The Algebra of Units: From Buckingham's Pi-grec Theorem to Latent-Variable Learning

2026-06-15 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Engineering & Applied Sciences, Mathematics & Computational Sciences · Depth: Expert, quick

Summary

A new method automatically discovers dimensionless numbers from data, bypassing the traditional requirement for expert physical insight. Building on the Buckingham Pi-grec theorem, this approach logarithmically transforms measurements, revealing a low-dimensional manifold whose geometry is defined by underlying dimensionless groups. Singular Value Decomposition (SVD) identifies this manifold, followed by an integer-exponent search and a repeating-variable filter to recover candidate quantities. The procedure successfully recovers familiar engineering groups, including the flow coefficient, head coefficient, and Mach number. Demonstrated on a synthetic compressor dataset of 16,000 measurements, the method accurately reproduces the compressor performance map with an error below 0.01%, highlighting a strong connection between classical dimensional analysis and modern data-driven learning.

Key takeaway

For engineers and research scientists modeling complex physical systems, this method offers a data-driven path to discover fundamental dimensionless groups. You can reduce reliance on extensive domain expertise for dimensional analysis, potentially accelerating model development and improving interpretability. Consider applying this technique to large datasets from experiments or simulations to automatically derive key scaling parameters and enhance predictive model accuracy.

Key insights

Dimensionless groups can be automatically discovered from data using algebraic methods, without prior physics knowledge.

Principles

• Logarithmic transformation reveals low-dimensional manifolds for scaled physical systems.
• SVD can directly identify these manifolds from raw dimensional data.
• Dimensionless groups are recoverable via integer-exponent combinations and filtering.

Method

Logarithmically transform measurements, then apply Singular Value Decomposition (SVD) to identify a low-dimensional manifold. Search integer-exponent combinations for candidate dimensionless quantities, filtering them with a repeating-variable criterion.

In practice

• Automatically recover engineering groups like Mach number.
• Reproduce system performance maps with high precision.
• Build interpretable and data-efficient physical models.

Topics

• Dimensional Analysis
• Buckingham Pi Theorem
• Latent Variable Learning
• Singular Value Decomposition
• Engineering Physics
• Data-driven Modeling

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• A Machine Learning Approach to the Nirenberg Problem — A physics-informed neural network can effectively solve the Nirenberg problem and distinguish between realisable and non-realisable curvatures.
• Discovering and decoding latent mean-field structure with variational autoencoders — A VAE's successful reconstruction of many-body systems reveals a latent mean-field theory, decodable from its structure.
• Engineering Hybrid Physics-Informed Neural Networks for Next-Generation Electricity Systems: A State-of-the-Art Review — Physics-informed machine learning integrates domain physics into models, enhancing accuracy and efficiency for data-scarce electricity systems.
• Reformulating Neural Operators in $d+1$ Dimensions for Embedding Evolution — Neural operators redefined in d+1 dimensions using Schrödingerisation better capture system evolution and achieve superior performance.
• A Differentiable Measure of Algebraic Complexity: Provably Exact Discovery of Group Structures — The HyperCube model's objective function inherently biases its optimization towards discovering group structures and their unitary representations.
• Exact equivariance, kept through training, buys zero-shot generalisation across the symmetry group — Exact equivariance, maintained through training, enables zero-shot generalization across symmetry groups.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.