Documentation

TauCeti.Geometry.Manifold.Morse.Cell.Attachment

Cell and handle attachment across a critical level #

Let f be a smooth function on a boundaryless finite-dimensional manifold, and let c be a critical value at which f has finitely many critical points, all nondegenerate, such that the band f⁻¹[c - δ, c + δ] is compact for some δ > 0. Milnor's Theorem 3.2 says that passing the level c attaches one cell at each of these critical points: for small ε > 0, the sublevel M_{c+ε} = {f ≤ c + ε} deformation retracts onto the lower sublevel M_{c-ε} = {f ≤ c - ε} with one λ-cell attached at each critical point of index λ.

The cells are explicit. In a normalized Morse chart ψ at a critical point of index λ,

f = c - Σ_{j < λ} yⱼ² + Σ_{j ≥ λ} yⱼ²,

and the cell is the disc {‖y‖² ≤ ε, yⱼ = 0 for j ≥ λ} of the negative coordinates, read in M. Its boundary sphere lies on the level f⁻¹(c - ε).

Following Milnor, write u for the negative and v for the positive coordinates of a chart. The function F = f - μ(‖u‖² + 2‖v‖²) agrees with f outside the ellipsoids ‖u‖² + 2‖v‖² ≤ 2ε, where μ is a smooth nonincreasing cutoff with μ = 2ε on [0, ε] and μ = 0 on [2ε, ∞) (TauCeti.exists_contDiff_antitone_cutoff). It has the same sublevel {F ≤ c + ε} = M_{c+ε}. When the band f⁻¹[c - ε, c + ε] is compact and its critical points lie in the charts, F has no critical points in the band F⁻¹[c - ε, c + ε]: outside the ellipsoids it agrees with f, whose critical points in the band are the chart centres; along the positive directions F strictly increases away from v = 0, and on v = 0 it lies below c - ε. So M_{c+ε} deformation retracts onto {F ≤ c - ε} (TauCeti.exists_sublevelDeformation), which in turn retracts onto M_{c-ε} with the cells attached (TauCeti.exists_inter_sublevel_union_cells_deformation), since lowering the positive coordinates lowers F. The two retractions compose.

Milnor's Remark 3.3 refines this to a handle attachment, up to a diffeomorphism rather than a homotopy. Since F has no critical points in the compact band, the descending flow of TauCeti.exists_diffeomorph_sublevel is a diffeomorphism of M carrying M_{c+ε} = {F ≤ c + ε} onto {F ≤ c - ε}, and the level f⁻¹(c + ε) onto F⁻¹(c - ε). For a cutoff positive on (-∞, 2ε), the sublevel {F ≤ c - ε} is M_{c-ε} together with one handle in each chart, the points of {F ≤ c - ε} on or above the level f = c - ε in Milnor's ellipsoid ‖u‖² + 2‖v‖² ≤ 2ε. In the coordinates (u, v) it is Milnor's handle TauCeti.quadraticHandle ε μ, which is homeomorphic to D^λ × D^(m-λ) with its attaching region {f = c - ε} carried onto ∂D^λ × D^(m-λ) (TauCeti.quadraticHandleHomeomorph).

Main results #

References #

theorem TauCeti.exists_sublevel_union_cells_deformation {E : Type u_1} {H : Type u_2} {M : Type u_3} [NormedAddCommGroup E] [NormedSpace ℝ E] [TopologicalSpace H] {I : ModelWithCorners ℝ E H} [TopologicalSpace M] [ChartedSpace H M] {ι : Type u_4} {n : ℕ} {f : M → ℝ} {c ε : ℝ} [FiniteDimensional ℝ E] [I.Boundaryless] [IsManifold I (↑⊤) M] [T2Space M] [Finite ι] (hfs : ContMDiff I (modelWithCornersSelf ℝ ℝ) (↑⊤) f) (ψ : ι → PartialDiffeomorph I (modelWithCornersSelf ℝ (EuclideanSpace ℝ (Fin n))) M (EuclideanSpace ℝ (Fin n)) ↑⊤) (hψ : Pairwise (Function.onFun Disjoint fun (i : ι) => (ψ i).source)) (k : ι → ℕ) (hε : 0 < ε) (hball : ∀ (i : ι), Metric.closedBall 0 √(2 * ε) ⊆ (ψ i).target) (hf : ∀ (i : ι), ∀ y ∈ (ψ i).source, f y = c + ∑ j : Fin n, (if ↑j < k i then -1 else 1) * (↑(ψ i).toPartialEquiv y).ofLp j ^ 2) (hcompact : IsCompact (f ⁻¹' Set.Icc (c - ε) (c + ε))) (hcrit : criticalPoints I f ∩ f ⁻¹' Set.Icc (c - ε) (c + ε) ⊆ ⋃ (i : ι), (ψ i).source) :
∃ (r : C(↑{x : M | f x ≤ c + ε}, ↑({x : M | f x ≤ c + ε} ∩ ({x : M | f x ≤ c - ε} ∪ ⋃ (i : ι), (ψ i).source ∩ ↑(ψ i).toPartialEquiv ⁻¹' {y : EuclideanSpace ℝ (Fin n) | ‖y‖ ^ 2 ≤ ε ∧ ∀ (j : Fin n), k i ≤ ↑j → y.ofLp j = 0})))), (∀ (x : ↑({x : M | f x ≤ c + ε} ∩ ({x : M | f x ≤ c - ε} ∪ ⋃ (i : ι), (ψ i).source ∩ ↑(ψ i).toPartialEquiv ⁻¹' {y : EuclideanSpace ℝ (Fin n) | ‖y‖ ^ 2 ≤ ε ∧ ∀ (j : Fin n), k i ≤ ↑j → y.ofLp j = 0}))), r ((ContinuousMap.inclusion ⋯) x) = x) ∧ Nonempty (((ContinuousMap.inclusion ⋯).comp r).HomotopyRel (ContinuousMap.id ↑{x : M | f x ≤ c + ε}) {x : ↑{x : M | f x ≤ c + ε} | ↑x ∈ {x : M | f x ≤ c - ε} ∪ ⋃ (i : ι), (ψ i).source ∩ ↑(ψ i).toPartialEquiv ⁻¹' {y : EuclideanSpace ℝ (Fin n) | ‖y‖ ^ 2 ≤ ε ∧ ∀ (j : Fin n), k i ≤ ↑j → y.ofLp j = 0}})

Cell attachment in disjoint Morse charts. Let f be smooth, and let ψ i be charts of M with pairwise disjoint sources, each containing the closed ball of radius √(2ε) about 0 in its target, in which f = c - Σ_{j < k i} yⱼ² + Σ_{j ≥ k i} yⱼ². Suppose the band f⁻¹[c - ε, c + ε] is compact and its critical points lie in the chart sources. The cell of the chart ψ i is the disc {‖y‖² ≤ ε, yⱼ = 0 for j ≥ k i} of the first k i coordinates, read in M.

Then the lower sublevel {f ≤ c - ε} with these cells attached is a strong deformation retract of {f ≤ c + ε}: there is a retraction of {f ≤ c + ε} onto it, and a homotopy in {f ≤ c + ε} from this retraction to the identity which fixes it pointwise. The cells lie in {f ≤ c}, so intersecting with {f ≤ c + ε} does not change the retract.

theorem TauCeti.exists_diffeomorph_sublevel_eq_union_handles {E : Type u_1} {H : Type u_2} {M : Type u_3} [NormedAddCommGroup E] [NormedSpace ℝ E] [TopologicalSpace H] {I : ModelWithCorners ℝ E H} [TopologicalSpace M] [ChartedSpace H M] {ι : Type u_4} {n : ℕ} {f : M → ℝ} {c ε : ℝ} [FiniteDimensional ℝ E] [I.Boundaryless] [IsManifold I (↑⊤) M] [T2Space M] [Finite ι] (hfs : ContMDiff I (modelWithCornersSelf ℝ ℝ) (↑⊤) f) (ψ : ι → PartialDiffeomorph I (modelWithCornersSelf ℝ (EuclideanSpace ℝ (Fin n))) M (EuclideanSpace ℝ (Fin n)) ↑⊤) (hψ : Pairwise (Function.onFun Disjoint fun (i : ι) => (ψ i).source)) (k : ι → ℕ) (hε : 0 < ε) (hball : ∀ (i : ι), Metric.closedBall 0 √(2 * ε) ⊆ (ψ i).target) (hf : ∀ (i : ι), ∀ y ∈ (ψ i).source, f y = c + ∑ j : Fin n, (if ↑j < k i then -1 else 1) * (↑(ψ i).toPartialEquiv y).ofLp j ^ 2) (hcompact : IsCompact (f ⁻¹' Set.Icc (c - ε) (c + ε))) (hcrit : criticalPoints I f ∩ f ⁻¹' Set.Icc (c - ε) (c + ε) ⊆ ⋃ (i : ι), (ψ i).source) :
∃ (H_1 : ι → Set M) (e : (i : ι) → ↑(H_1 i) ≃ₜ ↑(Metric.closedBall 0 1) × ↑(Metric.closedBall 0 1)) (Φ : Diffeomorph I I M M ↑⊤), (∀ (i : ι), H_1 i ⊆ (ψ i).source) ∧ (∀ (i : ι) (x : ↑(H_1 i)), c - ε ≤ f ↑x) ∧ (∀ (i : ι) (x : ↑(H_1 i)), f ↑x ≤ c - ε ↔ ‖↑((e i) x).1‖ = 1) ∧ (∀ (i : ι) (x : ↑(H_1 i)), ↑((e i) x).2 = 0 ↔ ∀ (j : Fin n), k i ≤ ↑j → (↑(ψ i).toPartialEquiv ↑x).ofLp j = 0) ∧ (∀ (x : M), Φ x ∈ {x : M | f x ≤ c - ε} ∪ ⋃ (i : ι), H_1 i ↔ f x ≤ c + ε) ∧ ∀ (x : M), f x = c + ε ↔ (∀ (i : ι) (h : Φ x ∈ H_1 i), ‖↑((e i) ⟨Φ x, h⟩).2‖ = 1) ∧ (Φ x ∉ ⋃ (i : ι), H_1 i → f (Φ x) = c - ε)

Handle attachment in disjoint Morse charts (Milnor, Remark 3.3). Let f be smooth, and let ψ i be charts of M with pairwise disjoint sources, each containing the closed ball of radius √(2ε) about 0 in its target, in which f = c - Σ_{j < k i} yⱼ² + Σ_{j ≥ k i} yⱼ². Suppose the band f⁻¹[c - ε, c + ε] is compact and its critical points lie in the chart sources.

Then there are handles H i in the chart sources, on or above the level f⁻¹(c - ε), and a diffeomorphism Φ of M carrying the upper sublevel {f ≤ c + ε} onto the lower sublevel {f ≤ c - ε} with the handles attached. Each handle is homeomorphic to the product D^{k i} × D^{n - k i} of closed unit balls in the first k i and in the remaining coordinates, by a homeomorphism e i carrying:

  • its intersection with {f ≤ c - ε} onto ∂D^{k i} × D^{n - k i};
  • its core, the points whose coordinates beyond the first k i vanish, onto D^{k i} × 0.

Φ carries the level f⁻¹(c + ε) onto the boundary of the attached space: the points on the top D^{k i} × ∂D^{n - k i} of every handle containing them, and on the level f⁻¹(c - ε) outside the handles.

theorem TauCeti.IsMorse.exists_sublevel_union_cells_deformation {E : Type u_1} {H : Type u_2} {M : Type u_3} [NormedAddCommGroup E] [NormedSpace ℝ E] [TopologicalSpace H] {I : ModelWithCorners ℝ E H} [TopologicalSpace M] [ChartedSpace H M] {f : M → ℝ} {c : ℝ} [FiniteDimensional ℝ E] [I.Boundaryless] [IsManifold I (↑⊤) M] [T2Space M] (hf : IsMorse I f) {δ : ℝ} (hδ : 0 < δ) (hcompact : IsCompact (f ⁻¹' Set.Icc (c - δ) (c + δ))) :
∃ (ε₀ : ℝ), 0 < ε₀ ∧ ε₀ ≤ δ ∧ ∃ (ψ : ↑(criticalPoints I f ∩ f ⁻¹' {c}) → PartialDiffeomorph I (modelWithCornersSelf ℝ (EuclideanSpace ℝ (Fin (Module.finrank ℝ E)))) M (EuclideanSpace ℝ (Fin (Module.finrank ℝ E))) ↑⊤), Pairwise (Function.onFun Disjoint fun (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})) => (ψ p).source) ∧ (∀ (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})), ↑p ∈ (ψ p).source ∧ ↑(ψ p).toPartialEquiv ↑p = 0 ∧ Metric.closedBall 0 √(2 * ε₀) ⊆ (ψ p).target ∧ ∀ y ∈ (ψ p).source, f y = c + ∑ i : Fin (Module.finrank ℝ E), (if ↑i < manifoldMorseIndex I f ↑p then -1 else 1) * (↑(ψ p).toPartialEquiv y).ofLp i ^ 2) ∧ ∀ (ε : ℝ), 0 < ε → ε ≤ ε₀ → ∃ (r : C(↑{x : M | f x ≤ c + ε}, ↑({x : M | f x ≤ c + ε} ∩ ({x : M | f x ≤ c - ε} ∪ ⋃ (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})), (ψ p).source ∩ ↑(ψ p).toPartialEquiv ⁻¹' {y : EuclideanSpace ℝ (Fin (Module.finrank ℝ E)) | ‖y‖ ^ 2 ≤ ε ∧ ∀ (i : Fin (Module.finrank ℝ E)), manifoldMorseIndex I f ↑p ≤ ↑i → y.ofLp i = 0})))), (∀ (x : ↑({x : M | f x ≤ c + ε} ∩ ({x : M | f x ≤ c - ε} ∪ ⋃ (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})), (ψ p).source ∩ ↑(ψ p).toPartialEquiv ⁻¹' {y : EuclideanSpace ℝ (Fin (Module.finrank ℝ E)) | ‖y‖ ^ 2 ≤ ε ∧ ∀ (i : Fin (Module.finrank ℝ E)), manifoldMorseIndex I f ↑p ≤ ↑i → y.ofLp i = 0}))), r ((ContinuousMap.inclusion ⋯) x) = x) ∧ Nonempty (((ContinuousMap.inclusion ⋯).comp r).HomotopyRel (ContinuousMap.id ↑{x : M | f x ≤ c + ε}) {x : ↑{x : M | f x ≤ c + ε} | ↑x ∈ {x : M | f x ≤ c - ε} ∪ ⋃ (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})), (ψ p).source ∩ ↑(ψ p).toPartialEquiv ⁻¹' {y : EuclideanSpace ℝ (Fin (Module.finrank ℝ E)) | ‖y‖ ^ 2 ≤ ε ∧ ∀ (i : Fin (Module.finrank ℝ E)), manifoldMorseIndex I f ↑p ≤ ↑i → y.ofLp i = 0}})

Milnor's cell attachment theorem. Let f be a Morse function on a Hausdorff manifold whose band f⁻¹[c - δ, c + δ] is compact, for some 0 < δ. Then there are a width 0 < ε₀ ≤ δ and normalized Morse charts ψ p with pairwise disjoint sources, indexed by the critical points p on the level f⁻¹(c), with ψ p p = 0, with the closed ball of radius √(2ε₀) in the target of ψ p, and in which f = c - Σ_{i < λ} yᵢ² + Σ_{i ≥ λ} yᵢ², where λ is the Morse index of f at p. For every width 0 < ε ≤ ε₀, the cell at p is the disc {‖y‖² ≤ ε, yᵢ = 0 for i ≥ λ} of its negative coordinates, read in M through ψ p.

The lower sublevel {f ≤ c - ε} with these cells attached is a strong deformation retract of {f ≤ c + ε}, for every such ε: there is a retraction of {f ≤ c + ε} onto it, and a homotopy in {f ≤ c + ε} from this retraction to the identity which fixes it pointwise.

theorem TauCeti.IsMorse.exists_diffeomorph_sublevel_eq_union_handles {E : Type u_1} {H : Type u_2} {M : Type u_3} [NormedAddCommGroup E] [NormedSpace ℝ E] [TopologicalSpace H] {I : ModelWithCorners ℝ E H} [TopologicalSpace M] [ChartedSpace H M] {f : M → ℝ} {c : ℝ} [FiniteDimensional ℝ E] [I.Boundaryless] [IsManifold I (↑⊤) M] [T2Space M] (hf : IsMorse I f) {δ : ℝ} (hδ : 0 < δ) (hcompact : IsCompact (f ⁻¹' Set.Icc (c - δ) (c + δ))) :
∃ (ε : ℝ), 0 < ε ∧ ε ≤ δ ∧ ∃ (ψ : ↑(criticalPoints I f ∩ f ⁻¹' {c}) → PartialDiffeomorph I (modelWithCornersSelf ℝ (EuclideanSpace ℝ (Fin (Module.finrank ℝ E)))) M (EuclideanSpace ℝ (Fin (Module.finrank ℝ E))) ↑⊤), Pairwise (Function.onFun Disjoint fun (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})) => (ψ p).source) ∧ (∀ (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})), ↑p ∈ (ψ p).source ∧ ↑(ψ p).toPartialEquiv ↑p = 0 ∧ Metric.closedBall 0 √(2 * ε) ⊆ (ψ p).target ∧ ∀ y ∈ (ψ p).source, f y = c + ∑ i : Fin (Module.finrank ℝ E), (if ↑i < manifoldMorseIndex I f ↑p then -1 else 1) * (↑(ψ p).toPartialEquiv y).ofLp i ^ 2) ∧ ∃ (H_1 : ↑(criticalPoints I f ∩ f ⁻¹' {c}) → Set M) (e : (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})) → ↑(H_1 p) ≃ₜ ↑(Metric.closedBall 0 1) × ↑(Metric.closedBall 0 1)) (Φ : Diffeomorph I I M M ↑⊤), (∀ (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})), H_1 p ⊆ (ψ p).source) ∧ (∀ (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})) (x : ↑(H_1 p)), c - ε ≤ f ↑x) ∧ (∀ (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})) (x : ↑(H_1 p)), f ↑x ≤ c - ε ↔ ‖↑((e p) x).1‖ = 1) ∧ (∀ (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})) (x : ↑(H_1 p)), ↑((e p) x).2 = 0 ↔ ∀ (i : Fin (Module.finrank ℝ E)), manifoldMorseIndex I f ↑p ≤ ↑i → (↑(ψ p).toPartialEquiv ↑x).ofLp i = 0) ∧ (∀ (x : M), Φ x ∈ {x : M | f x ≤ c - ε} ∪ ⋃ (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})), H_1 p ↔ f x ≤ c + ε) ∧ ∀ (x : M), f x = c + ε ↔ (∀ (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})) (h : Φ x ∈ H_1 p), ‖↑((e p) ⟨Φ x, h⟩).2‖ = 1) ∧ (Φ x ∉ ⋃ (p : ↑(criticalPoints I f ∩ f ⁻¹' {c})), H_1 p → f (Φ x) = c - ε)

Milnor's handle attachment theorem (Milnor, Remark 3.3). Let f be a Morse function on a Hausdorff manifold of dimension m whose band f⁻¹[c - δ, c + δ] is compact, for some 0 < δ. Then there are a width 0 < ε ≤ δ and normalized Morse charts ψ p with pairwise disjoint sources, indexed by the critical points p on the level f⁻¹(c), with ψ p p = 0, with the closed ball of radius √(2ε) in the target of ψ p, and in which f = c - Σ_{i < λ} yᵢ² + Σ_{i ≥ λ} yᵢ², where λ is the Morse index of f at p.

Attaching one handle D^λ × D^(m-λ) at each critical point turns the lower sublevel {f ≤ c - ε} into the upper one {f ≤ c + ε}, up to a diffeomorphism Φ of M: the conclusion of TauCeti.exists_diffeomorph_sublevel_eq_union_handles holds for these charts.