Constraint Satisfaction Problems (CSPs) are mathematical frameworks defined by a set of objects whose state must satisfy a number of strict limitations. Rather than finding a "best" solution, CSPs search for a state that exists within the intersection of all rules.
Every CSP consists of three fundamental elements. Understanding these is the key to seeing why certain problems are hard — and how geometry can make them tractable.
Why study puzzles? Because Sudoku is the perfect crystallization of a CSP. It strips away the noise of the real world and presents the raw structure of constraint logic.
Sudoku has 81 variables (the cells), 9-element domains (digits 1-9), and 27 all-different constraints (rows, columns, boxes). This creates a constraint graph where the geometry of the problem — the curvature of the solution space — directly determines solving difficulty.
When you see a "hard" Sudoku, what you're really seeing is a region of high curvature on the constraint manifold. The Davis Law C = τ/K tells us that high curvature regions have low capacity — they're bottlenecks where progress is slow. Understanding this geometry is the key to 40,000× speedups.
CSPs generally belong to the complexity class NP-Complete (or NP-Hard for optimization variants). This means that as you increase the number of variables, the time required to solve using standard algorithms grows exponentially.
Verification is easy. Finding solutions is hard. There is no known algorithm that can solve all CSPs efficiently. This computational "hardness" forces us to develop advanced heuristics, constraint propagation techniques, and backtracking strategies.
But the Davis Manifold reveals something important: the hardness is geometric. NP-Complete problems have regions of the solution space with infinite curvature — topological barriers that standard algorithms cannot navigate. Curvature-guided search finds paths around these barriers, explaining the massive speedups we observe.
CSPs are everywhere. Any problem that involves assigning values subject to rules is a CSP in disguise.
Assigning shifts to thousands of employees while respecting labor laws, skills, and availability.
Optimizing operating room usage, nurse rostering, and bed allocation under life-critical constraints.
The VRP: delivering goods with optimal routes while adhering to capacity and time windows.
Our Sudoku solver achieved a 40,128× speedup over Yann LeCun's Kona 1.0 — not through better hardware, but through understanding the geometry of constraint satisfaction.