Simple Algorithm
Quantum Computing · Rydberg State · 28 min

Uncertainty Relations

Robertson, Schrödinger and entropic uncertainty relations.

Heisenberg's microscope is a story; Robertson's inequality is a theorem. This lesson proves the variance-based uncertainty relations, shows where they fall silent, and replaces them with entropic bounds that hold for every state and underpin the security of quantum cryptography.

What you’ll learn

  • Prove the Robertson and Schrödinger uncertainty relations from the Cauchy–Schwarz inequality.
  • Show that for a qubit, Schrödinger's relation for XX and YY is equivalent to the Bloch-ball condition.
  • Derive the Maassen–Uffink entropic uncertainty relation by Riesz–Thorin interpolation.
  • State the uncertainty relation with quantum memory and use it to explain QKD key rates.
  • Distinguish preparation uncertainty from error–disturbance trade-offs.

Robertson and Schrödinger

DerivationProof via Cauchy–Schwarz
  1. ∣a⟩:=(A−⟨A⟩)∣ψ⟩,∣b⟩:=(B−⟨B⟩)∣ψ⟩,(ΔA)2=⟨a∣a⟩,(ΔB)2=⟨b∣b⟩\ket{a} := (A - \expval{A})\ket\psi, \quad \ket{b} := (B - \expval{B})\ket\psi, \qquad (\Delta A)^2 = \braket{a|a},\quad (\Delta B)^2 = \braket{b|b}

    Center both observables on the state.

  2. (ΔA)2(ΔB)2  ≥  ∣⟨a∣b⟩∣2=(Re⁡⟨a∣b⟩)2+(Im⁡⟨a∣b⟩)2(\Delta A)^2(\Delta B)^2 \;\ge\; \abs{\braket{a|b}}^2 = \big(\operatorname{Re}\braket{a|b}\big)^2 + \big(\operatorname{Im}\braket{a|b}\big)^2

    Cauchy–Schwarz.

  3. ⟨a∣b⟩=⟨AB⟩−⟨A⟩⟨B⟩,AB=12{A,B}+12[A,B]\braket{a|b} = \expval{AB} - \expval{A}\expval{B}, \qquad AB = \tfrac12\acomm{A}{B} + \tfrac12\comm{A}{B}

    Expand, and split ABAB into its Hermitian and anti-Hermitian parts, whose expectation values are real and imaginary respectively.

  4. Re⁡⟨a∣b⟩=12⟨{A,B}⟩−⟨A⟩⟨B⟩,Im⁡⟨a∣b⟩=12i⟨[A,B]⟩\operatorname{Re}\braket{a|b} = \tfrac12\expval{\acomm{A}{B}} - \expval{A}\expval{B}, \qquad \operatorname{Im}\braket{a|b} = \tfrac{1}{2i}\expval{\comm{A}{B}}

    Keeping both terms gives Schrödinger's relation; dropping the covariance term gives Robertson's. For mixed states, run the same argument with the semi-inner product (X,Y)↦Tr⁡(ρX†Y)(X, Y)\mapsto\Tr(\rho X^\dagger Y).

With [x,p]=iℏ\comm{x}{p} = i\hbar this becomes Kennard's Δx Δp≥ℏ/2\Delta x\,\Delta p \ge \hbar/2 (1927). Note what kind of statement it is: a constraint on preparations. No state has both spreads small. By itself it says nothing about one measurement disturbing another.

A qubit test case

For Pauli observables (ΔP)2=1−⟨P⟩2(\Delta P)^2 = 1-\expval{P}^2, with [X,Y]=2iZ\comm{X}{Y} = 2iZ and {X,Y}=0\acomm{X}{Y} = 0. Write (x,y,z)(x, y, z) for the Bloch vector of the state.

DerivationSchrödinger's relation for X and Y is the Bloch ball
  1. (1−x2)(1−y2)  ≥  x2y2+z2(1 - x^2)(1 - y^2) \;\ge\; x^2y^2 + z^2

    Schrödinger's relation, with covariance term (0−xy)2(0 - xy)^2 and commutator term z2z^2.

  2.   ⟺    1−x2−y2≥z2    ⟺    x2+y2+z2≤1\iff\; 1 - x^2 - y^2 \ge z^2 \;\iff\; x^2 + y^2 + z^2 \le 1

    Expand the left side; the x2y2x^2y^2 terms cancel.

  3. So for a qubit, Schrödinger's inequality for XX and YY is the statement that the state lies in the Bloch ball, with equality exactly for pure states. Robertson's weaker form, (1−x2)(1−y2)≥z2(1-x^2)(1-y^2)\ge z^2, is saturated by a pure state only when xy=0xy = 0, that is, in the xx–zz or yy–zz plane.

Entropic uncertainty: Maassen–Uffink

DerivationProof via Riesz–Thorin interpolation
  1. uj=⟨aj∣ψ⟩,vk=⟨bk∣ψ⟩=∑j⟨bk∣aj⟩ uj,∥v∥2=∥u∥2,∥v∥∞≤c ∥u∥1u_j = \braket{a_j|\psi},\quad v_k = \braket{b_k|\psi} = \sum_j\braket{b_k|a_j}\,u_j, \qquad \norm{v}_2 = \norm{u}_2,\quad \norm{v}_\infty \le c\,\norm{u}_1

    The change of basis T:u↦vT: u\mapsto v is unitary, and every entry of its matrix has modulus at most cc. These are the two endpoint bounds.

  2. ∥v∥p′  ≤  c (2−p)/p ∥u∥p,1≤p≤2,1p+1p′=1\norm{v}_{p'} \;\le\; c^{\,(2-p)/p}\,\norm{u}_p, \qquad 1\le p\le 2,\quad \tfrac1p + \tfrac1{p'} = 1

    The Riesz–Thorin theorem (for complex scalars) interpolates between ℓ1→ℓ∞\ell^1\to\ell^\infty with norm cc and ℓ2→ℓ2\ell^2\to\ell^2 with norm 11.

  3. 1−β2β Hβ(q)  ≤  1−ααlog⁡c+1−α2α Hα(p)\frac{1-\beta}{2\beta}\,H_\beta(q) \;\le\; \frac{1-\alpha}{\alpha}\log c + \frac{1-\alpha}{2\alpha}\,H_\alpha(p)

    Write pj=∣uj∣2p_j = \abs{u_j}^2, qk=∣vk∣2q_k = \abs{v_k}^2, α=p/2\alpha = p/2 and β=p′/2\beta = p'/2, so that 1α+1β=2\tfrac1\alpha+\tfrac1\beta = 2. Take logarithms and use log⁡∑jpjα=(1−α)Hα(p)\log\sum_j p_j^\alpha = (1-\alpha)H_\alpha(p).

  4. 1−ββ=−1−αα  ⟹  Hα(p)+Hβ(q)  ≥  −2log⁡c\frac{1-\beta}{\beta} = -\frac{1-\alpha}{\alpha} \;\Longrightarrow\; H_\alpha(p) + H_\beta(q) \;\ge\; -2\log c

    Multiply by 2α/(1−α)>02\alpha/(1-\alpha) > 0, valid for 12≤α<1\tfrac12 \le \alpha < 1. Letting α,β→1\alpha,\beta\to1 gives the Shannon version.

For mutually unbiased bases c=1/dc = 1/\sqrt d and the bound is log⁡2d\log_2 d, attained by any basis vector of either basis; for qubit ZZ and XX, H(Z)+H(X)≥1H(Z)+H(X) \ge 1 bit. Deutsch (1983) proved the first state-independent entropic bound, −2log⁡21+c2-2\log_2\tfrac{1+c}{2}; Kraus (1987) conjectured the sharper form that Maassen and Uffink proved.

Predict before you look

Take the qubit state halfway between ∣0⟩\ket0 and ∣+⟩\ket+ on the Bloch sphere (θ=π/4\theta = \pi/4, φ=0\varphi = 0). What is H(Z)+H(X)H(Z) + H(X), in bits?

Show the answer

1.2 bits

Both outcome distributions are (cos⁡2π8,sin⁡2π8)≈(0.854,0.146)\big(\cos^2\tfrac\pi8, \sin^2\tfrac\pi8\big) \approx (0.854, 0.146), each with entropy about 0.600.60 bits, so the sum is about 1.20≥11.20 \ge 1. The bound is attained by basis states (0+10 + 1), not by this "balanced" state: entropic and variance-based relations have different minimizers.

Uncertainty with quantum memory

Without correlations (ρAB=ρA⊗ρB\rho_{AB} = \rho_A\otimes\rho_B) this is the Maassen–Uffink relation, strengthened by H(A)≥0H(A) \ge 0. With a maximally entangled memory, H(A∣B)=−log⁡2dH(A|B) = -\log_2 d, and for mutually unbiased bases the bound drops to 00. Bob really can predict both outcomes: on ∣Φ+⟩\ket{\Phi^+}, X⊗X=Z⊗Z=+1X\otimes X = Z\otimes Z = +1. This is the EPR paradox, made quantitative.

Check your understanding

Alice and Bob share ∣Φ+⟩=12(∣00⟩+∣11⟩)\ket{\Phi^+} = \tfrac{1}{\sqrt2}(\ket{00}+\ket{11}), and Alice measures XX or ZZ on her qubit. What is H(X∣B)+H(Z∣B)H(X|B) + H(Z|B)?

  1. 00
  2. 11 bit
  3. 22 bits
  4. −1-1 bit
Show the answer

A. 00

Bob can measure the same observable on his half and obtain Alice's outcome with certainty, so both conditional entropies vanish. The bound is log⁡22+H(A∣B)=1−1=0\log_2 2 + H(A|B) = 1 - 1 = 0: it is tight here.

Preparation versus measurement uncertainty

Heisenberg's original argument was about measurement: locating an electron with a short-wavelength photon kicks its momentum. Robertson and Maassen–Uffink constrain preparations. Turning the measurement story into a theorem is subtle, because "error" and "disturbance" must first be defined.

RelationStatementStatus
Kennard (1927); Robertson (1929)ΔA ΔB≥12∣⟨[A,B]⟩∣\Delta A\,\Delta B \ge \tfrac12\abs{\expval{\comm{A}{B}}}Theorem about preparations; state-dependent
Białynicki-Birula & Mycielski (1975)h(x)+h(p)≥ln⁡(eπℏ)h(x) + h(p) \ge \ln(e\pi\hbar) for the differential entropiesTheorem; implies Kennard's relation, because Gaussians maximize entropy at fixed variance
Maassen & Uffink (1988)H(A)+H(B)≥−2log⁡2cH(A)+H(B) \ge -2\log_2 cTheorem about preparations; state-independent
Ozawa (2003)ε(A)η(B)+ε(A)ΔB+ΔA η(B)≥12∣⟨[A,B]⟩∣\varepsilon(A)\eta(B) + \varepsilon(A)\Delta B + \Delta A\,\eta(B) \ge \tfrac12\abs{\expval{\comm{A}{B}}}Theorem for state-dependent error ε\varepsilon and disturbance η\eta. The naive εη≥12∣⟨[A,B]⟩∣\varepsilon\eta \ge \tfrac12\abs{\expval{\comm{A}{B}}} can fail, as neutron-spin experiments confirmed (Erhart et al., 2012)
Busch, Lahti & Werner (2013)ε(Q) η(P)≥ℏ/2\varepsilon(Q)\,\eta(P) \ge \hbar/2Theorem with worst-case, state-independent error measures
Two of these "Heisenberg relations" seem to contradict each other. They do not: they define error differently. Which definitions are operationally appropriate is still debated.

Key takeaways

  • Robertson's and Schrödinger's relations follow from Cauchy–Schwarz. They are state-dependent and can be vacuous, for instance on eigenstates.
  • For a qubit, Schrödinger's relation for XX and YY is exactly the Bloch-ball condition, with equality for pure states.
  • Maassen–Uffink: H(A)+H(B)≥−2log⁡cH(A)+H(B)\ge-2\log c, a state-independent bound equal to log⁡d\log d for mutually unbiased bases, proved by Riesz–Thorin interpolation.
  • With quantum memory the bound gains H(A∣B)H(A|B), which entanglement can make negative. Its tripartite form powers QKD security proofs, such as the BB84 rate 1−2h(Q)1-2h(Q).
  • Preparation uncertainty and error–disturbance trade-offs are different statements; the latter depend on how error is defined.

Check yourself

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

Question 1

For the state ∣+i⟩\ket{+i}, what is Robertson's lower bound on ΔX ΔZ\Delta X\,\Delta Z, and is it attained?

  1. 00; not attained
  2. 12\tfrac12; attained
  3. 11; attained, since ΔX=ΔZ=1\Delta X = \Delta Z = 1
  4. 22; not attained
Show the answer

C. 11; attained, since ΔX=ΔZ=1\Delta X = \Delta Z = 1

[X,Z]=−2iY\comm{X}{Z} = -2iY and ⟨Y⟩=1\expval{Y} = 1 for ∣+i⟩\ket{+i}, so the bound is 12⋅2=1\tfrac12\cdot 2 = 1. Both XX and ZZ are coin flips for ∣+i⟩\ket{+i}, so ΔX=ΔZ=1\Delta X = \Delta Z = 1 and the bound is attained.

Question 2

For a qutrit measured in the computational basis and in the Fourier basis, what does the Maassen–Uffink relation guarantee?

  1. H(A)+H(B)≥1H(A) + H(B) \ge 1 bit
  2. H(A)+H(B)≥log⁡23≈1.58H(A) + H(B) \ge \log_2 3 \approx 1.58 bits
  3. H(A)+H(B)≥2log⁡23H(A) + H(B) \ge 2\log_2 3 bits
  4. Nothing: the relation only applies to qubits
Show the answer

B. H(A)+H(B)≥log⁡23≈1.58H(A) + H(B) \ge \log_2 3 \approx 1.58 bits

The Fourier basis is unbiased to the computational basis, ∣⟨j∣F∣k⟩∣=1/3\abs{\bra{j}F\ket{k}} = 1/\sqrt3, so −2log⁡2c=log⁡23-2\log_2 c = \log_2 3. A basis state attains it: 0+log⁡230 + \log_2 3.

Question 3

What is the main advantage of the Maassen–Uffink relation over Robertson's?

  1. It applies to continuous variables
  2. It uses variances, which are easier to measure
  3. It is always numerically larger
  4. Its bound depends only on the two bases, not on the state, so it stays informative even for eigenstates of one observable
Show the answer

D. Its bound depends only on the two bases, not on the state, so it stays informative even for eigenstates of one observable

Robertson's bound 12∣⟨[A,B]⟩∣\tfrac12\abs{\expval{\comm{A}{B}}} depends on the state and vanishes on eigenstates. The entropic bound −2log⁡c-2\log c is a property of the two bases alone, and it is strictly positive whenever no vector of one basis coincides with a vector of the other.

Question 4

In H(X∣B)+H(Z∣B)≥log⁡2(1/c2)+H(A∣B)H(X|B) + H(Z|B) \ge \log_2(1/c^2) + H(A|B), with mutually unbiased qubit bases, when can the right-hand side reach 00?

  1. When AA is maximally entangled with BB
  2. When AA is in a pure state
  3. When AA and BB are uncorrelated
  4. Never
Show the answer

A. When AA is maximally entangled with BB

For a maximally entangled pair of qubits H(A∣B)=−1H(A|B) = -1, which cancels log⁡2(1/c2)=1\log_2(1/c^2) = 1.

Further reading

  • H. P. Robertson, "The uncertainty principle", Physical Review 34 (1929).
  • H. Maassen & J. B. M. Uffink, "Generalized entropic uncertainty relations", Physical Review Letters 60 (1988).
  • M. Berta, M. Christandl, R. Colbeck, J. M. Renes & R. Renner, "The uncertainty principle in the presence of quantum memory", Nature Physics 6 (2010).
  • P. J. Coles, M. Berta, M. Tomamichel & S. Wehner, "Entropic uncertainty relations and their applications", Reviews of Modern Physics 89 (2017).
  • P. Busch, P. Lahti & R. F. Werner, "Proof of Heisenberg's error-disturbance relation", Physical Review Letters 111 (2013).