Quantum Mechanics
Weak Values and the AAV Amplification: Measuring Without Disturbing
In 1988 three physicists showed that a "measurement" of the spin of a spin-½ particle — a quantity that can only be ±½ℏ — could yield the value 100. The trick was not a paradox but a new observable: the weak value, obtained by coupling an apparatus so gently that it barely disturbs the system, then keeping only runs that end in a chosen final state. Because the outcome can lie far outside the operator's eigenvalue spectrum, a tiny physical coupling can produce a huge, easily-read pointer shift.
Weak-value amplification (WVA) turns this into a metrology tool: by pre- and post-selecting nearly orthogonal states, experimenters have resolved mirror tilts of 560 femtoradians, beam displacements of ~1 ångström, and piezo travel of 20 femtometers — all with modest laser hardware.
- RegimeWeak system–pointer coupling, g → 0
- Key relationA_w = ⟨ψ_f|Â|ψ_i⟩ / ⟨ψ_f|ψ_i⟩
- IntroducedAharonov, Albert & Vaidman, 1988 (PRL 60, 1351)
- First realizedRitchie, Story & Hulet, 1991 (PRL 66, 1107)
- Characteristic scale560 frad tilt; ~1 Å beam shift; 20 fm travel
- Matters forPrecision metrology, wavefunction tomography, foundations
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 a weak value is, and why it matters
A projective measurement of an observable  can only return one of its eigenvalues, and it collapses the state onto the corresponding eigenvector. Aharonov, Albert, and Vaidman (AAV, 1988) asked a different question: what does a very gentle measuring device read on average if you keep only the runs where the system is later found in a specific final state |ψ_f⟩? The answer is the weak value
A_w = ⟨ψ_f|Â|ψ_i⟩ / ⟨ψ_f|ψ_i⟩,
a generally complex number that need not lie between the smallest and largest eigenvalues. For a spin-½ component (eigenvalues ±½), suitably chosen |ψ_i⟩ and |ψ_f⟩ give A_w = 100 — the title of the founding paper. This matters for two reasons. Conceptually, it sharpens what "measuring without disturbing" can mean and how pre- and post-selection jointly constrain a quantum system. Practically, the anomalous magnitude of A_w is an amplifier: a coupling too small to see directly is multiplied into a large, readable pointer deflection, enabling ultra-sensitive metrology with ordinary optics.
The mechanism, step by step
Model the apparatus as a von Neumann pointer with position q and conjugate momentum p, coupled to the system by the interaction Hamiltonian H = g·Â⊗p̂ acting impulsively. Start with system in |ψ_i⟩ and pointer in a broad Gaussian of width σ. In a strong measurement the pointer shifts by the eigenvalue and the state collapses. In the weak limit g·(spread of Â) ≪ σ, the pointer barely moves and entanglement is tiny.
Now post-select on |ψ_f⟩. Expanding the joint evolution to first order in g, the surviving pointer wavefunction is displaced by g·A_w. Because A_w = ⟨ψ_f|Â|ψ_i⟩/⟨ψ_f|ψ_i⟩, choosing |ψ_f⟩ nearly orthogonal to |ψ_i⟩ sends the denominator ⟨ψ_f|ψ_i⟩ → 0 and blows up the ratio. The real part of A_w shifts the pointer's position; the imaginary part shifts its conjugate momentum (or, in optics, imprints a phase/transverse-momentum kick). The price is post-selection probability |⟨ψ_f|ψ_i⟩|², which shrinks exactly as the amplification grows — you throw away most events, keeping the rare, information-rich ones.
The key relation, scales, and the amplification tradeoff
The core formula is the AAV weak value, with the measured pointer shift δq = g·Re(A_w) and a conjugate shift δp = 2g·σ_p²·Im(A_w). Validity requires the weakness conditions g|A_w| ≪ σ and higher-order terms g²⟨²⟩_w to stay small — otherwise the linear expansion fails and the shift saturates.
Amplification A = A_w scales roughly as 1/⟨ψ_f|ψ_i⟩ = 1/sin(ε) for a pre/post-selection misalignment angle ε, while the retained fraction of events falls as sin²(ε). The signal-to-noise ratio for a shot-noise-limited pointer is, to leading order, independent of ε: the amplified signal and the reduced statistics cancel. WVA's real advantages are therefore practical — it moves a signal away from technical noise (1/f, jitter, saturation) into a clean regime, and shrinks required detector dynamic range. Characteristic achieved sensitivities: angular tilt ~560 femtoradians (5.6×10⁻¹³ rad), transverse beam displacement ~1 Å (10⁻¹⁰ m), and longitudinal travel ~20 fm, with SNR gains near 150 in interferometric implementations.
How it is realized and measured
The first realization came from Ritchie, Story, and Hulet in 1991 (PRL 66, 1107): a laser beam passed through a birefringent crystal that split the two linear polarizations by a distance far smaller than the beam waist (the weak measurement of polarization), then a polarizer post-selected a nearly crossed state, translating the beam centroid by an amount far larger than the birefringent splitting — the optical AAV effect, with photons (spin-1) standing in for AAV's spin-½.
The metrology era opened with Hosten and Kwiat (2008, Science 319, 787), who used pre/post-selected polarization to amplify the spin-Hall shift of light at an air–glass interface by nearly four orders of magnitude, reaching ~1 Å sensitivity. Dixon, Starling, Jordan, and Howell (2009, PRL 102, 173601) built a Sagnac interferometer whose which-path states play the role of pre/post-selection, measuring mirror tilts to 560 frad and 20 fm of piezo travel. The pointer is simply a split-detector or CCD reading the beam's transverse position or momentum.
Where it operates, and distinctions from related effects
WVA works in any system with a controllable two-level (or few-level) "which-observable" degree of freedom coupled weakly to a continuous pointer: photon polarization coupled to transverse position, atomic spin coupled to a light beam, superconducting-qubit readout, and neutron interferometry. It is not a violation of the uncertainty principle — each individual weak measurement carries almost no information and enormous variance; only the conditioned ensemble average reveals A_w.
Distinguish it from three neighbors. (1) Standard amplification boosts signal and noise together; WVA rearranges where noise sits. (2) Quantum non-demolition (QND) measurement preserves an eigenbasis exactly and can be strong; weak measurement is gentle but generic. (3) The quantum Zeno effect uses repeated strong measurements to freeze evolution — the opposite limit. Also, the anomalous magnitude and complexity of A_w are tied to contextuality: negative or out-of-range weak values are a proof-of-contextuality signature, formalized by Pusey and demonstrated experimentally (PRL 116, 180401, 2016).
Applications, open questions, and significance
Beyond ultra-sensitive deflection and phase metrology, weak values enabled direct measurement of the wavefunction: Lundeen and colleagues (2011, Nature 474, 188) weakly measured transverse position and post-selected zero momentum, making the real and imaginary parts of ψ appear directly on the meter. Kocsis et al. (2011, Science 332, 1170) reconstructed the average photon trajectories in a two-slit interferometer — Physics World's breakthrough of the year — by weakly measuring momentum and post-selecting position, operationally realizing Bohmian streamlines.
Open questions remain contentious. Because shot-noise-limited SNR is ε-independent, whether WVA gives a genuine metrological advantage over conventional strategies — or merely a practical convenience against technical noise and detector saturation — is debated (Ferrie–Combes, Jordan et al., 2014). Ongoing work probes technical-noise suppression, imaginary weak values for frequency/phase sensing, and the interpretation of anomalous values as contextuality witnesses. The concept endures as both a metrology technique and a lens on quantum measurement itself.
| Property | Standard projective measurement | Weak value measurement |
|---|---|---|
| Coupling strength g | Strong (fully entangles pointer & system) | Weak (g·σ ≪ pointer width) |
| Disturbance to state | Collapses to an eigenstate | Negligible per run; state nearly intact |
| Possible outcomes | Eigenvalues of  only | Complex A_w, can lie outside the spectrum |
| Selection | Pre-selection only | Pre- AND post-selection (conditioning) |
| Info per run | O(1) bit, one eigenvalue | Tiny; needs large ensemble to average |
| Pointer shift | δ = ⟨Â⟩ | δ = g·Re(A_w); Im(A_w) shifts conjugate variable |
Frequently asked questions
Does a weak value of 100 for a spin-½ component violate quantum mechanics?
No. A single weak measurement barely couples the pointer to the spin, so any one run yields almost no information and huge variance; no individual outcome is ever 100. The value 100 is the conditioned average over a large pre- and post-selected ensemble, and the formula A_w = ⟨ψ_f|Â|ψ_i⟩/⟨ψ_f|ψ_i⟩ can exceed the eigenvalue range precisely because it is not an eigenvalue but a weak value.
Why can the weak value be complex, and what does the imaginary part do?
Because A_w is a ratio of complex amplitudes, not an expectation value. The real part shifts the pointer's measured variable (e.g. position: δq = g·Re A_w). The imaginary part shifts the conjugate variable (momentum/phase: δp ∝ g·Im A_w). In optics, Im(A_w) shows up as a transverse-momentum or angular kick, which is what interferometric phase-based WVA schemes exploit.
Does weak-value amplification actually beat standard measurement in sensitivity?
For a shot-noise-limited pointer, the signal-to-noise ratio is essentially independent of the amplification because throwing away most events cancels the boosted signal. WVA's practical wins come from moving the signal into a low-technical-noise regime, avoiding detector saturation, and reducing required dynamic range. Whether it offers a fundamental advantage is genuinely debated (Ferrie–Combes vs. Jordan et al., 2014).
What is the role of post-selection, and what is its cost?
Post-selection conditions on finding the system in |ψ_f⟩ after the weak coupling. Choosing |ψ_f⟩ nearly orthogonal to |ψ_i⟩ drives the denominator ⟨ψ_f|ψ_i⟩ toward zero, amplifying A_w. The cost is the success probability |⟨ψ_f|ψ_i⟩|², which falls as the square of the same overlap — so amplification and event yield trade off directly, and most runs are discarded.
How does this differ from the quantum Zeno effect or QND measurement?
The Zeno effect uses frequent strong measurements to freeze evolution — the strong-coupling limit. QND measurement can be strong but is engineered to preserve a chosen eigenbasis exactly. Weak measurement is the opposite of Zeno: minimal coupling per interaction so the state is barely disturbed, with information extracted from the ensemble average rather than any single collapse.
Where was weak-value amplification first demonstrated experimentally?
Ritchie, Story, and Hulet (PRL 66, 1107, 1991) made the first realization using a birefringent crystal to weakly separate a laser beam's polarizations, then post-selecting a nearly crossed polarizer to shift the beam centroid far more than the actual splitting. Landmark metrology followed: Hosten–Kwiat's spin-Hall-of-light detection (2008, ~1 Å) and Dixon et al.'s 560-femtoradian Sagnac deflectometer (2009).