The word dimension in mathematics has many definitions, but all of them are trying to quantify our intuition that, for example, a sheet of paper has somehow one less dimension than a stack of papers.

One common way to define dimension is through some notion of a number of independent quantities needed to describe an element of an object. For example, it is natural to say that the sheet of paper is two-dimensional because one needs two real numbers to specify a position on the sheet, whereas the stack of papers is three-dimension because a position in a stack is specified by a sheet and a position on the sheet. Following this notion, in linear algebra the dimension of a vector space is defined as the minimal number of vectors such that every other vector in the vector space is representable as a sum of these. Similarly, the word rank denotes various dimension-like invariants that appear throughout the algebra.

However, if we try to generalize this notion to the mathematical objects that do not possess an algebraic structure, then we run into a difficulty. From the point of view of set theory there are as many real numbers as pairs of real numbers since there is a bijection from real numbers to pairs of real numbers. To distinguish a plane from a cube one needs to impose restrictions on the kind of mapping. Surprisingly, it turns out that the continuity is not enough as was pointed out by Peano. There are continuous functions that map a square onto a cube. So, in topology one uses another intuitive notion that in a high-dimensional space there are more directions than in a low-dimensional. Hence, the (Lebesgue covering) dimension of a topological space is defined as the smallest number $d$ such that every covering of the space by open sets can be refined so that no point is contained in more than $d+1$ sets. For example, no matter how one covers a sheet of paper by sufficiently small other sheets of paper such that two sheets can overlap each other, but cannot merely touch, one will always find a point that is covered by $2+1=3$ sheets.

Another definition of dimension rests on the idea that higher-dimensional objects are in some sense larger than the lower-dimensional ones. For example, to cover a cube with a side length $2$ one needs at least $2^3=8$ cubes with a side length $1$ , but a square with a side length $2$ can be covered by only $2^2=4$ unit squares. Let $N(\epsilon)$ be the minimal number of open balls in any covering of a bounded set $S$ by balls of radius $\epsilon$ . The Besicovitch-Hausdorff dimension of $S$ is defined as $-\lim_{\epsilon\to 0} \log_\epsilon N(\epsilon)$ . The Besicovitch-Hausdorff dimension is not always defined, and when defined it might be non-integral.