, ,

1 Introduction

Consider a type

I I_{1}

factor

M

with trace

τ

, and let

A \subset M

be a maximal abelian von Neumann subalgebra (MASA) of

M

. Following Dixmier ([JDIX] ), let

N_{M} (A) = {u \in A | u unitary, u A u^{*} = A}

be the group of normalizing unitaries of

A

. According to the size of

B = N_{M} (A)^{”}

M

, two extreme situations can occur:

B = M

, that is

B

has enough unitaries to generate

M

and in this case

A

is called regular or

C a r t a n

;

B = A

, in which case the only normalizing unitaries are the unitaries in

A

, and

B

is called singular. Regular/Cartan MASAs appear in a natural way in cross-products :

L^{\infty} [0, 1]

is such a Cartan MASA in its crossed product with some countable group

G

. A far reaching generalization of this is the von Neumann algebra of a countable measure preserving equivalence relation on a standard Borel probability space, which always contains the Cartan MASA

L^{\infty} [0, 1]

(see [FM77] ). On the other hand, until recently, singular MASAs where hard to find.

Dixmier, who started the investigation of MASAs, gave the example of the singular MASA inside the free group factor

ℒ (F_{2})

generated by one of the two generators of

F_{2}

inside the factor. Building on work of Pytlik and Popa ([SP] ,[SP2] ), Rădulescu [FR] showed, using the Pukansky invariant, that the Laplacian (a.k.a radial) subalgebra of

ℒ (F_{N})

is a singular MASA. Together with Boca ([BR] ), they gave more examples of singular MASAs in group von Neumann algebras of certain free products of groups.

The class of strongly singular MASAs, singled out by Sinclair and Smith in [ASRS2] turn out to be more tractable and a plethora of examples can be found now in [ASRS] ,[ASRS2] [GSS] , mostly in the context of group von Neumann algebras or crossed products wth groups. Further more, using the notion of asymptotic homomorphism and good criteria to detect such maps, examples were produced were abelian algebras turned out to be strongly singular (hence also singular) MASAs, with Popa invariant equal to 1 (for more on the Popa invariant see [SP] and section 3 ). In particular, among the first examples of strongly singular MASAs were the ones known to Dixmier (as singular MASAs) and the Laplacian subalgebra of the free group factor.

The aim of this paper is to present yet another set of examples of strongly singular MASAs with Popa invariant 1, living inside tensor products of

ℒ (F_{N})

with itself.

The examples are natural generalizations of the Laplacian subalgebra of

ℒ (F_{N})

, generated by the selfadjoint element

w = \sum_{i = 1}^{N} (g_{i} + g_{i}^{- 1}),

were we identified the generators

g_{1}, . . ., g_{N}

F_{N}

with the unitaries in

ℒ (F_{N})

given by the left regular representation. In

ℒ (F_{N}) {\underset{︸}{\bar{\otimes} . . . \bar{\otimes}}}_{k t i m e s} ℒ (F_{N})

we consider the abelian von Neumann subalgebra generated by

w^{(k)} = \sum_{i = 1}^{N} (g_{i}^{\otimes_{k}} + (g_{i}^{- 1})^{\otimes_{k}}),

and show that this is a strongly singular MASA for each

k \geq 1

The article is organized as follows: section 2 contains preliminaries and notations used in the paper; in section 3 we review the notion of strongly singular MASA and the characterizations providing the tools needed to prove our result. Next (section 4 ) we prove an essential lemma on free groups. In the last section we use this lemma to show the main result.

2 Preliminaries and notations

We denote by

F_{N}

the free group on a fixed set of

N \geq 2

generators

g_{1}, g_{2}, . . ., g_{N}

We will call

S_{N} = {g_{1}, g_{2}, . . ., g_{N}, g_{1}^{- 1}, g_{2}^{- 2}, . . ., g_{N}^{- 1}}

a generating set for

F_{N}

. We denote with

e

the empty word. On

F_{N}

we will consider the canonical length function

| \cdot |

, defined by

| e | = 0

and

| w | = | α_{1} | + | α_{2} | + . . . + | α_{p} |

w \in F_{N}

has the reduced form

g_{i_{1}}^{α_{1}} g_{i_{2}}^{α_{2}} . . . g_{i_{p}}^{α_{p}} (i_{j} \neq i_{j + 1})

The von Neumann group algebra

ℒ (F_{N})

is the norm closure of the left regular representation of

F_{N}

l^{2} (F_{N})

. It is well known that

ℒ (F_{N})

is a type

I I_{1}

factor acting standardly on

l^{2} (F_{N})

and that with this identifcation the

∥ ∥_{τ}

norm coincides with the usual norm

∥ ∥_{2}

l^{2} (F_{N})

. We will use the same notation

∥ \cdot ∥_{2}

for the norm

∥ ∥_{τ}

given by

∥ x ∥_{τ} = (τ (x^{*} x)^{1 / 2})

. We will identify elements in

F_{N}

with their left regular representations. We will also identify the elements of

ℒ (F_{N})

with the corresponding vectors in

L^{2} (ℒ (F_{N}), τ) = l^{2} (F_{N})

H

is a Hilbert space, and

k \geq 0

we denote by

H^{\otimes_{k}}

the

k

-folded tensor product of

H

with itself; for

k = 0

we set

H^{\otimes_{k}} : = H

. An element

v \in H

embeds as

v \otimes . . . \otimes v = : v^{\otimes_{k}} \in H^{\otimes_{k}}

. For a von Neumann algebra

M

M^{{\bar{\otimes}}_{k}}

denotes the

k

-folded von Neumann algebra tensor product of

M

with itself; again, for

k = 0

we set

M^{{\bar{\otimes}}_{k}} : = M

. If

x \in M

, then

x \otimes . . . \otimes x = : x^{\otimes_{k}} \in M^{{\bar{\otimes}}_{k}}

For

n \geq 0

we define

w_{n}^{(k)}

to be the sum of all

k

-folded tensors of reduced words in

F_{N}

of length

n

w_{0}^{(k)} = e^{\otimes_{k}},

w_{1}^{(k)} = \sum_{i = 1}^{N} g_{i}^{\otimes_{k}} + \sum_{i = 1}^{N} (g_{1}^{- 1})^{\otimes_{k}},

w_{n}^{(k)} = \sum_{| v | = n} v^{\otimes_{k}} .

The following relations are well known in

ℒ (F_{N})

and hold for any

k \geq 1

(w_{1}^{(k)})^{2} = w_{2}^{(k)} + 2 N w_{0}^{(k)},

\begin{matrix} w_{1}^{(k)} w_{n}^{(k)} = w_{n}^{(k)} w_{1}^{(k)} = w_{n + 1}^{(k)} + (2 N - 1) w_{n - 1}^{(k)} . \end{matrix}

(2.1)

These relations show that the von Neumann algebra

ℬ^{(k)}

generated by

w_{1}^{(k)}

contains each

w_{n}^{(k)}

, and that this algebra is the weak closure of the span of these elements:

ℬ^{(k)} : = \bar{S p (w_{n}^{(k)} | n \geq 0)^{w}} \subset ℒ (F_{N})^{{\bar{\otimes}}_{k}} .

We call

ℬ^{(k)}

the Laplacian or radial subalgebra, in analogy with the case

k = 0

. The words

w_{n}^{(k)}

are pairwise orthogonal with respect to the trace

τ^{\otimes_{k}}

, and a simple counting argument shows that :

\begin{matrix} ∥ w_{n}^{(k)} ∥_{2}^{2} = 2 N (2 N - 1)^{n - 1}, n \geq 1, \end{matrix}

(2.2)

which represents the number of words of length

n

. The previous remarks show that

{\frac{w_{n}^{(k)}}{∥ w_{n}^{(k)} ∥_{2}} | n \geq 0}

is an orthonormal basis for

L^{2} (ℬ^{(k)}, τ^{\otimes_{k}})

. Moreover, if

E_{k} : ℒ (F_{N})^{{\bar{\otimes}}_{k}} \to ℬ^{(k)}

is the unique conditional expectation satisfying

τ^{\otimes_{k}} \circ E_{k} = τ^{\otimes_{k}}

, then

\begin{matrix} E_{k} (x) = \sum_{n = 0}^{\infty} τ^{\otimes_{k}} (x w_{n}^{(k)}) \frac{w_{n}^{(k)}}{∥ w_{n}^{(k)} ∥_{2}^{2}} . \end{matrix}

(2.3)

In particular, if

x = v^{\otimes_{k}}

with

v

a reduced word of length

p

, then note the following relation for future use:

\begin{matrix} E_{k} (v^{\otimes_{k}}) = \frac{w_{p}^{(k)}}{∥ w_{p}^{(k)} ∥_{2}^{2}} . \end{matrix}

(2.4)

3 Strongly singular MASA's

A maximal abelian selfadjoint subalgebra

A

in a

I I_{1}

factor

ℳ

is singular if any unitary

u \in ℳ

which normalizes

A

(i.e.

u A u^{*} = A

), must lie in

A

. In [SP] , Sorin Popa introduced an invariant for masas of type

I I_{1}

factors. As pointed out in [ASRS] (see [SP] and [ASRS2] for details) a masa

A

has Popa invariant

δ (A) = 1

if and only if for each nilpotent partial isometry

v \in ℳ

with

v A v^{*}

and

v^{*} A v

contained in

A

, we have:

s u p {∥ (I - E_{A}) (x) ∥_{2} : x \in v A v^{*}, ∥ x ∥ \leq 1} \geq ∥ v v^{*} ∥_{2} .

To define the notion of strongly singular masa, consider a linear map

φ : ℳ_{1} \to ℳ_{2}

between two type

I I_{1}

factors. There are several norms for

φ

, depending on the norms considered on the two algebras. When

ℳ_{1}

has the operator norm and

ℳ_{2}

has the

∥ \cdot ∥_{2}

-norm given by the trace, we denote the resulting norm for

φ

∥ φ ∥_{\infty, 2}

, following [ASRS2] .

Definition 3.1. Suppose

A

is a masa in a type

I I_{1}

factor

ℳ

A

is called

α

-strongly singular (or simply strongly singular for

α = 1

) if

∥ E_{u A u^{*}} - E_{A} ∥_{\infty, 2} \geq α ∥ u - E_{A} (u) ∥_{2},

for all unitaries

u \in ℳ

2) The conditional expectation

E_{A}

is an asymptotic homomorphism if there is a unitary

u \in A

such that

{lim}_{| k | \to \infty} ∥ E_{A} (x u^{k} y) - E_{A} (x) E_{A} (y) u^{k} ∥_{2} = 0

for all

x, y \in ℳ

Remark 3.2. As abelian von Neumann algebras,

ℬ^{(k)} \subset B (L^{2} (ℬ^{(k)}, τ^{\otimes_{k}}))

and

ℬ^{(0)} \subset B (L^{2} (ℬ^{(0)}, τ))

are

*

-isomorphic. In [ASRS] it is shown that

E_{0}

is an asymptotic homomorphism and that

ℬ^{(0)}

is a strongly singular masa with Popa invariant

δ (ℬ^{(0)}) = 1

. We will show that this is true for all

k \geq 1

The following provides the criteria we will use in the last section to prove the main result of this paper.

Theorem 3.3. Let

A

be a abelian von Neumann subalgebra of a type

I I_{1}

factor

(ℳ, t r)

, and suppose that there is a

*

-isomorphism

π : A \to L^{\infty} [0, 1]

which induces an isometry from

L^{2} (A, t r)

onto

L^{2} [0, 1]

. Let

{v_{n} \in A | n \geq 0}

be an orthonormal bases for

L^{2} (A, t r)

, and let

Y \subseteq ℳ

be a set whose linear span is norm dense in

L^{2} (ℳ, t r)

. Let

E_{A} : ℳ \to A

be the unique conditional expectation satisfying

t r \circ E_{A} = t r

. If

\sum_{n = 0}^{\infty} ∥ E_{A} (x v_{n} y) - E_{A} (x) E_{A} (y) v_{n} ∥_{2}^{2} < \infty

for all

x, y \in Y

, then

E_{A}

is an asymptotic homomorphism,

A

is a strongly singular masa, and

δ (A) = 1

4 Counting words in $F_{N}$

In this section we recall the needed results from [ASRS] about enumeration of words in

F_{N}

and we prove a technical lemma about solutions of a conjugacy equation in

F_{N}

. For this purposes, let

S_{N} = {g_{1}, g_{2}, . . ., g_{N}, g_{1}^{- 1}, g_{2}^{- 1}, . . ., g_{N}^{- 1}}

be a generating set for the free group

F_{N}

a, b \in F_{N}

are two words such that their concatenation is a reduced word, we will denote this by

a \cdot b

. In general, the product (concatenation) will be denoted

a b

, meaning that cancellations may or may not take place.

For

x, y \in S_{N}

, we let

ν_{n} (x, y)

be the number of words of length

n

starting with

x

and ending with

y

. If

σ, ρ \subset S_{N}

, denote by

ν_{n} (σ, ρ)

the number of words of length

n

starting in

σ

and ending in

ρ

. Note that

\begin{matrix} ν_{n} (σ, ρ) = \sum_{x \in σ, y \in ρ} ν_{n} (x, y) . \end{matrix}

(4.1)

The next two results can be found in [ASRS] .

Proposition 4.1. There exists a constant

D_{N}

, depending only on

N

, such that

| ν_{n} (σ_{1}, ρ_{1}) - ν_{n} (σ_{2}, ρ_{2}) | \leq D_{N}

whenever

σ_{1}, ρ_{1}, σ_{2}, ρ_{2}

are non-empty subsets of

S_{N}

satisfying

| σ_{1} | = | σ_{2} |, | ρ_{1} | = | ρ_{2} | .

With the notations from section 2 , let

w_{n} = w_{n}^{(0)} = \sum_{| v | = n} v \in ℒ (F_{N}) .

Proposition 4.2. Let

x = x_{l} \cdot . . . \cdot x_{1}

and

y = y_{1} \cdot . . . \cdot y_{m}

be words in

F_{N}

, with

x_{i}, y_{j} \in S_{N}

. Let

μ (r, s, n; x, y)

be the number of reduced words in the product

x w_{n} y

which result from

r

cancellations on the left and

s

cancellations on the right. Then there exist subsets

σ_{r} (x), ρ_{s} (y)

S_{N}

, whose cardinalities depend only on

r

and

s

respectively, such that

μ (r, s, n; x, y) = ν_{n - r - s} (σ_{r} (x), ρ_{s} (y)),

for

n \geq l + m + 2, 0 \leq r \leq l, 0 \leq s \leq m

Lemma 4.3. Let

a, b \in F_{N}

be two distinct non-trivial words. For

k \geq 1

big enough, there are at most two

w \in F_{N}

of length

k

satisfying

w a = b w

Proof: Suppose

w a = b w

with

| w | = k \geq 1

case I: no cancellations, i.e.

w a = w \cdot a

and

b w = b \cdot w

. Then

| a | = | b |

, otherwise there is no solution. The equation becomes:

w \cdot a = b \cdot w .

$∙$ case I.1: $k \leq | a | = | b |$ . There must be a $b_{1} \in F_{N}$ such that $b = w \cdot b_{1}$ .
Hence $a = b_{1} \cdot w$ , and $w = b_{1}^{- 1} a$ is the unique solution. Further more, this implies restrictions on $a, b$ : $b = w \cdot b_{1} = (b^{- 1} a) \cdot a$ . But then $| b | = | b_{1}^{- 1} a | + | b_{1} |$ , so $| b^{- 1} a | = | a | - | b_{1} |$ . We conclude that in this case we have a solution if and only if $a = b_{1} \cdot a_{1}$ , $b = a_{1} \cdot b_{1}$ . The (unique) solution is $w = a_{1} = b_{1}^{- 1} a$ , and $| a_{1} | = | b_{1} |$ . In particular, $b_{1} = 1$ implies $a = b = w$ .
$∙$ case I.2 : $k > | a | = | b | .$ Then $w = b^{l} \cdot w_{1}$ with $l \geq 1$ maximal and hence $| w_{1} | < | b | = | a |$ . We have ${b̸}^{l} \cdot w_{1} \cdot a = {b̸}^{l} b \cdot w_{1}$ and again, as in case I.1, there is at most one $w_{1}$ (so at most one $w$ ) satisfying $w_{1} \cdot a = b \cdot w_{1}$ .

So for fixed length

k

, there is a most one word

w

so that

w \cdot a = b \cdot w

case II: there are cancellations. Take

k > | a | + | b |

. Then there are words

a_{1}, a_{2}, b_{1}, b_{2}

and

w_{1}

such that

a = a_{1} \cdot a_{2},

b = b_{2} \cdot b_{1},

w = b_{1}^{- 1} \cdot w_{1} \cdot a_{1}^{- 1}, | w_{1} | \geq 1,

and at least one of

a_{1}

b_{1}

is a nonempty word. Then

b_{1}^{- 1} \cdot w_{1} \cdot a_{2} = b_{2} \cdot w_{1} \cdot a_{1}^{- 1} \Rightarrow

\begin{matrix} | b_{1}^{- 1} | + | a_{2} | = | b_{2} | + | a_{1}^{- 1} | . \end{matrix}

(4.2)

Note also that 4.2

\begin{matrix} | a_{1} | + | a_{2} | = | a |, | b_{1} | + | b_{2} | = | b | \overset{()}{\Rightarrow} | a_{1} | + | b_{1} | = \frac{| a | + | b |}{2} = | a_{2} | + | b_{2} | . \end{matrix}

(4.3)

By hypothesis, we have

| a_{1} | > 0

| b_{1} | > 0

$∙$ case II.1 : $| a_{1} | = 0$ . Then $| b_{1} | > 0$ and $| b_{1} | = \frac{| b | - | a |}{2} < | b_{2} | = \frac{| b | + | a |}{2} .$ We have $w_{1} \cdot a = (b_{1} b_{2}) \cdot w_{1}$ . As in case II.1, $w_{1}$ is uniquely determined, so there is at most one solution for $w a = b w$ .
$∙$ case II.2 : $| b_{1} | = 0$ . This case is similar to previous one. We get at most one solution $w$ for $w a = b w$ .

∙

case II.3 :

| a_{1} | > 0

and

| b_{1} | > 0

\begin{matrix} claim: | a_{1} | = | a_{2} | and | b_{1} | = | b_{2} | . \end{matrix}

(4.4)

Suppose for example that

0 < | a_{1} | < | a_{2} |

. From relations ( 4.3 ) we get

0 < | a_{1} | < \frac{| a |}{2},

0 < | b_{1} | < | b_{2} |, hence 0 < | b_{1} | < \frac{| b |}{2} \Rightarrow

| a_{1} | + | b_{1} | < \frac{| a | + | b |}{2},

which contradicts ( 4.3 ). We conclude that in this case there are solutions only if

a, b

have even lengths. Since

a = a_{1} \cdot a_{2},

b = b_{2} \cdot b_{1},

with

| a_{1} | = | a_{2} |

and

| b_{1} | = | b_{2} |

and both

a, b

nonempty words, we cannot have

a_{2} a_{1} = e

nor

b_{1} b_{2} = e

. The initial equation

w a = b w

becomes

w_{1} \cdot (a_{2} a_{1}) = (b_{1} b_{2}) \cdot w_{1} .

Even though it does not affect the number of solutions, it is worth mentioning that this also implies

| a_{2} a_{1} | = | b_{1} b_{2} |

. As in case II.1, there is at most one solution of given length for the equation

w a = b w

We have finished all possible situations. To resume, here are the possible solutions:

$∙$ no cancellations: solutions only if $a = a_{1} \cdot b_{1}, b = b_{1} \cdot a_{1}$ . In particular, note that $| a | = | b |$ .
$∙$ only ”a” cancellations: solutions only if $| a | > | b |$ .
$∙$ only ”b” cancellations: solutions only if $| b | > | a |$ .
$∙$ ”a” and ”b” cancellations: solutions only if $| a |, | b |$ are both even, $a = a_{1} \cdot a_{2}$ with $| a_{1} | = | a_{2} |$ and $b = b_{2} \cdot b_{1}$ with $| b_{1} | = | b_{2} |$ . Further more $0 \neq | a_{2} a_{1} | = | b_{1} b_{2} |$ .

In all cases, there is at most one solution of prescribed length. To count solutions, we compare

| a |

and

| b |

$∙$ If $| a | \neq | b |$ then for fixed length $k$ there is at most 1 solution.
$∙$ If $| a | = | b |$ then for fixed length $k$ there are at most 2 possible solutions one without cancellations and one with ”a” and ”b” cancellations.

We conclude that for any given length

k

, there are at most two solutions

w

for

w a = b w

5 The case of tensors of the Free Group Factors

In what follows we will identify elements

g \in F_{N}

with their left regular representations

λ (g) \in ℒ (F_{N})

and we will use the notations from section 2 .

Lemma 5.1. For

k \geq 1

let

x_{1}, . . ., x_{k} \in F_{N}

. With the notations from section 2 ,

E_{k} (x_{1} \otimes . . . \otimes x_{k}) \neq 0 \Leftrightarrow x_{1} = . . . = x_{k} .

Proof: Recall from ( 2.3 ) that

E_{k} (x_{1} \otimes . . . \otimes x_{k}) = \sum_{n = 0}^{\infty} τ^{\otimes_{k}} (x_{1} \otimes . . . \otimes x_{k} w_{n}^{(k)}) \frac{w_{n}^{(k)}}{∥ w_{n}^{(k)} ∥_{2}^{2}} .

Therefore :

E_{k} (x_{1} \otimes . . . \otimes x_{k}) \neq 0 \Leftrightarrow

\exists n \geq 0 : τ^{\otimes_{k}} (x_{1} \otimes . . . \otimes x_{k} w_{n}^{(k)}) \neq 0 \Leftrightarrow

\exists n \geq 0 : \sum_{| v | = n} τ^{\otimes_{k}} (x_{1} v \otimes . . . \otimes x_{k} v) \neq 0 \Leftrightarrow

\exists n \geq 0 : \sum_{| v | = n} τ (x_{1} v) . . . τ (x_{k} v) \neq 0 \Leftrightarrow

\exists n \geq 0 : x_{1} = . . . = x_{k} = v .

The next technical lemma contains the calculations needed to prove the main result.

Lemma 5.2. Let

k \geq 1

and

w_{n}^{(k)} = \sum_{| v | = n} v^{\otimes_{k}}

. Let

x_{1}, . . ., x_{k}, y_{1}, . . ., y_{k} \in F_{N}

Then:

\begin{matrix} \sum_{n = 0}^{\infty} \frac{1}{∥ w_{n}^{(k)} ∥_{2}^{2}} ∥ E_{k} (x_{1} \otimes . . . \otimes x_{k} w_{n}^{(k)} y_{1} \otimes . . . \otimes y_{k}) - E_{k} (x_{1} \otimes . . . \otimes x_{k}) E_{k} (y_{1} \otimes . . . \otimes y_{k}) w_{n}^{(k)} ∥_{2}^{2} < \infty . \end{matrix}

(5.1)

Proof: case I:

E_{k} (x_{1} \otimes . . . \otimes x_{k}) E_{k} (y_{1} \otimes . . . \otimes y_{k}) \neq 0

. By the previous lemma 5.1 , it follows that

x_{1} = . . . = x_{k} = : x

and

y_{1} = . . . = y_{k} = : y

. Let

l = | x |, m = | y |

l = 1

m = 1

the inequality ( 5.1 ) is trivial. Indeed, suppose

l = 1

(the other case is even easier). Then we actually have

∥ E_{k} (x^{\otimes_{k}} w_{n}^{(k)} y^{\otimes_{k}}) - E_{k} (x^{\otimes_{k}}) E_{k} (y^{\otimes_{k}}) w_{n}^{(k)} ∥_{2}^{2} =

∥ E_{k} (w_{n}^{(k)} y^{\otimes_{k}}) - E_{k} (y^{\otimes_{k}}) w_{n}^{(k)} ∥_{2}^{2} =

∥ w_{n}^{(k)} E_{k} (y^{\otimes_{k}}) - E_{k} (y^{\otimes_{k}}) w_{n}^{(k)} ∥_{2}^{2} = 0,

because

ℬ^{(k)}

is abelian. We are left with the case

l, p \geq 1

. The estimations bellow can be found in [ASRS] , our proof is just a straightforward generalization.

We reproduce them here for convenience.

Let

z

be an arbitrary word of length

l

, and suppose that

n \geq l + m + 2

. Thus, with the notation of proposition 4.2 ,

E_{k} (x^{\otimes_{k}} w_{n}^{(k)} y^{\otimes_{k}}) = \sum_{r = 0}^{l} \sum_{s = 0}^{m} ν_{n - r - s} (σ_{r} (x), ρ_{s} (y)) w_{n + l + m - 2 (r + s)} ∥ w_{n + l + m - 2 (r + s)} ∥_{2}^{- 2} .

By propositions 4.1 , 4.2

\begin{matrix} | ν_{p} (σ_{r} (x), τ_{s} (y) - ν_{p} (σ_{r} (z), τ_{s} (y) | \leq D_{N} \end{matrix}

(5.2)

for

p \geq 2

. Thus there exist constants

λ_{r, s}

, uniformly bounded by

D_{N}

, such that

\begin{matrix} E_{k} (x^{\otimes_{k}} w_{n}^{(k)} y^{\otimes_{k}}) - E_{k} (z^{\otimes_{k}} w_{n}^{(k)} y^{\otimes_{k}}) = \sum_{r = 0}^{l} \sum_{s = 0}^{m} λ_{r, s} w_{n + l + m - 2 (r + s)} ∥ w_{n + l + m - 2 (r + s)} ∥_{2}^{- 2} . \end{matrix}

(5.3)

Since

∥ w_{p}^{(k)} ∥_{2}^{2}

is the number of terms in

w_{p}^{(k)}

, we have:

\begin{matrix} ∥ w_{p}^{(k)} ∥_{2}^{2} = 2 N (2 N - 1)^{p - 1} = (2 N - 1)^{p - n} ∥ w_{n}^{(k)} ∥_{2}^{2} . \end{matrix}

(5.4)

Thus

∥ w_{n + l + m - 2 (r + s)}^{(k)} ∥_{2}^{2} = (2 N - 1)^{l + m - 2 (r + s)} ∥ w_{n}^{(k)} ∥_{2}^{2},

∥ w_{n + l + m - 2 (r + s)}^{(k)} ∥_{2}^{2} \geq (2 N - 1)^{- (l + m)} ∥ w_{n}^{(k)} ∥_{2}^{2},

for

0 \leq r \leq l, 0 \leq s \leq m

. This last inequality and 5.3 then give the estimate

∥ E_{k} (x^{\otimes_{k}} w_{n}^{(k)} y^{\otimes_{k}}) - E_{k} (z^{\otimes_{k}} w_{n}^{(k)} y^{\otimes_{k}}) ∥_{2}^{2} \leq (l + 1) (m + 1) D_{N} (2 N - 1)^{(l + m) / 2} ∥ w_{n}^{(k)} ∥_{2}^{- 1} .

Let

H_{l, m, N}

be the constant on the right hand side of the above estimate. If we sum this inequality over all words

z

of lenght

l

(of which there are

∥ w_{l}^{(k)} ∥_{2}^{2}

), we obtain

∥ ∥ w_{l}^{(k)} ∥_{2}^{2} E_{k} (x^{\otimes_{k}} w_{n}^{(k)} y^{\otimes_{k}}) - E_{k} (w_{l}^{(k)} w_{n}^{(k)} y^{\otimes_{k}}) ∥_{2}^{2} \leq ∥ w_{l}^{(k)} ∥_{2}^{2} H_{l, m, N} ∥ w_{n}^{(k)} ∥_{2}^{- 1} .

Since

E_{k} (x^{\otimes_{k}}) = w_{l}^{(k)} ∥ w_{l}^{(k)} ∥_{2}^{- 2}

and

E_{k} (w_{l}^{(k)} w_{n}^{(k)} y) = w_{l}^{(k)} w_{n}^{(k)} E_{k} (y)

, it follows that

∥ E_{k} (x^{\otimes_{k}} w_{n}^{(k)} y^{\otimes_{k}}) - E_{k} (x^{\otimes_{k}}) E_{k} (y^{\otimes_{k}}) w_{n}^{(k)} ∥_{2} \leq H_{l, m, N} ∥ w_{n}^{(k)} ∥_{2}^{- 1} .

and so 2.2

\sum_{n = 0}^{\infty} \frac{1}{∥ w_{n}^{(k)} ∥_{2}^{2}} ∥ E_{k} (x^{\otimes_{k}} w_{n}^{(k)} y^{\otimes_{k}}) - E_{k} (x^{\otimes_{k}}) E_{k} (y^{\otimes_{k}}) w_{n}^{(k)} ∥_{2}^{2} \leq \sum_{n = 0}^{\infty} H_{l, m, N}^{2} ∥ w_{n}^{(k)} ∥_{2}^{- 4} \overset{()}{<} \infty .

case II:

E_{k} (x_{1} \otimes . . . \otimes x_{k}) E_{k} (y_{1} \otimes . . . \otimes y_{k}) = 0

. By the lemma 5.1 , there must be at least to distinct

x_{i_{1}}, x_{i_{2}}

and at least two distinct

y_{j_{1}}, y_{j_{2}}

. The nonzero terms in the series ( 5.1 )

\sum_{n = 0}^{\infty} \frac{1}{∥ w_{n}^{(k)} ∥_{2}^{2}} ∥ E_{k} (x_{1} \otimes . . . \otimes x_{k} w_{n}^{(k)} y_{1} \otimes . . . \otimes y_{k}) ∥_{2}^{2} =

\sum_{n = 0}^{\infty} \frac{1}{∥ w_{n}^{(k)} ∥_{2}^{2}} ∥ \sum_{| v | = n} E_{k} (x_{1} v y_{1} \otimes . . . \otimes x_{k} v y_{k}) ∥_{2}^{2}

are given by the solutions

v

of length

n \geq 0

of the equations:

x_{1} v y_{1} = x_{2} v y_{2} = . . . = x_{k} v y_{k} .

Remark that a solution exists only if

x_{i} = x_{j} \Leftrightarrow y_{i} = y_{j}

for

1 \leq i, j \leq k

Without loss of generality, assume

x_{1} \neq x_{2} (\Leftrightarrow y_{1} \neq y_{2})

. Any solution of length

n

for the above system of equations is a solution of

v (y_{1} y_{2}^{- 1}) = (x_{1}^{- 1} x_{2}) v

. Clearly

x_{1}^{- 1} x_{2} \neq e \neq y_{1} y_{2}^{- 1}

. By lemma 4.3 , for each fixed

n > m a x {| x_{i} | : 1 \leq i \leq k} + m a x {| y_{j} | : 1 \leq j \leq k} = : n_{0}

there are at most 2 solutions to this equation. So

\sum_{n = n_{0}}^{\infty} \frac{1}{∥ w_{n}^{(k)} ∥_{2}^{2}} ∥ \sum_{| v | = n} E_{k} (x_{1} v y_{1} \otimes . . . \otimes x_{k} v y_{k}) ∥_{2}^{2} \leq

\sum_{n = n_{0}}^{\infty} \frac{1}{∥ w_{n}^{(k)} ∥_{2}^{2}} ∥ 2 \frac{w_{n}^{(k)}}{∥ w_{n}^{(k)} ∥_{2}^{2}} ∥_{2}^{2} =

2.2

\sum_{n = n_{0}}^{\infty} \frac{2}{∥ w_{n}^{(k)} ∥_{2}^{4}} \overset{()}{<} \infty .

Theorem 5.3. For each

k \geq 0

,the conditional expectation

E_{k}

onto the laplacian subalgebra

ℬ^{(k)}

ℒ (F_{N})^{{\bar{\otimes}}_{k}}

, generated by

w_{1}^{(k)} = \sum_{v \in F_{N}, | v | = 1} v^{\otimes_{k}},

is an asymptotic homomorphism and

ℬ^{(k)}

is a strongly singular masa with Popa invariant

δ (ℬ^{(k)}) = 1

Proof: For

k = 0

ℒ (F_{N})^{{\bar{\otimes}}_{0}} : = ℒ (F_{N})

and the result was proved in [ASRS] .

For

k \geq 1

, apply the criteria given by theorem 3.3 . Use remark 3.2 and the proof for

k = 0

to verify the statement about the existence of the

*

-isometry. Use

Y = {x_{1} \otimes . . . \otimes x_{k} | x_{i} \in F_{N}, 1 \leq i \leq k}

and

v_{n} : = w_{n}^{(k)} / ∥ w_{n}^{(k)} ∥_{2}, n \geq 0

. The theorem is a direct consequence of the above technical lemma and theorem 3.3 .

Acknowledgements. First I want to thank once again my advisor, Florin Rădulescu, for his constant support and patience. I would also like to thank Ionuţ Chifan and Dorin Dutkay for valuable discussions and help.

References

Boca, F.; Rădulescu, F. Singularity of radial subalgebras in $I I_{1}$ factors asociated with free products of groups, J. Funct. Anal., 103 (1992), no. 1, 138–159.
Dixmier, J. Sous-anneaux amxmaux dans les facteurs de type fini, Ann. Math., 59 (1954), 279–286.
Feldman, J.; Moore, C. Ergodic equivalence relations, cohomology and von Neumann algebras I, II, Trans. Amer. Math. Soc., 234 (1977), 289–324 and 325-359.
Popa, S. Notes on Cartan subalgebras in type $I I_{1}$ factors, Math. Scand., 57 (1985), 171–188.
Popa, S. Singular maximal abelian $*$ -subalgebras in continuous von Neumann algebras, J. Funct. Anal., 103 (1983), 151–166.
Rădulescu, F. Singularity of the radial subalgebra of $ℒ (F_{N})$ and the Pukánsky invariant, Pacific J. Math. , 151, no.2 (1991), pp 297-306.
Robertson, G.; Sinclair, A. M.; Smith, R. R. Strong singularity for subalgebras of finite factors, Intern. J. Math., 14, No. 3 (2003) 235–258.
Sinclair, A. M.; Smith, R. R. The Laplacian MASA in a free group factor Trans. Amer. Math. Soc. 355 (2003), no. 2, 465–475.
Sinclair, A. M.; Smith, R. R. Strongly singular masas in type $I I_{1}$ factors Geom. Funct. Anal. 12 (2002), no. 1, 199–216.

Department of Mathematics, The University of Iowa, 14 MacLean Hall, Iowa City, IA 52242-1419, U.S.A., E-mail address : sbildea@math.uiowa.edu

Teodor Ş tefan Bîldea