Introduction
In the early 1990s, the field of machine discovery primarily focused on domains like physics and number theory. Plane geometry presented a unique challenge, requiring a bridge between spatial manipulation and formal symbolic logic. The Discovery System for Trigonometric Functions (DST) was developed to address this, aiming to "rediscover" classical theorems and trigonometric identities with minimal prior knowledge. Unlike systems that rely on vast libraries of pre-existing rules, DST autonomously explores geometric domains by modifying shapes and extracting mathematical laws through systematic experimentation.
System Architecture
The DST framework operates through a continuous feedback loop between two specialized subsystems: the Plane Geometry Analysis System (PGA) and the Mathematical Formula Transformation System (MFT).
Plane Geometry Analysis System (PGA)
The PGA acts as the "experimentalist" of the system. Starting with an arbitrary geometric figure, such as a triangle, it performs three types of actions: specialization (e.g., making it a right triangle), adding auxiliary lines (like perpendiculars from a vertex), and extracting relations. As illustrated in the following figure, the system identifies constituent lines, triangles, and angles to find congruencies, similarities, or adjacent relationships.
Mathematical Formula Transformation System (MFT)
The MFT serves as the "theoretician," processing the raw relations from the PGA through algebraic substitution and transposition. A critical technique employed is subproduct elimination. The system identifies auxiliary elements—"subproducts"—created during the construction phase (such as the intersection point of a perpendicular) and attempts to eliminate them from the final equations. A formula that remains valid after removing these temporary scaffolding variables is classified as a discovered theorem.
Key Insights and Results
DST successfully rediscovered several pillars of high school mathematics, including fundamental trigonometric identities like $\tan \theta = \frac{\sin \theta}{\cos \theta}$ and $\sin^2 \theta + \cos^2 \theta = 1$, as well as the properties of isosceles triangles.
One of the system's most significant achievements was the derivation of the Pythagorean theorem ($a^2 + b^2 = c^2$). Interestingly, DST reached this discovery purely through similarity ratios and algebraic substitution, bypassing the traditional area-based proofs often used in education. To arrive at these results, the MFT subsystem processed approximately 10,000 formulas, demonstrating the necessity of automated pruning to manage the combinatorial explosion of mathematical noise.
Conclusion and Future Work
While DST demonstrated the power of automated discovery, it faces limitations regarding "interest criteria"—the ability to distinguish which theorems are meaningful to humans. Currently, it uses expression complexity as a proxy for utility. Future work aims to extend the system's scope to circles and quadrilaterals and to improve its ability to recognize when internal "subproducts" (like a center of gravity) are themselves geometrically significant features rather than mere auxiliary scaffolding.
Summary updated