Fragment Shader as Observation Apparatus

Key Result

Abstract

We prove from first principles that a GPU fragment shader is not a visualization tool but a physical observation apparatus: when it writes a pixel value it performs a measurement, and the rendered texture is the computed result in categorical representation, not a picture of it. The argument rests on three pillars. First, the Oscillatory Necessity Theorem: every persistent dynamical system in bounded phase space necessarily oscillates. Second, the Triple Equivalence Theorem: oscillatory, categorical, and partitional descriptions yield identical state counts and entropies, connected by explicit bijective maps. Third, the Rendering-Measurement Identity: for a fragment shader implementing partition observation, rendering a texture is identical to measuring the categorical state. From these we derive: an O(1) GPU-memory streaming protocol reducing database search to ~13 MB working set independent of database size; a GPU-supervised training framework where physical observables serve as training signals without human labels; and proof that integrated GPUs with ~25 MB working set are sufficient for the complete pipeline.

Key Theorems

• 1Rendering-Measurement Identity: rendering a texture IS performing a measurement — mathematical identity, not analogy
• 2O(1) Memory Theorem: streaming observation with constant ~13 MB working set regardless of database size
• 3GPU-Supervised Training: partition sharpness, phase coherence, interference visibility as label-free training signals
• 4Integrated GPU Sufficiency: Intel UHD / AMD Radeon / Apple M-series at ~1-2 TFLOPS sufficient for complete pipeline

Validation Results

Figure Panels