SciPost Submission Page
QDFlow: A Python package for physics simulations of quantum dot devices
by Donovan L. Buterakos, Sandesh S. Kalantre, Joshua Ziegler, Jacob M Taylor, and Justyna P. Zwolak
Submission summary
| Authors (as registered SciPost users): | Justyna Zwolak |
| Submission information | |
|---|---|
| Preprint Link: | scipost_202510_00006v2 (pdf) |
| Code repository: | https://github.com/QDFlow/QDFlow-sim |
| Code version: | 1.0.1 |
| Code license: | GPL-2.0 |
| Date submitted: | Dec. 1, 2025, 8:01 p.m. |
| Submitted by: | Justyna Zwolak |
| Submitted to: | SciPost Physics Codebases |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approaches: | Theoretical, Experimental, Computational |
Abstract
Recent advances in machine learning (ML) have accelerated progress in calibrating and operating quantum dot (QD) devices. However, most ML approaches rely on access to large, representative datasets designed to capture the full spectrum of data quality encountered in practice, with both high- and low-quality data for training, benchmarking, and validation, with labels capturing key features of the device state. Collating such datasets experimentally is challenging due to limited data availability, slow measurement bandwidths, and the labor-intensive nature of labeling. QDFlow is an open-source physics simulator for multi-QD arrays that generates realistic synthetic data with ground-truth labels. QDFlow combines a self-consistent Thomas-Fermi solver, a dynamic capacitance model, and flexible noise modules to simulate charge stability diagrams and ray-based data closely resembling experiments. With an extensive set of parameters that can be varied and customizable noise models, QDFlow supports the creation of large, diverse datasets for ML development, benchmarking, and quantum device research.
Author comments upon resubmission
List of changes
- Added a full section and figure discussing benchmarks, testing, and limitations of QDFlow.
- Added several sentences to the introduction emphasizing prior work validating QDFlow data in experimental contexts.
- Added a link to the QDFlow API documentation in the introduction and in the code availability section.
- Added a sentence in the introduction mentioning the unit and benchmark tests in the QDFlow repository.
- Added a few sentences in the conclusion discussing planned future updates to QDFlow.
Current status:
Reports on this Submission
Report #1 by Anonymous (Referee 1) on 2025-12-22 (Invited Report)
The referee discloses that the following generative AI tools have been used in the preparation of this report:
Assisted by Gemini 3 Pro (accessed December 21, 2025) for language editing and preliminary verification of the final draft. The author takes full responsibility for the accuracy of the content.
Strengths
1- Proven research impact 2- Novel physics-informed method 3- Simplicity in comparison to previous methods 4- Flexible noise parameters 5- Utility in machine learning pipeline 6- Clear scalability benchmark, 7- Good contextualization
Weaknesses
1- Lack of quantum effects, like temperature, spin, tunnel coupling 2- Heuristic model of quantum dots 3-Uncertain applicability to experimentally relevant 2D quantum dot arrays
Report
I agree with their justification that for training machine learning models for autonomous tuning algorithms, capturing qualitative features (such as transition slopes and charging energies) is more critical than exact quantitative matching of capacitance matrices. The references provided regarding the experimental success of ML models trained on this data support this claim.
However, I stand by my previous comments that further progress in autonomous tuning depends on more accurate modeling of quantum effects, in particular tunnel couplings inducing state hybridization and anticrossings, which would be even more visible in 2D arrays. In those systems, larger crosstalk would most likely break the quasi-1D model and lead to additional charge transitions modifying the regular shapes of Coulomb diamonds. As an example, I am attaching a CSD of a 2x2 dot array. Simultaneously, I recognize the authors' empirical evidence for the model's utility for 2D array systems, and their honest discussion of the limitations in the newly added sentences.
The new benchmarking section clearly identifies the scalability bottlenecks, particularly the limit imposed by the integer optimization problem for arrays larger than ~20 dots. I have verified the documentation website, which now includes a detailed description of relevant functions, but still perhaps misses tutorial-style examples.
Overall, I believe the revised manuscript is much improved and provides a clear picture of the capabilities and limitations of QDFlow. After addressing a few minor comments below, I recommend acceptance of the manuscript in its revised form.
Requested changes
1- Could you discuss how the empirical argument underlying transfer learning holds for 2D arrays in the presence of crosstalk and charge hybridisation? Under what conditions can QDFlow be applied to heterostructure-based dots relevant to the community?
2- Following 1, I suggest adding more explicit references to other simulators or models that have attempted to model relevant quantum effects in 2D structures, for instance tunnel coupling. This would help contextualize the limitations of the current semiclassical approach.
3- I suggest improving the GitHub repository to include dependencies, installation instructions, and an example notebook that walks through generating a charge stability diagram. The notebook should be reflected in the documentation website as well, with a tutorial section.
Recommendation
Publish (easily meets expectations and criteria for this Journal; among top 50%)