# GPoS: Geospatially-aware Proof of Stake

Crypto.world

> PoS blockchains decide voting power purely by how many tokens a validator stakes, ignoring where in the world that validator sits — so voting power can cluster in a few regions, exposing the network…

**arXiv**: 2511.02034

**Published**: 2025-11-03

**Authors**: Shashank Motepalli, Naman Garg, Gengrui Zhang, Hans-Arno Jacobsen

## AI summary

## Overview

**Research area:** Distributed systems and blockchain consensus — specifically Proof of Stake (PoS) validator decentralization, decentralization measurement, and Byzantine Fault Tolerant (BFT) consensus protocols (HotStuff, CometBFT).

**Technical level:** Intermediate. The paper combines empirical data collection, a graph-theoretic metric (eigenvector centrality, Gini coefficient), and consensus-layer mechanism design, but each piece is explained in accessible terms.

**Scope:** The paper empirically measures how geographically concentrated validator voting power is across five major PoS blockchains (Aptos, Avalanche, Ethereum, Solana, Sui), proposes a new metric (Gini of Eigenvector Centrality, GEC) and a new consensus mechanism (GPoS) that blends stake with geospatial diversity, and evaluates the resulting decentralization gains and performance cost.

## What This Paper Is About

PoS blockchains decide voting power purely by how many tokens a validator stakes, ignoring where in the world that validator sits — so voting power can cluster in a few regions, exposing the network to regional regulation, outages, and latency-based unfairness. The authors first quantify how bad this clustering actually is using real validator location and stake data, then propose GPoS, a modification to PoS that gives validators extra voting power for being geographically distant from the rest of the validator set, while keeping stake the dominant factor.

## Key Contributions

1. **A geospatially-aware voting mechanism (GPoS).** Voting power is recomputed at each epoch as a normalized linear combination of normalized stake and a normalized Geospatial Diversity Index (GDI), governed by a tunable parameter λ, where λ = 1 reproduces traditional PoS.
2. **A public dataset of validator geolocation and stake** across Aptos, Avalanche, Ethereum, Solana and Sui, including raw geolocations, stakes, and pre-processing scripts (proximity merging, stake aggregation), released for reproducibility.
3. **A new decentralization metric, GEC (Gini of Eigenvector Centrality),** designed to satisfy three properties the authors require — quantifiability, inequality sensitivity, and geospatial awareness — which they show no single existing metric satisfies.
4. **An empirical trade-off study** between geospatial decentralization and consensus performance, evaluated by emulating validator distributions across HotStuff and CometBFT.

## Main Findings

- **Voting power is geographically concentrated today.** Eigenvector centrality scores across all five blockchains are heavily skewed, with mean values substantially above medians (Ethereum is given as the example), and wide interquartile ranges for Avalanche and Solana indicating that a few validators hold outsized influence.
- **Gini coefficients of centrality range from 0.527 to 0.941.** The authors report 0.941 for Ethereum and 0.804 for Avalanche as signs of severe inequality, and note that even the lowest observed value, 0.527, still indicates centralization.
- **No existing metric covers all three requirements.** Nakamoto coefficient, Gini coefficient and entropy are quantifiable and inequality-sensitive but not geospatially aware; KDE is geospatially aware but not a single scalar and not inequality-sensitive; Moran's I is quantifiable and geospatially aware but insensitive to voting-power disparity.
- **GPoS improves geospatial decentralization by 45% on average.** Moving from λ = 1 (traditional PoS) to λ = 0.5 reduced the Gini coefficient of eigenvector centrality across every blockchain, with individual reductions between 41.38% and 49.72%. λ was swept from 0.5 to 0.9 in increments of 0.1.
- **Overhead in BFT protocols is minimal.** The abstract reports minimal performance overhead in HotStuff and CometBFT as measured by throughput and latency; the truncated Section 6 content does not give the specific throughput and latency numbers here.
- **Computation is practical at realistic scale.** GDI computation is O(n² log n) in the worst case, but with symmetry, parallelization, caching and incremental updates, the authors compute GDI for n = 10,000 validators in under 60 seconds on commodity hardware — against epochs of roughly 24 hours and validator sets in the hundreds to low thousands.
- **λ ≥ 0.5 is a deliberate security bound.** Allowing the geospatial term to outweigh stake (λ < 0.5) could let an adversary holding less than 1/3 of stake amass more than 1/3 of voting power by optimizing validator locations.
- **Location spoofing is handled economically, not cryptographically.** A dispute mechanism requires challengers to post collateral of 10% of their stake; a proven false location slashes the offender's stake with 20% of the slashed amount paid to the challenger, while a failed challenge burns the collateral. The authors cite optimistic rollup data of roughly 25,000–300,000 gas (≈$0.01–$15) per dispute and disputes in fewer than 0.01% of transactions.
- **Data cleaning dropped meaningful stake in some chains.** Excluded stake and validator counts were 31.57% (482 validators) for Avalanche, 5.5% (49) for Aptos, and 0.07% (931) for Solana; validators within a 20 km radius were merged to avoid local artifacts such as neighboring data centers.

## Methodology in Plain English

The authors start by gathering what no one had assembled before: where validators actually are and how much they have staked. They pulled data from public explorer APIs for Solana, Aptos and Avalanche, scraped node data for Ethereum, and obtained Sui's data privately on request; where explicit locations were missing they used IP-based geolocation. Ethereum required special handling because full nodes and validators are not distinguishable from the outside — so they watched which short-lived beacon-node subnets validators subscribed to across epochs to estimate validators per node, assuming every validator holds 32 ETH.

They then cleaned the data: dropping validators with no location, and merging any validators within 20 km of each other (summing their stake and keeping one node) so that the analysis reflects global geography rather than neighboring data centers. Validator counts fell sharply in preprocessing, for example from 10,803 to 1,046 for Ethereum's beacon-subnet dataset and from 2,310 to 118 for Solana.

To measure decentralization they built a graph where each validator is a node, edge weights combine both validators' voting power with a distance-decayed proximity term (1 − Δ/Δ_max, using haversine distance in kilometers), and then computed eigenvector centrality — the idea being that a validator is influential if it is close to other high-stake validators, since proximity helps reach the two-thirds quorum. They then applied the Gini coefficient to those centrality scores, producing GEC. This lets them test one number per blockchain, and check the definition of "decentralized" against three explicit properties they set out.

For the mechanism, each validator's Geospatial Diversity Index is the summed distance to the nearest set of validators whose combined normalized stake, plus its own, reaches the two-thirds quorum — a high GDI means you are far from the quorum-forming cluster. GDI is normalized by the maximum, blended linearly with normalized stake via λ, and renormalized so all voting powers sum to one. The change is applied only at epoch boundaries through the existing reconfiguration path (in CometBFT, the validator updates returned in EndBlock), so implementation requires no protocol rewrite. Finally, they re-ran their centrality analysis on the collected real-world data for different λ values, and emulated validator distributions in HotStuff and CometBFT to measure throughput and latency.

## Why This Matters

**Impact on research.** The paper reframes decentralization as a multi-dimensional property rather than a single stake-distribution question, and shows that widely used metrics (Nakamoto coefficient, Gini, entropy) are blind to geography. It also provides an openly released dataset and a reproducible metric, which lets other researchers benchmark geospatial decentralization instead of arguing about it qualitatively.

**Real-world applications:**

- **Validator siting and staking policy:** staking providers and foundations can use GEC and GDI to see which regions are over-represented and to justify location-diversity requirements for their validator sets.
- **Regulatory risk assessment:** the paper's motivating example is the U.S. SEC asserting jurisdiction over Ethereum transactions based on validator locations, plus Treasury sanctions on addresses; GEC gives regulators and chain governors a quantitative handle on jurisdictional exposure.
- **Infrastructure resilience planning:** the 2021 AWS data center outage that affected Solana's liveness is the canonical case — geographic and provider diversity directly limits correlated failure.
- **Fairness and MEV:** because proximity confers latency advantages that feed front-running and MEV capture, reweighting voting power toward distant validators broadens who can participate in block production and MEV.

**Industry relevance.** The mechanism is deliberately incremental: it changes only how voting power ρ is computed, reuses existing epoch reconfiguration, and is claimed to be composable with external Proof-of-Location systems rather than replacing them. That makes it plausible as a parameter change for chains running CometBFT-style consensus, though the authors are explicit that formal safety and liveness proofs under the modified voting power distribution remain open.

## Future Directions

- **Formal safety and liveness proofs.** The authors state directly that a theoretical analysis of BFT safety and liveness under GPoS's modified voting power distribution is left for future work.
- **Game-theoretic analysis of validator behavior.** The paper mitigates location spoofing and Sybil attacks, but explicitly flags long-term strategic behavior under GPoS as an important open area — including whether relocation incentives produce the intended decentralization or something else.
- **Verified location attestation.** GPoS assumes accurate coordinates and depends on an external Proof-of-Location or topology-based geolocation mechanism (ping-latency triangulation, oracle services); integrating and hardening such a system is treated as complementary, out-of-scope work.
- **Broader performance characterization.** The reported overhead is limited to HotStuff and CometBFT emulations at the validator-set sizes studied; behavior under adversarial network conditions, larger validator sets, and other BFT protocols is not established in the content summarized here.

## Target Audience

Blockchain protocol researchers and consensus engineers working on PoS and BFT systems; staking providers and validator operators who need to reason about location-based risk; decentralization researchers who need a metric that accounts for geography; and policy or risk analysts assessing jurisdictional concentration of blockchain infrastructure. Readers with a basic grasp of staking and BFT quorums will follow it comfortably; the graph-theoretic portions (eigenvector centrality, the distance-weighted adjacency matrix) reward some linear-algebra familiarity but are not the barrier.

## Authors’ abstract

Geospatial decentralization is essential for blockchains, ensuring regulatory resilience, robustness, and fairness. We empirically analyze five major Proof of Stake (PoS) blockchains: Aptos, Avalanche, Ethereum, Solana, and Sui, revealing that a few geographic regions dominate consensus voting power, resulting in limited geospatial decentralization. To address this, we propose Geospatially aware Proof of Stake (GPoS), which integrates geospatial diversity with stake-based…

[The full abstract on arXiv](https://arxiv.org/abs/2511.02034)

[Read the original paper](https://arxiv.org/abs/2511.02034)

---

Canonical: https://crypto.world/paper/gpos-geospatially-aware-proof-of-stake-2511.02034
