Introduction
In the realm of mathematics, we often distinguish between a proof that merely demonstrates a fact and one that provides a deep explanation for why that fact is true. While a brute-force calculation might convince us of a result, a elegant theorem often provides the "Aha!" moment that reveals the underlying structure. However, formally defining what constitutes a "mathematical explanation" has long been a challenge for logicians and philosophers. This paper, authored by Joseph Y. Halpern, addresses this puzzle by extending the Halpern-Pearl (HP) definition of causal explanation into the domain of necessary truths.
The Problem of Necessary Truths
Standard causal modeling relies on counterfactual reasoning: we say $A$ caused $B$ if, in a world where $A$ did not happen, $B$ would not have occurred. This "but-for" logic works well for physical events, such as a match causing a fire. However, mathematical facts—such as $2+2=4$ or the primality of a specific number—are considered necessary truths. They are true in all possible worlds.
Under traditional logical frameworks, an "ideal agent" is assumed to be logically omniscient, meaning they already know all mathematical truths. This creates a paradox: if an agent already knows everything that is necessarily true, then no mathematical fact can ever serve as an explanation because it provides no new information. To solve this, Halpern introduces a framework that accounts for the epistemic limitations of real-world agents.
Methodology: Impossible Possible Worlds
The core innovation of this paper is the introduction of "impossible" possible worlds into the causal model. To explain why a mathematical statement is true, an agent must be able to conceive of a scenario where it could have been false, even if that scenario is mathematically inconsistent.
Epistemic States and Causal Settings
Explanation is defined relative to an agent’s knowledge. The agent considers a set of causal settings (worlds). By allowing these worlds to include mathematically inconsistent states—for example, a world where a prime number is not congruent to 1 mod 4—the framework can apply the same "but-for" logic used in physical causality.
Structural Equations for Mathematics
Halpern treats mathematical properties (e.g., "is a factor," "is prime," "is a sum of squares") as endogenous variables in a causal graph. Mathematical "laws" are then modeled as structural equations. This allows us to say, for instance, "But for the fact that $x-2$ is a factor of $f(x)$, $f(2)$ would not be $0$." By bridging the gap between physical laws and mathematical theorems, the HP model provides a unified language for explanation.
Evaluating and Ranking Explanations
Not all explanations are created equal. The paper introduces several criteria for ranking the quality of an explanation:
- Subjectivity: An explanation is only useful if it targets what the agent does not already know. If an agent is already aware that a number is prime, simply restating its primality offers no explanatory value.
- Probability: Explanations can be ranked based on their prior probability in the agent's epistemic state. A "better" explanation is one that the agent finds more plausible or that covers a wider range of the agent's uncertainty.
- Sufficiency: A deep theorem is often a better explanation than a specific calculation because the theorem provides a "sufficient cause" that applies to a broad class of instances, whereas a calculation only explains a single data point.
Case Study: Fermat’s Two-Squares Theorem
The paper applies this framework to a classic example from number theory: Why is 4373 the sum of two squares?
There are two potential explanations:
- The Brute-Force Explanation: $4373 = 62^2 + 23^2$.
- The Theoretical Explanation: Fermat’s Two-Squares Theorem, which states that an odd prime $p$ is a sum of two squares if and only if $p \equiv 1 \pmod 4$. Since 4373 is prime and $4373 = 1093 \times 4 + 1$, it must be a sum of two squares.
Halpern provides a formal logical proof that while both are valid HP explanations, the theorem-based explanation is "strictly better." It remains a valid cause even in "impossible" worlds where the specific arithmetic of $62^2 + 23^2$ might have been different, thus providing a more robust account of the truth.
Limitations and Future Directions
While the framework offers a powerful new tool for formal epistemology, Halpern notes several limitations:
- Heuristic Probabilities: Assigning specific probability values to "impossible" worlds is subjective and often heuristic.
- Logical Omniscience: While the model moves away from ideal agents, it does not yet provide a general theory for bounding an agent's reasoning power.
- Computational Complexity: Applying this to massive, multi-step proofs (such as the Classification of Finite Simple Groups) remains a daunting challenge.
Ultimately, this work suggests that mathematical and scientific explanations may not be as different as previously thought, potentially unifying them under a single causal umbrella.
Summary updated