Category Theory & Homological Algebra
The Five Lemma: When the Middle Map Must Be an Isomorphism
Give me a commutative ladder of two exact sequences and tell me that four of the five vertical maps are isomorphisms — the Five Lemma forces the fifth, the one in the middle, to be an isomorphism too, for free, without ever writing down a formula for it. This one diagram-chase is the workhorse that lets algebraic topologists prove two homology theories agree, that a long exact sequence pins down an unknown group, and that a map of complexes inducing isomorphisms on homology is a genuine equivalence.
Precisely: in an abelian category, given a commutative diagram with exact rows A₁→A₂→A₃→A₄→A₅ and B₁→B₂→B₃→B₄→B₅ and vertical maps f₁,…,f₅, if f₁, f₂, f₄, f₅ are isomorphisms then f₃ is an isomorphism.
- FieldHomological algebra / category theory
- SettingAbelian category (e.g. R-modules, abelian groups)
- Key hypothesisTwo exact rows, commuting squares, f₁ f₂ f₄ f₅ isos
- ConclusionMiddle map f₃ is an isomorphism
- Proof techniqueDiagram chase (element or arrow-theoretic)
- Sharper formFour Lemma splits it into surjective + injective halves
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
The precise statement
Work in an abelian category 𝒜 — concretely, think of left R-modules, abelian groups (R = ℤ), or vector spaces. Suppose you have a commutative diagram of the shape of two horizontal exact sequences stacked into a ladder:
A₁ → A₂ → A₃ → A₄ → A₅
↓f₁ ↓f₂ ↓f₃ ↓f₄ ↓f₅
B₁ → B₂ → B₃ → B₄ → B₅
Here both rows are exact (the image of each arrow equals the kernel of the next), and every square commutes (going right-then-down equals down-then-right). The Five Lemma asserts: if f₁, f₂, f₄, and f₅ are all isomorphisms, then f₃ is an isomorphism as well. Notably, the conclusion is not about the two ends — it is the map trapped in the center that is determined. What is remarkable is that we never construct f₃⁻¹ explicitly; exactness plus commutativity conspire to force it into existence.
The picture: exactness leaves f₃ no room to fail
Think of exactness as a bookkeeping constraint that ties each object tightly to its neighbours: A₃ is squeezed between A₂ (which surjects onto ker of A₃→A₄'s relevant piece) and A₄ (which receives the cokernel). An element a₃ ∈ A₃ is pinned down by two coordinates: where it maps to in A₄ (its 'derivative') and, via A₂, what its kernel-part is. The vertical maps f₂ and f₄ let you read both coordinates in the B-row perfectly. So any behaviour of f₃ — whether it kills something (failing injectivity) or misses something (failing surjectivity) — would have to show up as a defect in f₂ or f₄, since those neighbours faithfully translate the two coordinates. Because f₂ and f₄ are isomorphisms, there is nowhere for such a defect to hide. The outer maps f₁ and f₅ do the mopping up at the boundary: they guarantee the exactness constraints don't leak out the ends of the ladder.
Key idea of the proof: the diagram chase
The mechanism is a diagram chase. In R-modules you argue with elements (in a general abelian category, either embed via Freyd–Mitchell or chase with 'members'/generalized elements). Prove surjectivity of f₃: take b₃ ∈ B₃. Push to b₄ = (B₃→B₄)(b₃); since f₄ is iso, lift to a₄ ∈ A₄. Show a₄ maps to 0 in A₅ using that f₅ is injective and commutativity, so by exactness a₄ = (A₃→A₄)(a₃) for some a₃. Now f₃(a₃) and b₃ agree after mapping to B₄, so their difference comes from B₂ (exactness); pull that back through f₂ (surjective) into A₂, push into A₃, and correct a₃. Then f₃(corrected a₃) = b₃. Injectivity is the dual chase using f₂ mono, f₁ surjective, f₄ mono. Each step is forced: at every juncture exactness supplies an element and a vertical iso identifies it. The chase only ever uses f₂, f₄ isos, f₁ epi, f₅ mono — the sharp hypotheses.
Canonical special case: the Short Five Lemma
The cleanest instance takes both rows to be short exact sequences:
0 → A → B → C → 0
↓α ↓β ↓γ
0 → A' → B' → C' → 0
Here the outer zeros act as f₁ and f₅ (automatically isos), so the Five Lemma reduces to: if α and γ are isomorphisms, so is β. Concretely, suppose 0 → ℤ →×2→ ℤ → ℤ/2 → 0 and you compare it with an isomorphic presentation of an extension of ℤ/2 by ℤ; matching the sub (α on ℤ) and quotient (γ on ℤ/2) as isomorphisms forces the total object B ≅ B'. This is exactly how one proves that two extensions with the same sub and quotient, fitting in a commuting ladder, are isomorphic extensions — the backbone of the theory of Ext and the classification of group/module extensions.
Why the hypotheses matter — and what breaks
Every hypothesis is load-bearing, and the Four Lemma shows exactly how much each buys. To get f₃ injective you need f₂ injective, f₄ injective, and f₁ surjective. To get f₃ surjective you need f₂ surjective, f₄ surjective, and f₅ injective. Drop even one and it fails. Example: take the ladder with top row 0 → 0 → ℤ →id→ ℤ → 0 and bottom row 0 → ℤ →×0→ ℤ →id→ ℤ; if you break exactness or make f₁ merely injective (not surjective) the chase stalls and f₃ can fail to be mono. A vivid failure: two short exact sequences 0→ℤ→ℤ→ℤ/2→0 and 0→ℤ→ℤ→0→0 with f₁ = id on ℤ but f₃ (the quotients ℤ/2 → 0) not an iso — the middle ℤ→ℤ (multiplication by 2) is not surjective, precisely because the right-hand vertical map failed to be an iso. Commutativity is equally essential: without it the two 'coordinates' of a₃ aren't faithfully tracked and the argument collapses.
Why it matters: the engine of homological algebra
The Five Lemma is the tool that turns local isomorphism data into global equivalence. Its headline application is the comparison of homology/cohomology theories: given a long exact sequence (Mayer–Vietoris, the LES of a pair, the snake-lemma sequence), if a natural transformation of theories is an iso on all but one term, the Five Lemma promotes it to an iso there too — this is how Eilenberg–Steenrod-style uniqueness proofs proceed, and how one shows singular and simplicial homology agree. It underlies the proof that a chain map inducing isomorphisms on homology in a five-term window is invertible, feeds the theory of derived functors (Ext, Tor), spectral-sequence comparison theorems, and the '2-out-of-3' style arguments in model categories. First systematized in the 1940s–50s by Eilenberg and Steenrod and codified in Cartan–Eilenberg's Homological Algebra (1956) and Mac Lane's Homology (1963), it is, alongside the Snake Lemma, the first serious theorem every student of the subject internalizes.
| Variant | Hypotheses on f₁…f₅ | Conclusion on f₃ |
|---|---|---|
| Five Lemma | f₁,f₂,f₄,f₅ isomorphisms | f₃ isomorphism |
| Four Lemma (mono half) | f₂ mono, f₁ epi, f₄ mono | f₃ mono (injective) |
| Four Lemma (epi half) | f₄ epi, f₅ mono, f₂ epi | f₃ epi (surjective) |
| Short Five Lemma | 0→A→B→C→0 rows, outer two isos | middle iso |
| Sharp Five Lemma | f₁ epi, f₂ f₄ isos, f₅ mono | f₃ isomorphism |
Frequently asked questions
Do all four outer maps really need to be isomorphisms?
No — that is the point of the sharp form. The chase only uses f₂ and f₄ as full isomorphisms, f₁ as an epimorphism (surjection), and f₅ as a monomorphism (injection). So 'f₁ epi, f₂ iso, f₄ iso, f₅ mono' already forces f₃ to be an isomorphism. The symmetric 'all four are isos' version is just the memorable packaging.
Why does the conclusion land on the MIDDLE map and not an end?
Because exactness pins an element of the central object A₃ down by two independent coordinates: its image in A₄ and its relationship to A₂. The maps f₂ and f₄ translate exactly those two coordinates into the B-row. The middle object is the one fully surrounded by controlled neighbours, so it is the one forced to match; the ends only have a neighbour on one side and cannot be constrained this way.
How do you chase a diagram in an abelian category with no elements?
Two standard routes. First, the Freyd–Mitchell embedding theorem embeds any small abelian category fully-faithfully and exactly into R-Mod, so element chases are legitimate. Second, one uses 'members' (generalized elements), à la Mac Lane, where you argue with maps into objects and manipulate images/kernels arrow-theoretically. Both reproduce the same proof; the element language is just the most readable.
What is the difference between the Five Lemma and the Four Lemma?
The Four Lemma is the two asymmetric halves. One version ('surjective Four Lemma') concludes f₃ is epi from f₂,f₄ epi and f₅ mono; the dual concludes f₃ is mono from f₂,f₄ mono and f₁ epi. Composing the two mono/epi conclusions gives the full Five Lemma. Splitting this way reveals exactly which hypothesis each half of the conclusion depends on.
Does it hold outside abelian categories?
The proof needs kernels, cokernels, images, and the notion of exactness — i.e. an abelian (or at least suitable exact/Puppe-exact) category. It holds in R-modules, sheaves of abelian groups, chain complexes, and any abelian category. It also holds, more generally, in the category of all groups (with exactness taken in the sense of normal images) and in other semi-abelian / homological categories such as rings and Lie algebras: the same diagram chase goes through unchanged, and the Short Five Lemma for groups is a classical result used throughout extension theory and group cohomology. What genuinely requires the abelian setting is the free use of additive structure (subtracting parallel maps) in the most elementary form of the chase, but the theorem itself remains true for groups — no weakening is needed.
What famous theorem depends on the Five Lemma?
The uniqueness half of the Eilenberg–Steenrod axioms: any two homology theories satisfying the axioms agree on finite CW complexes, proved by inducting up skeleta and invoking the Five Lemma on the long exact sequences of pairs. It is also the key step showing a quasi-isomorphism-like map is an equivalence, and it powers the comparison theorem for spectral sequences and the classification of extensions via Ext.