The fundamental principles of reproducibility

Area of Science:

• Computational science
• Scientific methodology
• Research reliability

Background:

• The term 'reproducibility' is often used ambiguously in scientific literature.
• Existing literature on reproducibility and replication lacks a unified framework.
• The need for clear definitions is critical for advancing computational science.

Purpose of the Study:

• To fundamentally analyze and characterize the scientific method to establish a robust definition of reproducibility.
• To develop precise terminology for discussing reproducibility and replication in research.
• To propose a structured framework for understanding different facets of reproducibility.

Main Methods:

• Analysis of the scientific method to derive core principles.
• Survey of existing literature on reproducibility and replication.
• Modeling experiments as tasks and problem-solving methods.
• Application of machine learning as a case study.

Main Results:

• A fundamental definition of reproducibility is established.
• Three distinct degrees of reproducibility are specified.
• Four unique types of reproducibility are identified and detailed.
• The proposed framework is exemplified using machine learning.

Conclusions:

• The proposed definitions and classifications enhance clarity in scientific communication.
• This work provides a foundation for improving reliability and reproducibility in computational science.
• The framework aids in verification, validation, and uncertainty quantification in silico.