Reasoning with Orders of Magnitude and Approximate Relations

Michael L. Mavrovouniotis, George StephanopoulosAAAI 1987

Abstract

Order-of-magnitude reasoning is a fundamental tool for engineers, yet formalizing it for automated systems has historically been challenging. This paper presents O[M], a framework that defines seven primitive relations to bridge the gap between coarse qualitative signs and precise quantitative models. By employing both strict interval-based and heuristic fuzzy semantics, O[M] allows for rigorous inference using rough estimates.

Introduction

In the late 1980s, the field of Qualitative Reasoning (QR) faced a significant "resolution" problem. Traditional systems, which primarily tracked signs (+, -, 0), were often too coarse for engineering applications where the relative magnitudes of variables are critical. Conversely, full quantitative models require exact numerical data that is rarely available during the early stages of design or diagnosis.

The O[M] formalism was proposed to address this gap. Unlike previous attempts like FOG, which lacked formal semantics and bundled sign information with magnitude, O[M] provides a mathematically sound yet flexible framework for reasoning with "rough" estimates.

The O[M] Formalism

The core of the O[M] system is a set of seven primitive irreducible relations between the magnitudes of two quantities, $A$ and $B$. These relations range from "much smaller" to "much larger," allowing for a more nuanced representation than simple inequality.

Extracted figure from page 2 of Reasoning with Orders of Magnitude and Approximate Relations

These primitives can be combined into 21 compound relations, such as "less than or approximately equal to," providing a comprehensive vocabulary for approximate comparison.

Semantics and Interpretation

O[M] operates using two distinct types of semantics to balance mathematical rigor with human-like intuition:

  1. Strict Interpretation: Relations are mapped to non-overlapping intervals of the ratio $A/B$. This ensures certain inferences but can be overly rigid.
  2. Heuristic Interpretation: This mode allows for "fuzzy" boundaries between relations. It enables more aggressive inferences, mimicking how a human might conclude that if $A$ is slightly larger than $B$, and $B$ is slightly larger than $C$, then $A$ is likely moderately larger than $C$.
Fuzzy interval boundaries used for the heuristic interpretation of relations.

To prevent errors from compounding in heuristic mode, the system "shrinks" intervals when a conclusion is reused as a premise, effectively modeling decreasing certainty.

Implementation and Case Study

The system is implemented using an Assumption-based Truth Maintenance System (ATMS), which manages multiple hypothetical relations and detects contradictions. To manage computational complexity, O[M] leverages dimensional constraints, ensuring the system only compares quantities with compatible physical units.

The formalism was evaluated using a case study in biochemical pathways, specifically Glycolysis. By modeling the relative sizes and concentrations of enzymes and metabolites, the O[M] system successfully identified inconsistencies in hypotheses regarding "rate-limiting steps" without requiring exact numerical values.

Conclusion

O[M] demonstrates that order-of-magnitude reasoning is sufficient for high-level troubleshooting in complex domains. While the use of an ATMS presents scaling challenges for massive systems, the flexibility provided by the accuracy parameter ($e$)—which allows the logic to adapt to different professional contexts—makes O[M] a powerful tool for bridging the qualitative-quantitative divide.

Summary updated

BibTeX
@inproceedings{mavrovouniotis1987reasoning,
  author = {Michael L. Mavrovouniotis and George Stephanopoulos},
  title = {{Reasoning with Orders of Magnitude and Approximate Relations}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1987},
}
Figure