---
title: Von Neumann Unitaries & Their Properties
url: https://www.emergentmind.com/topics/von-neumann-unitaries
type: topic
---

# Von Neumann Unitaries & Their Properties

Von Neumann unitaries are the unitary elements of a von Neumann algebra \(M\subset B(H)\), organized either as the topological group
\[
U_M=\{\,u\in M\mid u^*u=uu^*=1\,\}
\]
or through special subclasses such as symmetries, namely self-adjoint unitaries. Two complementary lines of analysis are particularly prominent: periodicity properties of conjugation coefficient functions \(u\mapsto \langle uxu^*\xi\mid\eta\rangle\), which characterize finiteness and almost-periodicity phenomena of \(M\), and factorization properties expressing unitaries as products of symmetries, especially in type \(II_1\) von Neumann algebras [2006.08146]; [2204.00009].

## 1. The unitary group as a topological object

Let \(M\subset B(H)\) be a von Neumann algebra acting on a Hilbert space \(H\). Its unitary group \(U_M\) is considered as a topological group under either the weak operator topology or the strong operator topology, and these topologies agree on \(U_M\). A net \(u_i\to u\) in the weak operator topology satisfies
\[
\langle u_i\xi\mid\eta\rangle\longrightarrow \langle u\xi\mid\eta\rangle
\qquad (\forall\,\xi,\eta\in H),
\]
whereas a net \(u_i\to u\) in the strong operator topology satisfies
\[
\|u_i\xi-u\xi\|\longrightarrow 0
\qquad (\forall\,\xi\in H).
\]
These are Polish group topologies on \(U_M\) [2006.08146].

A basic family of functions on \(U_M\) is given by the conjugation coefficient functions
\[
(\xi\star T\star\eta)(u):=\langle uTu^*\xi\mid\eta\rangle,
\]
defined for \(T\in B(H)\) and \(\xi,\eta\in H\). When \(T=x\in M\), these functions encode the orbit of \(x\) under the inner automorphism action \(u\mapsto uxu^*\). In the setting of von Neumann algebras, this orbit structure becomes a bridge between operator-algebraic properties of \(M\) and dynamical properties of \(U_M\) as a topological group.

The same unitary group also supports a second, more algebraic viewpoint. In any von Neumann algebra \(\mathscr M\), a symmetry is a self-adjoint unitary \(S\in\mathscr M\) with \(S^*=S\) and \(S^2=I\). Products of such symmetries generate large parts of the unitary group, and in type \(II_1\) algebras they admit explicit finite-length factorization results [2204.00009].

## 2. Weakly almost periodic and almost periodic coefficient functions

For a topological group \(G\), let
\[
C_b(G)=\{\,f\colon G\to\mathbb C\mid f \text{ continuous and bounded}\,\},
\]
and let \(b,r(G)\subset C_b(G)\) denote the subalgebra of right-uniformly-continuous functions. A function \(f\in b,r(G)\) is weakly almost periodic if its left orbit \(\{g\cdot f\mid g\in G\}\) is relatively weakly compact in the Banach space \(b,r(G)\), and it is almost periodic if that orbit is norm-relatively compact. Equivalently, weak almost periodicity may be tested by the right orbit or by Grothendieck’s double-limit criterion:
\[
f\in WAP(G)\Longleftrightarrow
\forall (x_i),(y_j)\subset G \text{ with both } \lim_i\lim_j f(x_i y_j),\ \lim_j\lim_i f(x_i y_j) \text{ existing, they are equal.}
\]
The space \(WAP(G)\) is a unital \(C^*\)-subalgebra of \(b,r(G)\), invariant under left and right translations [2006.08146].

A crucial structural fact is the existence of a unique invariant mean on \(WAP(G)\). There is a unique state
\[
m:WAP(G)\longrightarrow \mathbb C
\]
such that \(m(f)\ge 0\) if \(f\ge 0\), \(m(1)=1\), and
\[
m(g\cdot f)=m(f)=m(f\cdot g)
\qquad (\forall\,g\in G).
\]
Moreover, \(m(f)\) lies in the closed convex hull of the left orbit of \(f\), and likewise of the right orbit. For \(G=U_M\), this invariant mean on \(WAP(U_M)\) is the main device in the analysis of conjugation coefficients and the operator systems they generate [2006.08146].

In this framework, the distinction between weak almost periodicity and almost periodicity is substantive rather than terminological. Weak compactness of orbits is sufficient to recover finite-type structure through invariant averaging, whereas norm-compactness forces a significantly stronger decomposition of the underlying algebra.

## 3. Finiteness characterized by weakly almost periodic conjugation

A central theorem identifies finiteness of a von Neumann algebra with weak almost periodicity of all conjugation coefficient functions. For \(M\subset B(H)\), the following are equivalent:

1. \(M\) is finite, i.e. it admits a faithful normal center-valued trace.
2. For every \(x\in M\) and \(\xi,\eta\in H\), the coefficient function
   \[
   u\longmapsto \langle uxu^*\xi\mid\eta\rangle
   \]
   belongs to \(WAP(U_M)\) [2006.08146].

When these conditions hold, the invariant mean yields a completely positive, unital, \(M'\)-bimodular conditional expectation
\[
E_M:C^*(M,M')\longrightarrow M'
\]
whose restriction to \(M\) is precisely the center-valued trace of \(M\). Thus the periodicity of scalar-valued coefficient functions on the unitary group recovers a canonical operator-algebraic averaging map.

The proof mechanism has two directions. If every \(x\in M\) has weakly almost periodic coefficients, then \(M\) sits inside the operator system \(wap_M\) of operators whose coefficients lie in \(WAP(U_M)\), and the invariant mean defines
\[
\langle E(T)\xi\mid\eta\rangle = m(\xi\star T\star\eta).
\]
Restricted to \(M\), this map is a unital \(M'\)-bimodule map satisfying \(E(uxu^*)=E(x)\), which forces \(E\) to agree with the unique center-valued trace. Conversely, if \(M\) is finite, then the group of inner automorphisms
\[
\operatorname{int}(M)=\{\,\operatorname{Ad}(u)\mid u\in U_M\}
\]
is relatively weak\(^*\)-compact in the Banach space \(B(M)\), and Grothendieck’s criterion shows that each conjugation coefficient lies in \(WAP(U_M)\) [2006.08146].

This theorem places finiteness in a dynamical form: it is detected not by traces alone, but by weak compactness of the scalar orbit data generated by conjugation under unitaries.

## 4. Almost periodicity, direct-sum structure, and minimal almost periodicity

The almost-periodic analogue is more restrictive. For \(M\subset B(H)\), the following are equivalent:

1. For every \(x\in M\) and \(\xi,\eta\in H\), the coefficient function
   \[
   u\mapsto \langle uxu^*\xi\mid\eta\rangle
   \]
   lies in \(AP(U_M)\).
2. \(M\) decomposes as a direct sum
   \[
   M\cong A\oplus\Bigl(\bigoplus_{k\ge 1} M_{n_k}(\mathbb C)\Bigr),
   \]
   where \(A\) is a diffuse abelian von Neumann algebra and each \(M_{n_k}(\mathbb C)\) is a finite-dimensional factor [2006.08146].

Equivalently stated in the source, all conjugation-coefficient functions are almost periodic if and only if no infinite faithful atomic part of \(M\) remains. The proof again proceeds through an expectation on the operator system of almost-periodic operators \(ap_M\). If \(M\subset ap_M\), then all central summands of type \(I_\infty\), type \(II\), or type \(III\) must vanish, and on the atomic type \(I_n\) part one recovers only finite-dimensional blocks. Conversely, for
\[
M\cong A\oplus\bigoplus_{k\ge1}M_{n_k}
\]
with \(A\) abelian and diffuse, each coefficient function is a uniform limit of matrix coefficients of finite-dimensional unitary representations of \(U_M\), hence is almost periodic [2006.08146].

A related but distinct notion is minimal almost periodicity. A topological group \(G\) is minimally almost periodic if its only continuous finite-dimensional irreducible unitary representation is the trivial one; equivalently, \(AP(G)=\mathbb C\). If \(M\) is a diffuse von Neumann algebra, then \(U_M\) is minimally almost periodic. The proof proceeds by showing first that a diffuse maximal abelian subalgebra \(A\cong L^\infty(X)\subset M\) has unitary group \(U_A\) with no non-trivial characters, and then passing from the abelian case to general diffuse \(M\) by restriction to maximal abelian subalgebras [2006.08146].

These statements isolate two different levels of periodicity. The source material implies that the condition \(AP(U_M)=\mathbb C\) for diffuse \(M\) concerns finite-dimensional unitary representations of the group itself, whereas the stronger requirement that all conjugation-coefficient functions be almost periodic forces a direct-sum decomposition with a diffuse abelian part and finite-dimensional factors.

## 5. Products of symmetries in von Neumann algebras

In a von Neumann algebra \(\mathscr M\), the set of symmetries is
\[
\mathscr S(\mathscr M)=\{\,S\in \mathscr U(\mathscr M)\mid S^*=S,\ S^2=I\,\},
\]
and for \(n\in\mathbb N\),
\[
\mathscr S(\mathscr M)^n
=
\{\,U\in\mathscr U(\mathscr M)\mid U=S_1S_2\cdots S_n,\ S_j\in\mathscr S(\mathscr M)\,\}.
\]
For type \(II_1\) von Neumann algebras, the factorization theory is especially explicit [2204.00009].

| Setting | Conclusion | Citation |
|---|---|---|
| \(\mathscr R\) type \(II_1\), arbitrary \(U\in \mathscr U(\mathscr R)\) | Every unitary is a product of six symmetries | [2204.00009] |
| \(\mathscr R\) type \(II_1\), \(U\) with finite spectrum | Every such unitary is a product of four symmetries | [2204.00009] |
| \(\mathscr R\) type \(II_1\) | \(\overline{\mathscr S(\mathscr R)^4}^{\|\cdot\|}=\mathscr U(\mathscr R)\) | [2204.00009] |
| Arbitrary von Neumann algebra \(\mathscr M\) | \(\overline{\mathscr S(\mathscr M)^3}^{\|\cdot\|}\subsetneqq \mathscr U(\mathscr M)\) | [2204.00009] |

The failure at three factors is not merely a failure of exact generation. The source exhibits a non-empty open set of unitaries—such as those whose spectrum sits inside one of the four open arcs of the circle obtained by removing \(\{1,i,-1,-1\}\)—none of which can be written as a product of three symmetries. Thus the three-symmetry obstruction is topologically stable in the norm topology [2204.00009].

This produces a sharp asymmetry between four and three factors in the type \(II_1\) setting: four symmetries are already norm-dense, while three are never norm-dense in any von Neumann algebra.

## 6. Decomposition mechanisms and classical placement

The six-symmetry theorem for type \(II_1\) algebras is proved by an infinite block-diagonal cutting argument. In a maximal abelian subalgebra \(\mathcal A\) containing a unitary \(U\), one chooses mutually orthogonal projections
\[
F^{(1)},F^{(2)},\dots
\qquad\text{with}\qquad
\tau(F^{(n)})=\tfrac{3}{4^n}1,\qquad \sum_{n\ge 1}F^{(n)}=I,
\]
where \(\tau\) is the center-valued trace. On each block \(F^{(n)}\mathscr R F^{(n)}\), one peels off four symmetries \(R_1^{(n)},\dots,R_4^{(n)}\), together with a decomposition
\[
F^{(n)}=E_1^{(n)}+E_2^{(n)}+E_3^{(n)}
\]
satisfying
\[
\tau(E_1^{(n)})=\tfrac23\tau(F^{(n)}),\qquad
\tau(E_2^{(n)})=\tfrac16\tau(F^{(n)}),
\]
and a unitary \(W^{(n)}\) on the remaining piece, so that
\[
U\,F^{(n)}
=
R_1^{(n)}R_2^{(n)}R_3^{(n)}R_4^{(n)}
\bigl(B_1^{(n)}E_1^{(n)}+B_2^{(n)}E_2^{(n)}+E_3^{(n)}\bigr).
\]
The remainders \(W^{(n)}\) are then reassembled into a single unitary \(W=\sum_n W^{(n)}\), and \(W\) is itself a product of two symmetries, yielding a total of six [2204.00009].

The finite-spectrum four-symmetry theorem is obtained by diagonalizing
\[
U=\sum_{k=1}^n \lambda_k E^{(k)}
\]
in a maximal abelian von Neumann algebra containing \(U\), then showing that each scalar unitary \(\lambda_kE^{(k)}\) is a product of four symmetries in the corresponding corner. Norm-density of \(\mathscr S(\mathscr R)^4\) follows by writing any \(U\in\mathscr U(\mathscr R)\) as \(U=e^{2\pi i H}\) for some self-adjoint \(H\) with spectrum in \([0,1]\), approximating \(H\) in norm by finite-spectrum self-adjoints \(H_n\), and setting \(U_n=e^{2\pi i H_n}\in \mathscr S(\mathscr R)^4\) with \(U_n\to U\) in norm. The obstruction at three factors is linked to the characterization of \(\mathscr S^2\): a unitary \(V\) is a product of two symmetries if and only if \(V\) is unitarily equivalent to its adjoint \(V^*\), so \(\operatorname{sp}(V)\) must be symmetric under complex conjugation about the real axis [2204.00009].

The type \(II_1\) results are situated in an existing classical picture. Halmos–Kakutani showed that in \(\mathcal B(\mathscr H)\), a type \(I_\infty\) factor, every unitary is a product of four symmetries and that three symmetries cannot generate all unitaries. Fillmore later extended the four-symmetry result to all properly infinite and type \(III\) von Neumann algebras. Broise, and later Dowerk–Thom, showed that in a \(II_1\)-factor every unitary is a product of finitely many symmetries; the Dowerk–Thom bound was \(16\) factors. The six-symmetry theorem sharpens this to six in arbitrary \(II_1\) von Neumann algebras, while the finite-spectrum bound drops to four. The resulting summary given in the source is:

- Type \(I_n\), \(I_\infty\), \(II_\infty\), \(III\): four symmetries suffice.
- Type \(II_1\): six in general; four for finite-spectrum unitaries; four are already norm-dense; three never suffice [2204.00009].

Taken together, these results present von Neumann unitaries as objects with simultaneously topological, representation-theoretic, and factorization-theoretic structure. Weak and norm compactness of conjugation orbits detect finiteness and direct-sum structure, while products of symmetries quantify how unitary groups in different von Neumann types are generated and approximated.

Source: https://www.emergentmind.com/topics/von-neumann-unitaries