arXiv reader

Divisible properties for asymptotically tracially approximation {\rm C^*}-algebras

Qingzhai Fan and Jiahui Wang

arXiv
2205.15106v2 · PDF · 5 April 2024
Categories
math.OA
Licence
CC BY 4.0 The site shows the text.

Abstract

We show that the following divisible properties of the {\rm C^*}-algebras in a class \mathcal{P} are inherited by simple unital {\rm C^*}-algebras in the class of asymptotically tracially in \mathcal{P}: (1) m-almost divisible, (2) weakly (m, n)-divisible.

Items 11

Definition

Definition 2.1

8RXM

([RT].) Let m\in \mathbb{N}. We say that A is m-almost divisible if for each a\in {\rm M}_{\infty}(A)_+, k\in \mathbb{N} and \varepsilon>0, there exists b\in{\rm M}_{\infty}(A)_+ such that k\langle b\rangle\leq \langle a\rangle and \langle (a-\varepsilon)_+\rangle\leq (k+1)(m+1)\langle b\rangle.

Definition

Definition 2.2

CW11

([KM].) Let A be unital {\rm C^*}-algebra. Let m, n\geq 1 be integers. We say A is weakly (m, n)-divisible, if for every u in {\rm W}(A), any \varepsilon>0, there exist elements x_1,x_2,\cdots, x_n\in {\rm W}(A), such that mx_j\leq u for all j=1,2,\cdots, n and (u-\varepsilon)_+\leq x_1+x_2+\cdots+ x_n.

Definition

Definition 2.3

SMKY

Let A and B be {\rm C^*}-algebra, let \varphi:A\to B be a map, let \mathcal{G}\subset A, and let \varepsilon>0. The map \varphi is called \mathcal{G}-\varepsilon-multiplicative, or called \varepsilon-multiplicative on \mathcal{G}, if for any x,y\in F, \|\varphi(xy)-\varphi(x)\varphi(y)\|<\varepsilon. If, in addition, for any x\in \mathcal{G}, |\|\varphi(x)\|-\|x\||<\varepsilon, then we say \varphi is an \mathcal{G}-\varepsilon-approximate embedding.

Definition

Definition 2.4

6ION

([FL].) Let A be a unital {\rm C^*}-algebra. Let \mathcal{P} be a class of {\rm C^*}-algebra. We say A is asymptotically tracially in \mathcal{P}, if for any \varepsilon>0, any finite subset F\subseteq A, and any non-zero element a\geq 0, there exist a {\rm C^*}-algebra B in \mathcal{P} and completely positive contractive linear maps \alpha:A\to B and \beta_n: B\to A, and \gamma_n:A\to A such that

(1) \|x-\gamma_n(x)-\beta_n(\alpha(x))\|<\varepsilon for all x\in F and for all n\in {\mathbb{N}},

(2) \alpha is an F-\varepsilon approximate embedding,

(3) \lim_{n\to \infty}\|\beta_n(xy)-\beta_n(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n(x)\|=\|x\| for all x,y\in B, and

(4) \gamma_n(1_A)\precsim a for all n\in \mathbb{N}.

Theorem

Theorem 2.5

XY7H

([FL].) Let \mathcal{P} be a class of {\rm C^*}-algebras. Let A be a simple unital {\rm C^*}-algebra which is asymptotically tracially in \mathcal{P}. Then the following conditions holds: For any \varepsilon>0, any finite subset F\subseteq A, and any non-zero element a\geq 0, there exist a {\rm C^*}-algebra B in \mathcal{P} and completely positive contractive linear maps \alpha:A\to B and \beta_n: B\to A, and \gamma_n:A\to A\cap\beta_n(B)^{\perp} such that

(1) the map \alpha is unital completely positive linear map, \beta_n(1_B) and \gamma_n(1_A) are projections and \beta_n(1_B)+\gamma_n(1_A)=1_A for all n\in \mathbb{N},

(2) \|x-\gamma_n(x)-\beta_n(\alpha(x))\|<\varepsilon for all x\in F and for all n\in {\mathbb{N}},

(3) \alpha is an F-\varepsilon-approximate embedding,

(4) \lim_{n\to \infty}\|\beta_n(xy)-\beta_n(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n(xy)\|=\|x\| for all x,y\in B, and

(5) \gamma_n(1_A)\precsim a for all n\in \mathbb{N}.

Lemma

Lemma 2.6

Y8WA

([FL].) If the class \mathcal{P} is closed under tensoring with matrix algebras and under passing to unital hereditary {\rm C^*}-subalgebras, then the class which is asymptotically tracially in \mathcal{P} is closed under tensoring with matrix algebras and under passing to unital hereditary {\rm C^*}-subalgebras.

Lemma

Lemma 2.7

XX2V

The m-almost divisible (or weakly (m, n)-divisible) is preserved under tensoring with matrix algebras and under passing to unital hereditary {\rm C^*}-subalgebras.

Used by Proof, Proof
Theorem

Theorem 3.1

K7O2

Let \mathcal{P} be a class of unital {\rm C^*}-algebras which are m-almost divisible. Let A be a unital separable simple {\rm C^*}-algebra. If A is asymptotically tracially in \mathcal{P}, then A is m-almost divisible.

Proof of Theorem 3.1

We need to show that there is b\in {\rm M}_{\infty}(A)_+ such that kb\precsim a and (a-\varepsilon)_+\precsim (k+1)(m+1)b for any a\in A_+, \varepsilon>0 and k\in \mathbb{N}. We may assume that \|a\|=1.

With F=\{a\}, any \varepsilon'>0 with \varepsilon' sufficiently small, since A is asymptotically tracially in \mathcal{P}, there exist a {\rm C^*}-algebra B in \mathcal{P} and completely positive contractive linear maps \alpha:A\to B and \beta_n: B\to A, and \gamma_n:A\to A\cap\beta_n(B)^{\perp} such that

(1) the map \alpha is unital completely positive linear map, \beta_n(1_B) and \gamma_n(1_A) are all projections, and \beta_n(1_B)+\gamma_n(1_A)=1_A for all n\in \mathbb{N},

(2) \|x-\gamma_n(x)-\beta_n(\alpha(x))\|<\varepsilon' for all x\in F and for all n\in {\mathbb{N}},

(3) \alpha is an F-\varepsilon'-approximate embedding,

(4) \lim_{n\to \infty}\|\beta_n(xy)-\beta_n(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n(x)\|=\|x\| for all x,y\in B.

Since B is m-almost divisible, and \alpha(a)\in B, there exists b_1\in B such that kb_1\precsim \alpha(a) and (\alpha(a)-\varepsilon')_+\precsim (k+1)(m+1)b_1.

We divide the proof into two cases.

Case (1), we assume that (\alpha(a)-\varepsilon')_+ is Cuntz equivalent to a projection.

(1.1), If (\alpha(a)-\varepsilon')_+ is not Cuntz equivalent to (k+1)(m+1)b_1.

By Theorem 2.1 (2) in [EFF], we may assume that there exist non-zero c\in B_+ such that (\alpha(a)-\varepsilon')_++ c\precsim (k+1)(m+1)b_1.

Since kb_1\precsim \alpha(a), there exist v\in M_k(B)_+ such that

\|v^*diag(\alpha(a), 0\otimes 1_{k-1})v-b_1\otimes 1_k\|<\bar{\varepsilon}.

We assume that \|v\|\leq M(\bar{\varepsilon}), by (4), there exists sufficiently large integer N_1 such that for any n>N_1, we have

\|\beta_n\otimes id_{M_{k}}(v^*)diag(\beta_n\alpha(a), 0\otimes 1_{k-1})\beta_n\otimes id_{M_{k}}(v)-\beta_n(b_1)\otimes 1_k\|<\varepsilon'.

Therefore we have

k(\beta_n(b_1)-4\varepsilon')_+\precsim (\beta_n\alpha(a)-2\varepsilon')_+.

Since (\alpha(a)-\varepsilon')_++ c\precsim (k+1)(m+1)b_1, there exist w\in M_{(k+1)(m+1)}(B)_+ such that

\|w^*(b_1\otimes 1_{(k+1))m+1)}w-diag((\alpha(a)-\varepsilon')_++ c, 0\otimes 1_{(k+1)(m+1)-1)})\|<\bar{\varepsilon}.

We assume that \|w\|\leq N(\bar{\varepsilon}), by (4), there exists sufficiently large integer N_2 such that for any N_2<n we have

\|\beta_n\otimes id_{M_{(k+1)(m+1)}}(w^*)\beta_n(b_1)\otimes 1_{(k+1)(m+1)}\beta_n\otimes id_{M_{(k+1)(m+1)}}(w)
-diag(\beta_n\alpha((a)-\varepsilon')_++ c, 0\otimes 1_{(k+1)(m+1)-1})\|<\varepsilon'.

Therefore we have

(\beta_n\alpha(a)-6\varepsilon')_++c\precsim (\beta_n(b_1)-2\varepsilon')_+.

For sufficiently large n>max\{N_1,N_2\}, with F=\{\gamma_n(a)\}, any \varepsilon''>0 with \varepsilon'' sufficiently small, let E=\gamma_n(1_A)A\gamma_n(1_A), by Lemma 2.7, E is asymptotically tracially in \mathcal{P}, there exist a {\rm C^*}-algebra D in \mathcal{P} and completely positive contractive linear maps \alpha':E\to D and \beta_n': D\to E, and \gamma_n':E\to E\cap\beta_n'(D)^{\perp} such that

(1)' the map \alpha' is unital completely positive linear map, \beta_n'(1_D) and \gamma_n'(1_E) are all projections, \beta_n'(1_D)+\gamma_n'(1_E)=1_E for all n\in \mathbb{N},

(2)' \|x-\gamma_n'(x)-\beta_n'(\alpha'(x))\|<\varepsilon'' for all x\in G and for all n\in {\mathbb{N}},

(3)' \alpha' is an G-\varepsilon''-approximate embedding,

(4)' \lim_{n\to \infty}\|\beta_n'(xy)-\beta_n'(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n'(x)\|=\|x\| for all x,y\in D, and

(5)' \gamma_n'\gamma(1_E)\precsim \beta_n(c) for all n\in \mathbb{N}.

Since D is m-almost divisible, and (\alpha'\gamma_n(a)-3\varepsilon')_+\in D, there exists b_2\in D_+ such that kb_2\precsim (\alpha'\gamma_n(a)-3\varepsilon')_+ and (\alpha'\gamma_n(a)-4\varepsilon')_+\precsim (k+1)(m+1)b_2.

With the same argument as above we have

k(\beta_n'(b_2)-2\varepsilon')_+\precsim (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+

and

(\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+\precsim (\beta_n'(b_2)-2\varepsilon')_+.

Therefore we have

\begin{aligned} &&k((\beta_n(b_1)-2\varepsilon')_+\oplus (\beta_n'(b_2)-2\varepsilon')_+) \thicksim k((\beta_n(b_1)-2\varepsilon')_++ (\beta_n'(b_2)-2\varepsilon')_+)\\ &&\precsim (\beta_n\alpha(a)-\varepsilon')_+\oplus(\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+ \\ &&\precsim (\beta_n\alpha(a)-\varepsilon')_+\oplus(\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+ +(\gamma_n'\gamma_n(a)-3\varepsilon)_+ \\ &&\precsim a, \end{aligned}

and we also have

\begin{aligned} &&(a-\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon')_+\oplus (\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+ \oplus (\gamma_n'\gamma_n(a)-4\varepsilon')_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon')_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon')_+ \oplus \gamma_n'\gamma_n(1_E)\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon')_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+ \oplus \beta_n(c)\\ &&\precsim (k+1)(m+1)(\beta_n(b_1)-2\varepsilon')_+\oplus (k+1)(m+1)(\beta_n(b_1)-2\varepsilon')_+.\\ &&\precsim (k+1)(m+1)(\beta_n(b_1)-2\varepsilon')_+\oplus (k+1)(m+1)(\beta_n(b_1)-2\varepsilon')_+. \end{aligned}

(1.2), If (\alpha(a)-\varepsilon')_+ is Cuntz equivalent to (k+1)(m+1)b_1.

Since k(b_1\oplus b_1)\precsim \alpha(a), there exist v\in M_k(B)_+ such that

\|v^*diag(\alpha(a), 0\otimes 1_{2k-1})v-b_1\otimes 1_{2k}\|<\bar{\varepsilon}.

We assume that \|v\|\leq M(\bar{\varepsilon}), by (4), there exists sufficiently large integer N_1 such that for any n>N_1, we have

\|\beta_n\otimes id_{M_{2k}}(v^*)diag(\beta_n\alpha(a), 0\otimes 1_{k-1}\beta_n\otimes id_{M_{2k}}(v)-\beta_n(b_1)\otimes 1_{2k}\|<\varepsilon'.

Therefore we have

k(\beta_n(b_1)\oplus\beta_n(b_1)-4\varepsilon')_+\precsim (\beta_n\alpha(a)-2\varepsilon')_+.

With the same argument, we have

(\beta_n\alpha(a)-6\varepsilon')_++c\precsim (\beta_n(b_1)-2\varepsilon')_+.

For sufficiently large n>max\{N_1,N_2\}, with F=\{\gamma_n(a)\}, any \varepsilon''>0 with \varepsilon'' sufficiently small, let E=\gamma_n(1_A)A\gamma_n(1_A), by Lemma 2.7, E is asymptotically tracially in \mathcal{P}, there exist a {\rm C^*}-algebra D in \mathcal{P} and completely positive contractive linear maps \alpha':E\to D and \beta_n': D\to E, and \gamma_n':E\to E\cap\beta_n'(D)^{\perp} such that

(1)' the map \alpha' is unital completely positive linear map, \beta_n'(1_D) and \gamma_n'(1_E) are all projections, \beta_n'(1_D)+\gamma_n'(1_E)=1_E for all n\in \mathbb{N},

(2)' \|x-\gamma_n'(x)-\beta_n'(\alpha'(x))\|<\varepsilon'' for all x\in G and for all n\in {\mathbb{N}},

(3)' \alpha' is an G-\varepsilon''-approximate embedding,

(4)' \lim_{n\to \infty}\|\beta_n'(xy)-\beta_n'(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n'(x)\|=\|x\| for all x,y\in D, and

(5)' \gamma_n'\gamma(1_A)\precsim \beta_n(b_1) for all n\in \mathbb{N}.

Since D is m-almost divisible, and (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+\in B, there exists b_2\in D_+ such that kb_2\precsim (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+ and (\beta_n'\alpha'\gamma_n(a)-2\varepsilon')_+\precsim (k+1)(m+1)b_2.

With the same argument as above we have

k(\beta_n'(b_2)-2\varepsilon')_+\precsim (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+

and

(\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+\precsim (\beta_n'((b_1-2\varepsilon')_+).

Therefore we have

\begin{aligned} &&k((\beta_n(b_1\oplus b_1)-2\varepsilon')_+\oplus (\beta_n'(b_2)-2\varepsilon')_+) \thicksim k((\beta_n(b_1)-2\varepsilon')_++ (\beta_n'(b_2)-2\varepsilon')_+)\\ &&\precsim (\beta_n\alpha(a)-\varepsilon')_+\oplus(\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+ \\ &&\precsim (\beta_n\alpha(a)-\varepsilon')_+\oplus(\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+ +(\gamma_n'\gamma_n(a)-3\varepsilon)_+ \\ &&\precsim a, \end{aligned}

and we also have

\begin{aligned} &&(a-\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon')_+\oplus (\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+ \oplus (\gamma_n'\gamma_n(a)-4\varepsilon')_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon')_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon')_+ \oplus \gamma_n'\gamma_n(1_E)\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon')_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+ \oplus \beta_n(c)\\ &&\precsim (k+1)(m+1)(\beta_n(b_1)-2\varepsilon')_+\oplus (k+1)(m+1)(\beta_n(b_1)-2\varepsilon')_+.\\ &&\precsim (k+1)(m+1)(\beta_n(b_1)-2\varepsilon')_+\oplus (k+1)(m+1)(\beta_n(b_1\oplus b_1)-2\varepsilon')_+. \end{aligned}

Case (2), we assume that (\alpha(a)-\varepsilon')_+ is not Cuntz equivalent to a projection.

By Theorem 2.1 (4) in [EFF], there is a non-zero positive element d such that (\alpha(a)-2\varepsilon')_++d\precsim (\alpha(a)-\varepsilon')_+.

Since (\alpha(a)-\varepsilon')_++ c\precsim (k+1)(m+1)b_1, there exist w\in M_{(k+1)(m+1)}(B)_+ such that

\|w^*b_1\otimes 1_{(k+1)(m+1)}w-diag(\alpha(a)-\varepsilon')_++ c, 0\otimes 1_{(k+1)(m+1)-1}\|<\bar{\varepsilon}.

We assume that \|w\|\leq N(\bar{\varepsilon}), by (4), there exists sufficiently large integer n such that

\|\beta_n(w^*)\beta_n(b_1)\otimes 1_{(k+1)(m+1)}\beta_n(w)-diag(\beta_n\alpha((a)-\varepsilon')_++ c), 0\otimes 1_{(k+1)(m+1)}\|<\bar{\varepsilon}.

Therefore we have

(\beta_n\alpha(a)-6\varepsilon')_++c)\precsim (\beta_n((b_1-2\varepsilon')_+).

Since kb_1\precsim \alpha(a), there exist v\in M_k(B)_+ such that

\|v^*diag(\alpha(a), 0\otimes 1_{k-1})v-b_1\otimes 1_k\|<\bar{\varepsilon}.

We assume that \|v\|\leq M(\bar{\varepsilon}), by (4), there exists sufficiently large integer N_1 such that for any n>N_1, we have

\|\beta_n\otimes id_{M_{k}}(v^*)diag(\beta_n\alpha(a), 0\otimes 1_{k-1})\beta_n\otimes id_{M_{k}}(v)-\beta_n(b_1)\otimes 1_k\|<\varepsilon'.

Therefore we have

k(\beta_n(b_1)-4\varepsilon')_+\precsim (\beta_n\alpha(a)-2\varepsilon')_+.

For sufficiently large n>max\{N_1,N_2\}, with F=\{\gamma_n(a)\}, any \varepsilon''>0 with \varepsilon'' sufficiently small, let E=\gamma_n(1_A)A\gamma_n(1_A), by Lemma 2.7, E is asymptotically tracially in \mathcal{P}, there exist a {\rm C^*}-algebra D in \mathcal{P} and completely positive contractive linear maps \alpha':E\to D and \beta_n': D\to E, and \gamma_n':E\to E\cap\beta_n'(D)^{\perp} such that

(1)' the map \alpha' is unital completely positive linear map, \beta_n'(1_D) and \gamma_n'(1_E) are all projections, \beta_n'(1_D)+\gamma_n'(1_E)=1_E for all n\in \mathbb{N},

(2)' \|x-\gamma_n'(x)-\beta_n'(\alpha'(x))\|<\varepsilon'' for all x\in G and for all n\in {\mathbb{N}},

(3)' \alpha' is an G-\varepsilon''-approximate embedding,

(4)' \lim_{n\to \infty}\|\beta_n'(xy)-\beta_n'(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n'(x)\|=\|x\| for all x,y\in D, and

(5)' \gamma_n'\gamma(1_A)\precsim \beta_n(b_1) for all n\in \mathbb{N}.

Since D is m-almost divisible, and (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+\in B, there exists b_2\in D_+ such that kb_2\precsim (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+ and (\beta_n'\alpha'\gamma_n(a)-2\varepsilon')_+\precsim (k+1)(m+1)b_2.

With the same argument as above we have

k(\beta_n'(b_2)-2\varepsilon')_+\precsim (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+

and

(\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+\precsim (\beta_n'((b_1-2\varepsilon')_+).

Therefore we have

\begin{aligned} &&k((\beta_n(b_1\oplus b_1)-2\varepsilon')_+\oplus (\beta_n'(b_2)-2\varepsilon')_+) \thicksim k((\beta_n(b_1)-2\varepsilon')_++ (\beta_n'(b_2)-2\varepsilon')_+)\\ &&\precsim (\beta_n\alpha(a)-\varepsilon')_+\oplus(\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+ \\ &&\precsim (\beta_n\alpha(a)-\varepsilon')_+\oplus(\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+ +(\gamma_n'\gamma_n(a)-3\varepsilon)_+ \\ &&\precsim a, \end{aligned}

and we also have

\begin{aligned} &&(a-\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon')_+\oplus (\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+ \oplus (\gamma_n'\gamma_n(a)-4\varepsilon')_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon')_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon')_+ \oplus \gamma_n'\gamma_n(1_E)\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon')_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+ \oplus \beta_n(c)\\ &&\precsim (k+1)(m+1)(\beta_n(b_1)-2\varepsilon')_+\oplus (k+1)(m+1)(\beta_n(b_1)-2\varepsilon')_+.\\ &&\precsim (k+1)(m+1)(\beta_n(b_1)-2\varepsilon')_+\oplus (k+1)(m+1)(\beta_n(b_1\oplus b_1)-2\varepsilon')_+. \end{aligned}
Theorem

Theorem 3.2

5S4Y

Let \mathcal{P} be a class of unital {\rm C^*}-algebras which are weakly (m, n)-divisible. Let A be a unital separable simple {\rm C^*}-algebra. If A is asymptotically tracially in \mathcal{P}, then A is weakly (m, n)-divisible.

Proof of Theorem 3.2

We need to show that for any a\in {\rm M}_{\infty}(A)_+, any \varepsilon>0, there exist x_1,x_2, \cdots, x_n \in {\rm M}_{\infty}(A)_+ such that x_j\oplus x_j\oplus \cdots \oplus x_j\precsim a for all 1\leq j\leq n, where x_j repeat m times, and (a-\varepsilon)_+\precsim \oplus_{i=1}^{n}x_i.

We may assume a\in A_+ and \|a\|\leq 1.

With F=\{a\}, any \varepsilon'>0 with \varepsilon' sufficiently small, since A is asymptotically tracially in \mathcal{P}, there exist a {\rm C^*}-algebra B in \mathcal{P} and completely positive contractive linear maps \alpha:A\to B and \beta_n: B\to A, and \gamma_n:A\to A\cap\beta_n(B)^{\perp} such that

(1) the map \alpha is unital completely positive linear map, \beta_n(1_B) and \gamma_n(1_A) are all projections, and \beta_n(1_B)+\gamma_n(1_A)=1_A for all n\in \mathbb{N},

(2) \|x-\gamma_n(x)-\beta_n(\alpha(x))\|<\varepsilon' for all x\in F and for all n\in {\mathbb{N}},

(3) \alpha is an F-\varepsilon'-approximate embedding,

(4) \lim_{n\to \infty}\|\beta_n(xy)-\beta_n(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n(x)\|=\|x\| for all x,y\in B.

Since B has weakly (m,n)-divisible, there exist x_1',x_2', \cdots,x_n' \in {\rm M}_{\infty}(B)_+ such that x_j'\oplus x_j'\oplus \cdots \oplus x_j'\precsim \alpha(a) where x_j' repeat m times and (\alpha(a)-\varepsilon')_+ \precsim \oplus_{i=1}^{n}x_i'.

We divide the proof into two cases.

Case (1), we assume that (\alpha(a)-\varepsilon')_+ is Cuntz equivalent to a projection.

(1.1), If (\alpha(a)-\varepsilon')_+ is not Cuntz equivalent to \oplus_{i=1}^{n}x_i'.

By Theorem 2.1 (2) in [EFF], we may assume that there exist non-zero c\in B_+ such that (\alpha(a)-\varepsilon')_++ c\precsim \oplus_{i=1}^{n}x_i'.

Since x_j'\oplus x_j'\oplus \cdots \oplus x_j'\precsim \alpha(a), there exist v\in M_k(B)_+ such that

\|v^*diag(\alpha(a), 0\otimes 1_{m-1})v-x_j'\otimes 1_m\|<\bar{\varepsilon}.

We assume that \|v\|\leq M(\bar{\varepsilon}), by (4), there exists sufficiently large integer N_1 such that for any n>N_1, we have

\|\beta_n\otimes id_{M_{m}}(v^*)diag(\beta_n\alpha(a), 0\otimes 1_{k-1})\beta_n\otimes id_{M_{k}}(v)-\beta_n(x_j')\otimes 1_m\|<\varepsilon'.

Therefore we have

k(\beta_n(x_j)\oplus\beta_n(x_j')\oplus\cdots\oplus\beta_n(x_j')-4\varepsilon')_+\precsim (\beta_n\alpha(a)-2\varepsilon')_+.

Since (\alpha(a)-\varepsilon')_++ c\precsim \oplus_{i=1}^{n}x_i' there exist w\in M_{n}(B)_+ such that

\|w^*(\oplus_{i=1}^{n}x_i')w-diag((\alpha(a)-\varepsilon')_++ c, 0\otimes 1_{n-1})\|<\bar{\varepsilon}.

We assume that \|w\|\leq N(\bar{\varepsilon}), by (4), there exists sufficiently large integer N_2 such that for any N_2<n we have

\|\beta_n\otimes id_{M_{n}}(w^*)\beta_n(\oplus_{i=1}^{n}x_i')\otimes 1_{n}\beta_n\otimes id_{M_{n}}(w)
-diag(\beta_n\alpha((a)-\varepsilon')_++ c, 0\otimes 1_{n})\|<\varepsilon'.

Therefore we have

(\beta_n\alpha(a)-6\varepsilon')_++c\precsim (\beta_n(\oplus_{i=1}^{n}x_i')-2\varepsilon')_+.

For sufficiently large n>max\{N_1,N_2\}, with F=\{\gamma_n(a)\}, any \varepsilon''>0 with \varepsilon'' sufficiently small, let E=\gamma_n(1_A)A\gamma_n(1_A), by Lemma 2.7, E is asymptotically tracially in \mathcal{P}, there exist a {\rm C^*}-algebra D in \mathcal{P} and completely positive contractive linear maps \alpha':E\to D and \beta_n': D\to E, and \gamma_n':E\to E\cap\beta_n'(D)^{\perp} such that

(1)' the map \alpha' is unital completely positive linear map, \beta_n'(1_D) and \gamma_n'(1_E) are all projections, \beta_n'(1_D)+\gamma_n'(1_E)=1_E for all n\in \mathbb{N},

(2)' \|x-\gamma_n'(x)-\beta_n'(\alpha'(x))\|<\varepsilon'' for all x\in G and for all n\in {\mathbb{N}},

(3)' \alpha' is an G-\varepsilon''-approximate embedding,

(4)' \lim_{n\to \infty}\|\beta_n'(xy)-\beta_n'(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n'(x)\|=\|x\| for all x,y\in D, and

(5)' \gamma_n'\gamma(1_E)\precsim \beta_n(c) for all n\in \mathbb{N}.

Since D is m-almost divisible, and (\alpha'\gamma_n(a)-3\varepsilon')_+\in D, there exists b_2\in D_+ such that kb_2\precsim (\alpha'\gamma_n(a)-3\varepsilon')_+ and (\alpha'\gamma_n(a)-4\varepsilon')_+\precsim (k+1)(m+1)b_2.

With the same argument as above we have

k(\beta_n'(\oplus_{i=1}^{n}x_i'')-2\varepsilon')_+\precsim (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+

and

(\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+\precsim (\beta_n'(\oplus_{i=1}^{n}x_i'')-2\varepsilon')_+.

Therefore we have

\begin{aligned} &&k((\beta_n(\oplus_{i=1}^{n}x_i')-2\varepsilon')_+\oplus (\beta_n'(\oplus_{i=1}^{n}x_i'')-2\varepsilon')_+) \thicksim k((\beta_n(b_1)-2\varepsilon')_++ (\beta_n'(b_2)-2\varepsilon')_+)\\ &&\precsim (\beta_n\alpha(a)-\varepsilon')_+\oplus(\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+ \\ &&\precsim (\beta_n\alpha(a)-\varepsilon')_+\oplus(\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+ +(\gamma_n'\gamma_n(a)-3\varepsilon)_+ \\ &&\precsim a, \end{aligned}

and we also have

\begin{aligned} &&(a-\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon')_+\oplus (\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+ \oplus (\gamma_n'\gamma_n(a)-4\varepsilon')_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon')_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon')_+ \oplus \gamma_n'\gamma_n(1_E)\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon')_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+ \oplus \beta_n(c)\\ &&\precsim (k+1)(m+1)(\beta_n(b_1)-2\varepsilon')_+\oplus (k+1)(m+1)(\beta_n(b_1)-2\varepsilon')_+.\\ &&\precsim (k+1)(m+1)(\beta_n(\oplus_{i=1}^{n}x_i')-2\varepsilon')_+\oplus (k+1)(m+1)(\beta_n(\oplus_{i=1}^{n}x_i')-2\varepsilon')_+. \end{aligned}

(1.1), If (\alpha(a)-\varepsilon')_+ is not Cuntz equivalent to \oplus_{i=1}^{n}x_i'.

With F=\{\gamma_n(a)\}, any \varepsilon''>0 with \varepsilon' sufficiently small, let E=\gamma_n(1_A)A\gamma_n(1_A), by Lemma 2.7, E is asymptotically tracially in \mathcal{P}, there exist a {\rm C^*}-algebra D in \mathcal{P} and completely positive contractive linear maps \alpha':E\to D and \beta_n': D\to E, and \gamma_n':E\to E\cap\beta_n'(D)^{\perp} such that

(1)' the map \alpha' is unital completely positive linear map, \beta_n'(1_D) and \gamma_n'(1_E) are all projections, \beta_n'(1_D)+\gamma_n'(1_E)=1_E for all n\in \mathbb{N},

(2)' \|x-\gamma_n'(x)-\beta_n'(\alpha'(x))\|<\varepsilon'' for all x\in G and for all n\in {\mathbb{N}},

(3)' \alpha' is an G-\varepsilon''-approximate embedding,

(4)' \lim_{n\to \infty}\|\beta_n'(xy)-\beta_n'(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n'(x)\|=\|x\| for all x,y\in D, and

(5)' \gamma_n'(\gamma(1_A))\precsim \beta_n\alpha(c) for all n\in \mathbb{N}.

Since D is weakly (m,n)-divisible, there exist x_1'',x_2'', \cdots,x_n'' \in {\rm M}_{\infty}(D)_+ such that x_j''\oplus x_j''\oplus \cdots \oplus x_j''\precsim (\beta_n'\alpha'\gamma_n(a)-2\varepsilon)_+ where x_j'' repeat m times and (\beta_n'\alpha'\gamma_n(a)-3\varepsilon)_+\precsim \oplus_{i=1}^{n}x_i''.

With the same argument as above we have

k(\beta_n'(b_2)-2\varepsilon')_+\precsim (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+

and

(\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+\precsim (\beta_n'((b_1-2\varepsilon')_+).

Therefore we have

\begin{aligned} &&((x_j'\oplus r)\oplus x_j''))\oplus((x_j'\oplus r)\oplus x_j''))\oplus\cdots \oplus((x_j'\oplus r)\oplus x_j''))\\ &&\precsim (\beta_n\alpha(a)-2\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-2\varepsilon)_+\\ &&\precsim a, \end{aligned}

and

\begin{aligned} &&(x_i'\oplus x_i'')\oplus(x_i'\oplus x_i'')\oplus\cdots \oplus(x_i'\oplus x_i'')\\ &&\precsim (\beta_n\alpha(a)-2\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-2\varepsilon)_+\\ &&\precsim a, \end{aligned}

for all i\neq j and 1\leq i\leq n where (x_i'\oplus x_i'') repeat m times.

We also have

\begin{aligned} &&(a-20\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-3\varepsilon)_+\oplus (\gamma_n'\gamma_n(a)-4\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-3\varepsilon)_+\oplus (\gamma_n'\gamma_n(1_A)-\varepsilon)_+\\ &&\precsim (a'-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-3\varepsilon)_+\oplus r\\ &&\precsim \oplus_{i=1,i\neq j}^{n}(x_i'\oplus x_i'')\oplus ((x_j'\oplus r)\oplus x_j''). \end{aligned}

(1.1.1.2), If x_1',x_2', \cdots,x_k' \in {\rm M}_{\infty}(B)_+ are all projections, and (a'-3\varepsilon)_+< \oplus_{i=1}^{k}x_i'. Then there exists a nonzero projection s such that (a'-3\varepsilon)_+\oplus s \precsim \oplus_{i=1}^{k}x_i'.

Since (\alpha(a)-3\varepsilon)_+ \precsim \oplus_{i=1}^{n}x_i', . Since kb_1\precsim \alpha(a), there exist v\in M_k(B)_+ such that

\|v^*diag(\alpha(a), 0\otimes 1_{k-1}v-b_1\otimes 1\|<\bar{\varepsilon}.

We assume that \|v\|\leq M(\bar{\varepsilon}), by (4), there exists sufficiently large integer n such that

\|\beta_n(v^*)diag(\beta_n\alpha(a), 0\otimes 1_{k-1}\beta_n(v)-\beta_n(b_1)\otimes 1\|<\varepsilon'.

Therefore we have

k(\beta_n(b_1)-2\varepsilon')_+\precsim (\beta_n\alpha(a)-\varepsilon')_+.

With F=\{\gamma_n(a)\}, any \varepsilon'>0 with \varepsilon' sufficiently small, let E=\gamma_n(1_A)A\gamma_n(1_A), by Lemma 2.7, E is asymptotically tracially in \mathcal{P}, there exist a {\rm C^*}-algebra D in \mathcal{P} and completely positive contractive linear maps \alpha':E\to D and \beta_n': D\to E, and \gamma_n':E\to E\cap\beta_n'(D)^{\perp} such that

(1)' the map \alpha' is unital completely positive linear map, \beta_n'(1_D) and \gamma_n'(1_E) are all projections, \beta_n'(1_D)+\gamma_n'(1_E)=1_E for all n\in \mathbb{N},

(2)' \|x-\gamma_n'(x)-\beta_n'(\alpha'(x))\|<\varepsilon'' for all x\in G and for all n\in {\mathbb{N}},

(3)' \alpha' is an G-\varepsilon''-approximate embedding,

(4)' \lim_{n\to \infty}\|\beta_n'(xy)-\beta_n'(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n'(x)\|=\|x\| for all x,y\in D, and

(5)' \gamma_n'(\gamma(1_A))\precsim \beta_n\alpha(c) for all n\in \mathbb{N}.

Since D is weakly (m,n)-divisible, there exist x_1'',x_2'', \cdots,x_n'' \in {\rm M}_{\infty}(D)_+ such that x_j''\oplus x_j''\oplus \cdots \oplus x_j''\precsim (\beta_n'\alpha'\gamma_n(a)-2\varepsilon)_+ where x_j'' repeat m times and (\beta_n'\alpha'\gamma_n(a)-3\varepsilon)_+\precsim \oplus_{i=1}^{n}x_i''. With the same argument as above we have

k(\beta_n'(b_2)-2\varepsilon')_+\precsim (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+

and

(\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+\precsim (\beta_n'((b_1-2\varepsilon')_+).

Therefore we have

\begin{aligned} &&(x_i'\oplus x_i'')\oplus(x_i'\oplus x_i'')\oplus\cdots \oplus(x_i'\oplus x_i'')\\ &&\precsim (a'-2\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-2\varepsilon)_+\\ &&\precsim a\oplus a, \end{aligned}

for 1\leq i\leq n where (x_i'\oplus x_i'') repeat m times.

We also have

\begin{aligned} &&(a-20\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-3\varepsilon)_+\oplus (\gamma_n'\gamma_n(a)-2\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-3\varepsilon)_+\oplus (\gamma_n'\gamma_n(1_A)-\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\oplus s\\ &&\precsim \oplus_{i=1}^{n}(x_i'\oplus x_i''). \end{aligned}

(1.1.1.3), we assume that there is a purely positive element x_1'. Since (a'-2\varepsilon)_+\precsim \oplus_{i=1}^{n}x_i', for any \varepsilon>0, there exists \delta>0, such that (a'-4\varepsilon)_+\precsim (x_1'-\delta)_+ \oplus_{i=2}^{n}x_i',

By Theorem 2.1 (4) in [EFF], there exists a nonzero positive element d such that (x_1'-\delta)_++d\precsim x_1'.

Since (\alpha(a)-3\varepsilon)_+ \precsim \oplus_{i=1}^{n}x_i', . Since kb_1\precsim \alpha(a), there exist v\in M_k(B)_+ such that

\|v^*diag(\alpha(a), 0\otimes 1_{k-1}v-b_1\otimes 1\|<\bar{\varepsilon}.

We assume that \|v\|\leq M(\bar{\varepsilon}), by (4), there exists sufficiently large integer n such that

\|\beta_n(v^*)diag(\beta_n\alpha(a), 0\otimes 1_{k-1}\beta_n(v)-\beta_n(b_1)\otimes 1\|<\varepsilon'.

Therefore we have

k(\beta_n(b_1)-2\varepsilon')_+\precsim (\beta_n\alpha(a)-\varepsilon')_+.

With F=\{\gamma_n(a)\}, any \varepsilon'>0 with \varepsilon' sufficiently small, let E=\gamma_n(1_A)A\gamma_n(1_A), by Lemma 2.7, E is asymptotically tracially in \mathcal{P}, there exist a {\rm C^*}-algebra D in \mathcal{P} and completely positive contractive linear maps \alpha':E\to D and \beta_n': D\to E, and \gamma_n':E\to E\cap\beta_n'(D)^{\perp} such that

(1)' the map \alpha' is unital completely positive linear map, \beta_n'(1_D) and \gamma_n'(1_E) are all projections, \beta_n'(1_D)+\gamma_n'(1_E)=1_E for all n\in \mathbb{N},

(2)' \|x-\gamma_n'(x)-\beta_n'(\alpha'(x))\|<\varepsilon'' for all x\in G and for all n\in {\mathbb{N}},

(3)' \alpha' is an G-\varepsilon''-approximate embedding,

(4)' \lim_{n\to \infty}\|\beta_n'(xy)-\beta_n'(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n'(x)\|=\|x\| for all x,y\in D, and

(5)' \gamma_n'(\gamma(1_A))\precsim \beta_n\alpha(c) for all n\in \mathbb{N}.

Since D is weakly (m,n)-divisible, there exist x_1'',x_2'', \cdots,x_n'' \in {\rm M}_{\infty}(D)_+ such that x_j''\oplus x_j''\oplus \cdots \oplus x_j''\precsim (\gamma_n'\gamma_n(a)-2\varepsilon)_+ where x_j'' repeat m times and (\gamma_n'\gamma_n(a)-3\varepsilon)_+\precsim \oplus_{i=1}^{n}x_i''. With the same argument as above we have

k(\beta_n'(b_2)-2\varepsilon')_+\precsim (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+

and

(\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+\precsim (\beta_n'((b_1-2\varepsilon')_+).

Therefore we have

\begin{aligned} &&(x_i'\oplus x_i'')\oplus(x_i'\oplus x_i'')\oplus\cdots \oplus(x_i'\oplus x_i'')\\ &&\precsim (\beta_n\alpha(a)-2\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-2\varepsilon)_+\\ &&\precsim a, \end{aligned}

for 1\leq i\leq n where (x_i'\oplus x_i'') repeat m times.

We also have

\begin{aligned} &&(a-20\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-3\varepsilon)_+\oplus (\gamma_n'\gamma_n(a)-2\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-3\varepsilon)_+\oplus (\gamma_n'\gamma_n(1_A)-\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\oplus d\\ &&\precsim \oplus_{i=1}^{n}(x_i'\oplus x_i''). \end{aligned}

(1.1.2) We assume that there exists nonzero projection r such that (a'-3\varepsilon)_++r\precsim (a'-2\varepsilon)_+.

Since (\alpha(a)-3\varepsilon)_+ \precsim \oplus_{i=1}^{n}x_i', . Since kb_1\precsim \alpha(a), there exist v\in M_k(B)_+ such that

\|v^*diag(\alpha(a), 0\otimes 1_{k-1}v-b_1\otimes 1\|<\bar{\varepsilon}.

We assume that \|v\|\leq M(\bar{\varepsilon}), by (4), there exists sufficiently large integer n such that

\|\beta_n(v^*)diag(\beta_n\alpha(a), 0\otimes 1_{k-1}\beta_n(v)-\beta_n(b_1)\otimes 1\|<\varepsilon'.

Therefore we have

k(\beta_n(b_1)-2\varepsilon')_+\precsim (\beta_n\alpha(a)-\varepsilon')_+.

With F=\{\gamma_n(a)\}, any \varepsilon'>0 with \varepsilon' sufficiently small, let E=\gamma_n(1_A)A\gamma_n(1_A), by Lemma 2.7, E is asymptotically tracially in \mathcal{P}, there exist a {\rm C^*}-algebra D in \mathcal{P} and completely positive contractive linear maps \alpha':E\to D and \beta_n': D\to E, and \gamma_n':E\to E\cap\beta_n'(D)^{\perp} such that

(1)' the map \alpha' is unital completely positive linear map, \beta_n'(1_D) and \gamma_n'(1_E) are all projections, \beta_n'(1_D)+\gamma_n'(1_E)=1_E for all n\in \mathbb{N},

(2)' \|x-\gamma_n'(x)-\beta_n'(\alpha'(x))\|<\varepsilon'' for all x\in G and for all n\in {\mathbb{N}},

(3)' \alpha' is an G-\varepsilon''-approximate embedding,

(4)' \lim_{n\to \infty}\|\beta_n'(xy)-\beta_n'(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n'(x)\|=\|x\| for all x,y\in D, and

(5)' \gamma_n'(\gamma(1_A))\precsim \beta_n\alpha(c) for all n\in \mathbb{N}.

Since D is weakly (m,n)-divisible, there exist x_1'',x_2'', \cdots,x_n'' \in {\rm M}_{\infty}(D)_+ such that x_j''\oplus x_j''\oplus \cdots \oplus x_j''\precsim (\beta_n'\alpha'\gamma_n(a)-2\varepsilon)_+ where x_j'' repeat m times and (\beta_n'\alpha'\gamma_n(a)-3\varepsilon)_+\precsim \oplus_{i=1}^{n}x_i''. With the same argument as above we have

k(\beta_n'(b_2)-2\varepsilon')_+\precsim (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+

and

(\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+\precsim (\beta_n'((b_1-2\varepsilon')_+).

Therefore we have

\begin{aligned} &&(y_i'\oplus x_i'')\oplus(y_i'\oplus x_i'')\oplus\cdots \oplus(y_i'\oplus x_i'')\\ &&\precsim (\beta_n\alpha(a)-\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-2\varepsilon)_+\\ &&\precsim a, \end{aligned}

for 1\leq i\leq n where (x_i'\oplus x_i'') repeat m times.

We also have

\begin{aligned} &&(a-20\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\oplus (\gamma_n'\gamma_n(a)-3\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\oplus (\gamma_n'\gamma_n(1_A)-\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\oplus r\\ &&\precsim (\beta_n\alpha(a)-2\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\\ &&\precsim \oplus_{i=1}^{n}(y_i'\oplus x_i''). \end{aligned}

(1.2) If (a'-3\varepsilon)_+ is not Cuntz equivalent to a projection.

By Theorem 2.1 (4) in [EFF], there is a non-zero positive element d such that (a'-4\varepsilon)_++d\precsim (a'-3\varepsilon)_+.

Since (\alpha(a)-3\varepsilon)_+ \precsim \oplus_{i=1}^{n}x_i', . Since kb_1\precsim \alpha(a), there exist v\in M_k(B)_+ such that

\|v^*diag(\alpha(a), 0\otimes 1_{k-1}v-b_1\otimes 1\|<\bar{\varepsilon}.

We assume that \|v\|\leq M(\bar{\varepsilon}), by (4), there exists sufficiently large integer n such that

\|\beta_n(v^*)diag(\beta_n\alpha(a), 0\otimes 1_{k-1}\beta_n(v)-\beta_n(b_1)\otimes 1\|<\varepsilon'.

Therefore we have

k(\beta_n(b_1)-2\varepsilon')_+\precsim (\beta_n\alpha(a)-\varepsilon')_+.

With F=\{\gamma_n(a)\}, any \varepsilon'>0 with \varepsilon' sufficiently small, let E=\gamma_n(1_A)A\gamma_n(1_A), by Lemma 2.7, E is asymptotically tracially in \mathcal{P}, there exist a {\rm C^*}-algebra D in \mathcal{P} and completely positive contractive linear maps \alpha':E\to D and \beta_n': D\to E, and \gamma_n':E\to E\cap\beta_n'(D)^{\perp} such that

(1)' the map \alpha' is unital completely positive linear map, \beta_n'(1_D) and \gamma_n'(1_E) are all projections, \beta_n'(1_D)+\gamma_n'(1_E)=1_E for all n\in \mathbb{N},

(2)' \|x-\gamma_n'(x)-\beta_n'(\alpha'(x))\|<\varepsilon'' for all x\in G and for all n\in {\mathbb{N}},

(3)' \alpha' is an G-\varepsilon''-approximate embedding,

(4)' \lim_{n\to \infty}\|\beta_n'(xy)-\beta_n'(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n'(x)\|=\|x\| for all x,y\in D, and

(5)' \gamma_n'(\gamma(1_A))\precsim \beta_n\alpha(c) for all n\in \mathbb{N}.

Since D is weakly (m,n)-divisible, there exist x_1'',x_2'', \cdots,x_n'' \in {\rm M}_{\infty}(D)_+ such that x_j''\oplus x_j''\oplus \cdots \oplus x_j''\precsim (\beta_n'\alpha'\gamma_n(a)-2\varepsilon)_+ where x_j'' repeat m times and (\beta_n'\alpha'\gamma_n(a)-3\varepsilon)_+\precsim \oplus_{i=1}^{n}x_i''. With the same argument as above we have

k(\beta_n'(b_2)-2\varepsilon')_+\precsim (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+

and

(\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+\precsim (\beta_n'((b_1-2\varepsilon')_+).

Therefore we have

\begin{aligned} &&(x_j'\oplus x_j''))\oplus(x_j'\oplus x_j''))\oplus\cdots \oplus(x_j'\oplus x_j''))\\ &&\precsim (a'-2\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-2\varepsilon)_+\\ &&\precsim a, \end{aligned}

for 1\leq j\leq n where (x_j'\oplus x_j'') repeat m times.

We also have

\begin{aligned} &&(a-20\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\oplus (\gamma_n'\gamma_n(a)-2\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\oplus (\gamma_n'\gamma_n(1_A)-\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-4\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\oplus d\\ &&\precsim (\beta_n\alpha(a)-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\\ &&\precsim \oplus_{i=1}^{n}(x_i'\oplus x_i''). \end{aligned}

Case (2), If (a'-2\varepsilon)_+ is not Cuntz equivalent to a projection.

By Theorem 2.1 (4) in [EFF], there is a non-zero positive element d such that (a'-3\varepsilon)_++d\precsim (a'-2\varepsilon)_+.

With F=\{\gamma_n(a)\}, any \varepsilon'>0 with \varepsilon' sufficiently small, let E=\gamma_n(1_A)A\gamma_n(1_A), by Lemma 2.7, E is asymptotically tracially in \mathcal{P}, there exist a {\rm C^*}-algebra D in \mathcal{P} and completely positive contractive linear maps \alpha':E\to D and \beta_n': D\to E, and \gamma_n':E\to E\cap\beta_n'(D)^{\perp} such that

(1)' the map \alpha' is unital completely positive linear map, \beta_n'(1_D) and \gamma_n'(1_E) are all projections, \beta_n'(1_D)+\gamma_n'(1_E)=1_E for all n\in \mathbb{N},

(2)' \|x-\gamma_n'(x)-\beta_n'(\alpha'(x))\|<\varepsilon'' for all x\in G and for all n\in {\mathbb{N}},

(3)' \alpha' is an G-\varepsilon''-approximate embedding,

(4)' \lim_{n\to \infty}\|\beta_n'(xy)-\beta_n'(x)\beta_n(y)\|=0 and \lim_{n\to \infty}\|\beta_n'(x)\|=\|x\| for all x,y\in D, and

(5)' \gamma_n'(\gamma(1_A))\precsim \beta_n\alpha(c) for all n\in \mathbb{N}.

Since D is weakly (m,n)-divisible, there exist x_1'',x_2'', \cdots,x_n'' \in {\rm M}_{\infty}(D)_+ such that x_j''\oplus x_j''\oplus \cdots \oplus x_j''\precsim (\beta_n'\alpha'\gamma_n(a)-2\varepsilon)_+ where x_j'' repeat m times and (\beta_n'\alpha'\gamma_n(a)-3\varepsilon)_+\precsim \oplus_{i=1}^{n}x_i''. With the same argument as above we have

k(\beta_n'(b_2)-2\varepsilon')_+\precsim (\beta_n'\alpha'\gamma_n(a)-\varepsilon')_+

and

(\beta_n'\alpha'\gamma_n(a)-6\varepsilon')_+\precsim (\beta_n'((b_1-2\varepsilon')_+).

Therefore we have

\begin{aligned} &&(y_i'\oplus x_i'')\oplus(y_i'\oplus x_i'')\oplus\cdots \oplus(y_i'\oplus x_i'')\\ &&\precsim (\beta_n\alpha(a)-2\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-2\varepsilon)_+\\ &&\precsim a, \end{aligned}

for 1\leq i\leq n where (x_i'\oplus x_i'') repeat m times.

We also have

\begin{aligned} &&(a-20\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\oplus (\gamma_n'\gamma_n(a)-2\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\oplus (\gamma_n'\gamma_n(1_A)-\varepsilon)_+\\ &&\precsim (\beta_n\alpha(a)-3\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\oplus d\\ &&\precsim (\beta_n\alpha(a)-2\varepsilon)_+\oplus (\beta_n'\alpha'\gamma_n(a)-4\varepsilon)_+\\ &&\precsim \oplus_{i=1}^{n}(y_i'\oplus x_i''). \end{aligned}

Source and licence

The text on this page is from arXiv:2205.15106v2 by Qingzhai Fan and Jiahui Wang, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/2205/2205.15106.jsonl.