Henceforth F denotes a J*-triple spanned by a non-nil weighted grid G:={gw: w∈W} with non-degenerate weight figure W. For short we write λef := [λ ∈ {0,±12,±1} : (e e∗)f =λf] fore, f ∈G. We use also the direct notation eRf (e, f ∈G, R = ⊤, ⊥,⊢ ,⊣,≈) for the COG relations of the elements of G(defined in terms of (8.2.1)).
Proposition 8.5.1. Let (wk : k ∈ Z) be an arithmetic sequence in W and suppose u, v ∈W with u−v = N(w1−w0). Then with the notations e := gu, f :=gv, an := gwn (n∈Z) and F0 := Spann∈Zan, for some ξ ∈C we have
e f∗|F0 =ξaN a∗0|F0 , f e∗|F0 =ξa0 a∗N|F0 (n∈Z) . Proof. By setting σ := sgn(a0) and ε:= sgn(a0)sgn(a1), we may assume
ak−ℓ+m =σεℓ{akaℓ∗am} (k, ℓ, m∈Z) , e f∗ :an7→ξnan+N , f e∗ :an7→ηnan−N for suitable coefficients ξn, ηn∈C (n∈Z). Thus
{ef∗{akaℓ∗am}} = {{ef∗ak}aℓ∗am} − {ak{f e∗aℓ}∗am}+{akaℓ∗{ef∗am}}
σεℓ{ef∗ak−ℓ+m} = ξk{ak+Naℓ∗am} −ηℓ{akaℓ−N∗am}+ξm{akaℓ∗am+N} ξk−ℓ+mσεℓak−ℓ+m+N = [ξkσεℓ−ηℓσεℓ−N +ξmσεℓ]ak−ℓ+m+N
ξk+ξm−ξk−ℓ+m = ηℓεN (k, ℓ, m∈Z) .
In particular (withℓ:= 0) we haveξk+ξm−ξk+m =η0εN (k, m∈Z). That isξk′+ξm′ =ξk+m′ (k, m∈Z) for the valuesξn′ :=ξn−η0εN. It follows by induction thatξ′n=nξ1′ (n ∈Z) and hence the sequence (ξn: n∈ Z) is arithmetic. On the other hand (withk =ℓ =m=:n) also ξn=ηnεN (n ∈Z). Thus for some α, β ∈C,
ξn=nα+β , ηn =εN(nα+β) (n ∈Z).
Since u−v =N(w1−w0), we have λge−λgf =N(λga1 −λga0) = 0 (g ∈G). Thuse ≈f and therefore λeee e∗ = λf ff f∗ [on the base space if we consider partial J*-triples]. It follows
[e f∗, f e∗] ={ef∗f} e∗−f {ee∗f}∗ =λf ee e∗−λeff f∗ =λf fe e∗−λeef f∗ = 0 . This means that {ef∗{f e∗an}} = {f e∗{ef∗an}} i.e. ηn{ef∗an−N} = ξn{f e∗an+N} or ηnξn−Nan=ξnηn+Nan (n∈Z). Thus
ηnξn−N =ξnηn+N (n∈Z)
εN(nα+β)[(n−N)α+β] = (nα+β)εN[(n+N)α+β]
n2|α|2+n[−N|α|2+ 2Reαβ] + [−N αβ+|β|2] =
= n2|α|2+n[N|α|2+ 2Reαβ] + [−N βα+|β|2]
which is possible only if α = 0. Therefore ξn = ξ0 and ηn = η0 = εNξ0 for every index n∈Z.
Definition 8.5.2. We introduce the notation
For every k, ℓ, m∈Z we have Using these relations, we evaluate the identity
{bkak+t∗{bkbk−t∗ai}} = {{bkak+t∗bk}bk−t∗ai} −
is independent of the index k and has absolute value 1. Thus
λtξ(k−i)−t=αβ(1 +σtτt)ξtξk−i−ξ(k−i)+t , Qbkak+t= Λktck−t =ασk+tλtck−t for any i, k, t∈Z. We complete the proof by substituting j :=k−i.
Remark 8.5.6. It is well-known from elementary linear algebra that the bilateral shift T : (zn: n∈Z)7→(zn+1 : n ∈Z)
on the sequence space S :={
(zn: n∈Z) : z0, z−1, z1, . . .∈C}
has the following spectral property: {
(nkωn: n∈Z) : ω∈C\ {0}, k= 0,1, . . .}
is a basis in S0
whereS0 :={
z ∈ S : ∃ ppolynomial p(T)z = 0}
. Moreover, each sequence (nkωk : n ∈ Z) is an eigenvector of orderk (with eigenvalue ω̸= 0).
Corollary 8.5.7. There exist ω1, ω2 ∈C\ {0} and δ ∈R such that ξn= αβ
2 (1 +iδ)ω1n+ αβ
2 (1−iδ)ω2n , λn= (ω1ω2)n (n∈Z) . The following alternatives hold
i) στ = 1, δ= 0 and |ω1|=|ω2|= 1, ii)στ = 1, δ= 0 and ω2 =ω1−1,
iii) στ =−1, δ ∈R arbitrary and ω1 =−ω2, |ω1|= 1.
Proof. By Proposition 8.5.5, ξj+2t−αβ(1 +σtτt)ξtξj+t+λtξj ≡0. Thus with the notation of the previous remark,
[T2t−αβ(1 +σtτt)ξtTt+λt
](ξn: n∈Z) = 0 (t ∈Z).
Letω1, ω2 denote the roots of the polynomial z2−αβ(1 +στ)ξ1z+λ1. Therefore we have only the possibilities
1)στ = 1 , ξn = (A+Bn)ωn with ω ∈ {ω1, ω2},
2)στ = 1 , ξn =Aω1n+Bω2n with ω1 ̸=ω2 , A, B ̸= 0 , 3)στ =−1, ξn =Aϱn+B(−ϱ)n with ϱ∈ {ω: ω2 =λ1} for some A, B ∈C.
Case 1. We show that necessarily A =αβ, B = 0, and λt=ω2t.
The condition ξ0 =αβ implies A=αβ. Hence the relation ξ−n= (στ)nξn=ξn means (αβ−nB)ω−n = (αβ+nB)ωn (n ∈Z).
Thus we must have ω−1 = ω and B = −B. That is |ω|2 = 1 and B ∈ iR. On the other hand
0 = ξn+2t−2αβξtξn+t+λtξn =
= [αβ+(n+2t)B]ωn+2t−2αβ(αβ+tB)[αβ+(n+t)B]ωn+2t+λt(αβ+nB)ωn=
= [
−αβω2t−2B(tω2t)−2αβB2(t2ω2t) +αβλt] ωn+ +B[
−ω2t−2αβB(tω2t) +λt] nωn .
For fixed t ∈ Z, both the coefficients of ωn and nωn should vanish in the last expression.
Hence B = 0 implies λt = ω2t. From the assumption B ̸= 0, we get the contradiction λt =ω2t+ 2αβB(tω2t)−2αβB2(t2ω2t)≡ω2t+ 2αβB(tω2t).
Case 2. For any fixed t∈Z,
0 = ξn+2t−2αβξtξn+t+λtξn =
= ω1nA[(
1−2αβA)
ω2t1 −2αβB(ω1ω2)t+λt] + +ω2nB[(
1−2αβB)ω2t2 −2αβA(ω1ω2)t+λt
] (n∈Z).
By assumptionA, B ̸= 0. Therefore, since the coefficients ofω1, ω2 must vanish, λt = 2αβB(ω1ω2)t+ (2αβA−1)(ω12)t=
= 2αβA(ω1ω2)t+ (2αβB−1)(ω22)t
for anyt ∈Z. By assumptionω1 ̸=ω2 and henceω1ω2 ̸=ω12,ω1ω2 ̸=ω22 in this case. Thus the coefficients of the terms (ω1ω2) should be the same i.e. A=B. SinceA+B =ξ0 =αβ, necessarily
A =B =αβ/2 , λt = (ω1ω2)t (t ∈Z).
Since |λt|= 1, also |ω1ω2|= 1. On the other handξ−n=ξn (n∈Z). This means ω1−n+ω2−n =ω1n+ω2n (n ∈Z) .
Since the sequences( ζn)
(ζ ∈C\ {0}) form a linearly independent family, it followsω1−1 ∈ {ω2−1, ω1, ω2}. However, ω1−1 ̸= ω−11 whence ω1−1 ∈ {ω1, ω2}. Similarly ω2−1 ∈ {ω1−1, ω1}. Since ω1 ̸=ω2, we have the subcases
2.1) ω−11 =ω1 and ω−21 =ω2 with |ω1|2 =|ω2|2 = 1 2.2) ω−11 =ω2 i.e. ω1 =ω, ω2 =ω−1 .
Case 3. Since |ϱ| =√
|λ1|= 1, the relations ξ0 =αβ, ξ−n= (στ)nξn = (−1)nξn imply A+B =αβ and
0 = [Aϱ−n+B(−ϱ)−n]−(−1)n[Aϱn+B(−ϱ)n] = (A−B)ϱn+ (B−A)(−ϱ)n for all n ∈Z. This is possible if and only if A= (1 +iδ)αβ/2, B = (1−iδ)αβ/2 for some constant δ∈R. In this case
0 = ξn+2t−[1 + (−1)t]αβξtξn+t+λtξn =
= αβ 2
((1 +iδ)ϱn+ (1−iδ)(−ϱ)n) [
λt−(−ϱ2)t]
(n∈Z) , and hence λt= (−ϱ2)t for any fixedt ∈Z.
Remark 8.5.8. Henceforth we assume W :=Z2 ⊕1 and we use the abbreviation gpq for the term g(p,q,1).
Lemma 8.5.9. Assume
{gpigpj∗gpk}= sgn(gpj)gp,i−j+k (i, j, k ∈Z, p= 0,1) , {giqgjq∗gkq}= sgn(gjq)gi−j+k,q (i, j, k, q ∈Z) .
Then also
{gpigpj∗gpk}= sgn(gpj)gp,i−j+k (i, j, k, p ∈Z) . Proof. Taking into account 8.2.8, it suffices to verify
sgn(gpk)gp,k±2 ={ap,k±1gpk∗gp,k±1}=Qgp,k±1(gpk) for all p, k ∈Z. We prove this statement by induction. By assumption
sgn(gpk)gp,k+2ζ=Qgp,k+ζ(gpk) for p= 0,1 with k= 0, ζ= 1 or k= 1, ζ=−1;
sgn(gjq)gj+2ε,q =Qgj+ε,q(gjq) for any q, j ∈Z and ε=±1 . Thus we can apply the following induction step:
For any j, k ∈Z, ε, ζ ∈ {±1}
sgn(gpk)gp,k+2ζ =Qgp,k+ζ(gpk) (p=j, j+ε, j+2ε) sgn(gjq)gj+2ε,q =Qgj+ε,q(gjq) (q=k, k+ζ)
}
⇒ sgn(gjk)gj+2ε,k+2ζ =
=Qgj+2ε,k+ζ(gj+2ε,k).
By setting
am :=gj+mε,k , bm :=gj+mε,k+ζ , cm :=gj+mε,k+2ζ =Qbmam (m= 0,1,2), we have to establish the relation Qc1c0 = sgn(c0)c2.
Since, for any p, q the subspaces Spanj ∈Zgpj, Spani ∈Zgiq are string triples, we have {akaℓ∗am}=ασℓ , {bkbℓ∗bm}=βτℓ (k, ℓ, m∈Z).
with some α, β, σ, τ ∈ {±1}. Thus we can apply Proposition 8.5.5 and its corollary to the strings (an), (bn), (cn). It follows in particular
Qb1a2 =λ1c0 , Qb1a0 =λ−1c2 =λ−11 =λ1c2
for some λ1 ∈ T. Notice that for any g ∈ G(= {gij : i, j ∈ Z}), Q2g = id because g g∗ =±id. Thus we complete the proof by the argument
Qc1c0 = Qc1(λ1Qb1a2) = λ1QQb
1a1Qb1a2 =
= λ1(Qb1Qa1Qb1)Qb1a2 =λ1Qb1Qa1a2 =
= λ1sgn(a2)Qb1a0 = sgn(c0)c2
since sgn(a2) = sgn(gj+2ε,k) = sgn(gjk) = sgn(gj,k+2η) = sgn(c0).
Remark 8.5.10. Gp :={gpk : k ∈Z}and G′q:={gjq : j ∈Z} in 8.5.9 are weighted grids with weight figure Z. Therefore (cf. 8.2.8) we can define an equivalent non-nil weighted grid G′ :={gpq′ : p, q ∈Z} such that
{g′pigpj′ ∗gpk′ }
= sgn(g′pj)gp,i−j+k , {g′iqgjq′ ∗gkq′ }
= sgn(gjq′ )gi−j+k,q (p, q, i, j, k ∈Z) by means of the following double recursion:
g′pq =gpq (p, q = 0,1), g′p,k+1 :=Qg′
pkg′p,k−1 (p= 0,1; k >1), gp,k′ −1 :=Qg′
pkgp,k+1′ (p= 0,1; k <0), g′ℓ+1,q :=Qg′
ℓ,qgℓ′−1,q (q∈Z; ℓ >1), g′ℓ−1,q :=Qg′
ℓ,qgℓ+1,q′ (q∈Z; ℓ <0).
Corollary 8.5.11. There exists an equivalent non-nil weighted grid {gkℓ′ : k, ℓ∈Z} such
that {
g′ugv′∗gw′ }
= sgn(gxy′ )g′u−v+w (u, v, w ∈Z2 lie in one straight line) .
Proof. As we have seen, a non-nil weighted grid G′ has the required property whenever Qugv = sgn(gv)g2u−v for all u, v ∈ Z2. By Lemma 8.5.9, we may assume without loss of generality that
{gpigpj∗gpk}= sgn(gpj)gp,i−j+k, {giqgjq∗gkq}= sgn(gjq)gi−j+k,q (p, q, i, j, k∈Z).
Since, in any case sgn(gpq) = sgn(gp+2,q) = sgn(gp,q+2) (p, q ∈Z), we can write sgn(gpq) =αµpνqκp·q (p, q ∈Z)
whereα := sgn(g00),µ:= sgn(g10)sgn(g00),ν := sgn(g01)sgn(g00) andκ:=∏1
k,ℓ=0sgn(gkℓ).
Given any x, y, q ∈ Z, we can apply Proposition 8.5.5 and its corollary to the strings ap :=gp,y , bp :=gp,y+q , cp :=gp,y+2q (p∈Z). Since
{akaℓ∗am}=αyqσyqℓ ak−ℓ+m, {bkbℓ∗bm}=βyqτyqℓ (k, ℓ, m∈Z)
where αyq := sgn(a0), βyq := sgn(b0), σyq := sgn(a0)sgn(a1) and τyq := sgn(b0)sgn(b1), it follows in particular
Qgx+p,y+qgxy = Qbx+pax = sgn(ax)λ−p,yqcx+2p = sgn(gx,y)ω−1,yqp ω2,yq−p gx+2p,y+2q =
= sgn(gxy)Ωpy,qgx+2p,y+2q (p∈Z)
with some constants Ωy,q ∈ T for fixedy, q ∈Z. Analogously, by arguing with the strings a′q :=gxq, b′q :=gx+p,q,c′q :=gx+2p,q (q∈Z), we get
Qgx+p,y+qgxy = sgn(gxy)(Ω′x,p)−qgx+2p,y+2q (q∈Z) with some Ω′x,p∈T for fixed x, p. Necessarily
(Ωy,q)p = (Ω′x,p)q (x, y, p, q ∈Z) .
Here Ωy,q = (Ωy,q)1 = (Ω′0,1)q (y, q ∈ Z). Similarly Ω′x,p = (Ω0,1)p (x, p ∈ Z). Hence Ω0,1 = Ω′0,1 and, by denoting this common value by Ω,
Qx+p,y+qgxy = sgn(gxy)Ωpqgx+2p,y+2q (x, y, p, q ∈Z). Forζ ∈T, consider the non-nil weighted grid
Gζ :={ζpqgpq : p, q ∈Z} (ζ ∈T) . Since
{(ζ(x+p)(y+q)gx+p,y+q)(ζxygxy)∗(ζ(x+p)(y+q)gx+p,y+q)}
=
=ζ(x+2p)(y+2q)−2pq{gx+p,y+qgxy∗gx+p,y+q}=
= (ζ−2Ω)pqζ(x+2p)(y+2q)sgn(gxy)gx+2p,y+2q (x, y, p, q ∈Z),
by taking a square root ζ ∈ {ω : ω2 = Ω}, the non-nil weighted grid G′ :=Gζ suits our requirements.
Definition 8.5.12. Henceforth we shall use the notation u∧v := det
(u1u2 v1v2
)
=u1v2−u2v1 for u:= (u1, u2), v := (v1, v2)∈Z2.
Lemma 8.5.13. Suppose {gugv∗gw} = sgn(gv)gu−v+w whenever the vectors u, v, w(∈ Z2) lie on one straight line. Then
{gz+ugz∗gz+w}= sgn(gz)ξu∧vgz+u+v = (u, v, z ∈Z2) where ξn := sgn(g01)sgn(g00)
[gn1 g∗01 gn0 g∗00 ]
(n∈Z) . Proof. Observe that
{gz+ugz∗gz+v} =
[ gz+u g∗z gz+u+v gz+v∗
]{gz+u+vgz+v∗gz+v}=
= Sz(u, v)gz+u+v for any z, u, v ∈Z2 where
Sz(u, v) := sgn(gz+v)
[ gz+u gz∗ gz+u+v g∗z+v
] . Since the triple product is symmetric in the outer variables,
Sz(u, v) = Sz(v, u) (z, u, v ∈Z2) .
If v ∈Z2 and the constant θ∈R is such that θv ∈Z, then for some integers k, n we have
whenever z, u, v, θv ∈ Z. Hence, as it is well-known from elementary linear algebra of determinants, the functionalsSz satisfies
Theorem 8.5.14. LetEbe a J*-triple spanned by a non-nil weighted gridG={gpq:p, q∈Z}
over the non-degenerate weight figure Z2⊕1 (notation see 8.5.8). If ∏4
k=1sgn( gu(k)
)= 1 whenever the vectorsu(1), . . . , u(4) form a parallelogram then there exists an equivalent non-nil weighted grid G′ :={g′pq : p, q ∈Z} of E along with a constant ω ∈ T∪R\ {0} such
that {
gz+u′ gz′∗g′z+v}
=[1
2ωu∧v+1
2ω−u∧v]
sgn(g′z)gz+u+v′ (z, u, v ∈Z2).
Otherwise there exists an equivalent non-nil weighted grid G′ := {g′pq : p, q ∈ Z} along with a constant δ ∈R such that
{gz+u′ gz′∗g′z+v}
= Re(
(1 +iδ)iu∧v)
sgn(g′z)gz+u+v′ (z, u, v ∈Z2).
All the above operations determine different J*-triple structure.
Proof. Let G′ be a non-nil weighted grid with the properties described in the previous lemma. For the sake of simplicity, we may assume G = G′ without danger of confusion.
We know already that we can parameterize the signs of the grid elements as sgn(gpq) =αµpνqκpq (p, q ∈Z)
and the triple product has the form {gz+ugz∗gz+v}= 1
2sgn(gz)[
(1 +iδ)ω1u∧v + (1−iδ)ω1−u∧v]
gz+u+v
on the grid G with suitable constants δ ∈ R, ω1, ω2 ∈ C\ {0}. It is straightforward to verify that
∏4 k=1
sgn(gvk) =κ(v1−v2)∧(v3−v2) whenever v4 =v1−v2+v3 . Furthermore we have established that only the following three cases can occur
1)κ= 1 =|ω1|=|ω2|, δ = 0; 2)κ= 1 =ω1ω2, δ = 0; 3)κ=−1, ω1=−ω2∈T. Moreover, since Qgx+p,y+qgxy= sgn(gxy)gx+2p,y+2q= sgn(gxy)(ω1ω2)−pqgx+2p,y+2q (p, q ∈ Z), we have
ω1ω2 = 1 .
Thus, in Case 1)ω2 =ω∈T; in Case 2)ω1, ω2 ∈Rwith ω2 =ω1−1; in Case 3)ω2 =−ω1 =
±i. Therefore we have actually the following two cases
(i) κ= 1 , ω1 =ω , ω2 =ω−1 , δ= 0 for some ω∈T∪R\ {0} , (ii) κ=−1 , ω1 =i , ω2 =−i .
To complete the proof, it suffices to check that the sesqui-trilinear extensions of the oper-ations
{gz+ugzgz+v}ωα,µ,ν,κ1,ω2,δ = 1
2αµz1νz2κz1z2[
(1 +iδ)ω1u∧v+ (1−iδ)ω−1u∧v]
gz+u+v
satisfy the Jordan identity whenever the parameters α, µ, ν, κ∈ {±1}, ω1, ω2∈C, δ∈R satisfy the relations described in Cases (i),(ii). Moreover it is enough to check the Jordan identity only for the grid elements. Since for the triple product{. . .}:={. . .}ωα,µ,ν,κ1,ω2,δ we have {gagb∗{gxgy∗gz}},{{gagb∗gx}gy∗gz},{gx{gbga∗gy}∗gz},{gxgy∗{gagb∗gz}} ∈ Cga−b+x−y+z, we have to prove that
Dωα,µ,ν,κ1,ω2,δ(a, b, x, y, z) = 0 (a, b, x, y, z ∈Z2) where the function Dα,µ,ν,κω1,ω2,δ is defined by the relation
1
4(αµb1νb2κb1b2)(αµy1νy2κy1y2)Dα,µ,ν,κω1,ω2,δ(a, b, x, y, z)ga−b+x−y+z :=
:={gugv∗{gxgy∗gz}} − {{gugv∗gx}gy∗gz}+{gx{gvgu∗gy}∗gz} − {gxgy∗{gugv∗gz}}
for { }:={ }ωα,µ,ν,κ1,ω2,δ. Let a, b, x, y, z ∈Z2 be arbitrarily fixed.
Case (i). We can write {gugvgz}ωα,µ,ν,κ1,ω2,δ =[1
2ω(u−v)∧(w−v)+1
2ω−(u−v)∧(w−v)]
αµv1νv2gu−v+w
with suitable 0̸=ω∈T∪R. Therefore, in this case Dα,µ,nu,1ω,1/ω,0 (a, b, x, y, z) =
= (
ω(x−y)∧(z−y)+ω−(x−y)∧(z−y))(
ω(a−b)∧(x−y+z−b)+ω−(a−b)∧(x−y+z−b))
−
−(
ω(a−b)∧(x−b)+ω−(a−b)∧(x−b))(
ωa−b+x−y)∧(z−y)+ω−(a−b+x−y)∧(z−y)) + +(
ω(b−a)∧y−a)+ω−(b−a)∧(y−a))(
ω(x−b+a−y)∧z−b+a−y)+ω−(x−ba−y)∧(z−b+a−y))
−
−(
ω(a−b)∧(z−b)+ω−(a−b)∧(z−b))(
ω(x−y)∧(a−b+z−y)+ω−(x−y)∧(a−b+z−y)) for ω∈T∪R\ {0} and a, b, x, y, z ∈Z2. Since
ωd+ω−d=ωd+ω−d (ω∈T∪R\ {0} , d∈R) , the identity Dω,1/ω,0α,µ,nu,1(a, b, x, y, z) = 0 holds. Namely, by setting
A := (x−y)∧(z−y) =x∧z−y∧z−x∧y , B := (a−b)∧(x−b) =a∧x−b∧x−a∧b , C := (b−a)∧(y−a) =b∧y−a∧y+a∧b , D := (a−b)∧(z−b) =a∧z−b∧z−a∧b , we have
(a−b)∧(x−y+z−b) = B+C+D , (z−y)∧(a−b+x−y) = −A−C−D , (x−b+a−y)∧(z−b+a−y) = A−B+D ,
(x−y)∧(a−b+z−y) = A−B−C .
Hence indeed
(ω(x−y)∧(z−y)+ω−(x−y)∧(z−y))(
ω(a−b)∧(x−y+z−b)+ω−(a−b)∧(x−y+z−b))
−
−(
ω(a−b)∧(x−b)+ω−(a−b)∧(x−b))(
ω(z−y)∧(a−b+x−y)+ω−(z−y)∧(a−b+x−y)) + +(
ω(b−a)∧(y−a)+ω−(b−a)∧(y−a))(
ω(x−b+a−y)∧(z−b+a−y)+ω−(x−ba−y)∧(z−b+a−y))
−
−(
ω(a−b)∧(z−b)+ω−(a−b)∧(z−b))(
ω(x−y)∧(a−b+z−y)+ω−(x−y)∧(a−b+z−y))
=
= (ωA+ω−A)(ωB+C+D+ω−B−C−D)−(ωB+ω−B)(ω−A−C−D +ωA+C+D) + +(ωC+ω−C)(ωA−B+D+ω−A+B−D)−(ωD+ω−D)(ωA−B−C+ω−A+B+C) = 0.
Case (ii). With someδ∈R we can write {gugvgw}ωα,µ,ν,κ1,ω2,δ = Re[
(1 +iδ)i(u−v)∧(w−v)]
αµv1νv2(−1)v1v2gu−v+w . Therefore, by settingγ := (1 +iδ)/2, with the same terms A, B, C, D as above
Dα,µ,ν,i,−i,δ−1 = 4Re(γi(x−y)∧z−y))Re(γi(a−b)∧(x−y+z−b))−
−4Re(γi(a−b)∧(x−b))Re(γi(a−b+x−y)∧(z−y)) +
+(−1)(b−a)∧(y−a)4Re(γi(b−a)∧(y−a))Re(γi−(x−b+a−y)∧(z−b+a−y))−
−4Re(γi(a−b)∧(z−b))Re(γi(x−y)∧(a−b+z−y)) =
= 2Re[
γiA(γiB+C+D+γi−B−C−D)−γiB(γiA+C+D+γi−A−C−D) + +γi−C(γiA−B+D+γi−A+B−D)−γiD(γiA−B−C +γi−A+B+C)]
=
= 2Re[
γ2(iA+B+C+D−iA+B+C+D+iA−B−C+D−iA−B−C+D) + +|γ|2(iA−B−C−D −i−A+B−C−D+i−A+B−C−D−i−A+B+C+D)]
=
= 2|γ|2Re(iA−B−C−D −i−A+B+C+D) = 0 .
Remark 8.5.15. The grid triples Fα,µ,ν,1ω,1/ω,0 with the triple products{..∗.}ω,1/ω,0α,µ,ν,1 are pair-wise non-isomorphic for different parameters ω ∈ T+ ∪ (0,1] where T+ := {ζ ∈ T : Re(ζ),Im(ζ)≥0}. On the other hand, Fα,µ,ν,1ω,1/ω,0, Fα,µ,ν,11/ω,ω,0 and Fα,µ,ν,1−ω,−1/ω,0 are isomorphic to each other for anyα, µ, ν ∈ {±1} and ω ∈T∪R\ {0}.
The grid triples Fα,µ,ν,i,−i,δ−1 with triple products {..∗.}i,α,µ,ν,−i,δ−1 are pairwise non-isomorphic for different parameters δ∈[0,∞). On the other hand, Fα,µ,ν,i,−i,δ−1, Fα,µ,ν,−i,i,δ−1 and Fα,µ,ν,i,−i,−−δ1 are isomorphic to each other for any α, µ, ν ∈ {±1}and δ∈R.
Remark 8.5.16. Givenδ ∈[0,∞), by settingθ:= arcotgδ, the triple product{..∗.}i,1,1,1,−i,δ−1 of the grid triple Eθ :=F1,1,1,i,−i,cotg−1 θ has the form
{gugv∗gw}i,1,1,1,−i,cotg−1 θ = (−1)v1v2sin[θ−(u−v)∧(w−v)π/2]
sinθ gu−v+w . Therefore the scaled operation
{gugv∗gw}θ := (−1)v1v2sin[(u−v)∧(w−v)π/2−θ]gu−v+w
is a triple product on Eθ for any θ ∈ (0, π/2]. Thus, by passing to the limit θ ↓ 0, the operation {..∗.}0 is also a well-defined non-trivial J*-triple product on the vector space Spanw∈Z2gw such that (gw gw∗)0 :={gwgw∗ · }0 = 0 (w∈Z2).
Chapter A Appendix
A.1 Finite dimensional unitary tensor operators
For the purposes of Step (1) of the proof Theorem 2.4.17, let H(1), . . . ,H(N) be fixed finite dimensional Hilbert spaces. We are aimed to describe the structure of the linear unitary operators in the space E := B(H(1), . . . ,H(N)) of N-linear functionals with unit ball denoted by B :=B(E) for short. We shall use the tensorial notations of 2.4.13 along with the dual objects B∗ :=B(E∗),
K :={Φ∈∂B:∃!F ∈∂B∗ ⟨F,Φ⟩= 1}, K∗ :={F ∈∂B∗ :∃Φ∈K ⟨F,Φ⟩= 1}. Lemma A.1.1. K∗ ={δe1,...,eN : e1 ∈∂B(H1), . . . ,eN ∈∂B(H(N))}.
Proof. Since dimE < ∞, B is compact thus for any n-linear functional Φ ∈ ∂B, one can find e1 ∈ ∂B(H(1)), . . . ,eN ∈ ∂B(H()N) with Φ(e1, . . . ,eN) = 1. Hence K∗ ⊂ {δe1,...,eN : ej ∈ ∂B(H(j)}. On the other hand, every E-unitary operator maps K onto itself and therefore also
U∗K∗ =K∗ for all E-unitary operators. (A.1.2) From the compactness of B it follows K ̸=∅ (indeed: for any smooth norm ∥.∥1 on E,
∅ ̸= {Φ ∈ ∂B : ∥F∥1 ≤ ∥G∥1∀G ∈ ∂B} ⊂ K) whence K∗ ̸= ∅. That is, for some unit vectors e01 ∈ H(1), . . . ,e0N ∈ H(N) we have δe0
1,...,e0n ∈ K∗. Now, from (A.1.2) we obtain δU1e0
1,...,Une0n = (U1⊗ · · · ⊗Un)∗δe0
1,...,e0n ∈K∗ whenever the Uj-s are H(j)-unitary operators.
Thus{δe1,...,eN :ej ∈∂B(H(j))} ⊃K∗.
Lemma A.1.3. Let ∆1 = δf1,...,fN, ∆2 := δg1,...,gN and ∆3 := δh1,...,hN where 0 ̸= fj,gj,hj ∈H(j) (j = 1, . . . , N) and assume ∆1+∆2 =∆3. Then there exists k such that for each j ̸=k we have fj∥gj (i.e. fj and gj are linearly dependent).
Proof. The statement holds obviously if for some index m, we have fj∥hj for all j ̸=m or fj∥gj for allj ̸=m. In the contrary casefk̸∥gkandfm̸∥hmfor some pair of indicesk ̸=m.
We may then suppose k = 1 and m = 2. First we show that in this case we have h1̸∥f1. Indeed: from h1 ̸∥f1 it follows that introducing the functional Φ :=e ge∗1 ⊗g∗2 ⊗ · · · ⊗gN∗ where eg1 := g1 − ∥f1∥−2⟨g1|f1⟩f2, the relations ⟨∆1,Φe⟩ = ⟨∆3,Φe⟩ = 0 ̸= ⟨∆2,Φe⟩ hold.
One can see in the same manner that h2̸∥ g2. Since h1̸∥ f1, there exists u1 ∈ H(1) with f1 ⊥u1 ̸⊥ h1 and since h2̸∥ g2, one can find u2 ∈ H(2) with g2 ⊥u2 ̸⊥ h2. But then the functional Θ :=u∗1⊗u∗2⊗h∗3⊗ · · · ⊗hN satisfies ⟨∆1,Θ⟩=⟨∆2,Θ⟩= 0̸=⟨∆3,Θ⟩ which is impossible.
Proposition A.1.4. Set rj = dimH(j)(j = 1, . . . , N) and let U ∈ L(E) be fixed so that U|B ∈aut(B). Then one can chooseH(j)-unitary operatorsUj such that U =U1⊗· · ·⊗Un. Proof. It is enough to prove the statement only forE-unitary operators lying in a suitable neighborhood of idE as it is well-known (see e.g. [32]). To do this, fix ε >0 such that the functionalsF :=δe1,...,eN,Fe :=δee1,...,eeN, G:=δf1,...,fN,Ge :=δef
1,...,efN(∈E∗) fulfill
∃k ek⊥eek,fk ⊥efk and ∀j̸=k ej∥eej, fj∥efj (A.1.5) whenever we have
F −F , Ge −Ge ∈K∗,∥F −Fe∥=∥G−Ge∥=√
2 , ∥F −G∥,∥Fe−Ge∥< ε, (A.1.6)
∥ej∥=∥eej∥=∥fj∥=∥efj∥= 1 (j = 1, . . . , N). (A.1.7) A valueε >0 with the above properties in fact exists. Otherwise there would be a sequence Fm := δem
1,...,emN, Fem := δeem
1 ,...,eemn, Gm =:= δfm
1 , . . . ,fNm, Gem := δefm
1 ,...,efNm (m = 1,2, . . .) satisfying (A.1.6),(A.1.7) for ε = m1 but without property (A.1.5). For a suitable index subsequence [
ms]∞
s=1 and for some unit vectors ej,eej,fj,efj we have emj s → ej, eemj s →eej, fjms → fj, efjms → efj (s → ∞, j = 1, . . . , N). Then the limits F,F , G,e Ge satisfy F = G, Fe = G,e ∥F −Fe∥ = ∥G−Ge∥ = √
2 and the contrary of (A.1.5). At the same time we also have F −F , Ge −Ge∈ K∗ because of the figure K∗ is closed. Thus by Lemma A.1.3,
∃!k0 ∀j̸=k0 ej∥eej. Since ∥F −Fe∥=√
2, hence ∥ek0 −eek0∥=√
2 i.e. ek0 ⊥eek0. Similarly
∃!ℓ0 fℓ0 ⊥efℓ0 and ∀j̸=ℓ0 fj∥efj. Since (A.1.5) does not hold, necessarily k0 ̸=ℓ0. However, the relations F =G, Fe=Ge entail k0 =ℓ0.
Now assume ∥U − idE∥ < ε. Fix orthonormed basis {ekj : j = 1, . . . , rk} ⊂ H(k) (k = 1, . . . , N), and let us write the functional U∗δe1
1,...,eN1 in the formU∗δe1
1,...,en1 =δf1
1,...,f1n
(cf. Lemma A.1.1) where f1k is a fixed unit vector in H(k) (k = 1, . . . , N). It follows from the choice of ε that for arbitrary indexk, the singleton {f1k} can be continued to an orthonormed basis{fjk : j = 1, . . . , rk} of H(k) in a unique way so that we have
U∗δe1
1,...,ek1−1,ekj,ek+11 ,...,eN1 =δf1
1,...,f1k−1,fjk,f1k+1,...,f1n (j = 1, . . . , rk).
Set I0 :={1, . . . ,1, j,1, . . . ,1) : k= 1, . . . , N; j= 1, . . . , rk}, I1 :=×nk=1{1, . . . , rk} and let a family I ⊂ I1 of multiindices be called thick if ∀i∈I ∀i′∈I1 i′≤i ⇒ i′ ∈ I. Observe that for any i:= (i1, . . . , iN)∈I1, there is a unique complex number κi ∈T such that
U∗δe1
i1,...,enin =κiδf1
i1,...,finn. (A.1.8)
Indeed: if not, we could find a minimal multiindex i∈ I1 (w.r.t ≤) not satisfying (A.1.8). linearity of the mapping U, (A.1.8) yields the statement of the proposition immediately.
Assume I1 \I ̸= ∅ and let j be minimal element of I1 \ I. Observation: ∀i ∈ I1
Throughout this section we use the notations and concepts established at the beginning of 5.5. In particular (F, F0,{..∗.}) denotes an arbitrarily fixed partial J*-triple over K. Definition A.2.1. An element 0̸= e ∈F0 is a tripotent with sign λ∈ K if {ee∗e} =λe.
Remark that the sign of a tripotent is unambiguously determined. We shall write sgn(e) :=
[λ∈K: e3 =λe]. Clearly, weighted grids consist of (signed) tripotents.
Lemma A.2.2. Supposee is a tripotent in F0 with λ := sgn(e)̸= 0. Then λ∈ ReK and F0 =⊕2k=0
{x∈F0 : (e e∗)x= (λ/2)x} .
Proof. It is well-known [10] that for any fixedc∈F0 the J*-tripleF0becomes a commutative Jordan algebra when equipped with the c-product x•c y:={xc∗y}. Hence, for any a∈ coefficientsand we reserve the notationλab, λbafor them. Notice that weighted grids consist of pairwise compatible tripotents.