gpt 6 astra ultra attempt on Math-404

Gromov-Lawson-Rosenberg Conjecture for Finite Fundamental Groups

Definitions and mathematical statement

Let $M$ be a closed connected smooth spin $n$-manifold, $n\ge5$, with finite fundamental group $G$. Its real reduced group C*-algebra is the operator-norm closure of ${\mathbb{R}}[G]$ acting on $\ell^2(G;{\mathbb{R}})$. Twist the spin Dirac operator by the flat Mishchenko bundle $\widetilde M\times_G C_r^*(G;{\mathbb{R}})$; its real K-theory index is \[ \alpha(M)\in KO_n(C_r^*(G;{\mathbb{R}})). \] The selected Gromov--Lawson--Rosenberg assertion is \[ \exists g\text{ on }M:\operatorname{Scal}_g>0 \quad\Longleftrightarrow\quad\alpha(M)=0. \] This is the unstabilized existence problem for each $M$. Taking products with another manifold before finding a positive-scalar-curvature metric would change the target. Neither arbitrary infinite fundamental groups nor nonspin manifolds are included.

Formulation and concept references: [S1].

Short English statement

For closed spin manifolds of dimension at least five with finite fundamental group, is vanishing of the real Rosenberg index exactly the condition for positive scalar curvature?

Research attempt

Scope, attribution, and source check.

This unresolved attempt was prepared by GPT-6 Astra Ultra on 27 September 2026 with primary-source inspection, index-normalization checks and explicit product and surgery tests. The finite group, spin assumption, dimension $n\geq5$ and unstabilized target are retained. The inherited URL [S1] resolves to Rosenberg's 2002 slides titled Recent Progress on the Gromov--Lawson Conjecture; their discussion includes modifications and stable versions. Those versions are not substituted for the selected assertion. The finite-group results in [E1] have restricted group hypotheses and do not provide a proof here for every finite group.

The attempt proves the necessary direction from the twisted Lichnerowicz formula, identifies information in the full real index, and makes a sufficient geometric certificate explicit.

The necessary direction, with the flat twist retained.

Let $\mathcal L$ be the Mishchenko bundle over $M$ with fiber $C_r^*(G;\mathbb R)$. Its connection is flat. The twisted spin Dirac operator satisfies \[ D_{\mathcal L}^2 =\nabla_{\mathcal L}^*\nabla_{\mathcal L} +\frac{\operatorname{Scal}}4. \] There is no additional curvature term from the twist. On a compact positive-scalar-curvature manifold, $\operatorname{Scal}\geq s_0>0$. Therefore, in the Hilbert-module quadratic-form sense, \[ D_{\mathcal L}^2\geq(s_0/4)I. \] The regular self-adjoint functional calculus gives an inverse for $D_{\mathcal L}$, with inverse norm at most $2/\sqrt{s_0}$. Its real index class is zero. This is the established index obstruction in [S1], with the spectral-gap implication spelled out: \[ \operatorname{Scal}>0\quad\Longrightarrow \quad\alpha(M)=0. \] The converse cannot be obtained by reversing this argument. Index zero says that an obstruction vanishes; it does not produce an invertible Dirac operator for a metric, much less a pointwise-positive scalar curvature term.

Why the finite-group index is more than one number.

Since $G$ is finite, the real group algebra is already finite-dimensional and complete in its operator norm. Real semisimple representation theory decomposes it as \[ \mathbb R[G]\cong \bigoplus_jM_{d_j}(\mathbb R)\ \oplus\ \bigoplus_kM_{e_k}(\mathbb C)\ \oplus\ \bigoplus_\ell M_{f_\ell}(\mathbb H). \] Morita invariance decomposes its real $K$-theory accordingly. Thus $\alpha(M)$ has several components, including real and quaternionic information and possible torsion. It must not be replaced by only an ordinary complex $\widehat A$ number.

For an explicit algebra test, if $G=(\mathbb Z/2)^2$ then $\mathbb R[G]\cong\mathbb R^4$, with primitive idempotents \[ e_\chi=\frac14\sum_{g\in G}\chi(g)g \] for its four real characters. They satisfy $e_\chi e_\psi=0$ for $\chi\ne\psi$, $e_\chi^2=e_\chi$ and $\sum_\chi e_\chi=1$. The index is the tuple of the four corresponding flat-line-bundle twists.

The ordinary index is its trivial-character component. The index on the finite cover is the sum of the four components, since the regular representation is their direct sum. These two homomorphisms do not algebraically determine a tuple. For example in degree $n\equiv1\pmod8$, where $KO_n(\mathbb R)=\mathbb Z/2$, the tuple $(0,1,1,0)$ has trivial first component and zero sum but is not zero. This is an abstract target-group calculation, not a claim that the tuple is realized by a particular spin manifold. Geometric realizability can impose additional constraints and would itself require proof. The example shows why inspecting only those two numerical images cannot replace inspection of the specified full index without such an argument.

Averaging metrics is not a converse construction.

If a positive-scalar-curvature metric is found on the universal cover and is invariant under the deck group, it descends to $M$ with the same local curvature. The invariance condition is essential. Averaging a metric over the finite deck group makes it invariant and positive definite as a tensor, but scalar curvature depends nonlinearly on the metric and its first two derivatives. No positivity-preserving identity follows from tensor averaging.

The index calculation on the cover therefore does not provide a descended metric. One would need a separate equivariant construction. This is a concrete missing compatibility condition, distinct from the vanishing of the ordinary index.

Products show why stabilization changes the problem.

For product metrics, \[ \operatorname{Scal}(g\oplus\varepsilon^2g_{S^2}) =\operatorname{Scal}(g)+2\varepsilon^{-2}. \] Every compact $(M,g)$ has scalar curvature bounded below. Hence $M\times S^2$ has positive scalar curvature for sufficiently small $\varepsilon$, regardless of whether $M$ does. The sphere is spin and has zero Dirac index, so multiplicativity gives $\alpha(M\times S^2)=0$ even when $\alpha(M)\ne0$. This product has erased the obstruction and changed the target.

The Bott stabilization in [S1] is subtler: the auxiliary eight-manifold has index the Bott generator, so multiplication is an isomorphism on the periodic index group. Even there, a metric on the product does not automatically yield a positive-scalar- curvature metric on the original factor. Restricting a metric to a slice has no general positivity implication for scalar curvature. Therefore stable existence, even with the correct Bott factor, still requires an unstabilization argument.

A concrete sufficient surgery certificate.

The classical positive-scalar-curvature surgery theorem in [S1] says that surgery in codimension at least three preserves existence of such a metric. Consequently the following data would certify the desired conclusion for a particular $M$: a closed spin $n$-manifold $N$ already carrying positive scalar curvature, and a cobordism from $N$ to $M$ decomposed, relative to $N$, into handles of indices at most $n-2$.

An index-$k$ handle changes the boundary by surgery on an embedded $S^{k-1}$ whose codimension in that boundary is $n-k+1\geq3$. Apply the surgery theorem successively to get a metric on the final boundary $M$. Disjoint starting components and connecting handles can be handled in the same way. If one wants to track the fundamental group and spin data through a bordism over $BG$, those data must also extend over the handles; the metric theorem alone does not choose them.

This is a verifiable sufficient geometric certificate, rather than a construction deduced from index zero. An arbitrary bordism may contain handles of forbidden indices. Eliminating them while respecting the classifying map is an additional topological problem. The vanishing index does not visibly hand over such a handle decomposition.

Round spherical space forms supply simple starting examples: a finite free isometric quotient of a round sphere has positive scalar curvature, and when it is spin it falls within the necessary-index calculation. They do not exhaust all finite fundamental groups or all spin bordism classes.

Verification, first gap, and outcome.

The scalar-curvature scaling and the surgery codimension were checked with their actual dimensions. The character idempotents were checked directly in the finite real group algebra. The torsion tuple is explicitly not asserted to be geometrically realizable. The Dirac argument uses a flat real twist, rather than a complex numerical substitute.

The first missing general step is to turn vanishing of the full real Rosenberg index into a geometric existence certificate for the original manifold, with no auxiliary product. That could involve a sufficiently strong spin-bordism generation theorem and the required surgery control, but those inputs are not proved here. UNRESOLVED. The obstruction direction, explicit index-component tests and sufficient surgery criterion are established. No unstabilized converse for all finite groups or counterexample in the stated class is claimed.

Sources

• S1 Jonathan Rosenberg, Modifications of the Gromov–Lawson Conjecture, conference slides. https://www.math.umd.edu/~jmr/LawsonConf.pdf. Formulation source.
• E1 Arjun Malhotra, The Gromov--Lawson--Rosenberg conjecture for some finite groups, arXiv:1305.0455 $2013$. https://arxiv.org/abs/1305.0455. Primary paper treating specified finite groups; consulted 27 September 2026. Its restricted scope is not promoted to the general unstabilized assertion.