Gleason’s Theorem

Gleason’s theorem characterizes consistent probability assignments to all sharp quantum events. In a standard complex-Hilbert-space form:

Let $\mathcal H$ be a separable complex Hilbert space with $\dim\mathcal H\geq3$. Suppose

$$\mu:\mathcal P(\mathcal H)\longrightarrow[0,1]$$

assigns a number to every orthogonal projection, satisfies $\mu(I)=1$, and is countably additive on mutually orthogonal projections. Then there is a unique positive trace-class operator $\rho$ with $\operatorname{Tr}\rho=1$ such that

$$\mu(P)=\operatorname{Tr}(\rho P)$$

for every projection $P$.

In finite dimension, finite orthogonal additivity suffices. The original theorem also includes real Hilbert spaces under the corresponding dimensional hypothesis. The complex form is the one used for quantum mechanics here.

The theorem does not derive quantum theory from no assumptions. It begins with Hilbert-space projections, one context-independent value for each projection, normalization, and orthogonal additivity. Within that structure, the density-operator Born form is forced when the dimension is at least three.

Required background. Projection-Valued Measures supplies countable orthogonal additivity; Projectors supplies sharp-event geometry. Positive trace-class operators are assumed at an introductory level.

Helpful background. The Born Rule identifies the state-event trace pairing; Expectation Values connects event probabilities to averages of sharp observables.

Measures on projections

Let $\mathcal P(\mathcal H)$ denote the set of orthogonal projections on $\mathcal H$. A normalized probability measure on projections satisfies

$$\mu(0)=0, \qquad \mu(I)=1, \qquad \mu(P)\geq0.$$

If $P_jP_k=0$ for $j\ne k$, then their strong sum is a projection and

$$\mu\!\left(\sum_{j=1}^{\infty}P_j\right) = \sum_{j=1}^{\infty}\mu(P_j).$$

For a complete orthogonal resolution of the identity,

$$\sum_jP_j=I,$$

the assigned numbers form a normalized classical probability distribution:

$$\sum_j\mu(P_j)=1.$$

The important compatibility condition is that $\mu(P)$ depends only on the projector $P$, not on which complete PVM contains it. The same yes-no subspace can occur in several measurement contexts, and the theorem assumes one value for it in all of them. This is the precise noncontextuality built into the measure.

The theorem concerns probabilities, not predetermined outcomes. A value such as $\mu(P)=0.37$ is not a hidden truth value, and no assumption says every projector must be assigned zero or one.

Frame-function form

For a unit vector $e$, let $P_e=|e\rangle\langle e|$ and define

$$f(e)=\mu(P_e).$$

The function is phase invariant:

$$f(e^{i\chi}e)=f(e).$$

For every orthonormal basis $\{e_j\}$,

$$\sum_j f(e_j)=1.$$

Such an $f$ is a nonnegative frame function of weight one. Gleason’s central mathematical conclusion is that, in dimension at least three, every such function has the quadratic form

$$f(e)=\langle e|\rho e\rangle$$

for one positive trace-one operator $\rho$.

Once the values on rank-one projectors are known, additivity gives the values on finite-rank projections. Countable additivity and trace-class control then extend the formula to arbitrary projections in the separable infinite-dimensional case.

Why the trace form is a measure

Let $\rho\geq0$ be trace class with $\operatorname{Tr}\rho=1$. Then

$$\mu_\rho(P)=\operatorname{Tr}(\rho P)$$

has every required property. Positivity follows because

$$\operatorname{Tr}(\rho P) = \operatorname{Tr}(\rho^{1/2}P\rho^{1/2}) \geq0.$$

Normalization follows from $P=I$. If $P_j$ are mutually orthogonal, normality of the trace pairing gives

$$\operatorname{Tr}\!\left( \rho\sum_jP_j \right) = \sum_j\operatorname{Tr}(\rho P_j).$$

Thus the theorem’s difficult direction is the converse: every normalized additive measure must arise this way.

The density operator is unique. If $\rho$ and $\sigma$ give the same value on every rank-one projector, then

$$\langle e|(\rho-\sigma)e\rangle=0$$

for every unit vector $e$. Homogeneity extends this to all vectors, and the polarization identity recovers every matrix element of $\rho-\sigma$ from its quadratic form. Hence $\rho=\sigma$.

Proof architecture and boundary

Gleason’s original proof is not a one-line application of linear algebra. A modern proof architecture can be summarized as follows.

1. Reduce to rank-one projections. Encode the measure as a bounded nonnegative frame function $f$ with constant sum over every orthonormal basis.
2. Solve the three-dimensional problem. Use overlapping orthogonal frames and rotations among them to prove the regularity and quadratic structure of $f$ on the unit sphere of a three-dimensional subspace. This is the hard geometric lemma.
3. Patch finite-dimensional subspaces. Restrictions to intersecting three-dimensional subspaces determine compatible quadratic forms. They combine into one positive operator on every finite-dimensional subspace.
4. Recover the global operator. Compatibility and boundedness define a positive operator $\rho$ whose quadratic form equals $f$.
5. Use countable additivity. In infinite dimension, countable orthogonal sums imply normality and trace-class normalization, extending the formula to all projections.

Different proofs reorganize the geometric lemma or use operator-algebraic extension theorems, but the dimension-three constraint enters at the same structural point: two-dimensional orthogonal frames do not overlap richly enough to force a quadratic function.

This page states the imported frame-function lemma rather than pretending to prove it in a few paragraphs. The trace-form consequences and the explicit two-dimensional failure below are proved directly.

The dimension-two exception

For a qubit, every rank-one projector has the form

$$P_{\mathbf n} = \frac12(I+\mathbf n\cdot\boldsymbol\sigma), \qquad |\mathbf n|=1,$$

and its only orthogonal rank-one complement is $P_{-\mathbf n}$. Orthogonal additivity therefore requires only

$$\mu(P_{\mathbf n})+\mu(P_{-\mathbf n})=1.$$

Define

$$\mu(P_{\mathbf n}) = \frac{1+n_z^3}{2}, \qquad \mu(0)=0, \qquad \mu(I)=1.$$

Because $-1\leq n_z^3\leq1$, the values lie in $[0,1]$, and

$$\mu(P_{\mathbf n})+\mu(P_{-\mathbf n}) = \frac{1+n_z^3}{2} + \frac{1-n_z^3}{2} = 1.$$

This is a normalized finitely additive measure on qubit projections. But a qubit density operator

$$\rho=\frac12(I+\mathbf r\cdot\boldsymbol\sigma)$$

would give

$$\operatorname{Tr}(\rho P_{\mathbf n}) = \frac{1+\mathbf r\cdot\mathbf n}{2},$$

which is affine linear in $\mathbf n$. The function $n_z^3$ is not linear on the sphere, so no density operator produces the displayed assignment. This is a concrete failure of the original projection theorem in dimension two.

Countable additivity in infinite dimension

In a finite-dimensional Hilbert space, an orthogonal family has only finitely many nonzero members, so finite additivity already covers every resolution.

In infinite dimension, finite additivity is weaker. States on the bounded- operator algebra can exist that are not normal and are not represented by a trace-class density operator. Their restrictions can be finitely additive on projections without respecting strong limits of countable orthogonal sums. The countable-additivity hypothesis excludes these singular states and is therefore consequential, not decorative.

Equivalently, the theorem’s probability measure is continuous from below for increasing projection sequences:

$$P_n\uparrow P \quad\Longrightarrow\quad \mu(P_n)\longrightarrow\mu(P).$$

This normality condition is what permits the trace-class representation in the separable infinite-dimensional setting.

Relation to the Born rule

Gleason’s conclusion has the same form as the Born rule for sharp events:

$$\Pr(P\mid\rho) = \operatorname{Tr}(\rho P).$$

It shows that once probabilities are assigned to every projection in a normalized, context-independent, orthogonally additive way, no different functional form is available in dimension at least three.

This is a structural derivation within the Hilbert-space event framework. The assumptions already encode:

• projections as sharp events;
• one probability for a projector across all PVM contexts;
• orthogonal alternatives as additive;
• normalization and, in infinite dimension, countable additivity.

The theorem does not explain why physical events are projections, why probability is the relevant predictive concept, or why those assumptions should hold. It also says nothing about selective state update, collapse, instruments, decoherence, or dynamics.

POVM variants and their stronger premise

Gleason-type theorems can include two-dimensional Hilbert spaces by assigning probabilities consistently to all effects or to outcomes across all POVMs, not only to projections. Under suitable additivity assumptions, these assignments again have the form

$$\mu(E)=\operatorname{Tr}(\rho E).$$

The qubit loophole closes because the effect space contains far more decompositions of the identity than antipodal projector pairs.

These are extensions, not the original theorem. Their premise is stronger: probabilities are assigned to generalized measurement effects and are consistent across POVM decompositions. They should not be advertised as obtaining the qubit result from the original projection assumptions alone.

What the theorem does not derive

• It does not derive the Hilbert-space projection structure.
• It does not remove the noncontextual assignment assumption.
• It does not cover the original qubit projection problem.
• It does not derive state update, collapse, or measurement dynamics.
• It does not force probabilities to be objective frequencies or subjective degrees of belief.
• It does not make finite additivity sufficient in infinite dimension.
• It does not classify deterministic hidden-variable models without further assumptions.

Common pitfalls

Forgetting the dimension hypothesis. The explicit qubit assignment shows why $\dim\mathcal H\geq3$ matters for projection measures.

Calling the result assumption free. Projection events, context-independent values, and orthogonal additivity are substantive premises.

Replacing countable additivity by finite additivity in infinite dimension. The weaker condition admits nonnormal states not represented by density operators.

Treating the theorem as a state-update rule. It constrains probabilities, not conditional post-measurement dynamics.

Calling POVM variants the original theorem. Generalized-effect premises are stronger and must be labeled explicitly.

Exercises

1. Verify trace-form additivity

Let $\rho$ be a density operator and $P_j$ mutually orthogonal projections. Show that $\mu(P)=\operatorname{Tr}(\rho P)$ is normalized and countably additive.

Solution

Normalization is $\mu(I)=\operatorname{Tr}\rho=1$. For the strong sum $P=\sum_jP_j$, normality of the trace-class pairing gives

$$\mu(P) = \operatorname{Tr}\!\left(\rho\sum_jP_j\right) = \sum_j\operatorname{Tr}(\rho P_j) = \sum_j\mu(P_j).$$

2. Audit the qubit counterassignment

Prove that $\mu(P_{\mathbf n})=(1+n_z^3)/2$ is a valid probability assignment on qubit projections but is not generated by any density operator.

Solution

Its values lie between zero and one, and replacing $\mathbf n$ by $-\mathbf n$ makes the pair sum to one, which checks every nontrivial qubit orthogonal resolution. A density operator would give $(1+\mathbf r\cdot\mathbf n)/2$, linear in $\mathbf n$. If it equaled the proposed function, then $\mathbf r\cdot\mathbf n=n_z^3$ for all unit vectors. Taking equatorial vectors forces $r_x=r_y=0$, while $\mathbf n=(0,0,1)$ forces $r_z=1$. But then a vector with $n_z=1/2$ gives $1/2$ on the linear side and $1/8$ on the cubic side, a contradiction.

3. Uniqueness of the density operator

Suppose $\operatorname{Tr}(\rho P)=\operatorname{Tr}(\sigma P)$ for every rank-one projector. Prove $\rho=\sigma$.

Solution

For $B=\rho-\sigma$, equality on rank-one projectors gives $\langle\psi|B\psi\rangle=0$ for every $\psi$. The complex polarization identity expresses $\langle\phi|B\psi\rangle$ as a linear combination of the four quadratic values on $\phi+i^k\psi$, $k=0,1,2,3$. Every matrix element vanishes, so $B=0$.

4. Locate noncontextuality

A rank-one projector $P$ occurs in two different orthonormal-basis PVMs. Which assumption says its probability is the same in both?

Solution

The measure is defined as one function $\mu(P)$ on the projector itself. It has no additional context argument. This context-independent assignment is the noncontextuality assumption; additivity then constrains its values within each PVM.

5. Finite versus countable additivity

Why do finite and countable orthogonal additivity coincide in a $d$-dimensional Hilbert space but not in infinite dimension?

Solution

In dimension $d$, at most $d$ mutually orthogonal nonzero subspaces can appear in a rank-one refinement, and every orthogonal family has only finitely many nonzero members. Countable sums therefore reduce to finite sums. Infinite dimension permits infinitely many nonzero orthogonal projections whose strong sum is another projection, so continuity under that limit is an additional condition.

References

• P. Busch, “Quantum states and generalized observables: A simple proof of Gleason’s theorem,” Physical Review Letters 91, 120403, 2003, doi:10.1103/PhysRevLett.91.120403.
• C. M. Caves, C. A. Fuchs, K. Manne, and J. M. Renes, “Gleason-type derivations of the quantum probability rule for generalized measurements,” Foundations of Physics 34, 193–209, 2004, doi:10.1023/B:FOOP.0000019581.00318.a5.
• A. M. Gleason, “Measures on the closed subspaces of a Hilbert space,” Journal of Mathematics and Mechanics 6, 885–893, 1957, doi:10.1512/iumj.1957.6.56050.
• N. P. Landsman, Foundations of Quantum Theory: From Classical Concepts to Operator Algebras, Springer, 2017.