Connected Theorems: A Graph-Based Approach to Evaluating Mathematical Results

Haocheng Ju, Tianyi Xu, Bin Dong

Abstract

The evaluation of mathematical results plays a central role in assessing researchers' contributions and shaping the direction of the field. Currently, such evaluations rely primarily on human judgment, whether through journal peer review or committees at research institutions. To complement these traditional processes, we propose a data-driven approach. We construct a hierarchical graph linking theorems, papers, and fields to capture their citation relationships. We then introduce a PageRank-style algorithm to compute influence scores for these entities. Using these scores, we analyze the evolution of field rankings over time and quantify the impact between fields. We hope this framework can contribute to the development of more advanced, quantitative methods for evaluating mathematical research and serve as a complement to expert assessment.

1. Introduction

In mathematics, measuring "impact" is a complex challenge. Unlike disciplines where citations accumulate rapidly, mathematical results often exhibit a "long tail," where a theorem may remain dormant for decades before becoming foundational to a new breakthrough. Traditional metrics like the h-index or journal impact factors often fail to capture the granular reality that a single theorem within a paper may be far more influential than the paper itself. This study introduces a quantitative, multi-level framework to map mathematical influence by looking beyond paper-to-paper citations to the underlying theorem-to-theorem relationships.

2. Methodology: The Hierarchical Graph

The core of this research is the construction of a hierarchical three-level directed graph that models the flow of mathematical knowledge.

The hierarchical graph structure illustrating the connections between the field, paper, and theorem levels.

As shown in the figure above, the graph consists of three layers:

  1. Theorem Level (Bottom): Nodes represent individual theorems, linked when the proof of one theorem cites another.
  2. Paper Level (Middle): Nodes represent published papers, connected by standard bibliographic citations.
  3. Field Level (Top): Nodes represent broad mathematical disciplines (e.g., Algebra, PDE, Statistics).

2.1 The Influence Algorithm

To calculate influence, we adapted the PageRank algorithm to propagate scores both within and across levels. A theorem gains influence if it is cited by other influential theorems or belongs to an influential paper. Conversely, a paper gains influence from its citations, its parent field, and its high-impact theorems. The algorithm incorporates a weighting system that penalizes self-citations and rewards external impact to ensure the metric reflects genuine scholarly influence.

3. Data Acquisition and Processing

The dataset was constructed by scraping over 500 journals, focusing on papers with arXiv preprints available in HTML via the ar5iv project. This format was essential for the automated extraction of theorem IDs and proof structures. The resulting graph includes 141,087 papers and 2,387,375 theorems, with over 1.46 million theorem-level citations. Papers were mapped to 13 primary fields using Mathematics Subject Classification (MSC) codes.

4. Key Insights and Results

4.1 Directional Knowledge Flow

The framework quantifies asymmetric influence between disciplines. For instance, Analysis exerts a 27% stronger influence on PDE than vice versa, and Numerical Analysis significantly influences Optimization. This highlights a "top-down" flow where theoretical fields provide the foundational tools for application-driven domains.

A heatmap illustrating the directional impact scores between different mathematical fields.

4.2 Field Connectivity and Stability

The data reveals "core clusters" where fields like Differential Geometry, Analysis, PDE, and Physics move in lockstep. Interestingly, while the proportion of papers in Probability has plateaued since 2013, its influence score has continued to rise, suggesting its results are becoming increasingly foundational to other areas. In contrast, fields like Statistics and Numerical Analysis consistently ranked lower, likely due to a lower frequency of citations from "pure" mathematics papers within the arXiv-based dataset.

5. Limitations and Future Directions

Several challenges remain, including the reliance on arXiv's HTML record, which misses historical and non-arXiv literature. Additionally, the current model assigns each paper to a single field, whereas mathematics is inherently interdisciplinary. Future work will explore the use of Large Language Models (LLMs) to better understand the context of citations—distinguishing between foundational tools and passing mentions—to further refine the measurement of mathematical impact.

Summary updated

Abstract

The evaluation of mathematical results plays a central role in assessing researchers' contributions and shaping the direction of the field. Currently, such evaluations rely primarily on human judgment, whether through journal peer review or committees at research institutions. To complement these traditional processes, we propose a data-driven approach. We construct a hierarchical graph linking theorems, papers, and fields to capture their citation relationships. We then introduce a PageRank-style algorithm to compute influence scores for these entities. Using these scores, we analyze the evolution of field rankings over time and quantify the impact between fields. We hope this framework can contribute to the development of more advanced, quantitative methods for evaluating mathematical research and serve as a complement to expert assessment.
BibTeX
@misc{ju2025connected,
  author = {Haocheng Ju and Tianyi Xu and Bin Dong},
  title = {{Connected Theorems: A Graph-Based Approach to Evaluating Mathematical Results}},
  year = {2025},
  eprint = {2508.17596},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2508.17596},
}
Figure