For a real or complex vector space \({V}\), we can generalize another Cartesian structure, the **inner product** (AKA scalar product, dot product). We define an **inner product space** as including a mapping from vectors to scalars denoted \({\left\langle v,w\right\rangle }\) (also denoted \({(v,w)}\) or \({v\cdot w}\)). The mapping must satisfy:

- \({\left\langle v,w\right\rangle =\left\langle w,v\right\rangle ^{*}}\) where \({^{*}}\) denotes complex conjugation
- \({\left\langle ax+y,bz\right\rangle =a^{*}b\left\langle x,z\right\rangle +b\left\langle y,z\right\rangle }\)
- \({\left\langle v,v\right\rangle >0}\) except if \({v=0}\), in which case it vanishes

The first requirement implies that \({\left\langle v,v\right\rangle }\) is real, and that the inner product is symmetric for real scalars. The second requirement can be phrased as saying that the inner product is **anti-linear** (AKA conjugate-linear) in its first argument and linear in its second, or **sesquilinear**, and the first and second requirements together define a **Hermitian form**. A real inner product is then **bilinear** or **multilinear**, meaning linear in each argument. The third requirement above makes the inner product **positive definite**.

Δ Sometimes the definitions of both inner product and sesquilinear are reversed to make the second argument anti-linear instead of the first. This is sometimes called the “mathematics” convention, while ours would then be the “physics” convention. |

Δ An inner product as we have defined it on a complex vector space is also called a Hermitian inner product, and a complex inner product space is sometimes called a Hermitian inner product space, Hermitian space, or unitary space. |

Two vectors are defined to be **orthogonal** if their inner product vanishes. The **orthogonal complement** of a subspace \({W}\) of \({V}\) is the subspace of all vectors orthogonal to every vector in \({W}\), i.e. \({W^{\perp}\equiv\left\{ v\in V\mid\forall w\in W\left\langle v,w\right\rangle =0\right\} }\). A basis of \({W}\) together with a basis for its orthogonal complement \({W^{\perp}}\) forms a basis for all of \({V}\).

An inner product defines a special class of bases, the **orthonormal bases** \({\hat{e}_{\mu}}\) with \({\left\langle \hat{e}_{\mu},\hat{e}_{\nu}\right\rangle =\delta_{\mu\nu}}\) (\({\equiv1}\) if \({\mu=\nu}\), \({0}\) otherwise). If we then write \({v=v^{\mu}\hat{e}_{\mu}}\) and \({w=w^{\mu}\hat{e}_{\mu}}\), we have

\(\displaystyle \left\langle v,w\right\rangle =\sum_{\mu}v^{\mu*}w^{\mu}=v^{\dagger}w, \)

where in the first expression we take the complex conjugate of the components \({v^{\mu}}\), and the second is common in linear algebra, where we treat the vectors as column matrices of components, and the inner product is formed by matrix multiplication after taking the adjoint (hermitian conjugate) of the first matrix. Thus every finite-dimensional real or complex inner product space is isomorphic to \({\mathbb{R}^{n}}\) or \({\mathbb{C}^{n}}\) with their standard component inner products as above.

The positive definite definition of a real inner product is sometimes relaxed to only require the product to be **nondegenerate** (AKA anisotropic), i.e. \({\left\langle v,w\right\rangle =0}\) for all \({w}\) only if \({v=0}\). We will instead refer to this type of form, a nondegenerate symmetric bilinear form, as a **pseudo inner product** (AKA pseudo-metric); it is characterized by the fact that \({\left\langle v,v\right\rangle }\) can be negative or vanish. A vector \({v}\) is called **isotropic** (AKA light-like) if \({\left\langle v,v\right\rangle =0}\). In the context of manifolds, an inner product is called a **metric**; and in the context of spacetime, variants of the inner product carry specific terminology. We will cover this terminology in the following.

A pseudo inner product does not yield a well-defined norm, but the (not necessarily real) quantity \({\sqrt{\left\langle v,v\right\rangle }}\) is nevertheless sometimes called the “length” of \({v}\). A pseudo inner product also defines orthonormal bases, with the definition modified to allow \({\left\langle \hat{e}_{\mu},\hat{e}_{\nu}\right\rangle =\pm\delta^{\mu}{}_{\nu}}\). The number of positive and negative values turns out to be independent of the choice of orthonormal basis, and this pair of integers \({\left(r,s\right)}\) is called the **signature**. We define \({\eta{}_{\mu\nu}\equiv \left\langle \hat{e}_{\mu},\hat{e}_{\nu}\right\rangle \equiv\eta^{\mu\nu}}\), so that it is \({\pm1}\) if \({\mu=\nu}\) (with \({r}\) positive values), \({0}\) otherwise. A general signature is called **pseudo-Riemannian** (AKA pseudo-Euclidean), a signature with \({s=0}\) is called **Riemannian** (AKA Euclidean), a signature with \({s=1}\) (or sometimes \({r=1}\)) is called **Lorentzian** (AKA Minkowskian), and the signature (3,1) is called **Minkowskian** (more specifically this is called the **“mostly pluses” signature** (AKA relativity, spacelike, or east coast signature), while the signature (1,3), the **“mostly minuses” signature** (AKA particle physics, timelike, or west coast signature), is also called Minkowskian).

Δ The term “signature” also sometimes refers to the integer \({r-s}\). |

Δ For an arbitrary basis, the number of positive and negative values of \({\left\langle e_{\mu},e_{\nu}\right\rangle}\) do not necessarily match the signature. |