# What Does the Rank Buy? A Spectral and Distributional Analysis of Low-Rank Adaptation

arXiv: https://arxiv.org/abs/2609.32002 · HTML: https://mlsanity.com/paper/2609.32002/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 32 of its 44 references that are in the map.
- First posted on arXiv 25 Sep 2026.

## Builds on (most cited first, 30 of 32)

- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Parameter-Efficient Transfer Learning for NLP](https://mlsanity.com/paper/1902.00751/) (2019)
- [Prefix-Tuning: Optimizing Continuous Prompts for Generation](https://mlsanity.com/paper/2101.00190/) (2021)
- [An Introduction to Matrix Concentration Inequalities](https://mlsanity.com/paper/1501.01571/) (2015)
- [LoraHub: Efficient Cross-Task Generalization via Dynamic LoRA Composition](https://mlsanity.com/paper/2307.13269/) (2023)
- [AdaLoRA: Adaptive Budget Allocation for Parameter-Efficient Fine-Tuning](https://mlsanity.com/paper/2303.10512/) (2023)
- [VeRA: Vector-based Random Matrix Adaptation](https://mlsanity.com/paper/2310.11454/) (2023)
- [ReLoRA: High-Rank Training Through Low-Rank Updates](https://mlsanity.com/paper/2307.05695/) (2023)
- [Scaling Down to Scale Up: A Guide to Parameter-Efficient Fine-Tuning](https://mlsanity.com/paper/2303.15647/) (2023)
- [A vector-contraction inequality for Rademacher complexities](https://mlsanity.com/paper/1605.00251/) (2016)
- [Theoretical Analysis of Domain Adaptation with Optimal Transport](https://mlsanity.com/paper/1610.04420/) (2016)
- [The Expressive Power of Low-Rank Adaptation](https://mlsanity.com/paper/2310.17513/) (2023)
- [LoRAMoE: Alleviate World Knowledge Forgetting in Large Language Models via MoE-Style Plugin](https://mlsanity.com/paper/2312.09979/) (2023)
- [A Survey on LoRA of Large Language Models](https://mlsanity.com/paper/2407.11046/) (2024)
- [MoRAL: MoE Augmented LoRA for LLMs' Lifelong Learning](https://mlsanity.com/paper/2402.11260/) (2024)
- [VB-LoRA: Extreme Parameter Efficient Fine-Tuning with Vector Banks](https://mlsanity.com/paper/2405.15179/) (2024)
- [K-LoRA: Unlocking Training-Free Fusion of Any Subject and Style LoRAs](https://mlsanity.com/paper/2502.18461/) (2025)
- [DoRA: Enhancing Parameter-Efficient Fine-Tuning with Dynamic Rank Distribution](https://mlsanity.com/paper/2405.17357/) (2024)
- [MoRA: High-Rank Updating for Parameter-Efficient Fine-Tuning](https://mlsanity.com/paper/2405.12130/) (2024)
- [Hyperbolic Fine-Tuning for Large Language Models](https://mlsanity.com/paper/2410.04010/) (2024)
- [Keeping Yourself is Important in Downstream Tuning Multimodal Large Language Model](https://mlsanity.com/paper/2503.04543/) (2025)
- [GoRA: Gradient-driven Adaptive Low Rank Adaptation](https://mlsanity.com/paper/2502.12171/) (2025)
- [SVFit: Parameter-Efficient Fine-Tuning of Large Pre-Trained Models Using Singular Values](https://mlsanity.com/paper/2409.05926/) (2024)
- [RefLoRA: Refactored Low-Rank Adaptation for Efficient Fine-Tuning of Large Models](https://mlsanity.com/paper/2505.18877/) (2025)
- [LoLDU: Low-Rank Adaptation via Lower-Diag-Upper Decomposition for Parameter-Efficient Fine-Tuning](https://mlsanity.com/paper/2410.13618/) (2024)
- [RaSA: Rank-Sharing Low-Rank Adaptation](https://mlsanity.com/paper/2503.12576/) (2025)
- [LoR2C : Low-Rank Residual Connection Adaptation for Parameter-Efficient Fine-Tuning](https://mlsanity.com/paper/2503.00572/) (2025)
- [Bernoulli-LoRA: A Theoretical Framework for Randomized Low-Rank Adaptation](https://mlsanity.com/paper/2508.03820/) (2025)
- [Localized LoRA-MoE: Block-wise Low-Rank Experts With Adaptive Routing](https://mlsanity.com/paper/2607.05114/) (2026)
- [LoRA vs. Full Fine-Tuning: A Theoretical Perspective](https://mlsanity.com/paper/2605.19018/) (2026)

## Abstract (arXiv)

The rank r in LoRA is widely treated as a capacity control: a smaller rank is assumed to yield a simpler model that generalizes better. We show that, under hard per-factor norm budgets---the idealization of the weight decay and norm control used in practice---this intuition breaks down. The reason is structural: under such budgets, the updates LoRA can reach are exactly the matrices of rank at most r inside a nuclear-norm ball, and every complexity and displacement functional we analyze is maximized over this set by a rank-one update---so the rank cap never binds. The consequences follow directly. The linear-readout model class we study is identical for every r ≥ 1, its Rademacher complexity carries no dependence on r, and the distance the adaptation can move the source distribution obeys a rank-independent upper bound that we show is sharp. If rank does not control capacity, where does it act? We identify two places. Statistically, replacing the per-factor budgets with a joint budget on the product restores a data-dependent, rank-sensitive complexity bound---though the gain appears only for well-spread feature distributions, and the worst case remains rank-free. Spectrally, rank sets the price of adaptation: canceling the leading singular directions of the pretrained weight requires both sufficient rank and sufficient budget. We bound the smallest rank achieving a desired source--target alignment, with upper and lower bounds that match under two-sided spectral decay. Together, these results recast rank as governing which updates are reachable and what cancellation costs---not how much capacity the model has.