ReLU-GNNs Prove More Expressive Than TrReLU-GNNs for Boolean Queries

Pablo Barcel\'o, Floris Geerts, Matthias Lanzinger, Klara Pakhomenko, Jan Van den Bussche· August 14, 2026 View original

Key takeaways

  • ReLU-GNNs are strictly more expressive than TrReLU-GNNs for Boolean queries on Boolean-featured graphs.
  • This settles a long-standing open problem in GNN expressivity research.
  • The choice of activation function significantly impacts a GNN's computational power.
  • ReLU's unique properties make it superior for certain logical reasoning tasks on graphs.

Who benefits

AI/ML DevelopmentCybersecurityDrug DiscoverySocial Network Analysis

Summary

Researchers proved that ReLU-MPLang is strictly more powerful than TrReLU-MPLang for expressing Boolean queries on Boolean-featured graphs, settling an open problem regarding GNN expressivity.

The expressiveness of Graph Neural Networks (GNNs) is a critical area of research, particularly concerning the power of different activation functions. This paper focuses on comparing the capabilities of ReLU (Rectified Linear Unit) activation functions against other eventually constant activation functions, specifically TrReLU (truncated ReLU), within the context of Message Passing Languages (MPLang) on finite simple undirected graphs with Boolean node features. The core finding is a proof demonstrating that the Boolean queries expressible in $\Sigma$-MPLang, where $\Sigma$ represents any collection of eventually constant activation functions with arbitrary real coefficients, form a strict subclass of those expressible in ReLU-MPLang. This definitively settles a recently posed open problem regarding whether ReLU-MPLang possesses greater power than TrReLU-MPLang for Boolean queries. Consequently, this implies that ReLU-GNNs are strictly more expressive than {TrReLU,id}-GNNs when it comes to handling Boolean queries on graphs characterized by Boolean features. This result provides a fundamental understanding of the computational capabilities of different GNN architectures, highlighting ReLU's unique strength in this specific domain.

Why it matters

Understanding the expressivity of GNNs with different activation functions is crucial for designing more powerful and efficient graph-based AI models, especially for tasks involving discrete or logical relationships.

How to implement this in your domain

  1. 1Prioritize ReLU activation functions in GNN architectures when designing models for tasks involving Boolean logic or discrete features.
  2. 2Re-evaluate existing GNN models that use non-ReLU activation functions for Boolean-featured graphs to assess potential expressivity limitations.
  3. 3Explore the implications of this finding for GNN applications in areas like knowledge graphs, circuit design, or logical reasoning.
  4. 4Contribute to research on the theoretical foundations of GNNs to further understand activation function choices.

Original post by Pablo Barcel\'o, Floris Geerts, Matthias Lanzinger, Klara Pakhomenko, Jan Van den Bussche

"arXiv:2608.12617v1 Announce Type: new Abstract: We prove that, on finite simple undirected graphs equipped with a single Boolean node feature, the Boolean queries expressible in $\Sigma$-MPLang, for any collection $\Sigma$ of eventually constant activation functions and with arbi…"

View on X

Originally posted by Pablo Barcel\'o, Floris Geerts, Matthias Lanzinger, Klara Pakhomenko, Jan Van den Bussche on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI Engineering & DevToolsAI ResearchAI Investing

FlowLOB Generates Realistic, Controllable Limit Order Books Efficiently

This paper introduces FlowLOB, a conditional flow-matching generator for Limit Order Book (LOB) trajectories that offers realistic market dynamics, efficient sampling, and controllable scenario generation, outperforming existing agent-based and deep generative simulators. FlowLOB achieves high fidelity with significantly fewer computational steps than diffusion models and transfers effectively to unseen instruments.

Zhuohan Wang, Andreea Bacalum, Ollie Olby, Carmine Ventre, Namid StillmanAug 14, 2026
AI Engineering & DevToolsAI Research

Auditing Reveals Bias in Neural Combinatorial Optimization Benchmarks

This paper audits test-time budget allocation in Neural Combinatorial Optimization (NCO) solvers, revealing that reported gains from non-uniform sampling often stem from "sampling luck" rather than true allocation benefits on in-distribution data. It proposes a correction procedure and demonstrates real gains under distribution shift, emphasizing the need for rigorous evaluation.

Jinhyung BaeAug 14, 2026
AI Engineering & DevToolsAI Research

Diffusion Models Solve Mixed-Integer Optimization Problems Faster

This paper introduces Constrained Graph Diffusion (CGD), a novel learning-based approach that uses a graph-based generative diffusion model to approximately solve mixed-integer optimization problems (MIPs). CGD integrates a training-free feasibility projection operator into the diffusion process, significantly improving solution quality and feasibility while achieving substantial speedups over traditional numerical solvers.

Vincenzo Di Vito, Mehdi Taghizadeh, Deepjyoti Deka, Kaarthik Sundar, Ferdinando FiorettoAug 14, 2026