) = I - R(a,b) + Q(a)Q(b).}

If a is invertible then

\displaystyle{B(a,b)=Q(a)Q(a^{-1}-b);}

while if b is invertible then

\displaystyle{B(a,b)=Q(a-b^{-1})Q(b).}

In fact if a is invertible

Q(a)Q(a−1b) = Q(a)[Q(a−1 − 2Q(a−1,b) + Q(b)]=I − 2Q(a)Q(a−1,b) + Q(a)Q(b)=IR(a,b) + Q(a)Q(b)

and similarly if b is invertible.

More generally the Bergman operator satisfies a version of the commutation or homotopy identity:

\displaystyle{B(a,b)Q(a)=Q(a)B(b,a)=Q(a-Q(a)b)}

and a version of the fundamental identity:

\displaystyle{Q(B(a,b)c)=B(a,b)Q(c)B(b,a).}

There is also a third more technical identity:

\displaystyle{2Q(B(a,b)c,a-Q(a)b) =B(a,b)(2Q(a,c) - R(c,b)Q(a))=(2Q(a,c)-Q(a)R(b,c))B(b,a).}

Quasi-invertibility

Let A be a finite-dimensional unital Jordan algebra over a field k of characteristic ≠ 2.[8] For a pair (a,b) with a and a−1b invertible define

\displaystyle{a^b= (a^{-1}-b)^{-1}.}

In this case the Bergman operator B(a,b) = Q(a)Q(a−1b) defines an invertible operator on A and

\displaystyle{a^b = B(a,b)^{-1}(a-Q(a)b).}

In fact

\displaystyle{B(a,b)^{-1}(a-Q(a)b)= Q(a^b)Q(a^{-1})(a-Q(a)b)=Q(a^b)(a^b)^{-1}=a^b.}

Moreover by definition a−1bc is invertible if and only if (ab)−1c is invertible. In that case

\displaystyle{a^{b+c}=(a^b)^c.}

Indeed

\displaystyle{a^{b+c}=((a^{-1} - b) -c)^{-1} = ((a^b)^{-1} -c)^{-1}=(a^b)^c.}

The assumption that a be invertible can be dropped since ab can be defined only supposing that the Bergman operator B(a,b) is invertible. The pair (a,b) is then said to be quasi-invertible. In that case ab is defined by the formula

\displaystyle{a^b = B(a,b)^{-1}(a-Q(a)b).}

If B(a,b) is invertible, then B(a,b)c = 1 for some c. The fundamental identity implies that B(a,b)Q(c)B(b,a) = I. So by finite-dimensionality B(b,a) is invertible. Thus (a,b) is invertible if and only if (b,a) is invertible and in this case

\displaystyle{a^b = a +Q(a)b^a.}

In fact

B(a,b)(a + Q(a)ba) = a − 2R(a,b)a + Q(a)Q(b)a + Q(a)(bQ(b)a) = aQ(a)b,

so the formula follows by applying B(a,b)−1 to both sides.

As before (a,b+c) is quasi-invertible if and only if (ab,c) is quasi-invertible; and in that case

\displaystyle{a^{b+c}=(a^b)^c.}

If k = R or C, this would follow by continuity from the special case where a and a−1b were invertible. In general the proof requires four identities for the Bergman operator:

  • \displaystyle{B(a,b)Q(a^b) = Q(a^b)B(b,a)=Q(a)}
  • \displaystyle{B(a,b)Q(a^b,c) + Q(a)R(b,c) =Q(a^b,c)B(b,a) + R(c,b)Q(a) =Q(a,c)}
  • \displaystyle{B(a,b)R(a^b,c)=R(a,c) -2Q(a)Q(b,c)}
  • \displaystyle{ B(a,b)B(a^b,c)=B(a,b+c)}

In fact applying Q to the identity B(a,b)ab = aQ(a)b yields

\displaystyle{B(a,b)Q(a^b)B(b,a)=B(a,b)Q(a)=Q(a)B(b,a).}

The first identity follows by cancelling B(a,b) and B(b,a). The second identity follows by similar cancellation in

B(a,b)Q(ab,c)B(b,a) = Q(B(a,b)ab,B(a,b)c) = Q(aQ(a)b,B(a,b)c) = B(a,b)(Q(a,c) − R(c,b)Q(a)) = (Q(a,c) − Q(a)R(b,c))B(b,a).

The third identity follows by applying the second identity to an element d and then switching the roles of c and d. The fourth follows because

B(a,b)B(ab,c) = B(a,b)(IR(ab,c) + Q(ab)Q(c)) = IR(a,b + c) + Q(a) Q(b + c) = B(a,b+c).

In fact (a,b) is quasi-invertible if and only if a is quasi-invertible in the mutation Ab. Since this mutation might not necessarily unital this means that when an identity is adjoint 1 − a becomes invertible in Abk1. This condition can be expressed as follows without mentioning the mutation or homotope:

(a,b) is quasi-invertible if and only if there is an element c such that B(a,b)c = aQ(a)b and B(a,b)Q(c)b = Q(a)b. In this case c = ab.

In fact if (a,b) is quasi-invertible, then c = ab satisfies the first identity by definition. The second follows because B(a,b)Q(ab) = Q(a). Conversely the conditions state that in Abk1 the conditions imply that 1 + c is the inverse of 1 − a. On the other hand, ( 1 − a) ∘ x = B(a,b)x for x in Ab. Hence B(a,b) is invertible.

Equivalence relation

Let A be a finite-dimensional unital Jordan algebra over a field k of characteristic ≠ 2.[9] Two pairs (ai,bi) with ai invertible are said to be equivalent if (a1)−1b1 + b2 is invertible and a2 = (a1)b1b2.

This is an equivalence relation, since if a is invertible a0 = a so that a pair (a,b) is equivalent to itself. It is symmetric since from the definition a1 = (a2)b2b1. It is transitive. For suppose that (a3,b3) is a third pair with (a2)−1b2 + b3 invertible and a3 = (a2)b2b3. From the above

\displaystyle{a_1^{-1} - b_1 + b_3= (a_1^{-1} - b_1 +b_2) -b_2 + b_3=a_2^{-1} -b_2+b_3}

is invertible and

\displaystyle{a_3=a_2^{b_2-b_3}=(a_1^{b_1-b_2})^{b_2-b_3}=a_1^{b_1-b_3}.}

As for quasi-invertibility, this definition can be extended to the case where a and a−1b are not assumed to be invertible.

Two pairs (ai,bi) are said to be equivalent if (a1, b1b2) is quasi-invertible and a2 = (a1)b1b2. When k = R or C, the fact that this more general definition also gives an equivalence relation can deduced from the invertible case by continuity. For general k, it can also be verified directly:

\displaystyle{B(a_1,b_1-b_3)=B(a_1,b_1-b_2)B(a_2,b_2-b_3),}
so that (a1,b1b3) is quasi-invertible with
\displaystyle{a_3=a_2^{b_2-b_3}=(a_1^{b_1-b_2})^{b_2-b_3}=a_1^{b_1-b_3}.}

The equivalence class of (a,b) is denoted by (a:b).

Structure groups

Let A be a finite-dimensional complex semisimple unital Jordan algebra. If T is an operator on A, let Tt be its transpose with respect to the trace form. Thus L(a)t = L(a), Q(a)t = Q(a), R(a,b)t = R(b,a) and B(a,b)t = B(b,a). The structure group of A consists of g in GL(A) such that

\displaystyle{Q(ga)=gQ(a)g^t.}

They form a group Γ(A). The automorphism group Aut A of A consists of invertible complex linear operators g such that L(ga) = gL(a)g−1 and g1 = 1. Since an automorphism g preserves the trace form, g−1 = gt.

The complex Jordan algebra A is the complexification of a real Euclidean Jordan algebra E, for which the trace form defines an inner product. There is an associated involution aa* on A which gives rise to a complex inner product on A. The unitary structure group Γu(A) is the subgroup of Γ(A) consisting of unitary operators, so that Γu(A) = Γ(A) ∩ U(A). The identity component of Γu(A) is denoted by K. It is a connected closed subgroup of U(A).

The structure group Γ(A) acts naturally on X.[10] For g in Γ(A), set

\displaystyle{g(a,b)=(ga,(g^t)^{-1}b).}

Then (x,y) is quasi-invertible if and only if (gx,(gt)−1y) is quasi-invertible and

\displaystyle{g(x^y)=(gx)^{(g^t)^{-1}y}.}

In fact the covariance relations for g with Q and the inverse imply that

\displaystyle{gB(x,y)g^{-1}=B(gx,(g^t)^{-1}y)}

if x is invertible and so everywhere by density. In turn this implies the relation for the quasi-inverse. If a is invertible then Q(a) lies in Γ(A) and if (a,b) is quasi-invertible B(a,b) lies in Γ(A). So both types of operators act on X.

The defining relations for the structure group show that it is a closed subgroup of \mathfrak{g}_0 of GL(A). Since Q(ea) = e2L(a), the corresponding complex Lie algebra contains the operators L(a). The commutators [L(a),L(b)] span the complex Lie algebra of derivations of A. The operators R(a,b) = [L(a),L(b)] + L(ab) span \mathfrak{g}_0 and satisfy R(a,b)t = R(b,a) and [R(a,b),R(c,d)]=R(R(a,b)c,d) − R(c,R(b,a)d).

Geometric properties of quotient space

Let A be a finite-dimensional complex unital Jordan algebra which is semisimple, i.e. the trace form Tr L(ab) is non-degenerate. Let X be the quotient of A×A by the equivalence relation. Let Xb be the subset of X of classes (a:b). The map φb:XbA, (a:b) ↦ a is injective. A subset U of X is defined to be open if and only if UXb is open for all b.

The transition maps of the atlas with charts φb are given by

\displaystyle{\varphi_{cb}=\varphi_c\circ\varphi_b^{-1}:\varphi_b(X_b\cap X_c)\rightarrow \varphi_c(X_b\cap X_c).}

and are injective and holomorphic since

\displaystyle{\varphi_{cb}(a)=a^{b-c}}

with derivative

\displaystyle{\varphi_{cb}^\prime(a)=B(a,b-c)^{-1}.}

This defines the structure of a complex manifold on X because φdc ∘ φcb = φdb on φb(XbXcXd).

Given a finite set of points (ai:bi) in X, they are contained in a common Xb.

Indeed, all the polynomial functions pi(b) = det B(ai,bib) are non-trivial since pi(bi) = 1. Therefore there is a b such that pi(b) ≠ 0 for all i, which is precisely the criterion for (ai:bi) to lie in Xb.

X is compact.

Loos (1977) uses the Bergman operators to construct an explicit biholomorphism between X and a closed smooth algebraic subvariety of complex projective space.[11] This implies in particular that X is compact. There is a more direct proof of compactness using symmetry groups.

Given a Jordan frame (ei) in E, for every a in A there is a k in U = Γu(A) such that a=k(∑ αi ei) with αi ≥ 0 (and αi > 0 if a is invertible). In fact, if (a,b) is in X then it is equivalent to k(c,d) with c and d in the unital Jordan subalgebra Ae = ⊕ Cei, which is the complexification of Ee = ⊕ Rei. Let Z be the complex manifold constructed for Ae. Because Ae is a direct sum of copies of C, Z is just a product of Riemann spheres, one for each ei. In particular it is compact. There is a natural map of Z into X which is continuous. Let Y be the image of Z. It is compact and therefore coincides with the closure of Y0 = AeA = X0. The set UY is the continuous image of the compact set U × Y. It is therefore compact. On the other hand, UY0 = X0, so it contains a dense subset of X and must therefore coincide with X. So X is compact.

The above argument shows that every (a,b) in X is equivalent to k(c,d) with c and d in Ae and k in Γu(A). The mapping of Z into X is in fact an embedding. This is a consequence of (x,y) being quasi-invertible in Ae if and only if it is quasi-invertible in A. Indeed, if B(x,y) is injective on A, its restriction to Ae is also injective. Conversely, the two equations for the quasi-inverse in Ae imply that it is also a quasi-inverse in A.

Möbius transformations

Let A be a finite-dimensional complex semisimple unital Jordan algebra. The group SL(2,C) acts by Möbius transformation on the Riemann sphere C ∪ {∞}, the one-point compactification of C. If g in SL(2,C) is given by the matrix

\displaystyle{g=\begin{pmatrix}\alpha & \beta \\ \gamma & \delta\end{pmatrix},}

then

\displaystyle{g(z)=(\alpha z +\beta)(\gamma z +\delta)^{-1}.}

There is a generalization of this action of SL(2,C) to A and its compactification X. In order to define this action, note that SL(2,C) is generated by the three subgroups of lower and upper unitriangular matrices and the diagonal matrices. It is also generated by the lower (or upper) unitriangular matrices, the diagonal matrices and the matrix

\displaystyle{J=\begin{pmatrix}0 & 1 \\ -1 & 0\end{pmatrix}.}

The matrix J corresponds to the Möbius transformation j(z) = −z−1 and can be written

\displaystyle{J=\begin{pmatrix}1 & 0 \\ -1 & 1\end{pmatrix}\begin{pmatrix}1 & 1 \\ 0 & 1\end{pmatrix}
\begin{pmatrix}1 & 0 \\ -1 & 1\end{pmatrix}.}

The Möbius transformations fixing ∞ are just the upper triangular matrices. If g does not fix ∞, it sends ∞ to a finite point a. But then g can be composed with an upper unitriangular to send a to 0 and then with J to send 0 to infinity.

For an element a of A, the action of g in SL(2,C) is defined by the same formula

\displaystyle{g(a)=(\alpha a +\beta 1)(\gamma a +\delta 1)^{-1}.}

This defines an element of C[a] provided that γa + δ1 is invertible in A. The action is thus defined everywhere on A if g is upper triangular. On the other hand the action on X is simple to define for lower triangular matrices.[12]

\displaystyle{g(a:0)=(a:-\gamma)= (a^{-\gamma}:0)= (a(\gamma a +1 )^{-1}:0)}
if γa + 1 is invertible, so this is an extension of the Möbius action.

In fact on the invertible elements in A, the operator j(a) = −a−1 satisfies j(ga) = (gt)−1j(a). To define a biholomorphism j on X such that jg = (gt)−1j, it is enough to define these for (a:b) in some suitable orbit of Γ(A) or Γu(A). On the other hand, as indicated above, given a Jordan frame (ei) in E, for every a in A there is a k in U = Γu(A) such that a=k(∑ αi ei) with αi ≥ 0.

The computation of j in the associative commutative algebra Ae is straightforward since it is a direct product. For c = ∑ αi ei and d = ∑ βi ei, the Bergman operator on Ae has determinant det B(c,d) = ∏(1 − αiβi)2. In particular det B(c,d − λ) ≠ 0 for some λ ≠ 0. So that (c,d) is equivalent to (x,λ). Let μ = −λ−1. On A, for a dense set of a, the pair (a,λ) is equivalent to (b,0) with b invertible. Then (−b−1,0) is equivalent to (μ − μ2a,μ). Since a ↦ μ − μ2a is holomorphic it follows that j has a unique continuous extension to X such that jg = (gt)−1j for g in Γ(A), the extension is holomorphic and for λ ≠ 0, μ = −λ−1

\displaystyle{j(a,\lambda)=(\mu -\mu^2a,\mu).}

The holomorphic transformations corresponding to upper unitriangular matrices can be defined using the fact that they are the conjugates by J of lower unitriangular matrices, for which the action is already known. A direct algebraic construction is given in Dineen, Mackey & Mellon (1999).

This action of SL(2,C) is compatible with inclusions. More generally if e1, ..., em is a Jordan frame, there is an action of SL(2,C)m on Ae which extends to A. If c = ∑ γiei and b = ∑ βiei, then S(c) and T(b) give the action of the product of the lower and upper unitriangular matrices. If a = ∑ αiei is invertible, the corresponding product of diagonal matrices act as W = Q(a).[13] In particular the diagonal matrices give an action of (C*)m and Tm.

Holomorphic symmetry group

Let A be a finite-dimensional complex semisimple unital Jordan algebra. There is a transitive holomorphic action of a complex matrix group G on the compact complex manifold X. Koecher (1967) described G analogously to SL(2,C) in terms of generators and relations. G acts on the corresponding finite-dimensional Lie algebra of holomorphic vector fields restricted to X0 = A, so that G is realized as a closed matrix group. It is the complexification of a compact Lie group without center, so a semisimple algebraic group. The identity component H of the compact group acts transitively on X, so that X can be identified as a Hermitian symmetric space of compact type.[14]

The group G is generated by three types of holomorphic transformation on X:

The operators W normalize the group of operators Sc. Similarly the operator j normalizes the structure group, jW = (Wt)−1j. The operators Tc = jScj also form a group of holomorphic transformations isomorphic to the additive group of A. They generalize the upper unitriangular subgroup of SL(2,C). This group is normalized by the operators W of the structure group. The operator Tc acts on A as aa + c. If c is a scalar the operators Sc and Tc coincide with the operators corresponding to lower and upper unitriangular matrices in SL(2,C). Accordingly there is a relation j = S1T1S1 and PSL(2,C) is a subgroup of G. Loos (1977) defines the operators Tc in terms of the flow associated to a holomorphic vector field on X, while Dineen, Mackey & Mellon (1999) give a direct algebraic description.

G acts transitively on X.

Indeed SbTa(0:0) = (a:b).

Let G−1 and G+1 be the complex Abelian groups formed by the symmetries Tc and Sc respectively. Let G0 = Γ(A).

\displaystyle{G=G_0G_{+1}G_{-1}G_{+1}=G_0G_{-1}G_{+1}G_{-1}.}

The two expressions for G are equivalent as follows by conjugating by j.

For a invertible, Hua's identity can be rewritten

\displaystyle{Q(a)=T_a \circ j \circ T_{a^{-1}}\circ j \circ T_a \circ j.}

Moreover j = S1T1S1 and Sc = jTcj.[15]

The convariance relations show that the elements of G fall into sets G0G1, G0G1jG1, G0G1jG1jG1, G0G1jG1jG1jG1. ... The first expression for G follows once it is established that no new elements appear in the fourth or subsequent sets. For this it suffices to show that[16]

jG1jG1jG0 G1jG1jG1.

For then if there are three or more occurrences of j, the number can be recursively reduced to two. Given a, b in A, pick λ ≠ 0 so that c = a − λ and d = b − λ−1 are invertible. Then

\displaystyle{j T_a j  T_b  j =  j  T_c  T_\lambda  j  T_{\lambda^{-1}}  T_d \circ j = \lambda^2 j T_c  j  T_{-\lambda}  j T_d \ j = \lambda^2 T_{-c^{-1}} j Q(c^{-1})T_{-c^{-1} -\lambda - d^{-1}} j Q(d^{-1})jT_{-d^{-1}},}

which lies in G0G1j G1jG1.

The stabilizer of (0:0) in G is G0G−1.

It suffices to check that if SaTb(0:0) = (0:0), then b = 0. If so (b:0) = (0: −a) = (0:0), so b = 0.

Exchange relations

G is generated by G±1.

For a invertible, Hua's identity can be rewritten

\displaystyle{Q(a)=T_a \circ j \circ T_{a^{-1}}\circ j \circ T_a \circ j.}

Since j = S1T1S1, the operators Q(a) belong to the group generated by G±1.[17]

For quasi-invertible pairs (a,b), there are the "exchange relations"[18]

SbTa = TabB(a,b)−1Sba.

This identity shows that B(a,b) is in the group generated by G±1. Taking inverses, it is equivalent to the identity TaSb = SbaB(a,b)Tab.

To prove the exchange relations, it suffices to check that it valid when applied to points the dense set of points (c:0) in X for which (a+c,b) is quasi-invertible. It then reduces to the identity:

\displaystyle{(a+c)^b= a^b + B(a,b)^{-1}c^{(b^a)}.}

In fact, if (a,b) is quasi-invertible, then (a + c,b) is quasi-invertible if and only if (c,ba) is quasi-invertible. This follows because (x,y) is quasi-invertible if and only if (y,x) is. Moreover the above formula holds in this case.

For the proof, two more identities are required:

\displaystyle{B(c+b,a)=B(c,a^b)B(b,a)}
\displaystyle{R(a,b)=R(a^b,b-Q(b)a)=R(a-Q(a)b,b^a)}

The first follows from a previous identity by applying the transpose. For the second, because of the transpose, it suffices to prove the first equality. Setting c = bQ(b)a in the identity B(a,b)R(ab,c) = R(a,c) − Q(a)Q(b,c) yields

B(a,b)R(ab,bQ(b)c) = B(a,b)R(a,b),

so the identity follows by cancelling B(a,b).

To prove the formula, the relations (a + c)b = B(a,c)−1(a + cQ(a + c)b) and ab + B(a,b)−1c(ba) = B(a + c,b)−1(B(c,ba) (aQ(a)b) + cQ(c)ba) show that it is enough to prove that

a + cQ(a + c)b = B(c,ba) (aQ(a)b) + cQ(c)ba.

Indeed B(c,ba) (aQ(a)b) + cQ(c)ba = a + cQ(a)b + 2R(c,ba)(aQ(a)b) − Q(c)[ baQ(ba)(aQ(a)b)]. On the other hand 2R(c,ba)(aQ(a)b) = 2R(c,aQ(a)b)ba = R(a,b)c = 2Q(a,c)b and baQ(ba)(aQ(a)b) = baQ(b)B(a,b)−1(aQ(a)b) = baQ(b)ab = b. So B(c,ba) (aQ(a)b) + cQ(c)ba = a + cQ(a)b − 2Q(a,c)bQ(c)b = a + cQ(a + c)b.

Now set Ω = G+1G0G−1. Then the exchange relations imply that Sb Ta lies in Ω if and only if (a,b) is quasi-invertible; and that g lies in Ω if and only if g(0:0) is in X0.[19]

In fact if Sb Ta lies in Ω, then (a,b) is equivalent to (x,0), so it a quasi-invertible pair; the converse follows from the exchange relations. Clearly Ω(0:0) = G1(0:0) = X0. The converse follows from G = G−1G1 G0G−1 and the criterion for Sb Ta to lie in Ω.

Lie algebra of holomorphic vector fields

The compact complex manifold X is modelled on the space A. The derivatives of the transition maps describe the tangent bundle through holomorphic transition functions Fbc:XbXc → GL(A). These are given by Fbc(a,b) = B(a,bc), so the structure group of the corresponding principal fiber bundle reduces to Γ(A), the structure group of A.[20] The corresponding holomorphic vector bundle with fibre A is the tangent bundle of the complex manifold X. Its holomorphic sections are just holomorphic vector fields on X. They can be determined directly using the fact that they must be invariant under the natural adjoint action of the known holomorphic symmetries of X. They form a finite-dimensional complex semisimple Lie algebra. The restriction of these vector fields to X0 can be described explicitly. A direct consequence of this description is that the Lie algebra is three-graded and that the group of holomorphic symmetries of X, described by generators and relations in Koecher (1967) and Loos (1979), is a complex linear semisimple algebraic group that coincides with the group of biholomorphisms of X.

The Lie algebras of the three subgroups of holomorphic automorphisms of X give rise to linear spaces of holomorphic vector fields on X and hence X0 = A.

Let

\displaystyle{\mathfrak{g}=\mathfrak{g}_{-1}\oplus\mathfrak{g}_0\oplus\mathfrak{g}_1.}

Then, defining \mathfrak{g}_i = (0) for i ≠ −1, 0, 1, \mathfrak{g} forms a complex Lie algebra with

\displaystyle{[\mathfrak{g}_p,\mathfrak{g}_q]\subseteq \mathfrak{g}_{p+q}}.

This gives the structure of a 3-graded Lie algebra. For elements (a,T,b) in \mathfrak{g}, the Lie bracket is given by

\displaystyle{[(a_1,T_1,b_1),(a_2,T_2,b_2)]=(T_1a_2-T_2a_1,[T_1,T_2]+R(a_1,b_2)-R(a_2,b_1),T_2^tb_1-T_1^tb_2)}

The group PSL(2,C) of Möbius transformations of X normalizes the Lie algebra \mathfrak{g}. The transformation j(z) = −z−1 corresponding to the Weyl group element J induces the involutive automorphism σ given by

\displaystyle{\sigma(a,T,b)=(b,-T^t,a).}

More generally the action of a Möbius transformation

\displaystyle{g=\begin{pmatrix}\alpha & \beta \\ \gamma & \delta\end{pmatrix}}

can be described explicitly. In terms of generators diagonal matrices act as

\displaystyle{\begin{pmatrix}\alpha & 0 \\ 0 & \alpha^{-1}\end{pmatrix}(a,T,b)=(\alpha^2 a,T,\alpha^{-2}b),}

upper unitriangular matrices act as

\displaystyle{\begin{pmatrix}1 & \beta \\ 0 & 1\end{pmatrix}(a,T,b)=(a +\beta T(1) -\beta^2 b,T - \beta L(a),b),}

and lower unitriangular matrices act as

\displaystyle{\begin{pmatrix}1 & 0 \\ \gamma & 1\end{pmatrix}(a,T,b)=(a,T-\gamma L(b),b-\gamma T^t(1) -\gamma^2 a).}

This can be written uniformly in matrix notation as

\displaystyle{\begin{pmatrix}g(T) & g(a) \\ g(b) & g(T)^t\end{pmatrix}= g \begin{pmatrix}T & a \\ b & T^t\end{pmatrix} g^{-1}.}

In particular the grading corresponds to the action of the diagonal subgroup of SL(2,C), even with |α| = 1, so a copy of T.

The Killing form is given by

\displaystyle{\mathbf{{B}}((a_1,T_1,b_1),(a_2,T_2,b_2))= (a_1,b_2) + (b_1,a_2) +  \beta(T_1,T_2),}

where β(T1,T2) is the symmetric bilinear form defined by

\displaystyle{\beta(R(a,b),R(c,d))=(R(a,b)c,d)=(R(c,d)a,b),}

with the bilinear form (a,b) corresponding to the trace form: (a,b) = Tr L(ab).

More generally the generators of the group G act by automorphisms on \mathfrak{g} as

The Killing form is nondegenerate on \mathfrak{g}.

The nondegeneracy of the Killing form is immediate from the explicit formula. By Cartan's criterion, \mathfrak{g} is semisimple. In the next section the group G is realized as the complexification of a connected compact Lie group H with trivial center, so semisimple. This gives a direct means to verify semisimplicity. The group H also acts transitively on X.

\mathfrak{g} is the Lie algebra of all holomorphic vector fields on X.

To prove that \mathfrak{g} exhausts the holomorphic vector fields on X, note the group T acts on holomorphic vector fields. The restriction of such a vector field to X0 = A gives a holomorphic map of A into A. The power series expansion around 0 is a convergent sum of homogeneous parts of degree m ≥ 0. The action of T scales the part of degree m by α2m − 2. By taking Fourier coefficients with respect to T, the part of degree m is also a holomorphic vector field. Since conjugation by J gives the inverse on T, it follows that the only possible degrees are 0, 1 and 2. Degree 0 is accounted for by the constant fields. Since conjugation by J interchanges degree 0 and degree 2, it follows that \mathfrak{g}_{\pm 1} account for all these holomorphic vector fields. Any further holomorphic vector field would have to appear in degree 1 and so would have the form aMa for some M in End A. Conjugation by J would give another such map N. Moreover etM(a,0,0)= (etMa,0,0). But then

\displaystyle{e^{tM}(0,0,b)=Je^{tN}J(0,0,b)=Je^{tN}(b,0,0)=(0,0,e^{tN}b).}

Set Ut = etM and Vt = etB. Then

\displaystyle{Q(U_ta)b = U_tQ(a)V_{-t}b.}

It follows that Ut lies in Γ(A) for all t and hence that M lies in \mathfrak{g}_0. So \mathfrak{g} is exactly the space of holomorphic vector fields on X.

Compact real form

The action of G on \mathfrak{g} is faithful.

Suppose g = WTxSy Tz acts trivially on \mathfrak{g}. Then Sy Tz must leave the subalgebra (0,0,A) invariant. Hence so must Sy. This forces y = 0, so that g = WTx + z. But then Tx+z must leave the subalgebra (A,0,0) invariant, so that x + z = 0 and g = W. If W acts trivially, W = I.[21]

The group G can thus be identified with its image in GL \mathfrak{g}.

Let A = E + iE be the complexification of a Euclidean Jordan algebra E. For a = x + iy, set a* = xiy. The trace form on E defines a complex inner product on A and hence an adjoint operation. The unitary structure group Γu(A) consists of those g in Γ(A) that are in U(A), i.e. satisfy gg*=g*g = I. It ia a closed subgroup of U(A). Its Lie algebra consists of the skew-adjoint elements in \mathfrak{g}_0. Define a conjugate linear involution θ on \mathfrak{g} by

\displaystyle{\theta(a,T,b)=(b^*,-T^*, a^*).}

This is a period 2 conjugate-linear automorphism of the Lie algebra. It induces an automorphism of G, which on the generators is given by

\displaystyle{\theta(S_a)=T_{a^*},\,\,\, \theta(j)=j,\,\,\, \theta(T_b)=S_{b^*},\,\,\, \theta(W)=(W^*)^{-1}.}

Let H be the fixed point subgroup of θ in G. Let \mathfrak{h} be the fixed point subalgebra of θ in \mathfrak{g}. Define a sesquilinear form on \mathfrak{g} by (a,b) = −B(a,θ(b)). This defines a complex inner product on \mathfrak{g} which restricts to a real inner product on \mathfrak{h}. Both are preserved by H. Let K be the identity component of Γu(A). It lies in H. Let Ke = Tm be the diagonal torus associated with a Jordan frame in E. The action of SL(2,C)m is compatible with θ which sends a unimodular matrix \begin{pmatrix} \alpha & \beta\\ \gamma & \delta\end{pmatrix} to \begin{pmatrix} \overline{\delta} & -\overline{\gamma}\\ -\overline{\beta} & \overline{\alpha}\end{pmatrix}. In particular this gives a homomorphism of SU(2)m into H.

Now every matrix M in SU(2) can be written as a product

\displaystyle{M = \begin{pmatrix} \zeta_1 & 0\\ 0 & \zeta_1^{-1}\end{pmatrix}
\begin{pmatrix} \cos \varphi & \sin \varphi\\ -\sin \varphi & \cos \varphi\end{pmatrix}
\begin{pmatrix} \zeta_2 & 0\\ 0 & \zeta_2^{-1}\end{pmatrix}.}

The factor in the middle gives another maximal torus in SU(2) obtained by conjugating by J. If a = ∑ αiei with |αi| = 1, then Q(a) gives the action of the diagonal torus T = Tm and corresponds to an element of KH. The element J lies in SU(2)m and its image is a Möbius transformation j lying in H. Thus S = jTj is another torus in H and TST coincides with the image of SU(2)m.

H acts transitively on X. The stabilizer of (0:0) is K. Furthermore H = KSK, so that H is a connected closed subgroup of the unitary group on \mathfrak{g}. Its Lie algebra is \mathfrak{h}.

Since Z = SU(2)m(0:0) for the compact complex manifold corresponding to Ae, if follows that Y = T S (0:0), where Y is the image of Z. On the other hand X = KY, so that X = KTS(0:0) = KS(0:0). On the other hand the stabilizer of (0:0) in H is K, since the fixed point subgroup of G0G−1 under θ is K. Hence H = KSK. In particular H is compact and connected since both K and S are. Because it is a closed subgroup of U \mathfrak{g}, it is a Lie group. It contains K and hence its Lie algebra contains the operators (0,T,0) with T* = −T. It contains the image of SU(2)m and hence the elements (a,0,a*) with a in Ae. Since A = KAe and (kt)−1(a*) = (ka)*, it follows that the Lie algebra \mathfrak{h}_1 of H contains (a,0,a*) for all a in A. Thus it contains \mathfrak{h}.

They are equal because all skew-adjoint derivations of \mathfrak{h} are inner. In fact, since H normalizes \mathfrak{h} and the action by conjugation is faithful, the map of \mathfrak{h}_1 into the Lie algebra \mathfrak{d} of derivations of \mathfrak{h} is faithful. In particular \mathfrak{h} has trivial center. To show that \mathfrak{h} equals \mathfrak{h}_1, it suffices to show that \mathfrak{d} coincides with \mathfrak{h}. Derivations on \mathfrak{h} are skew-adjoint for the inner product given by minus the Killing form. Take the invariant inner product on \mathfrak{d} given by −Tr D1D2. Since \mathfrak{h} is invariant under \mathfrak{d} so is its orthogonal complement. They are both ideals in \mathfrak{d}, so the Lie bracket between them must vanjsh. But then any derivation in the orthogonal complement would have 0 Lie bracket with \mathfrak{h}, so must be zero. Hence \mathfrak{h} is the Lie algebra of H. (This also follows from a dimension count since dim X = dim H − dim K.)

G is isomorphic to a closed subgroup of the general linear group on \mathfrak{g}.

The formulas above for the action of W and Sy show that the image of G0G−1 is closed in GL \mathfrak{g}. Since H acts transitively on X and the stabilizer of (0:0) in G is G0G−1, it follows that G = HG0G−1. The compactness of H and closedness of G0G−1 implies that G is closed in GL \mathfrak{g}.

G is a connected complex Lie group with Lie algebra \mathfrak{g}. It is the complexification of H.

G is a closed subgroup of GL \mathfrak{g} so a real Lie group. Since it contains Gi with i = 0 or ±1, its Lie algebra contains \mathfrak{g}. Since \mathfrak{g} is the complexification of \mathfrak{h}, like \mathfrak{h} all its derivations are inner and it has trivial center. Since the Lie algebra of G normalizes \mathfrak{g} and o is the only element centralizing \mathfrak{g}, as in the compact case the Lie algebra of G must be \mathfrak{g}. (This can also be seen by a dimension count since dim X = dim G − dim G0G−1.) Since it is a complex subspace, G is a complex Lie group. It is connected because it is the continuous image of the connected set H × G0G−1. Since \mathfrak{g} is the complexification of \mathfrak{h}, G is the complexification of H.

Noncompact real form

For a in A the spectral norm ||a|| is defined to be max αi if a = u ∑ αiei with αi ≥ 0 and u in K. It is independent of choices and defines a norm on A. Let D be the set of a with ||a|| < 1 and let H* be the identity component of the closed subgroup of G carrying D onto itself. It is generated by K, the Möbius transformations in PSU(1,1) and the image of SU(1,1)m corresponding to a Jordan frame. Let τ be the conjugate-linear period 2 automorphism of \mathfrak{g} defined by

\displaystyle{\tau(a,T,b)=(-a^*,-T^*,-b^*).}

Let \mathfrak{h}^* be the fixed point algebra of τ. It is the Lie algebra of H*. It induces a period 2 automorphism of G with fixed point subgroup H*. The group H* acts transitively on D. The stabilizer of 0 is K.[22]

The noncompact real semisimple Lie group H* acts on X with an open orbit D. As with the action of SU(1,1) on the Riemann sphere, it has only finitely many orbits. This orbit structure can be explicitly described when the Jordan algebra A is simple. Let X0(r,s) be the subset of A consisting of elements a = u ∑ αiai with exactly r of the αi less than one and exactly s of them greater than one. Thus 0 ≤ r + sm. These sets are the intersections of the orbits X(r,s) of H* with X0. The orbits with r + s = m are open. There is a unique compact orbit X(0,0). It is the Shilov boundary S of D consisting of elements eix with x in E, the underlying Euclidean Jordan algebra. X(p,q) is in the closure of X(r,s) if and only if pr and qs. In particular S is in the closure of every orbit.[23]

Jordan algebras with involution

The preceding theory describes irreducible Hermitian symmetric spaces of tube type in terms of unital Jordan algebras. In Loos (1977) general Hermitian symmetric spaces are described by a systematic extension of the above theory to Jordan pairs. In the development of Koecher (1969), however, irreducible Hermitian symmetric spaces not of tube type are described in terms of period two automorphisms of simple Euclidean Jordan algebras. In fact any period 2 automorphism defines a Jordan pair: the general results of Loos (1977) on Jordan pairs can be specialized to that setting.

Let τ be a period two automorphism of a simple Euclidean Jordan algebra E with complexification A. There are corresponding decompositions E = E+E and A = A+A into ±1 eigenspaces of τ. Let VAτ = A. τ is assumed to satisfy the additional condition that the trace form on V defines an inner product. For a in V, define Qτ(a) to be the restriction of Q(a) to V. For a pair (a,b) in V2, define Bτ(a,b) and Rτ(a,b) to be the restriction of B(a,b) and R(a,b) to V. Then V is simple if and only if the only subspaces invariant under all the operators Qτ(a) and Rτ(a,b) are (0) and V.

The conditions for quasi-invertibility in A show that Bτ(a,b) is invertible if and only if B(a,b) is invertible. The quasi-inverse ab is the same whether computed in A or V. A space of equivalence classes Xτ can be defined on pairs V2. It is a closed subspace of X, so compact. It also has the structure of a complex manifold, modelled on V. The structure group Γ(V) can be defined in terms of Qτ and it has as a subgroup the unitary structure group Γu(V) = Γ(V) ∩ U(V) with identity component Kτ. The group Kτ is the identity component of the fixed point subgroup of τ in K. Let Gτ be the group of biholomorphisms of Xτ generated by W in Gτ,0, the identity component of Γ(V), and the Abelian groups Gτ,−1 consisting of the Sa and Gτ,+1 consisting of the Tb with a and b in V. It acts transitively on Xτ with stabilizer Gτ,0Gτ,−1 and Gτ = Gτ,0Gτ,−1Gτ,+1Gτ,−1. The Lie algebra \mathfrak{g}_\tau of holomorphic vector fields on Xτ is a 3-graded Lie algebra,

\displaystyle{\mathfrak{g}_\tau = \mathfrak{g}_{\tau,+1} \oplus \mathfrak{g}_{\tau,0} \oplus
\mathfrak{g}_{\tau,-1}.}

Restricted to V the components are generated as before by the constant functions into V, by the operators Rτ(a,b) and by the operators Qτ(a). The Lie brackets are given by exactly the same formula as before.

The spectral decomposition in Eτ and V is accomplished using tripotents, i.e. elements e such that e3 = e. In this case f = e2 is an idempotent in E+. There is a Pierce decomposition E = E0(f) ⊕ E½(f) ⊕ E1(f) into eigenspaces of L(f). The operators L(e) and L(f) commute, so L(e) leaves the eigenspaces above invariant. In fact L(e)2 acts as 0 on E0(f), as 1/4 on E½(f) and 1 on E1(f). This induces a Pierce decomposition Eτ = Eτ,0(f) ⊕ Eτ,½(f) ⊕ Eτ,1(f). The subspace Eτ,1(f) becomes a Euclidean Jordan algebra with unit f under the mutation Jordan product xy = {x,e,y}.

Two tripotents e1 and e2 are said to be orthogonal if all the operators [L(a),L(b)] = 0 when a and b are powers of e1 and e2 and if the corresponding idempotents f1 and f2 are orthogonal. In this case e1 and e2 generate a commutative associative algebra and e1e2 = 0, since (e1e2,e1e2) =(f1,f2) =0. Let a be in Eτ. Let F be the finite-dimensional real subspace spanned by odd powers of a. The commuting self-adjoint operators L(x)L(y) with x, y odd powers of a act on F, so can be simultaneously diagonalized by an orthonormal basis ei. Since (ei)3 is a positive multiple of ei, rescaling if necessary, ei can be chosen to be a tripotent. They form an orthogonal family by construction. Since a is in F, it can be written a = ∑ αi ei with αi real. These are called the eigenvalues of a (with respect to τ). Any other tripotent e in F has the form a = ∑ εi ei with εi = 0, ±1, so the ei are up to sign the minimal tripotents in F.

A maximal family of orthogonal tripotents in Eτ is called a Jordan frame. The tripotents are necessarily minimal. All Jordan frames have the same number of elements, called the rank of Eτ. Any two frames are related by an element in the subgroup of the structure group of Eτ preserving the trace form. For a given Jordan frame (ei), any element a in V can be written in the form a = u ∑ αi ei with αi ≥ 0 and u an operator in Kτ. The spectral norm of a is defined by ||a|| = sup αi and is independent of choices. Its square equals the operator norm of Qτ(a). Thus V becomes a complex normed space with open unit ball Dτ.

Note that for x in E, the operator Q(x) is self-adjoint so that the norm ||Q(x)n|| = ||Q(x)||n. Since Q(x)n = Q(xn), it follows that ||xn|| = ||x||n. In particular the spectral norm of x = ∑ αi ei in A is the square root of the spectral norm of x2 = ∑ (αi)2 fi. It follows that the spectral norm of x is the same whether calculated in A or Aτ. Since Kτ preserves both norms, the spectral norm on Aτ is obtained by restricting the spectral norm on A.

For a Jordan frame e1, ..., em, let Ve = ⊕ C ei. There is an action of SL(2,C)m on Ve which extends to V. If c = ∑ γiei and b = ∑ βiei, then S(c) and T(b) give the action of the product of the lower and upper unitriangular matrices. If a = ∑ αiei with αi ≠ 0, then the corresponding product of diagonal matrices act as W = Bτ(a, ea), where e = ∑ ei.[24] In particular the diagonal matrices give an action of (C*)m and Tm.

As in the case without an automorphism τ, there is an automorphism θ of Gτ. The same arguments show that the fixed point subgroup Hτ is generated by Kτ and the image of SU(2)m. It is a compact connected Lie group. It acts transitively on Xτ; the stabilizer of (0:0) is Kτ. Thus Xτ = Hτ/Kτ, a Hermitian symmetric space of compact type.

Let Hτ* be the identity component of the closed subgroup of Gτ carrying Dτ onto itself. It is generated by Kτ and the image of SU(1,1)m corresponding to a Jordan frame. Let ρ be the conjugate-linear period 2 automorphism of \mathfrak{g}_\tau defined by

\displaystyle{\rho(a,T,b)=(-a^*,-T^*,-b^*).}

Let \mathfrak{h}^*_\tau be the fixed point algebra of ρ. It is the Lie algebra of Hτ*. It induces a period 2 automorphism of G with fixed point subgroup Hτ*. The group Hτ* acts transitively on Dτ. The stabilizer of 0 is Kτ*.[25] Hτ*/Kτ is the Hermitian symmetric space of noncompact type dual to Hτ/Kτ.

The Hermitian symmetric space of non-compact type have an unbounded realization, analogous the upper half-plane in C. Möbius transformations in PSL(2,C) corresponding to the Cayley transform and its inverse give biholomorphisms of the Riemann sphere exchanging the unit disk and the upper halfplane. When the Hermitian symmetric space is of tube type the same Möbius transformations map the disk D in A onto the tube domain T = E + iC were C is the open self-dual convex cone of squares in the Euclidean Jordan algebra E.

For Hermitian symmetric space not of tube type there is no action of PSL(2,C) on X, so no analogous Cayley transform. A partial Cayley transform can be defined in that case for any given maximal tripotent e = ∑ εi ei in Eτ. It takes the disk Dτ in Aτ = Aτ,1(f) ⊕ Aτ,½(f) onto a Siegel domain of the second kind.

In this case Eτ,1(f) is a Euclidean Jordan algebra and there is symmetric Eτ,1(f)-valued bilinear form on Eτ,½(f) such that the corresponding quadratic form q takes values in its positive cone Cτ. The Siegel domain consists of pairs (x + iy,u + iv) such that yq(u) − q(v) lies in Cτ. The quadratic form q on Eτ,½(f) and the squaring operation on Eτ,1(f) are given by xQτ(x)e. The positive cone Cτ corresponds to x with Qτ(x) invertible.[26]

Examples

For simple Euclidean Jordan algebras E with complexication A, the Hermitian symmetric spaces of compact type X can be described explicitly as follows, using Cartan's classification.[27]

Type In. A is the Jordan algebra of n × n complex matrices Mn(C) with the operator Jordan product xy = ½(xy + yx). It is the complexification of E = Hn(C), the Euclidean Jordan algebra of self-adjoint n × n complex matrices. In this case G = PSL(2n,C) acting on A with g=\begin{pmatrix}a & b \\ c & d\end{pmatrix} acting as g(z) = (az + b)(cz + d)−1. Indeed this can be verified directly for diagonal, upper and lower unitriangular matrices which correspond to the operators W, Sc and Tb. The subset Ω corresponds to the matrices g with d invertible. In fact consider the space of linear maps from Cn to C2n = CnCn. It is described by a pair (T1|T2) with Ti in Mn(C). This is a module for GL(2n,C) acting on the target space. There is also an action of GL(n,C) induced by the action on the source space. The space of injective maps U is invariant and GL(n,C) acts freely on it. The quotient is the Grassmannian M consisting of n-dimensional subspaces of C2n. Define a map of A2 into M by sending (a,b) to the injective map (a|Ibta). This map induces an isomorphism of X onto M.

In fact let V be an n-dimensional subspace of CnCn. If it is in general position, i.e. it and its orthogonal complement have trivial intersection with Cn ⊕ (0) and (0) ⊕ Cn, it is the graph of an invertible operator T. So the image corresponds to (a|Ibta) with a = I and bt = IT.

At the other extreme, V and its orthogonal complement U can be written as orthogonal sums V = V1V2, U = U1U2, where V1 and U1 are the intersections with Cn ⊕ (0) and V2 and U2 with (0) ⊕ Cn. Then dim V1 = dim U2 and dim V2 = dim U1. Moreover Cn ⊕ (0) = V1U1 and (0) ⊕ Cn = V2U2. The subspace V corresponds to the pair (e|Ie), where e is the orthogonal projection of Cn ⊕ (0) onto V1. So a = e and b = I.

The general case is a direct sum of these two cases. V can be written as an orthogonal sum V = V0V1V2 where V1 and V2 are the intersections with Cn ⊕ (0) and (0) ⊕ Cn and V0 is their orthogonal complement in V. Similarly the orthogonal complement U of V can be written U = U0U1U2. Thus Cn ⊕ (0) = V1U1W1 and (0) ⊕ Cn = V2U2W2, where Wi are orthogonal complements. The direct sum (V1U1) ⊕ (V2U2) ⊆ CnCn is of the second kind and its orthogonal complement of the first.

Maps W in the structure group correspond to h in GL(n,C), with W(a) = haht. The corresponding map on M sends (x|y) to (hx|(ht)−1y). Similarly the map corresponding to Sc sends (x|y) to (x|y + c), the map corresponding to Tb sends (x|y) to (x + b|y) and the map corresponding to J sends (x|y) to (y|x). It follows that the map corresponding to g sends (x|y) to (ax + by|cx + dy). On the other hand if y is invertible, (x|y) is equivalent to (xy−1|I), whence the formula for the fractional linear transformation.

Type IIIn. A is the Jordan algebra of n × n symmetric complex matrices Sn(C) with the operator Jordan product xy = ½(xy + yx). It is the complexification of E = Hn(R), the Euclidean Jordan algebra of n × n symmetric real matrices. On C2n = CnCn, define a nondegenerate alternating bilinear form by ω(x1y1, x2y2) = x1y2y1x2. In matrix notation if J=\begin{pmatrix} 0 & I \\ -I & 0\end{pmatrix},

\displaystyle{\omega(z_1,z_2)=zJz^t.}

Let Sp(2n,C) denote the complex symplectic group, the subgroup of GL(2n,C) preserving ω. It consists of g such that gJgt = J and is closed under ggt. If g=\begin{pmatrix}a & b \\ c & d\end{pmatrix} belongs to Sp(2n,C) then

\displaystyle{g^{-1}=\begin{pmatrix}d^t & -c^t \\ -b^t & a^t\end{pmatrix}.}

It has center I}. In this case G = Sp(2n,C)/{±I} acting on A as g(z) = (az + b)(cz + d)−1. Indeed this can be verified directly for diagonal, upper and lower unitriangular matrices which correspond to the operators W, Sc and Tb. The subset Ω corresponds to the matrices g with d invertible. In fact consider the space of linear maps from Cn to C2n = CnCn. It is described by a pair (T1|T2) with Ti in Mn(C). This is a module for Sp(2n,C) acting on the target space. There is also an action of GL(n,C) induced by the action on the source space. The space of injective maps U with isotropic image, i.e. ω vanishes on the image, is invariant. Moreover GL(n,C) acts freely on it. The quotient is the symplectic Grassmannian M consisting of n-dimensional Lagrangian subspaces of C2n. Define a map of A2 into M by sending (a,b) to the injective map (a|Iba). This map induces an isomorphism of X onto M.

In fact let V be an n-dimensional Lagrangian subspace of CnCn. Let U be a Lagrangian subspace complementing V. If they are in general position, i.e. they have trivial intersection with Cn ⊕ (0) and (0) ⊕ Cn, than V is the graph of an invertible operator T with Tt = T. So the image corresponds to (a|Iba) with a = I and b = IT.

At the other extreme, V and U can be written as direct sums V = V1V2, U = U1U2, where V1 and U1 are the intersections with Cn ⊕ (0) and V2 and U2 with (0) ⊕ Cn. Then dim V1 = dim U2 and dim V2 = dim U1. Moreover Cn ⊕ (0) = V1U1 and (0) ⊕ Cn = V2U2. The subspace V corresponds to the pair (e|Ie), where e is the projection of Cn ⊕ (0) onto V1. Note that the pair (Cn ⊕ (0), (0) ⊕ Cn) is in duality with respect to ω and the identification between them induces the canonical symmetric bilinear form on Cn. In particular V1 is identified with U2 and V2 with U1. Moreover they are V1 and U1 are orthogonal with respect to the symmetric bilinear form on (Cn ⊕ (0). Hence the idempotent e satisfies et = e. So a = e and b = I lie in A and V is the image of (a|Iba).

The general case is a direct sum of these two cases. V can be written as a direct sum V = V0V1V2 where V1 and V2 are the intersections with Cn ⊕ (0) and (0) ⊕ Cn and V0 is a complement in V. Similarly U can be written U = U0U1U2. Thus Cn ⊕ (0) = V1U1W1 and (0) ⊕ Cn = V2U2W2, where Wi are complements. The direct sum (V1U1) ⊕ (V2U2) ⊆ CnCn is of the second kind. It has a complement of the first kind.

Maps W in the structure group correspond to h in GL(n,C), with W(a) = haht. The corresponding map on M sends (x|y) to (hx|(ht)−1y). Similarly the map corresponding to Sc sends (x|y) to (x|y + c), the map corresponding to Tb sends (x|y) to (x + b|y) and the map corresponding to J sends (x|y) to (y|x). It follows that the map corresponding to g sends (x|y) to (ax + by|cx + dy). On the other hand if y is invertible, (x|y) is equivalent to (xy−1|I), whence the formula for the fractional linear transformation.

Type II2n. A is the Jordan algebra of 2n × 2n skew-symmetric complex matrices An(C) and Jordan product xy = −½(x J y + y J x) where the unit is given by J=\begin{pmatrix}0 & I \\ -I & 0\end{pmatrix}. It is the complexification of E = Hn(H), the Euclidean Jordan algebra of self-adjoint n × n matrices with entries in the quaternions. This is discussed in Loos (1977) and Koecher (1969).

Type IVn. A is the Jordan algebra CnC with Jordan product (x,α) ∘ (y,β) = (βx + αy,αβ + xy). It is the complexication of the rank 2 Euclidean Jordan algebra defined by the same formulas but with real coefficients. This is discussed in Loos (1977).

Type VI. The complexified Albert algebra. This is discussed in Faulkner (1972), Loos (1978) and Drucker (1981).

The Hermitian symmetric spaces of compact type X for simple Euclidean Jordan algebras E with period two automorphism can be described explicitly as follows, using Cartan's classification.[28]

Type Ip,q. Let F be the space of q by p matrices over R with pq. This corresponds to the automorphism of E = Hp + q(R) given by conjugating by the diagonal matrix with p diagonal entries equal to 1 and q to −1. Without loss of generality p can be taken greater than q. The structure is given by the triple product xytz. The space X can be identified with the Grassmannian of p-dimensional subspace of Cp + q = CpCq. This has a natural embedding in C2p = CpCp by adding 0's in the last pq coordinates. Since any p-dimensional subspace of C2p can be represented in the form [Iytx|x], the same is true for subspaces lying in Cp + q. The last pq rows of x must vanish and the mapping does not change if the last pq rows of y are set equal to zero. So a similar representation holds for mappings, but now with q by p matrices. Exactly as when p = q, it follows that there is an action of GL(p + q, C) by fractional linear transformations.[29]

Type IIn F is the space of real skew-symmetric m by m matrices. After removing a factor of √(-1), this corresponds to the period 2 automorphism given by complex conjugation on E = Hn(C).

Type V. F is the direct sum of two copies of the Cayley numbers, regarded as 1 by 2 matrices. This corresponds to the canonical period 2 automorphism defined by any minimal idempotent in E = H3(O).

See also

Notes

  1. See:
  2. See:
  3. See:
  4. See:
  5. See:
  6. McCrimmon 1978, pp. 616–617
  7. Loos 1975, pp. 20–22
  8. In the main application in Loos (1977), A is finite dimensional. In that case invertibility of operators on A is equivalent to injectivity or surjectivity. The general case is treated in Loos (1975) and McCrimmond (2004).
  9. Loos 1977
  10. Loos & 77 8.3-8.4
  11. Loos 1977, p. 7.1−7.15
  12. See:
  13. Loos 1977, pp. 9.4–9.5
  14. See:
  15. Koecher 1967, p. 144
  16. Koecher 1967, p. 145
  17. Koecher 1967, p. 144
  18. Loos 1977, p. 8.9-8.10
  19. Loos 1977
  20. See:
  21. Koecher 1967, p. 164
  22. See:
  23. See:
  24. Loos 1977, pp. 9.4–9.5
  25. See:
  26. Loos 1977, pp. 10.1–10.13
  27. Loos 1978, pp. 125–128
  28. Koecher 1969
  29. See:

References

This article is issued from Wikipedia - version of the Thursday, September 17, 2015. The text is available under the Creative Commons Attribution/Share Alike but additional terms may apply for the media files.