In mathematics, **4-manifold** is a 4-dimensional topological manifold. A **smooth 4-manifold** is a 4-manifold with a smooth structure. In dimension four, in marked contrast with lower dimensions, topological and smooth manifolds are quite different. There exist some topological 4-manifolds which admit no smooth structure and even if there exists a smooth structure it need not be unique (i.e. there are smooth 4-manifolds which are homeomorphic but not diffeomorphic).

4-manifolds are of importance in physics because, in General Relativity, spacetime is modeled as a pseudo-Riemannian 4-manifold.

The homotopy type of a simply connected compact 4-manifold only depends on the intersection form on the middle dimensional homology. A famous theorem of M. Freedman (1982) implies that the homeomorphism type of the manifold only depends on this intersection form, and on a **Z**/2**Z** invariant called the Kirby–Siebenmann invariant, and moreover that every combination of unimodular form and Kirby–Siebenmann invariant can arise, except that if the form is even then the Kirby–Siebenmann invariant must be the signature/8 (mod 2).

Examples:

In the special case when the form is 0, this implies the 4-dimensional topological Poincaré conjecture.
If the form is *E*_{8}, this gives a manifold called the E8 manifold, a manifold not homeomorphic to any simplicial complex.
If the form is **Z**, there are two manifolds depending on the Kirby–Siebenmann invariant: one is 2-dimensional complex projective space, and the other is a fake projective space, with the same homotopy type but not homeomorphic (and with no smooth structure).
When the rank of the form is greater than about 28, the number of positive definite unimodular forms starts to increase extremely rapidly with the rank, so there are huge numbers of corresponding simply connected topological 4-manifolds (most of which seem to be of almost no interest).
Freedman's classification can be extended to some cases when the fundamental group is not too complicated; for example, when it is **Z** there is a classification similar to the one above using Hermitian forms over the group ring of **Z**. If the fundamental group is too large (for example, a free group on 2 generators) then Freedman's techniques seem to fail and very little is known about such manifolds.

For any finitely presented group it is easy to construct a (smooth) compact 4-manifold with it as its fundamental group. As there is no algorithm to tell whether two finitely presented groups are isomorphic (even if one is known to be trivial) there is no algorithm to tell if two 4-manifolds have the same fundamental group. This is one reason why much of the work on 4-manifolds just considers the simply connected case: the general case of many problems is already known to be intractable.

For manifolds of dimension at most 6, any piecewise linear (PL) structure can be smoothed in an essentially unique way, so in particular the theory of 4 dimensional PL manifolds is much the same as the theory of 4 dimensional smooth manifolds.

A major open problem in the theory of smooth 4-manifolds is to classify the simply connected compact ones. As the topological ones are known, this breaks up into two parts:

- Which topological manifolds are smoothable?
- Classify the different smooth structures on a smoothable manifold.

There is an almost complete answer to the first problem of which simply connected compact 4-manifolds have smooth structures. First, the Kirby–Siebenmann class must vanish.

If the intersection form is definite Donaldson's theorem (Donaldson 1983) gives a complete answer: there is a smooth structure if and only if the form is diagonalizable.
If the form is indefinite and odd there is a smooth structure.
If the form is indefinite and even we may as well assume that it is of nonpositive signature by changing orientations if necessary, in which case it is isomorphic to a sum of *m* copies of II_{1,1} and 2*n* copies of E_{8}(−1) for some *m* and *n*. If *m* ≥ 3*n* (so that the dimension is at least 11/8 times the |signature|) then there is a smooth structure, given by taking a connected sum of *n* K3 surfaces and *m* − 3*n* copies of *S*^{2}×*S*^{2}. If *m* ≤ 2*n* (so the dimension is at most 10/8 times the |signature|) then Furuta proved that no smooth structure exists (Furuta 2001). This leaves a small gap between 10/8 and 11/8 where the answer is mostly unknown. (The smallest case not covered above has *n*=2 and *m*=5, but this has also been ruled out, so the smallest lattice for which the answer is not currently known is the lattice II_{7,55} of rank 62 with *n*=3 and *m*=7.) The "11/8 conjecture" states that smooth structures do not exist if the dimension is less than 11/8 times the |signature|.
In contrast, very little is known about the second question of classifying the smooth structures on a smoothable 4-manifold; in fact, there is not a single smoothable 4-manifold where the answer is known. Donaldson showed that there are some simply connected compact 4-manifolds, such as Dolgachev surfaces, with a countably infinite number of different smooth structures. There are an uncountable number of different smooth structures on **R**^{4}; see exotic **R**^{4}. Fintushel and Stern showed how to use surgery to construct large numbers of different smooth structures (indexed by arbitrary integral polynomials) on many different manifolds, using Seiberg–Witten invariants to show that the smooth structures are different. Their results suggest that any classification of simply connected smooth 4-manifolds will be very complicated. There are currently no plausible conjectures about what this classification might look like. (Some early conjectures that all simply connected smooth 4-manifolds might be connected sums of algebraic surfaces, or symplectic manifolds, possibly with orientations reversed, have been disproved.)

There are several fundamental theorems about manifolds that can be proved by low-dimensional methods in dimensions at most 3, and by completely different high-dimensional methods in dimension at least 5, but which are false in dimension 4. Here are some examples:

In dimensions other than 4, the Kirby–Siebenmann invariant provides the obstruction to the existence of a PL structure; in other words a compact topological manifold has a PL structure if and only if its Kirby–Siebenmann invariant in H^{4}(*M*,**Z**/2**Z**) vanishes. In dimension 3 and lower, every topological manifold admits an essentially unique PL structure. In dimension 4 there are many examples with vanishing Kirby–Siebenmann invariant but no PL structure.
In any dimension other than 4, a compact topological manifold has only a finite number of essentially distinct PL or smooth structures. In dimension 4, compact manifolds can have a countable infinite number of non-diffeomorphic smooth structures.
Four is the only dimension *n* for which **R**^{n} can have an exotic smooth structure. **R**^{4} has an uncountable number of exotic smooth structures; see exotic **R**^{4}.
The solution to the smooth Poincaré conjecture is known in all dimensions other than 4 (it is usually false in dimensions at least 7; see exotic sphere). The Poincaré conjecture for PL manifolds has been proved for all dimensions other than 4, but it is not known whether it is true in 4 dimensions (it is equivalent to the smooth Poincaré conjecture in 4 dimensions).
The smooth h-cobordism theorem holds for cobordisms provided that neither the cobordism nor its boundary has dimension 4. It can fail if the boundary of the cobordism has dimension 4 (as shown by Donaldson). If the cobordism has dimension 4, then it is unknown whether the h-cobordism theorem holds.
A topological manifold of dimension not equal to 4 has a handlebody decomposition. Manifolds of dimension 4 have a handlebody decomposition if and only if they are smoothable.
There are compact 4-dimensional topological manifolds that are not homeomorphic to any simplicial complex. In dimension at least 5 the existence of topological manifolds not homeomorphic to a simplicial complex was an open problem. In 2013 Ciprian Manolescu posted a preprint on the ArXiv showing that there are manifolds in each dimension greater than or equal to 5, that are not homeomorphic to a simplicial complex.
According to Frank Quinn, "Two *n*-dimensional submanifolds of a manifold of dimension 2*n* will usually intersect themselves and each other in isolated points. The "Whitney trick" uses an isotopy across an embedded 2-disk to simplify these intersections. Roughly speaking this reduces the study of *n*-dimensional embeddings to embeddings of 2-disks. But this not a reduction when the embedding is 4: the 2 disks themselves are middle-dimensional, so trying to embed them encounters exactly the same problems they are supposed to solve. This is the phenomenon that separates dimension 4 from others."