Quantum Computing For Logistics At IEEE Quantum Week 2026 – Quantum Zeitgeist

This guide follows nine papers and one panel on quantum computing for logistics and planning from IEEE Quantum Week 2026 in Toronto. Teams cut real jobs down until a quantum step could cope, and then said plainly where classical solvers still won.
In 1950 the Canadian National Exhibition in Toronto put on show Bertie the Brain, a machine that played tic-tac-toe, which Vikram Mulligan at QCE26 credited to Josef Kates. Seventy-six years later, in the same city, Mulligan had a D-Wave quantum annealer play the same game. Moving first, it drew 28 of 30 games against a perfect classical player. Tic-tac-toe is a fair warm-up for quantum computing for logistics. Its board is small and its rules are few.
The planning jobs that railways, water boards and databases face are the same kind of puzzle on a far larger board. Each is a long list of yes-or-no choices. A railway picks which train set runs which cycle of trips, and a water board picks which rivers get protection. Every choice added doubles the number of possible plans, so fifty choices give about a million billion, far too many to check one by one. That is why quantum computing for logistics draws so much hope, and why classical solvers that have been tuned for decades set such a high bar.
Two quantum methods compete for the work. An annealer, the kind of machine D-Wave builds, writes the problem into the energy of linked qubits and lets them settle toward a low-energy state that stands for a good plan. A gate-based machine runs QAOA instead, a quantum search method in which interference gathers weight on the better answers while a classical computer tunes the settings. Both want the problem as a QUBO, a score over yes-or-no choices. Real rules, such as a train’s maintenance visits or a seat that holds one guest, fit that form badly.
Taken together, the ten talks in Toronto made a case that runs against the hope. The more carefully a team fitted a real job to a quantum optimiser, the smaller the quantum part became, until the classical code around it was doing most of the planning.
The strongest case against that reading came on the Monday morning. Ibrahim Sabek of the University of Southern California was called by the chair the first person to attach a quantum computer to the Postgres database. Sabek reported gains of up to three times over the best classical and learned query planners. Immanuel Trummer of Cornell added that database researchers already write problems close to a QUBO. That gain is Sabek’s own claim, and it came from sending only the hardest fragment of the planning to the annealer, with classical code keeping the rest.
The rail and routing papers that afternoon followed the same pattern. Jiri Guth Jarkovsky of IQM, working with Deutsche Bahn, gave QAOA small patches of train cycles, and the team’s exact classical solver beat every quantum run. Camille de Valk of Universiteit Leiden and Capgemini gave the quantum step only the choice of which cuts to keep in a delivery routing problem. Done classically, that choice took under 1 percent of the runtime. De Valk called that placement of the quantum step a negative result.
Two Tuesday truck papers pushed back in part. IonQ reported that its warm starts always improved Einride’s own solver, and a D-Wave annealer beat simulated annealing on truck-and-drone routes, though its presenter said the split did most of the work.
Yet the most telling thing about these talks was their candour. Candour turns a failed pilot into evidence that others can use. Federico Spedalieri of Capital One, speaking personally, said years of tests on an annealer had produced one scaling win on a contrived problem. Mulligan’s annealer, which had drawn 28 of 30 games against perfect play when it moved first, could not keep wedding guests off shared seats from five guests upward, and released the seating problems as a public benchmark. A field that publishes its failures this plainly is drawing a map of where quantum optimisers break.
What it is. Railways, water boards, databases and builders face planning jobs made of many yes-or-no choices. Quantum annealers and QAOA promise better plans, and at QCE26 teams tested them on real cases.
The best claim. A database query planner that sends its hardest step to an annealer reported gains of up to three times over the best classical planners. Much of that gain was lost on the hop to the cloud.
The pattern. On trains, vehicle routes, cables and aquifer sites, classical solvers matched or beat the quantum runs. Splitting a job to fit the machine shrank the quantum share.
The warning. A D-Wave annealer that drew 28 of 30 tic-tac-toe games against perfect play kept putting several wedding guests on one seat from five guests upward. Mulligan blamed the dense links the seating rule demands.
Database systems solve hard planning jobs all day, such as choosing the order in which to join tables. Quantum optimisers want exactly such problems, so a Monday panel asked why the two fields still treat each other as niches.
Wolfgang Mauerer of OTH Regensburg chaired the panel. On stage were Hans-Arno Jacobsen of Toronto, Ibrahim Sabek of USC, Immanuel Trummer of Cornell and Federico Spedalieri of Capital One. A fifth, Valter Uotila of TU Delft, was listed but did not take part. Each of the four had a view on what the two fields could do for each other, and not all of them were hopeful.
Sabek sends the hard fragment of query planning to an annealer while a classical program keeps the rest. On a standard join-order benchmark Sabek reported an end-to-end gain of up to three times over the best classical and learned planners, though much of the benefit was lost on the trip to the cloud. Trummer told database researchers to start with annealing, since a QUBO is close to the models they already send to solvers such as CPLEX and Gurobi. Spedalieri was less hopeful, and said the annealer returns a decent answer at once and is then caught by classical solvers.
Jacobsen described a Canadian effort to pool quantum machines, along with Q-DICE, an open-source compiler that maps split circuits back onto one chip, and said a simulated anneal had matched the quantum hardware on the jobs the group tried. Trummer said bluntly that filters and joins make a poor target. Their pain starts only at billions of rows. A poll of the room ranked distributed quantum computing as the main place databases could help quantum machines in return.
Two Monday afternoon papers took on classic logistics jobs, and two Tuesday papers followed with trucks. The Monday pair split the work so that the quantum step faced only a small piece, and both were frank about the result.
Jiri Guth Jarkovsky of IQM presented work with Patricia Bickert, Elisabeth Wybo and Martin Leib. The authors come from IQM and Deutsche Bahn. The job was rolling stock planning, which assigns trains to a timetable so that every trip is covered, each train repeats a cycle, and every train reaches Hamburg for maintenance every 4,000 kilometres. The test case was 190 real trips among the five largest German cities.
Each possible cycle becomes a node, and two cycles are joined if they share a trip. The job is to find a heavy set of cycles with no two joined. In general the task is NP-hard. Even a two-day window gives tens of thousands of nodes, so Jarkovsky took the 20 heaviest, solved that patch, kept the chosen cycles and moved on to the next. Each patch went to an exact classical solver, a random guess and QAOA, run both in noiseless simulation and on IQM’s Emerald machine.
The exact classical solver won. Over about 50 seeds it gave the fewest trains and empty kilometres. The quantum runs landed level with a simple greedy sweep, a method that just grabs the heaviest cycle each time, and Jarkovsky suspected the small size of each patch was the reason. Larger patches, tried only with the exact solver, cut the empty distance further. For more on the company, see our guide to IQM.
Camille de Valk of Universiteit Leiden and Capgemini presented with Koen Reerink, Siert Sebus and Sébastian de Bone. De Valk contrasted two ways to look for a quantum use, starting from a famous algorithm and hoping for a problem, or starting from a real job and accepting that a classical solver may still win. Their job was vehicle routing, with depots, customers and vans of limited load.
The team’s pipeline split the problem in two. A master problem was solved first, and many smaller problems, which hundreds of cores can run side by side, each offered one cut to tighten it. De Valk said 10,000 or more candidate cuts fell to a master of about 800, and only the choice of which cuts to keep went to the quantum side, run mostly on emulators, with one step on IBM hardware.
A full toy route on the emulator took about 3,000 seconds, nearly all of it in choosing cuts, while a classical timing put that same step under 1 percent of the runtime. The toy case solves on a laptop. De Valk called the placement of their quantum step a negative result.
Miguel Angel Lopez-Ruiz, a staff scientist at IonQ, presented the paper with ten co-authors, from Daiwei Zhu to Martin Roetteler. Einride is a Swedish operator of electric truck fleets. When a shipment is cancelled a gap opens in the weekly schedule, and the job is to pick one sequence of shipments, or none, to fill each gap. Patching gaps one at a time, Lopez-Ruiz argued, misses how choices in different trucks affect each other.
The team wrote the job as a QUBO. It solved it with a form of QAOA that trains nothing, whose settings follow a fixed ramp set by one slope, while the best answers of each run seed the next. Einride supplied eight real weekly schedules, giving 32 cases of up to 130 qubits. On IonQ’s Forte machine, cases of 20 to 35 qubits ran circuits of up to 819 two-qubit gates and, after light classical clean-up, closely matched noiseless simulation.
Used as warm starts for Einride’s own planning solver, Lopez-Ruiz reported, the quantum answers always beat a cold start on its working measures. The classical solver SCIP did not always finish. It timed out after 48 hours on cases of up to 220 variables with strong links between pairs, and where it did finish, the refined quantum results almost all matched its answers.
Chia-Chen Yang of National Pingtung University presented with Chia-Ho Ou, Masayuki Ohzeki and Chih-Yu Chen. A truck carries drones for the last mile of delivery, and the team counts a drone as four times cleaner over the same distance. Written whole, the job is too big. It needs 795 physical qubits at five customers and passes the machine’s 5,627 at 12, and at eight customers only 30 percent of the annealer’s answers were valid.
So the team splits it. A greedy truck tour with drones added comes first, and then a window of the eight customers with the most to gain goes to the annealer as a small truck problem and a small drone problem. Reverse annealing starts from the current answer and searches nearby. From 5 to 35 customers it found valid plans in 30 to 128 seconds, with 2 to 8 percent less carbon than the greedy start, while Gurobi timed out beyond 30 customers.
With everything else fixed, the annealer beat simulated annealing at all seven sizes and a simple local search at six. Yang called the extra gain modest and real, said the split does most of the work, and noted that every test used made-up grids, with real road networks left for later.
Two Monday papers turned planning toward the climate. Both used QAOA with care, and both hit a size limit.
Lea Casse of Institut Polytechnique de Paris presented work from the TAIAO project, which links their school with two universities and New Zealand’s weather service, MetService. The problem is a water authority that can protect only some of its rivers. Seven New Zealand sites were the test. The first stage forecasts floods, feeding rainfall and river levels into a single qubit through a set of repeated turns.
A classical network with the same 72 settings served as the comparison. Casse said the quantum forecast tracked the peaks, including one flood the classical network missed. The second stage used QAOA to choose which sites to protect, aiming to cut the average loss in the worst cases, with a starting state and mixer that only allow plans of the right size.
Success fell as sites were added. From seven to 16 sites the rate dropped, though it stayed above what a random draw would give, and a one-qubit forecast run on real hardware sat close to the simulator’s answer. Asked for processor time, Casse had no figure.
Samar Aseeri of KAUST presented this paper, which won a best-paper award in the QECS track, and Julián Stiefkens spoke on the runs. Co-authors are Jeff Burgett, Tom Taylor and Adam Milewski. Managed aquifer recharge means choosing where to send water back underground. Twenty candidate sites give about a million choices, and fifty give a million billion, which Aseeri put at about 19.4 years of brute-force search.
Five open US data sets fed a QUBO that QAOA searched. A random start fell into a barren plateau, a flat region where the optimiser finds no direction, so the team solved the shallow circuit first and passed its angles to the deeper one. With that warm start the search matched the exhaustive answer at every depth tried on the real costs.
Angles learned at 20 sites carried over to 50 on a laptop. On IBM’s Kingston machine the 20-site search scored 0.994 on a flat cost and 0.972 on the real costs, which Aseeri called within 3 percent of the best plan. At 50 sites the circuit grew from 854 gates to 26,911, about 31 times bigger. The hardware result collapsed. Aseeri called it a coherence wall. Aseeri said simulated annealing and a genetic algorithm already match the quantum search at 20 sites.
Thursday afternoon closed the strand with three papers, two of them from one speaker. They gave the week’s clearest picture of where today’s annealers work and where they break.
Vikram Mulligan of the Flatiron Institute gave this paper with Tristan Zaborniak of the University of Victoria. Games, Mulligan said, have long pushed hardware forward. So the team let a D-Wave Advantage 2 annealer play out the rest of the game from the current board many times over, and then took the move that had won most often in all of those games.
The board used nine groups of nine qubits, one group per move and one qubit per square, with links that forbid two marks in one move or on one square. Extra qubits light up when three in a row appears. By biasing them, the machine can sample only wins for X, only wins for O or only draws.
Against a random player the annealer won all 30 games it started and 26 of 30 that the opponent started. Against a perfect player it drew 28 of 30 when it moved first and lost all 30 when it moved second, because it always plays for a win and never settles for a draw. Its odds for each opening tracked an exact count over roughly 250,000 possible games.
Paul-Niklas Kandora of the Karlsruhe Institute of Technology presented work with three KIT colleagues and two co-authors from VINCI Energies. Cables in buildings run along trays. High and low voltage cables may not share a segment except at their ends, and the team wrote the job with one yes-or-no choice for each cable on each segment.
They compared two ways of writing the score, solved all at once or one cable at a time, with QAOA and with VQE. On a room with 6 nodes, 7 segments and 2 cables, the global QUBO needed 22 qubits and the split version only 7. All were noiseless simulations with 15 seeds each.
Every split version found valid routes in all 15 seeds. But those routes cost 35 to 37 percent more than the Gurobi optimum, while one global version came within 2.4 percent. Kandora claimed no speedup. A classical solver handles the whole room in milliseconds, Kandora noted, which leaves a quantum step very little to win. On a reduced real layout D-Wave’s hybrid solver found valid routes in up to 49.6 percent of runs, though far from the best route.
Mulligan returned for a paper written with Karie Nicholas, their wife, who did not attend. Mulligan linked wedding seating to their day job in protein design, where each position on a protein needs one choice of shape. The open Masala libraries, which their group’s Rosetta software can call, send such problems to classical or quantum solvers, and in a design test the annealer’s proteins were predicted to fold in 4 percent of 250 cases.
To test that the code was general, Mulligan wrote seating plugins with guests in place of positions and seats in place of shapes, plus a rule that no two guests share a seat. The cases ran from four friends and enemies at one table to the couple’s own wedding. One rule kept a guest who could not take a booster shot away from vulnerable relatives.
The classical solver did so well that the couple used its chart on the day. The annealer could not seat the guests at all. From five guests upward it kept putting several guests on one seat, and changing the penalty did not help, which Mulligan blamed on the dense web of links that the one-seat rule demands of the qubits. The seating problems are now a public benchmark.
Planners should keep the classical solver. A rail planner, a utility or a database team weighing a quantum pilot should run any quantum step beside a classical one on the same case. Look first at how densely the rules tie the choices together, since that is what Mulligan blamed for the wedding chart, and check how fast the circuit grows, since that is what stopped the aquifer search.
Publishing the cases matters as much. Mulligan did, so others can check the result. The pilots worth paying for are the ones designed to find the limit, because Toronto showed that the limit arrives early and that much of it sits in the rules. Bertie the Brain only ever had nine squares to fill.
Quantum computing for logistics means using quantum annealers, QAOA and related methods on planning jobs such as train scheduling, delivery routes and cable layouts. These jobs are long lists of yes-or-no choices whose combinations grow very fast. The hope is better plans, or good plans sooner, than classical solvers find.
Mostly not at QCE26. On rolling stock, vehicle routing and cable routing, classical solvers matched or beat the quantum runs, and several speakers said so plainly. One database query planner reported gains of up to three times, with the annealer handling only the hardest fragment of the work, and IonQ said quantum warm starts always improved a truck operator’s own solver.
An annealer, such as D-Wave’s, writes a problem into the energy of linked qubits and lets them settle toward a low-energy state that stands for a good answer. QAOA runs on a gate-based machine and alternates circuit layers that favour good answers with layers that mix them, while a classical computer tunes the settings.
Both annealers and QAOA want problems as a QUBO, a score over yes-or-no choices that counts only single choices and pairs. Rules such as one guest per seat become penalties that link nearly every choice to every other. At QCE26 Vikram Mulligan blamed that dense web of links for a D-Wave annealer that kept seating several wedding guests on one seat.
Yes. At QCE26 Vikram Mulligan of the Flatiron Institute showed a D-Wave Advantage 2 annealer that samples whole future games to pick its moves. Moving first it drew 28 of 30 games against a perfect player, and moving second it lost all 30, because it always plays for a win.
IEEE Quantum Week 2026, also called QCE26, ran from 13 to 18 September at the Metro Toronto Convention Centre. This guide covers nine of its papers and one panel on planning problems, from railways and delivery routes to floods, aquifers, cables and games.
This guide is based on nine papers and one panel at IEEE Quantum Week 2026 (QCE26), held at the Metro Toronto Convention Centre from 13 to 18 September 2026, and on the published conference programme. Names follow the programme where it lists them, and affiliations not listed there are as speakers gave them. Results are stated as each speaker presented them, and most come from simulation or small test cases. Company claims are the companies’ own.


See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
Ivy Delaney has been working with neural networks and machine learning since the mid-nineties, back when a couple of hidden layers and a long afternoon of training counted as ambitious. She has watched the field go from academic curiosity to the thing quietly running underneath everything, and she brings that long view to quantum computing. For Quantum Zeitgeist she covers the ground where the two fields meet. That means quantum machine learning and the variational algorithms it leans on, and it also means the less glamorous but more interesting story of classical machine learning already doing real work inside quantum machines, decoding error-correcting codes, calibrating noisy hardware and learning the error models that simulators depend on. She writes about the hardware those algorithms have to run on too, and about the post-quantum cryptography scramble that the same hardware has set off. Her stories typically start with the paper, whether that is peer-reviewed work, conference proceedings or an arXiv preprint, with the source linked so you can hold a claim up against the research it came from. She is unimpressed by benchmarks that will not say what they beat, and by demonstrations that only work in the press release.
Advertisement
Quantum computing’s breakthroughs, deals, and research, delivered straight to your inbox.
Read by quantum researchers, engineers, developers, and business professionals worldwide.
The Quantum Ecosystem
Explore the Quantum Navigator
A living directory of the global quantum industry, plus public quantum stocks and the investors funding them.
Counts as of October 2026
Disclaimer: All material, including information from or attributed to Quantum Zeitgeist or individual authors of content on this website, has been obtained from sources believed to be accurate as of the date of publication. However, Quantum Zeitgeist makes no warranty of the accuracy or completeness of the information and Quantum Zeitgeist does not assume any responsibility for its accuracy, efficacy, or use. Any information on the website obtained by Quantum Zeitgeist from third parties has not been reviewed for accuracy.
Copyright 2019 to 2026 The Quantum Zeitgeist website is owned and operated by Hadamard LLC, a Wyoming limited liability company.
To provide the best experiences, we and our partners use technologies like cookies to store and/or access device information. Consenting to these technologies will allow us and our partners to process personal data such as browsing behavior or unique IDs on this site and show (non-) personalized ads. Not consenting or withdrawing consent, may adversely affect certain features and functions.
Click below to consent to the above or make granular choices. Your choices will be applied to this site only. You can change your settings at any time, including withdrawing your consent, by using the toggles on the Cookie Policy, or by clicking on the manage consent button at the bottom of the screen.

source
This is a newsfeed from leading technology publications. No additional editorial review has been performed before posting.