When nearby operator algebras are the same—and when they are not

← All digests

Family 289 · Operator algebras
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-25

1. The question

Suppose two collections of operators are so close that every unit-sized operator in either collection can be approximated in norm by one in the other. Must the collections describe the same algebra in slightly different coordinates?

The Kadison–Kastler perturbation problem asks this for operator algebras. Family 289 claims a positive answer, with a uniformly small coordinate change, for arbitrary unital von Neumann algebras. Its other two manuscripts show why nearby formulations have different answers: one-sided approximation need not admit a small spatial embedding, and arbitrarily close separable C*-algebras need not be spatially conjugate at all.

These are compatible conclusions. They change either the approximation hypothesis or the kind of closure defining the algebras. The family is therefore a study of the exact boundary between stable and unstable operator-algebraic structure.

2. Background and essential definitions

All algebras here act on a common complex Hilbert space $H$, with the same identity operator. A C*-algebra is closed under adjoints and operator-norm limits. A von Neumann algebra is also closed in the weak operator topology, which tests operators through their matrix coefficients. The latter closure can add many operators.

For an algebra $M$, write $M_1=\{x\in M:\|x\|\leq1\}$. The Kadison–Kastler distance is the Hausdorff distance of unit balls:

$$d(M,N)=\max\left\{ \sup_{x\in M_1}\inf_{y\in N_1}\|x-y\|, \sup_{y\in N_1}\inf_{x\in M_1}\|y-x\| \right\}.$$

Small distance gives approximants in both directions. It does not supply a preferred linear or multiplicative map matching the operators.

A spatial conjugacy is an equality $uMu^*=N$ for a unitary operator $u$ on the given Hilbert space. Requiring $\|u-I\|$ to be small is stronger: it says that the concrete positions of the algebras differ only slightly. An abstract *-isomorphism need not be implemented by such a unitary in the given representation.

For one-sided approximation, M2 uses

$$\gamma(M,N)=\sup_{x\in M_1}\inf_{y\in N}\|x-y\|.$$

Here the approximating $y$ need not be contractive, and there is no reverse condition. If $uMu^*\subseteq N$, then $\gamma(M,N)\leq2\|u-I\|$. The question is whether a small gap forces a converse with a small unitary.

An algebra is norm separable if a countable set is dense in its operator norm. This differs from having separable predual as a von Neumann algebra. M3’s norm-separability assumption is part of its stated counterexample.

3. Development history and previous partial results

Kadison and Kastler introduced the metric in 1972. Early work treated type I algebras, followed by the amenable case through results of Christensen, Johnson and Raeburn–Taylor. Christensen also established a one-sided embedding theorem when the source von Neumann algebra is amenable.

Other positive results required additional structure, such as placing both algebras inside one common finite von Neumann algebra. Merely knowing that each algebra is finite does not supply that common ambient hypothesis. Later work of Cameron and collaborators produced strongly stable nonamenable factors using crossed products and a hyperfinite tensor factor, and studied which structural properties transfer under small perturbations.

The C*-algebra story differs. Christensen, Sinclair, Smith, White and Winter proved spatial conjugacy for sufficiently close norm-separable nuclear algebras on a separable Hilbert space. Nuclearity is an approximation property that makes these algebras particularly tractable.

Earlier negative examples stopped short of M3’s target. Johnson constructed close separable nuclear algebras for which implementing unitaries cannot uniformly approach the identity, but spatial conjugacies still exist. Choi–Christensen constructed close nonisomorphic algebras using norm nonseparability. M3 claims failure of any spatial conjugacy while retaining norm separability.

Similarity theory supplies another part of the background. Bounds on commutators at every matrix level control the passage between representations, but do not themselves construct a multiplicative correspondence between nearby algebras. M1 develops that additional step. Its commutator argument is stated to be independent of the companion similarity manuscript in family 288.

4. What the new manuscripts claim

M1: universal strong stability for von Neumann algebras

For every $\varepsilon>0$, M1 states that there exists $\delta(\varepsilon)>0$ such that

$$d(M,N)<\delta(\varepsilon)$$

implies the existence of a unitary $u\in B(H)$ with

$$uMu^*=N,\qquad \|u-I\|<\varepsilon.$$

The tolerance is independent of the algebras, their types, their representations and the cardinalities of their Hilbert spaces and preduals. The hypothesis uses ordinary operator-norm unit-ball distance, not an assumed distance at every matrix level.

M2: one-sided gaps do not force small embeddings

M2 constructs pairs $M_n,N_n\subseteq B(H_n)$ on separable Hilbert spaces such that $\gamma(M_n,N_n)\to0$, but a fixed $\varepsilon_0>0$ satisfies

$$uM_nu^*\subseteq N_n\quad\Longrightarrow\quad \|u-I\|\geq\varepsilon_0$$

for all sufficiently large $n$. Spatial embeddings do exist in these examples. The failure concerns their unavoidable displacement from the identity, not nonexistence of embeddings. No reverse near inclusion is asserted.

M3: close separable C*-algebras without any spatial conjugacy

For every $\varepsilon>0$, M3 constructs unital norm-separable C*-algebras $A,B$ on a separable Hilbert space with

$$d(A,B)<\varepsilon,\qquad A''=B'',$$

but no unitary $u$ satisfying $uAu^*=B$. The notation $A''$ denotes the weak operator closure, equivalently the double commutant. Thus the two algebras even have the same von Neumann closure.

M3 does not assert that $A$ and $B$ are abstractly nonisomorphic. Failure of spatial conjugacy in a particular representation is its precise conclusion.

5. Why the advance matters

Importance: 5/5. If correct, M1 settles the unrestricted strong perturbation conjecture for von Neumann algebras. The uniform tolerance would make the result independent of representation size and algebra type, rather than a collection of separate stability theorems.

M2 and M3 make the boundary informative. Approximation of every source operator is insufficient if the target has uncontrolled extra structure. Likewise, norm-closed subalgebras may retain positional information that disappears entirely on taking weak closure.

The family therefore distinguishes three questions often compressed into “nearby means the same”: whether an isomorphism exists, whether it is spatial in the given Hilbert space, and whether its implementing unitary is close to the identity. The manuscripts answer different versions with carefully different hypotheses.

6. The main idea

Build the missing multiplicative map

For finite factors, M1 combines matrix-level commutator control with a construction of a near-identity isomorphism. After arranging a common tracial position and a shared hyperfinite subfactor, it chooses nearby unitaries. These choices almost multiply correctly, but need correction.

Boundary averaging first turns them into a measurable cocycle over an amenable action. This does not assume that the underlying group is amenable. An additional ergodicity argument yields a map preserving adjoints and squares; closeness rules out reversing multiplication. Commutator estimates then implement the resulting isomorphism by a small unitary on the original Hilbert space.

Infinite factors require modular theory, which associates a canonical time evolution to a faithful state. M1 compares suitably rescaled graphs of the operators defining that evolution, aligns the modular actions, and passes to crossed products. Finite corners bring the problem back to the finite-factor theorem. Completely positive maps, which preserve positivity at every matrix size, then permit descent to the original algebras. Centers and separable restrictions finish the arbitrary-Hilbert-space case.

Encode almost-commuting matrices

M2 starts with clock and shift matrices that nearly commute but remain a fixed distance from every commuting unitary pair. Their obstruction is topological, expressed through a winding number.

A family of free-group representations inserts those matrices as parameters. Uniform Fourier estimates turn their small commutator into a small one-sided gap for the entire source unit ball. If a near-identity spatial embedding existed, weak operator limits and compression would recover genuinely commuting matrices close to the original pair, contradicting the starting obstruction.

The construction thus transfers a finite-dimensional instability into a problem about infinite-dimensional operator algebras. Existence of some spatial embedding is retained throughout.

Record an inclusion at the end of a sequence

M3 starts with subalgebras $P,Q$ inside a common algebra $D$. It forms an algebra of norm-convergent sequences

$$E_D(P)=\{(x_n):x_n\in D,\ x_n\to p\in P\}.$$

Changing $P$ changes only which limits are allowed. Coordinate projections are intrinsic central projections, so any spatial conjugacy of sequence algebras must permute coordinates and normalize $D$ there.

A tensor-norm identity distinguishes the positions of the chosen close $P,Q$ inside $D$. That identity survives the coordinate normalizations and limiting operations forced by a hypothetical conjugacy. The sequence construction therefore upgrades a relative inclusion obstruction into failure of every spatial conjugacy, while preserving closeness and a common weak closure.

7. Scope, limitations, and open questions

The positive theorem requires two-sided distance and weakly closed algebras with common identity. M2 changes the first requirement; M3 changes the second. Neither is a counterexample to M1’s stated theorem.

M1 asserts a universal tolerance but does not provide an optimal modulus relating algebra distance to unitary displacement. Its independent commutator estimate is only one component; the multiplicative correction and return to the original representation are essential additional claims.

M2 does not rule out arbitrary spatial embeddings. M3 does rule out arbitrary spatial conjugacies, but makes no abstract nonisomorphism claim and does not apply to the nuclear class covered by earlier positive results. Determining sharper tolerances and identifying intermediate structural assumptions for other stability notions remain meaningful questions.

The three source versions also matter: M1 is dated September 23, while the two counterexample manuscripts used here are dated October 5. Their conclusions are reported with the scope of these pinned versions.

8. Sources and further reading

• Christensen and collaborators, Perturbations of nuclear C*-algebras, for the separable nuclear positive theorem.
• Cameron and collaborators, Kadison–Kastler stable factors and Type II1 factors satisfying the spatial isomorphism conjecture, for earlier nonamenable stability results.
• M2 explains its clock-and-shift obstruction; M3 distinguishes Johnson’s displacement examples from its stronger failure of spatial conjugacy.

Source manuscripts

• M1. OpenAI, Universal strong Kadison–Kastler stability (September 23, 2026).
• M2. OpenAI, Near Inclusions of von Neumann Algebras Without Small Spatial Embeddings (October 5, 2026).
• M3. OpenAI, Close Separable C*-Algebras Without Spatial Conjugacy (October 5, 2026).

Sources consulted on 25 October 2026 at repository revision 1581d4c. Download the manuscript citations.