# Capacity-Controlled Global Attention for Graph Transformers

arXiv: https://arxiv.org/abs/2604.17324 · HTML: https://mlsanity.com/paper/2604.17324/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 30 of its 50 references that are in the map.
- First posted on arXiv 19 Apr 2026.

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

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Graph Attention Networks](https://mlsanity.com/paper/1710.10903/) (2017)
- [Layer Normalization](https://mlsanity.com/paper/1607.06450/) (2016)
- [Neural Message Passing for Quantum Chemistry](https://mlsanity.com/paper/1704.01212/) (2017)
- [Open Graph Benchmark: Datasets for Machine Learning on Graphs](https://mlsanity.com/paper/2005.00687/) (2020)
- [GLU Variants Improve Transformer](https://mlsanity.com/paper/2002.05202/) (2020)
- [Deeper Insights into Graph Convolutional Networks for Semi-Supervised Learning](https://mlsanity.com/paper/1801.07606/) (2018)
- [Predict then Propagate: Graph Neural Networks meet Personalized PageRank](https://mlsanity.com/paper/1810.05997/) (2018)
- [On Layer Normalization in the Transformer Architecture](https://mlsanity.com/paper/2002.04745/) (2020)
- [Gated Graph Sequence Neural Networks](https://mlsanity.com/paper/1511.05493/) (2015)
- [DropEdge: Towards Deep Graph Convolutional Networks on Node Classification](https://mlsanity.com/paper/1907.10903/) (2019)
- [On the Bottleneck of Graph Neural Networks and its Practical Implications](https://mlsanity.com/paper/2006.05205/) (2020)
- [Measuring and Relieving the Over-smoothing Problem for Graph Neural Networks from the Topological View](https://mlsanity.com/paper/1909.03211/) (2019)
- [Graph Transformer Networks](https://mlsanity.com/paper/1911.06455/) (2019)
- [Principal Neighbourhood Aggregation for Graph Nets](https://mlsanity.com/paper/2004.05718/) (2020)
- [Rethinking Graph Transformers with Spectral Attention](https://mlsanity.com/paper/2106.03893/) (2021)
- [Residual Gated Graph ConvNets](https://mlsanity.com/paper/1711.07553/) (2017)
- [Graph Neural Networks with Learnable Structural and Positional Representations](https://mlsanity.com/paper/2110.07875/) (2021)
- [PairNorm: Tackling Oversmoothing in GNNs](https://mlsanity.com/paper/1909.12223/) (2019)
- [Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free](https://mlsanity.com/paper/2505.06708/) (2025)
- [DeeperGCN: All You Need to Train Deeper GCNs](https://mlsanity.com/paper/2006.07739/) (2020)
- [NodeFormer: A Scalable Graph Structure Learning Transformer for Node Classification](https://mlsanity.com/paper/2306.08385/) (2023)
- [Stabilizing Transformer Training by Preventing Attention Entropy Collapse](https://mlsanity.com/paper/2303.06296/) (2023)
- [Graph Inductive Biases in Transformers without Message Passing](https://mlsanity.com/paper/2305.17589/) (2023)
- [NAGphormer: A Tokenized Graph Transformer for Node Classification in Large Graphs](https://mlsanity.com/paper/2206.04910/) (2022)
- [SGFormer: Simplifying and Empowering Transformers for Large-Graph Representations](https://mlsanity.com/paper/2306.10759/) (2023)
- [Walking Out of the Weisfeiler Leman Hierarchy: Graph Learning Beyond Message Passing](https://mlsanity.com/paper/2102.08786/) (2021)
- [Replacing softmax with ReLU in Vision Transformers](https://mlsanity.com/paper/2309.08586/) (2023)
- [On the Stability of Expressive Positional Encodings for Graphs](https://mlsanity.com/paper/2310.02579/) (2023)
- [Graph Convolutions Enrich the Self-Attention in Transformers!](https://mlsanity.com/paper/2312.04234/) (2023)

## Built on by (most cited first, 2 of 2)

- [Gated Graph Attention Networks with Learnable Temperature](https://mlsanity.com/paper/2605.29803/) (2026)
- [Why don't Graph Transformers attend to the same token?](https://mlsanity.com/paper/2610.19324/) (2026)

## Abstract (arXiv)

Global self-attention drives modern graph transformers, yet the softmax at its core imposes a structural constraint rarely examined directly: every attention row is non-negative and sums to one, so each per-head output is a mass-conserving convex combination of value vectors. A node can never "attend to nothing." We argue this conservation constraint is a single root cause behind three pathologies usually studied in isolation: the collapse of node representations with depth (over-smoothing), a low-rank bottleneck on per-head outputs, and brittle optimization in deep stacks. Drawing on how sigmoid gating removes analogous attention sinks in language models, we introduce SigGate-GT, a graph transformer that applies a learned, per-head, input-conditioned sigmoid gate to the attention output inside the GraphGPS framework. The gate is a smooth, per-dimension "volume control" that can drive head outputs toward zero, relaxing the constraint without abandoning attention's probabilistic interpretation. Analytically and through synthetic experiments, we show the gate strictly increases the stable rank of per-head outputs, and connect this rank gain to all three manifestations. On five molecular and long-range benchmarks, SigGate-GT matches the prior best on ZINC (0.059 MAE), records the strongest result among the graph-transformer baselines we evaluate on ogbg-molhiv (82.47% ROC-AUC), and is competitive on ogbg-molpcba and the Long-Range Graph Benchmark, with statistically significant gains over GraphGPS on all five datasets (p < 0.05). Mechanism analyses confirm the diagnosis: gating slows over-smoothing (a 30% mean relative gain in representation diversity across 4-16 layers), keeps attention entropy from collapsing, and stabilizes training across a 10x learning-rate range, at about 1% parameter overhead on OGB and under 3% wall-clock cost.