Definition

Let MM be a with separable predual and let P(M)\mathcal P(M) be its projections. A dimension function on projections is a map

d:P(M)[0,]d:\mathcal P(M)\longrightarrow[0,\infty]

that is invariant under , additive on , faithful (d(p)=0p=0)(d(p)=0\Rightarrow p=0), and normal:

pipd(pi)d(p).p_i\uparrow p\quad\Longrightarrow\quad d(p_i)\uparrow d(p).

In a finite factor one requires d(1)=1d(1)=1. In an infinite semifinite factor a choice of scale is required; in a it is customary to give every dimension 11. These normalizations make dimension an extension of matrix rank or normalized rank.

Dimension scales and factor types

With the standard normalization, the possible finite values distinguish the factor types. They are {0,1/n,,1}\{0,1/n,\ldots,1\} for type In\mathrm I_n, [0,1][0,1] for type II1\mathrm{II}_1, {0,1,2,}\{0,1,2,\ldots\} for type I\mathrm I_\infty, and [0,)[0,\infty) for type II\mathrm{II}_\infty. For a the only scalar dimension function has values 00 and \infty. This is the continuous-dimension theory underlying the classification of factors Takesaki, Chapter V, §§1–2.

Finite factors and traces

If MM is finite, its unique normalized center-valued trace is scalar because MM is a factor. Its restriction to projections is the normalized dimension function. In a type II1\mathrm{II}_1 factor, this gives

pqd(p)=d(q),p\sim q\quad\Longleftrightarrow\quad d(p)=d(q),

and every value in [0,1][0,1] occurs. Thus dimension is a complete invariant of projection equivalence in this setting, not merely a numerical estimate.

Conventions and scope
References
  1. Masamichi Takesaki, Theory of Operator Algebras I, Springer, 1979. DOI record. Relevant: Chapter V, §§1–2 on equivalence, comparison, dimension, and factor types.
  2. Richard V. Kadison and John R. Ringrose, Fundamentals of the Theory of Operator Algebras, vol. II, American Mathematical Society, 1997. DOI record. Relevant: Chapter 6 on comparison theory and dimension of projections.