Introduction
In the mid-1980s, Qualitative Physics aimed to describe the physical world using simple signs ($+$, $-$, $0$) instead of complex differential equations. However, this approach often suffered from inherent ambiguity. For instance, in a collision between a massive truck and a small ball, standard qualitative equations cannot determine the final velocities because they involve subtracting unknown magnitudes. This paper introduces FOG (Formalisation du raisonnement sur l’Ordre de Grandeur), a system that resolves such ambiguities by formalizing "order of magnitude" reasoning—the intuitive ability of physicists to simplify systems by identifying dominant effects.
The FOG Formalism
The core of FOG is a set of qualitative operators that go beyond simple signs to describe relative scales:
- Ne (Negligible): $A$ is much smaller than $B$.
- Vo (Close to): $A$ and $B$ are approximately equal.
- Co (Comparable): $A$ and $B$ have the same sign and order of magnitude.
FOG utilizes 30 inference rules to derive new scale-based facts. Crucially, these rules are grounded in Non-Standard Analysis, where "negligible" values are mapped to infinitesimals. This provides a rigorous mathematical foundation for what was previously considered heuristic "hand-wavy" reasoning.
Resolving Physical Ambiguity: The Collision Benchmark
To demonstrate its power, the author applies FOG to a classic mechanics problem: an elastic collision between a large mass $M$ and a small mass $m$.
In standard qualitative physics, the final velocities $V_f$ and $v_f$ are ambiguous. However, by incorporating the knowledge that $m$ is negligible compared to $M$ ($m \text{ Ne } M$), FOG collapses the space of possible outcomes. It correctly infers that the large mass $M$ continues nearly unaffected ($V_f \text{ Vo } V_i$) while the small mass $m$ rebounds with a specific relative velocity.
Key Insights and Impact
FOG transforms the "quantity space" from a simple line of landmarks into a structured hierarchy. Values are grouped into equivalence classes and ordered by scale. This allows AI systems to mimic expert behavior, focusing on dominant physical processes while safely discarding secondary effects. Beyond mechanics, the system has been applied to analog circuit troubleshooting and macroeconomic modeling, proving that qualitative reasoning doesn't need exact numbers to be physically accurate—it just needs relative scales.
Limitations
While powerful, FOG depends on human experts to define initial relations (e.g., deciding if $A$ is truly negligible compared to $B$). Additionally, while the rules are mathematically sound for infinitesimals, real-world application requires careful consideration of the numerical boundaries where these qualitative assumptions remain valid.
Summary updated