# Orientability

In mathematics, **orientability** is a property of some topological spaces such as real vector spaces, Euclidean spaces, surfaces, and more generally manifolds that allows a consistent definition of "clockwise" and "counterclockwise".^{[1]} A space is **orientable** if such a consistent definition exists. In this case, there are two possible definitions, and a choice between them is an **orientation** of the space. Real vector spaces, Euclidean spaces, and spheres are orientable. A space is **non-orientable** if "clockwise" is changed into "counterclockwise" after running through some loops in it, and coming back to the starting point. This means that a geometric shape, such as , that moves continuously along such a loop is changed in its own mirror image . A Möbius strip is an example of a non-orientable space.

Various equivalent formulations of orientability can be given, depending on the desired application and level of generality. Formulations applicable to general topological manifolds often employ methods of homology theory, whereas for differentiable manifolds more structure is present, allowing a formulation in terms of differential forms. A generalization of the notion of orientability of a space is that of orientability of a family of spaces parameterized by some other space (a fiber bundle) for which an orientation must be selected in each of the spaces which varies continuously with respect to changes in the parameter values.

A surface *S* in the Euclidean space **R**^{3} is orientable if a two-dimensional figure (for example, ) cannot be moved around the surface and back to where it started so that it looks like its own mirror image (). Otherwise the surface is **non-orientable**. An abstract surface (i.e., a two-dimensional manifold) is orientable if a consistent concept of clockwise rotation can be defined on the surface in a continuous manner. That is to say that a loop going around one way on the surface can never be continuously deformed (without overlapping itself) to a loop going around the opposite way. This turns out to be equivalent to the question of whether the surface contains no subset that is homeomorphic to the Möbius strip. Thus, for surfaces, the Möbius strip may be considered the source of all non-orientability.

For an orientable surface, a consistent choice of "clockwise" (as opposed to counter-clockwise) is called an **orientation**, and the surface is called **oriented**. For surfaces embedded in Euclidean space, an orientation is specified by the choice of a continuously varying surface normal **n** at every point. If such a normal exists at all, then there are always two ways to select it: **n** or −**n**. More generally, an orientable surface admits exactly two orientations, and the distinction between an orient*ed* surface and an orient*able* surface is subtle and frequently blurred. An orientable surface is an abstract surface that admits an orientation, while an oriented surface is a surface that is abstractly orientable, and has the additional datum of a choice of one of the two possible orientations.

Most surfaces we encounter in the physical world are orientable. Spheres, planes, and tori are orientable, for example. But Möbius strips, real projective planes, and Klein bottles are non-orientable. They, as visualized in 3-dimensions, all have just one side. The real projective plane and Klein bottle cannot be embedded in **R**^{3}, only immersed with nice intersections.

Note that locally an embedded surface always has two sides, so a near-sighted ant crawling on a one-sided surface would think there is an "other side". The essence of one-sidedness is that the ant can crawl from one side of the surface to the "other" without going through the surface or flipping over an edge, but simply by crawling far enough.

In general, the property of being orientable is not equivalent to being two-sided; however, this holds when the ambient space (such as **R**^{3} above) is orientable. For example, a torus embedded in

Any surface has a triangulation: a decomposition into triangles such that each edge on a triangle is glued to at most one other edge. Each triangle is oriented by choosing a direction around the perimeter of the triangle, associating a direction to each edge of the triangle. If this is done in such a way that, when glued together, neighboring edges are pointing in the opposite direction, then this determines an orientation of the surface. Such a choice is only possible if the surface is orientable, and in this case there are exactly two different orientations.

If the figure can be consistently positioned at all points of the surface without turning into its mirror image, then this will induce an orientation in the above sense on each of the triangles of the triangulation by selecting the direction of each of the triangles based on the order red-green-blue of colors of any of the figures in the interior of the triangle.

This approach generalizes to any *n*-manifold having a triangulation. However, some 4-manifolds do not have a triangulation, and in general for *n* > 4 some *n*-manifolds have triangulations that are inequivalent.

If *H*_{1}(*S*) denotes the first homology group of a surface *S*, then *S* is orientable if and only if *H*_{1}(*S*) has a trivial torsion subgroup. More precisely, if *S* is orientable then *H*_{1}(*S*) is a free abelian group, and if not then *H*_{1}(*S*) = *F* + **Z**/2**Z** where *F* is free abelian, and the **Z**/2**Z** factor is generated by the middle curve in a Möbius band embedded in *S*.

Let *M* be a connected topological *n*-manifold. There are several possible definitions of what it means for *M* to be orientable. Some of these definitions require that *M* has extra structure, like being differentiable. Occasionally, *n* = 0 must be made into a special case. When more than one of these definitions applies to *M*, then *M* is orientable under one definition if and only if it is orientable under the others.^{[2]}^{[3]}

Orientability and orientations can also be expressed in terms of the tangent bundle. The tangent bundle is a vector bundle, so it is a fiber bundle with structure group GL(*n*, **R**). That is, the transition functions of the manifold induce transition functions on the tangent bundle which are fiberwise linear transformations. If the structure group can be reduced to the group GL^{+}(*n*, **R**) of positive determinant matrices, or equivalently if there exists an atlas whose transition functions determine an orientation preserving linear transformation on each tangent space, then the manifold *M* is orientable. Conversely, *M* is orientable if and only if the structure group of the tangent bundle can be reduced in this way. Similar observations can be made for the frame bundle.

Another way to define orientations on a differentiable manifold is through volume forms. A volume form is a nowhere vanishing section *ω* of ⋀^{n} *T*^{∗}*M*, the top exterior power of the cotangent bundle of *M*. For example, **R**^{n} has a standard volume form given by *dx*^{1} ∧ ⋯ ∧ *dx*^{n}. Given a volume form on *M*, the collection of all charts *U* → **R**^{n} for which the standard volume form pulls back to a positive multiple of *ω* is an oriented atlas. The existence of a volume form is therefore equivalent to orientability of the manifold.

Volume forms and tangent vectors can be combined to give yet another description of orientability. If *X*_{1}, …, *X*_{n} is a basis of tangent vectors at a point *p*, then the basis is said to be **right-handed** if ω(*X*_{1}, …, *X*_{n}) > 0. A transition function is orientation preserving if and only if it sends right-handed bases to right-handed bases. The existence of a volume form implies a reduction of the structure group of the tangent bundle or the frame bundle to GL^{+}(*n*, **R**). As before, this implies the orientability of *M*. Conversely, if *M* is orientable, then local volume forms can be patched together to create a global volume form, orientability being necessary to ensure that the global form is nowhere vanishing.

At the heart of all the above definitions of orientability of a differentiable manifold is the notion of an orientation preserving transition function. This raises the question of what exactly such transition functions are preserving. They cannot be preserving an orientation of the manifold because an orientation of the manifold is an atlas, and it makes no sense to say that a transition function preserves or does not preserve an atlas of which it is a member.

This question can be resolved by defining local orientations. On a one-dimensional manifold, a local orientation around a point *p* corresponds to a choice of left and right near that point. On a two-dimensional manifold, it corresponds to a choice of clockwise and counter-clockwise. These two situations share the common feature that they are described in terms of top-dimensional behavior near *p* but not at *p*. For the general case, let *M* be a topological *n*-manifold. A **local orientation** of *M* around a point *p* is a choice of generator of the group

Around each point of *M* there are two local orientations. Intuitively, there is a way to move from a local orientation at a point *p* to a local orientation at a nearby point *p*′: when the two points lie in the same coordinate chart *U* → **R**^{n}, that coordinate chart defines compatible local orientations at *p* and *p*′. The set of local orientations can therefore be given a topology, and this topology makes it into a manifold.

There is a canonical map π : *O* → *M* that sends a local orientation at *p* to *p*. It is clear that every point of *M* has precisely two preimages under π. In fact, π is even a local homeomorphism, because the preimages of the open sets *U* mentioned above are homeomorphic to the disjoint union of two copies of *U*. If *M* is orientable, then *M* itself is one of these open sets, so *O* is the disjoint union of two copies of *M*. If *M* is non-orientable, however, then *O* is connected and orientable. The manifold *O* is called the **orientation double cover**.

If *M* is a manifold with boundary, then an orientation of *M* is defined to be an orientation of its interior. Such an orientation induces an orientation of ∂*M*. Indeed, suppose that an orientation of *M* is fixed. Let *U* → **R**^{n}_{+} be a chart at a boundary point of *M* which, when restricted to the interior of *M*, is in the chosen oriented atlas. The restriction of this chart to ∂*M* is a chart of ∂*M*. Such charts form an oriented atlas for ∂*M*.

When *M* is smooth, at each point *p* of ∂*M*, the restriction of the tangent bundle of *M* to ∂*M* is isomorphic to *T*_{p}∂*M* ⊕ **R**, where the factor of **R** is described by the inward pointing normal vector. The orientation of *T*_{p}∂*M* is defined by the condition that a basis of *T*_{p}∂*M* is positively oriented if and only if it, when combined with the inward pointing normal vector, defines a positively oriented basis of *T*_{p}*M*.

A closely related notion uses the idea of covering space. For a connected manifold *M* take *M*^{∗}, the set of pairs (*x*, o) where *x* is a point of *M* and *o* is an orientation at *x*; here we assume *M* is either smooth so we can choose an orientation on the tangent space at a point or we use singular homology to define orientation. Then for every open, oriented subset of *M* we consider the corresponding set of pairs and define that to be an open set of *M*^{∗}. This gives *M*^{∗} a topology and the projection sending (*x*, o) to *x* is then a 2-to-1 covering map. This covering space is called the **orientable double cover**, as it is orientable. *M*^{∗} is connected if and only if *M* is not orientable.

Another way to construct this cover is to divide the loops based at a basepoint into either orientation-preserving or orientation-reversing loops. The orientation preserving loops generate a subgroup of the fundamental group which is either the whole group or of index two. In the latter case (which means there is an orientation-reversing path), the subgroup corresponds to a connected double covering; this cover is orientable by construction. In the former case, one can simply take two copies of *M*, each of which corresponds to a different orientation.

In Lorentzian geometry, there are two kinds of orientability: space orientability and time orientability. These play a role in the causal structure of spacetime.^{[6]} In the context of general relativity, a spacetime manifold is space orientable if, whenever two right-handed observers head off in rocket ships starting at the same spacetime point, and then meet again at another point, they remain right-handed with respect to one another. If a spacetime is time-orientable then the two observers will always agree on the direction of time at both points of their meeting. In fact, a spacetime is time-orientable if and only if any two observers can agree which of the two meetings preceded the other.^{[7]}

Formally, the pseudo-orthogonal group O(*p*,*q*) has a pair of characters: the space orientation character σ_{+} and the time orientation character σ_{−},

Their product σ = σ_{+}σ_{−} is the determinant, which gives the orientation character. A space-orientation of a pseudo-Riemannian manifold is identified with a section of the associated bundle

where O(*M*) is the bundle of pseudo-orthogonal frames. Similarly, a time orientation is a section of the associated bundle