What Constraints Does Torus Surgery Impose on Intermediate Covers of the Cartwright-Steger Surface?

Why torus surgery cannot simplify the surface

The Cartwright–Steger surface X is a compact quotient of the complex two-ball, hence aspherical with strictly negative sectional curvature, and its numerical invariants are e(X) = 3, σ(X) = 1, H₁(X;Z) ≅ Z². The paper's first question is how far torus surgery can change this topology. The answer is: not far enough to reach a rational homology CP².

The mechanism is Preissmann's theorem. In a closed aspherical manifold of strictly negative sectional curvature, every nontrivial abelian subgroup of the fundamental group is infinite cyclic, so the image of any map T² → X on π₁ is trivial or cyclic. The induced map on H₂ therefore factors through H₂(S¹;Z) = 0, giving Theorem 2.1: every continuous map T² → X sends the integral fundamental class to zero. Corollary 2.2 draws the geometric consequence — every smoothly embedded torus in X is integrally nullhomologous, and X contains no symplectically embedded torus, since a symplectic torus would have ∫ω > 0 while a nullhomologous class pairs to zero with ω.

The Betti number floor for fixed torus links

The obstruction is quantitative, not just qualitative. For a closed connected oriented four-manifold M and pairwise disjoint oriented embedded tori T₁,…,Tₙ with [Tᵢ] = 0 in H₂(M;Z), Lemma 3.1 gives an exact sequence 0 → Zⁿ → H₁(C;Z) → H₁(M;Z) → 0, where C is the complement of the tubular neighborhoods and the standard generators map to the meridians.

Theorem 3.2 then computes the effect of arbitrary torus fillings: H₁(N;Z) ≅ H₁(C;Z)/⟨γ₁,…,γₙ⟩, and over every field F, dim_F H₁(N;F) ≥ dim_F H₁(M;F). Euler characteristic and signature are unchanged, because the removed and attached pieces both have Euler characteristic and signature zero. Applied to X, where H₁(X;Z) ≅ Z², Corollary 3.3 gives dim_F H₁(N;F) ≥ 2 over every field. So N is neither simply connected nor a rational homology CP², and direct Luttinger surgeries on such a family cannot produce a standard or exotic CP².

The scope condition matters here. Remark 3.4 states that Corollary 3.3 concerns fillings of the common complement of a fixed disjoint torus link in X; it does not automatically apply to later surgeries on core tori created by previous fillings, or to newly selected tori whose nullhomology has not been verified in the manifold present at that stage. If d = b₁(N) − 2, then d ≥ 0 and the preserved Euler characteristic and signature give (b₂, b₂⁺, b₂⁻) = (5 + 2d, 3 + d, 2 + d).

Why the nodal curve controls the double cover

The paper's constructive half revisits the degree-two cover proposed in Remark 1 of the earlier Akhmedov–Sakallı–Yeung work, which outlined a minimal symplectic four-manifold with π₁ ≅ Z/2 and (σ, χ_h) = (0,8) but supplied neither the subgroup nor the surface-complement calculations. The present paper completes them.

The covering tower is organized by N = ker(Π → H₁(X;Z/2)), where Π = π₁(X). Since H₁(X;Z) ≅ Z², one has [Π : N] = 4 and Π/N ≅ (Z/2)². The fourfold cover X₄ and any intermediate double cover X₂ fit into X₄ → X₂ → X with π₁(X₄) = N ⊂ Π₂ = π₁(X₂) ⊂ Π and [Π₂ : N] = 2. The decisive point is that the image of the fundamental group of the nodal curve is precisely N. Consequently the fourfold cover contains four copies of the curve and each intermediate double cover contains two, each mapping homeomorphically to the original curve.

The nodal curve D ⊂ X has normalization of genus four and D² = −1. In each of the three possible intermediate covers one lifted component D₂ is chosen; its two local branches meet positively, and the node is smoothed symplectically by replacing {z₁z₂ = 0} with {z₁z₂ = ε} for 0 < |ε| ≪ 1. Because H₂(U, ∂U;Z) = H₂(B⁴, S³;Z) = 0, the local replacement does not change the global class: [D] = [D₂] in H₂(X₂;Z). The smoothing adds one handle to the genus-four normalization, giving g(D) = 5 and [D]² = −1. Blowing up a regular point q of D away from the smoothing chart produces a genus-five symplectic surface Σ₋ ⊂ X₂#CP² with [Σ₋]² = −2, and the exceptional sphere meets it once — the geometric dual that supplies the meridian relation.

The nodal-curve subgroup equals the mod-two kernel

The group-theoretic core is the identification B = N, where B = im(π₁(D) → Π) is the nodal-curve subgroup. Lemma 6.8 shows that, up to conjugacy, the image H of π₁(Σ₋) in Π₂ is B, so K = ⟨⟨B⟩⟩_{Π₂}. The proof uses a local model: in a small ball about the node the two branches are the coordinate disks {uv = 0}; replacing them by {uv = ε} attaches a one-handle to the normalization with small disks removed at the two preimages of the node. Closing an arc through the core of the smoothing annulus gives a generator whose image is r₁r₂⁻¹ = h, while a dual generator — the waist circle of the smoothing annulus — bounds a disk in the coordinate ball and maps trivially. Van Kampen applied to the smoothed surface therefore gives im(π₁(D) → Π₂) = ⟨A, h⟩ = B up to conjugacy, and blowing up the regular point q does not change the ambient fundamental group.

What the surface-sum quotient becomes

With B = N in hand, Proposition 6.7 computes the quotient for every fiber-reversing normal-bundle gluing: P ≅ Π₂/K. The proof writes μ_A and μ_B for the two meridians; Lemma 6.6 gives i_A(μ_A) = 1 in G_A, and a fiber-reversing gluing has Φ_*(μ_A) = μ_B⁻¹, so the amalgamation relation forces i_B(μ_B) = 1 in P. Since G_B is normally generated by μ_B, the homomorphism G_B → P is trivial, and the only remaining relations in G_A ≅ Π₂ are the normal closure of H. Proposition 6.3 gives B = N, so P ≅ Π₂/N ≅ Z/2, proving Theorem 4.3.

The numerical invariants of the resulting manifold Z = (X₂#CP²) #_{Σ₋ = F} X_{2,3} are e(Z) = 32, σ(Z) = 0, χ_h(Z) = 8, c₁²(Z) = 64. The double cover has e(X₂) = 6 and σ(X₂) = 2, so e(X₂#CP²) = 7 and σ(X₂#CP²) = 1; the second block has e(X_{2,3}) = 9 and σ(X_{2,3}) = −1; and e(F) = −8, so the fiber-sum formulas give e(Z) = 7 + 9 − 2(−8) = 32 and σ(Z) = 1 − 1 = 0. Rationally b₁(Z) = 0, b₂(Z) = 30, and b₂⁺(Z) = b₂⁻(Z) = 15, so Z has the rational homology of 15CP²#15CP̄². The integral homology is H₂(Z;Z) = Z³⁰ ⊕ Z/2, obtained from e(Z) = 32 together with H₁(Z;Z) = Z/2 and Poincaré duality.

Minimality is Theorem 5.1: for every symplectically admissible gluing, Z is minimal. The blowdown map p: X₂#CP² → X₂ has kernel ZE, and since X₂ is aspherical every embedded sphere maps nullhomotopically, so S² = −1 forces [S] = ±E; but E·[Σ₋] = 1, so no exceptional sphere is disjoint from Σ₋. The other block X_{2,3} is minimal, and the ruled exception in Usher's theorem does not occur because the two summands have signatures 1 and −1 rather than the signature zero required of an S²-bundle over a surface.

The universal cover is an exotic symplectic manifold

For compatible symplectic gluings, the universal double cover of Z is an exotic symplectic 31CP²#31CP̄². The paper separates the topological gluing statement from the symplectic assertion: the fundamental-group calculation applies to every orientation-reversing isomorphism of the normal circle bundles, while the symplectic statements require areas scaled to agree and a gluing symplectically admissible in Gompf's sense.

The universal double cover has e = 64, σ = 0, and b₂⁺ = b₂⁻ = 31. The paper is explicit about what it does not claim: the equivariant intersection form of the quotient is not computed, so no homeomorphism classification of Z itself is asserted. The w₂-type is determined for compatible gluings via Proposition 5.3, but that determination is tied to the compatibility condition on the gluing, not to arbitrary gluings.

The same argument on the original surface

The construction has three natural levels. The fourfold cover used in the earlier work gives the simply connected exotic manifold M_{2,10}, homeomorphic to 19CP²#17CP̄² with signature two and (e, σ, χ_h, c₁²) = (38, 2, 10, 82). The intermediate double cover gives the signature-zero manifold with π₁ ≅ Z/2 studied above. The paper then applies the same argument to the original Cartwright–Steger surface.

Smoothing the positive node of D ⊂ X gives an embedded genus-five symplectic surface D_V representing the same homology class, with [D_V]² = −1. Blowing up a regular point of D_V gives a symplectic surface Σ_V ⊂ X#CP² with g(Σ_V) = 5 and Σ_V² = −2, and the exceptional sphere is again a geometric dual. Using the square +2 surface F ⊂ X_{2,3}, define Z_V = (X#CP²) #_{Σ_V = F} X_{2,3}.

Theorem 7.1 states that for every symplectically admissible fiber-reversing normal-bundle gluing, Z_V is a minimal symplectic four-manifold with π₁(Z_V) ≅ (Z/2)² and (e, σ, χ_h, c₁²) = (29, −1, 7, 55), plus H₁(Z_V;Z) ≅ (Z/2)² and H₂(Z_V;Z) ≅ Z²⁷ ⊕ (Z/2)². The negative block has e(X#CP²) = 4 and σ(X#CP²) = 0, so e(Z_V) = 4 + 9 − 2(−8) = 29 and σ(Z_V) = 0 − 1 = −1, giving χ_h(Z_V) = 7 and c₁²(Z_V) = 55. Minimality follows the same route as Theorem 5.1, with the ruled exception excluded because the negative block's fundamental group has cohomological dimension four while a ruled surface's has cohomological dimension at most two. This Klein four-group construction does not appear in the earlier work.

For a symplectically admissible gluing compatible with the dual surfaces, Corollary 7.2 shows the universal fourfold cover of Z_V is an exotic symplectic 55CP²#59CP̄². The inverse image of a square −1 surface S ⊂ Z_V consists of four disjoint copies, so the cover is non-spin; it has e = 4e(Z_V) = 116, σ = 4σ(Z_V) = −4, hence b₂ = 114, b₂⁺ = 55, b₂⁻ = 59, with odd indefinite unimodular intersection form 55⟨1⟩ ⊕ 59⟨−1⟩. Freedman's theorem gives the homeomorphism, and Taubes's theorem gives a nonzero Seiberg–Witten invariant since b₂⁺ = 55 > 1, while the standard connected sum has vanishing Seiberg–Witten invariants — so the two smooth structures are not diffeomorphic.

What the exact computation certifies and what it does not

The subgroup identification rests on exact Todd–Coxeter enumeration in finitely presented groups, and the paper is careful about the division of labor. The ambient lattice is entered as Γ = ⟨u, v, b | u³, v⁴, b³, (uv)²(vu)⁻², [v, b], (buv)³, (buvu)²v⟩, with j = (uv)² and a₁, a₂, a₃ defined by the CKY words. The subgroup Π = ⟨a₁, a₂, a₃⟩ < Γ has index 864, computed by GAP and reproducing the CKY index.

The eight normalization words p₁,…,p₈ and the node-joining word h = r₁r₂⁻¹, with r₁ = a₃³a₁a₂⁻¹ and r₂ = a₂a₁⁻²a₃⁻¹a₁a₃⁻¹a₁⁻¹a₂⁻², define B = ⟨p₁,…,p₈, h⟩ < Γ. SymPy 1.14.0 computes [Γ : B] = 3456 by two independent strategies, and together with [Γ : Π] = 864 the index formula gives [Π : B] = 4 once B ⊂ Π is verified. GAP 4.12.1 checks that containment, and forms K = ⟨⟨B⟩⟩^Π, returning [Π : normal closure(B)] = 4, Size(Π/normal closure(B)) = 4, IsAbelian = true, Abelian invariants = [2, 2], so Π/K ≅ (Z/2)². Since [Π : B] = [Π : K] = [Π : N] = 4, one obtains B = K = N. Finally, for each of the three nonzero characters Π/N → Z/2, GAP verifies that all nine generators of B lie in Π₂ and that [Π₂ : ⟨⟨B⟩⟩^{Π₂}] = 2.

The paper states the boundary of this evidence explicitly: Lemma 6.1 identifies the entered word subgroup with the geometric nodal-curve subgroup, and the computer calculation is used only for the exact finite-index and normal-closure assertions. The ancillary files carry the presentations, words, commands, completion counts, and recorded output, with SHA-256 hashes recorded for the SymPy and GAP artifacts. The paper also discloses that LLM tools were used for grammar and language editing, with all mathematical ideas, arguments, and results attributed to the authors.

Sources used in this answer