Find the mechanism before the formula
Topology often asks what a removed set forces the surrounding space to do. Alexander duality turns that outside question into a cohomology calculation on the set itself, with a dimension shift that explains separation and linking.
We will state the theorem with its hypotheses, calculate complements of spheres and circles, turn a meridian into a linking obstruction, and show why the same homology can hide very different knot complements.
This guide belongs to the Gold Standard of Topology, the Gold Standard of Education and the How X Works hub.
Choose a reading route
Start with the question closest to the decision you need to make. Every route returns to the same worked evidence and transfer checks.
| Route | Chapters | Use it when |
|---|---|---|
| 1 | 1–3 | You need the theorem, dimension shift and reduced groups. |
| 2 | 4–5 | You want separation and sphere-complement examples. |
| 3 | 6–8 | You need linking and a changed-dimension transfer. |
| 4 | 9–10 | You must handle wild sets and theorem limits. |
| 5 | 11–12 | You want FAQs and a reproducible duality checklist. |
Chapter contents
Open the group you need, then use the chapter links. Teaching sections remain open on the page.
Build the model
Test the mechanism
1. State the compact-subset theorem first
For a compact, locally contractible subset K of Sn, Alexander duality gives natural isomorphisms H̃i(Sn−K;G)≅H̃n−i−1(K;G), with coefficients G chosen consistently.
The theorem exchanges homology of the complement with cohomology of K and reverses dimension. Reduced groups remove the routine connected component and basepoint terms that would otherwise obscure the clean statement.
Many problems begin in ℝn. Passing to the one-point compactification Sn makes the point at infinity part of the complement, so that step must be stated rather than silently ignored.
2. The n−i−1 shift carries the geometry
If H̃q(K) is nonzero, duality predicts nonzero H̃n−q−1 in the complement. High-dimensional cohomology on K therefore creates lower-dimensional cycles outside it.
For q=n−1, the complementary group is H̃0, which counts components minus one. This is the algebraic mechanism behind familiar separation results such as a circle dividing the sphere into two regions.
| Class on K | Predicted complement group | Geometric reading |
|---|---|---|
| H̃ⁿ⁻¹(K) | H̃₀ | Extra connected components |
| H̃ⁿ⁻²(K) | H̃₁ | Loops around K |
| H̃⁰(K) | H̃ₙ₋₁ | High-dimensional cycles around components |
3. Reduced groups prevent a component-counting trap
If a space has c path components, H̃0 has rank c−1 under ordinary tame hypotheses. Thus H̃0≅ℤ means two components, not one. Forgetting the tilde creates an off-by-one error at exactly the place where separation is being counted.
Similarly, a connected nonempty K has H̃0(K)=0. A disconnected set can create a top-dimensional complement class through H̃0. Coefficients can also reveal torsion that integral rank alone hides.
Write the unreduced interpretation after applying the reduced theorem, rather than rewriting the theorem from memory with ordinary groups.
4. Worked example: a circle in the two-sphere
Let K=S1 embedded tamely in S2. Since H̃1(S1;ℤ)=ℤ, Alexander duality with n=2 and i=0 gives H̃0(S2−S1;ℤ)=ℤ.
Therefore the complement has two components. This recovers the separation conclusion for a simple closed curve. The calculation knows the number of regions from the circle’s first cohomology and the dimension shift.
It does not describe the geometry of either region by itself. Additional theorems are needed to conclude more about tameness or homeomorphism type.
5. A k-sphere leaves one complementary homology degree
For K=Sk in Sn, the only nonzero reduced cohomology is H̃k≅ℤ. Duality gives H̃n−k−1(Sn−Sk)≅ℤ and zero reduced homology in other degrees.
The complement therefore has the homology of Sn−k−1. When k=n−1 this is S0, matching two components. When k=1 and n=3, the complement has a first-homology generator represented by a meridian loop.
The homology conclusion does not depend on whether the embedded circle in S³ is knotted. That strength is also a limitation: homology alone cannot distinguish all embeddings.
6. A meridian records a linking obstruction
Let K be an oriented circle in S³. Duality gives H1(S³−K;ℤ)≅ℤ. A small meridian around K represents a generator. Another disjoint oriented circle L defines a class in this complement, hence an integer multiple of the meridian class.
That integer, after consistent orientation choices, is the linking number. If it is nonzero, L cannot be contracted to a point while remaining in the complement of K, because its homology class is nonzero.
Zero linking number is not proof that two components are unlinked in every stronger sense. Homology detects an abelian obstruction; more subtle links can require the fundamental group or higher invariants.
7. The complement can remember more than duality reports
Every tame knot K⊂S³ has H1(S³−K)≅ℤ, whether K is an unknot or a trefoil. Their complement fundamental groups differ, and the geometry of the complements differs. Alexander duality correctly reports the shared homology but does not collapse those stronger distinctions.
This is a useful discipline: an invariant can obstruct one equivalence without classifying the object. A nonzero dual class proves that a cycle cannot vanish; identical dual groups do not prove that two complements are homeomorphic.
Use van Kampen, knot groups, covering spaces or polynomial invariants when the question lies beyond complement homology.
8. Transfer check: a two-sphere in four-space
Let K=S²⊂S⁴. The nonzero class H̃²(S²)=ℤ corresponds to H̃4−2−1(S⁴−S²)=H̃1≅ℤ. The complement is connected but contains a nontrivial one-dimensional cycle.
The calculation transfers by changing n and k in the same dimension shift. Copying the circle-in-S² conclusion would be wrong: codimension two creates a meridian loop, while codimension one creates separation.
A second check uses K=S⁰ in S². Since H̃⁰(S⁰)=ℤ, the complement has H̃₁=ℤ, consistent with a sphere minus two points being a cylinder-like space.
9. Wild embeddings expose the role of hypotheses
The theorem’s compactness and local contractibility conditions are not decorative. Wild subsets such as the Alexander horned sphere can have complements with unexpected fundamental-group behaviour even when their homology follows the duality pattern.
Hatcher notes that for compact sets without local contractibility, a version of Alexander duality uses Čech cohomology. Substituting singular cohomology for such sets can make the stated theorem false.
The lesson is not that wild examples invalidate duality. They show which cohomology theory and which category of subsets the duality theorem actually controls.
10. Limits and misconceptions
- The theorem uses reduced groups and a dimension shift of n−i−1.
- A complement-homology calculation does not classify embeddings.
- Passing from ℝⁿ to Sⁿ adds the point at infinity.
- Local contractibility cannot be omitted without changing the cohomology theory.
- A nonzero linking number obstructs unlinking, but zero does not settle every link.
- Coefficients and orientability choices must remain consistent.
Alexander duality is strongest when the question is precisely matched to what complement homology can say: components, cycles around the removed set and homological linking obstructions.
11. Frequently asked questions
Why cohomology on K rather than homology?
Duality reverses dimensions and naturally pairs cycles with cohomology classes. Over a field the vector-space distinction may look mild, but integral torsion and naturality make the cohomological statement important.
Does a sphere always separate?
An embedded Sⁿ⁻¹ in Sⁿ gives the expected two complementary components under the appropriate tame hypotheses. Wild embeddings can complicate the regions even while the component count remains.
Is linking number a complete link invariant?
No. It is a powerful first obstruction for oriented components, but different links can share the same pairwise linking numbers.
12. A reproducible duality checklist
| Stage | Question |
|---|---|
| Ambient | ℝⁿ or its compactification Sⁿ? |
| Subset | Compact and locally contractible? |
| Theory | Singular or Čech cohomology? |
| Degree | Did you use n−i−1? |
| Coefficients | Which group or field? |
| Claim | Components, homology, linking or full classification? |
The gold standard is to name the ambient compactification, verify the subset hypotheses, compute the correct reduced cohomology, and stop at the strength of the resulting complement invariant.
Reading routes · Previous chapter · Contents · Continue reading
Continue through the Gold Standard ecosystem
Return to the Gold Standard of Topology, the Gold Standard of Education parent, or the How X Works hub.
Read next in this subject
- The Alexander Horned Sphere, Wild Embeddings and Tame Complements
- Čech Cohomology, Leray Covers and Sheaf Acyclicity
- Antoine’s Necklace, Cantor Sets and Non-Simply-Connected Complements
- Lusternik–Schnirelmann Category, Cup-Length Bounds and Critical Points