Where mathematics becomes something you can play with.
Play online · Learn the math · Print a puzzle · Explore the solvers
No mathematics needed to play: open the puzzle in your browser, drag the pieces into the tray, and see if you can make them fit.
![]() |
![]() |
| Hill 16b-r, printed and partly assembled | Hill 9b in the browser, freshly scrambled |
MaxPlusLab is an open collaboration between mathematician Madeline V. Brandt and scientist and engineer Igor Kuvychko. We turn ideas from tropical geometry into interactive puzzles, 3D-printed objects, and computational experiments.
Our projects form a feedback loop: mathematics generates puzzles; people and algorithms solve them; what we learn helps us design new ones.
A tropical polynomial or rational function divides a plane into regions. We turn those regions into pieces of a geometric tiling puzzle. In flat mode, the challenge is to fit the outlines together. In relief mode, each piece has a sloping top, and the completed puzzle forms a continuous surface.
The tropical puzzles come in four families: Bowl and Hill (tropical polynomials), and Terrain and Corrugated (tropical rational functions). Alongside them, Pleat is a related experiment that is not tropical: a triangulated square lifted so that its corners stand at only two heights, which makes many edges look alike. Every puzzle can be played in the browser in both modes. You can also generate STL files, 3D-print the pieces and tray, and explore the mathematics behind their construction. Each puzzle card shows an experimental difficulty rating based on how hard a computer search has to work, not on how hard people find it.
Play the puzzles · How the mathematics works · 3D printing guide · Printables collection · Code and puzzle definitions
How do you solve a geometric puzzle without trying every possible position and rotation? And can computational search reveal what makes one puzzle more difficult than another?
Our tiling-solvers project combines a Rust geometry and search engine, a Python research harness, an OR-Tools CP-SAT formulation, reproducible benchmarks, and an independent solution validator. It lets us compare algorithms, geometric constraints, and assumptions on the same puzzles.
Solver overview and findings · Algorithms · Experiments · Source code
A little height information can change the entire search. For the 16-piece Hill 16a, a complete footprint search explored 1,064,878 nodes. With the pieces' sloped roofs included, the same search explored just 68. The roof profiles sharply restrict which edges can meet.
Structure can matter even more than an improved algorithm. Tropical subdivisions have an edge-to-edge structure. Restricting the same flat puzzle to edge-to-edge assemblies reduced its search to 49 nodes. This is an explicit restriction on the allowed assemblies, not a general-purpose shortcut: flat pieces can sometimes form other, T-junction assemblies.
Difficulty is not simply piece count. The 11-piece Terrain 11a has 101 distinct footprint assemblies in our search but only one with its roofs. Roofs can also be made to help less: the 32-piece Pleat 32a, whose corners stand at just two heights, takes 21,973 nodes with its roofs on, compared with 1,529 for the 27-piece Corrugated 27a. Pleat 32a is hard for this search, not for every method: a CP-SAT solver settles it without a single conflict.
Corrugated 27a-r at a dead end: an experiment in designing harder puzzles, not just solving existing ones.
These figures count nodes in a particular solver's search, not execution time or human difficulty. Results are computational evidence under stated tolerances, not mathematical proofs of uniqueness.
Can measurements from search help us design better difficulty ratings? Which geometric features make a puzzle hard for people, rather than just for an algorithm? Can we use the mathematics of tropical constructions to create new kinds of challenging, printable puzzles?
Next on our list: more puzzles and parametric families, touch input for phones and tablets, and further studies of what makes a puzzle difficult.
You don't need to know tropical geometry to participate. Play a puzzle, print one, or tell us what made it easy or difficult. If you're interested in mathematics, geometry, algorithms, or making things, explore the repositories. Feedback, ideas and questions are welcome as issues:
Tropical Puzzle Lab issues · Tiling Solvers issues
Both projects are open source and developed in public.


