On the Convergence Rates of Federated Q-Learning across Heterogeneous Environments
Pith reviewed 2026-05-23 20:37 UTC · model grok-4.3
The pith
In heterogeneous environments, federated Q-learning cannot achieve error decay faster than Θ(E/T) when agents average every E iterations.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The paper proves that in the presence of environmental heterogeneity, the ℓ_∞ norm of the error in synchronous federated Q-learning cannot decay faster than Θ(E/T) for a wide range of stepsizes, where E is the number of local iterations between averaging steps and T is the total number of iterations. This limit is fundamental and not due to analysis looseness. Additionally, the convergence exhibits a two-phase behavior with rapid initial decay followed by stabilization.
What carries the argument
Synchronous periodic averaging of local Q-estimates every E iterations, which interacts with persistent environmental heterogeneity to sustain differences across agents.
If this is right
- Sampling errors continue to enjoy linear speedup in the number of agents K.
- Larger averaging intervals E slow the overall error decay proportionally.
- The error still reaches zero as T grows, but only at the reduced rate set by E.
- Switching stepsizes after the observed phase-transition point improves total convergence speed.
Where Pith is reading between the lines
- Communication frequency may need to increase with the degree of environmental heterogeneity to preserve fast convergence.
- Analogous rate ceilings could appear in other federated reinforcement-learning methods that rely on periodic model averaging.
- Heterogeneity-aware schedules that adjust E over time might reduce the slowdown without raising communication cost.
Load-bearing premise
The analysis relies on synchronous periodic averaging where environmental heterogeneity creates persistent differences in local Q-estimates that infrequent communication does not fully cancel.
What would settle it
An experiment in which the ℓ_∞ error in a heterogeneous multi-agent Q-learning run with E>1 decays faster than order E/T for large T would contradict the proven lower bound.
Figures
read the original abstract
Large-scale multi-agent systems are often deployed across wide geographic areas, where agents interact with heterogeneous environments. There is an emerging interest in understanding the role of heterogeneity in the performance of the federated versions of classic reinforcement learning algorithms. In this paper, we study synchronous federated Q-learning, which aims to learn an optimal Q-function by having $K$ agents average their local Q-estimates per $E$ iterations. We observe an interesting phenomenon on the convergence speeds in terms of $K$ and $E$. Similar to the homogeneous environment settings, there is a linear speed-up concerning $K$ in reducing the errors that arise from sampling randomness. Yet, in sharp contrast to the homogeneous settings, $E>1$ leads to significant performance degradation. Specifically, we provide a fine-grained characterization of the error evolution in the presence of environmental heterogeneity, which decay to zero as the number of iterations $T$ increases. The slow convergence of having $E>1$ turns out to be fundamental rather than an artifact of our analysis. We prove that, for a wide range of stepsizes, the $\ell_{\infty}$ norm of the error cannot decay faster than $\Theta (E/T)$. In addition, our experiments demonstrate that the convergence exhibits an interesting two-phase phenomenon. For any given stepsize, there is a sharp phase-transition of the convergence: the error decays rapidly in the beginning yet later bounces up and stabilizes. Provided that the phase-transition time can be estimated, choosing different stepsizes for the two phases leads to faster overall convergence.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper analyzes synchronous federated Q-learning with K agents that average local Q-estimates every E iterations in heterogeneous MDPs. It claims linear speedup in K for sampling-induced errors (similar to homogeneous cases) but shows that E > 1 causes fundamental degradation due to persistent heterogeneity effects. The authors provide a fine-grained error evolution characterization that decays with T, prove that the ℓ_∞ error cannot decay faster than Θ(E/T) for a wide range of stepsizes, and report experiments exhibiting a two-phase convergence phenomenon (rapid initial decay followed by stabilization) that can be mitigated by stepsize scheduling across phases.
Significance. If the lower bound holds beyond the specific heterogeneity constructions examined, the result establishes a fundamental communication-frequency limit in heterogeneous federated RL, which is relevant for large-scale multi-agent deployments. The two-phase experimental observation and the explicit separation of sampling vs. heterogeneity errors are useful for algorithm design. The paper supplies both upper-bound analysis and matching lower bounds plus reproducible experiments, which strengthens its contribution relative to purely asymptotic claims.
major comments (2)
- [Lower-bound theorem] Lower-bound theorem (likely §4 or §5): the Θ(E/T) claim is presented as fundamental for heterogeneous environments, yet the construction appears to rely on MDPs where local optimal Q-functions differ by a fixed additive bias that periodic averaging cannot cancel. It is unclear whether the bound extends to general heterogeneous MDPs (e.g., those where optimal Q* coincide or where bias averages to zero under infrequent communication). This directly affects whether the result rules out faster rates under natural heterogeneity models.
- [Error decomposition] Error decomposition (likely Eq. (X) in the convergence analysis): the characterization separates sampling variance (which benefits from K) from a heterogeneity term that scales with E. The proof must explicitly verify that the heterogeneity term remains non-vanishing for the stated stepsize range; otherwise the lower bound reduces to a fitted quantity rather than an independent limit.
minor comments (2)
- [Abstract] The abstract states the lower bound holds 'for a wide range of stepsizes' without specifying the exact interval or dependence on E and heterogeneity parameters; this should be stated precisely in the theorem statement.
- [Experiments] Figure captions for the two-phase experiments should include the exact stepsize values, E, K, and MDP parameters used so that the phase-transition time can be reproduced.
Simulated Author's Rebuttal
We thank the referee for their thorough review and insightful comments on our manuscript. We address each major comment in detail below, providing clarifications and indicating where revisions will be made.
read point-by-point responses
-
Referee: [Lower-bound theorem] Lower-bound theorem (likely §4 or §5): the Θ(E/T) claim is presented as fundamental for heterogeneous environments, yet the construction appears to rely on MDPs where local optimal Q-functions differ by a fixed additive bias that periodic averaging cannot cancel. It is unclear whether the bound extends to general heterogeneous MDPs (e.g., those where optimal Q* coincide or where bias averages to zero under infrequent communication). This directly affects whether the result rules out faster rates under natural heterogeneity models.
Authors: Our lower bound is established for a specific class of heterogeneous MDPs in which the local optimal Q-functions differ by a constant additive bias that cannot be eliminated through periodic averaging. This construction is intended to capture scenarios where environmental heterogeneity leads to persistent discrepancies. We do not assert that the Θ(E/T) rate applies universally to all heterogeneous MDPs; indeed, when local optima coincide (reducing to the homogeneous case), faster convergence is possible. The result serves to highlight the potential fundamental limit imposed by communication frequency in the presence of non-cancellable heterogeneity. To address the concern, we will revise the manuscript to explicitly state the scope of the lower bound and provide a brief discussion on conditions under which faster rates might hold. revision: partial
-
Referee: [Error decomposition] Error decomposition (likely Eq. (X) in the convergence analysis): the characterization separates sampling variance (which benefits from K) from a heterogeneity term that scales with E. The proof must explicitly verify that the heterogeneity term remains non-vanishing for the stated stepsize range; otherwise the lower bound reduces to a fitted quantity rather than an independent limit.
Authors: The error decomposition in our analysis separates the sampling-induced error, which scales with 1/sqrt(KT) or similar, from the heterogeneity-induced term that scales with E/T. The proof shows that the heterogeneity term arises directly from the mismatch in local transition and reward functions and remains non-vanishing under the stepsize conditions considered (e.g., stepsizes of order 1/t). This is verified by bounding the bias term independently of the variance terms in the recursive error evolution. The lower bound then follows by showing that this term dominates and cannot be canceled. We will add an explicit statement or corollary confirming that the heterogeneity term is Ω(E/T) and does not vanish for the relevant stepsize range. revision: partial
Circularity Check
No significant circularity; lower bound is an independent proof
full rationale
The paper's central claim is a mathematical proof that the ℓ_∞ error cannot decay faster than Θ(E/T) under the stated heterogeneity model and synchronous averaging. No quoted step reduces the claimed rate to a fitted parameter, self-citation chain, or definitional tautology. The lower bound is presented as derived from the model assumptions rather than constructed from the target quantity itself. The derivation chain remains self-contained against external benchmarks, with the two-phase convergence observation also arising from direct analysis rather than renaming or smuggling.
Axiom & Free-Parameter Ledger
axioms (1)
- domain assumption Synchronous federated Q-learning with periodic averaging every E iterations across heterogeneous environments
Lean theorems connected to this paper
-
IndisputableMonolith/Cost/FunctionalEquation.leanwashburn_uniqueness_aczel unclear?
unclearRelation between the paper passage and the cited Recognition theorem.
We prove that, for a wide range of stepsizes, the ℓ∞ norm of the error cannot decay faster than Θ(E/T)
-
IndisputableMonolith/Foundation/RealityFromDistinction.leanreality_from_one_distinction unclear?
unclearRelation between the paper passage and the cited Recognition theorem.
the slow convergence of having E>1 turns out to be fundamental rather than an artifact of our analysis
What do these tags mean?
- matches
- The paper's claim is directly supported by a theorem in the formal canon.
- supports
- The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
- extends
- The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
- uses
- The paper appears to rely on the theorem as machinery.
- contradicts
- The paper's claim conflicts with a theorem or certificate in the canon.
- unclear
- Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.
Reference graph
Works this paper leans on
-
[1]
Reinforcement learning and optimal control, volume 1
Dimitri Bertsekas. Reinforcement learning and optimal control, volume 1. Athena Scientific, 2019
work page 2019
-
[2]
Dimitri Bertsekas and John N Tsitsiklis. Neuro-dynamic programming. Athena Scientific, 1996
work page 1996
-
[3]
A finite time analysis of temporal difference learning with linear function approximation
Jalaj Bhandari, Daniel Russo, and Raghav Singal. A finite time analysis of temporal difference learning with linear function approximation. In Conference on learning theory, pp.\ 1691--1692. PMLR, 2018
work page 2018
-
[4]
Bdfl: a byzantine-fault-tolerance decentralized federated learning method for autonomous vehicle
Jin-Hua Chen, Min-Rong Chen, Guo-Qiang Zeng, and Jia-Si Weng. Bdfl: a byzantine-fault-tolerance decentralized federated learning method for autonomous vehicle. IEEE Transactions on Vehicular Technology, 70 0 (9): 0 8639--8652, 2021
work page 2021
-
[5]
Adaptive personalized federated learning,
Yuyang Deng, Mohammad Mahdi Kamani, and Mehrdad Mahdavi. Adaptive personalized federated learning. arXiv preprint arXiv:2003.13461, 2020
-
[6]
Thinh Doan, Siva Maguluri, and Justin Romberg. Finite-time analysis of distributed td (0) with linear function approximation on multi-agent reinforcement learning. In International Conference on Machine Learning, pp.\ 1626--1635. PMLR, 2019
work page 2019
-
[7]
Federated learning for vehicular internet of things: Recent advances and open issues
Zhaoyang Du, Celimuge Wu, Tsutomu Yoshinaga, Kok-Lim Alvin Yau, Yusheng Ji, and Jie Li. Federated learning for vehicular internet of things: Recent advances and open issues. IEEE Open Journal of the Computer Society, 1: 0 45--61, 2020
work page 2020
-
[8]
Federated reinforcement learning with environment heterogeneity
Hao Jin, Yang Peng, Wenhao Yang, Shusen Wang, and Zhihua Zhang. Federated reinforcement learning with environment heterogeneity. In International Conference on Artificial Intelligence and Statistics, pp.\ 18--37. PMLR, 2022
work page 2022
-
[9]
Peter Kairouz, H. Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Bennis, Arjun Nitin Bhagoji, Kallista Bonawitz, Zachary Charles, Graham Cormode, Rachel Cummings, Rafael G. L. D’Oliveira, Hubert Eichner, Salim El Rouayheb, David Evans, Josh Gardner, Zachary Garrett, Adrià Gascón, Badih Ghazi, Phillip B. Gibbons, Marco Gruteser, Zaid Harchaoui, Cha...
work page 2021
-
[10]
Scaffold: Stochastic controlled averaging for federated learning
Sai Praneeth Karimireddy, Satyen Kale, Mehryar Mohri, Sashank Reddi, Sebastian Stich, and Ananda Theertha Suresh. Scaffold: Stochastic controlled averaging for federated learning. In International Conference on Machine Learning, pp.\ 5132--5143. PMLR, 2020
work page 2020
-
[11]
Federated reinforcement learning: Linear speedup under markovian sampling
Sajad Khodadadian, Pranay Sharma, Gauri Joshi, and Siva Theja Maguluri. Federated reinforcement learning: Linear speedup under markovian sampling. In International Conference on Machine Learning, pp.\ 10997--11057. PMLR, 2022
work page 2022
-
[12]
Deep reinforcement learning for autonomous driving: A survey
B Ravi Kiran, Ibrahim Sobh, Victor Talpaert, Patrick Mannion, Ahmad A Al Sallab, Senthil Yogamani, and Patrick P \'e rez. Deep reinforcement learning for autonomous driving: A survey. IEEE Transactions on Intelligent Transportation Systems, 23 0 (6): 0 4909--4926, 2021
work page 2021
-
[13]
Sample complexity of asynchronous q-learning: Sharper analysis and variance reduction
Gen Li, Yuting Wei, Yuejie Chi, Yuantao Gu, and Yuxin Chen. Sample complexity of asynchronous q-learning: Sharper analysis and variance reduction. Advances in neural information processing systems, 33: 0 7031--7043, 2020 a
work page 2020
-
[14]
Is q-learning minimax optimal? a tight sample complexity analysis
Gen Li, Changxiao Cai, Yuxin Chen, Yuting Wei, and Yuejie Chi. Is q-learning minimax optimal? a tight sample complexity analysis. Operations Research, 72 0 (1): 0 222--236, 2024
work page 2024
-
[15]
Federated optimization in heterogeneous networks
Tian Li, Anit Kumar Sahu, Manzil Zaheer, Maziar Sanjabi, Ameet Talwalkar, and Virginia Smith. Federated optimization in heterogeneous networks. Proceedings of Machine Learning and Systems, 2: 0 429--450, 2020 b
work page 2020
-
[16]
Fedvision: An online visual object detection platform powered by federated learning
Yang Liu, Anbu Huang, Yun Luo, He Huang, Youzhi Liu, Yuanyuan Chen, Lican Feng, Tianjian Chen, Han Yu, and Qiang Yang. Fedvision: An online visual object detection platform powered by federated learning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pp.\ 13172--13179, 2020
work page 2020
-
[17]
Communication-efficient learning of deep networks from decentralized data
Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas. Communication-efficient learning of deep networks from decentralized data. In Artificial intelligence and statistics, pp.\ 1273--1282. PMLR, 2017
work page 2017
-
[18]
Reinforcement learning environment for tactical networks
Thies M \"o hlenhof, Norman Jansen, and Wiam Rachid. Reinforcement learning environment for tactical networks. In 2021 International Conference on Military Communication and Information Systems (ICMCIS), pp.\ 1--8. IEEE, 2021
work page 2021
-
[19]
D \"i ot: A federated self-learning anomaly detection system for iot
Thien Duc Nguyen, Samuel Marchal, Markus Miettinen, Hossein Fereidooni, N Asokan, and Ahmad-Reza Sadeghi. D \"i ot: A federated self-learning anomaly detection system for iot. In 2019 IEEE 39th International conference on distributed computing systems (ICDCS), pp.\ 756--767. IEEE, 2019
work page 2019
-
[20]
A reinforcement learning approach to robust scheduling of semiconductor manufacturing facilities
In-Beom Park, Jaeseok Huh, Joongkyun Kim, and Jonghun Park. A reinforcement learning approach to robust scheduling of semiconductor manufacturing facilities. IEEE Transactions on Automation Science and Engineering, 17 0 (3): 0 1420--1431, 2019
work page 2019
-
[21]
Fedsplit: An algorithmic framework for fast federated optimization
Reese Pathak and Martin J Wainwright. Fedsplit: An algorithmic framework for fast federated optimization. Advances in neural information processing systems, 33: 0 7057--7066, 2020
work page 2020
-
[22]
Fingerprint policy optimisation for robust reinforcement learning
Supratik Paul, Michael A Osborne, and Shimon Whiteson. Fingerprint policy optimisation for robust reinforcement learning. In International Conference on Machine Learning, pp.\ 5082--5091. PMLR, 2019
work page 2019
-
[23]
Muzi Peng, Jiangwei Wang, Dongjin Song, Fei Miao, and Lili Su. Privacy-preserving and uncertainty-aware federated trajectory prediction for connected autonomous vehicles. In 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp.\ 11141--11147, 2023. doi:10.1109/IROS55552.2023.10341638
-
[24]
Federated learning in vehicular networks: Opportunities and solutions
Jason Posner, Lewis Tseng, Moayad Aloqaily, and Yaser Jararweh. Federated learning in vehicular networks: Opportunities and solutions. IEEE Network, 35 0 (2): 0 152--159, 2021
work page 2021
-
[25]
Federated reinforcement learning: techniques, applications, and open challenges
Jiaju Qi, Qihao Zhou, Lei Lei, and Kan Zheng. Federated reinforcement learning: techniques, applications, and open challenges. Intelligence & Robotics, 2021. doi:10.20517/ir.2021.02. URL https://doi.org/10.20517
-
[26]
Federated Learning for Emoji Prediction in a Mobile Keyboard
Swaroop Ramaswamy, Rajiv Mathews, Kanishka Rao, and Fran c oise Beaufays. Federated learning for emoji prediction in a mobile keyboard. arXiv preprint arXiv:1906.04329, 2019
work page internal anchor Pith review Pith/arXiv arXiv 1906
-
[27]
Micah J Sheller, G Anthony Reina, Brandon Edwards, Jason Martin, and Spyridon Bakas. Multi-institutional deep learning modeling without sharing patient data: A feasibility study on brain tumor segmentation. In Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries: 4th International Workshop, BrainLes 2018, Held in Conjunction with M...
work page 2018
-
[28]
A non-parametric view of fedavg and fedprox: Beyond stationary points
Lili Su, Jiaming Xu, and Pengkun Yang. A non-parametric view of fedavg and fedprox: Beyond stationary points. Journal of Machine Learning Research, 24 0 (203): 0 1--48, 2023
work page 2023
-
[29]
Richard S. Sutton and Andrew G. Barto. Chapter 2.5 Tracking a Nonstationary Problem, Reinforcement Learning: An Introduction, chapter 8, pp.\ 33. The MIT Press, 2018
work page 2018
-
[30]
High-dimensional probability: An introduction with applications in data science, volume 47
Roman Vershynin. High-dimensional probability: An introduction with applications in data science, volume 47. Cambridge university press, 2018
work page 2018
-
[31]
Atpfl: Automatic trajectory prediction model design under federated learning framework
Chunnan Wang, Xiang Chen, Junzhe Wang, and Hongzhi Wang. Atpfl: Automatic trajectory prediction model design under federated learning framework. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp.\ 6563--6572, June 2022 a
work page 2022
-
[32]
Han Wang, Aritra Mitra, Hamed Hassani, George J Pappas, and James Anderson. Federated temporal difference learning with linear function approximation under environmental heterogeneity. arXiv preprint arXiv:2302.02212, 2023
-
[33]
Tackling the objective inconsistency problem in heterogeneous federated optimization
Jianyu Wang, Qinghua Liu, Hao Liang, Gauri Joshi, and H Vincent Poor. Tackling the objective inconsistency problem in heterogeneous federated optimization. Advances in neural information processing systems, 33: 0 7611--7623, 2020
work page 2020
-
[34]
On the unreasonable effectiveness of federated averaging with heterogeneous data
Jianyu Wang, Rudrajit Das, Gauri Joshi, Satyen Kale, Zheng Xu, and Tong Zhang. On the unreasonable effectiveness of federated averaging with heterogeneous data. arXiv preprint arXiv:2206.04723, 2022 b
-
[35]
Christopher Watkins and Peter Dayan. Q-learning. Machine Learning, 8: 0 279--292, 1992. URL https://api.semanticscholar.org/CorpusID:208910339
work page 1992
-
[36]
The blessing of heterogeneity in federated q-learning: Linear speedup and beyond
Jiin Woo, Gauri Joshi, and Yuejie Chi. The blessing of heterogeneity in federated q-learning: Linear speedup and beyond. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett (eds.), Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machine Learning Resear...
work page 2023
-
[37]
FedKL : Tackling Data Heterogeneity in Federated Reinforcement Learning by Penalizing KL Divergence
Zhijie Xie and Shenghui Song. FedKL : Tackling Data Heterogeneity in Federated Reinforcement Learning by Penalizing KL Divergence . IEEE Journal on Selected Areas in Communications, 41 0 (4): 0 1227--1242, April 2023. ISSN 1558-0008. doi:10.1109/JSAC.2023.3242734. URL https://ieeexplore.ieee.org/abstract/document/10038492?casa_token=yGyMDlnL_FsAAAAA:hqNvz...
-
[38]
Experiments of federated learning for covid-19 chest x-ray images
Bingjie Yan, Jun Wang, Jieren Cheng, Yize Zhou, Yixian Zhang, Yifan Yang, Li Liu, Haojiang Zhao, Chunjuan Wang, and Boyi Liu. Experiments of federated learning for covid-19 chest x-ray images. In Advances in Artificial Intelligence and Security: 7th International Conference, ICAIS 2021, Dublin, Ireland, July 19-23, 2021, Proceedings, Part II 7, pp.\ 41--5...
work page 2021
-
[39]
Applied Federated Learning: Improving Google Keyboard Query Suggestions
Timothy Yang, Galen Andrew, Hubert Eichner, Haicheng Sun, Wei Li, Nicholas Kong, Daniel Ramage, and Fran c oise Beaufays. Applied federated learning: Improving google keyboard query suggestions. arXiv preprint arXiv:1812.02903, 2018
work page internal anchor Pith review Pith/arXiv arXiv 2018
-
[40]
Learning context-aware policies from multiple smart homes via federated multi-task learning
Tianlong Yu, Tian Li, Yuqiong Sun, Susanta Nanda, Virginia Smith, Vyas Sekar, and Srinivasan Seshan. Learning context-aware policies from multiple smart homes via federated multi-task learning. In 2020 IEEE/ACM Fifth International Conference on Internet-of-Things Design and Implementation (IoTDI), pp.\ 104--115. IEEE, 2020
work page 2020
-
[41]
Federated learning on the road autonomous controller design for connected and autonomous vehicles
Tengchan Zeng, Omid Semiari, Mingzhe Chen, Walid Saad, and Mehdi Bennis. Federated learning on the road autonomous controller design for connected and autonomous vehicles. IEEE Transactions on Wireless Communications, 21 0 (12): 0 10407--10423, 2022
work page 2022
-
[42]
Finite-time analysis of on-policy heterogeneous federated reinforcement learning
Chenyu Zhang, Han Wang, Aritra Mitra, and James Anderson. Finite-time analysis of on-policy heterogeneous federated reinforcement learning. In The Twelfth International Conference on Learning Representations, 2023 a
work page 2023
-
[43]
Finite-time analysis of on-policy heterogeneous federated reinforcement learning
Chenyu Zhang, Han Wang, Aritra Mitra, and James Anderson. Finite-time analysis of on-policy heterogeneous federated reinforcement learning. arXiv preprint arXiv:2401.15273, 2024
-
[44]
On the convergence of sarsa with linear function approximation
Shangtong Zhang, Remi Tachet Des Combes, and Romain Laroche. On the convergence of sarsa with linear function approximation. In International Conference on Machine Learning, pp.\ 41613--41646. PMLR, 2023 b
work page 2023
-
[45]
Federated q-learning: Linear regret speedup with low communication cost
Zhong Zheng, Fengyu Gao, Lingzhou Xue, and Jing Yang. Federated q-learning: Linear regret speedup with low communication cost. arXiv preprint arXiv:2312.15023, 2023
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.