gpt 6 astra pro attempt on Math-211

Hilbert-Smith Conjecture

Definitions and mathematical statement

A faithful continuous action of a locally compact Hausdorff group $G$ on a connected topological manifold $M$ is a jointly continuous map $G\times M\to M$ satisfying the action laws, whose only element fixing every point is the identity. The Hilbert--Smith assertion says every such $G$ is a finite-dimensional Lie group. Its standard obstruction formulation excludes faithful actions of \[ {\mathbb{Z}}_p=\varprojlim_m{\mathbb{Z}}/p^m{\mathbb{Z}} \] by homeomorphisms for every prime $p$. Here $M$ has the usual finite-dimensional manifold conventions, and ${\mathbb{Z}}_p$ is the compact additive group of $p$-adic integers, not the field ${\mathbb{Q}}_p$. No differentiability or Lipschitz regularity of the action is assumed.

Formulation and concept references: [S1].

Short English statement

Must every locally compact group acting continuously and faithfully on a connected manifold be a Lie group, even when its action is only by homeomorphisms?

Research attempt

Current frontier.

Pardon's formulation source explains both the reduction to ${\mathbb{Z}}_p$ and Newman's theorem excluding sufficiently small nontrivial compact Lie actions on a connected manifold; it also proves the two-dimensional case.[S1] Pardon's separate three-manifold theorem handles arbitrary continuous faithful locally compact group actions in dimension three, without smoothness.[E1] A recent different method is Shelukhin's 2024 Theorem A: no nontrivial continuous ${\mathbb{Z}}_p$ action by Hamiltonian homeomorphisms on a closed symplectically aspherical symplectic manifold. Its Floer-theoretic input belongs to that setting and is not available for arbitrary topological homeomorphisms.[E2] The attempt below uses no differentiability, invariant symplectic form, or assumed regularity of an orbit-space metric.

Attack route.

A regularity route would need control of an invariant metric beyond mere topological compatibility. An averaging route would need an injectivity statement not implied by uniform closeness to coordinates. A finite-group route seems attractive because Newman already gives an order-independent obstruction, but ${\mathbb{Z}}_p/p^n{\mathbb{Z}}_p$ acts naturally on an orbit quotient, not on the original manifold. The route pursued here makes that last issue explicit: construct increasingly accurate finite cyclic almost actions on the original manifold and identify the precise repair theorem that would convert them to genuine actions. The resulting conditional argument treats noncompact manifolds as well as compact ones.

Attempt.

1. The local obstruction already available from topology. We use the following established form of Newman's theorem. If $M$ is connected, $d$ is a compatible metric, and $U\subset M$ is a nonempty open set, then there exists $\varepsilon_U>0$ such that a compact Lie group action with \[ d(gx,x)\le\varepsilon_U\quad\text{for all }x\in U\text{ and }g \] is trivial on all of $M$. In particular the constant works for every finite cyclic group, irrespective of its order or whether its action is faithful. Pardon's source states and proves the version needed here.[S1] One cannot insert ${\mathbb{Z}}_p$ into this theorem as though it were a compact Lie group; that would assume the conclusion.

Fix a prime $p$ and suppose for the purpose of constructing an obstruction that $a:{\mathbb{Z}}_p\to\operatorname{Homeo}(M)$ is a faithful continuous action. Write $a_t(x)$ for the action of $t$. Choose a relatively compact nonempty open $U$ and a compact set $K_0$ containing $\overline U$. Joint continuity on compact sets gives \[ \sup_{t\in p^m{\mathbb{Z}}_p}\sup_{x\in K_0}d(a_t x,x)\longrightarrow0 \quad(m\to\infty). \tag{207.1} \] Indeed a neighborhood of $0$ that works uniformly on $K_0$ is obtained from a finite subcover of $\{0\}\times K_0$, and eventually contains $p^m{\mathbb{Z}}_p$. Fix $m$ for which this supremum is less than $\varepsilon_U/3$. Faithfulness still gives $a_{p^m}\ne\operatorname{id}$.

2. Finite cyclic almost actions without changing the phase space. For $n\ge1$ and $0\le j<p^n$, define \[ \rho_n(j)=a_{p^m j},\qquad \rho_n:{\mathbb{Z}}/p^n{\mathbb{Z}}\longrightarrow\operatorname{Homeo}(M). \tag{207.2} \] These are maps of sets with $\rho_n(0)=\operatorname{id}$, not generally homomorphisms. In a product $j+l$, the integer representative either stays in range or differs from its residue by $p^n$. Thus the defect is exactly a carry by $a_{p^{m+n}}$: \[ \rho_n(j)\rho_n(l)= \begin{cases} \rho_n(j+l),&j+l<p^n,\\ \rho_n(j+l-p^n)a_{p^{m+n}},&j+l\ge p^n. \end{cases} \tag{207.3} \] For every compact $K\subset M$ it follows that \[ \delta_n(K):= \max_{j,l\in{\mathbb{Z}}/p^n{\mathbb{Z}}}\sup_{x\in K} d\left(\rho_n(j)\rho_n(l)x,\rho_n(j+l)x\right) \longrightarrow0. \tag{207.4} \] To justify uniformity, consider $d(a_{t+s}x,a_t x)$ for $t\in p^m{\mathbb{Z}}_p$, $x\in K$ and $s=p^{m+n}\to0$. Continuity on this compact parameter set makes the supremum tend to zero. No quantitative modulus or Lipschitz bound is needed.

There is additional compactness that an eventual repair theorem should exploit. All maps in $207.2$, and their inverses, lie in the image of the compact group $p^m{\mathbb{Z}}_p$. Hence they are equicontinuous on compact sets. The union of their images of any compact $K$ is contained in the compact orbit saturation $a(p^m{\mathbb{Z}}_p\times K)$. This controls both the maps and where their images can go on a noncompact manifold.

3. The precise missing repair principle. Here is a sufficient hypothesis, deliberately stated independently of any assumed ${\mathbb{Z}}_p$ action.

Finite cyclic repair hypothesis $\mathrm{FCR}(M,p)$. Let $\rho_n:{\mathbb{Z}}/p^n{\mathbb{Z}}\to\operatorname{Homeo}(M)$ have identity at zero. Assume that their maps and inverses, taken together, are equicontinuous on compact sets; that the union of their images of each compact set has compact closure; and that the defects $207.4$ tend to zero for every compact set. Then there exist homomorphisms $\sigma_n:{\mathbb{Z}}/p^n{\mathbb{Z}}\to\operatorname{Homeo}(M)$ such that, for every compact $K$, \[ \max_{j\in{\mathbb{Z}}/p^n{\mathbb{Z}}}\sup_{x\in K} d\left(\sigma_n(j)x,\rho_n(j)x\right)\longrightarrow0. \tag{207.5} \] The convergence must be uniform in the whole finite cyclic group, whose order tends to infinity. The hypothesis is not asserted to be a known theorem or to have been proved in this notebook.

Lemma 207.1 (repair implies the obstruction formulation). If $\mathrm{FCR}(M,p)$ holds for a connected manifold $M$, there is no faithful continuous ${\mathbb{Z}}_p$ action on $M$.

Proof. Apply the repair hypothesis to $207.2$, whose hypotheses were verified in step 2. By $207.1$, every $\rho_n(j)$ moves every point of $U$ by less than $\varepsilon_U/3$. By $207.5$ on $K_0$, for all sufficiently large $n$, every $\sigma_n(j)$ moves those points by less than $2\varepsilon_U/3$. Each $\sigma_n$ is a genuine finite group action, so Newman's theorem makes it trivial on all of $M$. For any compact $K$, $207.5$ at $j=1$ now gives \[ \sup_{x\in K}d(a_{p^m}x,x) =\sup_{x\in K}d(\rho_n(1)x,\sigma_n(1)x) \longrightarrow0. \] The left side is independent of $n$. It is zero on every compact set, so $a_{p^m}=\operatorname{id}$ globally, contradicting faithfulness. $\square$

Consequently, $\mathrm{FCR}(M,p)$ for every connected finite-dimensional manifold and prime would imply the full target through the standard locally compact-group reduction.[S1] The proof has not restricted $M$ to be compact: the local Newman threshold propagates triviality globally, and $207.5$ is subsequently used on arbitrary compact sets. For manifolds with boundary, under conventions allowing them, a faithful action restricts faithfully to the interior (a homeomorphism fixing the dense interior fixes the boundary); the same argument is applied to that interior. A connected zero-dimensional manifold is a point and its faithful acting group is trivial.

4. Why repairing only one generator would be insufficient. The relation that fails in $207.2$ is $\rho_n(1)^{p^n}=a_{p^{m+n}}$, which approaches the identity on compact sets. Even finding a nearby periodic homeomorphism does not immediately supply $207.5$: one must control all its first $p^n$ powers. For illustration, on a compact phase space with a metric in which the original map $f$ is an isometry, a candidate map $g$ with $\sup_xd(gx,fx)\le\eta$ satisfies only the elementary estimate \[ \sup_xd(g^j x,f^j x)\le j\eta. \] Thus a bound $\eta_n\to0$ alone is inadequate; this estimate would need $p^n\eta_n\to0$. Joint continuity gives no such rate for the carry $a_{p^{m+n}}$. Even a modulus decaying like $1/n$ would be compatible with continuity but defeat that estimate. This is an obstruction to this proposed proof method, not a construction of an actual faithful manifold action with that modulus.

The compact-action invariant metric is easy to construct: \[ d_*(x,y)=\sup_{t\in{\mathbb{Z}}_p}d(a_t x,a_t y) \] for a bounded compatible metric $d$. It is a metric, is invariant under the action, and induces the original topology by compact-parameter uniform continuity. These facts alone do not make it a Riemannian metric or give Lipschitz regularity relative to a fixed smooth metric. An isometry-group theorem with such additional hypotheses therefore cannot be invoked on $d_*$ without proving them.

5. Testing the repair idea in a genuine non-manifold action. Take the phase space to be the Cantor group ${\mathbb{Z}}_p$ itself with its $p$-adic metric, acting on itself by addition. For $x\in{\mathbb{Z}}_p$ write $x=b+p^n y$, $0\le b<p^n$. The translations $\rho_n(j)x=x+j$, with $j$ represented in $[0,p^n)$, have defect at most $p^{-n}$. Define an exact action instead by \[ \sigma_n(j)(b+p^n y) =\left((b+j)\bmod p^n\right)+p^n y. \tag{207.6} \] This is a homeomorphism on each clopen residue cylinder and a permutation of those cylinders. Addition modulo $p^n$ proves the group law. Also \[ \sup_{j,x}|\sigma_n(j)x-(x+j)|_p\le p^{-n}. \] Hence the desired kind of correction can really occur; it is not merely a formal request. Yet the original faithful ${\mathbb{Z}}_p$ action certainly exists here. The distinction is exactly that Newman's manifold theorem fails on a Cantor space. Truncated translations by multiples of $p^m$ give nontrivial finite actions all of whose displacements are at most $p^{-m}$. The conditional proof therefore uses the manifold hypothesis at an identifiable step and does not incorrectly rule out known profinite actions on arbitrary compact spaces.

6. A second route stops at injectivity, not at averaging. On a coordinate region whose relevant small-subgroup orbits remain in the chart, average its coordinate functions over $p^m{\mathbb{Z}}_p$ using Haar probability. Compactness gives continuous averaged coordinates that are invariant under that subgroup and uniformly close to the original coordinates as $m\to\infty$. To turn this into an obstruction one would need strong information about their fibers, and ultimately about triviality of the action beyond that region.

Uniform closeness does not provide injectivity. Already the real maps \[ q_\varepsilon(x)=x+\varepsilon\sin(x/\varepsilon^2) \] are within $\varepsilon$ of the identity but have derivative $1+\varepsilon^{-1}\cos(x/\varepsilon^2)$ of both signs on any fixed interval when $\varepsilon$ is small. They are not injective there. This is not claimed to be an average of a hypothetical group action; it refutes the proposed inference from closeness alone. Degree-one or homotopy-to-identity information also does not imply injectivity. Additional orbit-fiber topology would still have to be proved, so this alternative does not fill the repair gap.

Stress test.

The finite quotients in step 2 were never treated as genuine actions. The carry formula records their exact failure. The repair hypothesis includes compact-image control as well as equicontinuity, and uniformity in all group elements rather than only a generator. Newman's constant is used only for the repaired finite actions; applying it to the original $p$-adic group would be circular.

Arzelà--Ascoli compactness does not establish $207.5$. A convergent subsequence of some maps can recover a compact group action, which is precisely the structure one started with, without producing finite-order homeomorphisms satisfying all the required relations. Neither a near-identity averaging map nor an invariant compatible metric supplies that missing algebraic correction. Finally, the scope of the symplectic theorem has not been enlarged to arbitrary homeomorphisms.[E2]

\Needspace{10\baselineskip}

Outcome.

Outcome: Conditional reduction.

The attempt gives an explicit family of equicontinuous finite cyclic almost actions arising from any hypothetical faithful ${\mathbb{Z}}_p$ action, computes its defect, and proves that the stated local-uniform repair principle would settle the obstruction formulation on all connected manifolds, including noncompact ones. The Cantor model provides an exact test of the construction and identifies where manifold topology is indispensable.

No finite cyclic repair theorem with the required variable-order uniformity has been established. Proving it would itself imply the full Hilbert--Smith conjecture, so the repair step is a major unresolved hypothesis, not a consequence of ordinary compactness or continuity. The invariant-metric and averaging routes were also investigated but stop at explicit regularity and injectivity gaps. No counterexample to the original manifold assertion was found.

\Needspace{8\baselineskip}

Sources

• S1 John Pardon, "Totally disconnected groups (not) acting on two-manifolds," manuscript of 6 July 2018 (surveying the Hilbert-Smith Conjecture; published in proceedings on Breadth in Contemporary Topology). https://www.math.stonybrook.edu/~jpardon/manuscripts/13_hs2d.pdf. Catalogue formulation source.
• E1 J. Pardon, The Hilbert--Smith conjecture for three-manifolds, Journal of the American Mathematical Society 26 $2013$, 879--899, DOI: 10.1090/S0894-0347-2013-00766-3; main theorem and local reduction. https://arxiv.org/abs/1112.2324.
• E2 E. Shelukhin, A symplectic Hilbert--Smith conjecture, arXiv:2403.07195v2 $2024$, Theorem A and Section 2, with appendix by L. Polterovich. The theorem used here assumes a closed symplectically aspherical manifold and Hamiltonian homeomorphisms. https://arxiv.org/abs/2403.07195.