2000 Mathematics Subject Classification. Primary 46L87; Secondary 46L07, 53C23, 58B34, 60B10.
<ph f="cmbx">Quantized Gromov-Hausdorff distance</ph>

wei wu

Department of Mathematics, East China Normal University, Shanghai 200062, P.R. China E-mail address : wwu@math.ecnu.edu.cn Current address : Department of Mathematics, University of California, Berkeley, CA 94720-3840 E-mail address : wwu@math.berkeley.edu

1 Introduction

Following up the compact metric spaces given by Connes in connection with his theory of quantum Riemannian geometry defined by Dirac operators [3, Rieffel defined the notion of a compact quantum metric space ( A , L A )   in [17as an order unit space A   equipped with a Lip-norm L A   , which is a generalization of the usual Lipschitz seminorm on functions which one associates to an ordinary metric. Many interesting examples of compact quantum metric space have been constructed [14, 16, 11, 10. Motivated by the type of convergence of spaces that has recently begun to play a central role in string theory, Rieffel introduces the quantum Gromov-Hausdorff distance for the compact quantum metric spaces as a quantum analogue of Gromov-Hausdorff distance, and shows that the basic theorems of the classical theory have natural quantum analogues.
In [20and [21, we formulated matrix Lipschitz seminorms on matrix order unit spaces. This operator space version of Lipschitz seminorm has many nice properties which are similar to those for ordinary metric spaces. These data may then be thought of as some `noncommutative metric spaces'. So it is natural to ask, as does Rieffel in [17, if it is possible to develop a corresponding operator space version of quantum Gromov-Hausdorff distance. This is the aim of the present article.
In contrast to the matricial quantum Gromov-Hausdorff distance in [8and operator Gromov-Hausdorff distance in [9, our quantized Gromov-Hausdorff distance operates entirely at the “matrix” level. Not only the matrix state spaces but also the matrix Lipschitz seminorms and the complete isometries are brought into our picture. This should be important in the background of operator systems.
The paper has eight sections. Section 2 contains preliminaries, mainly to fix some terminology and notation. In Section 3 we define quantized metric space and develop an operator “quotient”. Section 4 defines our quantized Gromov-Hausdorff distance, and we prove that it satisfies the triangle inequality. Section 5 deals with the operator Gromov-Hausdorff distance zero. We establish that it implies a complete isometry. Section 6 treats the completeness theorem of the complete isometry classes of quantized metric spaces. In Section 7 we show that a quantized metric space with 1-exact underlying matrix order unit space is a limit of matrix algebras with respect to quantized Gromov-Hausdorff distance. It is established in Section 8 that matrix algebras converge naturally to the sphere for quantized Gromov-Hausdorff distance.

2 Preliminaries

All vector spaces are assumed to be complex throughout this paper. Given a vector space V   , we let M m , n ( V )   denote the matrix space of all m   by n   matrices v = [ v i j ]   with v i j V   , and we set M n ( V ) = M n , n ( V )   . If V = C   , we write M m , n = M m , n ( C )   and M n = M n , n ( C )   , which means that we may identify M m , n ( V )   with the tensor product M m , n V   . We identify M m , n   with the normed space ( C n , C m )   .
We use the standard matrix multiplication and *-operation for compatible scalar matrices, and 1 n   for the identity matrix in M n   .
There are two natural operations on the matrix spaces. For v M m , n ( V )   and w M p , q ( V )   , the direct sum v w M m + p , n + q ( V )   is defined by letting v w = [ v 0 0 w ] ,   and if we are given α M m , p   , v M p , q ( V )   and β M q , n   , the matrix product α v β M m , n ( V )   is defined by α v β = [ k , l α i k v k l β l j ] .   A *-vector space V   is a complex vector space together with a conjugate linear mapping v v *   such that v * * = v   . A complex vector space V   is said to be matrix ordered if:
  • (1) V   is a *-vector space;
  • (2) each M n ( V )   , n N   , is partially ordered;
  • (3) γ * M n ( V ) + γ M m ( V ) +   if γ = [ γ i j ]   is any n × m   matrix of complex numbers.
A matrix order unit space ( V , 1 )   is a matrix ordered space V   together with a distinguished order unit 1   satisfying the following conditions:
  • (1) V +   is a proper cone with the order unit 1   ;
  • (2) each of the cones M n ( V ) +   is Archimedean.
Each matrix order unit space ( V , 1 )   may be provided with the norm v = inf { t R : [ t 1 v v * t 1 ] 0 } .   As in [17, we will not assume that V   is complete for the norm.
If V   and W   are *-vector spaces and φ : V W   is a linear mapping, we have a linear mapping φ * : V W   defined by φ * ( v ) = φ ( v * ) *   .
Given vector spaces V   and W   and a linear mapping φ : V W   and n N   , we have a corresponding φ n : M n ( V ) M n ( W )   defined by φ n ( [ v i j ] ) = [ φ ( v i j ) ] .   If V   and W   are vector spaces in duality, then they determine the matrix pairing , : M n ( V ) × M m ( W ) M n m ,   where [ v i j ] , [ w k l ] = [ < v i j , w k l > ]   for [ v i j ] M n ( V )   and [ w k l ] M m ( W )   .
A graded set S = ( S n )   is a sequence of sets S n ( n N )   . If V   is a locally convex topological vector space, then the canonical topology on M n ( V ) ( n N )   is that determined by the natural linear isomorphism M n ( V ) = V n 2   , that is, the product topology. A graded set S = ( S n )   with S n M n ( V )   is closed or compact if that is the case for each set S n   in the product topology in M n ( V )   . Given a vector space V   , we say that a graded set B = ( B n )   with B n M n ( V )   is absolutely matrix convex if for all m , n N  
  • (1) B m B n B m + n   ;
  • (2) α B m β B n   for any contractions α M n , m   and β M m , n   .
A matrix convex set in V   is a graded set K = ( K n )   of subsets K n M n ( V )   such that i = 1 k γ i * v i γ i K n   for all v i K n i   and γ i M n i , n   for i = 1 , 2 , , k   satisfying i = 1 k γ i * γ i = 1 n   .
Let V   and W   be vector spaces in duality, and let S = ( S n )   be a graded set with S n M n ( V )   . The absolute operator polar S = ( S n )   with S n M n ( W )   , is defined by S n = { w M n ( W ) : v , w 1 f o r a l l v S r , r N }   . The matrix polar S π = ( S n π )   with S n π M n ( W )   , is defined by S n π = { w M n ( W ) : R e v , w 1 r × n f o r a l l v S r , r N }   . Given a subset S V   , the absolute polar of S   is defined by S = { w W : | < v , w > | 1 for all v S }   .
A gauge on a vector space V   is a function g : V [ 0 , + ]   such that
  • (1) g ( v + w ) g ( v ) + g ( w )   ;
  • (2) g ( α v ) | α | g ( v )   ,
for all v , w V   and α C   . We say that a gauge g   is a seminorm on V   if g ( v ) < +   for all v V   . Given an arbitrary vector space V   , a matrix gauge G = ( g n )   on V   is a sequence of gauges g n : M n ( V ) [ 0 , + ]   such that
  • (1) g m + n ( v w ) = max { g m ( v ) , g n ( w ) }   ;
  • (2) g n ( α v β ) α g m ( v ) β   ,
for any v M m ( V )   , w M n ( V )   , α M n , m   and β M m , n   . A matrix gauge G = ( g n )   is a matrix seminorm on V   if for any n N , g n ( v ) < +   for all v M n ( V )   .
If each g n   is a norm on M n ( V )   , we say that G   is a matrix norm. An operator space is a vector space together with a matrix norm on it. For a matrix order unit space ( V , 1 )   , it is an operator space with the matrix norm determined by the matrix order on it.

3 Quantized metric space

First we recall the following definitions given in [20, 21:
Definition 3.1. Given a matrix order unit space ( V , 1 )   , a matrix Lipschitz seminorm   on ( V , 1 )   is a sequence of seminorms L n : M n ( V ) [ 0 , + )   such that
  • (1) the null space of each L n   is M n ( C 1 )   ;
  • (2) L m + n ( v w ) = max { L m ( v ) , L n ( w ) }   ;
  • (3) L n ( α v β ) α L m ( v ) β   ;
  • (4) L m ( v * ) = L m ( v )   ,
for any v M m ( V )   , w M n ( V )   , α M n , m   and β M m , n   .
Let ( V , 1 )   be a matrix order unit space. The matrix state space of ( V , 1 )   is the collection C S ( V ) = ( C S n ( V ) )   of matrix states C S n ( V ) = { φ : φ i s a u n i t a l c o m p l e t e l y p o s i t i v e l i n e a r m a p p i n g f r o m V i n t o M n } .   If = ( L n )   is a matrix Lipschitz seminorm on ( V , 1 )   , we have a collection D = ( D L n )   of metrics on C S ( V )   given by D L n ( φ , ψ ) = sup { φ , a ψ , a : a M r ( V ) , L r ( a ) 1 , r N } ,   for φ , ψ C S n ( V )   (notice that it may take value +   ). And in turn we obtain a sequence D = ( L D L n )   of gauges on ( V , 1 )   by L D L n ( a ) = sup { φ , a ψ , a D L r ( φ , ψ ) : φ , ψ C S r ( V ) , φ ψ , r N } ,   for all a M n ( V )   .
Definition 3.2. Let ( V , 1 )   be a matrix order unit space. By a matrix Lip-norm on ( V , 1 )   we mean a matrix Lipschitz seminorm = ( L n )   on ( V , 1 )   such that the D   -topology on C S ( V )   agrees with the BW-topology.
We are now prepared to make:
Definition 3.3. By a quantized metric space we mean a pair ( V , )   consisting of a matrix order unit space ( V , 1 )   with a matrix Lip-norm   defined on it.
Example 3.4. Let ( X , ρ )   be an ordinary compact metric space, let A   denote the set of Lipschitz functions on X   , and let L ρ   denote the Lipschitz seminorm on A   . Then A C ( X )   , and for f , g A   and α C   , we have L ρ ( f * ) = L ρ ( f ) , L ρ ( α f ) = | α | L ρ ( f ) , L ρ ( f + g ) L ρ ( f ) + L ρ ( g ) .   Thus A   is a self-adjoint linear subspace of C ( X )   which contains constant functions, and so A   is a matrix order unit space by Theorem 4.4 in [2.
Since L ρ   is lower semicontinuous, K = { f A : L ρ ( f ) 1 }   is an absolutely convex normed-closed (and hence is weakly closed) set in A   . K   determines a graded set K n = { K , i f n = 1 , { 0 } , i f n > 1 .   The minimal envelope K ^   of K   is the matrix bipolar K   of K   . K ^   is an absolutely matrix convex weakly closed graded set. We let ^ = ( L ^ n )   be the corresponding matrix gauge of K ^   . Since L ^ 1 = L ρ   is a Lipschitz seminorm, ^   is a matrix Lipschitz seminorm. ρ L ρ = ρ   implies that ^   is also a matrix Lip-norm(see Theorem 1.9 in [14and Proposition 7.5 in [21). Therefore, ( A , ^ )   is a quantized metric space. It is called the minimal quantized metric space of ( X , ρ )   . The maximal envelope K ˇ   of K   is the matrix polar ( K )   of K = ( K n )   , where K n = { K , i f n = 1 , { 0 } , i f n > 1 .   Similarly, K ˇ   is an absolutely matrix convex weakly closed graded set, and the corresponding matrix gauge ˇ   of K ˇ   makes A   into a quantized metric space. ( A , ˇ )   is called the maximal quantized metric space of ( X , ρ )   .
Moreover, if C = ( C n )   is an absolutely matrix convex weakly closed graded set with C 1 = K   , then K ^ C K ˇ ,   and the corresponding matrix gauge = ( L n )   satisfies L ˇ n L n L ^ n , n N ,   (see page 181 in [4). So ( A , )   is a quantized metric space. It is called a quantized metric space of ( X , ρ )   .
Example 3.5. Let ( A , L )   be a compact quantum metric space, that is, an order unit space ( A , e )   equipped with a seminorm L   , called Lip-norm, on A   such that L ( a ) = 0   if and only if a R e   , and the topology on the state space S ( A )   of A   from the metric ρ L ( μ , ν ) = sup { | μ ( a ) ν ( a ) | : L ( a ) 1 }   is the w *   -topology (see Definition 2.2 in [17). So ( S ( A ) , ρ L )   is an ordinary compact metric space. Let A   denote the set of Lipschitz functions on S ( A )   . By Example  3.4 , there exists a quantized metric space structure ( A , 1 )   of ( S ( A ) , ρ L )   , where 1 = ( L 1 , n )   . From Lemma 3.2 in [15, A A   and L 1 , 1 ( a ) L ( a )   for a A   . Let = ( n )   be the matrix norm determined by the matrix order on ( A , 1 )   . By the basic representation theorem of Kadison[7, we also have that a = a 1   for a A   . If L   is lower semicontinuous, the embedding of A   into A   is isometric, that is, a = a 1   and L ( a ) = L 1 , 1 ( a )   for all a A   , according to Theorem 4.1 in [15.
Set V = A ( A + i A ) .   We denote the restriction of 1   on V   by = ( L n )   . Then V   is a self-adjoint linear subspace of A   and contains the order unit of A   . So V   is a matrix order unit space. Because the D 1   -topology on C S ( A )   agrees with the BW-topology, the image of L 1 , 1 1 = { a A : L 1 , 1 ( a ) 1 }   in A ~ = A / C 1   is totally bounded for 1   by Theorem 5.3 in [20. Since L 1 1 L 1 , 1 1   , the image of L 1 1   in V ~ A ~   is totally bounded for 1   , and so, by Theorem 5.3 in [20, the D   -topology on C S ( V )   is the BW-topology. Therefore, ( V , )   is a quantized metric space, and the embedding of ( A , L )   into ( V , )   is an isometry if L   is lower semicontinuous.
Let ( V , 1 )   and ( W , 1 )   be matrix order unit spaces, and let φ : V W   be a unital completely positive linear mapping. Then we have the dual mapping φ : W * V *   determined by φ ( f ) ( v ) = f ( φ ( v ) )   . Let V ~ *   denote the dual space of V ~ = V / ( C 1 )   .
V ~ *   is just the subspace of V *   consisting of those f V *   such that f ( 1 ) = 0   . For any v M n ( V )   and g M m ( W * )   , we have g , φ n ( v ) = [ g k l ( φ ( v i j ) ) ] = [ ( φ ( g k l ) ) ( v i j ) ] = ( φ ) m ( g ) , v .   So φ c b = 1   , ( φ ) m ( M m ( W ~ * ) ) M m ( V ~ * )   and ( φ ) m ( C S m ( W ) ) C S m ( V )   .
Moreover, φ   is w *   -continuous. Let φ n c = ( φ ) n | C S n ( W )   for n N   . Then for v M n ( V )   , f i C S n i ( W )   and γ i M n i , m   satisfying i = 1 k γ i * γ i = 1 m   , we have
φ m c ( i = 1 k γ i * f i γ i ) , v = i = 1 k γ i * f i γ i , φ n ( v )
= i = 1 k ( γ i 1 n ) * f i , φ n ( v ) ( γ i 1 n )
= i = 1 k ( γ i 1 n ) * φ n i c ( f i ) , v ( γ i 1 n )
= i = 1 k γ i * φ n i c ( f i ) γ i , v .
So φ c = ( φ n c )   is a BW-continuous matrix affine mapping of C S ( W )   into C S ( V )   . In particular, φ c ( C S ( W ) ) = ( φ n c ( C S n ( W ) ) )   is a closed matrix convex subset of C S ( V )   .
Clearly φ n c   is injective if φ   is surjective.
Let   be a matrix Lipschitz seminorm on V   . On V *   , we define the matrix gauge = ( L n )   by L n ( f ) = sup { f , a : a L r 1 , r N } , f M n ( V * ) .   Then L n ( φ ψ ) = D L n ( φ , ψ )   for φ , ψ C S n ( V )   by Lemma 4.3 in [21.
Proposition 3.6. Let ( V , 1 )   and ( W , 1 )   be matrix order unit spaces, and let φ : V W   be a unital completely positive linear mapping which is surjective. Let   be a matrix Lipschitz seminorm on V   , and let W = ( L W , n )   be a sequence of the corresponding quotient seminorms on W   , defined by L W , n ( b ) = inf { L n ( a ) : φ n ( a ) = b } , b M n ( W ) .   Then
  • (1) W   is a matrix Lipschitz seminorm on W   ;
  • (2) φ   is a complete isometry for the matrix norms W   and   on W ~ *   and V ~ *   ;
  • (3) φ c   is a complete isometry for the corresponding matrix metrics D W   and D   ;
  • (4) If   is a matrix Lip-norm, then so is W   .
  • Proof. (1) For b 1 M m ( W ) , b 2 M n ( W )   , we have
    L W , m + n ( b 1 b 2 )
    = inf { L m + n ( a ) : φ m + n ( a ) = b 1 b 2 }
    inf { L m + n ( a 1 a 2 ) : φ m + n ( a 1 a 2 ) = b 1 b 2 }
    = inf { max { L m ( a 1 ) , L n ( a 2 ) } : φ m ( a 1 ) = b 1 , φ n ( a 2 ) = b 2 }
    = max { inf { L m ( a 1 ) : φ m ( a 1 ) = b 1 } , inf { L n ( a 2 ) : φ n ( a 2 ) = b 2 } }
    = max { L W , m ( b 1 ) , L W , n ( b 2 ) } .
    If α M m , n   , β M n , m   and b M n ( W )   , we have
    L W , m ( α b β ) = inf { L m ( a ) : φ m ( a ) = α b β }
    inf { L m ( α a β ) : φ n ( a ) = b }
    α β inf { L n ( a ) : φ n ( a ) = b }
    = α β L W , n ( b ) ,
    and
    L W , n ( b * ) = inf { L n ( a ) : φ n ( a ) = b * }
    = inf { L n ( a ) : φ n ( a * ) = b }
    = inf { L n ( a * ) : φ n ( a ) = b }
    = inf { L n ( a ) : φ n ( a ) = b }
    = L W , n ( b ) .
    Given [ λ i j ] M n   . We have
    L W , n ( [ λ i j 1 ] ) = inf { L n ( a ) : φ n ( a ) = [ λ i j 1 ] }
    L n ( [ λ i j 1 ] ) = 0 ,
    and so L W , n ( [ λ i j 1 ] ) = 0   . If b = [ b i j ] M n ( W )   with L W , n ( b ) = 0   , then L W , 1 ( b i j ) = 0 , i , j = 1 , 2 , , n .   Letting b i j = c i j + i d i j ( i , j = 1 , 2 , , n )   , where c i j * = c i j , d i j * = d i j   , we get L W , 1 ( c i j ) = 0 , L W , 1 ( d i j ) = 0 , i , j = 1 , 2 , , n .   Since L W , 1 ( c i j ) = inf { L 1 ( a ) : φ 1 ( a ) = c i j }   and φ   is positive, we have L W , 1 ( c i j ) = inf { L 1 ( a ) : φ 1 ( a ) = c i j , a = a * } .   Now by Proposition 3.1 in [17, there exists an α i j R   such that c i j = α i j 1   .
    Similarly, there exists a β i j R   such that d i j = β i j 1   . Therefore, b M n ( C 1 )   .
    Thus W   is a matrix Lipschitz seminorm on W   .
    (2) Let f M m ( W ~ * )   . For any a M n ( V )   we clearly have L W , n ( φ n ( a ) ) L n ( a )   , and so if L n ( a ) 1   we have ( φ ) m ( f ) , a = f , φ n ( a ) L W , m ( f ) .   Consequently, L m ( ( φ ) m ( f ) ) L W , m ( f )   . But let δ > 0   be given, and let b M n ( W )   with L W , m ( b ) 1   . Then there is an a M n ( V )   such that φ n ( a ) = b   and L n ( a ) 1 + δ   . Thus, L n ( a / ( 1 + δ ) ) 1   . Consequently,
    L m ( ( φ ) m ( f ) ) ( φ ) m ( f ) , a / ( 1 + δ )
    = f , φ n ( a ) / ( 1 + δ )
    = f , b / ( 1 + δ ) .
    Taking the supremum over b M n ( W )   with L W , n ( b ) 1   , we see that L m ( ( φ ) m ( f ) ) L W , m ( f ) / ( 1 + δ ) .   Since δ   is arbitrary, we obtain that L m ( ( φ ) m ( f ) ) L W , m ( f )   . Thus φ   is a complete isometry.
    (3) By (2), we have
    D L W , n ( φ , ψ ) = L W , n ( φ ψ )
    = L n ( ( φ ) n ( φ ψ ) )
    = D L n ( φ n c ( φ ) , φ n c ( ψ ) ) ,
    where φ , ψ C S n ( W )   , that is, φ c   is a complete isometry for the corresponding matrix metrics D W   and D   . (4) Suppose that   is a matrix Lip-norm. Since φ   is w *   -continuous, φ   is surjective, and C S n ( V )   is BW-compact, ( φ ) n   is a homeomorphism of C S n ( W )   onto ( φ ) n ( C S n ( W ) ) C S n ( V )   . Because D L n   gives the BW-topology on C S n ( V )   , D L n | ( φ ) n ( C S n ( W ) )   gives the relative topology of ( φ ) n ( C S n ( W ) )   . According to (3), D L W , n   gives the BW-topology on C S n ( W )   . Therefore, W   is a matrix Lip-norm.
Notation 3.7. Under the conditions of Propositions  3.6 we will say that   induces W   via φ   .
For a matrix convex set K   in a locally convex vector space, let A ( K )   be the set of all matrix affine mappings from K   to C   (see §6 in [21). On the other hand, we have
Proposition 3.8. Let ( V , 1 )   be a matrix order unit space, and let K = ( K n )   be a compact matrix convex subset of C S ( V )   . View the elements of V   as matrix affine mapping from C S ( V )   to C   (Proposition 6.1 in [21), and let W   consist of their restrictions to K   , with φ   the restriction mapping of V   onto W   . Then ( W , φ ( 1 ) )   is a matrix order unit space, and K = φ c ( C S ( W ) )   .
  • Proof. Clearly, with the natural matrix order structure on W   and the order unit φ ( 1 )   , ( W , φ ( 1 ) )   is a matrix order unit space.
    For φ K n C S n ( V )   , we define the mapping ψ : W M n   by ψ ( φ ( v ) ) = φ ( v )   .
    Then ψ C S n ( W )   and ( φ n c ( ψ ) ) ( v ) = ψ ( φ ( v ) ) = φ ( v )   for v V   . Thus K n φ n c ( C S n ( W ) )   .
    Suppose that φ 0 C S n ( V )   and φ 0 / K n   . By Theorem 1.6 in [19, there is a v = [ v i j ] M n ( V )   and a self-adjoint α = [ α i j ] M n   such that R e φ , v α 1 r ,   for all r N   , φ K r   , and R e φ 0 , v α 1 n .   So we obtain φ n ( R e [ α i j 1 v i j ] ) 0   for all r N   and φ K r   . Thus φ n ( R e [ α i j 1 v i j ] ) 0   in W   . If φ 0 = φ n c ( ψ 0 )   for some ψ 0 C S n ( W )   , we would then have that R e φ 0 , v = R e ψ 0 , φ n ( v ) = α 1 n R e ψ 0 , φ n ( [ α i j 1 v i j ] ) = α 1 n ψ 0 , φ n ( R e [ α i j 1 v i j ] ) α 1 n   . Thus, φ 0 / φ c ( C S n ( W ) )   . Therefore, K = φ c ( C S ( W ) )   .
Notation 3.9. We will call the matrix order unit space ( W , φ ( 1 ) )   in the Proposition  3.8 the quotient of ( V , 1 )   with respect to K   , and will identify C S ( W )   with K   . When ( V , )   is a quantized metric space, ( W , W )   is a quantized metric space by Proposition  3.6 . ( W , W )   is called the quotient space of ( V , )   with respect to K   and φ   .
Proposition 3.10. Let ( V 1 , 1 )   , ( V 2 , 1 )   and ( V 3 , 1 )   be matrix order unit spaces. Suppose that φ : V 1 V 2   and ψ : V 2 V 3   are unital completely positive linear mappings which are surjective. Denote φ = ψ φ   . If   is a matrix Lipschitz seminorm on V 1   , V 2   and V 3   are the induced matrix Lipschitz seminorms of   via φ   and φ   , respectively, and V 23   is the induced matrix Lipschitz seminorm of V 2   via ψ   , then V 23 = V 3   .
  • Proof. This follows by exactly the same argument used for quantum Gromov-Hausdorff distance in [17.

4 Quantized Gromov-Hausdorff distance

As in the situation of compact quantum metric spaces, we need a corresponding notion of bridge for estimating distance between quantized metric spaces.
Let ( V 1 , 1 )   and ( V 2 , 2 )   be two quantized metric spaces with the matrix norms 1 = ( 1 , n )   and 2 = ( 2 , n )   determined by their matrix orders on ( V 1 , 1 )   and ( V 2 , 1 )   , respectively. We form the direct sum V 1 V 2   of operator spaces (see §2.6 in [13). ( V 1 V 2 , ( 1 , 1 ) )   becomes a matrix order unit space.
Definition 4.1. Let ( V 1 , 1 )   and ( V 2 , 2 )   be quantized metric spaces.
A matrix bridge between ( V 1 , 1 )   and ( V 2 , 2 )   is a matrix seminorm N   on V 1 V 2   such that
  • (1) N   is matrix continuous for the matrix norm   on V 1 V 2   , that is, each N n   is continuous for n   on M n ( V 1 V 2 )   .
  • (2) N n ( ( a , b ) * ) = N n ( a , b )   for a M n ( V 1 )   and b M n ( V 2 )   and n N   .
  • (3) N 1 ( 1 , 1 ) = 0   but N 1 ( 1 , 0 ) 0   .
  • (4) For any n N   , a M n ( V 1 )   and ε > 0   , there is a b M n ( V 2 )   such that max { L 2 , n ( b ) , N n ( a , b ) } L 1 , n ( a ) + ε ,   and similarly for V 1   and V 2   interchanged.
Example 4.2. Suppose ( V 1 , 1 )   and ( V 2 , 2 )   are quantized metric spaces. Choose φ 1 C S 1 ( V 1 )   and ψ 1 C S 1 ( V 2 )   . For n N   , we define N n : M n ( V 1 V 2 ) [ 0 , + )   by N n ( a , b ) = φ 1 , a ψ 1 , b .   Then N = ( N n )   is a matrix seminorm on V 1 V 2   , and satisfies the conditions (1), (2) and (3) of Definition  4.1 . For any a M n ( V 1 )   and ε > 0   , choose b = [ φ 1 ( a i j ) 1 ] M n ( V 2 )   . Then we have max { L 2 , n ( b ) , N n ( a , b ) } = 0 L 1 , n ( a ) + ε ,   and similarly if we are given b M n ( V 2 )   . So N   is a matrix bridge between ( V 1 , 1 )   and ( V 2 , 2 )   .
Proposition 4.3. Let N   be a matrix bridge between quantized metric spaces ( V 1 , 1 )   and ( V 2 , 2 )   . Define = ( L n )   on V 1 V 2   by L n ( a , b ) = max { L 1 , n ( a ) , L 2 , n ( b ) , N n ( a , b ) } , a M n ( V 1 ) , b M n ( V 2 ) , n N .   Let π 1   and π 2   be the projections from V 1 V 2   onto V 1   and V 2   , respectively, which are unital completely positive linear surjective mappings. Then   is a matrix Lip-norm on ( V 1 V 2 , ( 1 , 1 ) )   , and it induces V 1   and V 2   via π 1   and π 2   , respectively. If 1   and 2   are lower semicontinuous, then so is   .
  • Proof. For a i M n ( V i )   and b i M m ( V i ) , i = 1 , 2   , we have
    L n + m ( a 1 b 1 , a 2 b 2 )
    = max { L 1 , n + m ( a 1 b 1 ) , L 2 , n + m ( a 2 b 2 ) , N n + m ( a 1 b 1 , a 2 b 2 ) }
    = max { max { L 1 , n ( a 1 ) , L 1 , m ( b 1 ) } , max { L 2 , n ( a 2 ) , L 2 , m ( b 2 ) } ,
    max { N n ( a 1 , a 2 ) , N m ( b 1 , b 2 ) } }
    = max { max { L 1 , n ( a 1 ) , L 2 , n ( a 2 ) , N n ( a 1 , a 2 ) } ,
    max { L 1 , m ( b 1 ) , L 2 , m ( b 2 ) , N m ( b 1 , b 2 ) } }
    = max { L n ( a 1 , a 2 ) , L m ( b 1 , b 2 ) } ,
    and
    L n ( ( a 1 , a 2 ) * ) = L n ( a 1 * , a 2 * )
    = max { L 1 , n ( a 1 * ) , L 2 , n ( a 2 * ) , N n ( a 1 * , a 2 * ) }
    = max { L 1 , n ( a 1 ) , L 2 , n ( a 2 ) , N n ( a 1 , a 2 ) }
    = L n ( a 1 , a 2 ) ,
    and for α M m , n   and β M n , m   , we have
    L m ( α ( a 1 , a 2 ) β )
    = L m ( α a 1 β , α a 2 β )
    = max { L 1 , m ( α a 1 β ) , L 2 , m ( α a 2 β ) , N m ( α a 1 β , α a 2 β ) }
    max { α L 1 , n ( a 1 ) β , α L 2 , n ( a 2 ) β , α N n ( a 1 , a 2 ) β }
    = α L n ( a 1 , a 2 ) β .
    Thus   is a matrix seminorm. Since L 1 ( a s t ) L n ( [ a i j ] ) i , j = 1 n L 1 ( a i j )   for s , t = 1 , 2 , , n   and [ a i j ] M n ( V 1 V 2 )   , L n ( [ λ i j ( 1 , 1 ) ] ) = 0   for [ λ i j ] M n   .
    If L n ( [ ( a i j , b i j ) ] ) = 0   , then L 1 , n ( [ a i j ] ) = L 2 , n ( [ b i j ] ) = 0   , and hence a i j = λ i j 1   and b i j = μ i j 1   , i , j = 1 , 2 , , n   , where λ i j , μ i j C   . From N n ( [ a i j , b i j ] ) = 0   and N 1 ( a s t , b s t ) N n ( [ a i j , b i j ] )   for s , t = 1 , 2 , , n   , we have N 1 ( λ i j 1 , μ i j 1 ) = 0 , i , j = 1 , 2 , , n ,   and so for i , j = 1 , 2 , , n   ,
    | λ i j μ i j | N 1 ( 1 , 0 ) = N 1 ( ( λ i j μ i j ) 1 , 0 )
    = N 1 ( ( λ i j 1 , μ i j 1 ) ( μ i j 1 , μ i j 1 ) )
    N 1 ( λ i j 1 , μ i j 1 ) + N 1 ( μ i j 1 , μ i j 1 )
    = 0 .
    Thus [ ( a i j , b i j ) ] = [ ( λ i j 1 , λ i j 1 ) ] = [ λ i j ( 1 , 1 ) ]   . So   is a matrix Lipschitz seminorm.
    Similar to the same argument used in Theorem 5.2 of [17, we have that   induces 1   and 2   via π 1   and π 2   , respectively. By Proposition 3.1 in [20, Proposition 7.5 in [21and Theorem 5.2 in [17(see also §2 in [17), the D   -topology on C S ( V 1 V 2 )   agrees with the BW-topology. Therefore,   is a matrix Lip-norm on ( V 1 V 2 , ( 1 , 1 ) )   .
    Suppose that 1   and 2   are lower semicontinuous. Clearly,   is lower semicontinuous since N   is matrix continuous.
Notation 4.4. We will denote by ( 1 , 2 )   the set of matrix Lip-norms on V 1 V 2   which induce both 1   and 2   via π 1   and π 2   , respectively.
By Proposition  4.3 and Example  4.2 , ( 1 , 2 )   . From Proposition  3.6 , we can view C S ( V 1 )   and C S ( V 2 )   as closed matrix convex subsets of C S ( V 1 V 2 )   .
Now we introduce our notion of distance for quantized metric spaces.
Definition 4.5. Let ( V 1 , 1 )   and ( V 2 , 2 )   be quantized metric spaces. We define the quantized Gromov-Hausdorff distance d i s t N C ( V 1 , V 2 )   between them by d i s t N C ( V 1 , V 2 ) = inf = ( L n ) ( 1 , 2 ) sup n N { n 2 d i s t H D L n ( C S n ( V 1 ) , C S n ( V 2 ) ) } ,   where d i s t H D L n ( C S n ( V 1 ) , C S n ( V 2 ) )   is the Hausdorff distance between C S n ( V 1 )   and C S n ( V 2 )   for D L n   .
Given a quantized metric space ( V , )   , we define its diameter d i a m ( V , )   to be the diameter of C S 1 ( V )   with respect to D L 1   . The following proposition indicates that the quantized Gromov-Hausdorff distance is always finite.
Proposition 4.6. Let ( V 1 , 1 )   and ( V 2 , 2 )   be quantized metric spaces.
Then d i s t N C ( V 1 , V 2 ) 2 ( d i a m ( V 1 , 1 ) + d i a m ( V 2 , 2 ) ) .  
  • Proof. Choosing arbitrarily α > 0   , φ 0 C S 1 ( V 1 ) , ψ 0 C S 1 ( V 2 )   , we set N n ( a , b ) = α 1 φ 0 , a ψ 0 , b , a M n ( V 1 ) , b M n ( V 2 ) , n N .   As Example  4.2 , N = ( N n )   is a matrix bridge between ( V 1 , 1 )   and ( V 2 , 2 )   . By Proposition  4.3 , = ( L n )   , where L n ( a , b ) = max { L 1 , n ( a ) , L 2 , n ( b ) , N n ( a , b ) } , a M n ( V 1 ) , b M n ( V 2 ) , n N ,   is a matrix Lip-norm in ( 1 , 2 )   . Then for φ C S n ( V 1 ) , ψ C S n ( V 2 )   , and ( a , b ) M n ( V 1 V 2 )   with L n ( a , b ) 1   , we have
    φ , a ψ , b
    φ , a φ 0 φ 0 n , a
    + φ 0 φ 0 n , a ψ 0 ψ 0 n , b
    + ψ 0 ψ 0 n , b ψ , b
    i , j φ i j δ i j φ 0 , a + α + i , j ψ i j δ i j ψ 0 , b
    If n = 1   , we get φ , a ψ , b d i a m ( V 1 , 1 ) + α + d i a m ( V 2 , 2 ) ,   by Proposition 3.1 in [20. If n > 1   , similar to the proof of Proposition 4.2 in [20, there are φ i j ( k ) C S 1 ( V 1 ) , k = 1 , 2 , 3 , 4   , such that φ i j δ i j φ 0 = φ i j ( 1 ) φ i j ( 2 ) + i ( φ i j ( 3 ) φ i j ( 4 ) ) .   Since L 1 , n ( a ) L n ( a , b ) 1   , we obtain
    i , j φ i j δ i j φ 0 , a
    i , j ( φ i j ( 1 ) , a φ i j ( 2 ) , a + φ i j ( 3 ) , a φ i j ( 4 ) , a )
    i , j ( D L 1 , 1 ( φ i j ( 1 ) , φ i j ( 2 ) ) + D L 1 , 1 ( φ i j ( 3 ) , φ i j ( 4 ) ) )
    2 n 2 d i a m ( V 1 , 1 ) .
    Applying the same argument, we have i , j ψ i j δ i j ψ 0 , b 2 n 2 d i a m ( V 2 , 2 ) .   Hence φ , a ψ , b 2 n 2 ( d i a m ( V 1 , 1 ) + α + d i a m ( V 2 , 2 ) ) .   The arbitrariness of α   implies that d i s t N C ( V 1 , V 2 ) 2 ( d i a m ( V 1 ) + d i a m ( V 2 ) )   by Proposition 3.1 in [20.
It is clear that the quantized Gromov-Hausdorff distance is symmetric in V 1   and V 2   . We come to prove that it satisfies the triangle inequality.
Theorem 4.7. If ( V 1 , 1 )   , ( V 2 , 2 )   and ( V 3 , 3 )   be quantized metric spaces, then d i s t N C ( V 1 , V 3 ) d i s t N C ( V 1 , V 2 ) + d i s t N C ( V 2 , V 3 ) .  
  • Proof. Given ε > 0   . Then there are 12 ( 1 , 2 )   and 23 ( 2 , 3 )   such that sup n N { n 2 d i s t H D L 12 , n ( C S n ( V 1 ) , C S n ( V 2 ) ) } d i s t N C ( V 1 , V 2 ) + ε   and sup n N { n 2 d i s t H D L 23 , n ( C S n ( V 2 ) , C S n ( V 3 ) ) } d i s t N C ( V 2 , V 3 ) + ε .   We define = ( L n )   on V 1 V 2 V 3   by L n ( a 1 , a 2 , a 3 ) = max { L 12 , n ( a 1 , a 2 ) , L 23 , n ( a 2 , a 3 ) } .   Then for a i M n ( V i )   and b i M m ( V i ) , i = 1 , 2 , 3   , we have
    L n + m ( a 1 b 1 , a 2 b 2 , a 3 b 3 )
    = max { L 12 , n + m ( a 1 b 1 , a 2 b 2 ) , L 23 , n + m ( a 2 b 2 , a 3 b 3 ) }
    = max { max { L 12 , n ( a 1 , a 2 ) , L 12 , m ( b 1 , b 2 ) } , max { L 23 , n ( a 2 , a 3 ) , L 23 , m ( b 2 , b 3 ) } }
    = max { max { L 12 , n ( a 1 , a 2 ) , L 23 , n ( a 2 , a 3 ) } , max { L 12 , m ( b 1 , b 2 ) , L 23 , m ( b 2 , b 3 ) } }
    = max { L n ( a 1 , a 2 , a 3 ) , L m ( b 1 , b 2 , b 3 ) } ,
    and
    L n ( ( a 1 , a 2 , a 3 ) * ) = L n ( a 1 * , a 2 * , a 3 * )
    = max { L 12 , n ( a 1 * , a 2 * ) , L 23 , n ( a 2 * , a 3 * ) }
    = max { L 12 , n ( a 1 , a 2 ) , L 23 , n ( a 2 , a 3 ) }
    = L n ( a 1 , a 2 , a 3 ) ,
    and for α M m , n   and β M n , m   , we have
    L m ( α ( a 1 , a 2 , a 3 ) β ) = L m ( α a 1 β , α a 2 β , α a 3 β )
    = max { L 12 , m ( α a 1 β , α a 2 β ) , L 23 , m ( α a 2 β , α a 3 β ) }
    max { α L 12 , n ( a 1 , a 2 ) β , α L 23 , n ( a 2 , a 3 ) β }
    = α L n ( a 1 , a 2 , a 3 ) β .
    L n ( a 1 , a 2 , a 3 ) = 0   if and only if L 12 , n ( a 1 , a 2 ) = 0   and L 23 , n ( a 2 , a 3 ) = 0   , and this is equivalent to that ( a 1 , a 2 , a 3 ) M n ( C ( 1 , 1 , 1 ) )   . Therefore,   is a matrix Lipschitz seminorm.
    Similar to the same argument used in Lemma 4.4 of [17, we have that   induces 12   , 23   , 1   , 2   and 3   for the evident quotient mappings by Proposition  3.10 .
    By Proposition 3.1 in [20, Proposition 7.5 in [21and Lemma 4.4 in [17(see also §2 in [17), the D   -topology on C S ( V 1 V 2 V 3 )   agrees with the BW-topology. So   is a matrix Lip-norm on ( V 1 V 2 V 3 , ( 1 , 1 , 1 ) )   .
    By Proposition  3.6 , we have sup n N { n 2 d i s t H D L n ( C S n ( V 1 ) , C S n ( V 2 ) ) } d i s t N C ( V 1 , V 2 ) + ε ,   sup n N { n 2 d i s t H D L n ( C S n ( V 2 ) , C S n ( V 3 ) ) } d i s t N C ( V 2 , V 3 ) + ε   and d i s t N C ( V 1 , V 3 ) sup n N { n 2 d i s t H D L n ( C S n ( V 1 ) , C S n ( V 3 ) ) } .   So
    d i s t N C ( V 1 , V 3 )
    sup n N { n 2 d i s t H D L n ( C S n ( V 1 ) , C S n ( V 3 ) ) }
    sup n N { n 2 d i s t H D L n ( C S n ( V 1 ) , C S n ( V 2 ) )
    + n 2 d i s t H D L n ( C S n ( V 2 ) , C S n ( V 3 ) ) }
    sup n N { n 2 d i s t H D L n ( C S n ( V 1 ) , C S n ( V 2 ) ) }
    + sup n N { n 2 d i s t H D L n ( C S n ( V 2 ) , C S n ( V 3 ) ) }
    d i s t N C ( C S n ( V 1 ) , C S n ( V 2 ) ) + d i s t N C ( C S n ( V 2 ) , C S n ( V 3 ) ) + 2 ε .
    Since ε   is arbitrary, we obtain d i s t N C ( V 1 , V 3 ) d i s t N C ( V 1 , V 2 ) + d i s t N C ( V 2 , V 3 ) .  
Proposition 4.8. Let ( V , )   be a quantized metric space, and let K ( 1 ) = ( K n ( 1 ) )   and K ( 2 ) = ( K n ( 2 ) )   be compact matrix convex subsets of C S ( V )   .
If ( V j , j )   is the quotient space of ( V , )   with respect to K ( j )   and φ ( j )   , j = 1 , 2   , then we have d i s t N C ( V 1 , V 2 ) sup k N { k 2 d i s t H D L k ( K k ( 1 ) , K k ( 2 ) ) } .  
  • Proof. Let p 1   and p 2   be the projections from V V   onto the first space V   and the second space V   , respectively. Denote G n ( j ) = ( p 1 c ) n ( K n ( j ) ) , H n ( j ) = ( p 2 c ) n ( K n ( j ) ) , j = 1 , 2 , n N ,   and set G ( j ) = ( G n ( j ) )   , ( j ) = ( H n ( j ) )   , j = 1 , 2   , and K = ( K n ) = m c o ¯ ( G ( 1 ) ( 2 ) )   , the BW-closed matrix convex hull of the graded set ( G n ( 1 ) H n ( 2 ) )   . Let ( W , φ ( 1 1 ) )   be the quotient of ( V V , 1 1 )   with respect to K   . Then K = φ c ( C S ( W ) )   by Proposition  3.8 .
    For ( a , b ) K e r φ   , we have φ ( a , b ) = 0   , that is, ( a , b ) , φ = 0 n   for φ K n   .
    This is equivalent to ( a , b ) , φ = 0 n   for φ G n ( 1 ) H n ( 2 )   , n N   since ( a , b ) A ( K )   . And this holds if and only if a , φ 1 = 0 n   and b , φ 2 = 0 n   for φ 1 G n ( 1 )   and φ 2 H n ( 2 )   , n N   , that is, if and only if a K e r φ ( 1 )   and b K e r φ ( 2 )   .
    So K e r φ = K e r φ ( 1 ) K e r φ ( 2 )   . And thus there is a complete order isomorphism Ψ   from W   onto V 1 V 2   .
    Given ε > 0   . We define a matrix seminorm N = ( N n )   on V V   by N n ( a , b ) = ε 1 a b n , a , b M n ( V ) .   Then N   is a matrix bridge between ( V , )   and ( V , )   , and Q = ( Q n ) ( , )   by Proposition  4.3 , where Q n ( a , b ) = max { L n ( a ) , L n ( b ) , N n ( a , b ) } , a , b M n ( V ) , n N .   Thus Q   is a matrix Lip-norm on ( V V , ( 1 , 1 ) )   . Let P = ( P n )   and ( W , P )   be the quotient space of ( V V , Q )   with respect to K   and φ   . Then P ( 1 , 2 )   by Propositon  3.10 .
    Since D P k ( φ 1 , φ 2 ) = D Q k ( φ k c ( φ 1 ) , φ k c ( φ 2 ) )   for φ 1 , φ 2 C S k ( W )   , we have that d i s t H D P k ( C S k ( V 1 ) , C S k ( V 2 ) ) = d i s t H D Q k ( G k ( 1 ) , H k ( 2 ) )   . For ψ K k ( 2 )   , we have
    D Q k ( ( p 1 c ) k ( ψ ) , ( p 2 c ) k ( ψ ) )
    = sup { ( p 1 c ) k ( ψ ) , ( a , b ) ( p 2 c ) k ( ψ ) , ( a , b ) : Q r ( a , b ) 1 , r N }
    sup { ψ , a ψ , b : N r ( a , b ) 1 , r N }
    ε ,
    that is, d i s t H D Q k ( G k ( 2 ) , H k ( 2 ) ) ε   . Because Q ( , )   , we get that d i s t H D Q k ( G k ( 1 ) , G k ( 2 ) ) = d i s t H D L k ( K k ( 1 ) , K k ( 2 ) ) .   So
    d i s t N C ( V 1 , V 2 ) sup k N { k 2 d i s t H D P k ( C S k ( V 1 ) , C S k ( V 2 ) ) }
    = sup k N { k 2 d i s t H D Q k ( G k ( 1 ) , H k ( 2 ) ) }
    = sup k N { k 2 d i s t H D Q k ( G k ( 1 ) , G k ( 2 ) ) }
    + sup k N { k 2 d i s t H D Q k ( G k ( 2 ) , H k ( 2 ) ) }
    sup k N { k 2 d i s t H D Q P k ( G k ( 1 ) , G k ( 2 ) ) + k 2 ε }
    sup k N { k 2 d i s t H D Q k ( G k ( 1 ) , G k ( 2 ) ) } + ε
    = sup k N { k 2 d i s t H D L k ( K k ( 1 ) , K k ( 2 ) ) } + ε
    Since ε   is arbitrary, we obtain the desired inequality.
Let ( A , L A )   and ( B , L B )   be compact quantum metric spaces. The quantum Gromov-Hausdorff distance between them is defined by d i s t q ( A , B ) = inf d i s t H ρ L ( S ( A ) , S ( B ) ) ,   where the infimum is taken over all Lip-norms L   on A B   which induce L A   and L B   (see Definition 4.2 in [17).
Proposition 4.9. Let ( A j , L j )   for j = 1 , 2   be compact quantum metric spaces, and let ( V j , j )   be an associated quantized metric space of ( A j , L j )   (see Example  3.5 ). Then d i s t q ( A 1 , A 2 ) d i s t N C ( V 1 , V 2 ) .  
  • Proof. Suppose Q ( 1 , 2 )   . Then Q V j , 1 = L j , 1   for j = 1 , 2   and L j , 1 ( a ) = L j s ( a )   for a A j   , where L j s = L ρ L j   (see Example  3.5 ). So for a A 1   , we have
    L 1 s ( a ) = L 1 , 1 ( a ) = Q V j , 1 ( a )
    = inf { Q 1 ( a 1 , b 1 ) : π 1 ( a 1 , b 1 ) = a , ( a 1 , b 1 ) = V 1 V 2 }
    = inf { Q 1 ( a , b 1 ) : b 1 V 2 }
    inf { Q 1 ( a , b ) : b A 2 }
    = inf { R ( a , b ) : b A 2 }
    = R A 1 ( a ) ,
    where π 1   is the projection from V 1 V 2   onto V 1   and R   is the restriction of Q 1   to A 1 A 2   . Denote c = inf { Q 1 ( a , b ) : b V 2 }   . Let ε > 0   be given. Then there is a y V 2   such that Q 1 ( a , y ) c + ε   . Setting x = 1 2 ( y + y * )   , we have that x A 2   and
    R ( a , x ) = Q 1 ( a , x )
    = Q 1 ( a , 1 2 ( y + y * ) )
    1 2 Q 1 ( ( a , y ) + ( a , y ) * )
    1 2 ( Q 1 ( a , y ) + Q 1 ( ( a , y ) * ) )
    = Q 1 ( a , y )
    c + ε .
    Thus L 1 s ( a ) = R A 1 ( a )   for a A 1   . Similarly, we have that L 2 s ( b ) = R A 2 ( b )   for b A 2   . So R ( L 1 s , L 2 s )   .
    For φ C S 1 ( V 1 )   and ψ C S 1 ( V 2 )   , let φ 1 = φ | A 1   and ψ 1 = ψ | A 2   . Then φ 1 S ( A 1 )   and ψ 1 S ( A 2 )   . Since Q 1 ( ( a , b ) * ) = Q 1 ( a , b )   , we obtain
    D Q 1 ( φ , ψ ) = sup { | φ ( c ) ψ ( d ) | : Q 1 ( c , d ) 1 , ( c , d ) V 1 V 2 }
    = sup { | φ ( c ) ψ ( d ) | : Q 1 ( c , d ) 1 , ( c , d ) = ( c , d ) * V 1 V 2 }
    = sup { | φ 1 ( c ) ψ 1 ( d ) | : R ( c , d ) 1 , ( c , d ) A 1 A 2 }
    = ρ R ( φ 1 , ψ 1 ) ,
    (see §2 in [17). So d i s t H ρ R ( S ( A 1 ) , S ( A 2 ) ) = d i s t H D Q 1 ( C S 1 ( V 1 ) , C S 1 ( V 2 ) ) .   Therefore, by Theorem 4.3 and Proposition 7.1 in [17, we have
    d i s t q ( A 1 , A 2 ) d i s t q ( ( A 1 , L 1 ) , ( A 1 , L 1 s ) ) + d i s t q ( ( A 1 , L 1 s ) , ( A 2 , L 2 s ) )
    + d i s t q ( ( A 2 , L 2 s ) , ( A 2 , L 2 ) )
    = d i s t q ( ( A 1 , L 1 s ) , ( A 2 , L 2 s ) )
    d i s t H ρ R ( S ( A 1 ) , S ( A 2 ) )
    = d i s t H D Q 1 ( C S 1 ( V 1 ) , C S 1 ( V 2 ) )
    sup n N { n 2 d i s t H D Q n ( C S n ( V 1 ) , C S n ( V 2 ) ) } .
    Consequently, d i s t q ( A 1 , A 2 ) d i s t N C ( V 1 , V 2 )   .

5 Distance zero

In this section, we show that d i s t N C ( V 1 , V 2 ) = 0   is equivalent to the existence of a complete isometry between them in the following sense.
If ( V , )   is a quantized metric space, then D   is the largest lower semicontinuous matrix Lip-norm smaller than   by Corollary 4.5 in [21. From Proposition 7.1 in [21, D   extends uniquely to a closed matrix Lip-norm c   on the subspace V c = { a V ¯ : L 1 c ( a ) < + }   , where V ¯   is the completion of V   for its matrix norm.
Definition 5.1. Let ( V 1 , 1 )   and ( V 2 , 2 )   be quantized metric spaces. By a complete isometry from ( V 1 , 1 )   onto ( V 2 , 2 )   we mean a unital complete order isomorphism Φ   from V 1 c   onto V 2 c   such that 1 c = 2 c Φ   , that is, L 1 , n c = L 2 , n c Φ n   for all n N   .
Lemma 5.2. Let ( V , )   be a quantized metric space. Then d i s t N C ( V , V c ) = 0 , d i s t N C ( ( V , ) , ( V , D ) ) = 0 .  
  • Proof. Let ε > 0   be given, and define N n ( a 1 , a 2 ) = ε 1 a 1 a 2 n ,   for a 1 M n ( V )   , a 2 M n ( V c )   and n N   . Clearly N = ( N n )   is a matrix continuous matrix seminorm on V V c   , and N 1 ( 1 , 1 ) = 0   and N 1 ( 1 , 0 ) = ε 1 0   .
    For n N   and a 1 M n ( V )   and δ > 0   , setting a 2 = a 1 M n ( V c )   , we have max { L n c ( a 2 ) , N n ( a 1 , a 2 ) } = L n c ( a 2 ) = L D L n ( a 2 ) L n ( a 2 ) < L n ( a 2 ) + δ   by Proposition 3.6, Proposition 7.1 and Proposition 3.4 in [21. Given n N   and a 2 M n ( V c )   and δ > 0   . By Lemma 7.3 in [21, there is a sequence { a 1 ( k ) }   of elements in M n ( V )   such that L n ( a 1 ( k ) ) L n c ( a 2 )   and { a 1 ( k ) }   converges to a 2   in norm. Consequently, we can find an a 1 ( k 0 )   such that ε 1 a 1 ( k 0 ) a 2 n L n c ( a 2 ) + δ   .
    So max { L n ( a 1 ( k 0 ) ) , N n ( a 1 ( k 0 ) , a 2 ) } L n c ( a 2 ) + δ   . Thus N   is a matrix bridge between ( V , )   and ( V c , c )   .
    Define L n ( a 1 , a 2 ) = max { L n ( a 1 ) , L n c ( a 2 ) , N n ( a 1 , a 2 ) }   for a 1 M n ( V )   , a 2 M n ( V c )   and n N   . By Proposition  4.3 , = ( L n ) ( V , V c )   . For n N   and φ C S n ( V c )   , we have that ψ = φ | V C S n ( V )   , and hence
    D L n ( ψ , φ )
    = sup { ( π 1 ) n c ( ψ ) , ( a 1 , a 2 ) ( π 2 ) n c ( φ ) , ( a 1 , a 2 ) :
    L r ( a 1 , a 2 ) 1 , ( a 1 , a 2 ) M r ( V V c ) , r N }
    = sup { ψ , a 1 φ , a 2 :
    L r ( a 1 , a 2 ) 1 , ( a 1 , a 2 ) M r ( V V c ) , r N }
    = sup { φ , a 1 a 2 :
    L r ( a 1 , a 2 ) 1 , ( a 1 , a 2 ) M r ( V V c ) , r N }
    sup { a 1 a 2 r : L r ( a 1 , a 2 ) 1 , ( a 1 , a 2 ) M r ( V V c ) , r N }
    ε ,
    where π 1   and π 2   are the projections from V V c   onto V   and V c   , respectively. For n N   and φ C S n ( V )   , there is a ψ C S n ( V c )   such that ψ | V = φ   by Arveson's extension theorem. So
    D L n ( φ , ψ )
    = sup { ( π 1 ) n c ( φ ) , ( a 1 , a 2 ) ( π 2 ) n c ( ψ ) , ( a 1 , a 2 ) :
    L r ( a 1 , a 2 ) 1 , ( a 1 , a 2 ) M r ( V V c ) , r N }
    = sup { φ , a 1 ψ , a 2 :
    L r ( a 1 , a 2 ) 1 , ( a 1 , a 2 ) M r ( V V c ) , r N }
    = sup { ψ , a 1 a 2 :
    L r ( a 1 , a 2 ) 1 , ( a 1 , a 2 ) M r ( V V c ) , r N }
    sup { a 1 a 2 r : L r ( a 1 , a 2 ) 1 , ( a 1 , a 2 ) M r ( V V c ) , r N }
    ε .
    Thus d i s t H D L n ( C S n ( V ) , C S n ( V c ) ) ε   for n N   , and so sup n N { d i s t H D L n ( C S n ( V ) , C S n ( V c ) ) } ε .   Therefore, d i s t N C ( V , V c ) ε   . By the arbitrariness of ε   , we obtain d i s t N C ( V , V c ) = 0 .   By Proposition 3.4 in [21and the proof of Theorem 4.4 in [21, we can prove that d i s t N C ( ( V , ) , ( V , D ) ) = 0   similarly.
Theorem 5.3. Suppose ( V 1 , 1 )   and ( V 2 , 2 )   are quantized metric spaces.
If there exists a complete isometry Φ   from ( V 1 , 1 )   onto ( V 2 , 2 )   , then d i s t N C ( V 1 , V 2 ) = 0 .  
  • Proof. For ε > 0   , we define N n ( a 1 , a 2 ) = ε 1 Φ n ( a 1 ) a 2 n ,   for a 1 M n ( V 1 c )   , a 2 M n ( V 2 c )   and n N   . Clearly N = ( N n )   is a matrix seminorm on V 1 c V 2 c   and N n ( ( a 1 , a 2 ) * ) = N n ( a 1 , a 2 )   for a 1 M n ( V 1 c )   , a 2 M n ( V 2 c )   and n N   . And we have that N 1 ( 1 , 1 ) = ε 1 Φ ( 1 ) 1 1 = 0   and N 1 ( 1 , 0 ) = ε 1 Φ ( 1 ) 0 1 = ε 1   . If { a 1 ( k ) } M n ( V 1 c )   and { a 2 ( k ) } M n ( V 2 c )   with lim k a 1 ( k ) = a 1 M n ( V 1 c )   and lim k a 2 ( k ) = a 2 M n ( V 2 c )   , we have that lim k N n ( a 1 ( k ) , a 2 ( k ) ) = lim k ε 1 Φ n ( a 1 ( k ) ) a 2 ( k ) n = ε 1 Φ n ( a 1 ) a 2 n = N n ( a 1 , a 2 )   since Φ   is completely bounded(see Proposition 3.5 in [12).
    Given a 1 M n ( V 1 c )   and δ > 0   . Taking a 2 = Φ n ( a 1 )   , we have
    max { L 2 , n c ( a 2 ) , N n ( a 1 , a 2 ) } = max { L 2 , n c ( Φ n ( a 1 ) ) , ε 1 Φ n ( a 1 ) a 2 }
    = L 1 , n c ( a 1 ) < L 1 , n c ( a 1 ) + δ .
    While if a 2 M n ( V 2 c )   and δ > 0   , we can take a 1 M n ( V 1 c )   such that Φ n ( a 1 ) = a 2   , and hence we have
    max { L 1 , n c ( a 1 ) , N n ( a 1 , a 2 ) } = max { L 2 , n c ( Φ n ( a 1 ) ) , ε 1 Φ n ( a 1 ) a 2 }
    = L 2 , n c ( a 2 ) < L 2 , n c ( a 2 ) + δ .
    Therefore, N   is a matrix bridge between ( V 1 c , 1 c )   and ( V 2 c , 2 c )   . Define L n ( a 1 , a 2 ) = max { L 1 , n c ( a 1 ) , L 2 , n c ( a 2 ) , N n ( a 1 , a 2 ) } ,   for a 1 M n ( V 1 c )   , a 2 M n ( V 2 c )   and n N   . By Proposition  4.3 , = ( L n ) ( 1 c , 2 c )   . For n N   and φ C S n ( V 2 c )   , we have that φ Φ C S n ( V 1 c )   , and so
    D L n ( φ Φ , φ )
    = sup { φ Φ , a 1 φ , a 2 :
    L r ( a 1 , a 2 ) 1 , ( a 1 , a 2 ) M r ( V 1 c V 2 c ) , r N }
    = sup { φ , Φ r ( a 1 ) a 2 :
    L r ( a 1 , a 2 ) 1 , ( a 1 , a 2 ) M r ( V 1 c V 2 c ) , r N }
    sup { Φ r ( a 1 ) a 2 r : L r ( a 1 , a 2 ) 1 , ( a 1 , a 2 ) M r ( V 1 c V 2 c ) , r N }
    ε ,
    Similarly, for n N   and ψ C S n ( V 1 c )   , we have that D L n ( ( ψ , ψ Φ 1 ) ε   . Thus we obtain that d i s t H D L n ( C S n ( V 1 c ) , C S n ( V 2 c ) ) ε   for n N   , and so sup n N { n 2 d i s t H D L n ( C S n ( V 1 c ) , C S n ( V 2 c ) ) } ε .   Therefore, d i s t N C ( V 1 c , V 2 c ) ε   . Since ε   is arbitrary, we conclude d i s t N C ( V 1 c , V 2 c ) = 0 .   Now, by Theorem  4.7 and Lemma  5.2 we have d i s t N C ( V 1 , V 2 ) d i s t N C ( V 1 , V 1 c ) + d i s t N C ( V 1 c , V 2 c ) + d i s t N C ( V 2 c , V 2 ) = 0 .   So d i s t N C ( V 1 , V 2 ) = 0   .
Given a quantized metric space ( V , )   . From Proposition 6.1 in [21and the proof of Proposition 3.5 in [19, the mapping Ψ : V A ( C S ( V ) )   , defined by Ψ ( a ) ( φ ) = φ ( a )   for φ C S r ( V )   , is a unital complete order isomorphism from V   into A ( C S ( V ) )   , and Ψ   can be extended to a unital complete order isomorphism Ψ ¯   from the completion V ¯   of V   onto A ( C S ( V ) )   . Define L D L n ( F ( n ) ) = sup { F r ( n ) ( φ ) F r ( n ) ( ψ ) D L r ( φ , ψ ) : φ ψ , φ , ψ C S r ( V ) , r N } ,   where F ( n ) A ( C S ( V ) , M n )   . Then D = ( L D L n )   is a matrix gauge on A ( C S ( V ) )   .
Denote K n = { F ( n ) A ( C S ( V ) , M n ) : L D L n ( F ( n ) ) < + } .   Let L n 1 = { a V : L n ( a ) 1 }   and L ¯ n 1   be the norm closure of L n 1   in V ¯   . Denote 1 = ( L n 1 )   and ¯ 1 = ( L ¯ n 1 )   . The matrix gauge ¯ = ( L ¯ n )   on ( V ¯ , 1 )   determined by ¯ 1   is called the closure of   .   is closed if = ¯   on the subspace where ¯   is finite (see Definition 7.2 in [21).
Lemma 5.4. If   is closed, then Ψ n ( M n ( V ) ) = K n   for n N   .
  • Proof. Denote M n 1 = { Ψ n ( a ) : a M n ( V ) , L n ( a ) 1 } , n N ,   L D L n 1 = { F ( n ) A ( C S ( V ) , M n ) : L D L n ( F ( n ) ) 1 } , n N ,   and set 1 = ( M n 1 )   . Define L n ( f ) = sup { f , Ψ r ( a ) : L r ( a ) 1 , a M r ( V ) , r N } ,   for f M n ( ( A ( C S ( V ) ) / ( C I ) ) * )   , where I   is the order unit of A ( C S ( V ) )   . Here we view M n ( ( A ( C S ( V ) ) / ( C I ) ) * )   as the subspace of M n ( ( A ( C S ( V ) ) ) * )   consisting of those f M n ( ( A ( C S ( V ) ) ) * )   with f ( a ) = 0 n   for a C I   . Clearly, = ( L n )   is a matrix gauge on ( A ( C S ( V ) ) / ( C I ) ) *   and L n ( f * ) = L n ( f )   for all f M n ( ( A ( C S ( V ) ) / ( C I ) ) * )   , The generalized bipolar theorem says that ( 1 )   is the smallest weakly closed absolutely matrix convex set containing 1   (see Proposition 4.1 in [4). Since = ( L n )   is a matrix gauge and Ψ   is a unital complete order isomorphism, 1   is absolutely matrix convex. The closeness of   implies that 1   is normed-closed by Lemma 7.4 in [21. So 1   is weakly closed. Thus ( 1 ) = 1 .   For n N   , we have
    ( M n 1 ) = { Ψ n ( a ) : a M n ( V ) , L n ( a ) 1 }
    = { f M n ( ( A ( C S ( V ) ) ) * ) : f , Ψ r ( a ) 1
    f o r a l l a M r ( V ) , L r ( a ) 1 , r N }
    = { f M n ( ( A ( C S ( V ) ) / ( C I ) ) * ) : f , Ψ r ( a ) 1
    f o r a l l a M r ( V ) , L r ( a ) 1 , r N }
    = { f M n ( ( A ( C S ( V ) ) / ( C I ) ) * ) : L n ( f ) 1 }
    and
    ( M n 1 )
    = { f M n ( ( A ( C S ( V ) ) / ( C I ) ) * ) : L n ( f ) 1 }
    = { F ( n ) M n ( A ( C S ( V ) ) ) : f , F ( n ) 1
    f o r a l l f M r ( ( A ( C S ( V ) ) / ( C I ) ) * ) , L r ( f ) 1 , r N } .
    So F ( n ) ( M n 1 )   if and only if f , F ( n ) L r ( f )   for all f M r ( ( A ( C S ( V ) ) / ( C I ) ) * )   and r N   .
    Suppose that f , F ( n ) L r ( f )   for all f = f * M r ( ( A ( C S ( V ) ) / ( C I ) ) * )   and r N   . Then for g M r ( ( A ( C S ( V ) ) / ( C I ) ) * )   , we have
    g , F ( n ) = [ 1 0 ] [ 0 g , F ( n ) g * , F ( n ) 0 ] [ 0 1 ]
    [ 0 g g * 0 ] , F ( n )
    L 2 r ( [ 0 g g * 0 ] )
    L 2 r ( [ g 0 0 g * ] [ 0 1 1 0 ] )
    L 2 r ( [ g 0 0 g * ] )
    = L r ( g ) ,
    Thus F ( n ) ( M n 1 )   exactly if f , F ( n ) L r ( f )   for all f = f * M r ( ( A ( C S ( V ) ) / ( C I ) ) * )   and r N   . According to Lemma 4.1 in [21, F ( n ) ( M n 1 )   exactly if φ , F ( n ) ψ , F ( n ) L r ( φ ψ )   for all φ , ψ C S r ( A ( C S ( V ) ) )   and r N   . So F ( n ) ( M n 1 )   exactly if
    φ Ψ 1 , F ( n ) ψ Ψ 1 , F ( n )
    L r ( φ Ψ 1 ψ Ψ 1 )
    = sup { ( φ ψ ) Ψ 1 , Ψ k ( a ) : L k ( a ) 1 , a M k ( V ) , k N }
    = sup { φ ψ , a : L k ( a ) 1 , a M k ( V ) , k N }
    = D L r ( φ , ψ ) .
    for all φ , ψ C S r ( A ( C S ( V ) ) )   and r N   . Because φ Ψ 1 , F ( n ) ψ Ψ 1 , F ( n ) = F r ( n ) ( φ ) F r ( n ) ( ψ )   , F ( n ) ( M n 1 )   if and only if F r ( n ) ( φ ) F r ( n ) ( ψ ) D L r ( φ , ψ )   for all φ , ψ C S r ( A ( C S ( V ) ) )   and r N   . And this says exactly that F ( n ) L D L n 1   . Therefore, M n 1 = L D L n 1   for n N   . So Ψ n ( M n ( V ) ) = K n   for n N   .
Lemma 5.5. Let ( V 1 , 1 )   and ( V 2 , 2 )   be quantized metric spaces such that 1   and 2   are closed. For every matrix affine mapping α = ( α n )   from C S ( V 1 )   onto C S ( V 2 )   which is completely isometric for D 1   and D 2   , there is a unital complete order isomorphism Ψ   from V 1   onto V 2   such that 1 = 2 Ψ   .
  • Proof. Define Φ : A ( C S ( V 2 ) ) A ( C S ( V 1 ) )   by ( Φ ( F ( 1 ) ) ) r ( φ ) = F r ( 1 ) ( α r ( φ ) )   for F ( 1 ) A ( C S ( V 2 ) )   and φ C S r ( V 1 )   . Since α   is isometric and matrix affine, Φ   is well-defined. Clearly, Φ   is unital and surjective. On the level of matrices, we have ( Φ n ( F ( n ) ) ) r ( φ ) = F r ( n ) ( α r ( φ ) )   for F ( n ) M n ( A ( C S ( V 2 ) ) )   and φ C S r ( V 1 )   . Since F ( n ) 0   in M n ( A ( C S ( V 2 ) ) )   if and only if F r ( n ) ( φ ) 0   for all r N   and φ C S r ( V 2 )   , Φ   is a unital complete order isomorphism from A ( C S ( V 2 ) )   onto A ( C S ( V 1 ) )   . Since 1   and 2   are closed, Φ   is a unital complete order isomorphism from V 2   onto V 1   by Lemma  5.4 . That α   is completely isometric for D 1   and D 2   implies that L D L 1 , n ( Φ n ( a 2 ) ) = L D L 2 , n ( a 2 )   for all a 2 M n ( V 2 )   and n N   . Because 1   and 2   are closed, they are lower semicontinuous, so that D 1 = 1   on V 1   by Theorem 4.4 in [21, and similarly for V 2   . Thus Φ 1   is a unital complete order isomorphism from V 1   onto V 2   such that 1 = 2 Ψ   .
Theorem 5.6. Suppose ( V 1 , 1 )   and ( V 2 , 2 )   are quantized metric spaces.
If d i s t N C ( V 1 , V 2 ) = 0 ,   then there exists a complete isometry Φ   from ( V 1 , 1 )   onto ( V 2 , 2 )   .
  • Proof. Since d i s t N C ( V 1 , V 2 ) = 0   , we have d i s t N C ( V 1 c , V 2 c ) = 0   by Lemma  5.2 and Theorem  4.7 . From that d i s t N C ( V 1 c , V 2 c ) = 0   , there is a sequence { ( k ) } ( V 1 c V 2 c )   of matrix Lip-norms such that sup n N { n 2 d i s t H D L n ( k ) ( C S n ( V 1 c ) , C S n ( V 2 c ) ) } < 1 k .   So for each n N   , we have n 2 d i s t H D L n ( k ) ( C S n ( V 1 c ) , C S n ( V 2 c ) ) < 1 k .   And for φ , ψ C S n ( V i c )   , i = 1 , 2   , by Proposition  3.6 we have D L n ( k ) ( ( π i ) n c ( φ ) , ( π i ) n c ( ψ ) ) = D L i , n c ( φ , ψ ) ,   where π i , i = 1 , 2   , is the projection from V 1 c V 2 c   onto V i c   . Therefore, for each n N   we get d i s t G H ( C S n ( V 1 c ) , C S n ( V 2 c ) ) = 0 ,   where d i s t G H ( C S n ( V 1 c ) , C S n ( V 2 c ) )   is the Gromov-Hausdorff distance(see Definition 3.4 in [6) between ( C S n ( V 1 c ) , D L 1 , n c )   and ( C S n ( V 2 c ) , D L 2 , n c )   . As in the proofs of Theorem 7.6 and Theorem 7.7 in [17, there is a subsequence { D L 1 ( k j 1 ) }   which converges uniformly on the disjoint union C S 1 ( V 1 c ) C S 1 ( V 2 c )   to a semi-metric σ 1   and σ 1   determines an isometry α 1   from C S 1 ( V 1 c )   onto C S 1 ( V 2 c )   by the condition that σ 1 ( φ , α 1 ( φ ) ) = 0   . Similarly, there is a subsequence { D L 2 ( k j 1 , j 2 ) }   of { D L 2 ( k j 1 ) }   which converges uniformly on C S 2 ( V 1 c ) C S 2 ( V 2 c )   to a semi-metric σ 2   and σ 2   determines an isometry α 2   from C S 2 ( V 1 c )   onto C S 2 ( V 2 c )   by the condition that σ 2 ( φ , α 2 ( φ ) ) = 0   . In general, once { D L 2 ( k j 1 ) } , { D L 2 ( k j 1 , j 2 ) } , , { D L 2 ( k j 1 , j 2 , , j n ) }   have been choosen, there is a subsequence { D L 2 ( k j 1 , j 2 , , j n , j n + 1 ) }   of { D L 2 ( k j 1 , j 2 , , j n ) }   which converges uniformly on C S n + 1 ( V 1 c ) C S n + 1 ( V 2 c )   to a semi-metric σ n + 1   and σ n + 1   determines an isometry α n + 1   from C S n + 1 ( V 1 c )   onto C S n + 1 ( V 2 c )   by the condition that σ n + 1 ( φ , α n + 1 ( φ ) ) = 0   .
    Given φ i C S n i ( V 1 c )   and γ i M n i , n   , i = 1 , 2 , , m   , satisfying i = 1 m γ i * γ i = 1 n   , and ε > 0   . Let s = max { n , n 1 , , n m }   . We can find K N   such that if k j 1 , , j s > K   then σ l D L l ( k j 1 , , j s ) < ε 2   for l = n , n 1 , , n m   . Now for k j 1 , , j s > K   we have
    σ n ( i = 1 m γ i * φ i γ i , i = 1 m γ i * α n i ( φ i ) γ i )
    D L n ( k j 1 , , j s ) ( i = 1 m γ i * φ i γ i , i = 1 m γ i * α n i ( φ i ) γ i ) + ε 2
    D L n 1 + + n m ( k j 1 , , j s ) ( φ 1 φ m , α n 1 ( φ 1 ) α n m ( φ m ) ) + ε 2
    = max { D L n 1 ( k j 1 , , j s ) ( φ 1 , α n 1 ( φ 1 ) ) , , D L n m ( k j 1 , , j s ) ( φ m , α n m ( φ m ) ) } + ε 2
    < max { σ n 1 ( φ 1 , α n 1 ( φ 1 ) ) + ε 2 , , σ n m ( φ m , α n m ( φ m ) ) + ε 2 } + ε 2
    = ε .
    Since ε   is arbitrary, we have σ n ( i = 1 m γ i * φ i γ i , i = 1 m γ i * α n i ( φ i ) γ i ) = 0   . But σ n ( i = 1 m γ i * φ i γ i , α n ( i = 1 m γ i * φ i γ i ) ) = 0 .   By Lemma 7.4 in [17, we obtain α n ( i = 1 m γ i * φ i γ i ) = i = 1 m γ i * α n i ( φ i ) γ i .   So α = ( α n )   is matrix affine.
    Now, by Lemma  5.5 we conclude that there exists a unital complete order isomorphism Φ   from V 1 c   onto V 2 c   such that 1 c = 2 c Φ   , that is, Φ   is a complete isometry from ( V 1 , 1 )   onto ( V 2 , 2 )   .

6 Completeness

For the metric space of complete isometry classes of quantized metric spaces with the quantized Gromov-Hausdorff distance, we show in this section that it is complete.
Let { ( V i , 1 ) }   be a sequence of matrix order unit space. We will denote by i N V i   the operator space direct sum that is formed of all sequences { a i }   with a i V i   and sup i N a i < +   , and by i = 1 n V i   the operator space direct sum of V 1 , V 2 , , V n   (see §2.6 in [13). Then ( i N V i , { 1 } )   and ( i = 1 n V i , ( 1 , , 1 n ) )   are matrix order unit spaces.
Suppose we have a sequence { ( V i , i ) }   of quantized metric spaces. Suppose further that we have a sequence { i }   of matrix Lip-norms with i ( i , i + 1 )   .
Define Q = ( Q k )   on i N V i   , the full product, by Q k ( { a i } ) = sup i N { M i , k ( a i , a i + 1 ) } , { a i } M k ( i N V i ) ,   and set 1 = { { a i } i N V i : Q 1 ( { a i } ) < + } .   It is easy to check that 1   is a self-adjoint subspace of i N V i   containing { 1 }   , and so is a matix order unit space, and that Q   is a matrix Lipschitz seminorm on 1   .
For the evident identifications, we have C S ( V i ) C S ( j = 1 i V j ) C S ( j = 1 n V j ) C S ( 1 ) , 0 i n .   Given a family of graded sets S i = ( S i , n ) , i I   . We denote by i I S i   the graded set ( i I S i , n )   . If S = ( S n )   is a graded set in a vector space, we denote by m c o ( S )   the matrix convex hull of S   . Let Z = ( Z n ) = m c o ( i N C S ( V i ) )   , U = n N C S ( j = 1 n V j )   and W n = m c o ( i = 1 n C S ( V i ) )   .
Proposition 6.1. Z   and U   are BW-dense in C S ( 1 )   . In particular, W n   is BW-dense in C S ( j = 1 n V j )   for n N   .
  • Proof. Since the matrix polar
    Z n π = { a M n ( 1 ) : R e a , , φ 1 r × n for all φ Z r , r N }
    = { a M n ( 1 ) : φ n ( R e ( a ) 1 r × n for all φ Z r , r N }
    = { a M n ( 1 ) : φ n ( 1 n R e ( a ) ) 0 for all φ Z r , r N }
    = { a M n ( 1 ) : 1 n R e ( a ) 0 }
    = { a M n ( 1 ) : a + 1 } ,
    we have
    Z n π π = { f M n ( 1 * ) : R e f , a 1 r × n
    when 1 r R e ( a ) 0 , a M r ( i N V i ) , r N } .
    For f Z n π π , λ R   and a = a * M r ( 1 )   , we have that 1 r R e ( i λ a ) 0   , and so R e i λ a , f 1 r × n   . Thus 0 = R e i a , f = R e ( i a , f ) = I m a , f ,   that is, I m a , f = 0   . If a M r ( 1 ) , a 0   and λ 0   , then 1 r R e ( λ a ) 0   and so λ a , f = R e λ a , f 1 r × n   . Thus a , f 0   . Clearly, 1 r , f 1 r × n   . Therefore, Z n π π = { f M n ( 1 * ) : f is completely positive and f ( 1 ) 1 n } .   By the bipolar theorem in matrix convexity (see Corollary 5.5 in [5), m c o ( Z { 0 } )   is BW-dense in Z π π = ( Z n π π )   . Evaluting the completely positive mappings at 1   , we see that Z   is BW-dense in C S ( 1 )   . Because Z U C S ( 1 )   , U   is BW-dense in C S ( 1 )   .
Define P n = ( P n , k )   on i = 1 n V i   by P n , k ( a 1 , , a n ) = max { M i , k ( a i , a i + 1 ) : 1 i n 1 } ,   for ( a 1 , , a n ) M k ( i = 1 n V i )   . Similar to the proof of Proposition  4.3 , we have
Proposition 6.2. P n   is a matrix Lip-norm on i = 1 n V i   , and induces j , 1 j n   and i   and P i   , 1 i n 1   , via the evident projections.
For b M i ( j = 1 n V j )   and ε > 0   , set b n = b   . Since P n + 1   induces P n   , we can find b n + 1 M i ( j = 1 n + 1 V j )   such that ( π n ) i ( b n + 1 ) = b n   and P n + 1 , i ( b n + 1 ) < P n , i ( b n ) + ε 2 n   , where π n   is the evident projection from j = 1 n + 1 V j   onto j = 1 n V j   . Similarly, we can find b n + 2 M i ( j = 1 n + 2 V j )   such that ( π n + 1 ) i ( b n + 2 ) = b n + 1   and P n + 2 , i ( b n + 2 ) < P n + 1 , i ( b n + 1 ) + ε 2 n + 1   . Continuing in this way, for t n   we get b t + 1 M i ( j = 1 t + 1 V j )   such that ( π t ) i ( b t + 1 ) = b t   and P t + 1 , i ( b t + 1 ) < P t , i ( b t ) + ε 2 t   . We let c = { c j }   be the unique element of M i ( j N V j )   such that ( π t ) i ( c ) = b t   for t n   . Then Q i ( c ) P n , i ( b ) + ε   . So, P n , i ( b ) = Q j = 1 n V j , i ( b )   . Set d k = ( c 1 , , c k ) , k N   .
Since P n   induces i ( 1 i n 1 )   , via the evidence projections, d i s t H D P n , i ( C S i ( V k ) , C S i ( V k + 1 ) ) = d i s t H D M k , i ( C S i ( V k ) , C S i ( V k + 1 ) ) .   For m , n N   with m < n   and φ n C S i ( V n )   , we can find φ n 1 C S i ( V n 1 )   with D P n , i ( φ n 1 , φ n ) d i s t H D M n 1 , i ( C S i ( V n 1 ) , C S i ( V n ) ) .   Similarly, we can find φ n 2 C S i ( V n 2 )   with D P n , i ( φ n 2 , φ n 1 ) d i s t H D M n 2 , i ( C S i ( V n 2 ) , C S i ( V n 1 ) ) .   Inductively, we can find φ m , , φ n 1   with φ k C S i ( V k )   and D P n , i ( φ k , φ k + 1 ) d i s t H D M k , i ( C S i ( V k ) , C S i ( V k + 1 ) ) ,   for m k n 1   . Consequently, D P n , i ( φ m , φ n ) j = m n 1 d i s t H D M j , i ( C S i ( V j ) , C S i ( V j + 1 ) ) , 2 k n 1 .   Similarly, for φ m C S i ( V m )   we can find a φ n C S i ( V n )   such that the inequality above holds. Thus by Proposition 6.1, we have d i s t H D P n , i ( C S i ( j = 1 m V j ) , C S i ( V n ) ) j = m n 1 d i s t H D M j , i ( C S i ( V j ) , C S i ( V j + 1 ) ) .  
Proposition 6.3. For m < n   , we have d i s t H D P n , i ( C S i ( j = 1 m V j ) , C S i ( j = 1 n V j ) ) k = 1 i j = m n 1 d i s t H D M j , k ( C S k ( V j ) , C S k ( V j + 1 ) ) .  
  • Proof. For φ C S i ( j = 1 m V j )   , we can find a ψ C S i ( V n ) C S i ( j = 1 n V j )   such that D P n , i ( φ , ψ ) k = 1 i j = m n 1 d i s t H D M j , k ( C S k ( V j ) , C S k ( V j + 1 ) )   from the discussion before the proposition. Suppose φ C S i ( j = 1 n V j )   and ε > 0   . For each i N   , j = 1 n C S i ( V j )   is a BW-closed subset of C S i ( j = 1 n V j )   , and γ * ( j = 1 n C S i ( V j ) ) γ j = 1 n C S k ( V j )   for all isometries γ M i , k   . From Proposition  6.1 , the BW-closure m c o ¯ ( j = 1 n C S i ( V j ) )   of m c o ( j = 1 n C S i ( V j ) )   is C S ( j = 1 n V j )   , and so by Theorem 4.6 and Theorem 4.3 in [19, there exist φ k C S l k ( V j k )   and γ k M l k , i   for k = 1 , 2 , , s , 1 j k n , 1 l k i   satisfying k = 1 s γ k * γ k = 1 i   such that D P n , i ( φ , k = 1 s γ k * φ k γ k ) < ε .   For each φ k   , we can find ψ k C S l k ( j = 1 m V j )   so that D P n , l k ( φ k , ψ k ) j = m n 1 d i s t H D M j , l k ( C S l k ( V j ) , C S l k ( V j + 1 ) ) .   Thus
    D P n , i ( φ , k = 1 s γ k * ψ k γ )
    D P n , i ( φ , k = 1 s γ k * φ k γ ) + D P n , i ( k = 1 s γ k * φ k γ , k = 1 s γ k * ψ k γ )
    < ε + D P n , i ( k = 1 s γ k * φ k γ , k = 1 s γ k * ψ k γ )
    ε + D P n , k = 1 s l k ( φ 1 φ s , ψ 1 ψ s )
    = ε + max { D P n , l 1 ( φ 1 , ψ 1 ) , , D P n , l s ( φ s , ψ s ) }
    ε + k = 1 i j = m n 1 d i s t H D M j , k ( C S k ( V j ) , C S k ( V j + 1 ) ) ,
    because D P n = ( D P n , k )   is a matrix metric (see Example 5.2 in [21). Since ε   is arbitrary, we obtain the desired inequality.
Now for φ , ψ C S i ( j = 1 n V j )   , there are φ 1 , ψ 1 C S i ( V 1 )   with D P n , i ( φ 1 , φ ) j = 1 n 1 d i s t H D M j , i ( C S i ( V j ) , C S i ( V j + 1 ) ) ,   and D P n , i ( ψ 1 , ψ ) j = 1 n 1 d i s t H D M j , i ( C S i ( V j ) , C S i ( V j + 1 ) ) .   So
D P n , i ( φ , ψ ) D P n , i ( φ , φ 1 ) + D P n , i ( φ 1 , ψ 1 ) + D P n , i ( ψ 1 , ψ )
d i a m ( C S i ( V 1 ) , D L 1 , i ) + 2 j = 1 n 1 d i s t H D M j , i ( C S i ( V j ) , C S i ( V j + 1 ) )
= h i ,
where d i a m ( C S i ( V 1 ) , D L 1 , i )   is the diameter of C S i ( V 1 )   with respect to D L 1 , i   .
By Proposition 5.2 and Proposition 3.8 in [20, we have d ~ n i h i P n , i ( d n ) h i Q i ( c ) < h i Q i ( c ) + ε ,   where ε > 0   . So there is an α n , i = [ α s t ( n , i ) ] M i   such that d n [ α s t ( n , i ) ( 1 , , 1 ) ] i h i Q i ( c ) + ε , n N .   Set G n , i = { β n , i = [ β s t ( n , i ) ] M i : d n [ β s t ( n , i ) ( 1 , , 1 ) ] i h i Q i ( c ) + ε } .   Then G n , i   is a non-empty closed bounded subset of M i   . Clearly, G n + 1 , i G n , i   .
So there exists a β 0 n = 1 G n , i   . We have d n i β 0 + h i Q i ( c ) + ε , n N .   Thus c M i ( j N V j )   , and we obtain
Proposition 6.4. For n N   , Q   induces P n   via the evident projection.
Theorem 6.5. The metric space   of complete isometry classes of quantized metric spaces, with the metric d i s t N C   , is complete.
  • Proof. Let { ( V n , n ) }   be a sequence in   which is Cauchy with respect to the quantized Gromov-Hausdorff distance d i s t N C   . To show that { ( V n , n ) }   converges it suffices to show that a subsequence converges. Since { ( V n , n ) }   is Cauchy, we can choose a subsequence, still denoted by { ( V n , n ) }   , such that d i s t N C ( V n , V n + 1 ) < 1 2 n ,   for all n N   . By definition, there exist n = ( M n , k ) ( n , n + 1 )   with sup k N { k 2 d i s t H D M n , k ( C S k ( V n ) , C S k ( V n + 1 ) ) } < 1 2 n ,   for all n N   . It follows that n = 1 sup k N { k 2 d i s t H D M n , k ( C S k ( V n ) , C S k ( V n + 1 ) ) } < + .   Let ε > 0   be given. Then there is an m N   such that n = m sup k N { k 2 d i s t H D M n , k ( C S k ( V n ) , C S k ( V n + 1 ) ) } < ε .   By Proposition  6.2 , Proposition  6.3 and Proposition  6.4 , we have
    d i s t H D Q i ( C S i ( j = 1 m V j ) , C S i ( j = 1 n V j ) )
    k = 1 i j = m n 1 d i s t H D M j , k ( C S k ( V j ) , C S k ( V j + 1 ) )
    k = 1 i k 2 j = m n 1 sup k N { k 2 d i s t H D M j , k ( C S k ( V j ) , C S k ( V j + 1 ) ) }
    < ( k = 1 i k 2 ) ε ,
    for n > m   . This says that C S i ( j = 1 m V j )   is ( k = 1 i k 2 ) ε   -dense for D Q i   in Z i   . But C S i ( j = 1 m V j )   is BW-compact for the topology from D Q i = D P m , i   by Proposition  6.2 . Thus C S i ( j = 1 m V j )   is totally bounded for D Q i   , and so Z i   is totally bounded for D Q i   . Let Z ^ = ( Z ^ n )   be the completion of Z   for D Q   . We let D Q   denote also the extension of D Q   to Z ^   . Then Z ^   is a compact matrix convex set. For { a i } M n ( 1 )   , we have j = 1 m γ j * φ j γ j , { a i } = j = 1 m ( γ j 1 n ) * φ j , { a i } ( γ j 1 n ) ,   and
    j = 1 m γ j * φ j γ j , { a i } k = 1 p λ k * ψ k λ k , { a i }
    L D Q n ( { a i } ) D Q r ( j = 1 m γ j * φ j γ j , k = 1 p λ k * ψ k λ k )
    Q n ( { a i } ) D Q r ( j = 1 m γ j * φ j γ j , k = 1 p λ k * ψ k λ k ) ,
    where φ j C S n j ( V q j )   , ψ k C S m k ( V l k )   , and γ j M n j , r   , and λ k M m k , r   satisfying j = 1 m γ j * γ j = 1 r   and k = 1 p λ k * λ k = 1 r   . So the map Φ : 1 A ( Z )   , given by ( Φ ( { a i } ) ) ( j = 1 m γ j * φ j γ j ) = j = 1 m γ j * φ j ( a q j ) γ j ,   for { a i } 1   , φ j C S n j ( V q j )   and γ j M n j , r   satisfying j = 1 m γ j * γ j = 1 r   , is well-defined and Φ ( { a i } )   can be extended to an element Φ ( { a i } ) ^ A ( Z ^ )   . Moreover if { a i } 0   in 1   then Φ ( { a i } ) ^ 0   in A ( Z ^ )   and Φ ( { 1 } ) ^ = I   . Thus 1   can be regarded as a matrix order unit subspace of A ( Z ^ )   .
    Define the map Ψ r : Z ^ r C S r ( 1 ) , r N   , by Ψ r ( z ) ( { a i } ) = Ψ ( { a i } ) ^ ( z ) ,   for z Z ^ r   and { a i } 1   . Clearly, Ψ   is continuous. For z = j = 1 m γ j * φ j γ j Z r   with φ j C S n j ( V q j )   , γ j M n j , r   satisfying j = 1 m γ j * γ j = 1 r   , we have Ψ r ( z ) ( { a i } ) = Ψ ( { a i } ) ^ ( z ) = Ψ ( { a i } ) ( z ) = z ( { a i } ) ,   that is, Ψ r ( z ) = z   . Since Z r   is dense in C S r ( 1 )   and Z ^ r   is compact, we obtain that Ψ r ( Z ^ r ) = C S r ( 1 )   .
    If z 1 , z 2 Z ^ r   with z 1 z 2   and k = D Q r ( z 1 , z 2 )   , we can find y 1 , y 2 Z r   such that D Q r ( z i , y i ) < k 4 , i = 1 , 2   . Thus D Q r ( y 1 , y 2 ) > k 2   . So we can find { w i } M r ( 1 )   with Q r ( { w i } ) 1   and { w i } , y 1 { w i } , y 2 > k 2   . But L D Q r ( Φ r ( { w i } ) ^ 1   so that Φ r ( { w i } ) ^ , z i Φ r ( { w i } ) ^ , y i < k 4 , i = 1 , 2   . Thus Φ r ( { w i } ) ^ , z 1 Φ r ( { w i } ) ^ , z 2 > 0   . Therefore, Ψ r   is injective. So Ψ r   is a homeomorphism of Z ^ r   onto C S r ( 1 )   for r N   . From this we see that the D Q   -topology on C S ( 1 )   agrees with the BW-topology. Hence Q   is a matrix Lip-norm on 1   .
    By Proposition  6.2 and Proposition  6.4 , we obtain n = 1 sup k N { k 2 d i s t H D Q k ( C S k ( V n ) , C S k ( V n + 1 ) ) } < + ,   which indicate that, for k N   , { C S k ( V n ) }   is a Cauchy sequence for d i s t H D Q k   , and has a limit K k C S k ( 1 )   . Clearly K = ( K k )   is a compact matrix convex set.
    Because 1   is completely order isomorphic to a dense subspce of A ( C S ( 1 ) )   (Proposition 6.1(1) in [21), we can view 1   as a dence subspace of A ( C S ( 1 ) )   . Let φ   be the map which restricts the elements of A ( C S ( 1 ) )   to K   and V = φ ( 1 )   . Then ( V , Q V )   is a quantized metric space.
    Given ε > 0   . Then there is an N N   such that j = n sup k N { k 2 d i s t H D Q k ( C S k ( V j ) , C S k ( V j + 1 ) ) } < ε , n N .   For k , p N   , we have
    k 2 d i s t H D Q k ( C S k ( V n ) , C S k ( V n + p ) )
    j = n n + p 1 k 2 d i s t H D Q k ( C S k ( V j ) , C S k ( V j + 1 ) )
    j = n sup k N { k 2 d i s t H D Q k ( C S k ( V j ) , C S k ( V j + 1 ) ) }
    < ε ,
    for n N   . Letting p +   , we obtain k 2 d i s t H D Q k ( C S k ( V n ) , K k ) ε ,   for k N   , and so sup k N { k 2 d i s t H D Q k ( C S k ( V n ) , K k ) } ε   . By Proposition  4.8 , for n N   we have d i s t N C ( V n , V ) sup k N { k 2 d i s t H D Q k ( C S k ( V n ) , K k ) } ε .   Therefore, lim n d i s t N C ( V n , V ) = 0   .

7 Matrix approximability

In this section, we establish a matrix approximability theorem for 1   -exact matrix order unit spaces.
Lemma 7.1. Let ( V , )   be a quantized metric space and let x = [ x s t ] M k ( V )   , x s t = x s t ( 1 ) + i x s t ( 2 )   with ( x s t ( p ) ) * = x s t ( p )   for p = 1 , 2 , s , t = 1 , 2 , , k   .
Suppose λ s t ( p ) σ ( x s t ( p ) )   for p = 1 , 2 , s , t = 1 , 2 , , k   . Then x [ ( λ s t ( 1 ) + i λ s t ( 2 ) ) 1 ] k 2 k 2 L k ( x ) d i a m ( V , )   .
  • Proof. By Proposition 2.11 in [8, we have
    x [ ( λ s t ( 1 ) + i λ s t ( 2 ) ) 1 ] k [ x s t ( 1 ) λ s t ( 1 ) 1 ] k + [ x s t ( 2 ) λ s t ( 2 ) 1 ] k
    p = 1 2 s , t = 1 k x s t ( p ) λ s t ( p ) 1 1
    p = 1 2 s , t = 1 k L 1 ( x s t ( p ) ) d i a m ( V , )
    s , t = 1 k 2 L k ( x ) d i a m ( V , )
    = 2 k 2 L k ( x ) d i a m ( V , ) .
An operator space X   is said to be 1   -exact if for every finite-dimensional subspace X   and λ > 1   there is an isomorphism α   from   onto a subspace of a matrix algebra such that α c b α 1 c b λ   . A matrix order unit space ( V , 1 )   is said to be 1   -exact if it is 1   -exact as an operator space.
Theorem 7.2. Let ( V , )   be a quantized metric space. If V   is 1   -exact, then for every ε > 0   , there is a quantized metric space ( M n λ ε , N )   such that d i s t N C ( V , M n λ ε ) < ε .  
  • Proof. Since V   is 1   -exact, by Lemma 5.1 in [9there is a unital complete order embedding ι : V ( )   and a net V φ λ M n λ ψ λ ( )   of unital completely positive mappings through matrix algebras such that ψ λ φ λ   converges pointwise to ι   . Given ε > 0   . By Lemma 7.2, we have L 1 1 = B 1 2 d i a m ( V , ) + C 1 ,   where B 1 2 d i a m ( V , ) = { a V : L 1 ( a ) 1 , a 1 2 d i a m ( V , ) }   . From Proposition 7.5 in [21, B 1 2 d i a m ( V , )   is totally bounded for 1   . So there is a λ ε   such that ( ψ λ ε φ λ ε ) ( x ) x < ε 5 , x L 1 1 .   Denote W = φ λ ε ( V )   and Q k ( y ) = inf { L k ( x ) : ( φ λ ε ) k ( x ) = y }   for y M k ( W )   and k N   .
    We define N k ( x , y ) = 5 ε ( φ λ ε ) k ( x ) y k , ( x , y ) M k ( V W ) , k N .   It is clear that N = ( N k )   is a matrix seminorm on V W   and satisfies the conditions (1), (2) and (3) of Definition  4.1 . For x M k ( V )   and δ > 0   , we can choose y = ( φ λ ε ) k ( x )   . Then max { Q k ( y ) , N k ( x , y ) } = Q k ( y ) L k ( x ) L k ( x ) + δ .   For y M k ( W )   and δ > 0   , we can take x M k ( V )   such that y = ( φ λ ε ) k ( x )   and L k ( x ) Q k ( y ) + δ   . Then max { L k ( x ) , N k ( x , y ) } = L k ( x ) Q k ( y ) + δ .   So N   is a matrix bridge between ( V , )   and ( W , Q )   . Define P k ( x , y ) = max { L k ( x ) , Q k ( y ) , N k ( x , y ) } , ( x , y ) M k ( V W ) , k N .   Then P = ( P k ) ( , Q )   by Proposition  4.3 . If f C S k ( W )   , we have f φ λ ε C S k ( V )   and
    D P k ( f , f φ λ ε )
    = sup { f , y f φ λ ε , x : P k ( x , y ) 1 , ( x , y ) M k ( V W ) }
    = sup { f , y ( φ λ ε ) k ( x ) : P k ( x , y ) 1 , ( x , y ) M k ( V W ) }
    sup { y ( φ λ ε ) k ( x ) k : P k ( x , y ) 1 , ( x , y ) M k ( V W ) }
    ε 5
    On the other hand, if g C S k ( V )   , g   can be extended to a g ¯ C S k ( ( ) )   by Arveson's extension theorem. We have g ¯ ψ λ ε C S k ( W )   and
    D P k ( g , g ¯ ψ λ ε )
    = sup { g , x g ¯ ψ λ ε , y : P k ( x , y ) 1 , ( x , y ) M k ( V W ) }
    = sup { g ¯ , x ( ψ λ ε ) k ( y ) : P k ( x , y ) 1 , ( x , y ) M k ( V W ) }
    sup { g ¯ , x ( ψ λ ε φ λ ε ) k ( x ) + g ¯ ψ λ ε , ( φ λ ε ) k ( x ) y :
    P k ( x , y ) 1 , ( x , y ) M k ( V W ) }
    sup { x ( ψ λ ε φ λ ε ) k ( x ) k + ( φ λ ε ) k ( x ) y k :
    P k ( x , y ) 1 , ( x , y ) M k ( V W ) }
    sup { i , j = 1 k x i j ( ψ λ ε φ λ ε ) ( x i j ) 1 : P k ( x , y ) 1 ,
    ( x , y ) M k ( V W ) } + ε 5
    ( k 2 + 1 ) ε 5
    So we obtain that d i s t N C ( V , W ) < ε 2   .
    Since W M n λ ε   is finite-dimensional, K = Q 1 1   is a normed-closed (and hence weakly closed) absolutely convex set in M n λ ε   , and Q 1 = ( Q k 1 )   is a normed-closed (and hence weakly closed) absolutely matrix convex set in M n λ ε   . Then for the corresponding matrix seminorm ˇ = ( R ˇ k )   of the maximal envelope K ˇ   of K   in M n λ ε   (see Example  3.4 ), we have R ˇ 1 1 = Q 1 1 , R ˇ k | M k ( W ) Q k | M k ( W ) , k N ,   (see page 181 in [4). It is clear that ˇ   is a matrix Lipschitz seminorm. Since the image of Q 1 1 = R ˇ 1 1   in W / ( C 1 )   is totally bounded for   and W M n λ ε   , the image of R ˇ 1 1   in M n λ ε / ( C 1 )   is totally bounded for   . By Theorem 5.3 in [20, D   -topology on C S ( M n λ ε )   agrees with the BW-topology. So ˇ   is a matrix Lip-norm on ( M n λ ε , 1 )   . By Lemma 3.2.3 in [1, there is a (real linear) projection T   from ( M n λ ε ) s a   onto ( W ) s a   with T n λ ε   . We define S : M n λ ε W   by S ( a + i b ) = T ( a ) + i T ( b )   for a , b ( M n λ ε ) s a   . Then S   is a bounded linear mapping with S 2 n λ ε   . Define N k ( x ) = max { Q k ( S k ( x ) ) , R ˇ k ( x ) , 4 ε x S k ( x ) k } , x M k ( M n λ ε ) , k N .   It is clear that N = ( N k )   is a matrix Lip-norm on M n λ ε   since R ˇ k N k   for all k N   and ˇ   is a matrix Lip-norm. And for y M k ( W )   , we have
    N k ( y ) = max { Q k ( S k ( y ) ) , R ˇ k ( y ) , 4 ε y S k ( y ) k }
    = max { Q k ( y ) , R ˇ k ( y ) }
    = Q k ( y ) .
    Define X k ( x , y ) = 4 ε y S k ( x ) k , ( x , y ) M k ( M n λ ε W ) , k N .   It is clear that N = ( N k )   is a matrix seminorm on M n λ ε W   and satisfies the conditions (1), (2) and (3) of Definition  4.1 . For x M k ( M n λ ε )   and δ > 0   , we choose y = S k ( x )   . Then we have max { Q k ( y ) , X k ( x , y ) } = Q k ( S k ( x ) ) N k ( x ) N k ( x ) + δ .   For y M k ( W )   and δ > 0   , we choose x = y   . Then we have max { N k ( x ) , X k ( x , y ) } = N k ( y ) = Q k ( y ) Q k ( x ) + δ .   So X = ( X k )   is a matrix bridge between ( M k ( M n λ ε ) , N )   and ( W , Q )   . Define Y k ( x , y ) = max { N k ( x ) , Q k ( y ) , X k ( x , y ) } , ( x , y ) M k ( M n λ ε W ) , k N .   Then Y = ( Y k ) ( N , Q )   by Proposition  4.3 . For φ C S k ( M n λ ε )   , ψ = φ | W C S k ( W )   and
    D Y k ( φ , ψ )
    = sup { φ , x ψ , y : ( x , y ) M k ( M n λ ε W ) , Y k ( x , y ) 1 }
    sup { φ , x φ , S k ( x ) + φ , S k ( x ) φ , y :
    ( x , y ) M k ( M n λ ε W ) , Y k ( x , y ) 1 }
    sup { x S k ( x ) k + S k ( x ) y k : ( x , y ) M k ( M n λ ε W ) , Y k ( x , y ) 1 }
    ε 2 .
    For ψ C S k ( W )   , ψ   can be extended to a φ C S k ( M n λ ε )   by Arveson's extension theorem. We have
    D Y k ( φ , ψ )
    = sup { φ , x ψ , y : ( x , y ) M k ( M n λ ε W ) , Y k ( x , y ) 1 }
    sup { φ , x φ , S k ( x ) + φ , S k ( x ) φ , y :
    ( x , y ) M k ( M n λ ε W ) , Y k ( x , y ) 1 }
    sup { x S k ( x ) + S k ( x ) y : ( x , y ) M k ( M n λ ε W ) , Y k ( x , y ) 1 }
    ε 2 .
    So d i s t N C ( W , M n λ ε ) < ε 2   . Therefore, d i s t N C ( V , M n λ ε ) d i s t N C ( V , W ) + d i s t N C ( W , M n λ ε ) < ε .  

8 Sphere as the limit of matrix algebras

Let G   be a connected compact semisimple Lie group with a continuous length function l   on G   , which satisfies the additional condition l ( x y x 1 ) = l ( y )   for all x , y G   . Fix an irreducible unitary representation ( U , )   of G   . Then ( U , )   have a highest weight vector ξ   , of norm 1   , unique up to a scalar multiple. Let P   be the rank-one projection for ξ   . Denote by H   the stability subgroup for P   under α   . For any n N   , we form the n t h   inner tensor power ( U n , n )   of ( U , )   .
Let ( U ( n ) , ( n ) )   denote the subrepresentation generated by ξ ( n ) = ξ n   . Then ( U ( n ) , ( n ) )   is irreducible with ξ ( n )   as highest weight vector. We let ( n ) = ( ( n ) )   .
The action of G   on ( n )   by conjugation by U ( n )   is denoted by α ( n )   . We let λ   denote the action of G   on G / H   , and so on A = C ( G / H )   , by left-translation. We denote the corresponding Lip-norm for α ( n )   and l   on ( n )   by L ( n )   , that is, L ( n ) ( T ) = sup { α x ( n ) ( T ) T l ( x ) : x e , x G } , T ( n ) ,   and we denote the Lip-norm for λ   and l   on A   by L   , that is, L ( f ) = sup { λ x ( f ) f l ( x ) : x e , x G } , f A ,   here we view C ( G / H )   as a subalgebra of C ( G )   . By Theorem 3.2 in [18, the quantum metric spaces ( ( n ) , L ( n ) )   converge to ( A , L )   for quantum Gromov-Hausdorff distance as n   goes to   . In this section, a more general statement is established.
Let = ( , k )   be the matrix norm on A   . Set ( n ) = ( L k ( n ) )   , where L k ( n ) ( T ) = sup { [ α x ( n ) ( T i j ) T i j ] l ( x ) : x e , x G } , T = [ T i j ] M k ( ( n ) ) , k N ,   and = ( L k )   , where L k ( f ) = sup { [ λ x ( f i j ) f i j ] , k l ( x ) : x e , x G } , f = [ f i j ] M k ( A ) , k N .   Then ( ( n ) , ( n ) )   and ( A , )   are quantized metric spaces for all n N   by Example 6.5 in [21. As in [18, we will not restrict   to the Lipschitz functions. Let P ( n )   denote the rank-one projection for ξ ( n )   . We denote the corresponding Berezin symbol mapping from ( n )   to A   by σ ( n )   . Then σ ( n )   is unital, positive, norm-nonincreasing and α ( n )   - λ   -equivariant (see page 73 in [18). For k N   and T = [ T i j ] M k ( ( n ) )   , define σ T ( n ) ( x ) = [ σ T i j ( n ) ( x ) ] , x G .   For ε > 0   , define N k ( f , T ) = ε 1 f σ T ( n ) , k , f M k ( A ) , T M k ( ( n ) ) ,   and denote N = ( N k )   .
Lemma 8.1. For any T M k ( ( n ) )   , we have L k ( σ T ( n ) ) < L k ( n ) ( T ) + ε .  
  • Proof. Since σ ( n )   is a unital positive mapping from ( n )   to A   , σ ( n )   is unital completely positive and hence σ ( n ) c b = 1   by Theorem 3.8 and Proposition 3.5 in [12. So we have
    L k ( σ T ( n ) )
    = sup { [ λ x ( σ T i j ( n ) ) σ T i j ( n ) ] , k l ( x ) : x e , x G }
    = sup { [ σ ( α x ( n ) ( T i j ) T i j ) ( n ) ] , k l ( x ) : x e , x G }
    sup { [ α x ( n ) ( T i j ) T i j ] l ( x ) : x e , x G }
    = L k ( n ) ( T )
    < L k ( n ) ( T ) + ε ,
    by the α ( n )   - λ   -equivariation of σ ( n )   .
Put on A   the inner product from L 2 ( G / H )   , while on ( n )   its Hilbert-Schmidt inner product. Then the mapping σ ( n )   from ( n )   to A   has an adjoint operator σ ˘ ( n )   from A   to ( n )   . For any T ( n )   , a function f A   such that σ ˘ f ( n ) = T   is called a Berezin contravariant symbol for T   . Moreover, σ ˘ ( n )   is unital, positive, norm-nonincreasing, and λ   - α ( n )   -equivariant (see page 75 in [18). From Theorem 3.10 and Proposition 3.5 in [12, σ ˘ ( n )   is unital completely positive and σ ˘ ( n ) c b = 1   .
So by the same argument as in the proof of Lemma  8.1 , we obtain:
Lemma 8.2. For any f = [ f i j ] M k ( A )   , we have L k ( n ) ( σ ˘ f ( n ) ) < L k ( f ) + ε ,   where σ ˘ f ( n ) = [ σ ˘ f i j ( n ) ] M k ( ( n ) )   .
Denote D L k ( φ , ψ ) = sup { f , φ f , ψ : L r ( f ) 1 , f M r ( A ) , r N } ,   for φ , ψ C S k ( A )   , k N   , and h P ( n ) ( x ) = d ( n ) τ ( n ) ( P ( n ) α x ( n ) ( P ( n ) ) ) , x G / H ,   where τ ( n )   denotes the usual (un-normalized) trace on ( n )   and d ( n )   is the dimension of ( n )   . Set γ ( n ) = G / H D L 1 ( e ^ , y ^ ) h P ( n ) ( y ) d y ,   where every y G / H   is naturally identified with an element y ^   of C S 1 ( A )   . Then:
Lemma 8.3. For f M k ( A )   , we have f σ ( n ) ( σ ˘ f ( n ) ) , k γ ( n ) L k ( f ) .  
  • Proof. Suppose f = [ f i j ]   . Then for any x G / H   , we have
    f ( x ) ( σ ( n ) ( σ ˘ f ( n ) ) ) ( x )
    = [ G / H ( f i j ( x ) f i j ( y ) ) h P ( n ) ( y 1 x ) d y ]
    = G / H [ f i j ( x ) f i j ( y ) ] h P ( n ) ( y 1 x ) d y
    G / H [ f i j ( x ) f i j ( y ) ] h P ( n ) ( y 1 x ) d y
    L k ( f ) G / H D L 1 ( x ^ , y ^ ) h P ( n ) ( y 1 x ) d y
    = L k ( f ) G / H D L 1 ( e ^ , y ^ ) h P ( n ) ( y ) d y
    γ ( n ) L k ( f ) ,
    by the formula (2.2) in [18. So f σ ( n ) ( σ ˘ f ( n ) ) , k = max { f ( x ) ( σ ( n ) ( σ ˘ f ( n ) ) ) ( x ) : x G / H } γ ( n ) L k ( f ) .  
Since the sequence { γ ( n ) }   converges to 0   as n   (see page 80 in [18), there is an N 1 N   such that γ ( n ) < ε 2   for n > N 1 ( n )   . So we obtain:
Proposition 8.4. For n > N 1   , N   is a matrix bridge between ( ( n ) , ( n ) )   and ( A , )   , and hence Q = ( Q k ) ( ( n ) , )   , where Q k ( f , T ) = max { L k ( n ) ( T ) , L k ( f ) , N k ( f , T ) } , ( f , T ) M r ( ( n ) A ) .  
From Theorem 6.1 in [18, we have:
Lemma 8.5. There is an N 2 N   such that T σ ˘ ( n ) ( σ T ( n ) ) < ε 2 L 1 ( n ) ( T ) ,   for all T ( n )   and n > N 2   .
Theorem 8.6. With notation as above, the quantized metric spaces ( ( n ) , ( n ) )   converge to ( A , )   for quantized Gromov-Hausdorff distance as n   goes to   .
  • Proof. Given ε > 0   . Choose N = max { N 1 , N 2 }   . Then for n > N   , we have that Q ( ( n ) , )   by Proposition  8.4 . Given φ C S k ( A )   . we have φ σ ( n ) C S k ( ( n ) )   , and
    D L k ( φ , φ σ ( n ) )
    = sup { φ , f φ σ ( n ) , T : L r ( f , T ) 1 ,
    ( f , T ) M r ( A ( n ) ) , r N }
    = sup { φ , f σ T ( n ) : L r ( f , T ) 1 , ( f , T ) M r ( A ( n ) ) , r N }
    sup { f σ T ( n ) , r : L r ( f , T ) 1 , ( f , T ) M r ( A ( n ) ) , r N }
    ε .
    On the other hand, if ψ C S k ( ( n ) )   , then ψ σ ˘ ( n ) C S k ( A )   , and
    D L k ( ψ σ ˘ ( n ) , ψ )
    = sup { ψ σ ˘ ( n ) , f ψ , T : L k ( f , T ) 1 ,
    ( f , T ) M k ( A ( n ) ) }
    = sup { ψ , σ ˘ f ( n ) T : L k ( f , T ) 1 , ( f , T ) M k ( A ( n ) ) }
    sup { σ ˘ f ( n ) T : L k ( f , T ) 1 , ( f , T ) M k ( A ( n ) ) }
    sup { σ ˘ f ( n ) σ ˘ ( n ) ( σ T ( n ) ) + σ ˘ ( n ) ( σ T ( n ) ) T : L k ( f , T ) 1 ,
    ( f , T ) M k ( A ( n ) ) }
    f σ T ( n ) , k + sup { σ ˘ ( n ) ( σ T ( n ) ) T : L k ( f , T ) 1 ,
    ( f , T ) M k ( A ( n ) ) }
    ε 2 + 1 2 k 2 ε
    k 2 ε ,
    by Lemma  8.5 . Therefore, for n > N   , we have d i s t N C ( ( n ) , A ) ε ,   that is, lim n d i s t N C ( ( n ) , A ) = 0   .

Acknowledgements

I am grateful to Marc Rieffel for valuable discussions. I also thank Hanfeng Li for helpful comments. This research was partially supported by Shanghai Priority Academic Discipline, China Scholarship Council, and National Natural Science Foundation of China. References

  1. B. Blackadar, E. Kirchberg, Generalized inductive limits of finite-dimensional C *   -algebras, Math.Ann. 307 (1997), 343–380.
  2. M. -D. Choi, E. G. Effros, Injectivity and operator spaces, J. Funct. Anal., 24 (1977), 156–209.
  3. A. Connes, Compact metric spaces, Fredholm modules, and hyperfiniteness, Ergodic Theory Dynam. Systems,9(2) (1989), 207–220.
  4. E. G. Effros and C. Webster, Operator analogues of locally convex spaces, Operator algebras and applications(Samos, 1996), 163–207, NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., 495, Kluwer Acad. Publ., Dordrecht, 1997.
  5. E. G. Effros, S. Winkler, Matrix convexity: operator analogues of the Bipolar and Hahn-Banach theorems, J. Funct. Anal., 144 (1997), 117–152.
  6. M. Gromov, Metric structures for Riemannian and non-Riemannian spaces, Birkhäuser Boston Inc., Boston, MA, 1999.
  7. R. V. Kadison, A representation theory for commutative topological algebra. Mem. Amer. Math. Soc., 7, 1951.
  8. D. Kerr, Matricial quantum Gromov-Hausdorff distance. J. Funct. Anal., 205 (2003), 132–167.
  9. D. Kerr, H. Li, On Gromov-Hausdorff convergence for operator metric spaces, arXiv:math.OA/0411157 v2, 2004.
  10. H. Li, θ   -deformations as compact quantum metric spaces, to appear in Comm. Math. Phys., arXiv:math.OA/0311500.
  11. N. Ozawa, M. A. Rieffel, Hyperbolic group C *   -algebras and free-product C *   -algebras as compact quantum metric spaces, to appear in Canad. J. Math., arXiv:math.OA/0302310.
  12. V. I. Paulsen, Completely bounded maps and dilations, Pitman Research Notes in Mathematics Series, 146. Longman Scientific & Technical, Harlow; John Wiley & Sons, Inc., New York, 1986.
  13. G. Pisier, Introduction to operator space theory, London Mathematical Society Lecture Note Series, 294. Cambridge University Press, Cambridge, 2003.
  14. M. A. Rieffel, Metrics on states from actions of compact groups, Doc. Math., 3 (1998), 215–229.
  15. M. A. Rieffel, Metrics on state spaces, Doc. Math., 4 (1999), 559–600.
  16. M. A. Rieffel, Group C *   -algebras as compact quantum metric spaces, Doc. Math., 7 (2002), 605–651.
  17. M. A. Rieffel, Gromov-Hausdorff distance for quantum metric spaces, Mem. Amer. Math. Soc. 168 (2004), 1–65.
  18. M. A. Rieffel, Matrix algebras converge to the sphere for quantum Gromov-Hausdorff distance, Mem. Amer. Math. Soc. 168 (2004), 67–91.
  19. C. Webster, S. Winkler, The Krein-Milman theorem in operator convexity, Trans. Amer. Math. Soc., 351 (1999), 307–322.
  20. W. Wu, Non-commutative metric topology on matrix state spaces, to appear in Proc. Amer. Math. Soc., arXiv:math.OA/0410587.
  21. W. Wu, Non-commutative metrics on matrix state spaces, arXiv:math.OA/0411475.

Department of Mathematics, East China Normal University, Shanghai 200062, P.R. China E-mail address : wwu@math.ecnu.edu.cn Current address : Department of Mathematics, University of California, Berkeley, CA 94720-3840 E-mail address : wwu@math.berkeley.edu