Q-GAIN: A Python Package for Machine Learning and Physically Informed Analysis Applications

A. R. Fritsch; I. B. Spielman; J. P. Zwolak; L. Ritter; M. Doris; S. Guo; S. M. Koh; S. Mukherjee

arxiv: 2607.02413 · v1 · pith:GYCUGEHFnew · submitted 2026-07-02 · ❄️ cond-mat.quant-gas · cs.LG

Q-GAIN: A Python Package for Machine Learning and Physically Informed Analysis Applications

M. Doris , S. Guo , S. M. Koh , L. Ritter , A. R. Fritsch , S. Mukherjee , I. B. Spielman , J. P. Zwolak This is my paper

Pith reviewed 2026-07-03 02:34 UTC · model grok-4.3

classification ❄️ cond-mat.quant-gas cs.LG

keywords quantum gasesBose-Einstein condensatesmachine learningimage analysisobject detectionsoliton detectionvortex detection

0 comments

The pith

The Q-GAIN Python package combines machine learning feature detection with conventional physics analysis for images of Bose-Einstein condensates.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a Python package that supplies ready-to-use modules for applying machine learning to data from cold-atom experiments. It supplies classification, object detection, and physics-informed metrics aimed at images of atomic Bose-Einstein condensates. The design organizes work into a sequence of data loading and preprocessing, machine-learning feature identification, and standard analysis steps. Three concrete configurations illustrate the pattern: digit classification on a standard dataset, soliton detection in time-of-flight images, and vortex detection in ring-shaped condensates. A sympathetic reader would expect the package to let experimenters add modern image tools to existing analysis pipelines without starting from scratch each time.

Core claim

Q-GAIN implements classification, object detection, and physics-informed metrics for feature detection in images of atomic Bose-Einstein condensates, structured as a module-based workflow from data loading and preprocessing through ML-based feature identification to conventional analysis techniques.

What carries the argument

The modular workflow that sequences data loading and preprocessing, ML-based feature identification, and conventional analysis techniques.

If this is right

The same framework can classify handwritten digits from the MNIST dataset as a basic demonstration.
The framework can be reconfigured to detect and analyze solitonic excitations in time-of-flight data.
An object-detection module can locate quantized vortices in images of ring-shaped BECs.
Users obtain a consistent structure that begins with data handling, moves to ML identification, and ends with conventional analysis.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same modular separation could be applied to image-analysis tasks in other quantum-gas or ultracold-molecule experiments that already use time-of-flight imaging.
Adding hooks for direct import of raw camera files from common laboratory hardware would reduce the preprocessing step still further.
A natural next test would be whether the same workflow supports supervised training on simulated BEC images before deployment on real data.

Load-bearing premise

The modular workflow described actually reduces the effort required to combine ML feature identification with conventional analysis for real experimental BEC data.

What would settle it

A timed comparison in which an experienced user implements a new feature-detection task both with and without the package and measures the difference in total lines of code and wall-clock time.

Figures

Figures reproduced from arXiv: 2607.02413 by A. R. Fritsch, I. B. Spielman, J. P. Zwolak, L. Ritter, M. Doris, S. Guo, S. M. Koh, S. Mukherjee.

**Figure 2.** Figure 2: The directory and file structure for Q-GAIN. (a) Each experimental directory contains a data subdirectory with the data roster stored in data_info and data files stored in data_files, and a models subdirectory containing trained ML models and fitting files. All exported data is also stored in the experiment directories. (b) The basic structure of the HDF roster file system. Each entry points to its corre… view at source ↗

**Figure 3.** Figure 3: An overview of the performance of the modern SolDet sub-package. The [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗

**Figure 4.** Figure 4: Example synthetic time-of-flight images of ring BECs with quantized vor [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗

read the original abstract

Here we describe the quantum gas analysis and inference (Q-GAIN) Python package, which enables rapid deployment of machine learning (ML) and physics-informed analysis techniques for cold-atom experiments. Out of the box, Q-GAIN implements classification, object detection, and physics-informed metrics for feature detection in images of atomic Bose-Einstein condensates (BECs). Q-GAIN encourages a natural, module-based workflow: starting with data loading and preprocessing, followed by ML-based feature identification, and ending with conventional analysis techniques. We demonstrate this modularity by configuring Q-GAIN for three ML tasks. First, we demonstrate the basic workflow of the Q-GAIN framework by implementing the standard task of classifying handwritten digits from the MNIST dataset. Then, we re-implement our earlier soliton detection (SolDet) package in the Q-GAIN framework, enabling the detection and analysis of solitonic excitations in time-of-flight data. Finally, we develop an object-detection tool that identifies quantized vortices in images of ring-shaped BECs.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Q-GAIN wraps standard ML pipelines for BEC images into one framework but shows no measured reduction in effort or error rates on real data.

read the letter

The core of this paper is a Python package that chains data loading, ML-based feature finding, and conventional physics analysis for cold-atom images. The authors configure it for MNIST classification, a reimplementation of their prior SolDet soliton detector, and vortex detection in ring BECs. That modular structure is the main thing on offer.

What works is the consistent interface across those steps. Putting the old SolDet code inside the new framework at least demonstrates that prior tools can be brought in without starting from scratch. The workflow description is straightforward and matches how many experimental groups already move from raw images to extracted quantities.

The soft spot is the missing evidence on the central claim. The abstract and demonstrations establish that the modules exist and can be run, but there are no numbers on lines of code saved, wall-clock time to set up a new analysis, or accuracy on held-out experimental images compared with a direct implementation. The three examples stay close to standard benchmarks or internal reuse; none test whether the package actually shortens the loop for typical time-of-flight or in-situ BEC data. That leaves the “rapid deployment” benefit as an assertion rather than a measured result.

This is for experimental quantum-gas groups that already handle image data and want a single place to plug in ML models before doing their usual fitting or counting. A reader looking for new algorithms or first-principles derivations will not find them. The thinking is clear and the code appears to be the deliverable, so the paper is coherent on its own terms.

I would bring it to a reading group only if the group cares about analysis tools for cold atoms. I would not cite it in my own work. It is worth sending to peer review for a methods or software track so that users can comment on the interface before wider adoption.

Referee Report

1 major / 1 minor

Summary. The manuscript introduces the Q-GAIN Python package for applying machine learning and physics-informed analysis to cold-atom BEC images. It describes a modular workflow (data loading/preprocessing → ML feature identification → conventional analysis) and demonstrates the framework via three configurations: MNIST digit classification, a re-implementation of the SolDet soliton detector for time-of-flight data, and an object-detection tool for quantized vortices in ring BECs.

Significance. If the modular interfaces prove usable on real experimental data, the package could modestly aid reproducibility in the quantum gases community by providing ready-to-configure ML components alongside physics metrics. The work supplies no performance benchmarks, error rates on held-out BEC images, or effort metrics, so its practical significance remains unquantified.

major comments (1)

[Abstract, §3 (demonstrations)] Abstract and demonstration sections: the central claim that the module-based workflow 'enables rapid deployment' and reduces effort is unsupported by any quantitative evidence (development time, LOC counts, accuracy on experimental TOF/in-situ images, or A/B comparison against non-Q-GAIN pipelines). The three examples only confirm that the modules can be configured; they supply no metrics on actual BEC data.

minor comments (1)

[Abstract] The abstract states that Q-GAIN 'implements classification, object detection, and physics-informed metrics' but provides no description of the specific ML architectures, loss functions, or physics-informed terms used in the vortex or soliton modules.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed review. We agree that the manuscript's claims about rapid deployment lack supporting quantitative evidence and will revise the text accordingly.

read point-by-point responses

Referee: [Abstract, §3 (demonstrations)] Abstract and demonstration sections: the central claim that the module-based workflow 'enables rapid deployment' and reduces effort is unsupported by any quantitative evidence (development time, LOC counts, accuracy on experimental TOF/in-situ images, or A/B comparison against non-Q-GAIN pipelines). The three examples only confirm that the modules can be configured; they supply no metrics on actual BEC data.

Authors: We acknowledge that this assessment is correct. The three demonstrations illustrate workflow modularity (MNIST classification, SolDet re-implementation for TOF soliton detection, and vortex object detection in ring BECs) but provide no development-time metrics, line counts, accuracy figures on held-out experimental images, or direct comparisons to non-Q-GAIN pipelines. The soliton and vortex examples use real BEC data types, yet the sections do not report performance numbers. We will revise the abstract and §3 to remove or qualify the 'rapid deployment' and 'reduces effort' phrasing, framing the examples strictly as illustrations of configurability rather than evidence of efficiency gains. revision: yes

Circularity Check

0 steps flagged

No circularity: software package description with no derivations or predictions

full rationale

The paper describes the Q-GAIN Python package and its modular workflow for ML tasks on BEC images, demonstrated via MNIST classification, SolDet re-implementation, and vortex detection. It contains no equations, physical predictions, fitted parameters, or derivation chains. Claims about modularity and rapid deployment are software-interface assertions verified by code examples rather than self-referential logic. No patterns (self-definitional, fitted-input prediction, self-citation load-bearing, etc.) apply, as there are no load-bearing steps that reduce to the paper's own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper is a software description and introduces no physical model, free parameters, axioms, or new entities.

pith-pipeline@v0.9.1-grok · 5743 in / 1054 out tokens · 31399 ms · 2026-07-03T02:34:53.987134+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

18 extracted references · 14 canonical work pages

[1]

J. A. Armstrong and L. Fletcher, Fast Solar Image Classification Using Deep Learning and Its Importance for Automation in Solar Physics , Solar Physics 294(6), 80 (2019), doi:10.1007/s11207-019-1473-z

work page doi:10.1007/s11207-019-1473-z 2019
[2]

Carrasquilla and R

J. Carrasquilla and R. G. Melko, Machine learning phases of matter, Nature Physics 13(5), 431 (2017), doi:10.1038/nphys4035, Publisher: Nature Publishing Group

work page doi:10.1038/nphys4035 2017
[3]

Stanev, C

V. Stanev, C. Oses, A. G. Kusne, E. Rodriguez, J. Paglione, S. Curtarolo and I. Takeuchi, Machine learning modeling of superconducting critical temperature, npj Computational Materials 4(1), 29 (2018), doi:10.1038/s41524-018-0085-8, Publisher: Nature Publishing Group

work page doi:10.1038/s41524-018-0085-8 2018
[4]

L. v. d. Maaten and G. Hinton, Visualizing Data using t- SNE , Journal of Machine Learning Research 9(86), 2579 (2008)

2008
[5]

V AMPnets for deep learning of molecular kinetics.Nature Communications, 9(1):5, 2018

A. Mardt, L. Pasquali, H. Wu and F. Noé, VAMPnets for deep learning of molecular kinetics , Nature Communications 9(1), 5 (2018), doi:10.1038/s41467-017-02388-1, Publisher: Nature Publishing Group

work page doi:10.1038/s41467-017-02388-1 2018
[6]

Ziatdinov, A

M. Ziatdinov, A. Ghosh, C. Y. Wong and S. V. Kalinin, AtomAI framework for deep learning analysis of image and spectroscopy data in electron and scanning probe microscopy , Nature Machine Intelligence 4(12), 1101 (2022), doi:10.1038/s42256-022-00555-8, Publisher: Nature Publishing Group

work page doi:10.1038/s42256-022-00555-8 2022
[7]

LeCun, C

Y. LeCun, C. Cortes and C. Burges, Mnist handwritten digit database, ATT Labs [Online]. Available: http://yann.lecun.com/exdb/mnist 2 (2010)

2010
[8]

S. Guo, S. M. Koh, A. R. Fritsch, I. B. Spielman and J. P. Zwolak, Soldet: Solitonic feature detection package, doi:10.18434/mds2-2641, URL: https://github.com/usnistgov/SolDet (2022)

work page doi:10.18434/mds2-2641 2022
[9]

F. Metz, J. Polo, N. Weber and T. Busch, Deep learning based quantum vortex detection in atomic Bose - Einstein condensates , Mach. Learn.: Sci. Technol. 2(3), 035019 (2021), doi:10.1088/2632-2153/abea6a

work page doi:10.1088/2632-2153/abea6a 2021
[10]

Doris, L

M. Doris, L. Ritter, S. Guo, S. M. Koh, A. R. Fritsch, I. B. Spielman and J. P. Zwolak, Q-GAIN: Quantum Gas Analysis and Inference Library , URL: https://github.com/Q-GAIN/Q-GAIN (2026)

2026
[11]

P. T. Starkey, C. J. Billington, S. P. Johnstone, M. Jasperse, K. Helmerson, L. D. Turner and R. P. Anderson, A scripted control system for autonomous hardware-timed experiments, Rev. Sci. Instrum. 84(8), 085111 (2013), doi:10.1063/1.4817213

work page doi:10.1063/1.4817213 2013
[12]

J. D. Hunter, Matplotlib: A 2d graphics environment, Computing in Science & Engineering 9(3), 90 (2007), doi:10.1109/MCSE.2007.55

work page doi:10.1109/mcse.2007.55 2007
[13]

Ziegler, F

J. Ziegler, F. Luthi, M. Ramsey, F. Borjans, G. Zheng and J. P. Zwolak, Automated Extraction of Capacitive Coupling for Quantum Dot Systems , Physical Review Applied 19(5), 054077 (2023), doi:10.1103/PhysRevApplied.19.054077

work page doi:10.1103/physrevapplied.19.054077 2023
[14]

Pedregosa, G

F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos et al., Scikit-learn: Machine learning in P ython , Journal of Machine Learning Research 12, 2825 (2011)

2011
[15]

S. Guo, A. R. Fritsch, C. Greenberg, I. B. Spielman and J. P. Zwolak, Machine-learning enhanced dark soliton detection in Bose – Einstein condensates , Mach. Learn.: Sci. Technol. 2(3), 035020 (2021), doi:10.1088/2632-2153/abed1e, Publisher: IOP Publishing

work page doi:10.1088/2632-2153/abed1e 2021
[16]

J. P. Zwolak, S. Guo, A. R. Fritsch and I. B. Spielman, Dark solitons in BECs dataset 2.0 , doi:10.18434/MDS2-2363 (2021)

work page doi:10.18434/mds2-2363 2021
[17]

A. R. Fritsch, S. Guo, S. M. Koh, I. B. Spielman and J. P. Zwolak, Dark solitons in Bose - Einstein condensates: a dataset for many-body physics research , Mach. Learn.: Sci. Technol. 3, 047001 (2022), doi:10.1088/2632-2153/ac9454

work page doi:10.1088/2632-2153/ac9454 2022
[18]

S. Guo, S. M. Koh, A. R. Fritsch, I. B. Spielman and J. P. Zwolak, Combining machine learning with physics: A framework for tracking and sorting multiple dark solitons , Phys. Rev. Res. 4(2), 023163 (2022), doi:10.1103/PhysRevResearch.4.023163

work page doi:10.1103/physrevresearch.4.023163 2022

[1] [1]

J. A. Armstrong and L. Fletcher, Fast Solar Image Classification Using Deep Learning and Its Importance for Automation in Solar Physics , Solar Physics 294(6), 80 (2019), doi:10.1007/s11207-019-1473-z

work page doi:10.1007/s11207-019-1473-z 2019

[2] [2]

Carrasquilla and R

J. Carrasquilla and R. G. Melko, Machine learning phases of matter, Nature Physics 13(5), 431 (2017), doi:10.1038/nphys4035, Publisher: Nature Publishing Group

work page doi:10.1038/nphys4035 2017

[3] [3]

Stanev, C

V. Stanev, C. Oses, A. G. Kusne, E. Rodriguez, J. Paglione, S. Curtarolo and I. Takeuchi, Machine learning modeling of superconducting critical temperature, npj Computational Materials 4(1), 29 (2018), doi:10.1038/s41524-018-0085-8, Publisher: Nature Publishing Group

work page doi:10.1038/s41524-018-0085-8 2018

[4] [4]

L. v. d. Maaten and G. Hinton, Visualizing Data using t- SNE , Journal of Machine Learning Research 9(86), 2579 (2008)

2008

[5] [5]

V AMPnets for deep learning of molecular kinetics.Nature Communications, 9(1):5, 2018

A. Mardt, L. Pasquali, H. Wu and F. Noé, VAMPnets for deep learning of molecular kinetics , Nature Communications 9(1), 5 (2018), doi:10.1038/s41467-017-02388-1, Publisher: Nature Publishing Group

work page doi:10.1038/s41467-017-02388-1 2018

[6] [6]

Ziatdinov, A

M. Ziatdinov, A. Ghosh, C. Y. Wong and S. V. Kalinin, AtomAI framework for deep learning analysis of image and spectroscopy data in electron and scanning probe microscopy , Nature Machine Intelligence 4(12), 1101 (2022), doi:10.1038/s42256-022-00555-8, Publisher: Nature Publishing Group

work page doi:10.1038/s42256-022-00555-8 2022

[7] [7]

LeCun, C

Y. LeCun, C. Cortes and C. Burges, Mnist handwritten digit database, ATT Labs [Online]. Available: http://yann.lecun.com/exdb/mnist 2 (2010)

2010

[8] [8]

S. Guo, S. M. Koh, A. R. Fritsch, I. B. Spielman and J. P. Zwolak, Soldet: Solitonic feature detection package, doi:10.18434/mds2-2641, URL: https://github.com/usnistgov/SolDet (2022)

work page doi:10.18434/mds2-2641 2022

[9] [9]

F. Metz, J. Polo, N. Weber and T. Busch, Deep learning based quantum vortex detection in atomic Bose - Einstein condensates , Mach. Learn.: Sci. Technol. 2(3), 035019 (2021), doi:10.1088/2632-2153/abea6a

work page doi:10.1088/2632-2153/abea6a 2021

[10] [10]

Doris, L

M. Doris, L. Ritter, S. Guo, S. M. Koh, A. R. Fritsch, I. B. Spielman and J. P. Zwolak, Q-GAIN: Quantum Gas Analysis and Inference Library , URL: https://github.com/Q-GAIN/Q-GAIN (2026)

2026

[11] [11]

P. T. Starkey, C. J. Billington, S. P. Johnstone, M. Jasperse, K. Helmerson, L. D. Turner and R. P. Anderson, A scripted control system for autonomous hardware-timed experiments, Rev. Sci. Instrum. 84(8), 085111 (2013), doi:10.1063/1.4817213

work page doi:10.1063/1.4817213 2013

[12] [12]

J. D. Hunter, Matplotlib: A 2d graphics environment, Computing in Science & Engineering 9(3), 90 (2007), doi:10.1109/MCSE.2007.55

work page doi:10.1109/mcse.2007.55 2007

[13] [13]

Ziegler, F

J. Ziegler, F. Luthi, M. Ramsey, F. Borjans, G. Zheng and J. P. Zwolak, Automated Extraction of Capacitive Coupling for Quantum Dot Systems , Physical Review Applied 19(5), 054077 (2023), doi:10.1103/PhysRevApplied.19.054077

work page doi:10.1103/physrevapplied.19.054077 2023

[14] [14]

Pedregosa, G

F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos et al., Scikit-learn: Machine learning in P ython , Journal of Machine Learning Research 12, 2825 (2011)

2011

[15] [15]

S. Guo, A. R. Fritsch, C. Greenberg, I. B. Spielman and J. P. Zwolak, Machine-learning enhanced dark soliton detection in Bose – Einstein condensates , Mach. Learn.: Sci. Technol. 2(3), 035020 (2021), doi:10.1088/2632-2153/abed1e, Publisher: IOP Publishing

work page doi:10.1088/2632-2153/abed1e 2021

[16] [16]

J. P. Zwolak, S. Guo, A. R. Fritsch and I. B. Spielman, Dark solitons in BECs dataset 2.0 , doi:10.18434/MDS2-2363 (2021)

work page doi:10.18434/mds2-2363 2021

[17] [17]

A. R. Fritsch, S. Guo, S. M. Koh, I. B. Spielman and J. P. Zwolak, Dark solitons in Bose - Einstein condensates: a dataset for many-body physics research , Mach. Learn.: Sci. Technol. 3, 047001 (2022), doi:10.1088/2632-2153/ac9454

work page doi:10.1088/2632-2153/ac9454 2022

[18] [18]

S. Guo, S. M. Koh, A. R. Fritsch, I. B. Spielman and J. P. Zwolak, Combining machine learning with physics: A framework for tracking and sorting multiple dark solitons , Phys. Rev. Res. 4(2), 023163 (2022), doi:10.1103/PhysRevResearch.4.023163

work page doi:10.1103/physrevresearch.4.023163 2022