Abstract index notation is a mathematical notation for tensors and spinors that uses indices to indicate their types, rather than their components in a particular basis. The indices are mere placeholders, not related to any fixed basis, and in particular are non-numerical. The notation was introduced by Roger Penrose as a way to use the formal aspects of the Einstein summation convention in order to compensate for the difficulty in describing contractions and covariant differentiation in modern abstract tensor notation, while preserving the explicit covariance of the expressions involved.

Let V be a vector space, and V its dual. Consider, for example, a rank-2 covariant tensor . Then h can be identified with a bilinear form on V. In other words, it is a function of two arguments in V which can be represented as a pair of slots:

Abstract index notation is merely a labelling of the slots by Latin letters, which have no significance apart from their designation as labels of the slots (i.e., they are non-numerical):

A contraction between two tensors is represented by the repetition of an index label, where one label is contravariant (an upper index corresponding to a tensor in V) and one label is covariant (a lower index corresponding to a tensor in V). Thus, for instance,

is the trace of a tensor t = t_ab over its last two slots. This manner of representing tensor contractions by repeated indices is formally similar to the Einstein summation convention. However, as the indices are non-numerical, it does not imply summation: rather it corresponds to the abstract basis-independent trace operation (or duality pairing) between tensor factors of type V and those of type V.