Computational framework for quantum state tomography of spin ensembles

AI summary

Overview

Research area: Quantum information science / nuclear magnetic resonance (NMR) quantum computing, specifically quantum state tomography (QST) and scientific software engineering. Published in Computer Physics Communications (arXiv:2609.11868v1, 10 Sep 2026).

Technical level: Intermediate. The paper assumes familiarity with density matrices, product operators, and NMR spectroscopy, but the software itself is designed to be usable by experimentalists rather than theorists.

Scope: The paper presents and validates Tomography-NMR, an open-source Python package that turns raw NMR spectrometer data into reconstructed two-qubit density matrices, documenting every intermediate analysis step that is typically hidden inside proprietary software.

What This Paper Is About

Quantum state tomography lets researchers reconstruct a quantum system's density matrix from measurements, but published papers usually report only the final matrices, omitting the practical steps (phase correction, peak integration, parameter choices) needed to get there. Those steps are typically buried in closed-source commercial software such as Bruker TopSpin, which is built for chemistry rather than precision quantum-state characterization. The authors build a fully transparent, reproducible Python pipeline that connects raw time-domain signals to validated density matrices, and they benchmark it on 20 experimentally prepared two-qubit states measured on a two-qubit NMR processor.

Key Contributions

1. An end-to-end open-source pipeline (Tomography-NMR, MIT license, Python 3.11+) that goes from raw Bruker-format time-domain data through Fourier transformation, phase correction, peak integration, coefficient extraction, and density matrix reconstruction, with visualization at every stage.

2. Three interchangeable peak-integration methods for different experimental situations: direct peak height, fixed-parameter Simpson integration over user-chosen windows, and a systematic grid-search optimization against a known reference state — the first two require no knowledge of the target state, the third reaches the highest fidelity.

3. A documented mapping from spectra to density matrix coefficients, encoded as a length-28 tomography lookup table derived from the product operator formalism, together with a check_basis_complete() utility that verifies all 15 non-identity coefficients are covered.

4. Experimental validation on 20 two-qubit states (computational basis states, Hadamard outputs, CNOT outputs, Bell states, and Hadamard + T outputs) prepared on two coupled ³¹P nuclei in ADP, plus a prototype three-qubit operator module (63 non-identity product operators) as a starting point for extension.

Main Findings

• Reconstruction quality depends strongly on integration strategy. For the |00⟩ example state, peak height gave fidelity 0.980, fixed-parameter Simpson integration also gave 0.980, and optimized integration gave 0.998 — with computation times of under 1 s, under 1 s, and roughly 2 min respectively.

• Reference-free methods reach roughly 98% fidelity. Direct peak height and fixed-parameter numerical integration achieve fidelities of approximately 98% on the benchmark set of known states without any theoretical reference.

• Reference-based optimization exceeds 99%. Systematic parameter optimization against a known target state achieves reconstruction fidelities exceeding 99% on the same benchmark states.

• Average fidelity across 20 experimental states is high. The abstract and introduction report average reconstruction fidelities ranging from 0.975 to 0.995; the program summary separately states that reconstruction fidelities on the 20 experimental states range from 0.948 to 0.999.

• A minimal measurement set of 9 spectra suffices. Although the full protocol yields 28 real-valued spectra (7 readout pulses × 2 readout channels × 2 real/imaginary parts) and up to 56 peak measurements for only 15 independent coefficients, 9 carefully chosen spectra cover all of them; the redundancy in the full set improves accuracy via averaging.

• Deviation matrices need physical cleanup. The raw reconstructed matrix is traceless with negative eigenvalues because NMR measures only the deviation density matrix; adding (1/4)𝟙 restores unit trace, and optional truncation of negative eigenvalues enforces positive semi-definiteness (described as the recommended mode).

• Per-category validation results are not available in the supplied text. The paper's validation section is truncated mid-sentence at "All reconstructions use the optimized integration method with positive semi-def…", so the breakdown of fidelity by state category is not reported in the provided content.

Methodology in Plain English

The physical setup. Two phosphorus-31 nuclei in adenosine 5'-diphosphate (ADP) dissolved in D₂O act as two qubits, coupled by a scalar J-coupling of approximately 22 Hz. Measurements were made at T = 310 K in a B₀ = 14.09 T field on a Bruker spectrometer.

The measurement protocol. Seven readout pulse sequences (no pulse, √X₁, √Y₁, √X₂, √Y₂, √X₁√X₂, √X₁√Y₂) rotate the state into different measurement bases before acquisition. Each sequence produces a free induction decay that is Fourier-transformed into a spectrum containing two doublets — one per qubit. Splitting each complex spectrum into real and imaginary parts gives 28 real spectra.

From peaks to numbers. In a coupled two-spin system each peak is a doublet with left (L) and right (R) components. The real and imaginary parts of L+R and L−R each correspond to one specific product-operator coefficient, as set by the preparatory pulse. A lookup table encodes this correspondence, so measured peak intensities become the 15 non-identity density-matrix coefficients.

Three ways to measure a peak. Peak height reads a single frequency point — fast but noise-sensitive. Fixed-parameter integration applies Simpson's rule over a user-chosen window (for example w = 4 Hz with frequency offsets of 1.5 Hz for Q1 and 0.5 Hz for Q2 in the worked example). Optimized integration performs a grid search over frequency offsets, window width w, and a coupling correction δJ swept over ±1 Hz around the measured splitting, evaluating on the order of 10³ parameter combinations in typically 2–5 minutes and caching results.

Getting a physical state. Redundant measurements of the same coefficient are averaged with standard deviations used as uncertainties. The matrix is assembled, the identity component is added for unit trace, and negative eigenvalues are optionally truncated. Quality is scored with either Fortunato's normalized projection or Jozsa's fidelity, with uncertainty propagated through the pipeline.

Why This Matters

Impact on research. The paper targets a reproducibility gap: published QST results usually omit phase-correction procedures, integration choices, and parameter optimization, and the tools that implement them are proprietary. Because "even subtle choices in spectral processing, such as the width of an integration window or the method of phase correction, can significantly affect the final result" — a serious issue for applications demanding fidelities above 99% — an open, auditable implementation lets independent groups verify results and compare methodologies. The package complements the authors' earlier PULSEE software: PULSEE solves the forward problem of simulating spin dynamics from known pulse sequences, while Tomography-NMR solves the inverse problem of reconstructing states from measured spectra.

Real-world applications:

• Benchmarking quantum gates and verifying entangled-state preparation on NMR and other qubit platforms.
• Characterizing decoherence and systematic error in quantum hardware.
• Training students and new practitioners who currently must reverse-engineer workflows from incomplete descriptions.
• Serving as a template for adapting tomography to other spectroscopic platforms, since only the isolated data-import module is Bruker-specific.

Industry relevance. Closed-source commercial NMR software is optimized for chemistry and biology rather than quantum-state characterization, and quantum computing vendors and metrology labs need transparent, scriptable validation tooling. An MIT-licensed, NumPy/SciPy/QuTiP-based package lowers the barrier to building certified characterization pipelines and makes cross-lab comparisons possible.

Future Directions

• Scaling beyond two qubits. The three-qubit operator module is a prototype not used in the current pipeline; extending to more spins will require replacing the exhaustive grid search, whose cost "grows quickly" with additional swept parameters.

• Better search strategies for optimized integration. The authors explicitly suggest that larger systems would likely need a more efficient search than the current systematic grid search.

• Portability to other platforms and protocols. Because only the preprocessing module is tied to the Bruker data layout, adapting to other spectrometers or measurement modalities is a natural next step, as is supporting alternative measurement protocols.

• Modeling systematic error more completely. The reported uncertainty captures statistical scatter among redundant measurements but not systematic effects such as slow phase drift, which shifts repeated measurements consistently; folding such effects into error bars remains open.

Target Audience

Experimental quantum information researchers using NMR (and, with modest adaptation, other spectroscopic platforms), graduate students and newcomers who need a documented worked example of the full tomography workflow, and software-minded spectroscopists who want a reproducible alternative to closed-source analysis packages. Theorists interested in the product-operator-to-density-matrix mapping will also find the formalism and lookup-table construction useful, though readers without any quantum information or NMR background will need supplementary material.

Authors’ abstract

We present Tomography-NMR, an open-source Python package that reconstructs quantum density matrices from spectroscopic measurement data. The package implements a complete analysis pipeline for two-qubit quantum state tomography based on the product operator formalism: raw time-domain signals are Fourier-transformed into frequency-domain spectra, spectral peak intensities are mapped to expansion coefficients of the density matrix, and the full quantum state is reconstructed. Three integration methods are provided for different use cases: direct peak height measurement and fixed-parameter numerical integration require no theoretical reference and are suited to unknown states, achieving fidelities of approximately 98\% on a benchmark set of known states, while a systematic parameter optimization against a known target state achieves reconstruction fidelities exceeding 99\% for the same benchmark states. While the detailed theoretical framework for quantum state tomography is well established, the practical procedures for extracting density matrices from experimental spectra are often inadequately documented in the literature and obscured within proprietary software. This package addresses that gap by providing a fully transparent, reproducible implementation of every analysis step, from spectral preprocessing to density matrix visualization. The software has been validated on experimentally prepared two-qubit states measured via nuclear magnetic resonance (NMR) spectroscopy of coupled $^{31}$P nuclei. Average reconstruction fidelities range from 0.975 to 0.995 across a benchmark set of 20 two-qubit states, including the computational basis states, Bell states, and the outputs of three fundamental quantum gates (CNOT, H, and T). Although developed for NMR, the modular architecture facilitates adaptation to other spectroscopic platforms and alternative measurement protocols.