Skip to main content
Runs in this browser

Settle a problem to its lowest energy, then check the answer.

Write a problem as a score where lower is better, settle it to a low point in this tab, and check that point against every possible answer.

Some problems can be written as an energy: a score for every way of arranging the answer, where lower is better. The settling model finds a low point the way a ball comes to rest in a valley: each variable moves to the value its neighbours pull it towards, until nothing moves.

This page runs a classical emulator of that model on your computer, then checks the answer against every possible arrangement where the problem is small enough. Every answer says what it is, in these words:

A classical emulator of the settling model of a planned energy-based processor, running on your machine. Energies are computed exactly. The joule line is this run's operation count priced at rates measured on an FPGA, with a published figure for one integer add beside it for scale. No photonic hardware, custom chip or quantum effect is involved in this run.

The emulator is the browser build of open-standards commit 003be45e37b6. Before it runs, this page checks the engine file's SHA-256 against that build's. Nothing you run here is sent anywhere.

Press one button.

Twelve talks, two rooms. Twelve talks go into room A or room B, or are left out. Talks that share an audience must not be in the same room. Talks in a series belong in the same room. Every talk that runs takes a slot, so a talk with little reason to run stays out. There are 531,441 possible arrangements. The emulator settles into a low point, then checks all of them to see whether that low point is the lowest.

What the words mean
Energy
A score the program gives every arrangement. Lower is better. It is a whole number, computed exactly, so anyone can work it out again from the program.
Arrangement
One value for every variable: +, - or 0. Twelve variables have 3 to the 12th power of them.
Every arrangement checked
Up to 16 variables, the page works out the energy of every arrangement, so it knows the lowest one and can say whether the run found it.
Accumulate
Adding one link's pull to one variable's total. It is the unit of work, and the joule line prices it.
Priced joules
The run's count of accumulates multiplied by a rate per accumulate. It is a price for the work, not a reading from a meter: nothing measures this run.
Temperature
How much randomness the settling allows. At 0 it is deterministic and can get stuck; above 0 it can climb out of a poor low point.
Replay digest
A short code for the whole run. The same program, seed and temperature give the same code on any machine, so anyone can run it again and compare.

Worked examples, including the ones that fail.

Each example says what it shows. Those marked “Fails, on purpose” show what this model cannot do, and they fail here. Pick one, change it if you like, and press Run.

Twelve talks go into room A or room B, or are left out. Talks that share an audience must not be in the same room. Talks in a series belong in the same room. Every talk that runs takes a slot, so a talk with little reason to run stays out.

Sweep the temperature.

The program in the editor, twenty times at each of several temperatures, with the same budget of work at each.

The same program on classical solvers.

Simulated annealing, tabu search and greedy descent, on the program in the editor, twenty runs each, each run given the same number of operations as the emulator's.

How each one scales: the published table.

Four families of problem, from 8 to 128 variables, 100 runs of each solver on each, every run in a row given the same budget of operations. The table is published as it came out, including where the classical solvers lead.

Against the best of the three classical solvers, the emulator's default start reached the target more often in 2 of 68 rows, as often in 45, and less often in 21. Simulated annealing reached it more often than the default start in 20 rows, and the default start more often than simulated annealing in 2. Started hot, at the largest row of couplings, the emulator was ahead in 0, level in 38 and behind in 30.

  • clusters: sparse signed clustering, about degree 4, one link in eight contradicting the rest, cost 1.
  • sparse-glass: random signs on a sparse graph, about degree 6, cost 1.
  • dense-glass: random signs on every pair, cost 0.
  • selection: a hidden group of a quarter of the variables, all linked by sign, cost 1, with a known answer.
Runs that reached the target, of 100, at 40 sweeps a start; the most in each row in bold
Family Variables Target Operations a run Emulator, default startEmulator, hot startEmulator, deterministicSimulated annealingTabu searchGreedy descent
clusters 8 exact 12,800 10010060100100100
clusters 12 exact 19,200 1009755100100100
clusters 16 exact 25,600 100974100100100
clusters 24 best known 38,400 100942010010099
clusters 32 best known 51,200 9773199910082
clusters 48 best known 76,800 96460989062
clusters 64 best known 102,400 94350976022
clusters 96 best known 153,600 564086183
clusters 128 best known 204,800 46007970
sparse-glass 8 exact 17,920 10010021100100100
sparse-glass 12 exact 26,880 10010027100100100
sparse-glass 16 exact 35,840 1009027100100100
sparse-glass 24 best known 53,760 1006801009790
sparse-glass 32 best known 71,680 100743999896
sparse-glass 48 best known 107,520 77282825838
sparse-glass 64 best known 143,360 6820902812
sparse-glass 96 best known 215,040 37007641
sparse-glass 128 best known 286,720 36007150
dense-glass 8 exact 20,480 10010088100100100
dense-glass 12 exact 46,080 10010078100100100
dense-glass 16 exact 81,920 10010039100100100
dense-glass 24 best known 184,320 1009525100100100
dense-glass 32 best known 327,680 945349995100
dense-glass 48 best known 737,280 9422310092100
dense-glass 64 best known 1,310,720 8140975491
selection 8 exact 8,960 100100100100100100
selection 12 exact 10,240 10010099100100100
selection 16 exact 11,520 100100100100100100
selection 24 planted 17,280 100100100100100100
selection 32 planted 28,160 100100100100100100
selection 48 planted 57,600 100100100100100100
selection 64 planted 97,280 100100100100100100
selection 96 planted 207,360 100100100100100100
selection 128 planted 358,400 100100100100100100
The same table at 320 sweeps a start
Runs that reached the target, of 100, at 320 sweeps a start; the most in each row in bold
Family Variables Target Operations a run Emulator, default startEmulator, hot startEmulator, deterministicSimulated annealingTabu searchGreedy descent
clusters 8 exact 102,400 10010060100100100
clusters 12 exact 153,600 10010055100100100
clusters 16 exact 204,800 1001004100100100
clusters 24 best known 307,200 10010020100100100
clusters 32 best known 409,600 1009619100100100
clusters 48 best known 614,400 9910009910098
clusters 64 best known 819,200 100930999440
clusters 96 best known 1,228,800 9880099688
clusters 128 best known 1,638,400 9167094260
sparse-glass 8 exact 143,360 10010021100100100
sparse-glass 12 exact 215,040 10010027100100100
sparse-glass 16 exact 286,720 10010027100100100
sparse-glass 24 best known 430,080 1001000100100100
sparse-glass 32 best known 573,440 1001003100100100
sparse-glass 48 best known 860,160 85832877991
sparse-glass 64 best known 1,146,880 948701006542
sparse-glass 96 best known 1,720,320 876909397
sparse-glass 128 best known 2,293,760 7045084140
dense-glass 8 exact 163,840 10010088100100100
dense-glass 12 exact 368,640 10010078100100100
dense-glass 16 exact 655,360 10010039100100100
dense-glass 24 best known 1,474,560 10010025100100100
dense-glass 32 best known 2,621,440 98944100100100
dense-glass 48 best known 5,898,240 100963100100100
dense-glass 64 best known 10,485,760 9690010087100
selection 8 exact 71,680 100100100100100100
selection 12 exact 81,920 10010099100100100
selection 16 exact 92,160 100100100100100100
selection 24 planted 138,240 100100100100100100
selection 32 planted 225,280 100100100100100100
selection 48 planted 460,800 100100100100100100
selection 64 planted 778,240 100100100100100100
selection 96 planted 1,658,880 100100100100100100
selection 128 planted 2,867,200 100100100100100100

The target is the exact lowest energy where every arrangement can be checked, a planted answer the problem was built around, or else the best energy long runs of simulated annealing and tabu search found, which proves nothing about the lowest. One operation is one coupling read, one cell updated or one move considered, whoever does it. The count says nothing of what an operation costs on any device, or of cells that settle side by side. The table is table.json at commit 003be45e37b6, rebuilt by wai-quantum landscape table.

Run it on your own machine.

The command line at the same commit gives the same answers, byte for byte. With Rust installed through rustup, this runs the one-button problem; its replay digest is the one this page shows for seed 1.

git clone https://git.transaction.science/Transaction-Science/open-standards
cd open-standards
git checkout 003be45e37b6f35f5bcd3e795f66cd3909087bae
cd wai/wai-quantum-cli
cargo run --release --bin wai-quantum -- landscape run --example talks --seed 1

wai-quantum landscape examples lists the worked examples, sweep and compare do what the buttons above do, and replay runs many seeds, one line each. The specification is the extension text, and the results, with the script that reproduces them, are in quantum-conformance/landscape. This page's engine was built with rustc 1.100.0 and wasm-pack 0.15.0; its file's SHA-256 is 237cf854af7f7d3d77cbb7adb5bc865c41f9aa622b56022c4a9dcc7b4ca9c970.

Where to go from here