---
title: Dimensionless Learning Framework
url: https://www.emergentmind.com/topics/dimensionless-learning
type: topic
---

# Dimensionless Learning Framework

Dimensionless learning is a data‑driven framework for discovering dimensionless groups, scaling laws, and related low‑dimensional representations directly from data while enforcing dimensional consistency. In one core formulation, it “automatically discover[s] dominant and unique dimensionless numbers and scaling laws from data” by embedding “the principle of dimensional invariance into a two‑level machine learning scheme” [2111.03583]. Across the broader literature, the term also denotes learning in a dimensionless representation so that “the learned knowledge applies unchanged to any physically similar system,” exact “units equivariant machine learning” performed in a dimensionless space, and information‑theoretic or symbolic methods for selecting the dimensionless variables with the highest predictive power [2512.08667][2204.00887][2504.03927]. This suggests that dimensionless learning is not a single algorithm but a family of methods organized around the same principle: rewrite physical prediction, control, and inference problems in terms of dimensionless quantities, then use data to identify the combinations that are most informative, most parsimonious, or most transferable.

## 1. Conceptual basis and relation to classical dimensional analysis

Classical dimensional analysis guarantees that if variables are built from a set of independent base dimensions, physically meaningful relations can be written in terms of dimensionless groups. It also has a central limitation: “the Buckingham–\(\pi\) theorem tells us the *dimension* of the space of possible dimensionless groups,” but “there are infinitely many valid choices of \(\pi\) groups,” it “provides no guidance on *which* groups are most relevant to an actual dataset,” and “it does not supply the functional relationship \(F\) among them” [2111.03583]. Closely related statements recur throughout the literature: Buckingham–\(\Pi\) gives a basis of admissible groups, but “this set is not unique” and classical analysis alone does not identify the “most physically meaningful dimensionless basis” for a given dataset [2202.04643].

In the data‑driven formulation, a generic physical relation
\[
y = f(p_1, p_2, \dots, p_n)
\]
is re‑expressed as
\[
\Pi_{\text{out}} = F(\Pi_1,\dots,\Pi_k),
\]
where each \(\Pi\) is a monomial
\[
\Pi = \prod_{i=1}^n p_i^{\alpha_i}.
\]
The aim is to discover “dominant dimensionless numbers” and “scaling laws” directly from measurements, including “noisy experimental data,” while keeping the learned representation “scale‑free and unit‑independent” [2111.03583].

Several later formulations widen the scope of the term. “Dimensionless machine learning” is defined as imposing “exact units equivariance” by first constructing “a dimensionless version of its inputs using classic results from dimensional analysis, and then perform[ing] inference in the dimensionless space” [2204.00887]. “Dimensionless learning based on information” treats the best dimensionless groups as those with “the highest predictive power by measuring their shared information content” [2504.03927]. “Hierarchical Dimensionless Learning (Hi‑\(\pi\))” describes a “physics-data hybrid-driven approach for discovering dimensionless parameter combinations” that balances “accuracy and complexity” [2507.18332].

## 2. Mathematical structure and dimensionless parameterization

The common algebraic structure begins with the dimension matrix. If \(p_i\) has dimension vector \(\mathbf{d}_i\), then
\[
\mathbf{D} = [\mathbf{d}_1\ \mathbf{d}_2\ \dots\ \mathbf{d}_n]
\]
collects all dimensional information, and a candidate dimensionless monomial with exponent vector \(\mathbf{w}\) must satisfy
\[
\mathbf{D}\mathbf{w} = \mathbf{0}.
\]
Hence all admissible exponent vectors lie in the null space of \(\mathbf{D}\), with
\[
m = n - \operatorname{rank}(\mathbf{D}).
\]
If \(\mathbf{w}_{b1},\dots,\mathbf{w}_{bm}\) is a basis of that null space, then any admissible exponent vector can be written as
\[
\mathbf{w} = \mathbf{W}_b\,\boldsymbol{\gamma},
\]
where \(\mathbf{W}_b = [\mathbf{w}_{b1}\ \dots\ \mathbf{w}_{bm}]\) and \(\boldsymbol{\gamma}\) are basis coefficients to be learned from data [2111.03583].

A closely related representation appears in “Dimensionally Consistent Learning with Buckingham Pi”:
\[
\pi_j = \prod_{i=1}^{d} \tilde p_i^{\Phi_{ij}},
\qquad
\mathbf{D}\mathbf{\Phi} = \mathbf{0},
\]
with the values of the \(\Pi\) groups over a dataset written as
\[
\boldsymbol{\Pi} = \exp\left( \log(\tilde{\mathbf P})\, \mathbf \Phi \right).
\]
This makes the multiplicative structure linear in log space and turns dimensionless learning into a constrained search over the null space of the dimensional matrix [2202.04643].

A more geometric version is given in “The Algebra of Units: From Buckingham’s \(\Pi\) Theorem to Latent‑Variable Learning.” If \(q_j>0\) and \(y_j=\log q_j\), then for an exponent vector \(a\in\ker(D^\top)\),
\[
\log \Pi = a^\top y.
\]
The paper states that, after logarithmic transformation, measurements “lie on a low-dimensional manifold whose geometry is determined by the underlying dimensionless groups,” and that singular value decomposition identifies this manifold directly from data [2606.16737]. This suggests a direct connection between dimensional analysis and latent‑variable methods such as PCA, SVD, and POD.

## 3. Learning formulations and algorithmic families

A central formulation is the “two‑level optimization” of dimensionless learning. **Level 1** searches over admissible dimensionless groups, i.e. over \(\boldsymbol{\gamma}\), and **Level 2** fits a scaling relation and evaluates predictive quality. For each candidate \(\boldsymbol{\gamma}\), the method computes the corresponding \(\Pi\)-groups, fits a regressor such as a polynomial or XGBoost model, and evaluates a score such as \(R^2\) on held‑out data [2111.03583]. Practical search strategies include grid search, pattern search, and “gradient‑based schemes,” while parsimony is encouraged by limiting the number of groups \(k\), constraining coefficients to small ranges such as \([-2,2]\), and favoring “small rational numbers” for the exponents [2111.03583].

Other papers instantiate the same principle with different learners. “Dimensionally Consistent Learning with Buckingham Pi” develops three techniques: a constrained optimization with kernel ridge regression, **BuckiNet**, and a **dimensionless SINDy** procedure. BuckiNet constrains its first layer to implement
\[
\boldsymbol{\Pi}_p = \exp\big(\log(\mathbf P)\, \mathbf\Phi_p\big),
\]
and adds the soft dimensional-consistency penalty
\[
\mathcal L_{\text{null}} = \|\mathbf D_p \mathbf\Phi_p\|_2^2.
\]
The SINDy-based variant searches over candidate \(\Pi\)-groups and time scales to discover sparse, dimensionless dynamical equations [2202.04643].

Two later developments emphasize model selection rather than only prediction. IT‑\(\pi\) defines an information‑theoretic irreducible error bound
\[
\min_f \|\Pi_o - f(\boldsymbol{\Pi})\|_p
\ge e^{-I_\alpha(\Pi_o;\boldsymbol{\Pi})}\,c(\alpha,p,h_{\alpha,o})
\equiv \epsilon_{LB},
\]
and chooses the dimensionless variables that minimize that lower bound, or equivalently maximize the shared information between \(\Pi_o\) and \(\boldsymbol{\Pi}\) [2504.03927]. Hi‑\(\pi\) uses dimensional analysis to generate initial dimensionless variables, symbolic regression to search for nonlinear recombinations, and a polynomial map
\[
\Pi \approx F_{\text{poly}}(\mathbf{T},\boldsymbol{\beta})
\]
to score “accuracy and complexity” [2507.18332].

A different branch of the literature formulates the problem as exact symmetry enforcement. “Dimensionless machine learning: Imposing exact units equivariance” represents changes of units as a group action and defines a model to be units equivariant if
\[
f(g \cdot x) = g \cdot f(x).
\]
Its prescription is to construct dimensionless invariants first, learn only in that invariant space, and then decode back to dimensional outputs [2204.00887].

## 4. Empirical discoveries and representative physical applications

The original “dimensionless learning” paper demonstrates the framework on “noisy experimental measurements (not synthetic data)” from several engineering problems [2111.03583]. In turbulent Rayleigh–Bénard convection, the method discovers the unique optimum at \(\gamma_1=\gamma_2=\gamma_3=1\), yielding
\[
\Pi = \frac{g \alpha \Delta T\, h^3}{\nu \kappa} = \mathrm{Ra},
\]
and learns
\[
\mathrm{Nu} = f(\mathrm{Ra})
\]
with a 5th‑order polynomial, collapsing the data in log–log space with \(R^2 \approx 0.999\) [2111.03583]. In laser–metal interaction, it recovers the “keyhole number”
\[
\mathrm{Ke} = \frac{\eta P}{(T_l - T_0)\,\pi \rho C_p \sqrt{\alpha V_s r_0^3}}
\]
and a nearly linear law
\[
e^* = 0.12\,\mathrm{Ke} - 0.30,
\]
with \(R^2 \approx 0.98\) for keyhole aspect ratio [2111.03583]. In metal 3D printing porosity, the method shows that a naive energy density
\[
E = \frac{\eta_m P}{V_s d^2}
\]
gives only \(R^2 \approx 0.13\), whereas two learned dimensionless groups, including
\[
\mathrm{NED} = \frac{\eta_m P}{V_s \rho C_p (T_l - T_0) H L}
\]
and
\[
\mathrm{NEP} = \frac{\eta_m P}{V_s \rho C_p (T_l - T_0) d^2},
\]
allow XGBoost models to reach \(R^2\) up to about \(0.95\) on test data [2111.03583].

“Dimensionally Consistent Learning with Buckingham Pi” recovers classical variables in several canonical examples. For the bead on a rotating hoop, BuckiNet identifies
\[
\gamma = \frac{R\omega^2}{g},
\qquad
\epsilon = \frac{m^2 g R}{b^2},
\]
and dimensionless SINDy recovers the associated dimensionless ODE structure [2202.04643]. For the Blasius boundary layer, constrained optimization discovers the similarity variable
\[
\eta = y\sqrt{\frac{U_\infty}{\nu x}}
\]
to numerical precision from sparse dimensional data [2202.04643]. For Rayleigh–Bénard convection, the same paper identifies the inverse Rayleigh number as the best control parameter for a Landau‑type normal form near onset [2202.04643].

Hi‑\(\pi\) extends this program to multi‑parameter discovery. In Rayleigh–Bénard convection it “accurately extracted two intrinsic dimensionless parameters: the Rayleigh number and the Prandtl number,” and on a dataset spanning small and large \(\mathrm{Pr}\) it reports interpolation relative error \(1.29\%\) with \(R^2=0.998\) and extrapolation relative error \(1.53\%\) with \(R^2=0.996\) [2507.18332]. In viscous pipe flow it “automatically discovers two optimal dimensionless parameters: the Reynolds number and relative roughness,” and with a 4th‑order polynomial reports relative error \(0.71\%\) and \(R^2=0.999\) [2507.18332]. In compressibility correction for subsonic flow, it extracts the Prandtl–Glauert‑like combination
\[
C_{p,PG} = \frac{C_{p,0}}{\sqrt{1-M^2}}
\]
as a useful intermediate variable for recovering the Karman–Tsien structure [2507.18332].

The LPBF literature uses a related workflow to learn reduced process descriptions from high‑fidelity simulation. A thermo‑fluid finite‑element model is non‑dimensionalized using quantities such as
\[
\mathrm{Pe} = \frac{l_p \nu_p}{\alpha},
\qquad
\mathrm{Ma} = \frac{\dfrac{d\gamma}{dT} \, l_p \Delta T}{\mu \alpha},
\qquad
\mathrm{Ste} = \frac{c(T_l - T_s)}{L},
\]
and a learned heat‑absorption index
\[
\hat{\mathbf{U}} = a_0 + a_1 \mathbf{E} + a_2 \mathrm{Pe},
\qquad
\mathbf{E} = \frac{P}{k (T_l - T_\infty)} = \mathrm{Pe}\,\mathrm{Q},
\]
is then correlated with meltpool morphology, temperature gradients, and cooling rate [2211.06312].

## 5. Control, units equivariance, and transfer across scales

In control, dimensionless learning is used in a different but related sense: “learning control laws, policies, and tuning parameters in a dimensionless space, so that the learned knowledge applies unchanged to any physically similar system” [2512.08667]. The core construction starts from
\[
\dot{x} = f(x,u;p)
\]
and introduces scalings
\[
x = M_x \tilde{x}, \qquad u = M_u \tilde{u}, \qquad t = m_t \tilde{t},
\]
yielding the dimensionless dynamics
\[
\frac{d\tilde{x}}{d\tilde{t}} = \tilde{f}(\tilde{x}, \tilde{u}; \tilde{p}).
\]
After discretization,
\[
\tilde{x}_{k+1} = \tilde{f}_d(\tilde{x}_k,\tilde{u}_k;\tilde{p}),
\]
and, for similar systems, the dimensionless model is identical [2512.08667].

This structure is extended to dimensionless MDPs and a dimensionless MPC problem in which costs and constraints are also non‑dimensionalized:
\[
\tilde{L}(\tilde{x},\tilde{u}) = m_L^{-1} L(M_x\tilde{x},M_u\tilde{u}),
\qquad
\tilde{h}(\tilde{x},\tilde{u}) = M_h^{-1} h(M_x\tilde{x},M_u\tilde{u}) \le 0.
\]
The key claim is that if two systems are dynamically similar and share the same dimensionless parameters and sampling time, then “their *dimensionless MPC problems are identical*,” so the tuned controller transfers directly [2512.08667].

The paper demonstrates this on cartpole swing‑up and race‑car lap time minimization. For the cartpole, it uses
\[
\Pi_1 = \frac{m_p}{m_c},
\qquad
\Pi_2 = \frac{\mu_f}{m_c} \sqrt{\frac{l}{g}},
\]
and defines the state and input scalings
\[
M_s = \mathrm{diag}\left(l,\; 1,\; \sqrt{gl},\; \sqrt{g/l}\right), \quad M_a = m_c g,\quad m_t = \sqrt{\frac{l}{g}}.
\]
The result is that with “dimensionless MPC,” reinforcement learning continues smoothly after switching to a dynamically similar system of different physical scale, whereas with a “purely dimensional MPC,” performance “nearly resets when the scale changes” [2512.08667]. For the race car, Bayesian optimization on a small‑scale car transfers to a full‑size car under preserved \(\Pi\)-groups, without re‑tuning [2512.08667].

A broader machine‑learning treatment of the same idea appears in “Dimensionless machine learning: Imposing exact units equivariance.” There, the units rescaling group is \(G=\mathbb{R}_{>0}^k\), a units‑equivariant model satisfies
\[
f(g\cdot x)=g\cdot f(x),
\]
and dimensionless invariants are constructed by solving
\[
\sum_{i=1}^{d} \alpha_i \mathbf{u}_i = \mathbf{0}
\]
for the units vectors \(\mathbf{u}_i\) [2204.00887]. In the springy double pendulum benchmark, the dimensionless Hamiltonian neural network has in‑distribution relative error \(0.0061 \pm 0.0024\), compared with \(0.0055 \pm 0.0030\) for a dimensional baseline, but under units‑rescaled OOD testing the dimensionless model remains at \(0.0089 \pm 0.0034\) while the baseline rises to \(0.3669 \pm 0.0050\) [2204.00887]. This suggests that exact units equivariance and scale transfer are practically significant consequences of learning in a dimensionless space.

## 6. Information-theoretic, geometric, and domain-specific extensions

Several recent papers generalize dimensionless learning beyond regression over hand‑selected features. IT‑\(\pi\) defines optimal dimensionless variables as those minimizing a model‑free lower bound on the prediction error based on Rényi mutual information [2504.03927]. It states that the method can “rank variables by predictability, identify distinct physical regimes, uncover self-similar variables, determine the characteristic scales of the problem, and extract its dimensionless parameters” [2504.03927]. It also defines “model efficiency”
\[
\eta(f) = \frac{\epsilon_{LB}^*}{\epsilon_f},
\]
thereby comparing any fitted model with the information‑theoretic lower bound [2504.03927].

A geometric extension appears in “The Algebra of Units: From Buckingham’s \(\Pi\)-grec Theorem to Latent-Variable Learning,” where gauge variation across scaled realizations leads to a within‑cluster deviation matrix
\[
\Delta Y = U \Sigma V^\top,
\]
and the last \(k-n\) right singular vectors span \(\ker(D^\top)\) exactly under the stated assumptions [2606.16737]. On a synthetic compressor dataset of 16,000 measurements, the method recovers the flow coefficient, head coefficient, and Mach number to numerical precision, with “error below \(0.01\%\)” on the performance map [2606.16737].

Other domain‑specific examples show that the same dimensionless perspective can organize inference in very different fields. “Dimensionless cosmology” argues that “only dimensionless combinations of ‘fundamental constants’ can be operationally defined and compared,” and recasts Big Bang nucleosynthesis, recombination, and CMB inference in terms of
\[
\alpha_{\rm em},\quad \alpha_{\rm g},\quad \mu,\quad \theta
\]
rather than varying dimensional constants such as \(G\) directly [1109.0492]. In neutron‑star structure, the TOV equations are recast using
\[
\widehat{\varepsilon}=\frac{\varepsilon}{\varepsilon_c},\qquad
\widehat{P}=\frac{P}{\varepsilon_c},\qquad
\widehat{r}=\frac{r}{Q},\qquad
\widehat{M}=\frac{M}{W},
\]
leading to EOS‑independent mass–radius–compactness scalings and an “Intrinsic and Perturbative Analyses of the Dimensionless (IPAD) TOV equations” program for inferring core EOS structure from observables [2501.18676]. In reacting porous‑flow systems, a “dimensionless framework” maps a particle size distribution into distributions of Damköhler numbers such as
\[
\mathrm{Da}_{\mathrm{ext}}(d_p)=\frac{K_{\mathrm{ext}}}{d_p},
\qquad
\mathrm{Da}_{\mathrm{diff}}(d_p)=\frac{K_{\mathrm{diff}}}{d_p^2},
\]
and uses these distributions to compare column and heap leaching across scale [2601.14529]. These cases suggest that dimensionless learning can denote both explicit learning algorithms and a broader strategy of expressing governing relations, control laws, and inference problems in a reduced dimensionless representation.

## 7. Limitations, misconceptions, and open directions

A persistent misconception is that dimensional analysis alone identifies the physically relevant variables. The literature consistently states the opposite: Buckingham–\(\pi\) provides a null space, not a unique answer, and does not identify the functional relationship or the most predictive groups [2111.03583][2202.04643]. Dimensionless learning addresses that ambiguity, but it does not remove all assumptions.

Most formulations require that “relevant variables and their dimensions are correctly specified,” and that “the true physics admits a description by dimensionless monomials” or by a dimensionless representation of the chosen form [2111.03583]. If important variables are missing, “no high‑quality scaling can be recovered,” although the resulting low maximal \(R^2\) can reveal that omission [2111.03583]. The black‑body example in units‑equivariant machine learning shows the same issue differently: without including Planck’s constant \(h\), dimensional analysis yields only the Rayleigh–Jeans form and cannot recover Planck’s law [2204.00887].

Another limitation is computational. Search over \(\boldsymbol{\gamma}\)-space or over candidate exponent vectors can be expensive in high‑dimensional null spaces [2111.03583][2202.04643]. Multiple‑group extraction “needs more data” and is more sensitive to sample quality than one‑group discovery [2507.18332]. In the tutorial literature, key open problems include “managing the computational cost of identifying multiple dimensionless groups,” “automating the selection of relevant input variables,” understanding the role of sample size and data distribution, and developing “user-friendly tools for experimentalists” [2512.15760].

The effect of noise is a recurrent concern. The tutorial on dimensionless learning studies “measurement noise and discrete sampling,” and reports that a “quantization regularizer” pushing learned coefficients toward integers or half‑integers improves robustness and interpretability [2512.15760]. IT‑\(\pi\) adds a different perspective by replacing model‑specific error with a bound determined by shared information; this suggests a route for deciding whether poor predictive performance is due to the model class or to information missing from the inputs [2504.03927].

Taken together, these developments suggest several research directions already present in the literature: integrating symbolic regression, sparse regression, and deeper nonlinear learners while preserving dimensional consistency; building dimensionless surrogates for PDEs and control; exploiting SVD and latent‑variable geometry in log space; using information‑theoretic objectives to rank or select \(\Pi\)-groups; and extending dimensionless learning from single numbers to distributions of dimensionless quantities in heterogeneous systems [2202.04643][2606.16737][2504.03927][2601.14529]. In essence, the field treats dimensionless representation not merely as a preprocessing step, but as a principled way to constrain learning, organize physical interpretation, and improve extrapolation across scales.

Source: https://www.emergentmind.com/topics/dimensionless-learning