Papers
Topics
Authors
Recent
Search
2000 character limit reached

Weak Values in Quantum Mechanics

Updated 25 October 2026
  • Weak values are complex, context-dependent quantities in quantum mechanics, often used in weak measurement techniques and defined for nonorthogonal states and a quantum observable with the equation{\displaystyle}A_w={\frac{\{\langle}\psi_f|{\hat A}|{ \psi_i}\rangle\}{\{\langle}\psi_f|{ \psi_i}\rangle\}}.{It is a transition amplitude-alike operator response that can be amplified by post-selection.
  • Complications from weak value arising from the evolution of it within the Hilbert space, Therefore without implementing different kind of technics weak values are always subject to interpretation debate, it is defined at length on each quantum mechanics platform to adjust to any observable.
  • In optical interferometry, weak values determine the transverse position and momentum shifts that are necessary to prevent substantially interfering with the quantum system used in measurement.

Weak values are complex, generally context-dependent quantities associated with a quantum observable, a pre-selected state, and a post-selected state. For nonorthogonal states ∣ψi⟩\lvert\psi_i\rangle and ∣ψf⟩\lvert\psi_f\rangle, the standard weak value of an operator A^\hat A is

Aw=⟨ψf∣A^∣ψi⟩⟨ψf∣ψi⟩.A_w=\frac{\langle\psi_f|\hat A|\psi_i\rangle}{\langle\psi_f|\psi_i\rangle}.

The numerator is a transition matrix element and the denominator is the transition amplitude for the post-selection. A weak value need not be real, need not lie within the spectrum of a Hermitian operator, and is not generally an eigenvalue or an ordinary expectation value. Its operational manifestations include conditioned pointer shifts, post-selection-probability responses, phase responses, phase-space displacements, local momentum fields, and effective variables in reduced quantum dynamics. The meaning of these manifestations remains the subject of sustained interpretational debate.

1. Definition, terminology, and algebraic structure

The weak value is defined whenever ⟨ψf∣ψi⟩≠0\langle\psi_f|\psi_i\rangle\neq0. If the post-selected state is an eigenstate of A^\hat A,

A^∣a⟩=a∣a⟩,\hat A\lvert a\rangle=a\lvert a\rangle,

then

⟨a∣A^∣ψi⟩⟨a∣ψi⟩=a.\frac{\langle a|\hat A|\psi_i\rangle}{\langle a|\psi_i\rangle}=a.

In this special case the weak value coincides with an eigenvalue. If the post-selection is on an eigenstate ∣bi⟩\lvert b_i\rangle of another, noncommuting operator B^\hat B, then

∣ψf⟩\lvert\psi_f\rangle0

which is generally neither an eigenvalue nor a positive-probability average of eigenvalues.

For a complete orthonormal set ∣ψf⟩\lvert\psi_f\rangle1,

∣ψf⟩\lvert\psi_f\rangle2

Thus an ordinary expectation value can be represented as a weighted sum of transition amplitudes. The algebraic identity does not, by itself, establish that the summands are conditional probabilities. When the relevant projectors do not commute, products such as ∣ψf⟩\lvert\psi_f\rangle3 are not generally projectors, and their associated ratios need not lie between zero and one (Mochizuki, 2013).

The terminology is not uniform. “Weak value” originally referred to the Aharonov–Albert–Vaidman (AAV) protocol involving a minimally disturbing measurement followed by post-selection. Algebraically, however, the ratio can be defined independently of a particular measurement regime. In the Clifford- and Moyal-algebra treatment, the term is explicitly used for a transition probability amplitude, with “weak” not denoting a special measurement strength (Hiley, 2011). Exact generalized weak-value representations also exist for arbitrary coupling, mixed or entangled initial states, and generalized postselections (Dressel et al., 2012).

For density operators, a generalized weak value is commonly written

∣ψf⟩\lvert\psi_f\rangle4

For genuinely mixed pre- and post-selected states, one form is

∣ψf⟩\lvert\psi_f\rangle5

These generalizations distinguish mixed-state weak values from weak values associated with a pure two-state vector or a generalized two-state vector embedded in a larger pure description (Vaidman et al., 2016).

2. Weak-measurement protocol and detector response

The standard implementation uses a von Neumann interaction between a system observable ∣ψf⟩\lvert\psi_f\rangle6 and a meter observable. With pointer position ∣ψf⟩\lvert\psi_f\rangle7 and conjugate momentum ∣ψf⟩\lvert\psi_f\rangle8,

∣ψf⟩\lvert\psi_f\rangle9

and

A^\hat A0

For small coupling, the unitary is expanded as

A^\hat A1

Post-selection on A^\hat A2 gives an unnormalized meter state

A^\hat A3

For a suitable Gaussian pointer, A^\hat A4 produces a position shift and A^\hat A5 produces a conjugate-momentum shift or a change in the pointer distribution. The first-order position response has the form

A^\hat A6

while the momentum response depends on the pointer state and its position-momentum correlations.

The weak approximation is not simply the condition A^\hat A7. It requires the interaction-induced correction to remain small over the relevant pointer support. For nearly orthogonal pre- and post-selections, A^\hat A8 can be large, so the coupling must be reduced correspondingly. A representative Gaussian-meter condition is

A^\hat A9

where Aw=⟨ψf∣A^∣ψi⟩⟨ψf∣ψi⟩.A_w=\frac{\langle\psi_f|\hat A|\psi_i\rangle}{\langle\psi_f|\psi_i\rangle}.0 is the meter width (Knee et al., 2018). The post-selection probability is simultaneously reduced:

Aw=⟨ψf∣A^∣ψi⟩⟨ψf∣ψi⟩.A_w=\frac{\langle\psi_f|\hat A|\psi_i\rangle}{\langle\psi_f|\psi_i\rangle}.1

The measurement is not disturbance-free. The first-order interaction both produces the signal and alters the post-selection statistics. One analysis identifies

Aw=⟨ψf∣A^∣ψi⟩⟨ψf∣ψi⟩.A_w=\frac{\langle\psi_f|\hat A|\psi_i\rangle}{\langle\psi_f|\psi_i\rangle}.2

where Aw=⟨ψf∣A^∣ψi⟩⟨ψf∣ψi⟩.A_w=\frac{\langle\psi_f|\hat A|\psi_i\rangle}{\langle\psi_f|\psi_i\rangle}.3 is the post-selection probability after the weak coupling. On this account, the real weak value is an indicator of the first-order back-action of the weak measurement on the selected ensemble (Mochizuki, 2013). The imaginary part modifies pointer momentum or width and is likewise associated with a disturbance-sensitive response.

The distinction between weak values and weak measurements is important. A weak value is fixed by the observable and the pre- and post-selected states; a weak measurement is one possible method for determining it. Exact conditioned detector averages at arbitrary coupling can be expressed through generalized weak values defined on the joint system–detector Hilbert space:

Aw=⟨ψf∣A^∣ψi⟩⟨ψf∣ψi⟩.A_w=\frac{\langle\psi_f|\hat A|\psi_i\rangle}{\langle\psi_f|\psi_i\rangle}.4

where

Aw=⟨ψf∣A^∣ψi⟩⟨ψf∣ψi⟩.A_w=\frac{\langle\psi_f|\hat A|\psi_i\rangle}{\langle\psi_f|\psi_i\rangle}.5

For example,

Aw=⟨ψf∣A^∣ψi⟩⟨ψf∣ψi⟩.A_w=\frac{\langle\psi_f|\hat A|\psi_i\rangle}{\langle\psi_f|\psi_i\rangle}.6

These relations hold for arbitrary coupling strength, arbitrary detector states, mixed or entangled initial states, and generalized postselections (Dressel et al., 2012).

3. Real and imaginary parts: interference, phase space, and local quantities

Weak values are ratios of coherent amplitudes rather than averages with positive classical weights. If

Aw=⟨ψf∣A^∣ψi⟩⟨ψf∣ψi⟩.A_w=\frac{\langle\psi_f|\hat A|\psi_i\rangle}{\langle\psi_f|\psi_i\rangle}.7

then

Aw=⟨ψf∣A^∣ψi⟩⟨ψf∣ψi⟩.A_w=\frac{\langle\psi_f|\hat A|\psi_i\rangle}{\langle\psi_f|\psi_i\rangle}.8

The denominator can be small because of destructive interference, producing an anomalously large value. The associated real conditional quasiprobability can be written

Aw=⟨ψf∣A^∣ψi⟩⟨ψf∣ψi⟩.A_w=\frac{\langle\psi_f|\hat A|\psi_i\rangle}{\langle\psi_f|\psi_i\rangle}.9

so that

⟨ψf∣ψi⟩≠0\langle\psi_f|\psi_i\rangle\neq00

The quasiprobability is normalized but may be negative. Its complex counterpart is associated with the Kirkwood–Dirac distribution, while the real form is a Terletsky–Margenau–Hill distribution (Dressel, 2014).

For a position post-selection, the momentum weak value is

⟨ψf∣ψi⟩≠0\langle\psi_f|\psi_i\rangle\neq01

Writing

⟨ψf∣ψi⟩≠0\langle\psi_f|\psi_i\rangle\neq02

gives

⟨ψf∣ψi⟩≠0\langle\psi_f|\psi_i\rangle\neq03

Hence

⟨ψf∣ψi⟩≠0\langle\psi_f|\psi_i\rangle\neq04

The real part is the Bohmian or phase-gradient momentum, while the imaginary part is an amplitude-gradient or osmotic contribution. In the ⟨ψf∣ψi⟩≠0\langle\psi_f|\psi_i\rangle\neq05 convention, the corresponding expression is

⟨ψf∣ψi⟩≠0\langle\psi_f|\psi_i\rangle\neq06

The same decomposition can be obtained from Clifford density elements, left- and right-acting algebraic operations, and Moyal star products (Hiley, 2011).

In phase space, the weak value corresponds to a conditional average over a quasiprobability. The Moyal formulation separates the real and imaginary momentum contributions through Baker and Moyal brackets. For the Wigner–Moyal distribution ⟨ψf∣ψi⟩≠0\langle\psi_f|\psi_i\rangle\neq07,

⟨ψf∣ψi⟩≠0\langle\psi_f|\psi_i\rangle\neq08

and

⟨ψf∣ψi⟩≠0\langle\psi_f|\psi_i\rangle\neq09

The complementary Moyal-bracket contribution is proportional to A^\hat A0.

Second-order weak values similarly yield kinetic and quantum-potential terms. For a general mass,

A^\hat A1

The two terms are the Bohm kinetic energy and quantum potential,

A^\hat A2

In general,

A^\hat A3

In particular, the square of a weak value is not generally the weak value of the squared operator (Oianguren-Asua et al., 2023).

For qubits, the geometry is explicit. A Hermitian operator can be written

A^\hat A4

while pure pre- and post-selected states have Bloch vectors A^\hat A5 and A^\hat A6. The weak value is

A^\hat A7

The real part is controlled by the component of A^\hat A8 in the plane spanned by A^\hat A9 and A^∣a⟩=a∣a⟩,\hat A\lvert a\rangle=a\lvert a\rangle,0; the imaginary part is controlled by the perpendicular component A^∣a⟩=a∣a⟩,\hat A\lvert a\rangle=a\lvert a\rangle,1. The plane is called the pre-/post-selection, or PPS, plane. A traceless observable has a real weak value exactly when its representative lies in this plane (Farinholt et al., 2015).

4. Applications and physical implementations

Weak values have been applied to local quantum fields, optical interferometry, collision theory, quantum-state tomography, quantum metrology, and weakly coupled internal and motional degrees of freedom.

In a Mach–Zehnder interferometer, a path projector such as

A^∣a⟩=a∣a⟩,\hat A\lvert a\rangle=a\lvert a\rangle,2

can be weakly coupled to a transverse probe coordinate. Conditioned on output port A^∣a⟩=a∣a⟩,\hat A\lvert a\rangle=a\lvert a\rangle,3,

A^∣a⟩=a∣a⟩,\hat A\lvert a\rangle=a\lvert a\rangle,4

The real part is inferred from the average transverse position shift, while the imaginary part is inferred from the transverse momentum shift. Combining position and momentum shifts from both output ports gives a quantum-Bayes reconstruction of weak values and of the relative amplitudes of the initial interferometer state (Hosoya, 2015).

The quantum-Bayes relation for a projection operator can be expressed as

A^∣a⟩=a∣a⟩,\hat A\lvert a\rangle=a\lvert a\rangle,5

Taking real parts gives a formal analogue of Bayes’ rule,

A^∣a⟩=a∣a⟩,\hat A\lvert a\rangle=a\lvert a\rangle,6

with generalized probabilities that may be negative.

In collision theory, a weak Stern–Gerlach interaction includes both the kinetic energy and three-dimensional motion. For spin post-selection, the relevant quantity is the weak vector

A^∣a⟩=a∣a⟩,\hat A\lvert a\rangle=a\lvert a\rangle,7

The outgoing translational state is shifted according to

A^∣a⟩=a∣a⟩,\hat A\lvert a\rangle=a\lvert a\rangle,8

where, for a spatially confined apparatus, A^∣a⟩=a∣a⟩,\hat A\lvert a\rangle=a\lvert a\rangle,9. Higher orders generate weak tensors

⟨a∣A^∣ψi⟩⟨a∣ψi⟩=a.\frac{\langle a|\hat A|\psi_i\rangle}{\langle a|\psi_i\rangle}=a.0

The time-dependent Dyson and stationary scattering approaches recover the same first-order weak displacement (Castro et al., 2018).

Mass–energy equivalence provides another proposed coupling. For an atom with internal excited-state projector

⟨a∣A^∣ψi⟩⟨a∣ψi⟩=a.\frac{\langle a|\hat A|\psi_i\rangle}{\langle a|\psi_i\rangle}=a.1

the internal energy contribution to the rest mass modifies the center-of-mass kinetic energy:

⟨a∣A^∣ψi⟩⟨a∣ψi⟩=a.\frac{\langle a|\hat A|\psi_i\rangle}{\langle a|\psi_i\rangle}=a.2

Post-selection of the internal state produces a weak value that alters the external momentum distribution. Its imaginary part changes the relative populations of momentum components and therefore the momentum uncertainty (Zhang, 2016).

Weak values also support local-state reconstruction. For a position-postselected momentum weak value,

⟨a∣A^∣ψi⟩⟨a∣ψi⟩=a.\frac{\langle a|\hat A|\psi_i\rangle}{\langle a|\psi_i\rangle}=a.3

the measured position distribution supplies ⟨a∣A^∣ψi⟩⟨a∣ψi⟩=a.\frac{\langle a|\hat A|\psi_i\rangle}{\langle a|\psi_i\rangle}=a.4 and the real weak momentum supplies ⟨a∣A^∣ψi⟩⟨a∣ψi⟩=a.\frac{\langle a|\hat A|\psi_i\rangle}{\langle a|\psi_i\rangle}=a.5. Together they determine the wavefunction up to an additive global phase. In many-body systems, analogous weak values can be conditioned on configurations or particle-agnostic position data (Oianguren-Asua et al., 2023).

5. Arbitrary-strength measurements and metrology

The formal weak value does not require weak coupling. In an exact two-qubit protocol, a system Pauli observable ⟨a∣A^∣ψi⟩⟨a∣ψi⟩=a.\frac{\langle a|\hat A|\psi_i\rangle}{\langle a|\psi_i\rangle}=a.6 interacts with a meter qubit through

⟨a∣A^∣ψi⟩⟨a∣ψi⟩=a.\frac{\langle a|\hat A|\psi_i\rangle}{\langle a|\psi_i\rangle}=a.7

After pre- and post-selection, the meter state contains the exact weak value

⟨a∣A^∣ψi⟩⟨a∣ψi⟩=a.\frac{\langle a|\hat A|\psi_i\rangle}{\langle a|\psi_i\rangle}=a.8

Differences between meter-basis probabilities reconstruct

⟨a∣A^∣ψi⟩⟨a∣ψi⟩=a.\frac{\langle a|\hat A|\psi_i\rangle}{\langle a|\psi_i\rangle}=a.9

The formulas are valid at arbitrary interaction strength except at singular or uninformative values such as ∣bi⟩\lvert b_i\rangle0. An IBM quantum-computing implementation found that reconstruction quality varies nonmonotonically with interaction strength: very weak interactions produce small difference signals, while some stronger interactions provide better precision and accuracy; however, other strengths again become ill-conditioned (Paredes et al., 2023).

In ideal metrology, post-selection does not create additional Fisher information. For a Gaussian meter, the conditional Fisher information can scale as ∣bi⟩\lvert b_i\rangle1, but the total information per initial trial is multiplied by the success probability:

∣bi⟩\lvert b_i\rangle2

In the model considered,

∣bi⟩\lvert b_i\rangle3

where ∣bi⟩\lvert b_i\rangle4 is the largest-magnitude eigenvalue of ∣bi⟩\lvert b_i\rangle5. Thus weak-value amplification does not exceed the optimal standard strategy under ideal detection (Knee et al., 2018).

Its practical role can instead arise from detector imperfections. Weak-value protocols can reduce post-selected intensity while retaining a large conditional displacement, thereby avoiding detector saturation or limited dynamic range. They are less effective against detector jitter, whose Fisher-information efficiency is

∣bi⟩\lvert b_i\rangle6

independent of the displacement magnitude. They generally provide little systematic protection against pixelation. Weak values may nevertheless redirect information into a more stable phase-space quadrature.

Some proposals emphasize effects canceled in the total ensemble. For a qubit interaction with ∣bi⟩\lvert b_i\rangle7, the first-order changes in two orthogonal post-selected sectors can have opposite signs:

∣bi⟩\lvert b_i\rangle8

∣bi⟩\lvert b_i\rangle9

so that

B^\hat B0

Post-selection then separates contributions that cancel in the chosen unconditioned observable. This is a narrower claim than universal metrological superiority and does not establish a general Fisher-information advantage (Zhang, 2015).

6. Interpretations, controversies, and limitations

Weak values admit several non-equivalent interpretations. Operationally, they are conditional quantities that govern pointer responses, post-selection-probability slopes, phase shifts, or reduced-state dynamics. Formally, they are ratios of transition amplitudes. In some approaches they are treated as effective dynamical variables or as properties of a pure pre- and post-selected system; in others they are regarded as ensemble-level response coefficients or relational partial descriptions.

One influential position holds that a pure pre- and post-selected system acts on an additional, initially unentangled system as if the observable were replaced by B^\hat B1 during an infinitesimal interaction. The distinction from an ordinary expectation value is expressed through the pointer state: a pre- and post-selected weak value can reproduce the pointer response of an equal numerical eigenvalue to first order, whereas a pre-selected superposition with the same expectation value generally produces a mixture of eigenvalue responses. This eigenvalue-like interpretation has been defended for pure two-state vectors and generalized two-state vectors, but not for genuinely mixed pre- and post-selections (Vaidman et al., 2016).

A contrasting analysis argues that weak values are not ordinary conditional probabilities. If the relevant projectors do not commute, their products do not represent simultaneous propositions, and values outside the interval B^\hat B2 are not negative probabilities. On this account, B^\hat B3 can instead be interpreted as the first-order back-action of the weak interaction on the probability of the later post-selection (Mochizuki, 2013). A related property-ascription analysis treats weak values as contextual, post-selection-dependent partial descriptions rather than elements of reality in the eigenstate–eigenvalue sense (Matzkin, 2018).

The debate also concerns whether weak values are genuinely quantum. Epistemically restricted classical mechanics can reproduce the real and imaginary meter displacements of Gaussian weak-measurement protocols. In that theory, B^\hat B4 is a conditional phase-space expectation, while B^\hat B5 measures how post-selection biases the initial pointer momentum distribution for a finite disturbance. Such models reproduce Gaussian quantum mechanics but do not reproduce anomalous weak values associated with contextuality (Karanjai et al., 2015). Other analyses argue that reproducing an isolated anomalous conditional average classically is insufficient: a complete classical model would also need to reproduce arbitrarily small quantum disturbance, complex weak values, all postselections, quasiprobability negativity, and the full interference structure (Romito et al., 2015, Dressel, 2014).

The two-state-vector interpretation assigns a forward-evolving pre-selected state and a backward-evolving post-selected state to the intermediate interval. In that framework, weak values can be described as properties of an individual pre- and post-selected system, although their measurement requires an ensemble. This position is defended as an interpretation rather than as a consequence accepted by all treatments (Vaidman, 2017). A more recent operational formulation avoids a probe system altogether by defining a weak value as the first-order sensitivity of the post-selection amplitude to a small transformation:

B^\hat B6

Its real part is obtained from the normalized post-selection-probability slope,

B^\hat B7

while its imaginary part is obtained from an interferometric phase response (Ogawa et al., 2019).

A further critique distinguishes operational measurability from intrinsic reality. The weak value depends on a chosen pre-selection, post-selection, and ensemble-filtering procedure; it is not generally an eigenvalue, an expectation value of a single self-adjoint operator, or a quantity defined for every pair of states. This critique identifies three alleged reasoning errors: the ensemble fallacy, the post-selection fallacy, and the measurementist fallacy. It accepts weak values as useful for signal amplification, tomography, and detector readout while rejecting the inference that a reproducible pointer shift establishes an exotic property possessed by each individual system (Barandes, 10 Feb 2026).

Several limitations recur across these discussions. Weak values require nonzero pre-/post-selection overlap, and near-orthogonality increases both amplification and statistical cost. Weak measurements require controlled coupling and pointer calibration; the disturbance is reduced but not eliminated. Functions of weak values are not generally weak values of corresponding operator functions. Arbitrary-strength protocols can reconstruct the same formal quantity but become singular or poorly conditioned at particular couplings. Finally, operational agreement among standard quantum mechanics, Bohmian mechanics, stochastic mechanics, and epistemically restricted classical models does not by itself select a unique ontology.

Weak values are therefore best characterized as post-selection-dependent complex response quantities with multiple experimentally equivalent representations. Their established content is operational: they predict conditional detector statistics, probability responses, phase shifts, local fields, and reduced-state dynamics. Whether they are ensemble quantities, effective eigenvalue-like properties, contextual quasi-probabilities, or partial descriptions of individual systems remains an interpretational question rather than a settled consequence of the formalism.

Topic to Video (Beta)

No one has generated a video about this topic yet.

Whiteboard

No one has generated a whiteboard explanation for this topic yet.

Follow Topic

Get notified by email when new papers are published related to Weak Values.