What's in a Smoothness Constant? Tighter Rates for Local SGD with Bounded Second-order Heterogeneity
Kumar Kshitij Patel, Rustem Islamov, Sebastian U Stich, Aurelien Lucchi, Eduard Gorbunov, Lingxiao Wang
cs.LG, math.OC, stat.ML
Submitted: 2026-07-16
License: http://creativecommons.org/licenses/by/4.0/
The gist: Local SGD, also known as Federated Averaging, is a widely used distributed optimization algorithm.
Terminology
Abstract
Local SGD, also known as Federated Averaging, is a widely used distributed optimization algorithm. Although Local SGD often outperforms alternatives such as Mini-batch SGD in practice, theory still only partially explains when and why local updates help under realistic data heterogeneity. Recent work by [Patel et al., 2025] shows that a bounded second-order heterogeneity assumption captures the efficiency of Local SGD for strongly convex objectives, and conjectures that the same principle extends to the general convex setting. In this paper, we prove this conjecture by establishing an improved convergence guarantee for Local SGD on general convex objectives under bounded second-order heterogeneity. We also improve the best-known lower bounds for Local SGD in this setting, showing that our upper bounds are nearly tight. Together, these results provide a sharper, more fine-grained convergence theory for Local SGD. As a further application of our techniques, we provide a lower bound for serial SGD with replacement, showing how second-order heterogeneity captures the impact of rare high-curvature clients.
Sources
- Communication-Efficient Learning of Deep Networks from Decentralized Data
- DiLoCo: Distributed Low-Communication Training of Language Models
- Advances and Open Problems in Federated Learning
- A Field Guide to Federated Optimization
- Local SGD Converges Fast and Communicates Little
- Mime: Mimicking Centralized Stochastic Algorithms in Federated Learning
- Bias-Variance Reduced Local SGD for Less Heterogeneous Federated Learning
- Parallel SGD: When does averaging help?
- The Minimax Complexity of Distributed Optimization
- On the Unreasonable Effectiveness of Federated Averaging with Heterogeneous Data
- What Makes Local Updates Effective: The Role of Data Heterogeneity and Smoothness
- Revisiting LocalSGD and SCAFFOLD: Improved Rates and Missing Analysis
- Understanding Outer Optimizers in Local SGD: Learning Rates, Momentum, and Acceleration
- Communication-Efficient Language Model Training Scales Reliably and Robustly: Scaling Laws for DiLoCo
- MuLoCo: Muon is a practical inner optimizer for DiLoCo
- Double Momentum and Error Feedback for Clipping with Fast Rates and Differential Privacy
- Smoothed Normalization for Efficient Distributed Private Optimization
- On the Outsized Importance of Learning Rates in Local Update Methods
- Don't Use Large Mini-Batches, Use Local SGD
- Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks