Weak Value in Quantum Measurements

• Weak Value is a complex number defined for quantum observables between pre- and post-selected states and can exceed the observable's eigenvalue spectrum.
• It provides operational insights via pointer shifts in weak von Neumann measurements, linking transition amplitudes with measurable physical effects.
• Geometric formulations in qubit systems and metrological amplification demonstrate its practical applications despite ongoing interpretational controversies.

Searching arXiv for recent and foundational papers on weak values to ground the article. Search query: weak value quantum measurement weak measurement postselection site:arXiv.org A weak value is the complex number assigned to an observable $A$ for a pre-selected state and a post-selected state, defined whenever the overlap of those states is nonzero by

$$A_w=\frac{\langle \psi_f|A|\psi_i\rangle}{\langle \psi_f|\psi_i\rangle}.$$

It first appeared in the research literature in 1987 and 1988, and was originally introduced through weak, minimally disturbing measurements combined with post-selection. In the special case $|\psi_f\rangle=|\psi_i\rangle$, the weak value reduces to the ordinary expectation value, while if both states coincide with an eigenvector of $A$, it reproduces the corresponding eigenvalue. In general, however, $A_w$ can be complex and can lie outside the eigenspectrum of $A$, which is the source of both its utility and its controversy (Barandes, 10 Feb 2026).

1. Definition within pre- and post-selected quantum mechanics

The standard setting is a two-state description: a system is prepared at an initial time in a state $|\psi_i\rangle$, later post-selected in a state $|\psi_f\rangle$, and an observable is considered at an intermediate time. In the presence of nontrivial dynamics, the weak value is written in the time-symmetric form

$${}_{f}A_{i}^{w} = \frac{\langle\psi_f|\,U(t_f,t)\,A\,U(t,t_i)\,|\psi_i\rangle} {\langle\psi_f|\,U(t_f,t_i)\,|\psi_i\rangle},$$

provided the denominator is nonzero. This formulation makes explicit that the quantity depends on both forward and backward evolution, not on the initial state alone (Shikano, 2011).

Several elementary identities delimit the formal status of the weak value. If the post-selection equals the pre-selection, then $A_w=\langle A\rangle$. If the post-selected state is an eigenstate of $$A_w=\frac{\langle \psi_f|A|\psi_i\rangle}{\langle \psi_f|\psi_i\rangle}.$$0, then the weak value equals the corresponding eigenvalue. Averaging weak values over a complete set of post-selections reproduces the ordinary expectation value, which places weak values within the standard Hilbert-space formalism rather than outside it (Svensson, 2013).

Because the denominator $$A_w=\frac{\langle \psi_f|A|\psi_i\rangle}{\langle \psi_f|\psi_i\rangle}.$$1 can be made arbitrarily small by choosing nearly orthogonal pre- and post-selections, weak values need not be bounded by the spectrum of $$A_w=\frac{\langle \psi_f|A|\psi_i\rangle}{\langle \psi_f|\psi_i\rangle}.$$2. This is not a convex average over eigenvalues; it is a ratio of transition amplitudes. That algebraic feature underlies anomalous real weak values, complex weak values, and the amplification effects used in precision measurement (Vaidman, 2017).

2. Measurement theory and operational formulations

In the original Aharonov–Albert–Vaidman scheme, the observable is coupled weakly to a pointer through a von Neumann interaction such as

$$A_w=\frac{\langle \psi_f|A|\psi_i\rangle}{\langle \psi_f|\psi_i\rangle}.$$3

For a broad initial pointer state and sufficiently small coupling, first-order expansion gives an effective translation generated by the weak value. The mean shift of the pointer position is proportional to $$A_w=\frac{\langle \psi_f|A|\psi_i\rangle}{\langle \psi_f|\psi_i\rangle}.$$4, while the conjugate momentum shift is proportional to $$A_w=\frac{\langle \psi_f|A|\psi_i\rangle}{\langle \psi_f|\psi_i\rangle}.$$5. In one common Gaussian-pointer formulation,

$$A_w=\frac{\langle \psi_f|A|\psi_i\rangle}{\langle \psi_f|\psi_i\rangle}.$$6

This is the canonical operational route by which weak values are inferred from conditioned pointer statistics (Vaidman, 2017).

Later work generalized this picture beyond the usual weak-coupling approximation. In conditioned von Neumann measurements with arbitrary coupling strength $$A_w=\frac{\langle \psi_f|A|\psi_i\rangle}{\langle \psi_f|\psi_i\rangle}.$$7, arbitrary initial states, and arbitrary conditioning, detector averages can be written exactly in terms of generalized weak values in the joint Hilbert space. In particular,

$$A_w=\frac{\langle \psi_f|A|\psi_i\rangle}{\langle \psi_f|\psi_i\rangle}.$$8

and higher detector moments admit analogous expansions. In that sense, weak-value structure is not confined to infinitesimal coupling; it organizes conditioned von Neumann measurements more broadly (Dressel et al., 2012).

A distinct development is the probe-free operational formulation. Instead of introducing an ancillary pointer, one applies a small transformation directly to the system,

$$A_w=\frac{\langle \psi_f|A|\psi_i\rangle}{\langle \psi_f|\psi_i\rangle}.$$9

and studies the normalized post-selection amplitude

$|\psi_f\rangle=|\psi_i\rangle$0

Its first-order coefficient yields

$|\psi_f\rangle=|\psi_i\rangle$1

Within this formulation, the weak value is the first-order sensitivity of the post-selection amplitude to an infinitesimal transformation applied to the system itself. The real part is obtained from the slope of the normalized post-selection probability,

$|\psi_f\rangle=|\psi_i\rangle$2

while the imaginary part appears as an interferometric phase shift linear in $|\psi_f\rangle=|\psi_i\rangle$3. No probe system is required (Ogawa et al., 2019).

3. Geometric structure, especially for qubits

For qubits, weak values admit a particularly explicit geometric description. Let $|\psi_f\rangle=|\psi_i\rangle$4 denote the real vector space of $|\psi_f\rangle=|\psi_i\rangle$5 Hermitian matrices. Given a pre- and post-selected pair $|\psi_f\rangle=|\psi_i\rangle$6, one defines the linear “weak function”

$|\psi_f\rangle=|\psi_i\rangle$7

which maps Hermitian operators to complex numbers. Decomposing $|\psi_f\rangle=|\psi_i\rangle$8 into trace and trace-zero parts shows that the nontrivial structure resides in the three-dimensional subspace

$|\psi_f\rangle=|\psi_i\rangle$9

In Pauli coordinates, each $A$0 corresponds to a vector in $A$1, and $A$2 carries the Euclidean inner product

$A$3

This gives a Bloch-space formulation of the weak-value map (Farinholt et al., 2015).

The pre- and post-selected pure states determine a two-dimensional plane in $A$4,

$A$5

Any trace-zero Hermitian operator can be decomposed uniquely into a component in this “PPS-plane” and a component orthogonal to it. For $A$6, where $A$7 is associated with a state mutually unbiased to both $A$8 and $A$9, the weak value splits as

$A_w$0

The geometric consequence is that the in-plane component contributes only to the real part, whereas the perpendicular component contributes only to the imaginary part (Farinholt et al., 2015).

This geometry also yields extremal statements. Restricting to rank-one projectors on the Bloch sphere, the extremal real weak values are

$A_w$1

while the extremal imaginary parts are $A_w$2. This makes the dependence on small post-selection overlap geometrically transparent: nearly orthogonal pre- and post-selections enlarge the accessible weak-value range (Farinholt et al., 2015).

4. Anomalous weak values and canonical examples

The defining anomaly of weak values is that they can exceed eigenvalue bounds or become complex. In a spin-$A_w$3 example with $A_w$4, choosing

$A_w$5

gives

$A_w$6

which becomes arbitrarily large as $A_w$7. The mechanism is the small denominator $A_w$8 with finite numerator $A_w$9 (Vaidman, 2017).

Projector weak values provide the most widely discussed paradoxical cases. In the three-box setup with

$A$0

the projector weak values are

$A$1

In the probe-free operational formulation, inserting a tiny attenuation on box $A$2 yields a slope proportional to $A$3, which is operationally described as if “minus one particle” were affecting the post-selection probability (Svensson, 2013).

Weak values also admit a direct connection to interference. In the double-slit analysis, the momentum weak value takes the form

$A$4

Its imaginary part functions as an interference index: it vanishes at bright fringes and diverges at dark fringes. By contrast, the real part of the position weak value $A$5 behaves as a conditioned trajectory variable. The paper interprets this split by associating the real part with the particle aspect and the imaginary part with the wave aspect, while explicitly stating that this involves no conflict with complementarity (Mori et al., 2014).

A classical stochastic analogue clarifies the algebraic source of these anomalies. In a two-time-conditioned Markov process satisfying detailed balance, the analogue of the weak value is the two-time conditional expectation

$A$6

When expressed in an eigenbasis of the real symmetric generator, the conditioned weights can become negative, and $A$7 can exceed the spectral range of $A$8. In that construction, the origin of the anomaly is entirely real: two-time conditioning, sign cancellations under basis change, and division by a small overlap (Tomita, 2012).

5. Amplification, metrology, and noise

Weak-value amplification exploits the fact that a tiny physical coupling $A$9 can be converted into a much larger pointer displacement $|\psi_i\rangle$0 on the post-selected subensemble. This strategy underlies optical experiments such as the spin Hall effect of light via polarization post-selection, Sagnac-interferometer beam-deflection measurements, and phase estimation with white light. In such settings, the mean position shift tracks $|\psi_i\rangle$1, the conjugate variable tracks $|\psi_i\rangle$2, and large anomalous weak values can amplify small effects above a technical-noise floor (Vaidman, 2017).

A substantial literature has nevertheless shown that amplification per detected event does not, by itself, imply superior estimation performance. Fisher-information analyses summarized in the survey literature establish that, under ideal detection and standard resource accounting, the total Fisher information obtained after post-selection cannot exceed that of the corresponding optimal non-postselected strategy. In that framework, post-selection reduces the number of retained trials, and the apparent gain in per-event sensitivity is offset by the success probability (Knee et al., 2014).

More refined analyses distinguish noise models. In phase-space language, weak-value protocols rigidly shift the Wigner distribution of the meter, which allows detector imperfections to be studied through “Fisher information efficiency” functions. That approach shows why weak-value techniques are especially effective against detector saturation, less effective against transverse jitter, and only modestly different under pixelation. The central point is not a violation of the quantum Fisher-information bound under ideal detection, but a redistribution of practically accessible information under detector nonidealities (Knee et al., 2018).

Recent work also establishes a strict robustness advantage when the dominant noise acts on the primary system rather than on the detector. For unital noise channels, the weak-value measurement protocol is proved to be quadratically more robust to noise than strong measurements: the weak-value bias can be reduced to $|\psi_i\rangle$3, whereas strong-measurement protocols retain an $|\psi_i\rangle$4 bias floor. For amplitude-damping and phase-damping channels, the same work shows a quadratic advantage even over strong measurement protocols that are themselves allowed to use postselection (Schwartzman-Nowik et al., 2024).

Application-specific studies continue to exploit these features. In a Faraday magneto-optic scheme using a modified Sagnac interferometer with the probe in momentum space, numerical results show that weak-value amplification is effective and feasible for detecting weak magnetic fields with magnetic-field intensity lower than $|\psi_i\rangle$5, given appropriate pre-selection, post-selection, optical structure, and spectrometer constraints (Huang et al., 2021).

6. Generalizations, analogues, and alternative protocols

Weak values extend naturally beyond pure-state pre- and post-selection. For a mixed pre-selected state $|\psi_i\rangle$6 and mixed backward state $|\psi_i\rangle$7, one form given in the literature is

$|\psi_i\rangle$8

while a related expression for mixed pre-selection is

$|\psi_i\rangle$9

These generalizations are useful for diagnosing noise and for distinguishing pure two-state-vector behavior from genuinely statistical mixtures (Vaidman et al., 2016).

Alternative experimental and conceptual frameworks also broaden the notion. In collision-theory treatments of the Stern–Gerlach apparatus, inclusion of translational dynamics and kinetic energy leads to a “weak vector”

$|\psi_f\rangle$0

so that the beam momentum shift becomes a full vector displacement rather than a single-component pointer translation. Higher-rank “weak tensors” arise at higher order in the perturbative expansion (Castro et al., 2018).

A distinct two-step protocol is the null weak value. There the first measurement is not weak in the usual von Neumann sense; it is strong but rare, implemented with small probability $|\psi_f\rangle$1. Correlating the first click probability with a null outcome in the final strong postselection yields

$|\psi_f\rangle$2

The proposal is motivated by signal-to-noise considerations different from those of standard weak values, and it exemplifies how post-selection-based amplification can persist outside the Aharonov–Albert–Vaidman protocol (Zilberberg et al., 2013).

Even ontological reinterpretations have been explored in restricted regimes. In a quantum-optical experiment involving post-selected photon-number weak values, a stochastic-optics toy model reproduces the observed amplified conditional intensity shift as a conditional mean of an underlying field intensity, but only in a regime where the weak value is not truly anomalous. The same model breaks down precisely when the quantum weak value predicts negative or otherwise strongly anomalous results (Sinclair et al., 2018).

7. Interpretational status and controversy

The central interpretational dispute concerns whether a weak value should be regarded as a bona fide property of an individual pre- and post-selected system or only as an ensemble-level quantity extracted from a particular protocol. One influential position argues that weak values are robust single-system properties. In that view, during an infinitesimal interaction a system with weak value $|\psi_f\rangle$3 affects a pointer as if it were in an eigenstate of $|\psi_f\rangle$4 with eigenvalue $|\psi_f\rangle$5, and this differs sharply from the behavior of a merely pre-selected state with the same ordinary expectation value. Experimental comparisons based on photon polarization and Bures distance are presented in support of that claim (Vaidman et al., 2016).

A closely related line of argument emphasizes the operational formulation without probes. There, the weak value is the derivative of the post-selection amplitude under an infinitesimal system transformation, so anomalous weak values are interpreted as unusual sensitivities of $|\psi_f\rangle$6 to the generator $|\psi_f\rangle$7. This makes strange values appear as over-sensitive or under-sensitive transition amplitudes within the system’s own Hilbert space, rather than as obscure artifacts of an auxiliary pointer (Ogawa et al., 2019).

Critics reject the move from operational measurability to single-system ontology. One objection is formal: standard quantum axioms single out self-adjoint operators, spectra, Born probabilities, and density operators, but not amplitude ratios such as $|\psi_f\rangle$8, as intrinsic properties of individual systems. Another objection is contextual: the value changes with the chosen post-selection. More recent criticism systematizes these concerns as an ensemble fallacy, a post-selection fallacy, and a measurementist fallacy, arguing that the measurability of a quantity on a post-selected ensemble does not by itself warrant its reification as a new single-system observable (Svensson, 2013).

The controversy therefore has two layers. Operationally, weak values are well-defined and experimentally accessible; they organize weak measurements, conditioned measurements, interferometric phase shifts, and several metrological protocols. Ontologically, however, no universal consensus exists. Some authors treat weak values as physical properties of single pre- and post-selected systems, while others regard them as structured conditional amplitude ratios whose striking features are inseparable from ensemble averaging and post-selection (Vaidman, 2017).