Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2610.14177

Compass in the Mirror: Quantum Backpropagation with the Hopf Ansatz

Ruge Lin, Ranyiliu Chen

quant-phSubmitted Oct 9, 2026• Announced Oct 13, 2026• 0 citations
arXivHTMLPDF

Abstract

A variational quantum state may have many control angles, and its complete gradient contains one derivative for each. The task is to turn access to selected derivatives into a shared finite-shot record for the complete vector. The Hopf ansatz supplies the setting: it constructs any normalized pure state from nested two-way magnitude splits and, for complex states, leaf phases, with one directly programmed angle per split or phase. Earlier Hopf constructions prepare the state, recover coordinates from known amplitudes, and provide indexed derivative access. With a phase-calibrated controlled reflection, the inverse Hopf frame maps each normalized magnitude direction to a computational label, so every all-$X$ outcome contributes to every magnitude derivative. A separate signed leaf record supplies the complex phase block. A checkpoint variant reverses only the suffix below a selected depth and returns that layer's gradient block. For a complete chart with $M$ real-valued parameters, sufficient execution counts are $\mathcal{O}(\log\log M)$ for the global method and $\mathcal{O}(\log M\log\log M)$ for the complete checkpoint schedule at fixed additive coordinate accuracy and confidence. The global saving is record reuse, not a reduction of the $M$-entry output. Under the designated direct-angle schedules, the global method's assigned Hopf CNOT work and input/output-inclusive storage have at most logarithmic overhead relative to a matched scalar evaluation. The same Hopf geometry thus carries a state forward and the objective's complete coordinate response backward.

AI Summary

The paper asks whether measurements can be reused to estimate the complete gradient of a variational expectation objective, rather than measuring each derivative separately. It treats balanced-tree Hopf state coordinates: $M=2^n-1$ real parameters for real states and $M=2^{n+1}-1$ for complex states. Objective access requires an exact, phase-calibrated controlled Hermitian unitary $O$, with $O^2=I$. The construction places the prepared state and its orthonormal magnitude-tangent directions into computationally addressed columns of a unitary frame. After controlled-$O$ interference and the inverse frame, one all-$X$ outcome supplies parity samples for every magnitude derivative. A signed histogram and Walsh–Hadamard transform decode these samples. Complex phase derivatives use a separate ancilla-$Y$/system-$Z$ signed leaf histogram. A checkpoint variant reverses only the suffix below a selected tree depth and estimates that layer’s gradient block. Unbiasedness follows from interference identities and the Hopf differential geometry. Each depth record has Euclidean norm two; bounded-difference concentration and a union bound over depths give simultaneous additive coordinate error $\varepsilon$ with confidence $1-\delta$ using $O([1+\log(n/\delta)]/\varepsilon^2)$ global executions, versus $n$ times this order for complete checkpointing. At fixed accuracy and confidence, these are $O(\log\log M)$ and $O(\log M\log\log M)$. Under designated direct-angle schedules, global Hopf CNOT work is $O(M\log M\log\log M)$; decoding costs $O(M\log M)$ and storage is $O(M)$. Guarantees concern raw absolute-coordinate accuracy, not relative, directional, or metric-rescaled accuracy. The output remains $M$-dimensional, circuits are assumed exact, and reflection-sum objectives incur a squared coefficient-$\ell_1$-norm sampling factor.

Likely largely AI-generatedAI use disclosedStrong assumptionsMissing comparison
ChatGPT assisted in exploring constructions, developing arguments, checking implementations, and shaping the manuscript. GPT-6 was also used during the final revision phase.

Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.

Related papers

Quantum angle encoding with learnable rotation applied to quantum–classical convolutional neural networks

Emmanuel Ovalle-Magallanes, Dora Elisa Alvarado-Carrillo, Juan Gabriel Avina‐Cervantes +2 more

Applied Soft Computing•Apr 18, 2023•67 citations

A Compass on the Quantum State Sphere: The Hopf Ansatz for Arbitrary Pure-State Optimization

Ruge Lin, Guangxi Li

Optimizing an arbitrary pure state requires more than a circuit that can prepare it: one also needs coordinates, local distance information, and measurable derivative directions. We introduce the Hopf ansatz, a binary-tree circuit for arbitrary normalized real and complex state vectors. Internal angles split probability between subtrees, while leaf phases supply the complex degrees of freedom. The same tree recovers physical angles from known amplitudes, gives an analytic diagonal metric on the state-vector sphere, and prepares each nonzero coordinate tangent as a normalized state. For Hamiltonian objectives and other costs with the same local transition-moment form, each gradient component is a known metric factor times a transition moment between the current state and a tangent state. Signed branch states express these moments through expectation-value measurements. Indexed preparation groups magnitude tangents by tree depth and collects complex phase tangents into one leaf-indexed block. The number of compiled gradient-access circuit families is logarithmic in Hilbert-space dimension; measurement repetitions needed for a prescribed accuracy remain a separate statistical cost. In deterministic real-state benchmarks with exact objectives and gradients and known global optima, metric-aware Hopf updates reach numerical-precision median gaps. Relative to coordinate-Adam baselines, they yield smaller mean final gaps in variational eigensolver tests and concentrate more runs at numerical precision in metrology-inspired tests. The Hopf ansatz thus supplies a compass for arbitrary pure-state optimization, with coordinates, geometry, and measurable directions in one circuit structure.

quant-pharXiv:2607.14231•Jul 17, 2026•1 citation

Artian's Completed-Event Bell Reconstruction: Real Source Geometry, Access, and Durable Records

Attar Ali

From a finite event record to Bell correlations and their loss under recording A pair is prepared together, read at two separated instruments, and allowed to leave a material record. This paper builds the correlation law from a real source geometry in Artian's Quantum Traction Theory. Its source-to-access equation is \[\boxed{p_K(r,s\mid\mathbf a,\mathbf b)=\frac{C\!\left[(\Pi_r(\mathbf a)\otimes\Pi_s(\mathbf b)\otimes I)K\Omega\right]}{C(K\Omega)}}\] Here C is an additive, basis-invariant event capacity, Omega is the normalized orientation-neutral two-channel source, K is a specified record contact, and the projectors select the two instrument outputs. A real quarter-turn and a qualified shared-reference composition generate the readout algebra. Capacity symmetry fixes its quadratic weight. A separately stated capacity-to-frequency condition connects that weight to laboratory counts. The correlation follows before the Bell comparison For a phase-compensated record contact with overlap g between its two normalized record profiles, the construction gives \[E_K(\mathbf a,\mathbf b)=-a_zb_z-g(a_xb_x+a_yb_y),\qquad 0\leq g\leq1.\] The intact-source limit is E = -a dot b. The same calculation yields the fixed-setting and optimized Bell responses: \[\boxed{S_{45}=\sqrt{2}(1+g),\qquad S_{\mathrm{opt}}=2\sqrt{1+g^2}.}\] Fresh independent record contacts multiply their overlaps. The paper derives the repeated-contact law, a capacity-based overlap calibration, uncertainty bounds, and a protocol that separates calibration from the Bell observations being judged. A source-first account with an explicit dependency map A1 supplies event-time labels and the clock-angle correspondence; A4 supplies the real dial; A5-X supplies completed-event support, with its inherited A6/A7 conditions. The two-channel preparation, balanced composition, capacity symmetry, frequency bridge, and record coupling are printed separately. Each theorem states exactly which conditions it uses. The Access uncertainty commutator is treated as a distinct inherited algebraic premise, rather than inferred from Bell correlations. When a receiver follows the source Version 2.0 constructs a matched receiver rather than assigning its transformation by name. Preserving every probability of the stated bilinear contact forces its transport to be the conjugate of the source transport, up to an overall phase: \[\boxed{|r^{\mathrm T}V^{\mathrm T}Us|^2=|r^{\mathrm T}s|^2\ \text{for all }r,s\quad\Longrightarrow\quad V=e^{i\alpha}\overline U.}\] A finite set of basis and sum probes certifies this relation. A robust extension converts imperfect probe probabilities into an operator-error bound. Finite clock and shift operations then produce the comparison loop \[\boxed{\mathcal L=M+\omega(I-M),\qquad \omega=e^{2\pi i/d}.}\] The matched sector cancels the comparison phase while the unmatched sector retains it. This is a joint comparison law; each local readout can remain incompatible. The paper identifies the binary loop's ambiguity, provides additional probes that distinguish it, and transfers the Bell correlations into a complete detector record when the original carrier is consumed. One contact, two observable roles A specified marker contact supplies both the Bell record overlap and a resolved weak-measurement instrument: \[\begin{aligned}\gamma&=\sin\theta,\\K_0&=\operatorname{diag}(\cos(\theta/2),\sin(\theta/2)),\\K_1&=\operatorname{diag}(\sin(\theta/2),\cos(\theta/2)).\end{aligned}\] An independent replay executes this instrument on all 40,032 tree records in the public Feng and colleagues archive, reproducing every stored decoder success/failure score. The main five-setting batch contains 32,000 records; the complete archive has six settings. The reproduced job-cluster residual probabilities are p = 0.0188753 and p = 0.0442603, respectively. These diagnose modest tension with the ideal instrument under the stated retrospective covariance. The matched QTT and quantum representations make the same archival predictions, so this replay is not a significance test between them. The package preserves the complete source-first Bell construction and adds finite receiver certificates, physical transport-error bounds, a common-contact derivation, five vector figures, a twelve-item achievements ledger, and offline reproducibility materials. The next operational question is precise: which independently qualified physical contact fixes both the record overlap and the receiver comparison response? Dependencies: QTT Main Book; finite address-capacity counting; real source histories and composition; A1-CHSH spinor character; Observation as Access.

•Sep 25, 2026•2 citations

Toward Fully Unitary Quantum Backpropagation

Fu-Te Wong

•Jan 1, 2026•0 citations

Feedback and Robust Local Readout of a Fixed Four-State Quantum Source

Jeremy Rodgers

This paper develops local quantum-reader decision problems within Shadow Theory’s source–readout programme. It asks when feedback improves access to a target distinction in a fixed quantum source, and how reliably that distinction can be read under detector noise and calibration uncertainty. The source supplies independent copies conditional on one hidden label, held fixed throughout the experiment, from four equatorial qubit preparations. The task is to identify their common positive or negative sign, rather than reconstruct the full preparation label. Although all four preparations have identical separate-mode intensities, phase-sensitive local measurements provide access to the target. The analysis compares adaptive readers, whose measurement settings depend on earlier records, with randomized nonadaptive readers, whose complete schedules are selected before outcomes are observed. Each measurement consumes one fresh copy. Every protocol must return a sign at a fixed deadline, retaining all settings, outcomes, nulls and specified timing records. Neither class uses collective measurements or coherent quantum memory between specimens. The main results establish equal optimal worst-case error for both reader classes through three copies across the shared efficiency and visibility model. At four copies, explicit rational feedback policies outperform every randomized nonadaptive schedule over the full continuous equatorial measurement menu, in both the ideal model and a specified noisy model. The comparison is supported by continuous-menu lower certificates rather than measurement-grid searches alone. For a one-percent worst-case error target, the minimum copy count is narrowed to either 13 or 14 for each reader class in the nominal model, and either 16 or 17 under the specified asymmetric gain and phase uncertainty. These adjacent thresholds remain unresolved. Further results give an exact robust repeated-angle frontier, extend the comparison to randomized choices of a fixed repeated angle, and identify convex-source conditions under which one reader is globally sufficient. The accompanying technical supplement and reproducibility materials provide measurement-allocation benchmarks, expanded finite-menu comparisons, compact feedback policies, exact optimization certificates, and timing and calibration interfaces. The results quantify target access under explicitly stated preparation, measurement and record-access conditions. They use standard quantum probabilities within Shadow Theory’s research programme and distinguish copy savings from total apparatus, control or calibration savings.

•Oct 28, 2026•0 citations

Quantum gradient descent in graph state tomography

J.L.E. da Silva, Hitalo Joseferson Batista Nascimento

Journal of Computational and Applied Mathematics•May 29, 2026•0 citations
A variational quantum state may have many control angles, and its complete gradient contains one derivative for each. The task is to turn access to selected derivatives into a shared finite-shot record for the complete vector. The Hopf ansatz supplies the setting: it constructs any normalized pure state from nested two-way magnitude splits and, for complex states, leaf phases, with one directly programmed angle per split or phase. Earlier Hopf constructions prepare the state, recover coordinates from known amplitudes, and provide indexed derivative access. With a phase-calibrated controlled reflection, the inverse Hopf frame maps each normalized magnitude direction to a computational label, so every all-XXX outcome contributes to every magnitude derivative. A separate signed leaf record supplies the complex phase block. A checkpoint variant reverses only the suffix below a selected depth and returns that layer's gradient block. For a complete chart with MMM real-valued parameters, sufficient execution counts are O(log⁡log⁡M)\mathcal{O}(\log\log M)O(loglogM) for the global method and O(log⁡Mlog⁡log⁡M)\mathcal{O}(\log M\log\log M)O(logMloglogM) for the complete checkpoint schedule at fixed additive coordinate accuracy and confidence. The global saving is record reuse, not a reduction of the MMM-entry output. Under the designated direct-angle schedules, the global method's assigned Hopf CNOT work and input/output-inclusive storage have at most logarithmic overhead relative to a matched scalar evaluation. The same Hopf geometry thus carries a state forward and the objective's complete coordinate response backward.
The paper asks whether measurements can be reused to estimate the complete gradient of a variational expectation objective, rather than measuring each derivative separately. It treats balanced-tree Hopf state coordinates: M=2n−1M=2^n-1M=2n−1 real parameters for real states and M=2n+1−1M=2^{n+1}-1M=2n+1−1 for complex states. Objective access requires an exact, phase-calibrated controlled Hermitian unitary OOO, with O2=IO^2=IO2=I.
The construction places the prepared state and its orthonormal magnitude-tangent directions into computationally addressed columns of a unitary frame. After controlled-OOO interference and the inverse frame, one all-XXX outcome supplies parity samples for every magnitude derivative. A signed histogram and Walsh–Hadamard transform decode these samples. Complex phase derivatives use a separate ancilla-YYY/system-ZZZ signed leaf histogram. A checkpoint variant reverses only the suffix below a selected tree depth and estimates that layer’s gradient block.
Unbiasedness follows from interference identities and the Hopf differential geometry. Each depth record has Euclidean norm two; bounded-difference concentration and a union bound over depths give simultaneous additive coordinate error ε\varepsilonε with confidence 1−δ1-\delta1−δ using O([1+log⁡(n/δ)]/ε2)O([1+\log(n/\delta)]/\varepsilon^2)O([1+log(n/δ)]/ε2) global executions, versus nnn times this order for complete checkpointing. At fixed accuracy and confidence, these are O(log⁡log⁡M)O(\log\log M)O(loglogM) and O(log⁡Mlog⁡log⁡M)O(\log M\log\log M)O(logMloglogM).
Under designated direct-angle schedules, global Hopf CNOT work is O(Mlog⁡Mlog⁡log⁡M)O(M\log M\log\log M)O(MlogMloglogM); decoding costs O(Mlog⁡M)O(M\log M)O(MlogM) and storage is O(M)O(M)O(M). Guarantees concern raw absolute-coordinate accuracy, not relative, directional, or metric-rescaled accuracy. The output remains MMM-dimensional, circuits are assumed exact, and reflection-sum objectives incur a squared coefficient-ℓ1\ell_1ℓ1​-norm sampling factor.
Optimizing an arbitrary pure state requires more than a circuit that can prepare it: one also needs coordinates, local distance information, and measurable derivative directions. We introduce the Hopf ansatz, a binary-tree circuit for arbitrary normalized real and complex state vectors. Internal angles split probability between subtrees, while leaf phases supply the complex degrees of freedom. The same tree recovers physical angles from known amplitudes, gives an analytic diagonal metric on the state-vector sphere, and prepares each nonzero coordinate tangent as a normalized state. For Hamiltonian objectives and other costs with the same local transition-moment form, each gradient component is a known metric factor times a transition moment between the current state and a tangent state. Signed branch states express these moments through expectation-value measurements. Indexed preparation groups magnitude tangents by tree depth and collects complex phase tangents into one leaf-indexed block. The number of compiled gradient-access circuit families is logarithmic in Hilbert-space dimension; measurement repetitions needed for a prescribed accuracy remain a separate statistical cost. In deterministic real-state benchmarks with exact objectives and gradients and known global optima, metric-aware Hopf updates reach numerical-precision median gaps. Relative to coordinate-Adam baselines, they yield smaller mean final gaps in variational eigensolver tests and concentrate more runs at numerical precision in metrology-inspired tests. The Hopf ansatz thus supplies a compass for arbitrary pure-state optimization, with coordinates, geometry, and measurable directions in one circuit structure.
From a finite event record to Bell correlations and their loss under recording A pair is prepared together, read at two separated instruments, and allowed to leave a material record. This paper builds the correlation law from a real source geometry in Artian's Quantum Traction Theory. Its source-to-access equation is pK(r,s∣a,b)=C ⁣[(Πr(a)⊗Πs(b)⊗I)KΩ]C(KΩ)\boxed{p_K(r,s\mid\mathbf a,\mathbf b)=\frac{C\!\left[(\Pi_r(\mathbf a)\otimes\Pi_s(\mathbf b)\otimes I)K\Omega\right]}{C(K\Omega)}}pK​(r,s∣a,b)=C(KΩ)C[(Πr​(a)⊗Πs​(b)⊗I)KΩ]​​ Here C is an additive, basis-invariant event capacity, Omega is the normalized orientation-neutral two-channel source, K is a specified record contact, and the projectors select the two instrument outputs. A real quarter-turn and a qualified shared-reference composition generate the readout algebra. Capacity symmetry fixes its quadratic weight. A separately stated capacity-to-frequency condition connects that weight to laboratory counts. The correlation follows before the Bell comparison For a phase-compensated record contact with overlap g between its two normalized record profiles, the construction gives EK(a,b)=−azbz−g(axbx+ayby),0≤g≤1.E_K(\mathbf a,\mathbf b)=-a_zb_z-g(a_xb_x+a_yb_y),\qquad 0\leq g\leq1.EK​(a,b)=−az​bz​−g(ax​bx​+ay​by​),0≤g≤1. The intact-source limit is E = -a dot b. The same calculation yields the fixed-setting and optimized Bell responses: S45=2(1+g),Sopt=21+g2.\boxed{S_{45}=\sqrt{2}(1+g),\qquad S_{\mathrm{opt}}=2\sqrt{1+g^2}.}S45​=2​(1+g),Sopt​=21+g2​.​ Fresh independent record contacts multiply their overlaps. The paper derives the repeated-contact law, a capacity-based overlap calibration, uncertainty bounds, and a protocol that separates calibration from the Bell observations being judged. A source-first account with an explicit dependency map A1 supplies event-time labels and the clock-angle correspondence; A4 supplies the real dial; A5-X supplies completed-event support, with its inherited A6/A7 conditions. The two-channel preparation, balanced composition, capacity symmetry, frequency bridge, and record coupling are printed separately. Each theorem states exactly which conditions it uses. The Access uncertainty commutator is treated as a distinct inherited algebraic premise, rather than inferred from Bell correlations. When a receiver follows the source Version 2.0 constructs a matched receiver rather than assigning its transformation by name. Preserving every probability of the stated bilinear contact forces its transport to be the conjugate of the source transport, up to an overall phase: ∣rTVTUs∣2=∣rTs∣2 for all r,s⟹V=eiαU‾.\boxed{|r^{\mathrm T}V^{\mathrm T}Us|^2=|r^{\mathrm T}s|^2\ \text{for all }r,s\quad\Longrightarrow\quad V=e^{i\alpha}\overline U.}∣rTVTUs∣2=∣rTs∣2 for all r,s⟹V=eiαU.​ A finite set of basis and sum probes certifies this relation. A robust extension converts imperfect probe probabilities into an operator-error bound. Finite clock and shift operations then produce the comparison loop L=M+ω(I−M),ω=e2πi/d.\boxed{\mathcal L=M+\omega(I-M),\qquad \omega=e^{2\pi i/d}.}L=M+ω(I−M),ω=e2πi/d.​ The matched sector cancels the comparison phase while the unmatched sector retains it. This is a joint comparison law; each local readout can remain incompatible. The paper identifies the binary loop's ambiguity, provides additional probes that distinguish it, and transfers the Bell correlations into a complete detector record when the original carrier is consumed. One contact, two observable roles A specified marker contact supplies both the Bell record overlap and a resolved weak-measurement instrument: γ=sin⁡θ,K0=diag⁡(cos⁡(θ/2),sin⁡(θ/2)),K1=diag⁡(sin⁡(θ/2),cos⁡(θ/2)).\begin{aligned}\gamma&=\sin\theta,\\K_0&=\operatorname{diag}(\cos(\theta/2),\sin(\theta/2)),\\K_1&=\operatorname{diag}(\sin(\theta/2),\cos(\theta/2)).\end{aligned}γK0​K1​​=sinθ,=diag(cos(θ/2),sin(θ/2)),=diag(sin(θ/2),cos(θ/2)).​ An independent replay executes this instrument on all 40,032 tree records in the public Feng and colleagues archive, reproducing every stored decoder success/failure score. The main five-setting batch contains 32,000 records; the complete archive has six settings. The reproduced job-cluster residual probabilities are p = 0.0188753 and p = 0.0442603, respectively. These diagnose modest tension with the ideal instrument under the stated retrospective covariance. The matched QTT and quantum representations make the same archival predictions, so this replay is not a significance test between them. The package preserves the complete source-first Bell construction and adds finite receiver certificates, physical transport-error bounds, a common-contact derivation, five vector figures, a twelve-item achievements ledger, and offline reproducibility materials. The next operational question is precise: which independently qualified physical contact fixes both the record overlap and the receiver comparison response? Dependencies: QTT Main Book; finite address-capacity counting; real source histories and composition; A1-CHSH spinor character; Observation as Access.
This paper develops local quantum-reader decision problems within Shadow Theory’s source–readout programme. It asks when feedback improves access to a target distinction in a fixed quantum source, and how reliably that distinction can be read under detector noise and calibration uncertainty. The source supplies independent copies conditional on one hidden label, held fixed throughout the experiment, from four equatorial qubit preparations. The task is to identify their common positive or negative sign, rather than reconstruct the full preparation label. Although all four preparations have identical separate-mode intensities, phase-sensitive local measurements provide access to the target. The analysis compares adaptive readers, whose measurement settings depend on earlier records, with randomized nonadaptive readers, whose complete schedules are selected before outcomes are observed. Each measurement consumes one fresh copy. Every protocol must return a sign at a fixed deadline, retaining all settings, outcomes, nulls and specified timing records. Neither class uses collective measurements or coherent quantum memory between specimens. The main results establish equal optimal worst-case error for both reader classes through three copies across the shared efficiency and visibility model. At four copies, explicit rational feedback policies outperform every randomized nonadaptive schedule over the full continuous equatorial measurement menu, in both the ideal model and a specified noisy model. The comparison is supported by continuous-menu lower certificates rather than measurement-grid searches alone. For a one-percent worst-case error target, the minimum copy count is narrowed to either 13 or 14 for each reader class in the nominal model, and either 16 or 17 under the specified asymmetric gain and phase uncertainty. These adjacent thresholds remain unresolved. Further results give an exact robust repeated-angle frontier, extend the comparison to randomized choices of a fixed repeated angle, and identify convex-source conditions under which one reader is globally sufficient. The accompanying technical supplement and reproducibility materials provide measurement-allocation benchmarks, expanded finite-menu comparisons, compact feedback policies, exact optimization certificates, and timing and calibration interfaces. The results quantify target access under explicitly stated preparation, measurement and record-access conditions. They use standard quantum probabilities within Shadow Theory’s research programme and distinguish copy savings from total apparatus, control or calibration savings.