UCT Problem for Separable Nuclear Complex C*-Algebras

UCT Problem for Separable Nuclear Complex C*-Algebras

Definitions and mathematical statement

For separable complex $C^*$-algebras $A,B$, let $K_i$ be operator $K$-theory and $KK_j(A,B)$ Kasparov's bivariant groups, graded modulo two. Nuclearity of $A$ means its identity admits point-norm approximations by completely positive contractions factoring through finite-dimensional matrix algebras. Define \[ H_j=\bigoplus_{i=0,1}\operatorname{Hom}_{{\mathbb{Z}}}(K_i(A),K_{i+j}(B)),\quad E_j=\bigoplus_{i=0,1}\operatorname{Ext}^1_{{\mathbb{Z}}}(K_i(A),K_{i+j+1}(B)). \] The problem asks whether every separable nuclear $A$ has the canonical exact sequence $0\to E_j\to KK_j(A,B)\to H_j\to0$ for every separable $B$ and each $j$. The maps are the Rosenberg--Schochet maps, including the canonical kernel identification; an abstract short exact sequence with arbitrarily chosen maps is insufficient. Neither unitality nor a natural splitting is required.

Formulation and concept references: [S1].

Short English statement

Does operator K-theory, together with its extension term, compute KK-theory through the canonical universal coefficient sequence for every separable nuclear complex C*-algebra?

Research attempt

Current frontier.

The unrestricted UCT is still listed as Problem II in the May 2026 revision of Schafhauser--Tikuisis--White; neither nuclearity nor the classification results quoted there remove the UCT hypothesis.[E1] Barlak--Li prove the UCT for separable nuclear algebras with a Cartan subalgebra, and more generally for the nuclear twisted groupoid algebras in their Theorem 1.1.[S1] Kirchberg's reduction permits concentrating on unital Kirchberg algebras with vanishing $K$-theory. Here "Kirchberg" includes separability, nuclearity, simplicity, and pure infiniteness.[E2] Controlled $KK$-theory supplies an inverse-limit description with a potentially nonzero $\lim^1$ term; simply discarding that term replaces $KK$ by a quotient and loses information.[E3]

Attack route.

Three routes suggest different missing inputs. A Cartan construction would need a regular maximal abelian subalgebra with faithful conditional expectation, not merely a large commutative subalgebra.[S1] A finite-complexity decomposition would need compatible pieces and overlap control of the kind used by Willett--Yu, rather than arbitrary nuclear factorizations.[E4] The route pursued below instead asks whether a nuclear approximation can be promoted to a genuine $KK$-factorization whose error is nilpotent. The algebraic correction is exact; the substantial question is whether the required classes and nilpotence can be constructed. Two explicit tests then examine the two most tempting shortcuts: converting completely positive maps into $KK$-classes, and identifying classes from all their finite-stage restrictions.

Attempt.

1. A finite algebraic correction, with all coefficient algebras retained. Write $\mathcal B$ for the bootstrap category in $KK$. We use its standard closure under $KK$-retracts: a localizing triangulated category with countable coproducts is closed under retracts, by the mapping telescope of an idempotent. For separable nuclear algebras, membership in $\mathcal B$ is equivalent to the canonical UCT in the question, for all separable coefficient algebras.[S1][E1]

Lemma 203.1 (nilpotent-error domination). Suppose $A$ is separable nuclear, $D\in\mathcal B$, and there are even classes \[ x\in KK_0(A,D),\qquad y\in KK_0(D,A),\qquad e=1_A-x\otimes_Dy\in KK_0(A,A) \] such that $e^N=0$ for some finite $N\ge1$. Then $A$ satisfies the exact canonical UCT stated above.

Proof. All powers and products in $KK_0(A,A)$ are Kasparov products. Set \[ q=1_A+e+\cdots+e^{N-1},\qquad y'=y\otimes_Aq. \] Associativity and the finite geometric-series identity give \[ x\otimes_Dy'=(1_A-e)q=1_A-e^N=1_A. \] Thus $A$ is a $KK$-retract of $D$, hence belongs to $\mathcal B$. Applying the bootstrap UCT gives precisely its natural map on $K$-theory and its canonical kernel identification, simultaneously for $j=0,1$ and every separable $B$. No splitting is chosen or claimed to be natural. This argument uses neither unitality of $A,B$ nor nuclearity of $B$. $\square$

The useful feature is that an exact factorization of $1_A$ is unnecessary at the construction stage; a nilpotent error would suffice. The lemma is elementary categorical algebra, not a claim of a new permanence theorem. In particular, "small in norm" is not a substitute for nilpotence in this generally non-normed group.

2. Nuclear approximations do not yet supply those classes. There is a particularly transparent exact completely positive retraction. Define unital completely positive maps \[ \phi:{\mathbb{C}}^2\longrightarrow M_3({\mathbb{C}}),\quad \phi(a,b)=\operatorname{diag}\left(a,b,(a+b)/2\right), \qquad \psi(T)=(T_{11},T_{22}). \] The diagonal coordinates of $\phi$ are positive functionals, and the coordinates of $\psi$ are vector states, so complete positivity follows directly. Moreover $\psi\phi=\operatorname{id}_{{\mathbb{C}}^2}$. Nevertheless, for the projection $p=(1,0)$, \[ \phi(p)-\phi(p)^2=\operatorname{diag}(0,0,1/4),\qquad \|\phi(p)-\phi(p)^2\|=1/4. \] Even an exact completely positive retraction therefore does not force its outward leg to be multiplicative. Repeating this pair at every stage also refutes the assertion that improving the composite alone forces asymptotic multiplicativity of each leg.

A second test keeps the topology visible. For increasingly fine finite sets $\{z_{N,j}\}_{j=1}^{r_N}\subset S^1$, choose subordinate nonnegative partitions of unity $\{h_{N,j}\}$. Put \[ \phi_N(f)=(f(z_{N,j}))_j,\qquad \psi_N((a_j))=\sum_j a_jh_{N,j}. \] These maps factor through ${\mathbb{C}}^{r_N}$, and uniform continuity gives $\psi_N\phi_N(f)\to f$ in norm for every $f\in C(S^1)$. Suppose, incorrectly, that one could regard both legs as even $KK$-classes and obtain a product inducing the identity on $K$-theory. The resulting map on $K_1(C(S^1))={\mathbb{Z}}$ would factor through $K_1({\mathbb{C}}^{r_N})=0$, and hence be zero. This is a contradiction. The example is not a counterexample to the UCT: $C(S^1)$ is already commutative. It identifies exactly why completely positive approximation alone cannot prove Lemma 203.1's hypothesis.

3. A coherent-limit obstruction that survives every finite test. Let \[ A_n=C(S^1),\qquad \alpha_n(f)(z)=f(z^2),\qquad A=\varinjlim(A_n,\alpha_n). \] All connecting maps are injective. The limit is commutative and is in the bootstrap category. The standard Milnor sequence, in the form recalled in Willett--Yu, applies here; it is not being assumed for arbitrary completely positive approximations.[E3] For $B={\mathbb{C}}$ it gives \[ 0\longrightarrow\varprojlim{}^{1} KK_1(A_n,{\mathbb{C}}) \longrightarrow KK_0(A,{\mathbb{C}}) \longrightarrow\varprojlim KK_0(A_n,{\mathbb{C}})\longrightarrow0. \tag{203.1} \] The finite-stage groups are $KK_0(A_n,{\mathbb{C}})={\mathbb{Z}}$ and $KK_1(A_n,{\mathbb{C}})={\mathbb{Z}}$. Pullback by $\alpha_n$ is the identity in degree zero and multiplication by two in degree one. These assertions follow from the degree of $z\mapsto z^2$ and the UCT for the circle, not from the UCT being investigated for a general $A$.

Here the obstruction can be computed without an abstract vanishing claim. For the inverse system ${\mathbb{Z}}\xleftarrow{2}{\mathbb{Z}}\xleftarrow{2}\cdots$, index stages by $n\ge0$ and use the usual concrete definition \[ \varprojlim{}^{1}=\operatorname{coker}\Delta,\qquad \Delta:\prod_{n\ge0}{\mathbb{Z}}\longrightarrow\prod_{n\ge0}{\mathbb{Z}}, \quad\Delta(b)_n=b_n-2b_{n+1}. \] Let ${\mathbb{Z}}_2$ denote the $2$-adic integers. The convergent $2$-adic series \[ \Phi(a)=\sum_{n\ge0}2^n a_n\pmod{{\mathbb{Z}}} \] induces an isomorphism \[ \operatorname{coker}\Delta\ \cong\ {\mathbb{Z}}_2/{\mathbb{Z}}. \tag{203.2} \] Indeed, telescoping gives $\Phi(\Delta b)=b_0\pmod{\mathbb{Z}}=0$; binary digits show surjectivity. If the series equals an ordinary integer $b_0$, define \[ b_n=2^{-n}\left(b_0-\sum_{i<n}2^ia_i\right). \] Its numerator is divisible by $2^n$ in ${\mathbb{Z}}$, so each $b_n$ is an integer, and $b_n-2b_{n+1}=a_n$. This proves the kernel assertion.

The sequence $a=(1,0,1,0,\ldots)$ maps to $1+4+16+\cdots=-1/3\in{\mathbb{Z}}_2$, which is not an ordinary integer. It therefore represents a nonzero class in the left term of $203.1$. Its image in $KK_0(A,{\mathbb{C}})$ restricts to zero on every $A_n$. Consequently agreement on every finite stage does not imply equality in $KK$. This example demonstrates a genuine nonzero phantom class, not a numerical failure to detect a class.

For comparison, if an inverse sequence has surjective bonding maps, $\Delta$ is surjective: choose $b_0$, then solve recursively $f_n(b_{n+1})=b_n-a_n$. This gives a valid mechanism for eliminating a $\lim^1$ obstruction. In the displayed example multiplication by two is not surjective and its images do not stabilize. A proposed construction must supply such compatibility when it uses it; surjectivity cannot be inferred from nuclearity. Nor should all extension information be eliminated: here the nonzero term is precisely compatible with the UCT extension contribution from $K_1(A)={\mathbb{Z}}[1/2]$.

4. Testing the nilpotence step against the minimal reduction. Consider the remaining Kirchberg class from the established reduction: $A$ unital with $K_0(A)=K_1(A)=0$.[E2] The identity $1_A\in KK_0(A,A)$ is then invisible to $K$-theory. Suppose one proposed the general rule that every such invisible endomorphism is nilpotent. Applied to $1_A$, this says \[ 0=(1_A)^N=1_A. \] This would imply $KK_j(A,B)=0$ for every $B,j$, since any class is its product with $1_A$. Both $E_j$ and $H_j$ vanish, so the full canonical UCT follows for this class, and the reduction then settles the original problem. Thus this apparently technical "nilpotence of the invisible error" is already as deep as the unrestricted UCT in the critical case. It is not a consequence of $K_*(A)=0$ that may be used for free.

Stress test.

The finite geometric-series argument has no convergence issue and retains both parities and all separable coefficient algebras. The first failure occurs earlier: no $KK$-factorization of the stipulated kind has been obtained from a nuclear approximation. The exact ${\mathbb{C}}^2$ example rules out one possible justification, while the circle example rules out replacing its intermediate algebra by a finite-dimensional $KK$-model.

The solenoid calculation rules out a separate inference from finite-stage agreement to global equality. It does not contradict controlled $KK$-theory; it illustrates why its derived-limit term is needed. Finally, invoking Kirchberg--Phillips classification with the UCT hypothesis to identify an arbitrary $K$-trivial Kirchberg algebra with $\mathcal O_2$ would be circular. No such identification was used.

\Needspace{10\baselineskip}

Outcome.

Outcome: Conditional reduction.

A genuine bootstrap factorization with nilpotent error is sufficient for the exact UCT, and the correction is explicitly given. The notebook also provides two completely positive countertests and computes a nonzero finite-stage-invisible $KK$-class. These are diagnostic deductions, not a claim that a previously untreated nuclear algebra has entered the bootstrap category.

What remains is a construction of $D,x,y$ with the required nilpotence, or another argument carrying the missing coherent extension information. The $K$-trivial Kirchberg test shows that a blanket nilpotence assertion would already resolve the full problem. The conditional criterion is therefore not presented as an established solution or as an independently verified new reduction in the literature.

\Needspace{8\baselineskip}

Sources

• S1 Selcuk Barlak and Xin Li, Cartan Subalgebras and the UCT Problem, arXiv1511.02697v3,21June2017,18pages. https://arxiv.org/pdf/1511.02697v3. Catalogue formulation source.
• E1 C. Schafhauser, A. Tikuisis and S. White, Nuclear $C^*$-algebras: 99 problems, arXiv:2506.10902v2, 8 May 2026 (manuscript dated 11 May 2026), Section 3, especially Problem II and the discussion of the bootstrap category. https://arxiv.org/abs/2506.10902v2.
• E2 N. P. Brown, S. L. Browne, R. Willett and J. Wu, The UCT problem for nuclear $C^*$-algebras, arXiv:2005.03184v3 $2021$, Theorems 2.1--2.2 and 4.6; Rocky Mountain Journal of Mathematics 52 $2022$, beginning at p. 817. https://arxiv.org/abs/2005.03184.
• E3 R. Willett and G. Yu, Controlled $KK$-theory, and a Milnor exact sequence, arXiv:2011.10906, Theorems 1.1--1.2 and Section 1.2 (on Schochet's Milnor sequence and the $\lim^1$ term). https://arxiv.org/abs/2011.10906.
• E4 R. Willett and G. Yu, The Universal Coefficient Theorem for $C^*$-Algebras with Finite Complexity, Memoirs of the European Mathematical Society 8 $2024$, 108 pp.; introduction and main UCT theorem. DOI: 10.4171/MEMS/8. https://ems.press/books/mems/280.