Projections, traces, and factor types
Compare projections by partial isometries and use minimality, finiteness, and normal traces to distinguish factor types.
On this page
The question for today
What you’ll understand
- Verify Murray–von Neumann equivalence using initial and final projections.
- Distinguish minimal projections from finite projections.
- Classify the stated matrix and shift examples without mistaking a normalized trace for type II₁.
A projection represents a yes-or-no event, but operator algebras compare such events by more than their numerical probabilities. A partial isometry can identify their ranges internally, producing a noncommutative notion of equal size. Factor types organize what kinds of projection sizes and traces are possible. Throughout this lesson M is a complex von Neumann algebra, and a factor means its center is exactly the scalar multiples of its identity. A general algebra with a larger center requires a decomposition rather than one factor label.
Partial isometries implement internal equivalence
A projection p satisfies p-star=p=p squared. Projections p and q are Murray–von Neumann equivalent when a partial isometry v inside M has v-star v=p and vv-star=q. In matrices this means equal rank, but the definition works without counting a finite basis. The membership of v in M matters: equivalence in a larger ambient algebra is not automatically the same internal relation. A faithful normal trace takes equal values on equivalent projections by its cyclic property.
A projection is finite if it is not equivalent to any proper subprojection of itself. This is not the same as being finite rank in every representation. A projection is minimal if it is nonzero and its corner pMp contains only scalar multiples of p. Matrix rank-one projections are minimal and finite. In a type II₁ factor every projection is finite, yet no nonzero projection is minimal. Keeping these concepts separate is essential before reading the type classification.
Traces measure sizes without choosing coordinates
A normal trace is positive, preserves increasing positive suprema, and is invariant under exchanging the order of products in its finite-valued setting. A finite factor has a unique normalized faithful normal tracial state. In M_n(C) this is ordinary matrix trace divided by n. Infinite semifinite algebras can instead have a faithful normal semifinite trace with infinite value on the identity. Semifinite means every nonzero positive element dominates some nonzero positive element of finite trace, after requiring faithfulness.
Type I factors are B(K) for a complex Hilbert space K, with minimal projections present. Finite-dimensional ones are type I_n; the separable infinite-dimensional example is type I-infinity. Type II factors have no nonzero minimal projections and admit a faithful normal semifinite trace. Type II₁ has finite identity and a normalized trace, while type II-infinity has infinite identity. Type III factors have no nonzero finite projections and admit no faithful normal semifinite trace. These are structural assertions, not labels determined just by Hilbert-space dimension.
Infinite projections do not imply type III
On ℓ² of the nonnegative integers, the unilateral shift S sends the jth basis vector to the next. It is an isometry, so S-star S is the identity, while SS-star misses the first basis vector. The identity is therefore equivalent to a proper subprojection and is infinite. Nevertheless B(ℓ²) contains rank-one minimal finite projections and has its ordinary semifinite trace. It is type I-infinity, not type III.
A type II₁ factor has a continuous range of projection traces from zero to one; those projections are not obtained by dividing a finite matrix rank by a fixed integer. One important example arises by taking the von Neumann closure of an increasing tensor product of two-by-two matrix algebras in its tracial GNS representation. This example is stated as a construction roadmap, not proved here. The difference between its norm completion and von Neumann closure again matters, as does the representation chosen for that closure.
Work it through
Equal-size projections in a type I₄ factor
In M₄(C), let p=diag(1,1,0,0), q=diag(1,0,1,0), and v=e₁₁+e₃₂. Check equivalence and normalized traces.
The adjoint is v*=e₁₁+e₂₃. Matrix-unit multiplication gives v*v=e₁₁+e₂₂=p.
Multiplying in the reverse order gives vv*=e₁₁+e₃₃=q. Thus v is a partial isometry implementing p~q inside the matrix algebra.
Both projections have rank two, so normalized trace is 2/4=1/2. A rank-one projection has trace 1/4 and cannot be equivalent to either, because a trace would have to take the same value.
The two different projections are internally equivalent and have the same trace. The ambient factor is type I₄, despite having a normalized faithful normal trace.
The living notebook
Explore the model
Interactive lab · probability
Trace as a projection probability
Pause, reset, then set p=0.5 and trials n=1. Predict the mean 0.5 and variance 0.25 for a yes-or-no measurement of either worked projection in the normalized trace state. The binomial instrument records only that probability, not projection rank or partial isometries; equal scalar probabilities in other settings do not by themselves prove internal equivalence.
- Mean · np
- 5
- Variance · np(1−p)
- 2.5
- Total probability
- 1
These are exact binomial probabilities for independent trials with one fixed success probability. Playback tours the distribution; it does not resample outcomes. Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.
Look forA trace gives a numerical shadow of projection size, while equivalence is implemented by an operator in the algebra.
Pause & reason
Check your understanding
Your turn
Try it yourself
Using the unilateral shift S, verify that I is infinite and that the first-coordinate projection p₀ is nevertheless finite and minimal in B(ℓ²).
Take a moment to make a prediction or work it out on paper. Open a hint when you need one.
Give me a hint
Compare S*S and SS*, then inspect the one-dimensional corner p₀B(ℓ²)p₀.
Compare with the solution
S*S=I and SS*=I−p₀, so S implements equivalence between I and its proper subprojection I−p₀. Thus I is infinite. The corner p₀B(ℓ²)p₀ is Cp₀ because its range is one-dimensional, so p₀ is minimal. It has no nonzero proper subprojection and cannot be equivalent to zero, making it finite. Infinite identity therefore coexists with finite nonzero projections.
Keep the idea
Take it with you
- Projection equivalence is implemented by a partial isometry inside the chosen algebra.
- Finite and minimal projections are distinct notions, especially in type II₁ factors.
- An infinite identity or a normalized trace alone does not determine a factor's type.
Read further
Original sources and deeper explanations to follow at your own pace.
Optional progress, saved in this browser when available. Revisit any lesson at any time.
Lesson not marked complete.