Quantum Probability-Flow in Dynamical Systems

• Quantum probability-flow is a framework unifying multiple probability dynamics, encompassing current, phase-space, and geometric formulations in quantum mechanics.
• It provides a basis for analyzing phenomena like quantum backflow and Wigner shear while also inspiring learning algorithms in quantum simulation and neural models.
• Studies reveal its significance in connecting static quantum state descriptions with time-evolving probability dynamics across physical, computational, and interpretive contexts.

Searching arXiv for the provided topic and cited papers to ground the article in current arXiv records. “Quantum Probability-Flow Principle” denotes, in the cited arXiv literature, a family of related but nonidentical constructions in which quantum probabilities are treated as being transported, generated, constrained, or optimized under a specified dynamics. The exact phrase is used explicitly for leakage-based learning in associative memory (Ohzeki, 1 Jun 2026), while close conceptual analogues appear in current-based studies of quantum backflow (Bracken et al., 2016), phase-space analyses of regional probability transport (Goussev, 2020), coherent-state reinterpretations of flux (Mason et al., 2012), geometric reconstructions of quantum mechanics as flow on a statistical manifold (Caticha, 2021), and Husimi-fluid variational formulations (Zhdanov et al., 2021). This suggests that the term functions less as a single formal doctrine than as a unifying label for several probability-dynamical viewpoints.

1. Conceptual range and core definitions

One important usage is explicitly current-based. In that setting, the local probability flow is represented by the standard current

$$\vec j(\vec r)=\frac{\hbar}{2mi}\left(\Psi^*(\vec r)\nabla\Psi(\vec r)-\Psi(\vec r)\nabla\Psi^*(\vec r)\right),$$

or equivalently by a flux operator whose expectation value yields the textbook current (Mason et al., 2012). Negative or anomalous flow is then formulated as a sign reversal of this current, or of the integrated probability transferred through a boundary (Bracken et al., 2016).

A second usage is not built around current at all. “Principles and Dynamics of Quantum Mechanics” states that “the probabilities of physical outcomes are obtained from the intermediate states and processes of the interacting particles, considered as happening concurrently,” so probability flow is understood there as a dynamical principle of probability generation through superposition, not as a continuity equation or transport field (0904.0960). In that formulation, detection at $(r,t)$ occurs with probability $|\psi(r,t)|^2$, momentum-energy measurements occur with probability $|a(p,E)|^2$, and Schrödinger’s equation is presented as the local condition the wavefunction must satisfy at each point so that the total energy relation is fulfilled.

A third usage is geometric. “Quantum Mechanics as Hamilton-Killing Flows on a Statistical Manifold” starts from the simplex

$S=\{p \mid p_i>0,\ \sum_{i=1}^n p_i=1\},$

constructs its cotangent bundle, and seeks flows that preserve both the symplectic and metric structures. The resulting Hamilton-Killing flows lead to amplitudes $\psi_i=\sqrt{p_i}e^{iT_i}$, a Hermitian generator, the Schrödinger equation in matrix form, and the Born rule $p_i=|\psi_i|^2$ (Caticha, 2021).

These formulations share the idea that probabilities are dynamically structured, but they differ on what is primary: local current, phase-space transport, superposition of intermediate processes, or geometry of probability space.

2. Current-based formulations: backflow and classically forbidden transfer

The most developed current-based literature concerns quantum backflow. For a free non-relativistic particle on the $x$-axis, a state with momentum-space support restricted to $p_0<p<\infty$ can still exhibit probability flow opposite to the momentum direction (Bracken et al., 2016). The integrated transfer is represented by a Hermitian probability-flow operator $\hat Q$,

$(r,t)$0

and the maximal backflow is the largest eigenvalue of the corresponding integral operator. In the standard $(r,t)$1 case, the largest positive eigenvalue is

$(r,t)$2

about $(r,t)$3 of the total probability on the line (Bracken et al., 2016).

That analysis also shows that the maximal backflow decreases monotonically as $(r,t)$4 increases through positive values but never reaches $(r,t)$5, and increases monotonically as $(r,t)$6 decreases through negative values but never reaches $(r,t)$7. The paper interprets both trends as non-classical and uses an effective $(r,t)$8 scaling to recover the discontinuous classical limit (Bracken et al., 2016).

A related development establishes that quantum backflow and classically forbidden probability flow in diffraction in time are mathematically equivalent. In one problem, a state built only from nonnegative momenta yields $(r,t)$9; in the other, a wave packet initially confined to $|\psi(r,t)|^2$0 evolves freely and can yield $|\psi(r,t)|^2$1 at a point $|\psi(r,t)|^2$2. After change of variables, both reduce to the same integral eigenvalue problem with the same kernel structure, so the maximal transferred probability is governed by the same constant $|\psi(r,t)|^2$3 in the appropriate limit (Goussev, 2019).

The correlated-state generalization is stronger still. “Probability backflow for correlated quantum states” treats standard quantum backflow and quantum reentry as special cases of states constrained by positivity of $|\psi(r,t)|^2$4, or more generally $|\psi(r,t)|^2$5, and shows that the Bracken–Melloy constant is not the maximal wrong-way transfer for all such correlation classes. For nonlinear position-momentum correlations, the generalized supremum $|\psi(r,t)|^2$6 can exceed $|\psi(r,t)|^2$7, and the paper reports examples exceeding $|\psi(r,t)|^2$8 transferred probability (Goussev, 2020).

A common misconception is that anomalous negative current requires left-moving momentum components. The backflow literature directly contradicts that: strictly positive momentum support is compatible with temporary leftward probability transport (Bracken et al., 2016).

3. Phase-space transport, classical counterparts, and the limits of “quantum” anomaly

The most direct challenge to a purely quantum reading of reversed probability flow is the Wigner-phase-space analysis of a free Gaussian wave packet. For the packet

$|\psi(r,t)|^2$9

with

$|a(p,E)|^2$0

the probability of being to the right of a fixed point $|a(p,E)|^2$1,

$|a(p,E)|^2$2

can temporarily decrease even when the packet center moves left to right (Goussev, 2020).

In Wigner representation,

$|a(p,E)|^2$3

and for a free particle,

$|a(p,E)|^2$4

For the Gaussian packet, $|a(p,E)|^2$5 is everywhere positive, so it may be interpreted as an ordinary classical phase-space density. The central claim is therefore that the “negative flow of probability” is not intrinsically quantum in this case: the same effect occurs for a classical ensemble with the same Gaussian phase-space distribution because free evolution shears the distribution in phase space (Goussev, 2020).

In dimensionless variables, the Wigner function becomes

$|a(p,E)|^2$6

a Gaussian ellipse whose tilt is controlled by $|a(p,E)|^2$7. Over a small interval $|a(p,E)|^2$8, the evolution decomposes into a rigid shift and a shear. The shift contribution to $|a(p,E)|^2$9 is always positive, but the shear contribution can be negative if the ellipse is sufficiently tilted; the net decrease occurs when the negative shear outweighs the positive shift. The resulting condition,

$S=\{p \mid p_i>0,\ \sum_{i=1}^n p_i=1\},$0

is equivalent to Villanueva’s condition

$S=\{p \mid p_i>0,\ \sum_{i=1}^n p_i=1\},$1

(Goussev, 2020).

A further limitation of naive flow language appears in Husimi-space hydrodynamics. In the Husimi representation, a closed quantum system can be treated as a flow of multidimensional probability fluid in phase space, with continuity equation

$S=\{p \mid p_i>0,\ \sum_{i=1}^n p_i=1\},$2

and an action principle for elementary fluid parcels (Zhdanov et al., 2021). However, the paper emphasizes Skodje flux gauge fixing: the Husimi action is not unique, and different gauge choices can dramatically alter flux trajectories while leaving the density evolution unchanged. This is an objective caution against treating any single current visualization as uniquely physical (Zhdanov et al., 2021).

Taken together, these results show that the sign or geometry of a probability current is not, by itself, a reliable diagnostic of uniquely quantum behavior.

4. Coherent-state and Husimi-resolved reformulations

A different strand of the literature attempts to refine rather than discard current-based intuition. “Quantum Flux and Reverse Engineering of Quantum Wavefunctions” starts from the conventional flux operator

$S=\{p \mid p_i>0,\ \sum_{i=1}^n p_i=1\},$3

then replaces the point-localized Dirac basis by Gaussian states centered at $S=\{p \mid p_i>0,\ \sum_{i=1}^n p_i=1\},$4,

$S=\{p \mid p_i>0,\ \sum_{i=1}^n p_i=1\},$5

to define a smeared flux operator (Mason et al., 2012).

The corresponding flux eigenstates are Gaussian-plus-derivative combinations,

$S=\{p \mid p_i>0,\ \sum_{i=1}^n p_i=1\},$6

with eigenvalues

$S=\{p \mid p_i>0,\ \sum_{i=1}^n p_i=1\},$7

This makes explicit that the current is generated locally by interference between the amplitude and its first derivative (Mason et al., 2012).

The same paper identifies these structures with the first-order expansion of a coherent state

$S=\{p \mid p_i>0,\ \sum_{i=1}^n p_i=1\},$8

For small $S=\{p \mid p_i>0,\ \sum_{i=1}^n p_i=1\},$9, coherent-state projections become directional probes of local flow, and the current may be reconstructed from asymmetries of opposite-momentum Husimi weights. The extended flux viewpoint is summarized by

$\psi_i=\sqrt{p_i}e^{iT_i}$0

so the current is the first moment of a coherent-state momentum distribution (Mason et al., 2012).

This leads to the “processed Husimi” representation, which computes many Husimi projections at each spatial point and infers dominant ray directions. The method is designed for situations where the standard current is uniformly zero or misleading. In a circular billiard without magnetic field, the usual flux vanishes because of time-reversal symmetry, whereas processed Husimi flow reveals the classical bouncing-ray structure. In a magnetic circular billiard, it recovers the correct cyclotron radius and separates distinct underlying trajectories that an ordinary current map averages together (Mason et al., 2012).

The significance of this reformulation is not that it abolishes the standard current, but that it separates net current from component currents and thereby reintroduces directional content that cancellation had obscured.

5. Geometric, variational, and probabilistic reconstructions

Several papers elevate probability flow from a descriptive tool to a foundational principle. In the statistical-manifold reconstruction, probabilities $\psi_i=\sqrt{p_i}e^{iT_i}$1 are the primitive coordinates, the cotangent bundle $\psi_i=\sqrt{p_i}e^{iT_i}$2 supplies conjugate variables $\psi_i=\sqrt{p_i}e^{iT_i}$3, and dynamics are required to preserve both the symplectic form and the information-geometric metric. Once normalization and gauge equivalence $\psi_i=\sqrt{p_i}e^{iT_i}$4 are imposed, the amplitudes

$\psi_i=\sqrt{p_i}e^{iT_i}$5

emerge naturally, the allowed generators become bilinear and Hermitian, and the evolution equation

$\psi_i=\sqrt{p_i}e^{iT_i}$6

takes the Schrödinger form (Caticha, 2021). In that reconstruction, quantum mechanics is a constrained flow of probabilities on a statistical manifold, and the Born rule follows from $\psi_i=\sqrt{p_i}e^{iT_i}$7.

A distinct but related reconstruction is world-line based. “Quantum Action Principle in Relativistic Mechanics (II)” replaces the ordinary wavefunction by a wave functional $\psi_i=\sqrt{p_i}e^{iT_i}$8 on world lines, interprets $\psi_i=\sqrt{p_i}e^{iT_i}$9 as the probability density for a world line, and states that probability “flows” in Minkowski space with respect to an inner parameter $p_i=|\psi_i|^2$0 (Gorobey et al., 2010). In this formulation, the upper limit $p_i=|\psi_i|^2$1 of the parameter interval is interpreted as the particle lifetime, and it is fixed not by integrating over all $p_i=|\psi_i|^2$2 but by a stationarity condition on the quantum action,

$p_i=|\psi_i|^2$3

Other papers reconstruct the probability rule itself. “A Suggestive Way of Deriving the Quantum Probability Rule” assumes equal a priori probabilities for final states $p_i=|\psi_i|^2$4, where the post-measurement state after outcome $p_i=|\psi_i|^2$5 is written as

$p_i=|\psi_i|^2$6

and the allowed $p_i=|\psi_i|^2$7 region for each outcome is a disk of radius proportional to $p_i=|\psi_i|^2$8. Integrating a uniform density over that disk yields

$p_i=|\psi_i|^2$9

(Sutherland, 2020). “Quantum Probability as an Application of Data Compression Principles” instead uses algorithmic probability and interprets amplitudes as coefficients in an optimal lossy data compression; the transition weight is

$x$0

while basis-independence and the Gleason-Busch theorem force the conserved measure to reduce to norm-squared weight $x$1 (Randall, 2016).

These reconstructions are not identical. One is geometric, one variational and relativistic, one state-counting in an enlarged final-state space, and one algorithmic. Their common feature is that probability flow is treated as structurally prior to the usual Hilbert-space formalism or to the Born rule as an isolated axiom.

6. Learning principles and quantum simulation of probability flows

The exact phrase “quantum probability-flow principle” appears explicitly in a learning-theoretic setting in “Attention-Like Hebbian Learning from Quantum Probability Flow and Quantum-Annealer Tests” (Ohzeki, 1 Jun 2026). There, the principle is presented as a quantum extension of minimum probability flow: one chooses dynamics that expose local leakage channels from a data state $x$2, then learns by minimizing the probability that the system escapes into nearby error states under short evolution and measurement. With driver Hamiltonian

$x$3

the local energy gap for a spin flip is

$x$4

and the local loss is

$x$5

In the imaginary-time, dephased regime, the leakage weights are

$x$6

so the loss becomes a log-sum-exp over local energy gaps. The minimizing distribution over channels is a softmax,

$x$7

and the gradient yields a softmax-weighted Hebbian update. At high temperature, the rule reduces to ordinary Hebbian learning; at low temperature, it concentrates on the smallest gap. In the real-time regime, the weighting becomes algebraic rather than exponential, with robust objectives proportional to $x$8, and updates behave as a power law in the small-$x$9 limit (Ohzeki, 1 Jun 2026).

The same paper also reports D-Wave standard- and fast-anneal tests of a one-hot attention forward map. Conditioned on valid one-hot samples, an ideal thermal sampler gives

$p_0<p<\infty$0

and the measured distributions were fitted by both effective softmax and power-law/Lorentzian surrogates. The reported result is that the softmax fit had smaller KL divergence than the power-law fit; for example, under standard anneal,

$p_0<p<\infty$1

(Ohzeki, 1 Jun 2026).

A separate computational development maps classical flow models directly into quantum dynamics. If a density $p_0<p<\infty$2 evolves by the continuity equation

$p_0<p<\infty$3

then the qsample amplitude

$p_0<p<\infty$4

obeys a Schrödinger equation with the continuity Hamiltonian

$p_0<p<\infty$5

For conservative fields $p_0<p<\infty$6, this becomes

$p_0<p<\infty$7

(Layden et al., 9 Oct 2025). The paper proves efficient digital simulation with explicit spatial and temporal error bounds and uses this to prepare approximate qsamples of distributions learned by continuous flow models, including flow matching and probability-flow ODE formulations of diffusion models.

In both cases, probability flow becomes algorithmic: in one, as a local leakage objective for learning; in the other, as a continuity equation translated into a quantum state-preparation procedure.

7. Interpretive extensions and contested usages

The vocabulary of quantum probability flow also appears in more interpretive or domain-transferred settings. “Laser agitates probability flow in atoms to form alternating current and its peak-dip phenomenon” uses a trajectory-based, de Broglie–Bohm framework in which a laser drives an atomic probability current

$p_0<p<\infty$8

that oscillates with angular frequency

$p_0<p<\infty$9

(Cui, 2016). The paper interprets the oscillatory current as an atomic alternating current connecting electromagnetic driving to state evolution and introduces additional assumptions such as coherent self-overlap of matter waves, finite overlap number $\hat Q$0, and a Bohmian-style radial “quantum force.” It also attributes selection-rule-like behavior to orbital phase conditions and links finite-$\hat Q$1 peak-dip structures to orbit splitting and spectral consequences (Cui, 2016).

An even broader transfer appears in “The probability flow in the Stock market and Spontaneous symmetry breaking in Quantum Finance,” which rewrites Black–Scholes and Merton–Garman equations in Hamiltonian form, interprets the martingale condition as a vacuum state satisfying

$\hat Q$2

and relates non-Hermiticity to information or probability leakage across market boundaries (Arraut et al., 2022). In that framework, Hermiticity corresponds to preserved information/probability and a unique vacuum, while non-Hermiticity corresponds to open-market behavior and vacuum degeneracy. The paper treats “probability flow” as the market analogue of information transport rather than as a standard quantum-mechanical current (Arraut et al., 2022).

These extensions are significant mainly because they show how elastic the phrase has become. In some works it refers to rigorously defined current or phase-space transport; in others it denotes a broader dynamical, variational, or even metaphorical principle. A recurrent source of confusion is to assume that all such uses are equivalent. They are not. Current-based backflow, Wigner shear, Husimi-fluid action principles, statistical-manifold reconstructions, leakage-based learning rules, Bohmian transition pictures, and financial Hamiltonian analogies share a vocabulary of flow, but they do not share a single formalism.

The most stable cross-cutting conclusion is narrower. Across these literatures, “quantum probability-flow” marks the attempt to understand quantum behavior through the dynamics of probabilities or amplitudes rather than through static state descriptions alone. What varies is the carrier of that dynamics: current density, Wigner or Husimi phase-space density, probability simplex coordinates, world-line functionals, leakage channels in Ising memories, or coherent encodings of classical flow models.