arxiv: 2604.16718 · v1 · submitted 2026-04-17 · 💻 cs.ET

Recognition: unknown

Potential Energy Savings from Quantum Computing-Based Route Optimization

Adriana Caraeni, Ayush Nadiger, Katie Schouten

Authors on Pith no claims yet

Pith reviewed 2026-05-10 06:24 UTC · model grok-4.3

classification 💻 cs.ET

keywords QAOAvehicle routingenergy consumptionroute optimizationquantum computingQUBOlogisticssustainable transport

0 comments

The pith

QAOA provides higher-quality route solutions than classical methods while consuming three orders of magnitude less energy on small instances.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper benchmarks the Quantum Approximate Optimization Algorithm for solving the traveling salesman and vehicle routing problems encoded as quadratic unconstrained binary optimization tasks. On graphs with 5 to 20 nodes, QAOA achieves better approximation ratios, runs 2 to 3 times faster than simulated annealing, and uses far less energy than either simulated annealing or genetic algorithms. These advantages, if they extend to larger real-world problems, point to meaningful reductions in fuel use and carbon emissions for logistics operations.

Core claim

By implementing QAOA circuits of depth 3 to 5 on Euclidean graphs of sizes 5, 10, and 20, the work shows that QAOA attains approximation ratios of 0.953, 0.921, and 0.903 respectively. These results outperform simulated annealing and genetic algorithms by 2.7 to 4.4 percent in solution quality. The quantum approach also completes calculations 2 to 3 times faster and reduces energy consumption by three orders of magnitude to the picojoule range. Extrapolating an 8.2 percent gain in routing efficiency to U.S. logistics yields annual savings of 2.62 exajoules of fuel and avoidance of 194 million tonnes of CO2 emissions.

What carries the argument

QAOA circuits of depth p=3-5 applied to QUBO formulations of traveling salesman and vehicle routing problems

If this is right

QAOA achieves higher solution quality than classical heuristics on small Euclidean route graphs.
Wall-clock runtimes are 2-3 times shorter than for simulated annealing.
Energy use per solution falls into the picojoule range, three orders below classical methods.
An 8.2% improvement in routing efficiency could save 2.62 EJ of fuel per year in the United States.
Such efficiency would avoid approximately 194 million tonnes of CO2 emissions annually.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Similar energy advantages might appear in other combinatorial optimization tasks beyond routing if the same QUBO encoding is used.
Integration into hybrid quantum-classical fleet systems could become practical once hardware supports larger instances.
Real-world validation would require testing on graphs with hundreds of nodes to confirm the scaling of both solution quality and energy metrics.
The projected emissions reductions assume that the efficiency gain applies directly without accounting for quantum hardware overheads in larger systems.

Load-bearing premise

That performance advantages and energy reductions measured on Euclidean graphs with N ≤ 20 will scale proportionally to real-world vehicle routing problems with hundreds of nodes and that the picojoule energy figures will remain representative when executed on future large-scale quantum hardware.

What would settle it

Running QAOA on vehicle routing instances with 100 or more nodes and measuring whether approximation ratios stay above those of simulated annealing and genetic algorithms while energy per solution remains in the picojoule range on physical quantum hardware.

Figures

Figures reproduced from arXiv: 2604.16718 by Adriana Caraeni, Ayush Nadiger, Katie Schouten.

**Figure 2.** Figure 2: compares execution times. QAOA runtimes scale sub-quadratically: 3.2 s for N = 5, 7.4 s for N = 10, and 18 s for N = 20 - while SA runtimes grow roughly linearly in the number of required iterations: 9.8 s, 25 s, and 48 s, respectively. Thus, QAOA runs approximately 2–3× faster than SA across these mid-scale instances. This efficiency stems from the fixed-depth quantum circuit and parallel state evolution … view at source ↗

**Figure 3.** Figure 3: Estimated energy consumption as a function of [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

read the original abstract

We investigate the potential of the Quantum Approximate Optimization Algorithm (QAOA) for reducing energy consumption in route planning, a key challenge in logistics due to the NP-hard nature of the Traveling Salesman and Vehicle Routing Problems. By encoding route optimization as a Quadratic Unconstrained Binary Optimization (QUBO) problem and implementing QAOA circuits at depth p = 3-5 alongside classical baselines of Simulated Annealing (SA) and Genetic Algorithms (GA), we perform systematic benchmarks on Euclidean graphs of sizes N = 5, 10, and 20. Our results demonstrate that QAOA attains higher solution quality with approximation ratios of 0.953 (N = 5), 0.921 (N = 10), and 0.903 (N = 20), outperforming SA and GA by 2.7-4.4%. Wall-clock runtimes for QAOA are 2-3x faster than SA across all tested sizes, and energy consumption measurements reveal a three-order-of-magnitude reduction, remaining in the picojoule range versus nanojoules for classical methods. Translating these gains to real-world logistics suggests an 8.2% improvement in routing efficiency could save approximately 2.62 EJ of fuel annually in the U.S., avoiding nearly 1.94 x 10^8 tonnes of CO2 emissions. These findings highlight QAOA's promise as a fast, energy-efficient optimizer for sustainable logistics applications and underscore its potential role in next-generation fleet-management systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

QAOA edges out classical methods on tiny Euclidean TSPs up to N=20 but the 8.2% efficiency gain and 2.62 EJ savings are a straight rescaling of those same small results with no scaling checks.

read the letter

The main thing to know is that this paper runs QAOA on small TSP instances and gets modestly better solutions, faster runtimes, and lower modeled energy than SA or GA, but then turns those results into large US fuel-savings numbers without testing whether the edge survives at realistic sizes. The benchmarks themselves are straightforward and internally consistent. They encode the problems as QUBO, use QAOA at depths 3-5 on Euclidean graphs of N=5, 10, and 20, report approximation ratios that fall from 0.953 to 0.903, and show QAOA beating the classical baselines by 2.7-4.4% while using less wall-clock time and far less energy in their circuit model. That part gives usable small-scale data. The weakness is the jump to real-world impact. The 8.2% routing improvement and the 2.62 EJ annual savings figure come directly from applying the small-N outperformance to national logistics statistics. There is no larger-instance testing, no analysis of how circuit depth or noise would change at hundreds of nodes, and the picojoule energy numbers rest on an idealized model that skips cryogenic cooling, control electronics, and readout costs. The paper already notes the approximation ratio declining with N, which undercuts the assumption that the advantage scales linearly. The core QAOA-for-TSP technique is not new, so the added value is the energy measurements and the savings projection. The projection, however, loops back to the same small benchmarks rather than providing an independent check. This work is mainly useful to people who want preliminary comparisons of quantum optimizers on logistics-style problems. A referee could usefully ask for scaling discussion or separation of the benchmark results from the policy claims. I would send it to review rather than desk reject because the small-scale data is there and the topic is relevant, even though the headline numbers need more support.

Referee Report

2 major / 1 minor

Summary. The manuscript benchmarks QAOA (p=3-5) on Euclidean TSP instances of size N=5, 10, and 20, reporting approximation ratios 0.953/0.921/0.903 that outperform SA and GA by 2.7-4.4%. QAOA is also reported to run 2-3x faster in wall-clock time and to consume three orders of magnitude less energy (picojoules versus nanojoules). These small-instance results are extrapolated to claim an 8.2% real-world routing-efficiency gain that would save 2.62 EJ of U.S. fuel annually and avoid 1.94×10^8 tonnes of CO2.

Significance. If the observed quality, runtime, and energy advantages were shown to persist at realistic VRP scales, the work would be significant for energy-efficient optimization in logistics. The systematic small-N benchmarks and direct energy measurements constitute a clear, internally consistent data point for near-term quantum heuristics; the explicit comparison against SA and GA is a strength.

major comments (2)

[Abstract and real-world translation paragraph] Abstract and the paragraph translating benchmark gains to real-world logistics: the 8.2% efficiency uplift (and the derived 2.62 EJ / 1.94×10^8 t CO2 figures) is obtained by rescaling the 2.7-4.4% outperformance measured on the same N≤20 instances; no scaling analysis, larger-instance validation, or justification for proportionality at hundreds of nodes is supplied, despite the observed decline in approximation ratio with N.
[Energy consumption analysis] Energy-consumption section (benchmarks and discussion): the picojoule figures rest on idealized circuit models that omit cryogenic cooling, control electronics, and readout overheads; this directly affects the three-order-of-magnitude reduction claim and the subsequent global energy-savings extrapolation.

minor comments (1)

[Abstract] The abstract could state the per-size outperformance margins explicitly rather than the aggregate 2.7-4.4% range.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and have revised the manuscript to strengthen the presentation of limitations while preserving the core contributions of the small-scale benchmarks.

read point-by-point responses

Referee: Abstract and the paragraph translating benchmark gains to real-world logistics: the 8.2% efficiency uplift (and the derived 2.62 EJ / 1.94×10^8 t CO2 figures) is obtained by rescaling the 2.7-4.4% outperformance measured on the same N≤20 instances; no scaling analysis, larger-instance validation, or justification for proportionality at hundreds of nodes is supplied, despite the observed decline in approximation ratio with N.

Authors: We agree that the 8.2% projection is obtained by assuming the observed relative outperformance (2.7-4.4%) scales proportionally to larger instances, and that no explicit scaling analysis or validation on N>>20 is provided. The manuscript already reports the decline in absolute approximation ratio with N, yet the margin over SA and GA remains consistent across the tested sizes. We have revised the abstract and the real-world translation paragraph to frame the 8.2% figure explicitly as an illustrative projection based on small-N trends, to state the proportionality assumption, and to add a forward-looking caveat on the necessity of larger-scale studies. These changes make the claims more precise without altering the reported benchmark data. revision: yes
Referee: Energy-consumption section (benchmarks and discussion): the picojoule figures rest on idealized circuit models that omit cryogenic cooling, control electronics, and readout overheads; this directly affects the three-order-of-magnitude reduction claim and the subsequent global energy-savings extrapolation.

Authors: We acknowledge that the reported picojoule-scale figures are obtained from an idealized gate-level energy model that does not incorporate cryogenic cooling, control electronics, or readout overheads. This model is standard for theoretical assessments of quantum algorithm energy but limits the direct applicability of the three-order-of-magnitude claim to full-system comparisons. We have revised the energy-consumption section to (i) detail the model assumptions, (ii) qualify the picojoule versus nanojoule comparison as applying to algorithmic execution energy, and (iii) discuss the potential impact of overheads on the overall advantage. The global fuel-savings extrapolation has been updated with corresponding language indicating that the figures represent potential benefits under the stated modeling assumptions. revision: yes

Circularity Check

0 steps flagged

No circularity detected; benchmarks and extrapolation remain independent

full rationale

The paper's core derivation consists of encoding TSP/VRP instances as QUBO, running QAOA at fixed depth p=3-5 on Euclidean graphs with N=5/10/20, and directly measuring approximation ratios, wall-clock times, and energy per solution. These quantities are obtained from explicit circuit execution and classical baselines on the same small instances; they are not defined in terms of the later global savings figure. The 8.2% routing-efficiency delta and the resulting 2.62 EJ / 1.94e8 tCO2 estimates are downstream linear projections that apply the observed deltas to external logistics statistics. Because the measured performance numbers are not forced by, or algebraically equivalent to, the extrapolated totals, no self-definitional, fitted-input-renamed-as-prediction, or self-citation-load-bearing reduction occurs. The derivation chain is therefore self-contained against the supplied benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The headline energy-savings projection rests on two untested domain assumptions about scalability and hardware energy behavior rather than on new free parameters or invented entities.

free parameters (2)

circuit depth p
Chosen range p=3-5; performance depends on this hyperparameter.
real-world efficiency uplift
8.2% figure obtained by extrapolating small-N outperformance percentages.

axioms (2)

domain assumption Small-instance QAOA advantages on Euclidean graphs generalize linearly to large-scale logistics networks
Required to convert 2.7-4.4% benchmark gains into the 8.2% real-world claim.
domain assumption Picojoule energy figures measured on small QAOA circuits remain representative at practical deployment scales
Underpins the three-order-of-magnitude energy advantage used for sustainability projections.

pith-pipeline@v0.9.0 · 5571 in / 1789 out tokens · 103846 ms · 2026-05-10T06:24:08.265641+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

35 extracted references · 3 canonical work pages · 1 internal anchor

[1]

G. Laporte. Fifty years of vehicle routing.Transportation Science, 2009

2009
[2]

M. O. Ball. Recent developments in the vehicle routing problem.Operations Research, 1994

1994
[3]

G. B. Dantzig and J. H. Ramser. The truck dispatching problem.Management Science, 1959

1959
[4]

Department of Transportation

U.S. Department of Transportation. Fuel efficiency and emission reduction strategies, 2018. Online resource

2018
[5]

Transportation and environmental sustainability, 2020

Environmental Protection Agency (EPA). Transportation and environmental sustainability, 2020. Online resource

2020
[6]

Global energy review: Transport sector, 2019

International Energy Agency (IEA). Global energy review: Transport sector, 2019. Online resource

2019
[7]

T. G. Crainic and G. Laporte.Fleet Management and Logistics. Springer, 1997

1997
[8]

Toth and D

P. Toth and D. Vigo.The Vehicle Routing Problem. SIAM, 2002

2002
[9]

Applegate, R

D. Applegate, R. Bixby, V . Chvátal, and W. Cook.The Traveling Salesman Problem: A Computational Study. Princeton University Press, 2007

2007
[10]

G. Reinelt. Tsplib—a traveling salesman problem library.ORSA Journal on Computing, 1991

1991
[11]

M. R. Garey and D. S. Johnson.Computers and Intractability: A Guide to the Theory of NP-Completeness. W. H. Freeman, 1979

1979
[12]

Dorigo and G

M. Dorigo and G. Di Caro. The ant colony optimization meta-heuristic. InNew Ideas in Optimization. 1999

1999
[13]

Kirkpatrick, C

S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi. Optimization by simulated annealing.Science, 1983

1983
[15]

Quantum Boltzmann Machine,

M. Benedetti, J. Realpe-Gómez, R. Biswas, and A. Perdomo-Ortiz. Quantum-assisted learning of hardware- embedded probabilistic graphical models, 2016. arXiv:1601.02036

work page arXiv 2016
[16]

D. E. Goldberg.Genetic Algorithms in Search, Optimization, and Machine Learning. Addison-Wesley, 1989

1989
[17]

M. A. Nielsen and I. L. Chuang.Quantum Computation and Quantum Information. Cambridge University Press, 2010

2010
[18]

Tsplib—a traveling salesman problem library, 1991

Gerhard Reinelt. Tsplib—a traveling salesman problem library, 1991. Online resource

1991
[19]

Openstreetmap, 2004

OpenStreetMap contributors. Openstreetmap, 2004. Online resource

2004
[20]

Society for Industrial and Applied Mathematics, 2002

Paolo Toth and Daniele Vigo, editors.The Vehicle Routing Problem. Society for Industrial and Applied Mathematics, 2002

2002
[21]

and Mario P

Scott Kirkpatrick, Daniel Gelatt, C. and Mario P. Vecchi. Optimization by simulated annealing.Science, 220(4598):671–680, 1983

1983
[22]

Goldberg, David˙Genetic algorithms in search, optimization, and machine learning

E. Goldberg, David˙Genetic algorithms in search, optimization, and machine learning. 1989

1989
[23]

DEAP: Evolutionary algorithms made easy.Journal of Machine Learning Research, 13:2171–2175, 2012

Francois-Alexandre Fortin, Françoise-Marie De Rainville, Marc-Alexandre Gardner, Marc Parizeau, and Christian Gagné. DEAP: Evolutionary algorithms made easy.Journal of Machine Learning Research, 13:2171–2175, 2012

2012
[24]

Simanneal: A python library for simulated annealing, 2015

Alan Weiss. Simanneal: A python library for simulated annealing, 2015. GitHub repository

2015
[25]

Ising formulations of many np problems.Frontiers in Physics, 2:5, 2014

Andrew Lucas. Ising formulations of many np problems.Frontiers in Physics, 2:5, 2014

2014
[26]

Nielsen, Michael and L

A. Nielsen, Michael and L. Chuang, Isaac˙Quantum Computation and Quantum Information: 10th Anniversary Edition. Cambridge University Press, 2010. 7

2010
[27]

A direct search optimization method that models the objective and constraint functions by linear interpolation.Advances in Optimization and Numerical Analysis, pages 51–67, 1994

Michael Powell. A direct search optimization method that models the objective and constraint functions by linear interpolation.Advances in Optimization and Numerical Analysis, pages 51–67, 1994

1994
[28]

Spall, James˙Implementation of the simultaneous perturbation algorithm for stochastic optimization.IEEE Transactions on Aerospace and Electronic Systems, 34(3):817–823, 1998

C. Spall, James˙Implementation of the simultaneous perturbation algorithm for stochastic optimization.IEEE Transactions on Aerospace and Electronic Systems, 34(3):817–823, 1998

1998
[29]

Qiskit: An open-source framework for quantum computing, 2021

Qiskit Development Team. Qiskit: An open-source framework for quantum computing, 2021. Software

2021
[30]

Aspen-m-3 quantum processor, 2024

Rigetti Computing. Aspen-m-3 quantum processor, 2024. Hardware documentation

2024
[31]

Individual comparisons by ranking methods.Biometrics Bulletin, 1(6):80–83, 1945

Frank Wilcoxon. Individual comparisons by ranking methods.Biometrics Bulletin, 1(6):80–83, 1945

1945
[32]

Nicholas Moll, Pavlos Barkoutsos, Miles Reed, Patrick Vernaz-Gris, Pierre Ollitrault, C. McKay, David Frank Enderud, Eirik Haug, Kahshali Mitra, Daniel Lubinski, and Giacomo Guerreschi, Gian˙Quantum optimization using variational algorithms on near-term quantum devices.Quantum Science and Technology, 3(3):030503, 2018

2018
[33]

A Quantum Approximate Optimization Algorithm

Edward Farhi, Joel Goldstone, and Sam Gutmann. A quantum approximate optimization algorithm.arXiv preprint arXiv:1411.4028, 2014

work page internal anchor Pith review Pith/arXiv arXiv 2014
[34]

A practical method for finding optimal parameter settings in a quantum approximate optimization algorithm.NPJ Quantum Information, 2:16003, 2016

Marcello Benedetti, Diego Garcia-Pintos, Alejandro Perdomo-Ortiz, and Naoki Yamamoto. A practical method for finding optimal parameter settings in a quantum approximate optimization algorithm.NPJ Quantum Information, 2:16003, 2016

2016
[35]

Farhi and A

Edward Farhi, Joel Goldstone, and Sam Gutmann. Quantum approximate optimization algorithm for maxcut: Performance and analysis.arXiv preprint arXiv:1602.07674, 2016

work page arXiv 2016
[36]

Gambardella, Luca˙Ant algorithms for discrete optimization.Artificial Life, 5(2):137–172, 1999

Marco Dorigo, Gianni Di Caro, and M. Gambardella, Luca˙Ant algorithms for discrete optimization.Artificial Life, 5(2):137–172, 1999. 8

1999