Quantum Mechanics
The Leggett-Garg Inequality: Testing Macroscopic Realism in Time
Measure a single quantum object at three different moments in time and the arithmetic of your correlations can crawl to K₃ = 3/2 — half a unit past the ceiling of 1 that any classical, common-sense object is supposed to respect. The Leggett-Garg inequality, proposed in 1985 by Anthony Leggett and Anupam Garg, is the temporal cousin of Bell's inequality: rather than correlating two particles across space, it correlates one system across time, and asks whether a macroscopic object ever fails to have a definite property between our glances at it.
The inequality is a testable consequence of "macroscopic realism" — the intuition that a superconducting loop, a flux qubit, or a dust grain is in one definite state at every instant, whether or not anyone looks. Quantum mechanics predicts violations, and experiments now routinely see them, pushing the question of when (or whether) the classical world crystallizes out of quantum superposition.
- RegimeTemporal quantum coherence; quantum-to-classical transition
- Key relationK₃ = C₁₂ + C₂₃ − C₁₃ ≤ 1 (macrorealism); quantum max 3/2 for a qubit
- Proposed1985, by Anthony J. Leggett & Anupam Garg
- Characteristic scaleViolation window ~ measurement-interval τ set by system's coherence time T₂
- Realized inSuperconducting transmon and flux qubits, NV centers, photons, nuclear spins, IBM quantum processors
- Matters forMacroscopic superpositions, quantum foundations, weak-measurement metrology
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
What it is and why it matters
The Leggett-Garg inequality (LGI) asks a deceptively simple question: does a macroscopic system possess a definite value of some property at every instant, independent of measurement? This worldview — macroscopic realism — is the everyday intuition that the Moon is there when nobody looks. Leggett and Garg realized in 1985 that this belief, combined with the assumption that you can measure without disturbing, forces a bound on how correlated a single observable can be with itself at different times.
Quantum mechanics violates that bound. This makes the LGI the temporal analog of Bell's theorem: where Bell rules out local hidden variables using two entangled particles, Leggett-Garg targets a single system evolving in time, probing whether the object is in a superposition of, say, clockwise and counter-clockwise supercurrent between our measurements. It is one of the few quantitative, experimental handles on the quantum-to-classical transition — the boundary where superpositions stop and definite classical facts begin.
The mechanism, assumption by assumption
The derivation rests on three postulates. (A1) Macrorealism per se: a system with two or more macroscopically distinct states is, at all times, in exactly one of them. (A2) Non-invasive measurability: one can determine which state it is in without disturbing its subsequent evolution — in principle via an ideal negative-result measurement, learning the state from a detector's silence. (A3) Induction / arrow of time: a measurement's outcome depends only on the past, not the future.
Grant these, and each measurement outcome Qᵢ = ±1 is a pre-existing property. Consider three successive measurements at t₁ < t₂ < t₃. For any single realization where Q₁, Q₂, Q₃ each equal ±1, the algebraic identity Q₁Q₂ + Q₂Q₃ − Q₁Q₃ can only take the values +1 or −3 (check the eight sign combinations). Averaging over an ensemble therefore cannot exceed +1. Quantum mechanics breaks this because Q₂ is not a pre-existing fact: the intermediate measurement projects the state, and the coherent evolution between measurements builds correlations that no fixed set of values can reproduce.
The K₃ criterion, its bounds and scales
Define the two-time correlator Cᵢⱼ = ⟨Q(tᵢ)Q(tⱼ)⟩. The three-term Leggett-Garg function is
K₃ = C₁₂ + C₂₃ − C₁₃, with macrorealism requiring −3 ≤ K₃ ≤ 1.
For a qubit precessing at Rabi/Larmor frequency Ω with equal time steps τ, projective measurements give Cᵢⱼ = cos(Ω·Δt), so K₃ = 2cos(Ωτ) − cos(2Ωτ). Maximizing yields K₃ = 3/2 at Ωτ = π/3 — the temporal Tsirelson bound for a two-level system. This is the quantum ceiling for a qubit; the purely algebraic maximum is 3.
The characteristic scale is set by the coherence time: the measurement interval τ must be short compared to T₂ (dephasing time), so that genuine superposition survives between glances. In higher-dimensional systems (qutrits and beyond) with cleverly chosen state-update rules, violations can be pushed beyond 3/2, approaching the algebraic limit of 3 — a signature that dimensionality, not just 'quantumness,' sets the temporal correlation budget.
How it is measured: the experimental realizations
The landmark test was Palacios-Laloy, Mallet, Vion, Esteve et al. (2010), who coupled a superconducting transmon qubit to a microwave resonator and read it out with a weak continuous measurement. From the noise spectrum of the continuously monitored signal they reconstructed the two-time correlators and found K₃ ≈ 1.5, comfortably above the classical bound of 1. Weak measurement is essential: it extracts correlation information while only gently perturbing the qubit, sidestepping full projection.
Since then the LGI has been violated on many platforms: nitrogen-vacancy (NV) centers in diamond, nuclear spins probed by NMR, single photons, phosphorus donors in silicon, neutrino oscillation data, and IBM's cloud superconducting processors. A recurring strategy for closing loopholes is the ideal negative-result measurement: place a detector so that a 'click' would indicate one state, and record only the runs where it stays silent — inferring the state from absence of interaction, thereby defending non-invasiveness. Knee et al. (2016) implemented such a test with a flux qubit to confront the clumsiness loophole head-on.
Where it operates, and distinctions from related effects
The LGI applies to any system with a dichotomic (±1) observable evolving coherently in time — spins, superconducting circuits, photon polarization, atomic energy levels. It is not the same as Bell's inequality: Bell forbids enforcing its key assumption (locality) by spacelike separation, whereas Leggett-Garg's key assumption (non-invasiveness) cannot be geometrically enforced. This is the clumsiness loophole: a violation might mean the world is non-macrorealistic, or merely that your measurement was invasive due to experimental clumsiness. That gap is why the LGI, unlike Bell, cannot be made fully assumption-free.
It also differs from decoherence and from wavefunction collapse per se: the LGI does not describe how coherence is lost, it tests whether definite-value realism holds at all over an interval. And unlike the uncertainty principle, which constrains simultaneous incompatible observables, the LGI constrains the same observable at different times. Complementary formulations include the four-term K₄ inequality and 'no-signalling-in-time' conditions that sharpen the macrorealism criterion.
Applications, significance and open questions
The deepest motivation remains Leggett's original one: how large — how macroscopic — a superposition can be sustained before the world becomes stubbornly classical. Each larger, warmer, more macroscopic system that violates the LGI extends the domain where quantum coherence provably reigns, testing collapse models (like GRW/CSL) that predict a breakdown at large mass or complexity. So far, no such breakdown has been seen.
Beyond foundations, the LGI framework informs weak-measurement metrology, quantum-coherence witnessing in transport and biological systems, benchmarking of qubit non-classicality on real hardware, and even probes of quantum phase transitions via time correlations. Open questions are sharp: Can the clumsiness loophole ever be fully closed, or only bounded by adversarial 'more-invasive' control experiments? What is the correct macrorealism criterion — the original LGI, no-signalling-in-time, or a generalized-probabilistic-theory reformulation? And how far up the mass scale can violations be demonstrated before environmental decoherence, rather than any fundamental collapse, erases the effect?
| Feature | Leggett-Garg inequality | Bell / CHSH inequality |
|---|---|---|
| What is correlated | One system, measured at different times t₁,t₂,t₃ | Two systems, measured at spacelike-separated locations |
| Assumption tested | Macroscopic realism + non-invasive measurability | Local realism (locality + realism) |
| Loophole-free enforcement | Non-invasiveness cannot be enforced by geometry (clumsiness loophole) | Locality enforced by spacelike separation |
| Classical bound | −3 ≤ K₃ ≤ 1 | |S| ≤ 2 |
| Quantum maximum | 3/2 for a qubit (temporal Tsirelson bound); up to 3 in higher-d systems | 2√2 ≈ 2.83 (Tsirelson bound) |
| First convincing experiment | Palacios-Laloy et al., 2010 (transmon qubit + resonator) | Aspect et al., 1982 (entangled photons) |
Frequently asked questions
How is the Leggett-Garg inequality different from Bell's inequality?
Bell's inequality correlates two spatially separated particles and tests local realism, enforcing the locality assumption via spacelike separation. Leggett-Garg correlates a single system measured at different times and tests macroscopic realism plus non-invasive measurability. Crucially, non-invasiveness cannot be enforced geometrically, so the Leggett-Garg test carries the 'clumsiness loophole' that Bell tests avoid.
Why is the classical bound K₃ ≤ 1 and the quantum maximum 3/2?
If each outcome Q₁,Q₂,Q₃ is a pre-existing ±1 value, the combination Q₁Q₂ + Q₂Q₃ − Q₁Q₃ evaluates to either +1 or −3 for every sign assignment, so its ensemble average obeys −3 ≤ K₃ ≤ 1. A qubit under unitary evolution gives Cᵢⱼ = cos(ΩΔt); maximizing K₃ = 2cos(Ωτ) − cos(2Ωτ) yields 3/2 at Ωτ = π/3, the temporal Tsirelson bound.
What is the clumsiness loophole?
A violation of the LGI could mean the system genuinely lacks macroscopic realism, OR that it is macrorealistic but your measurement was invasive simply because of imperfect experimental technique ('clumsiness'). Because non-invasiveness is an assumption that cannot be geometrically guaranteed, skeptics can always attribute a violation to disturbance. Experiments address this using ideal negative-result measurements and adversarial control tests, but it cannot be fully eliminated.
What was the first convincing experimental violation?
Palacios-Laloy, Mallet, Vion, Esteve and collaborators in 2010 used a superconducting qubit coupled to a microwave resonator, read out by weak continuous measurement. They reconstructed the two-time correlators from the measurement noise spectrum and found K₃ close to the quantum bound of 3/2, exceeding the macrorealist limit of 1.
Can the Leggett-Garg violation exceed 3/2?
For a two-level system with projective measurements, 3/2 is the maximum (the temporal Tsirelson bound). But in three-level (qutrit) and higher-dimensional systems, using suitable projective measurements and state-update rules, the three-term K₃ can be pushed beyond 3/2, approaching the algebraic maximum of 3. This shows the temporal correlation budget grows with Hilbert-space dimension.
Does violating the LGI prove the macroscopic world is quantum?
It proves that macroscopic realism plus non-invasive measurability cannot both hold for that system in that regime. It rules out the classical picture of definite-valued, undisturbed evolution over the tested interval, extending the domain of demonstrated quantum coherence. But because the non-invasiveness assumption is not loophole-free, it is a strong constraint rather than an absolute proof, and it tests collapse models by pushing to ever larger, more macroscopic systems.