SciELO - Scientific Electronic Library Online

 
vol.46 número2Recent progress in the well-posedness of the Benjamin-Ono equationGeometric Methods in Wavelet Theory índice de autoresíndice de materiabúsqueda de artículos
Home Pagelista alfabética de revistas  

Servicios Personalizados

Revista

Articulo

Indicadores

  • No hay articulos citadosCitado por SciELO

Links relacionados

  • No hay articulos similaresSimilares en SciELO

Compartir


Revista de la Unión Matemática Argentina

versión impresa ISSN 0041-6932versión On-line ISSN 1669-9637

Rev. Unión Mat. Argent. v.46 n.2 Bahía Blanca jul./dic. 2005

 

Infinitely many minimal curves joining arbitrarily close points in a homogeneous space of the unitary group of a C*-algebra

Esteban Andruchow, Luis E. Mata-Lorenzo, Lázaro Recht, Alberto Mendoza and Alejandro Varela

Dedicated to the memory of Ángel Rafael Larotonda (Pucho).

 

Abstract: We give an example of a homogeneous space of the unitary group of a C*-algebra which presents a remarkable phenomenon, in its natural Finsler metric there are infinitely many minimal curves joining arbitrarily close points.

1. Introduction

In this paper, we give an example of a homogeneous space of the unitary group of a C* -algebra which presents a remarkable phenomenon. Namely, in its natural Finsler metric there are infinitely many minimal curves joining arbitrarily close points. More precisely the homogeneous space will be called P . The unitary group U  of a C* -algebra A  acts transitively on the left on P . The action is denoted by Lg ρ  , for g ∈ U and ρ ∈ P . The isotropy I ρ = {g ∈ U ∕ Lg ρ = ρ} will be the unitary group of a C * -subalgebra B ⊂  A . The Finsler norm in P  is naturally defined by ∣∣X ∣∣ρ = infb∈Bah ∣∣Z + b∣∣ , for X  ∈ (T P)ρ  where Z ∈ Aah  projects to X  in the quotient Aah ∕Bah  which is identified to the tangent space (T P)      ρ  . These definitions and notation are borrowed from [1].

This work is part of a forthcoming paper by the same authors which will contain additional results about minimal vectors. We call an element Z  ∈ Aah  minimal vector if ∥Z ∥ ≤ ∥Z  + V ∥,  for all V ∈ Bah.

2. The Minimality Theorem

In [1] the following theorem is proven.

Theorem 2.1. Let P be a homogeneous space of the unitary group of a C* -algebra A . Consider ρ ∈ P and X ∈ (T P) ρ   . Suppose that there exists Z ∈ Aah   which is a minimal vector i.e. ∣∣Z ∣∣ = ∣∣X ∣∣ρ   . Then the oneparameter curve γ(t)  given by γ(t) = LetZρ  has minimal length in the class of all curves in P joining γ(0)  to γ(t)  for each t  with ∣t∣ ≤ --π-      2∣∣Z∣∣ .

We will use the following notation. Let A be a unital C* -algebra, and 1 ∈ B  ⊂ A a C* -subalgebra. Denote Ah  and Bh  (resp. Aah  and Bah  ) the sets of selfadjoint (resp. anti-hermitian) elements of A and B . Let H be a Hilbert space, B(H)  the algebra of bounded operators acting on H , and Gl(H)  the group of invertible operators.

We call an element Z ∈ Aah  a minimal vector if

∥Z ∥ ≤ ∥Z  + V ∥,  for all V ∈ Bah.

Note that since for any operator, ∥Im(X)  ∥ ≤ ∥X ∥ , it follows that Z  ∈ Aah  is minimal if and only if

∥Z ∥ ≤ ∥Z +  B ∥,  for all B ∈ B.

In view of the purpose of this paper stated in the introduction and the previous theorem, to look for minimal curves we have to find minimal vectors and therefore the following theorem is relevant.

Theorem 2.2. An element Z ∈ Aah   is minimal if and only if there exists a representation ρ  of A in a Hilbert space H and a unit vector ξ  such that

  1.     2           2 ρ(Z  )ξ = - ∥Z ∥ ξ  .
  2. 〈ρ(Z) ξ,ρ(B) ξ〉 = 0  for all B ∈ B .

Proof. The if part is trivial. Suppose that there exist ρ,H, ξ  as above. Then if B  ∈ B ,

∥Z  + B∥2 ≥ ∥ρ(Z + B) ξ∥2 = ∥ρ(Z) ξ∥2 + ∥ρ(B) ξ∥2 ≥ ∥ρ(Z) ξ∥2 = - 〈ρ(Z2)ξ,ξ〉 = ∥Z∥2.

Suppose now that Z  is minimal. Denote by S the closed (real) linear span of Z2 +  ∥Z∥2I  and the operators of the form ZB  - B *Z  for all possible B ∈ B . Note that Z2 +  ∥Z∥2I  is positive and ZB  - B *Z  is selfadjoint, i.e. S  ⊂ Ah  .

Denote by C the cone of positive and invertible elements of A . We claim that the minimality condition implies that S ∩ C = ∅ . Indeed, otherwise, since C is open, there would exist s ∈ IR  and B ∈ B such that

s(Z2 + ∥Z ∥2I) + ZB  - B *Z ≥ rI,   with r > 0.

We may suppose that s > 0  , so that dividing by s  we get that for given B  ∈ B , r > 0  ,

Z2 + ∥Z ∥2I + ZB  - B *Z ≥  rI.

(2.1)

Also note that Z2 + ∥Z ∥2I ≥  0  , then for n ≥ 1  ,

n(Z2 + ∥Z ∥2) + ZB  - B *Z ≥  Z2 + ∥Z ∥2I + ZB  - B *Z ≥  rI.

Or equivalently, dividing by n  ,

                (    )    (     )   2      2        1-        1- *        ′ Z  + ∥Z ∥ I + Z   n B   -   nB    Z  ≥ r I.

In other words, one can find B  ∈ B with arbitrarily small norm such that inequality 2.1 holds.

This inequality clearly implies that

sp(Z2 + ZB  - B *Z) ⊂  (- ∥Z ∥2,+ ∞).

On the other hand, since B  can be chosen with arbitrarily small norm, and Z2   is non positive, it is clear that one can choose B  in order that sp(Z2 + ZB  - B *Z) ⊂  (- ∞,  ∥Z∥2)  . Therefore there exists B  ∈ B such that ∥Z2 +  ZB  - B *Z∥ < ∥Z ∥2   . Let us show that this contradicts the minimality of Z  , and thus proves our claim. Indeed, this is stated in lemma 5.3 of [1]:

Lemma 2.3. If ∥Z + B ∥ ≥ ∥Z ∥ for all B  ∈ B , then also ∥Z2  + ZB  - B *Z∥ ≥  ∥Z ∥2   .

We include its proof. Consider for t > 0  , f(t) = Z2 + 1((Z +  tB) *(Z + tB) -  Z2)             t  . Note that ∥f (t)∥ ≥ ∥Z ∥2   . Otherwise ∥f(t)∥ < ∥Z ∥2   and then the convex combination                 2 tf(t) + (1 - t)Z   has norm strictly smaller than     2 ∥Z ∥   for 0 < t < 1  . Note that

tf(t) + (1 - t)Z2 = (Z + tB) *(Z + tB).

That is ∥Z + tB ∥2 = ∥(Z + tB) *(Z + tB) ∥ < ∥Z ∥2   , which contradicts the hypothesis, and the lemma is proven, as well as our claim.

We have that S ∩ C =  ∅ , with S a closed (real) linear submanifold of Ah  and C open and convex in Ah  . By the Hahn-Banach theorem, there exists a bounded linear functional φ0   in Ah  such that

φ0(S)  = 0  and φ0(C) > 0.

The functional φ0   has a unique selfadjoint extension to A , let φ  be the normalization of this functional. Then clearly φ  is a state which vanishes on S . Let ρ,H, ξ  be the GNS triple associated to this state. Note that since   2       2 Z  +  ∥Z∥ I ∈ S ,      2             2         2 〈ρ(Z  )ξ,ξ >=  φ(Z  ) = - ∥Z ∥   , and therefore, by the equality part in the Cauchy-Schwartz inequality, it follows that

    2           2 ρ(Z  )ξ = - ∥Z∥ ξ.

Moreover, 0 = φ(ZB   - B *Z + Z2 +  ∥Z∥2I) =  φ(ZB  - B *Z)  . Since φ  is selfadjoint, this means Re( φ(ZB))  = 0  for all B ∈ B . Putting iB  in the place of B  , one has that in fact φ(ZB)  = 0  for all B ∈  B . Then,

0 =  〈ρ(ZB)  ξ,ξ〉 = 〈ρ(B) ξ,ρ(Z)ξ〉,

which concludes the proof. □

3. Infinitely many minimal curves joining arbitrarily close points

In this example the homogeneous space P is the flag manifold of 4-tuples of mutually orthogonal lines in ℂ4   (1-dimensional complex subspaces). The group of unitary operators in   4 ℂ   acts on the left in P by sending each complex line to its image by the unitary operator (thus preserving the orthogonality of the new 4-tuple complex lines). Consider the canonical flag pe = (sp {e1},sp {e2},sp {e3},sp {e4})  where sp {ei} is the complex line spanned by the canonical vector ei  in ℂ4   . The isotropy of the canonical flag pe  is the subgroup of 'diagonal' unitary operators.

We consider now the submanifold P   d  of P  given by

Pd = {(l1,l2,l3,l4) ∈ P ∣ sp{l1,l2}= sp{e1,e2}}

Notice that Pd = W  ×  W where W is the flag manifold of couples of mutually orthogonal 1-dimensional complex lines in ℂ2   . Notice also that an ordered pair of mutually orthogonal 1-dimensional complex lines in ℂ2   is totally determined by the first complex line of the pair, hence W  = ℂP (2)  . Furthermore ℂP (2) = RS , the Riemann Sphere, hence W =RS .

The minimal curves presented in this example shall be constructed in Pd  . For a better geometrical view of those curves we shall identify RS , via stereographic projection, with the unit sphere S2    in ℝ3   , hence we shall make the identification Pd =  S2 × S2   .

3.1. A description of the minimal curves. Let                      2    2 N =  (N, N ) ∈ Pd = S  × S   be the point whose coordinates are both the North Pole,       2 N ∈  S   . Let                  2    2 Q = (Q1, Q2) ∈ S  ×  S   be any point of Pd   such that Q1   has higher latitude than Q2   in S2    (Q1   is closer to N  than Q2   ).

We will fix Q so that Q2   is above the equator line (and Q1   is even higher) and present a family of minimal curves Γ β(t) = (γ1,β(t),γ2(t)), for t ∈ [0,1]  , all joining N to Q .
  • The curve γ (t)  2  in S2    will trace the smaller arc of the great circle that contains N  and Q2   .
  • The family of curves γ1,β(t)  will vary continuously with the parameter β  .
  • Each of the curves of the family γ   (t)  1,β  will parametrize the smaller arc of some circle in   2 S    that joins N  to Q1   ; the arcs will not be great circles but for β = 0  .

(  || | || || | |(

PIC

)  || | || || | |)

3.2. A precise description of the minimal curves. To present the curves drawn above we give a more manageable description of P . We consider the unitary subgroup U  = U (4)  of the   * C -algebra A =  M4( ℂ)  of 4 × 4  complex matrices, and denote with B the subalgebra of diagonal matrices in A . The homogeneous space P is given by the quotient U ∕D , where D =  U ∩ B is the subgroup of the diagonal unitary matrices. The group U acts on P (on the left). The tangent space at 1 (the identity class) is the subspace of anti-hermitian matrices in A  with zeroes on the diagonal.

We construct Pd ⊂  P as follows. First consider the subgroup SU (2) × SU (2) ⊂ U of special unitary matrices build with two, 2 × 2  , blocks on the diagonal. We set Pd ⊂ P as the quotient of SU  (2) × SU (2)  by the subgroup D of diagonal special unitary matrices. This submanifold is in itself a product of two copies of the quotient W of SU (2)  by the subgroup of diagonal matrices in SU (2)  . For the relations among the different groups here mentioned we suggest [2]. We write Pd = W  × W and a point of Pd   is a class (in a quotient) which in itself has two components which are also classes. We shall use the notation [U ] = ([u1],[u2]) ∈ Pd = W  × W .

The minimal curves starting at 1 ∈ Pd  are of the form        [   ] γ(t) =  etZ where the matrices Z  are anti-hermitian matrices with zero trace in A  built with two blocks of anti-hermitian 2 × 2  matrices on the diagonal (each one with zero trace).

The minimality of the curves is granted by 2.1 for the matrices Z  shall be minimal vectors according to theorem 2.2. In fact, we shall consider Z ∈ Aan  of the form

    (  Z    0  ) Z =      1         0  Z2

where Z   1   and Z   2   are anti-hermitian 2 × 2  matrices of the form

          (                                             )                      zi           r(- sin(α) + i cos(α)) Z1   =                                                    ,         (3.2)             r(sin(α) + i cos(α))           - z i     and           (  0    w ) Z2   =        --                                                    (3.3)             - w   0

(3.2)



(3.3)

where z, r,α  ∈ ℝ  , and w  ∈ ℂ  .

The minimality of these matrices Z  is assured in the case where ∣w ∣2 ≥ z2 + r2   . In such case, ∣∣Z ∣∣2 = ∣w ∣2   and, in relation to theorem 2.2, just consider the operator representation ρ  of the   * C -algebra A  = M4( ℂ)  on  4 ℂ   , together with the unit vector                   4 ξ = (0,0,0, 1) ∈ ℂ   .

3.2.1. The two components of the curves in Pd   . The curve γ(t) = [etZ]=  ([etZ1],  [etZ2])  in Pd   has two components (in W ).

We shall regard the Riemann Sphere RS  as the complex plane ℂ  with the point "∞ " added. Consider a matrix u  in SU (2)

    (      - )       a   -b                            2     2 u =    b  a-   ,  where a, b ∈ ℂ and ∣a∣ + ∣b∣ =  1

We consider the mapping L  from SU (2)  to RS  is given by

L(u) =  a, if b ⁄= 0, else L(u) = ∞.         b

It is clear that this mapping induces an explicit diffeomorphism from the quotient of SU  (2)  by its diagonal matrices to the Riemann Sphere RS .

Consider the unit sphere   2 S    in  3 ℝ   , and let the equatorial plane, ℂ  , represent the "finite" part of the Riemann Sphere RS . We set             2 φ : RS →  S   to be the stereographic projection given as by:

       (   2ζ    ∣ζ∣2 - 1) φ(ζ) =   --2----,--2----  ∈ ℂ × ℝ = ℝ3, for ζ ∈ ℂ, and φ(∞)  = (0,0,1) = N  ∈ S2 ⊂ ℝ3          ∣ζ∣ + 1  ∣ζ∣ +  1

Notice that in the class b ⁄= 0  , if ζ = L(u) =  a∈  ℂ             b  , then φ(ζ) = (2ab-, ∣a∣2 - ∣b∣2) . If b = 0  , then ∣a∣ = 1  , hence ζ =  L(u) = ∞ , then φ(ζ) = (0 , 0 , 1)  .

Via a composition of two maps, we define the diffeomorphism Ψ   from W onto   2 S   : for      [(       - )]          a  - b [u] =     b   a-    ∈ W we set,

                    (   -           )   (  -          ) Ψ([u]) = φ (L(u)) =  2a b, ∣a∣2 - ∣b∣2 = 2ab, 1 - 2∣b∣2  ∈ S2

Considering the curve q(t) = etZ1   in SU  (2)  with Z1   as in (3.2) above, and setting       ------- λ =  √r2 + z2   , it can be verified that L(q(t)) ∈ RS is given by,

pict

(3.4)



(3.5)

Notice then that L(q(t))  parametrizes a straight line lq  in RS . Hence the curve

Ψ([q(t)]) = φ(L(q(t)))

is an arc of a circle in S2    (not necessarily a great circle) contained in the plane in   3 ℝ   that contains both the line lq  , in the equatorial plane, and the North Pole N  , in  2 S   . It can be verified that this plane has unit normal vectors given by:

± (cos(β)cos(α) , cos(β) sin( α), sin(β))

where                                      ------- cos(β) =  r,  sin(β) =  z, with λ = √ r2 + z2           λ            λ   .

3.2.2. Some observations on the curves Ψ ([etZ1]) and Ψ ([etZ2]) in W . Let            (     ) γ1,β(t) = Ψ  [etZ1] , where          r            z           √ ------- cos(β) = --, sin(β) =  --, with λ =  r2 + z2          λ            λ   , and let γ (t) = Ψ ([etZ2])  2

  • In the constructions above, the curve γ1,β(t)  runs over a great circle in S2    if and only if β = 0  (equivalently z = 0  ).
  • The curve γ2(t)  runs over a great circle in  2 S    (Z2   has parameter z = 0  ).
  • The curve γ1,β(t)  varies continuously with the parameter β  .
  • The curve γ   (t)   1,β  starts at N  ∈ S2   and returns to that point exactly for     { kπ       } t ∈   λ ∣ k ∈ ℤ .
  • The curve γ1,β(t)  has constant speed 2λ cos(β)  in S2   .
  • The curve γ2(t)  has constant speed 2r  in S2   .

3.2.3. The curves Γ (t) = (Ψ ([etZ1]), Ψ ([etZ2]))  β in P  d   . Lets give explicit values of the "parameters" z, α, r ∈ ℝ  and w ∈  ℂ  that define Z1   and Z2   (according to formulas (3.2) and (3.3)), so that for t ∈ [0,1]  , the curves            (     ) γ1,β(t) = Ψ  [etZ1] and           (    ) γ2(t) = Ψ  [etZ2] join the point N  to Q1   and Q2   respectively.

Suppose that the distances from N  to Q1   and Q2   in S2    are 2φ1   and 2φ2   respectively (with φ  <  φ   1    2   ).

By means of some rotation of the sphere   2 S    we may suppose that Q1   is in the plane generated by ˆj  and ˆk  , as in figure 1 below, and we have, Q1  = (0,sin(2 φ1),cos(2φ1)) and Q2 = (sin(2φ2)cos(θ2),sin(2φ2)sin(θ2),cos(2φ2))  .

 

Figure 1
Figure 1: We suppose that Q1   is in the plane generated by ˆj  and ˆ k.

 

For Z2   we set w = φ2 (- sin(θ2) + i cos(θ2))  so that           (  Z2) γ2(1) = Ψ  [e  ] = Q2   .

We have to choose the values z, α, r ∈ ℝ  and w ∈  ℂ  that define Z1   . This is equivalent to chose β, α, λ ∈ ℝ  via the change of variables given by the equations

         -r           z-           √ -2----2 cos(β) = λ , sin(β) = λ , with λ =   r +  z .

The parameters α  and β  are shown in figure 1, with the only restriction that the vector

⃗n = (cos(β) cos(α) , cos(β) sin(α), sin(β))

is orthogonal to a plane π  α,β  that contains N  and Q   1   .

The parameter λ  is determined after choosing α  and β  so that the short arc joining N  and Q1   , in the intersection of the plane π α,β  with the sphere   2 S   as in figure 1, has length ℓ  equal to 2λ cos(β)  , from where the value of λ  is drawn.

References

[1]    Durán, C. E., Mata-Lorenzo, L. E. and Recht, L., Metric geometry in homogeneous spaces of the unitary group of a C* -algebra: Part I-minimal curves, Adv. Math. 184 No. 2 (2004), 342-366.         [ Links ]

[2]    Whittaker, E. T. "A Treatise on the Analytical Dynamics of Particles and Rigid Bodies", Cambridge University Press, London 1988.         [ Links ]

Esteban Andruchow
Instituto de Ciencias,
Universidad Nacional de Gral. Sarmiento,
J. M. Gutierrez (1613) Los Polvorines, Argentina
eandruch@ungs.edu.ar

Luis E. Mata-Lorenzo
Universidad Simón Bolívar, Apartado 89000,
Caracas 1080A, Venezuela
lmata@usb.ve

Lázaro Recht
Universidad Simón Bolívar, Apartado 89000,
Caracas 1080A, Venezuela, and
Instituto Argentino de Matemática, CONICET, Argentina
recht@usb.ve

Alberto Mendoza
Universidad Simón Bolívar, Apartado 89000,
Caracas 1080A, Venezuela
jacob@usb.ve

Alejandro Varela
Instituto de Ciencias,
Universidad Nacional de Gral. Sarmiento,
J. M. Gutierrez (1613) Los Polvorines, Argentina
avarela@ungs.edu.ar

Recibido: 23 de marzo de 2006
Aceptado: 7 de agosto de 2006

Creative Commons License Todo el contenido de esta revista, excepto dónde está identificado, está bajo una Licencia Creative Commons