A genuine A-B slice solution for the generalized Borromean rings: combining Freedman–Krushkal's link-homotopy+ stabilization with Whitehead doubling

The A-B slice problem is Freedman's reformulation of the four-dimensional topological surgery conjecture for free groups — the critical case for arbitrary fundamental groups — as a question about standard decompositions of the 4-ball. Freedman and Krushkal (2017) proved that it admits a link-homotopy+ solution for a coinitial family of generalized Borromean rings (GBRs): an explicit stabilization and band-sums whose result is homotopically trivial+ (h-trivial+), i.e. all Milnor $\bar{\mu}$-invariants with at most two repeating indices vanish. Because untwisted Whitehead doubles of h-trivial+ links are topologically slice, the strongest possible A-B slice problem is solvable for h-trivial+ links, and the paper asks whether the two stabilizations can be combined into a genuine (embedded, standard) A-B slice solution for GBRs — equivalent to four-dimensional surgery for free groups and hence for all groups. The problem remains open.

Background

Four-dimensional topological surgery. The topological surgery conjecture in dimension four states that a degree-one normal map $(M,\partial M)\to(X,\partial X)$ — a map from a topological 4-manifold to a 4-dimensional Poincaré pair (a space satisfying Poincaré duality) that carries the fundamental class to the fundamental class and is covered by a map of stable normal bundles — which is a $\mathbb{Z}[\pi_1 X]$-homology equivalence over the boundary, is normally bordant to a homotopy equivalence whenever the Wall surgery obstruction (the algebraic $L$-theory obstruction, which in dimensions at least five is the only one) vanishes; it is the engine of the topological classification of 4-manifolds. It was established in the simply connected case and is known for a class of so-called good fundamental groups (elementary amenable groups and groups of subexponential growth, among others), but for arbitrary groups it remains open. Freedman's geometric reformulation (1986) reduces the general conjecture to a family of universal model problems whose fundamental groups are free; free groups are therefore the critical case.

The A-B slice problem. Freedman's reformulation (1986; elaborated by Freedman–Lin 1989) recasts the free-group surgery problem as a question about decompositions of the 4-ball. A decomposition is a pair of compact codimension-zero submanifolds $A,B\subset D^4$ with $A\cup B=D^4$, $A\cap B=\partial^-A=\partial^-B$, and $\partial^+A\cup\partial^+B=S^3$ the standard genus-one Heegaard decomposition of $S^3=\partial D^4$; the attaching curves (cores of the two solid tori) form a Hopf link. A link $L=(l_1,\dots,l_k)\subset S^3$ with untwisted parallel copy $L'$ is A-B slice if there exist decompositions $(A_i,B_i)$ of $D^4$ and self-homeomorphisms $\varphi_i,\psi_i$ of $D^4$ such that all the images $\varphi_1(A_1),\dots,\varphi_k(A_k),\psi_1(B_1),\dots,\psi_k(B_k)$ are disjoint, $\varphi_i(\partial^+A_i)$ is a tubular neighborhood of $l_i$, and $\psi_i(\partial^+B_i)$ a tubular neighborhood of $l_i'$. The requirement that the embeddings be restrictions of self-homeomorphisms of $D^4$ (standardness: the complement of each $\varphi_i(A_i)$ is homeomorphic to $B_i$) is essential — without it, every link with vanishing pairwise linking numbers is weakly A-B slice (Krushkal 2008), so no link-homotopy invariant can obstruct the weak version.

Generalized Borromean rings (GBRs). Surgery for free groups is equivalent to the A-B slice problem for the GBRs: the links obtained from the Hopf link by any nontrivial amount of iterated ramified Bing doubling, where Bing doubling replaces a link component by a specific pair of components in its tubular neighborhood and the ramified variant places the pair on the boundary of that neighborhood; the Borromean rings arise as the first ramified Bing double of the Hopf link. GBRs carry a partial order in which more ramification and more Bing doubling counts as smaller, and any coinitial subset (one containing, below every GBR, some member of the subset) is universal for surgery: solving the A-B slice problem for such a subset solves surgery for free groups, hence for all groups.

Stabilization and the relative slice problem. The handle structure of a decomposition, read as a Kirby diagram (a convention encoding the handles of a 4-manifold as a framed link in its boundary), produces a stabilization of $L$: a pair of links $(J,K)$ in $S^3=\partial D^4_0$, where $D^4_0$ is a smaller 4-ball obtained by removing collars on all attaching regions, $J$ collects the attaching curves of the 2-handles of both sides, and $K$ the zero-framed curves corresponding to the 1-handles; duality between the 1- and 2-handles of the two sides forces the components of $L$ (respectively $L'$) to be paired with links in solid-torus neighborhoods of meridians (respectively of parallels), the two tori exchanging meridian and longitude. The pair $(J,K)$ is relatively slice if the components of $J$ bound disjoint smoothly embedded disks in the handlebody $H_K=D^4_0\cup$(zero-framed 2-handles attached along $K$). A link is A-B slice precisely when an associated stabilization, after band-sums (fusing or augmenting components by connected sum along embedded bands in $S^3$), is relatively slice with the standardness condition — this reduces the A-B slice problem to a link-slicing question.

Milnor invariants and link homotopy. Two links are link-homotopic if connected by a 1-parameter family of maps in which distinct components stay disjoint (self-intersections of a component are allowed). Milnor's $\bar{\mu}$-invariants (Milnor 1954) are link-homotopy invariants: coefficients of the Magnus expansion (the homomorphism sending each meridian $g\mapsto 1+x$ into non-commuting power series) of longitudes expressed in meridians; the multi-index of a coefficient may repeat components. The relevant quotient is the Milnor group: $\pi_1(S^3\setminus L)$ modulo the normal subgroup generated by commutators $[g_i,g_i^y]$ of conjugates of the same meridian. A link is homotopically trivial (h-trivial) iff its Milnor group is the free Milnor group, iff all non-repeating $\bar{\mu}$-invariants vanish. A link is h-trivial+ if adding a parallel copy of any single component yields an h-trivial link; equivalently, all $\bar{\mu}$-invariants with at most two repeating indices vanish.

Engel relations. The 2-Engel relation is the group identity $[[y,x],x]=1$; a 2-Engel group satisfies it identically. The key algebraic fact (classical work of Burnside, Hopkins and Levi, stated as Lemma 3.1 of Freedman–Krushkal (2017)) is that every 2-Engel group is nilpotent of class at most 3 — a bound independent of the number of generators, unlike the free Milnor group on $n$ generators, which is nilpotent of class $n$. Freedman–Krushkal (2016) discovered the geometric power of this relation in 4-manifold topology, realizing commutator structure in links so that handle slides force h-triviality. Freedman–Krushkal (2017) refines it to a '(2,2)-Engel' relation — commutators $[[g_i,g_i^{y_1}]^{z_1},[g_j,g_j^{y_2}]^{z_2}]$ trivial modulo the 2-Engel relation in two independent ways — from which h-triviality+ follows geometrically.

State of the art before the question. Freedman–Krushkal (2016) produced a link-homotopy solution: for each GBR in a coinitial family, a stabilization and band-sums giving an h-trivial link (non-repeating $\bar{\mu}$-invariants vanish). Freedman–Krushkal (2017) strengthens this to a link-homotopy+ solution (its Theorem 1): for each GBR in a more Bing-doubled, still coinitial family, a stabilization and band-sums give an h-trivial+ link — all $\bar{\mu}$-invariants with at most two repeating indices vanish. A $+k$ extension for fixed $k$ exists but requires input links of growing complexity (simplest representative with $2^{2k}+1$ components). Independently, Freedman–Teichner (1995) proved that untwisted Whitehead doubles of h-trivial+ links are topologically slice (the Whitehead double of a link replaces each component by the clasp-shaped curve tracing the Whitehead link pattern on the boundary of its tubular neighborhood, and is untwisted when given zero framing), so for an h-trivial+ link a stabilization exists that makes the relative-slice problem genuinely solvable by embedded disks — 'the strongest possible version of the A-B slice problem' is solvable for such links. Theorem 1 manufactures h-trivial+ links from GBRs by stabilization; the question is whether the two stabilizations compose.

Problem Statement

The problem posed by Freedman–Krushkal (2017):

Primary question. Can the link-homotopy+ stabilization of Theorem 1 — which, for each generalized Borromean ring in an explicitly described coinitial family (universal for surgery; the construction is presented for a concrete simplest representative — the GBR obtained from the Hopf link by keeping one component intact, Bing-doubling the other twice, and then Bing-doubling the four resulting components twice more — and, in the paper's words, the argument 'applies to GBRs that are more Bing-doubled and ramified', a still-coinitial family), produces a stabilization $K_0$ and band-sums whose stabilized link is h-trivial+, i.e. all Milnor $\bar{\mu}$-invariants with at most two repeating indices vanish — be combined with the untwisted-Whitehead-doubling stabilization, or with any other further geometric operations, to give a genuine A-B slice solution for GBRs? Concretely: do there exist, for each GBR in some coinitial family, decompositions $D^4=A_i\cup B_i$ with standard disjoint embeddings (Definition 2.1 of the source, with the standardness condition, Condition 2.4 in the homotopy setting) such that the associated stabilization pair $(J,K)$ is relatively slice — the components of $J$ bound disjoint smoothly embedded disks in the handlebody $H_K$, not merely maps of disks? An affirmative answer for any coinitial family is equivalent to an affirmative resolution of the four-dimensional topological surgery conjecture for free groups, and hence for all fundamental groups.

The source explicitly poses the following component questions — the identified routes to, and intermediate open fragments of, the primary question:

1. (All Milnor invariants.) Does there exist a stabilization of GBRs giving a link with all $\bar{\mu}$-invariants vanishing — for multi-indices with arbitrarily many repeating indices, not merely at most two? If so, can it be obtained without the Bing-doubling growth that the $+k$ extension requires (simplest representative link with $2^{2k}+1$ components)?

2. (Higher Engel control.) Can the geometric 2-Engel and (2,2)-Engel machinery be generalized or replaced by algebraic and geometric control of $n$-Engel relations [[y,\underbrace{x],x],\dots,x]=1 for $n>2$, to obtain stronger vanishing patterns more efficiently — in particular, is the nilpotency class of the free Milnor group modulo the $n$-Engel relation bounded independently of the number of generators for fixed $n\geq 3$ (Question A.1 of the companion paper)?

3. (Obstruction or construction.) The two-sided top-level form: determine whether there remains any obstruction to A-B sliceness of GBRs in terms of nilpotent invariants of links (Milnor's $\bar{\mu}$-invariants, including repeating ones). An obstruction would refute four-dimensional topological surgery — and the s-cobordism conjecture (the assertion that a 5-dimensional cobordism between closed 4-manifolds which is a homotopy product relative to one boundary is homeomorphic to a product) — for free groups; a sufficiently improved homotopy solution would resolve surgery affirmatively for all groups.

Scope of the question. The primary question is two-sided: an affirmative resolution would consist of an explicit genuine A-B slice construction for a coinitial GBR family, while a negative resolution would be a proof that no such solution exists — an obstruction valid for all stabilizations. Items 1–3 are separately posed questions in the source. The target is the standard A-B slice problem, in which the standardness condition is required, rather than the weak A-B slice problem (standardness dropped) or the model decompositions of Freedman–Lin.

References

1. Michael Freedman and Vyacheslav Krushkal, A homotopy+ solution to the A-B slice problem, Journal of Knot Theory and Its Ramifications 26 (2017), no. 2, article 1740018, DOI 10.1142/S0218216517400181, https://doi.org/10.1142/S0218216517400181 (arXiv:1601.05066, https://arxiv.org/abs/1601.05066)
2. Michael Freedman and Vyacheslav Krushkal, Engel relations in 4-manifold topology, Forum of Mathematics Sigma 4 (2016), e22, DOI 10.1017/fms.2016.20, https://doi.org/10.1017/fms.2016.20 (arXiv:1412.5024, https://arxiv.org/abs/1412.5024)
3. Michael Freedman and Vyacheslav Krushkal, Engel groups and universal surgery models, Journal of Topology 13 (2020), 1302–1316, DOI 10.1112/topo.12155, https://doi.org/10.1112/topo.12155 (arXiv:1707.07800, https://arxiv.org/abs/1707.07800)
4. Michael Freedman and Vyacheslav Krushkal, Universal surgery problems with trivial Lagrangian, Mathematical Research Letters 26 (2019), no. 6, 1587–1601, DOI 10.4310/MRL.2019.v26.n6.a2, https://doi.org/10.4310/MRL.2019.v26.n6.a2 (arXiv:1901.05951, https://arxiv.org/abs/1901.05951)
5. Jae Choon Cha, Min Hoon Kim, and Mark Powell, A family of freely slice good boundary links, Mathematische Annalen 376 (2020), 1009–1030, DOI 10.1007/s00208-019-01907-3, https://doi.org/10.1007/s00208-019-01907-3 (arXiv:1807.05474, https://arxiv.org/abs/1807.05474)
6. Min Hoon Kim, Mark Powell, and Peter Teichner, Round handle problem, Pure and Applied Mathematics Quarterly 17 (2021), no. 1, 237–247, DOI 10.4310/pamq.2021.v17.n1.a6, https://doi.org/10.4310/pamq.2021.v17.n1.a6 (arXiv:1706.09571, https://arxiv.org/abs/1706.09571; the arXiv version is titled 'The round handle problem')
7. James Conant, Rob Schneiderman, and Peter Teichner, Clasper concordance, Whitney towers and repeating Milnor invariants, Journal of Knot Theory and Its Ramifications 34 (2025), no. 4, article 2550013, 35 pp., DOI 10.1142/S0218216525500130, https://doi.org/10.1142/S0218216525500130 (arXiv:2005.05381, https://arxiv.org/abs/2005.05381)
8. Michael Freedman, A geometric reformulation of 4-dimensional surgery, Topology and its Applications 24 (1986), 133–141, DOI 10.1016/0166-8641(86)90054-4, https://doi.org/10.1016/0166-8641(86)90054-4
9. Michael Freedman, Are the Borromean rings A-B-slice?, Topology and its Applications 24 (1986), 143–145, DOI 10.1016/0166-8641(86)90055-6, https://doi.org/10.1016/0166-8641(86)90055-6
10. Michael Freedman and Xiao-Song Lin, On the (A,B)-slice problem, Topology 28 (1989), 91–110, DOI 10.1016/0040-9383(89)90034-7, https://doi.org/10.1016/0040-9383(89)90034-7
11. Michael Freedman and Peter Teichner, 4-Manifold topology II: Dwyer's filtration and surgery kernels, Inventiones Mathematicae 122 (1995), 531–557, DOI 10.1007/BF01231455, https://doi.org/10.1007/BF01231455
12. Vyacheslav Krushkal, A counterexample to the strong version of Freedman's conjecture, Annals of Mathematics (2) 168 (2008), 675–693, DOI 10.4007/annals.2008.168.675, https://doi.org/10.4007/annals.2008.168.675 (arXiv:math/0610865, https://arxiv.org/abs/math/0610865)
13. John Milnor, Link groups, Annals of Mathematics (2) 59 (1954), 177–195, DOI 10.2307/1969685, https://doi.org/10.2307/1969685
14. Michael Freedman, Vyacheslav Krushkal, Christopher J. Leininger, and Alan W. Reid, Filling links and spines in 3-manifolds, Communications in Analysis and Geometry 31 (2023), no. 10, 2307–2333, DOI 10.4310/CAG.2023.v31.n10.a1, https://doi.org/10.4310/CAG.2023.v31.n10.a1 (arXiv:2010.15644, https://arxiv.org/abs/2010.15644)