Simple Algorithm
Quantum Computing · Rydberg State · 30 min

Gleason's Theorem: Why the Born Rule?

The Born rule is the only consistent probability rule (d ≥ 3).

Why should probabilities be squared amplitudes, rather than absolute values or fourth powers? Gleason's theorem gives a startling answer: in any dimension of three or more, as soon as the probability of an outcome depends only on its projector, the Born rule is the only consistent possibility.

What you’ll learn

  • State Gleason's theorem precisely, both for frame functions and for probability measures on projections.
  • Explain the role of non-contextuality, and construct a non-Born rule and a deterministic hidden-variable model for d=2d = 2.
  • Prove that every continuous frame function on the unit sphere of R3\R^3 is a quadratic form, using spherical harmonics and Schur's lemma.
  • Prove the effect-space (POVM) version of Gleason's theorem, valid in every dimension.
  • Derive the obstruction to non-contextual definite values as a corollary.

From probability rules to frame functions

Suppose all you know is that a system has a Hilbert space H\Hil and that every orthonormal basis defines a possible measurement. A probability rule (for a fixed preparation) assigns a number in [0,1][0,1] to each outcome. Make two minimal assumptions:

  1. Non-contextuality. The probability of the outcome "∣x⟩\ket{x}" depends only on the ray of ∣x⟩\ket{x}, i.e. on the projector ∣x⟩ ⁣⟨x∣\dyad{x}, not on which basis it was measured in.
  2. Normalization. In every orthonormal basis the probabilities add up to one.

Read the conclusion slowly. The Born rule p(x)=∣⟨x∣ψ⟩∣2p(x) = \abs{\braket{x|\psi}}^2 is the special case ρ=∣ψ⟩ ⁣⟨ψ∣\rho = \dyad\psi. Mixed states appear automatically, without being postulated. And nothing else is possible. The only inputs were the lattice of projections and non-contextual additivity.

Why dimension two is special

In C2\C^2 every orthonormal basis is a pair of antipodal points {n^,−n^}\{\hat n, -\hat n\} on the Bloch sphere, and each ray belongs to exactly one basis. The frame condition therefore says only

f(n^)+f(−n^)=1,f(\hat n) + f(-\hat n) = 1,

so any f=12(1+g)f = \tfrac12(1 + g) with gg odd, g(−n^)=−g(n^)g(-\hat n) = -g(\hat n), and ∣g∣≤1\abs{g} \le 1 is allowed. The Born rule is the linear choice g(n^)=r⃗⋅n^g(\hat n) = \vec r\cdot\hat n, where r⃗\vec r is the Bloch vector of the state. Nothing singles it out.

Check your understanding

For a qubit with Bloch vector r⃗≠0\vec r \neq 0 and measurement direction n^\hat n, which rule is a non-negative frame function of weight 1 that is not of Born form?

  1. f(n^)=12(1+r⃗⋅n^)f(\hat n) = \tfrac12(1 + \vec r\cdot\hat n)
  2. f(n^)=12(1+(r⃗⋅n^)2)f(\hat n) = \tfrac12\big(1 + (\vec r\cdot\hat n)^2\big)
  3. f(n^)=12(1+(r⃗⋅n^)3)f(\hat n) = \tfrac12\big(1 + (\vec r\cdot\hat n)^3\big)
  4. f(n^)=∣r⃗⋅n^∣f(\hat n) = \abs{\vec r\cdot\hat n}
Show the answer

C. f(n^)=12(1+(r⃗⋅n^)3)f(\hat n) = \tfrac12\big(1 + (\vec r\cdot\hat n)^3\big)

(r⃗⋅n^)3(\vec r\cdot\hat n)^3 is odd in n^\hat n and bounded by 11, so f(n^)+f(−n^)=1f(\hat n)+f(-\hat n) = 1 and 0≤f≤10 \le f \le 1. It is not of the form ⟨n^∣ρ′∣n^⟩=12(1+r⃗ ′⋅n^)\bra{\hat n}\rho'\ket{\hat n} = \tfrac12(1+\vec r\,'\cdot\hat n) for any r⃗ ′\vec r\,', because it is not affine in n^\hat n. Gleason's theorem says nothing like this exists when d≥3d \ge 3.

DerivationA deterministic hidden-variable model for a qubit
  1. λ∼Uniform(S2),vλ(n^)=sign⁡((r⃗+λ)⋅n^)∈{±1}\lambda \sim \text{Uniform}(S^2), \qquad v_\lambda(\hat n) = \operatorname{sign}\big((\vec r + \lambda)\cdot\hat n\big) \in \{\pm1\}

    A hidden variable λ\lambda fixes the outcome of every possible measurement in advance. Since vλ(−n^)=−vλ(n^)v_\lambda(-\hat n) = -v_\lambda(\hat n), each basis gets exactly one "+1+1" (ties have probability zero).

  2. Pr⁡[λ⋅n^≤t]=1+t2,t∈[−1,1]\Pr\big[\lambda\cdot\hat n \le t\big] = \frac{1+t}{2}, \qquad t\in[-1,1]

    Archimedes' hat-box theorem: for λ\lambda uniform on the sphere, the projection λ⋅n^\lambda\cdot\hat n is uniform on [−1,1][-1, 1].

  3. Pr⁡[vλ(n^)=+1]=Pr⁡[λ⋅n^>−r⃗⋅n^]=1+r⃗⋅n^2=⟨n^∣ρ∣n^⟩\Pr\big[v_\lambda(\hat n) = +1\big] = \Pr\big[\lambda\cdot\hat n > -\vec r\cdot\hat n\big] = \frac{1 + \vec r\cdot\hat n}{2} = \bra{\hat n}\rho\ket{\hat n}

    Averaging over λ\lambda reproduces the Born rule for every state (∣r⃗∣≤1\abs{\vec r}\le 1) and every measurement.

  4. So for a single qubit, quantum statistics can be mimicked by pre-existing, context-independent values. Explicit models of this kind go back to Bell (1966) and to Kochen and Specker (1967). Gleason's theorem, and the obstruction to such models, genuinely need d≥3d\ge 3.

Proof, part one: continuous frame functions

Gleason's proof has three parts. (i) Every continuous frame function on the unit sphere S2⊂R3S^2\subset\R^3 is a quadratic form. (ii) Every non-negative frame function on R3\R^3 is continuous. (iii) A frame function on a real or complex space of dimension ≥3\ge 3 restricts to a frame function on every real three-dimensional slice spanned by orthonormal vectors, and regularity on all slices implies regularity. Part (i) is a gem of harmonic analysis.

DerivationContinuous frame functions on the 2-sphere are quadratic
  1. f=∑ℓ evenfℓ,fℓ∈Hℓ,dim⁡Hℓ=2ℓ+1f = \sum_{\ell\ \text{even}} f_\ell, \qquad f_\ell \in \mathcal{H}_\ell,\quad \dim\mathcal{H}_\ell = 2\ell+1

    Evenness. {e1,e2,e3}\{e_1,e_2,e_3\} and {−e1,e2,e3}\{-e_1,e_2,e_3\} are both orthonormal, so f(−x)=f(x)f(-x) = f(x). In the expansion of ff in spherical harmonics only even degrees ℓ\ell occur.

  2. (Tf)(R)=∑i=13f(Rei),T(f∘g−1)(R)=(Tf)(g−1R)(Tf)(R) = \sum_{i=1}^{3} f(Re_i), \qquad T\big(f\circ g^{-1}\big)(R) = (Tf)(g^{-1}R)

    An intertwiner. Identify orthonormal frames with rotations R∈SO(3)R\in SO(3) and define the frame-sum map TT. It commutes with rotations.

  3. Tf≡W  ⟹  Tfℓ=0  (ℓ≥2),ker⁡(T∣Hℓ)∈{0, Hℓ}Tf \equiv W \;\Longrightarrow\; Tf_\ell = 0\ \ (\ell\ge 2), \qquad \ker\big(T|_{\mathcal{H}_\ell}\big)\in\{0,\ \mathcal{H}_\ell\}

    Schur. The frame condition says TfTf is constant, i.e. lies in the trivial representation. TT maps Hℓ\mathcal{H}_\ell into the ℓ\ell-isotypic part of L2(SO(3))L^2(SO(3)), so every component with ℓ≥1\ell \ge 1 must vanish. Since Hℓ\mathcal{H}_\ell is irreducible, TT restricted to it is either injective or zero.

  4. (TZℓ)(I)=Pℓ(0)+Pℓ(0)+Pℓ(1)=1+2Pℓ(0)(TZ_\ell)(\id) = P_\ell(0) + P_\ell(0) + P_\ell(1) = 1 + 2P_\ell(0)

    A test vector. Evaluate TT on the zonal harmonic Zℓ(x)=Pℓ(x⋅e3)Z_\ell(x) = P_\ell(x\cdot e_3) at R=IR = \id, where PℓP_\ell is the Legendre polynomial.

  5. 1+2P2(0)=0,1+2P4(0)=74,1+2P6(0)=38,1+2P8(0)=9964, …1 + 2P_2(0) = 0, \qquad 1 + 2P_4(0) = \tfrac74,\quad 1 + 2P_6(0) = \tfrac38,\quad 1 + 2P_8(0) = \tfrac{99}{64},\ \dots

    For ℓ=2\ell = 2, TT annihilates H2\mathcal{H}_2, as it must: degree-2 harmonics are traceless quadratic forms xTAxx^{\mathsf T}Ax, and ∑i(Rei)TA(Rei)=Tr⁡A=0\sum_i (Re_i)^{\mathsf T}A(Re_i) = \Tr A = 0. For even ℓ≥4\ell\ge4, ∣Pℓ(0)∣<12\abs{P_\ell(0)} < \tfrac12, so TT is injective on Hℓ\mathcal{H}_\ell and therefore fℓ=0f_\ell = 0.

  6. f(x)=c+xTAx=xT(c I+A) x=:xTQx,Tr⁡Q=3c=Wf(x) = c + x^{\mathsf T}Ax = x^{\mathsf T}\big(c\,\id + A\big)\,x =: x^{\mathsf T}Qx, \qquad \Tr Q = 3c = W

    Only ℓ=0\ell = 0 and ℓ=2\ell = 2 survive, so ff is a quadratic form whose trace is the weight. If f≥0f \ge 0 then Q≥0Q \ge 0, and ρ=Q/W\rho = Q/W is a density matrix.

Gleason in every dimension: effects instead of projectors

What if the rule must also assign probabilities to the elements of generalized measurements (POVMs), the effects 0≤E≤I0 \le E \le \id? Then non-contextual additivity has many more instances to bite on. The theorem becomes nearly elementary, and it holds for qubits too.

DerivationProof
  1. v(knE)=k v(1nE)=kn v(E)v\big(\tfrac{k}{n}E\big) = k\,v\big(\tfrac{1}{n}E\big) = \tfrac{k}{n}\,v(E)

    Rational homogeneity. Additivity gives v(0)=0v(0) = 0 and n v(E/n)=v(E)n\,v(E/n) = v(E), hence v(qE)=q v(E)v(qE) = q\,v(E) for every rational q∈[0,1]q\in[0,1].

  2. q1≤λ≤q2  ⟹  q1 v(E)≤v(λE)≤q2 v(E)q_1 \le \lambda \le q_2 \;\Longrightarrow\; q_1\,v(E) \le v(\lambda E) \le q_2\,v(E)

    Monotonicity, then real homogeneity. If E≤FE \le F then v(F)=v(E)+v(F−E)≥v(E)v(F) = v(E) + v(F-E) \ge v(E). Squeezing a real λ∈[0,1]\lambda\in[0,1] between rationals gives v(λE)=λ v(E)v(\lambda E) = \lambda\,v(E).

  3. v(A+B)=v(A)+v(B),v(tA)=t v(A)(t∈R)v(A+B) = v(A) + v(B), \qquad v(tA) = t\,v(A) \quad (t\in\R)

    Linear extension. For A≥0A \ge 0 set v(A):=s v(A/s)v(A) := s\,v(A/s) with any s≥∥A∥s \ge \norm{A} (independent of ss by homogeneity); for Hermitian A=A+−A−A = A_+ - A_- set v(A)=v(A+)−v(A−)v(A) = v(A_+) - v(A_-). Additivity on the positive cone makes this well defined and R\R-linear.

  4. v(A)=Tr⁡(ρA)for all Hermitian Av(A) = \Tr(\rho A)\quad\text{for all Hermitian } A

    Riesz representation. Every real-linear functional on the Hermitian d×dd\times d matrices is of this form for a unique Hermitian ρ\rho, because the Hilbert–Schmidt inner product is non-degenerate.

  5. ⟨x∣ρ∣x⟩=v(∣x⟩ ⁣⟨x∣)≥0,Tr⁡ρ=v(I)=1■\bra{x}\rho\ket{x} = v\big(\dyad{x}\big) \ge 0, \qquad \Tr\rho = v(\id) = 1 \qquad\blacksquare

    Positivity and trace make ρ\rho a density operator.

Check your understanding

Busch's proof takes a few lines; Gleason's takes many pages. Why doesn't the effect version make Gleason's theorem redundant?

  1. It needs continuity of vv as an extra assumption
  2. It assumes additivity on all effects, a far larger family than projectors, so its non-contextuality assumption is much stronger
  3. It only works for qubits
  4. It produces a different probability rule from Gleason's
Show the answer

B. It assumes additivity on all effects, a far larger family than projectors, so its non-contextuality assumption is much stronger

An effect such as 12I\tfrac12\id can be split into smaller effects in countless ways, and demanding consistency across all of them is what makes linearity easy to extract. Gleason uses only projectors, which split only into orthogonal projectors, and still reaches the same conclusion when d≥3d\ge3. In d=2d=2 the deterministic model above satisfies Gleason's hypotheses but cannot be extended additively to effects.

Consequences and open questions

Bell (1966) drew this conclusion from Gleason's theorem. Kochen and Specker (1967) proved it with a finite set of 117 directions in R3\R^3, which turns it into a statement about finitely many experiments. That story continues in Commuting Observables & Contextuality.

Route to the Born ruleKey assumptionScope
Gleason (1957)Non-contextual, additive probabilities on projectorsd≥3d \ge 3
Busch (2003); Caves, Fuchs, Manne & Renes (2004)Non-contextual, additive probabilities on all effectsEvery d≥2d \ge 2
Deutsch (1999); Wallace (2012)Rational decision-making by agents in a branching (Everettian) universeContested: is it circular?
Zurek (2003–2005)Symmetries of entangled states ("envariance")Assumptions debated
Masanes, Galley & Müller (2019)The other postulates of quantum theory plus mild operational assumptionsMeasurement postulates shown to be redundant
Different derivations trade different assumptions. None of them explains why a measurement has an outcome in the first place.

Key takeaways

  • Gleason: in d≥3d\ge3, every non-negative frame function (a non-contextual, additive probability rule on projectors) is x↦⟨x∣ρ∣x⟩x\mapsto\bra x\rho\ket x. The Born rule and density operators are forced.
  • d=2d = 2 is exceptional: bases only pair antipodes, non-Born rules exist, and deterministic hidden-variable models reproduce all single-qubit statistics.
  • For continuous frame functions, harmonic analysis kills every spherical harmonic except ℓ=0,2\ell = 0, 2. The hard step is deriving continuity from positivity alone.
  • Requiring additivity on all effects makes the derivation elementary and valid in every dimension (Busch; Caves, Fuchs, Manne & Renes).
  • Corollary: no {0,1}\{0,1\} value assignments exist in d≥3d\ge3, the seed of the Kochen–Specker theorem.

Check yourself

In the interactive lesson these are scored with your confidence, and what you miss comes back for review.

Question 1

Why does Gleason's theorem require dim⁡H≥3\dim\Hil \ge 3?

  1. Density matrices are only defined for d≥3d \ge 3
  2. In d=2d = 2 each ray lies in exactly one orthonormal basis, so the frame condition only pairs antipodal Bloch points
  3. The proof uses the cross product, which only exists in R3\R^3
  4. Single qubits violate the Born rule
Show the answer

B. In d=2d = 2 each ray lies in exactly one orthonormal basis, so the frame condition only pairs antipodal Bloch points

With only the constraint f(n^)+f(−n^)=1f(\hat n)+f(-\hat n)=1, almost anything goes, including deterministic hidden-variable models. In d≥3d \ge 3 every ray belongs to infinitely many interlocking bases, and the frame condition becomes rigid.

Question 2

Which assumption about the probability rule does Gleason's theorem use?

  1. It is continuous in the measured vector
  2. It is a quadratic function of the amplitudes
  3. It is defined only for pure states
  4. The probability of a projector does not depend on the other projectors measured with it, and is additive over orthogonal decompositions
Show the answer

D. The probability of a projector does not depend on the other projectors measured with it, and is additive over orthogonal decompositions

Gleason assumes only non-contextuality and additivity, plus non-negativity. Continuity and the quadratic form are conclusions, and density operators emerge rather than being assumed.

Question 3

In the spherical-harmonic proof for continuous frame functions on S2S^2, which degrees ℓ\ell can appear?

  1. Only ℓ=0\ell = 0 and ℓ=2\ell = 2
  2. All even ℓ\ell
  3. Only ℓ=0\ell = 0
  4. ℓ=0\ell = 0, 11 and 22
Show the answer

A. Only ℓ=0\ell = 0 and ℓ=2\ell = 2

Odd ℓ\ell are excluded by evenness, ℓ=0\ell = 0 carries the weight, ℓ=2\ell = 2 is annihilated by the frame sum (traceless quadratic forms), and every even ℓ≥4\ell \ge 4 is excluded because 1+2Pℓ(0)≠01 + 2P_\ell(0) \neq 0.

Question 4

Gleason's theorem implies that in d≥3d \ge 3 no frame function of weight 1 takes only the values 00 and 11. What is the key step?

  1. {0,1}\{0,1\}-valued assignments can never be additive over orthogonal decompositions
  2. Gleason's theorem assumes continuity, and {0,1}\{0,1\}-valued functions are discontinuous
  3. By Gleason the function is x↦⟨x∣ρ∣x⟩x \mapsto \bra{x}\rho\ket{x}, which is continuous; a continuous {0,1}\{0,1\}-valued function on the connected unit sphere is constant
  4. It would violate the uncertainty principle
Show the answer

C. By Gleason the function is x↦⟨x∣ρ∣x⟩x \mapsto \bra{x}\rho\ket{x}, which is continuous; a continuous {0,1}\{0,1\}-valued function on the connected unit sphere is constant

Regularity gives continuity, connectedness gives constancy, and a constant function sums to 00 or dd over a basis, never 11.

Further reading

  • A. M. Gleason, "Measures on the closed subspaces of a Hilbert space", Journal of Mathematics and Mechanics 6 (1957).
  • R. Cooke, M. Keane & W. Moran, "An elementary proof of Gleason's theorem", Mathematical Proceedings of the Cambridge Philosophical Society (1985).
  • P. Busch, "Quantum states and generalized observables: a simple proof of Gleason's theorem", Physical Review Letters 91 (2003).
  • C. M. Caves, C. A. Fuchs, K. K. Manne & J. M. Renes, "Gleason-type derivations of the quantum probability rule for generalized measurements", Foundations of Physics 34 (2004).
  • J. S. Bell, "On the problem of hidden variables in quantum mechanics", Reviews of Modern Physics 38 (1966).