Classical Measurement Models

• Classical measurement models are formal frameworks that use analytical white-box equations, state-space representations, and stochastic processes to describe measurement dynamics.
• They underpin applications in metrology, quantum system identification, and statistical inference by providing clear, interpretable models for uncertainty and error propagation.
• These models bridge classical and quantum paradigms by enabling deterministic analysis alongside Bayesian conditioning, ensuring traceability and robust parameter identification.

Classical measurement models denote a family of formalisms in which the measurement layer is represented through analytic relations, commuting observables, classical stochastic dynamics, or classical conditional probabilities rather than through an irreducibly quantum measurement device. Across metrology, control and identification, quantum foundations, and statistical inference, the term is used in several domain-specific senses: physics-based white-box measurement equations, augmented classical state-space descriptions of measured dynamics, operationally classical devices without superposition, Hilbert-space theories of classical apparatuses and observables, and probabilistic models for measurement error or Bayesian conditioning (Schneider et al., 2024, Tan et al., 2019, Cobucci et al., 19 Feb 2026, Dellaporta et al., 2023, Nahum et al., 2 Apr 2025).

1. Scope and principal meanings

Across the cited literatures, the term does not have a single universal definition. In metrology it usually denotes an analytical, physics-based relation between a measurand and input quantities. In quantum measurement theory it can denote a convex mixture of commuting projective devices. In system identification it often denotes a standard classical state-space model, possibly augmented by a dynamical model of measurement noise. In statistics it denotes models in which observed covariates are corrupted by classical additive noise. In classical statistical mechanics it can denote a posterior, or conditioned ensemble, obtained by Bayesian updating after measurement (Schneider et al., 2024, Tan et al., 2019, Cobucci et al., 19 Feb 2026, Dellaporta et al., 2023, Nahum et al., 2 Apr 2025).

Domain	Classical measurement model	Representative structure
Metrology	White-box measurement equation	$Y = f(X_1,\dots,X_n)$
Quantum-system identification	Augmented classical LTI model	$$\dot{\check{x}}=\check{A}\check{x},\ \tilde{y}=\check{C}\check{x}$$
Operational quantum measurement theory	Mixture of commuting devices	$$M_{a|x}=\int d\lambda\,q(\lambda)\sum_k p(a|x,k,\lambda)E_{k|\lambda}$$
Statistical inference	Classical measurement-error model	$W = X + N$
Classical lattice models	Conditioned ensemble	$P(S|M)\propto e^{-\mathcal{H}[S]}P(M|S)$

A common source of confusion is the word classical itself. In some papers it means physics-based and interpretable; in others it means commutative; in others it means describable by classical probability and stochastic processes; and in still others it means lacking superposition properties. The resulting literature is best read as a set of related but nonidentical traditions.

2. Analytical measurement equations in metrology

In metrology, a measurement model is the mathematical relationship linking the measurand $Y$ to observed or otherwise characterized input quantities $X_1,\dots,X_n$, typically written $Y=f(X_1,\dots,X_n)$. The same literature distinguishes a forward model, built from the sensor chain and physical laws, from the inverse model used for evaluating the measurement; the inverse model is also called the measurement equation, evaluation model, or measurement model (Schneider et al., 2024). In this usage, classical measurement models are explicitly identified with analytical or white-box models: variables correspond to physical quantities such as pressure, temperature, flow, irradiance, or angle, the structure follows known physical laws or system theory, and parameters have direct physical meaning.

Their construction is described as a systematic process comprising delimitation, abstraction, reduction, decomposition and aggregation, parameter identification, and verification. Delimitation fixes system boundaries and the validity range. Abstraction replaces a specific device by a parameterized class of systems. Reduction neglects negligible influences and represents unknown or unobservable effects as random variables. Decomposition and aggregation model subsystems such as sensors, amplifiers, and filters and then reconnect them through block diagrams, port-Hamiltonian descriptions, or finite elements. Parameters may come from design, standards, calibration, system identification, or experiment. Verification compares model predictions with measurement data and refines the structure when deviations are significant (Schneider et al., 2024).

The same framework underpins uncertainty evaluation. The inverse model is the starting point for the methods of the Guide to the Expression of Uncertainty in Measurement. Each input quantity is assigned a probability distribution, and the model propagates those uncertainties to the measurand. The paper explicitly highlights Taylor-series-based propagation, yielding the standard linearized relation

$$u^2(y) = \sum_i \left(\frac{\partial f}{\partial x_i}\right)^2 u^2(x_i) + 2\sum_{i<j}\frac{\partial f}{\partial x_i}\frac{\partial f}{\partial x_j}u(x_i,x_j).$$

From this one obtains the combined standard uncertainty $u_c(y)$ and the expanded uncertainty $$\dot{\check{x}}=\check{A}\check{x},\ \tilde{y}=\check{C}\check{x}$$0. Classical measurement models are therefore not merely descriptive; they define the dependency structure, sensitivity coefficients, calibration chain, and traceability path needed for metrological decision-making (Schneider et al., 2024).

This metrological usage also supplies the main contrast class. White-box models are interpretable, domain-knowledge intensive, and often valid across broader operating domains, whereas black-box models infer $$\dot{\check{x}}=\check{A}\check{x},\ \tilde{y}=\check{C}\check{x}$$1 from data and may generalize poorly outside the training distribution. The paper’s conclusion is explicit that white- and grey-box models are well established and are not being replaced; rather, they remain the backbone for traceability, uncertainty, and validation even when hybrid and data-driven methods are introduced (Schneider et al., 2024).

3. Classical state-space and stochastic trajectory models

A second major usage treats classical measurement models as ordinary control-theoretic or stochastic-process models wrapped around measured dynamics. In quantum Hamiltonian identification with classical colored measurement noise, the quantum system is first rewritten in coherence-vector form so that the expectation values of chosen observables satisfy a reduced linear time-invariant state-space model

$$\dot{\check{x}}=\check{A}\check{x},\ \tilde{y}=\check{C}\check{x}$$2

Measurement noise is then modeled as the output of a stable classical LTI filter driven by white noise,

$$\dot{\check{x}}=\check{A}\check{x},\ \tilde{y}=\check{C}\check{x}$$3

and, because the identification uses ensemble-average traces, its expectation evolves deterministically. Augmenting the quantum and noise states produces a purely classical continuous-time LTI model

$$\dot{\check{x}}=\check{A}\check{x},\ \tilde{y}=\check{C}\check{x}$$4

which is then discretized, realized by the Eigenstate Realization Algorithm, and matched through transfer functions to recover both Hamiltonian and noise parameters (Tan et al., 2019). Here, “classical measurement model” means that the measurement apparatus and its colored noise are described by standard classical system-identification machinery even though the underlying plant is quantum.

A related but more explicitly stochastic usage appears in continuously monitored many-body systems. For a Bose-Einstein condensate in an optical cavity under photon detection, the exact master equation is mapped to a Wigner phase-space representation, truncated to a Fokker–Planck equation, and unraveled into stochastic differential equations for a complex cavity amplitude $$\dot{\check{x}}=\check{A}\check{x},\ \tilde{y}=\check{C}\check{x}$$5 and a c-number atomic field $$\dot{\check{x}}=\check{A}\check{x},\ \tilde{y}=\check{C}\check{x}$$6. The measurement record for a single run is represented by the trajectory-dependent photon counting rate

$$\dot{\check{x}}=\check{A}\check{x},\ \tilde{y}=\check{C}\check{x}$$7

while the backaction of measurement appears as classical noise terms that induce phase diffusion in the atomic field (Lee et al., 2014). In this setting, classical stochastic measurement trajectories are not merely approximations to ensemble averages; each trajectory is intended to represent an individual experimental run conditioned on a particular detection record.

These two traditions share a structural idea. The measured system may be quantum, but the measurement channel is represented through state augmentation, spectral factorization, phase-space Fokker–Planck dynamics, or stochastic differential equations familiar from classical control and signal processing. That shift is especially useful when measurement noise is colored, when measurements are continuous, or when the relevant observables are well approximated by classical dynamics under monitoring (Tan et al., 2019, Lee et al., 2014).

4. Classicality inside quantum measurement theory

Within quantum theory proper, one influential line of work defines classical observables by the suppression of quantum fluctuations on a designated family of states. For a density matrix $$\dot{\check{x}}=\check{A}\check{x},\ \tilde{y}=\check{C}\check{x}$$8, the classicality of an operator $$\dot{\check{x}}=\check{A}\check{x},\ \tilde{y}=\check{C}\check{x}$$9 is

$$M_{a|x}=\int d\lambda\,q(\lambda)\sum_k p(a|x,k,\lambda)E_{k|\lambda}$$0

with $$M_{a|x}=\int d\lambda\,q(\lambda)\sum_k p(a|x,k,\lambda)E_{k|\lambda}$$1. For a time-evolved pure state $$M_{a|x}=\int d\lambda\,q(\lambda)\sum_k p(a|x,k,\lambda)E_{k|\lambda}$$2, the time-averaged classicality is

$$M_{a|x}=\int d\lambda\,q(\lambda)\sum_k p(a|x,k,\lambda)E_{k|\lambda}$$3

When $$M_{a|x}=\int d\lambda\,q(\lambda)\sum_k p(a|x,k,\lambda)E_{k|\lambda}$$4 is close to $$M_{a|x}=\int d\lambda\,q(\lambda)\sum_k p(a|x,k,\lambda)E_{k|\lambda}$$5, the observable has large signal-to-noise ratio,

$$M_{a|x}=\int d\lambda\,q(\lambda)\sum_k p(a|x,k,\lambda)E_{k|\lambda}$$6

and behaves as a classical pointer variable on the relevant time window. In this approach, classical observables are constructed by solving a generalized eigenvalue problem determined by the Hamiltonian spectrum and the initial state, and Schrödinger-cat branches are identified with orthogonal subspaces that continue to support such classical observables when the full Hilbert space no longer does (Wouters, 2014).

A different route models the emergence of a classical apparatus through a continuous interpolation between quantum and classical wave dynamics. The unified equation

$$M_{a|x}=\int d\lambda\,q(\lambda)\sum_k p(a|x,k,\lambda)E_{k|\lambda}$$7

reduces to the Schrödinger equation at $$M_{a|x}=\int d\lambda\,q(\lambda)\sum_k p(a|x,k,\lambda)E_{k|\lambda}$$8 and to a classical wave equation at $$M_{a|x}=\int d\lambda\,q(\lambda)\sum_k p(a|x,k,\lambda)E_{k|\lambda}$$9. The associated commutator becomes

$W = X + N$0

When the measuring apparatus is treated in the $W = X + N$1 limit, its pointer states are classical, objective, and noninterfering, so the paper argues that definite outcomes emerge without collapse or thermodynamic pragmatism (Ghose, 2017).

More recently, the phrase “classical measurement model” has been given a precise operational meaning for sets of POVMs. A set $W = X + N$2 admits a classical model if there exist a classical random variable $W = X + N$3, commuting rank-1 projectors $W = X + N$4, and classical post-processings $W = X + N$5 such that

$W = X + N$6

Equivalently, one may write $W = X + N$7 where, for fixed $W = X + N$8, all projective measurements $W = X + N$9 commute. This notion is stronger than mere commutativity for a single realized POVM but weaker than joint measurability in general: classical models imply joint measurability, yet the converse fails, with the trine POVM serving as an explicit counterexample. For the set of all projective measurements in dimension $P(S|M)\propto e^{-\mathcal{H}[S]}P(M|S)$0, the exact depolarization threshold is

$P(S|M)\propto e^{-\mathcal{H}[S]}P(M|S)$1

where $P(S|M)\propto e^{-\mathcal{H}[S]}P(M|S)$2 is the $P(S|M)\propto e^{-\mathcal{H}[S]}P(M|S)$3-th harmonic number; below this threshold, all projective measurements admit a classical model. The same work also shows that classical models imply sequential implementation without disturbance when classical side-information is available (Cobucci et al., 19 Feb 2026).

5. Hilbert-space, field-theoretic, and operator-algebraic formalisms

Another tradition seeks classical measurement models that preserve the operator and Hilbert-space machinery usually associated with quantum theory. In Koopman–von Neumann classical mechanics, the Hilbert space is $P(S|M)\propto e^{-\mathcal{H}[S]}P(M|S)$4, classical observables $P(S|M)\propto e^{-\mathcal{H}[S]}P(M|S)$5 act as multiplication operators, and dynamics is generated by the Liouvillian

$P(S|M)\propto e^{-\mathcal{H}[S]}P(M|S)$6

through a Schrödinger-like equation. A classical analog of the von Neumann measurement coupling is obtained from the Hamiltonian $P(S|M)\propto e^{-\mathcal{H}[S]}P(M|S)$7, giving a Liouvillian $P(S|M)\propto e^{-\mathcal{H}[S]}P(M|S)$8. This correlates system position with apparatus pointer position and system momentum with apparatus momentum, producing a measurement state and a recording state inside a fully classical Hilbert-space formalism. Because $P(S|M)\propto e^{-\mathcal{H}[S]}P(M|S)$9, the corresponding error–disturbance inequalities for physical observables collapse to trivial bounds, and state update is interpreted as Bayesian conditioning rather than collapse (Katagiri, 2019).

A more algebraic extension adds noncommutativity to classical measurement theory through the Poisson bracket. Starting from multiplication operators $Y$0 for classical observables $Y$1 and generators $Y$2 of canonical transformations, one has

$Y$3

The extended classical measurement algebra is then the group algebra generated by these operations. In this framework, classical measurement incompatibility can be modeled without quantization, and one may distinguish thermal noise, represented by Gibbs states, from Poincaré-invariant quantum noise, characterized by $Y$4 rather than $Y$5. The paper’s claim is that once noncommutativity and Poincaré-invariant noise are admitted, classical and quantum measurement theories can be discussed within a single algebraic structure (Morgan, 2022).

Field-based models make a similar move at the level of classical random fields. One construction shows that the Hilbert spaces of the free quantized electromagnetic field and of the global-$Y$6-invariant sector of the Dirac field can be reproduced by classical random fields acting on a vacuum state in a Lorentz-invariant way. The algebra of beables remains commutative, but the algebra of classical observables associated with coordinate transformations is noncommutative, making it meaningful to speak of eigenstates, projectors, and even entangled versus mixed states within a classical random-field setting (Morgan, 2017). A different proposal, prequantum classical statistical field theory, models quantum systems as classical random fields whose covariance operator $Y$7 defines a density operator $Y$8; threshold detectors then generate discrete clicks, recover Born-rule probabilities $Y$9, and suppress coincidences so that $X_1,\dots,X_n$0 can be obtained for sufficiently large thresholds (Khrennikov, 2012).

Operator-algebraic measurement models push the classical outcome structure into the apparatus algebra itself. In one such model, the apparatus is a unital separable non–type I nuclear simple $X_1,\dots,X_n$1-algebra equipped with a pure state and an asymptotically inner unital endomorphism. After the unitary system–apparatus interaction, the endomorphism magnifies microscopic information to the classical level by producing a non-factorial state whose central decomposition yields a superposition of phases with weights. The ensuing phase selection plays the role of classical outcome realization (Kishimoto, 2013).

6. Statistical and Bayesian classical measurement models

In statistics, classical measurement models usually refer to additive noise corrupting observed covariates. The classical measurement-error model is

$X_1,\dots,X_n$2

where $X_1,\dots,X_n$3 is the latent true covariate, $X_1,\dots,X_n$4 the observed one, and $X_1,\dots,X_n$5 the measurement error. This is distinct from the Berkson model $X_1,\dots,X_n$6. For the classical case, naive regression on $X_1,\dots,X_n$7 induces bias, including attenuation in the scalar linear Gaussian setting. A recent Bayesian nonparametric approach places a Dirichlet-process prior on the conditional law $X_1,\dots,X_n$8, centered on a model-derived prior $X_1,\dots,X_n$9, and then defines a loss-based posterior through either total least squares or maximum mean discrepancy. The resulting posterior bootstrap samples latent covariate distributions $Y=f(X_1,\dots,X_n)$0 and computes

$Y=f(X_1,\dots,X_n)$1

The associated MMD generalization bounds explicitly separate sampling error, prior specification error, and the magnitude of measurement error, and the method is designed to avoid assuming known error variances, replicated measurements, or instruments (Dellaporta et al., 2023).

A broader Bayesian notion of classical measurement appears in classical lattice models. Starting from a Boltzmann prior

$Y=f(X_1,\dots,X_n)$2

one introduces measurement outcomes $Y=f(X_1,\dots,X_n)$3 through $Y=f(X_1,\dots,X_n)$4. The joint law is written

$Y=f(X_1,\dots,X_n)$5

and Bayes’ theorem gives the conditioned ensemble

$Y=f(X_1,\dots,X_n)$6

For Gaussian measurements of local observables, the measurement strength is $Y=f(X_1,\dots,X_n)$7, and averaging over outcomes leads to replica field theories in the $Y=f(X_1,\dots,X_n)$8 limit rather than the $Y=f(X_1,\dots,X_n)$9 limit of quenched disorder. These conditioned ensembles exhibit new phase transitions, new renormalization-group fixed points, and direct connections to Nishimori-line inference problems, partial quenches, and measurement of classical stochastic processes (Nahum et al., 2 Apr 2025).

These statistical usages shift attention from device physics to inference structure. A classical measurement model becomes a model of corrupted observations or a posterior over hidden configurations. The update rule is Bayes’ theorem; the central questions concern identifiability, generalization error, and the phase structure induced by conditioning.

7. Classical limits, computational analogues, and broader implications

The relation between classical and quantum measurement models is especially transparent in work on quantum backflow. There the flux operator

$$u^2(y) = \sum_i \left(\frac{\partial f}{\partial x_i}\right)^2 u^2(x_i) + 2\sum_{i<j}\frac{\partial f}{\partial x_i}\frac{\partial f}{\partial x_j}u(x_i,x_j).$$0

can have negative expectation values for positive-momentum states, so flux is not always an operational probability. The paper studies several measurement models—direct measurements on separate ensembles, sequential projective measurements, and absorbing complex-potential detectors—and then introduces smeared quasi-projectors with width $$u^2(y) = \sum_i \left(\frac{\partial f}{\partial x_i}\right)^2 u^2(x_i) + 2\sum_{i<j}\frac{\partial f}{\partial x_i}\frac{\partial f}{\partial x_j}u(x_i,x_j).$$1. The resulting classical-limit parameter

$$u^2(y) = \sum_i \left(\frac{\partial f}{\partial x_i}\right)^2 u^2(x_i) + 2\sum_{i<j}\frac{\partial f}{\partial x_i}\frac{\partial f}{\partial x_j}u(x_i,x_j).$$2

suppresses the most negative eigenvalue as $$u^2(y) = \sum_i \left(\frac{\partial f}{\partial x_i}\right)^2 u^2(x_i) + 2\sum_{i<j}\frac{\partial f}{\partial x_i}\frac{\partial f}{\partial x_j}u(x_i,x_j).$$3, restoring a classical flux picture when measurement resolution is coarse compared with quantum scales (Yearsley et al., 2012). Here, classical measurement models arise as coarse-grained limits of sharper quantum schemes.

A different operational analogue appears in measurement-based classical computation. MBCC is defined by access to a single sample from an $$u^2(y) = \sum_i \left(\frac{\partial f}{\partial x_i}\right)^2 u^2(x_i) + 2\sum_{i<j}\frac{\partial f}{\partial x_i}\frac{\partial f}{\partial x_j}u(x_i,x_j).$$4-bit probability distribution, followed by linear Boolean post-processing using XOR and NOT gates only. For fixed-basis measurement-based quantum computations, the joint outcome distribution can be dephased to a purely classical distribution without changing the output statistics, so every such non-adaptive MBQC instance has an MBCC analogue. Nevertheless, when the resource distributions are efficiently quantumly preparable, there exist uniform MBCC families whose exact classical simulation would collapse the polynomial hierarchy to the third level (Hoban et al., 2013). The resulting distributions violate no Bell inequality, but they retain computational non-classicality as an imprint of their quantum origin.

Taken together, these developments show that classical measurement models are not a single theory but a stratified vocabulary for several problems: uncertainty propagation and traceability, state-space representation of measurement channels, operational classicality of POVMs, Bayesian conditioning, classical limits of quantum observables, and classical analogues of quantum information-processing tasks. Current work extends this vocabulary in at least three directions: GUM-compatible uncertainty methods for data-driven and hybrid metrology, resource-theoretic analyses of superposition in measurement devices, and replica or hydrodynamic field theories for measured classical systems and inference problems (Schneider et al., 2024, Cobucci et al., 19 Feb 2026, Nahum et al., 2 Apr 2025).