---
title: Oversquashing in Graph Neural Networks
url: https://www.emergentmind.com/topics/oversquashing
type: topic
---

# Oversquashing in Graph Neural Networks

Oversquashing is a failure mode of message-passing graph neural networks in which information from many nodes, often but not only at long range, is compressed into fixed-size intermediate representations as it traverses a graph. In the standard formulation, a node representation after \(L\) layers depends on its \(L\)-hop computation tree, so exponentially many signals may be forced through a sparse set of graph pathways and summarized in a constant-width vector; the resulting loss of influence is the canonical oversquashing picture [2208.03471]. Subsequent work has broadened this picture: oversquashing has been tied to information contraction, graph expansion, effective resistance, commute time, conductance, and Jacobian sensitivity; it has also been shown to admit a short-range bottleneck form that is not captured by long-range gradient-decay explanations alone [2302.06835], [2405.13526], [2603.26140], [2511.20406].

## 1. Conceptual definition and scope

At the level of a standard message-passing GNN, node representations evolve as
\[
h_{v}^{(l)}=f_{l}\bigl(h_{v}^{(l-1)},\{h_{u}^{(l-1)} : u \in N_{G}(v)\}\bigr),
\]
so \(h_v^{(L)}\) can only depend on nodes within \(L\) hops [2208.03471]. In this setting, oversquashing is the failure mode in which exponentially many long-range signals are squashed into fixed-size node vectors, causing the model to lose information needed for long-range tasks [2208.03471].

This description is distinct from, but often entangled with, two other limitations. Underreaching refers to the fact that a depth-\(K\) message-passing GNN cannot propagate information beyond \(K\) hops; oversmoothing refers to the convergence of node embeddings toward increasingly similar representations under repeated propagation [2312.16560], [2603.26140]. Much of the oversquashing literature studies the regime in which depth is sufficient to reach distant nodes but insufficient to preserve their influence.

A central refinement is that oversquashing is not limited to long-range tasks. “Short-Range Oversquashing” constructs tasks with radius \(2\) in which any exact MPNN solution requires hidden dimension growing with graph size, and separates two mechanisms: a bottleneck phenomenon, which can arise even in low-range settings, and a vanishing gradient phenomenon, which is closely associated with long-range tasks [2511.20406]. This suggests that “long-range dependency” is not a complete synonym for oversquashing.

## 2. Formal lenses and quantitative proxies

One influential lens is information contraction. For a Markov chain \(U \rightarrow X \rightarrow Y\), the data processing inequality gives
\[
I(U;Y) \le I(U;X),
\]
and the strong data processing inequality quantifies strict loss through contraction coefficients [2208.03471]. In this view, a deep message-passing computation graph resembles a noisy computation graph in which repeated local transformations quench distant signals.

A second lens is Jacobian sensitivity. A common operational quantity is
\[
\left\|\frac{\partial h_u^{(r)}}{\partial x_v}\right\|,
\]
or its normalized variant, which measures how much source node \(v\) can influence target node \(u\) after \(r\) layers [2302.06835], [2508.09265]. One topology-focused approximation defines
\[
\mathcal{\tilde{J}_\ell(v, u) = \frac{\mathbf{\tilde{A}^{\ell}_{uv}}{\sum_{k}{\mathbf{\tilde{A}^{\ell}_{kv}}},
\qquad \mathbf{\tilde A}=\mathbf A+\mathbf I,
\]
and models pairwise oversquashing through an exponential decay rate \(k_{vu}\) fitted from \(\ln \mathcal{\tilde J}_\ell(v,u)\) across depth [2508.09265]. This produces four graph-level statistics: prevalence, intensity, variability, and extremity [2508.09265].

A third lens is graph connectivity via effective resistance. For a connected graph with Laplacian \(L\),
\[
R_{u,v}=(\mathbf{1}_u-\mathbf{1}_v)^TL^+(\mathbf{1}_u-\mathbf{1}_v),
\]
and high effective resistance indicates weak communication through the graph [2302.06835]. In the same line of work, total effective resistance \(R_{\mathrm{tot}}\) is proposed as a graph-level proxy for total oversquashing, with rewiring aimed at minimizing it [2302.06835].

A related spectral-topological view uses commute time and conductance. In the virtual-node analysis, commute time is proportional to effective resistance,
\[
\tau(i,j)=2|E|\,R(i,j),
\]
and bounded-depth MPNNs require depth at least \(m \ge \tau(i,j)/8\) to realize graph functions with sufficiently strong mixing between nodes \(i\) and \(j\) [2405.13526]. In a more cut-theoretic formulation, conductance
\[
\phi(G)=\min_{S\subset V,\ 0<\operatorname{vol}(S)\le \operatorname{vol}(V)/2}
\frac{|\partial S|}{\min(\operatorname{vol}(S),\operatorname{vol}(V\setminus S))}
\]
serves as a proxy for the worst bottleneck in the graph [2603.26140].

## 3. Structural explanations and unsettled interpretations

The classical structural explanation ties oversquashing to sparse cuts, weak expansion, and bottlenecks. The Cheeger constant, spectral gap, and expander-graph intuition all encode the idea that too much information may have to cross too few edges [2208.03471], [2210.11790]. In this framing, poor expansion creates bottlenecks; bottlenecks induce information contraction; repeated contraction causes signal quenching [2208.03471].

Curvature-based explanations refine this by localizing bottlenecks. Ollivier–Ricci, Forman, balanced Forman, and related curvatures are used as edge-local indicators of poor transport geometry, and several rewiring methods target highly negatively curved edges [2411.17429]. Yet this interpretation is not universal. “Expander Graph Propagation” argues that propagation over sparse expander templates can alleviate bottlenecks while the propagation graph remains negatively curved everywhere; it further shows that for large sparse graphs, bounded degree, uniform expansion, and non-negative Ollivier curvature cannot all coexist [2210.02997]. A plausible implication is that negative curvature is not, by itself, a sufficient diagnosis of harmful oversquashing.

Another unsettled point is whether sensitivity alone is an adequate metric. “Adaptive Message Passing” argues that rewiring may increase
\[
\left\|\frac{\partial \mathbf h_v^L}{\partial \mathbf x_u}\right\|_1
\]
while also sending more information into the same bottleneck, so sensitivity can improve even when the compression burden worsens [2312.16560]. “Beyond Oversquashing” pushes this further by reframing the problem as one of signal routing rather than all-pairs communication: a model should move signal from task-relevant sources to task-relevant destinations while keeping it concentrated, rather than merely increasing generic diffusion [2605.13383].

## 4. Rewiring and structural transformation

Graph rewiring is the dominant mitigation family. A central idea is to alter topology so that long-range information need not cross narrow cuts. “Oversquashing in GNNs through the lens of information contraction and graph expansion” proposes Random Local Edge Flip and its greedy variant G-RLEF, a local degree-preserving rewiring scheme inspired by expander constructions; it preserves node degrees exactly, never disconnects the graph, and improves expansion while targeting bottleneck-like edges with few triangles [2208.03471].

Spectral rewiring methods pursue the same goal through global expansion proxies. FoSR adds the edge predicted to most improve the graph’s spectral gap using a first-order perturbation approximation, and combines this with a relational architecture that distinguishes original edges from added edges to control oversmoothing [2210.11790]. Effective-resistance rewiring instead selects edges that maximally decrease total effective resistance, thereby lowering a global oversquashing proxy [2302.06835]. Conductance-based optimal rewiring, however, is computationally intractable in the worst case: exact optimization of conductance under bounded edge edits is NP-hard, and the decision version is NP-complete under the paper’s formulation [2603.26140].

Other structural transformations depart from ordinary graph rewiring. One line lifts graphs to hypergraphs using Forman–Ricci curvature and turns curvature-identified backbones or bridge structures into hyperedges, thereby “virtually shortening the distance between potentially relevant nodes” [2508.11390]. Another uses fixed sparse expander templates as propagation graphs rather than graph-specific rewiring [2210.02997]. Both are motivated by the same anti-bottleneck principle, but they modify the communication domain more radically than edge addition alone.

## 5. Architectural remedies beyond explicit rewiring

Several approaches attempt to mitigate oversquashing without changing the original graph globally. Virtual nodes add a hub connected to every node, so the augmented graph \(G_{\mathsf{vn}}\) changes commute times and effective resistances in a topology-dependent way rather than uniformly helping every graph [2405.13526]. The same work shows that classical virtual nodes often induce uniform sensitivity to distant nodes, whereas a reordered variant, \( \mathrm{VN}_G \), can produce graph-structured heterogeneous sensitivity at the same computational complexity [2405.13526].

A different line modifies the propagation operator itself. “Graph Unitary Message Passing” proposes applying a unitary adjacency matrix for message passing; the abstract explicitly frames this as an alternative to rewiring techniques that improve graph spectrums but disrupt structural bias and have limited improvement on oversquashing in terms of oversquashing measure [2403.11199]. “SWAN” uses antisymmetry in both feature and spatial operators to obtain local and global non-dissipativity, with the stated consequence that the information propagation rate among graph nodes is constant, independent of time \(t\) [2405.01009].

Selective or asynchronous propagation is another remedy. “Adaptive Message Passing” learns both the effective depth and a soft layer-dependent filtering tensor \(\mathbf F\), so the model can suppress irrelevant outgoing messages rather than broadcasting everything synchronously [2312.16560]. “Asynchronous Message Passing” updates only centrality-ordered node batches at each layer; the framework is explicitly motivated by the claim that synchronous message passing compresses too much information at once through fixed-capacity channels [2509.06777].

Hierarchical and higher-order methods address oversquashing by changing the unit of communication. “Cross-attentive Cohesive Subgraph Embedding” builds \(k\)-core-derived cohesive subgraphs, pools them, lets them communicate via cross-attention, and fuses the resulting subgraph representations back into node embeddings; this creates representation-level alternative pathways rather than explicit new edges [2603.27529]. In topological deep learning, oversquashing has been generalized to simplicial and cellular message passing by representing such models as relational structures and analyzing them through an induced influence graph [2506.06582].

Redundancy control offers yet another angle. “On the Two Sides of Redundancy in Graph Neural Networks” argues that repetitive exchange and re-encoding of identical information amplifies oversquashing; it replaces unfolding trees with pruned neighborhood trees and computes embeddings on a merged DAG, yielding the ordering
\[
I_{\text{MPNN}(v,u)\leq I_{\text{TPT}(v,u) \leq I_{1\text{NT}(v,u) \leq I_{0\text{NT}(v,u),
\]
which the paper interprets as progressively less oversquashing under stronger pruning [2310.04190].

## 6. Empirical patterns, diagnostics, and open questions

Across the literature, rewiring and alternative architectures frequently improve long-range synthetic tasks and several real benchmarks, but the relationship between oversquashing reduction and downstream accuracy is not monotone. The causal analysis of rewiring strategies reports that most graph-classification datasets suffer from over-squashing, that rewiring often mitigates it, and that the degree of mitigation varies by dataset and method; it also reports that over-squashing is less notable in node-classification datasets, where rewiring often increases over-squashing and performance variations are uncorrelated with over-squashing changes [2508.09265]. This suggests that rewiring is most useful when over-squashing is substantial and corrected with restraint.

A recurrent empirical theme is that aggressive global densification and restrained topological surgery behave differently. DIGL can reduce topology-based oversquashing metrics strongly while failing to translate that reduction into better accuracy, whereas more restrained methods such as FoSR or BORF can align mitigation more closely with performance [2508.09265]. A plausible implication is that anti-oversquashing interventions must be evaluated jointly with their effects on oversmoothing, locality, and inductive bias.

The field also lacks a single accepted definition. Some papers privilege sensitivity decay, others effective resistance or commute time, others conductance or curvature, and still others selective routing or dynamical non-dissipativity [2302.06835], [2405.01009], [2605.13383]. The short-range results further show that existing long-range diagnostics do not capture every bottleneck mechanism [2511.20406]. An objective synthesis is therefore that oversquashing is best understood not as one invariant scalar, but as a family of related failures in which graph structure, propagation dynamics, representational width, and aggregation geometry jointly determine whether distant information remains influential.

Source: https://www.emergentmind.com/topics/oversquashing