Quantum Brushes: Cross-Domain Quantum Visualizations

• Quantum Brushes are cross-domain constructs that reveal quantum state geometry through digital painting tools, entoptic visual effects, interactive simulations, and effective field corrections.
• In digital painting frameworks, color attributes are encoded into quantum states and evolved via circuit dynamics, enabling real-time, measurement-driven rendering on noisy intermediate-scale devices.
• Variational quantum brushes extend this approach with steerable algorithms and chemical models to optimize quantum state blending, while effective theories use 'quantum hair' to imprint interior source details onto exterior fields.

Searching arXiv for the specified paper and closely related “Quantum Brush/Quantum Brushes” works to ground the article in current arXiv records. Quantum Brushes denotes several distinct but related constructions in recent arXiv literature. In computational arts, quantum brushes are digital paintbrushes whose “bristle physics” is a quantum circuit: color patches or stroke samples are encoded as quantum states, evolved with unitary dynamics, and decoded from expectation values back into pixels (Lu et al., 30 Dec 2025, Ferreira et al., 1 Sep 2025). In visual psychophysics, “Quantum Brushes” names the topologically expanded, multi-lobed entoptic percept obtained when Boehm’s brushes are driven by spin–orbit structured light (Pushin et al., 3 Nov 2025). The phrase is also used metaphorically for interactive quantum-state visualization in Virtual Lab by Quantum Flytrap and for quantum-corrected exterior fields that encode classically hidden information about sources, described as a “quantum brush” in effective electrodynamics and gravity (Migdał et al., 2022, Calmet et al., 2022).

1. Multiple usages and conceptual commonalities

Recent arXiv usage assigns “Quantum Brushes” to four domains: computational art software, entoptic perception of structured light, interactive visualization of quantum states, and effective-field-theoretic “quantum hair.” In the first case, the term refers to executable painting tools; in the second, to a retinal percept; in the third, to a visualization metaphor; in the fourth, to exterior fields and radiation that “brush over” space with profile-dependent information.

Domain	Object denoted by “Quantum Brushes”	Operative mechanism
Digital painting	Quantum or variational paintbrushes	State encoding, circuit evolution, measurement-based pixel update
Vision science	Multi-lobed Boehm’s-brush entoptic percept	Spin–orbit structured light and polarization-sensitive retinal scattering
Interactive simulation	Visual and mathematical tools in Virtual Lab	Real-time state, operator, and entanglement visualization
Effective theory	Quantum-corrected exterior fields/radiation	Nonlocal kernels, vacuum polarization, and loop corrections

A common structural feature is that each usage makes an otherwise abstract quantum object perceptible. In painting systems, superposition, interference, entanglement, controllable dynamics, and measurement outcomes become color, texture, and stroke history. In the entoptic setting, the retina reads out the topology of a spin–orbit field. In Virtual Lab, amplitudes, operators, and entanglement are rendered as manipulable visual objects. In the effective-action setting, the “brush” is the profile-dependent exterior imprint of a source.

This suggests a family resemblance rather than a single technical definition. The phrase consistently denotes a visible or directly inspectable manifestation of quantum-state geometry, quantum dynamics, spin–orbit topology, or quantum-corrected field structure.

2. Quantum Brush as a digital painting framework

“Quantum Brush” is an open-source digital painting tool in which each stroke is turned into a small, executable quantum algorithm (Ferreira et al., 1 Sep 2025). The tool includes four different brushes—Aquarela, Heisenbrush, Smudge, and Collage—and is explicitly designed to be compatible with current noisy intermediate-scale quantum devices, as demonstrated by execution on IQM’s Sirius device. The paper reports no error mitigation and treats measurement-induced randomness and hardware noise as part of the aesthetic output.

The core encoding uses a general HL→Bloch-sphere mapping. From HSL, only hue and luminosity are encoded, with saturation preserved to avoid mixed-state preparation on NISQ hardware:

$$|\psi(\theta,\phi)\rangle = R_z(\phi)R_y(\theta)|0\rangle,\qquad \phi=2\pi H,\quad \theta=\pi L.$$

For regional averages, luminosity is averaged arithmetically and hue uses a circular mean. Readout relies on single-qubit tomography:

$$\tan \phi = \frac{\langle Y\rangle}{\langle X\rangle},\qquad \tan \theta = \frac{\sqrt{\langle X\rangle^2+\langle Y\rangle^2}}{\langle Z\rangle}.$$

This establishes a direct color-to-state-to-color pipeline.

Brush	Physical principle	Visual role
Aquarela	Entangling blend with a shared ancilla	Path-dependent watercolor-like blending
Heisenbrush	Spin-1/2 Heisenberg-chain time evolution	Color modulation by $\langle M_z\rangle(t)$
Smudge	Amplitude damping or pumping with a non-reset ancilla	Cascaded smear with ancilla-mediated correlations
Collage	Universal asymmetric quantum cloner	Copy/degradation trade-off from no-cloning

Aquarela sequentially couples a “brush” ancilla to segment qubits, steering each segment toward the brush color while interleaving a counter-rotation on the ancilla. Heisenbrush maps stroke segments to timesteps of a first-order Trotterized Heisenberg evolution, with local terms

$$H = \tfrac12 \sum_n [-X_nX_{n+1}-Y_nY_{n+1}-Z_nZ_{n+1}+X_n+Z_n]$$

and color driven by

$$\langle M_z\rangle(t)=\frac1N\sum_i \langle Z_i\rangle_t.$$

Smudge applies an ancilla-mediated amplitude-damping or pumping interaction without resetting the ancilla, so only the first qubit experiences true damping or pumping while subsequent qubits inherit correlations. Collage attempts to copy a source patch through a universal asymmetric quantum cloner, making the no-cloning trade-off directly visible in the source and paste regions.

The framework’s production orientation is equally central. A Stroke Manager snapshots the canvas, queues strokes, executes them asynchronously, and pastes only the changed pixels upon completion. This decouples slow or variable QPU runtimes from the act of drawing and turns repeated sampling into an explicit part of the workflow. The resulting system is not a classical style-transfer pipeline: colors are encoded as quantum states, transformed by physical circuits, and read back from measured observables rather than from pseudo-random seeds or pretrained deep features.

3. Variational Quantum Brushes: Steerable and Chemical

“Variational Quantum Brushes” extends the original framework with two brushes based on variational quantum algorithms, Steerable and Chemical, and makes both available in an implementation fully compatible with the original Quantum Brush application (Lu et al., 30 Dec 2025). The architecture is organized as encoding, evolution, and decoding. A selected image patch is aggregated into a small feature vector so that the number of qubits remains low and computations remain interactive.

Two encodings are used. The brush-agnostic one-qubit HL encoding prepares a state from hue–luminosity angles $(\phi,\theta)$ by applying $R_Z(\phi)$ and $R_Y(\theta)$ to $|0\rangle$. Steerable additionally introduces an RGBA SVD encoding: given an $m\times d$ color matrix $$\tan \phi = \frac{\langle Y\rangle}{\langle X\rangle},\qquad \tan \theta = \frac{\sqrt{\langle X\rangle^2+\langle Y\rangle^2}}{\langle Z\rangle}.$$0 of a patch with $$\tan \phi = \frac{\langle Y\rangle}{\langle X\rangle},\qquad \tan \theta = \frac{\sqrt{\langle X\rangle^2+\langle Y\rangle^2}}{\langle Z\rangle}.$$1 channels, compute

$$\tan \phi = \frac{\langle Y\rangle}{\langle X\rangle},\qquad \tan \theta = \frac{\sqrt{\langle X\rangle^2+\langle Y\rangle^2}}{\langle Z\rangle}.$$2

then form feature vectors from the diagonal entries of $$\tan \phi = \frac{\langle Y\rangle}{\langle X\rangle},\qquad \tan \theta = \frac{\sqrt{\langle X\rangle^2+\langle Y\rangle^2}}{\langle Z\rangle}.$$3 after logarithmic scaling and from $$\tan \phi = \frac{\langle Y\rangle}{\langle X\rangle},\qquad \tan \theta = \frac{\sqrt{\langle X\rangle^2+\langle Y\rangle^2}}{\langle Z\rangle}.$$4 for $$\tan \phi = \frac{\langle Y\rangle}{\langle X\rangle},\qquad \tan \theta = \frac{\sqrt{\langle X\rangle^2+\langle Y\rangle^2}}{\langle Z\rangle}.$$5. For $$\tan \phi = \frac{\langle Y\rangle}{\langle X\rangle},\qquad \tan \theta = \frac{\sqrt{\langle X\rangle^2+\langle Y\rangle^2}}{\langle Z\rangle}.$$6 qubits this yields 4, 8, or 16 features, respectively. Decoding for one-qubit states uses

$$\tan \phi = \frac{\langle Y\rangle}{\langle X\rangle},\qquad \tan \theta = \frac{\sqrt{\langle X\rangle^2+\langle Y\rangle^2}}{\langle Z\rangle}.$$7

with recovered angles

$$\tan \phi = \frac{\langle Y\rangle}{\langle X\rangle},\qquad \tan \theta = \frac{\sqrt{\langle X\rangle^2+\langle Y\rangle^2}}{\langle Z\rangle}.$$8

Steerable uses quantum geometric control to merge artworks. Given source and target patches $$\tan \phi = \frac{\langle Y\rangle}{\langle X\rangle},\qquad \tan \theta = \frac{\sqrt{\langle X\rangle^2+\langle Y\rangle^2}}{\langle Z\rangle}.$$9 and $\langle M_z\rangle(t)$0, it learns time-dependent controls in a bilinear Hamiltonian

$\langle M_z\rangle(t)$1

with Heisenberg nearest-neighbor drift

$\langle M_z\rangle(t)$2

and single-qubit Pauli controls distributed cyclically among $\langle M_z\rangle(t)$3, $\langle M_z\rangle(t)$4, and $\langle M_z\rangle(t)$5. The state obeys the Schrödinger equation

$\langle M_z\rangle(t)$6

and optimization minimizes

$\langle M_z\rangle(t)$7

Because the chosen Pauli strings generate a Lie algebra isomorphic to $\langle M_z\rangle(t)$8, the system is exactly controllable in principle, although the paper stresses that practical controllability is limited by discretization, circuit depth, and runtime.

Time evolution is implemented by a second-order symmetric splitting. Controls are represented by a small neural network, trained via backpropagation through PennyLane using JAX. The user-facing parameter is $\langle M_z\rangle(t)$9, which selects an intermediate point on the learned path; $$H = \tfrac12 \sum_n [-X_nX_{n+1}-Y_nY_{n+1}-Z_nZ_{n+1}+X_n+Z_n]$$0 extrapolates beyond the target and can produce nonclassical blends. Runtime defaults are $$H = \tfrac12 \sum_n [-X_nX_{n+1}-Y_nY_{n+1}-Z_nZ_{n+1}+X_n+Z_n]$$1 qubits, $$H = \tfrac12 \sum_n [-X_nX_{n+1}-Y_nY_{n+1}-Z_nZ_{n+1}+X_n+Z_n]$$2 controls, and $$H = \tfrac12 \sum_n [-X_nX_{n+1}-Y_nY_{n+1}-Z_nZ_{n+1}+X_n+Z_n]$$3 timesteps, with total depth approximately $$H = \tfrac12 \sum_n [-X_nX_{n+1}-Y_nY_{n+1}-Z_nZ_{n+1}+X_n+Z_n]$$4. Reported examples include Renoir’s Bal du moulin de la Galette steered toward a red parrot, Warhol’s Marilyn Diptych, and Miró’s El Jardín. The paper notes a “gray drift” near $$H = \tfrac12 \sum_n [-X_nX_{n+1}-Y_nY_{n+1}-Z_nZ_{n+1}+X_n+Z_n]$$5, likely due to the Hamiltonian choice or clipping; increasing controls can mitigate it, but at the cost of faster mid-range blending.

Chemical adopts a VQE-inspired workflow. It takes a molecule’s electronic Hamiltonian

$$H = \tfrac12 \sum_n [-X_nX_{n+1}-Y_nY_{n+1}-Z_nZ_{n+1}+X_n+Z_n]$$6

maps it to Pauli strings with the Jordan–Wigner transform, and uses the chemically motivated disentangled unitary coupled cluster ansatz. The variational objective is

$$H = \tfrac12 \sum_n [-X_nX_{n+1}-Y_nY_{n+1}-Z_nZ_{n+1}+X_n+Z_n]$$7

with generic parameter-shift derivative

$$H = \tfrac12 \sum_n [-X_nX_{n+1}-Y_nY_{n+1}-Z_nZ_{n+1}+X_n+Z_n]$$8

In the brush, optimization is not performed during painting. Instead, converged circuit families for $$H = \tfrac12 \sum_n [-X_nX_{n+1}-Y_nY_{n+1}-Z_nZ_{n+1}+X_n+Z_n]$$9 are precomputed offline for 1000 bond distances $$\langle M_z\rangle(t)=\frac1N\sum_i \langle Z_i\rangle_t.$$0 and stored as JSON; at paint time, the user-selected bond distance is snapped to the nearest precomputed set.

During painting, neighboring pixels along a stroke are aggregated into HL angles and encoded by $$\langle M_z\rangle(t)=\frac1N\sum_i \langle Z_i\rangle_t.$$1 then $$\langle M_z\rangle(t)=\frac1N\sum_i \langle Z_i\rangle_t.$$2. Circuit blocks are distributed along the stroke, with a “repetitions” parameter repeating portions of the sequence to smooth transitions. After each circuit application, Pauli expectations are decoded back to HL and written to the canvas. Reported examples include Anita Malfatti’s Portrait of Mário de Andrade and Tân Têng-pho’s Taipei Bridge. Smaller bond distance, around $$\langle M_z\rangle(t)=\frac1N\sum_i \langle Z_i\rangle_t.$$3, produced richer variation; larger radius made the aggregation disks visible. The paper restricts current support to $$\langle M_z\rangle(t)=\frac1N\sum_i \langle Z_i\rangle_t.$$4 and identifies extension to larger molecules as nontrivial because of slow convergence and long circuit sequences.

4. Entoptic Quantum Brushes and spin–orbit topology

A distinct usage of “Quantum Brushes” appears in the study of Boehm’s brushes under spin–orbit structured illumination (Pushin et al., 3 Nov 2025). Boehm’s brushes are a faint, bowtie-like entoptic pattern in peripheral vision attributed to polarization-sensitive scattering within the inner retina, notably the inner plexiform and ganglion cell layers. They differ mechanistically from Haidinger’s brushes, which arise in central vision from dichroic absorption by macular pigment molecules in the fovea. The paper’s claim is that when the retina is illuminated with spin–orbit structured light, the classical two-lobed Boehm’s pattern expands into a multi-lobed entoptic structure whose number and location encode the topology of the beam.

The stimulus is a non-separable superposition of circular polarization and orbital angular momentum:

$$\langle M_z\rangle(t)=\frac1N\sum_i \langle Z_i\rangle_t.$$5

or, in Jones-vector form,

$$\langle M_z\rangle(t)=\frac1N\sum_i \langle Z_i\rangle_t.$$6

For equal-amplitude components, the Stokes parameters satisfy

$$\langle M_z\rangle(t)=\frac1N\sum_i \langle Z_i\rangle_t.$$7

so the local linear polarization orientation becomes

$$\langle M_z\rangle(t)=\frac1N\sum_i \langle Z_i\rangle_t.$$8

with spatial index $$\langle M_z\rangle(t)=\frac1N\sum_i \langle Z_i\rangle_t.$$9. A q-plate optimized for 532 nm generated $(\phi,\theta)$0, and a downstream half-wave plate flipped the spin and yielded $(\phi,\theta)$1 at the retina when needed.

The retinal response is modeled phenomenologically as orientation-sensitive scattering. One expression given in the paper is

$(\phi,\theta)$2

and an alternative Stokes-form response is

$(\phi,\theta)$3

A single-brush envelope reproducing the bowtie lobe shape is

$(\phi,\theta)$4

The percept is the incoherent sum of these local responses over the annulus. Empirically and in simulation, the number of bright lobes obeys

$(\phi,\theta)$5

For $(\phi,\theta)$6 the modulation appears outside the stimulus ring, whereas for $(\phi,\theta)$7 it shifts inward.

The psychophysical protocol recruited sixteen participants, of whom eleven completed all procedures and were included in analysis. Retinal eccentricities ranged from $(\phi,\theta)$8 to $(\phi,\theta)$9 using six annular apertures of width 0.5 mm, mapped to retinal angle with an imaging insert and fundus photographs. The wavelength was 530 nm with a 10 nm FWHM bandpass. The task was a 2AFC rotation-direction discrimination using interleaved 2-up/1-down staircases converging at 70.7% correct. Contrast thresholds decreased with eccentricity and were fit by

$R_Z(\phi)$0

with group-averaged fit

$R_Z(\phi)$1

Individual fits yielded mean decay constant $R_Z(\phi)$2 with 95% CI $R_Z(\phi)$3, and the 50% threshold eccentricity was

$R_Z(\phi)$4

The effect therefore became perceptually robust at approximately $R_Z(\phi)$5 eccentricity, supporting a peripheral scattering mechanism.

The paper is explicit that the beams are bright classical vector beams rather than single-photon states. “Quantum” refers to quantized OAM, topological phase singularities, and classical nonseparability between polarization and OAM, not to nonlocal quantum entanglement of photons. This resolves a likely misconception: the observed multi-lobed brushes are topological and polarization-structured, but the illumination remains classical.

5. Quantum brushes as interactive visualization

Virtual Lab by Quantum Flytrap proposes another extension of the term: it asks that the platform be thought of as a set of quantum brushes, an interactive palette of visual, mathematically faithful tools with which users “paint” quantum states on an optical table (Migdał et al., 2022). The system is a no-code online laboratory with real-time simulation of an optical table, supports up to three entangled photons, and is available through open-source TypeScript packages including Quantum Tensor and BraKetVue.

The central contribution is visual methodology. Pure states are displayed as interactive ket notation,

$R_Z(\phi)$6

with basis switching among H/V, D/A, and L/R, and with complex numbers shown in polar, Cartesian, $R_Z(\phi)$7-based, or color-circle formats. Operators are rendered as arrays of colored disks whose radius encodes magnitude and hue encodes phase. For a matrix element $R_Z(\phi)$8, the disk radius is proportional to $R_Z(\phi)$9 and the disk hue to $R_Y(\theta)$0.

Entanglement is visualized in two distinct ways. The first is coordinated blinking of amplitudes generated by conditional projections: for a two-particle pure state $R_Y(\theta)$1, a random one-particle state $R_Y(\theta)$2 is sampled uniformly on the Bloch sphere, then

$R_Y(\theta)$3

and the two single-particle states blink in synchrony. The second is an entanglement graph driven by Rényi-2 entropy,

$R_Y(\theta)$4

which continuously changes as optical elements are added or rotated.

The simulation itself uses pure state vectors, tensor products, local unitaries, and POVMs. Free propagation is represented by

$R_Y(\theta)$5

and local optics act through

$R_Y(\theta)$6

Measurements are formulated as POVMs with post-measurement state

$R_Y(\theta)$7

The platform demonstrates teleportation, E91 quantum cryptography, Bell inequality violation, and the Deutsch–Jozsa algorithm, while the visualization layer remains general enough to apply to any discrete finite-dimensional quantum system.

In this usage, the “brush” is neither a paintbrush nor an entoptic lobe. It is a manipulable visual operator: every drag, rotation, and parameter change recomputes the quantum state and repaints the corresponding ket, operator matrix, entanglement graph, or multiverse tree. The metaphor emphasizes a direct, real-time coupling between mathematical formalism and visual form.

6. Quantum brush as quantum hair in electrodynamics and gravity

In effective electrodynamics and gravity, the phrase appears in still another sense. “Quantum hair” denotes quantum corrections to the classical exterior field of a compact source that encode information about the source’s internal structure, and the paper explicitly states, “This is the quantum brush: exterior fields/radiation ‘brush over’ space with patterns that carry interior information” (Calmet et al., 2022). The classical expectation from Gauss’s law in electrodynamics, or from no-hair expectations in gravity, is that exterior fields depend only on global charges. The quantum claim is that loop corrections and nonlocal kernels violate this simplification.

In QED, integrating out the electron yields the Euler–Heisenberg effective Lagrangian

$R_Y(\theta)$8

together with derivative terms such as

$R_Y(\theta)$9

The modified Maxwell equation in momentum space is

$|0\rangle$0

with

$|0\rangle$1

For a spherically symmetric source with charge density $|0\rangle$2 and total charge $|0\rangle$3, classical Gauss’s law gives

$|0\rangle$4

whereas the quantum-corrected potential obeys

$|0\rangle$5

Because $|0\rangle$6 depends on the full density profile rather than only on $|0\rangle$7, the exterior field after Fourier transform depends on the interior profile. The paper highlights both the nonlocal Uehling correction and inverse-power corrections such as

$|0\rangle$8

The same logic applies to radiation. In a uniform electric field, Schwinger pair production has rate density

$|0\rangle$9

For a charged ball with effective field

$m\times d$0

the locally constant field approximation yields

$m\times d$1

Since $m\times d$2 is exponentially sensitive to $m\times d$3 and $m\times d$4 depends on the source profile through nonlocal kernels, outgoing pair flux and spectra encode classically hidden information about the source.

In gravity, the paper uses the unique effective action with local quadratic curvature terms and nonlocal logarithmic terms:

$m\times d$5

The resulting field equations involve retarded kernels of the form

$m\times d$6

so the exterior metric depends on integrals over the interior energy-momentum distribution and its past history. A typical leading exterior correction for a compact source can scale as $m\times d$7, and the paper argues that Hawking radiation, like Schwinger radiation in the QED analog, can in principle carry interior information that classical no-hair expectations would forbid.

The effective-theory usage is therefore the most abstract of the four. Nothing is painted, and no percept is involved. Yet the underlying image is consistent with the other literatures: a quantum-corrected structure leaves a visible or measurable trace at the exterior, as though the source had brushed its internal profile into the surrounding field.

This suggests that “Quantum Brushes” has become a cross-domain label for situations in which quantum structure is made directly legible—on a digital canvas, on the retina, in an interactive operator display, or in the asymptotic field and radiation of an effective theory.