betaFeedback

WorksPhysics and Astronomy

Dimensionless learning based on information

Abstract

Dimensional analysis is one of the most fundamental tools for understanding physical systems. However, the construction of dimensionless variables, as guided by the Buckingham-ππ theorem, is not uniquely determined. Here, we introduce IT-ππ, a model-free method that combines dimensionless learning with the principles of information theory. Grounded in the irreducible error theorem, IT-ππ identifies dimensionless variables with the highest predictive power by measuring their shared information content. The approach is able to rank variables by predictability, identify distinct physical regimes, uncover self-similar variables, determine the characteristic scales of the problem, and extract its dimensionless parameters. IT-ππ also provides a bound of the minimum predictive error achievable across all possible models, from simple linear regression to advanced deep learning techniques, naturally enabling a definition of model efficiency. We benchmark IT-ππ across different cases and demonstrate that it offers superior performance and capabilities compared to existing tools. The method is also applied to conduct dimensionless learning for supersonic turbulence, aerodynamic drag on both smooth and irregular surfaces, magnetohydrodynamic power generation, and laser-metal interaction.

Introduction

Physical laws and models must adhere to the principle of dimensional homogeneity [1, 2], i.e., they must be independent of the units used to express their variables. A similar idea was first introduced by Newton in his Principia, under the term “dynamically similar systems” [3], although Galileo had already employed the notion of similar systems when discussing pendulum motions. Over the following centuries, the concept of similar systems was loosely applied in a variety of fields, including engineering (Froude, Bertrand, Reech), theoretical physics (van der Waals, Onnes, Lorentz, Maxwell, Boltzmann), and theoretical and experimental hydrodynamics (Stokes, Helmholtz, Reynolds, Prandtl, Rayleigh) [3]. The approach was formally articulated in the early 20th century, laying the foundation for what is known today as dimensional analysis [1, 4, 5].

The applications of dimensional analysis extend beyond the construction of dimensionally consistent physical laws. It provides the conditions under which two systems share identical behavior (dynamic similarity), allowing predictions from laboratory experiments to be extended to real-world applications [3, 6]. Dimensional analysis also facilitates dimensionality reduction, i.e., simplifying physical problems to their most fundamental forms, thus decreasing the amount of data required for analysis [7, 8]. Another application is the discovery of self-similar variables, for which systems exhibit invariant solutions under appropriate scaling [9]. Finally, dimensional analysis can reveal the physical regimes in which different phenomena dominate (e.g., incompressible versus compressible flow), enabling researchers to identify the most influential physical mechanisms in a given context [10].

A key landmark of dimensional analysis is the Buckingham-π\pi theorem [1], which offers a systematic framework for deriving dimensionless variables. However, the solution is not unique, as there are infinitely many possible ways to construct these variables. To address this limitation, recent studies have developed data-driven tools to identify unique dimensionless variables that minimize the error for a given model structure, particularly in the data-rich context enabled by modern simulations and experiments. These methods combine dimensional analysis with machine learning techniques to identify dimensionless variables using multivariate linear regression [11], polynomial regression [12, 13, 14], ridge regression [15], hypothesis testing [16], Gaussian process regression [17], neural networks [13, 15, 18, 19], sparse identification of nonlinear dynamics [12, 15], clustering [20], symbolic regression [13, 21, 22] and entropic principles [23, 24]. Table 1 offers a non-exhaustive comparative overview of data-driven methods for discovering dimensionless variables, highlighting their respective capabilities. These include applicability to ordinary and partial differential equations (ODEs/PDEs), the ability to rank dimensionless variables by predictability (and thus relative importance), identification of distinct physical regimes, detection of self-similar behavior, and extraction of characteristic scales (e.g., length and time scales of the system). Another key capability is whether the method can determine a bound on the minimum possible error across all models—which in turn enables the definition of a model efficiency. The comparison in Table 1 indicates that, although many approaches incorporate several of these properties, no single method currently supports all of these capabilities simultaneously. One notable shortcoming of previous data-driven methods is that they are not model-free; i.e., the discovery of dimensionless variables relies on a predefined model structure (e.g., linear regressions, neural networks,...). This can lead to potentially biased results, as the dimensionless variables identified may not be optimal for other models. A detailed account of the challenges in dimensionless learning can be found in the Supplementary Materials.

MethodODE/PDEModel-FreeInput RankingRegime DetectionSelf-SimilarCharacteristic ScalesIrreducible ErrorModel Efficiency
Scaling LAWs [11]××✓×––××
Active Subspaces [17]✓×✓×––××
AI Feynman [13]××××––××
Clustering [20]✓×✓✓––××
PyDimension [12]✓×××✓✓××
BuckiNet [15]✓×✓×✓✓××
BSM [21]✓×××✓✓××
IT-π (Current)✓✓✓✓✓✓✓✓

Table 1. Overview of some data-driven dimensionless learning methods and their capabilities. These capabilities include whether each method is applicable to ODEs/PDEs, operates in a model-free manner, ranks inputs by predictability, identifies distinct physical regimes, uncovers self-similarity, determines characteristic scales and dimensionless parameters, provides a bound on the irreducible error, and evaluates model efficiency. Entries marked with “–” indicate that although the method could potentially be extended to infer the corresponding physical property after deriving the dimensionless variables, the authors did not explicitly perform this step.

In this work, we present IT-π\pi, an information-theoretic, model-free method for dimensionless learning. An overview of the method is shown in Figure 2. Our approach is motivated by a fundamental question: Which dimensionless variables are best suited for predicting a quantity of interest, independent of the modeling approach? IT-π\pi addresses this question by unifying the Buckingham-π\pi theorem with the irreducible error theorem. The central idea is that the predictive capability of a model is bounded by the amount of information shared between the input (dimensionless variables) and the output (dimensionless quantity of interest) [25, 26].

Figure 2. Illustration of the dimensionless learning capabilities enabled by IT-π\pi. (a) Ranking of candidate dimensionless variables based on their predictability. (b) Example of regime detection for the skin friction coefficient CfC_f in a wall-bounded turbulent flow, with Reynolds number (Re) and Mach number (Ma) as dimensionless inputs. The score RReR_{\mathrm{Re}} quantifies the relative importance of Re in predicting CfC_f. (c) Degree of dynamic similarity (DoS) between the lift coefficient of a wind tunnel–scaled aircraft model and that of the corresponding full-scale vehicle. (d) Characteristic length scale (SlS_l) and time scale (StS_t) identified for a simple pendulum. (e) Model efficiency in predicting a target quantity Πo\Pi_o using a neural network with inputs Π\Pi and prediction error ϵf\epsilon_f. Additional details and discussion of the examples are provided in the text.

This manuscript is organized as follows. We begin by introducing the information-theoretic irreducible error theorem, which establishes the minimum achievable error across all possible models. This theorem serves as the foundation for a generalized formulation of the Buckingham-π\pi theorem, while enabling all capabilities outlined in Table 1. We then apply IT-π\pi to a broad set of validation and application cases. The validation cases, which have known analytical solutions, are used to benchmark the performance of IT-π. In contrast, the application cases—where no optimal solution is known—are used to discover new dimensionless variables governing the underlying physics. Finally, we compare the performance of IT-π with other dimensionless learning methods across all studied cases.

Results

Dimensionless learning based on information

A physical model (or law) aims to predict a dimensional quantity qoq_o using a set of nn dimensional input variables, q=[q1,q2,…,qn]\mathbf{q} = [q_1, q_2, \ldots, q_n], through the relation q^o=F(q)\hat{q}_o = \mathcal{F}(\mathbf{q}), where q^o\hat{q}_o is an estimate of qoq_o. As an example, consider the prediction of the gravitational force between two objects, qo=Fgq_o = F_g, which depends on q=[m1,m2,r,G]\mathbf{q} = [m_1, m_2, r, G], where m1m_1 and m2m_2 are the masses of the objects, rr is the distance between their centers of mass, and GG is the gravitational constant. According to the Buckingham-π theorem (1), physical models can be reformulated in a dimensionless form as Π^o=f(Π)\hat{\Pi}_o = f(\mathbf{\Pi}), where Π^o\hat{\Pi}_o denotes the predicted dimensionless output, and Π=[Π1,Π2,…,Πl]\mathbf{\Pi} = [\Pi_1, \Pi_2, \ldots, \Pi_l] is the set of dimensionless input variables. Each dimensionless variable has the form Πi=q1ai1q2ai2⋯qnain\Pi_i = q_1^{a_{i1}}q_2^{a_{i2}}\cdots q_n^{a_{in}} and the number of required dimensionless inputs is upper bounded by l=n−nul = n - n_u, where nun_u is the number of fundamental units involved in the problem (e.g., length, time, mass, electric current, temperature, amount of substance, and luminous intensity). For a given LpL_p-norm, the success of the model is measured by the error ϵf=∥Πo−Π^o∥p\epsilon_f = \lVert\mathbf{\Pi}_o - \hat{\mathbf{\Pi}}_o \rVert_p.

Irreducible error as lack of information. Our approach is grounded in the information-theoretic irreducible error theorem [see proof in the Supplementary Materials] [27, 28]. The key insight is that prediction accuracy of any model is fundamentally limited by the amount of information the input contains about the output, where information here is defined within the framework of information theory [29]. More precisely, the error across all possible models ff is lower-bounded by

ϵf≥e−Iα(Πo;Π)⋅c(α,p,hα,o)≡ϵLB,(1)\epsilon_f \ge e^{-I_\alpha(\Pi_o;\Pi)} \cdot c(\alpha,p,h_{\alpha,o}) \equiv\epsilon_{\mathrm{LB}}, \tag*{(1)}

where Iα(Πo;Π)I_\alpha(\Pi_o;\Pi) is the Rényi mutual information of order α\alpha [30], which measures the shared information between Πo\Pi_o and Π\Pi. The value of c(α,p,hα,o)c(\alpha,p,h_{\alpha,o}) depends on the LpL_p-norm, α\alpha and the information content of Πo\Pi_o, denoted by hα,oh_{\alpha,o} [see Methods]. The irreducible (lower bound) error is denoted as ϵLB\epsilon_{\mathrm{LB}}. The inequality in Eq. (1) represents a fundamental mathematical constraint that holds regardless of the complexity of the statistical relationship between input and output, and must therefore be satisfied by any predictive model, irrespective of its form. When an exact functional relationship exists between the input and the output, the information measure Iα(Πo;Π)I_\alpha(\Pi_o;\Pi) converges to infinity, indicating that an exact model is possible (ϵLB=0\epsilon_{\mathrm{LB}}=0). In contrast, if some of the variables influencing Πo\Pi_o are inaccessible or unmeasurable, Iα(Πo;Π)I_\alpha(\Pi_o;\Pi) remains finite, leading to an irreducible error (ϵLB>0\epsilon_{\mathrm{LB}}>0) that cannot be eliminated. It is interesting to note that the inequality in Eq. (1) holds for a range of values of α\alpha. However, the most useful case occurs when ϵLB\epsilon_{\mathrm{LB}} is maximized, yielding the tightest bound. A detailed analysis of the role of α\alpha in Eq. (1) is provided in the Supplementary Material.

The irreducible error ϵLB\epsilon_{\mathrm{LB}} has several useful properties. First, it is independent of any particular model ff. Second, it is invariant under bijective transformations of inputs when the output is fixed, reflecting the principle that such transformations produce alternative yet equivalent model formulations. Third, it is sensitive to the choice of the LpL_p-norm for the error. For example, predicting extreme events (captured by high LpL_p-norms) may be more challenging and require different variables than predicting weaker, common events (captured by low LpL_p-norms) [see example in the Supplementary Materials]. Finally, Eq. (1) naturally leads to the definition of the normalized irreducible error ϵ~LB=e−Iα(Πo;Π)\tilde{\epsilon}_{\mathrm{LB}}=e^{-I_\alpha(\Pi_o;\Pi)}, which ranges from 0—when exact predictions are possible—to 1—when predictions are essentially random guesses. Occasionally, we will refer to the percentage form of ϵ~LB\tilde{\epsilon}_{\mathrm{LB}}, defined as %ϵ~LB=ϵ~LB×100\%\tilde{\epsilon}_{\mathrm{LB}}=\tilde{\epsilon}_{\mathrm{LB}}\times100.

Information-theoretic Buckingham-π\pi theorem (IT-π\pi). Following Eq. (1), we define the optimal dimensionless inputs Π∗=[Π1∗,Π2∗,…,Πk∗∗]\Pi^*=[\Pi_1^*,\Pi_2^*,\ldots,\Pi_{k^*}^*] and dimensionless output Πo∗\Pi_o^* for a given LpL_p-norm as those satisfying

Π∗, Πo∗=arg⁡min⁡Π,Πo max⁡α[ϵLB].(2)\Pi^*,\ \Pi_o^*=\underset{\Pi,\Pi_o}{\arg\min}\ \underset{\alpha}{\max}[\epsilon_{\mathrm{LB}}]. \tag*{(2)}

Figure 1 illustrates the optimization process from Eq. (2). This model-free formulation ensures that the identified dimensionless variables yield the highest predictive capabilities irrespective of the modeling approach. If desired, the output can be fixed in dimensionless form, requiring only Π∗\Pi^* to be discovered. The irreducible error using the optimal dimensionless inputs Π∗\Pi^* is denoted by ϵLB∗=ϵLB(Π∗)\epsilon_{\mathrm{LB}}^*=\epsilon_{\mathrm{LB}}(\Pi^*). It is satisfied that ϵf≥ϵLB≥ϵLB∗\epsilon_f\ge\epsilon_{\mathrm{LB}}\ge\epsilon_{\mathrm{LB}}^*. The optimization problem from Eq. (2) can be efficiently solved by employing the covariance matrix adaptation evolution strategy (CMA-ES) [31] constrained to the dimensionless candidates for Π\Pi and Πo\Pi_o from the (classical) Buckingham-π\pi theorem [see Methods]. Next, we discuss the different capabilities enabled by IT-π\pi, with illustrative examples provided in Figure 2.

References

  1. [1]E. Buckingham, On physically similar systems; illustrations of the use of dimensional equations. Phys. Rev. 4 (4), 345 (1914).DOI
  2. [2]T. Szirtes, Applied Dimensional Analysis and Modeling (Butterworth-Heinemann) (2007).
  3. [3]S. G. Sterrett, Physically similar systems-a history of the concept. Springer handbook of model-based science pp. 377–411 (2017).DOI
  4. [4]A. C. Palmer, Dimensional analysis and intelligent experimentation (World Scientific) (2008).DOI
  5. [5]B. Lysik, M. Rybaczuk, Dimensional Analysis in the Identification of Mathematical Models (World Scientific Publishing) (1990).
  6. [6]O. Reynolds, An experimental investigation of the circumstances which determine whether the motion of water shall be direct or sinuous, and of the law of resistance in parallel channels. Philos. Trans. R. Soc. Lond. 174, 935–982 (1883).DOI
  7. [7]A. A. Sonin, A generalization of the Π-theorem and dimensional analysis. Proc. Natl. Acad. Sci. U.S.A. 101 (23), 8525–8526 (2004).DOI
  8. [8]L. A. Segel, Simplification and Scaling. SIAM Rev. 14 (4), 547–571 (1972).DOI
  9. [9]G. I. Barenblatt, Scaling, Self-similarity, and Intermediate Asymptotics, no. 14 in Cambridge Texts Appl. Math. (Cambridge Univ. Press) (1996).DOI
  10. [10]J. L. Callaham, J. V. Koch, B. W. Brunton, J. N. Kutz, S. L. Brunton, Learning dominant physical processes with data-driven balance models. Nat. Commun. 12 (1), 1016 (2021).
  11. [11]P. F. Mendez, F. Ordonez, Scaling laws from statistical data and dimensional analysis. J. Appl. Mech. 72 (5), 648–657 (2005).DOI
  12. [12]X. Xie, A. Samaei, J. Guo, W. K. Liu, Z. Gan, Data-driven discovery of dimensionless numbers and governing laws from scarce measurements. Nat. Commun. 13 (1), 7562 (2022).DOI
  13. [13]S.-M. Udrescu, M. Tegmark, AI Feynman: A physics-inspired method for symbolic regression. Sci. Adv. 6 (16), eaay2631 (2020).DOI
  14. [14]L. Jofre, Z. R. del Rosario, G. Iaccarino, Data-driven dimensional analysis of heat transfer in irradiated particle-laden turbulent flow. Int. J. Multiph. Flow 125, 103198 (2020).DOI
  15. [15]J. Bakarji, J. Callaham, S. L. Brunton, J. N. Kutz, Dimensionally consistent learning with Buckingham Pi. Nat. Comput. Sci. 2 (12), 834–844 (2022).arxiv.org/abs/2202.04643
  16. [16]Z. del Rosario, M. Lee, G. Iaccarino, Lurking variable detection via dimensional analysis. SIAM/ASA J. Uncertain. Quantif. 7 (1), 232–259 (2019).arxiv.org/abs/1711.03918
  17. [17]P. G. Constantine, Z. del Rosario, G. Iaccarino, Data-driven dimensional analysis: algorithms for unique and relevant dimensionless groups. arXiv preprint arXiv:1708.04303 (2017).arxiv.org/abs/1708.04303
  18. [18]Z. Xu, X. Zhang, S. Wang, G. He, Artificial neural network based response surface for data-driven dimensional analysis. J. Comput. Phys. 459, 111145 (2022).DOI
  19. [19]R. Watanabe, T. Ishii, Y. Hirono, H. Maruoka, Data-driven discovery of self-similarity using neural networks. Phys. Rev. E. 111 (2), 024301 (2025).DOI
  20. [20]L. Zhang, Z. Xu, S. Wang, G. He, Clustering dimensionless learning for multiple-physical-regime systems. Comput. Methods Appl. Mech. Eng. 420, 116728 (2024).DOI
  21. [21]N. Bempedelis, L. Magri, K. Steiros, Extracting self-similarity from data. arXiv preprint arXiv:2407.10724 (2024).DOI
  22. [22]M. Schmidt, H. Lipson, Distilling free-form natural laws from experimental data. Science 324 (5923), 81–85 (2009).DOI
  23. [23]R. Ma, et al., Building-block-flow model for LES of high-speed flows, in Center for Turbulence Research Proceedings of the Summer Program (2024).
  24. [24]L. Zhang, G. He, Mutual-information-based dimensional learning: Objective algorithms for identification of relevant dimensionless quantities. Computer Methods in Applied Mechanics and Engineering 440, 117922 (2025).DOI
  25. [25]A. Lozano-Durán, G. Arranz, Information-theoretic formulation of dynamical systems: causality, modeling, and control. Phys. Rev. Res. 4 (2), 023195 (2022).
  26. [26]Y. Yuan, A. Lozano-Durán, Limits to extreme event forecasting in chaotic systems. Phys. D. p. 134246 (2024).arxiv.org/abs/2401.16512
  27. [27]E. Lutwak, D. Yang, G. Zhang, Cramér-Rao and moment-entropy inequalities for Rényi entropy and generalized Fisher information. IEEE Trans. Inf. Theory. 51 (2), 473–478 (2005).DOI
  28. [28]T. M. Cover, Elements of information theory (John Wiley & Sons) (1999).
  29. [29]C. E. Shannon, A mathematical theory of communication. Bell Syst. Tech. J. 27 (3), 379–423 (1948).
  30. [30]A. Rényi, On measures of entropy and information, in Proc. 4th Berkeley Symp. Math. Stat. Probab. (Univ. California Press) (1961), pp. 547–562.
  31. [31]N. Hansen, S. D. Müller, P. Koumoutsakos, Reducing the time complexity of the derandomized evolution strategy with covariance matrix adaptation (CMA-ES). Evol. Comput. 11 (1), 1–18 (2003).pubmed.ncbi.nlm.nih.gov/12804094
  32. [36]J. Jiménez, Turbulent flows over rough walls. Annu. Rev. Fluid Mech. 36 (1), 173–196 (2004).
  33. [37]D. Chung, N. Hutchins, M. P. Schultz, K. A. Flack, Predicting the drag of rough surfaces. Annu. Rev. Fluid Mech. 53 (1), 439–471 (2021).DOI
  34. [38]S. H. Strogatz, Nonlinear dynamics and chaos: with applications to physics, biology, chemistry, and engineering (CRC Press) (2018).
  35. [39]E. N. Lorenz, Deterministic Nonperiodic Flow. J. Atmos. Sci. 20 (2), 130–141 (1963).
  36. [40]E. F. Spina, A. J. Smits, S. K. Robinson, The physics of supersonic turbulent boundary layers. Annu. Rev. Fluid Mech. 26 (1), 287–319 (1994).
  37. [41]A. Trettel, J. Larsson, Mean velocity scaling for compressible wall turbulence with heat transfer. Phys. Fluids. 28 (2) (2016).DOI
  38. [42]D. Modesti, S. Pirozzoli, Direct numerical simulation of supersonic pipe flow at moderate Reynolds number. Int. J. Heat Fluid Flow. 76, 100–112 (2019).arxiv.org/abs/1810.03747
  39. [43]T. Kármán, et al., Mechanische ähnlich­keit und turbulenz. Nachr. Ges. Wiss. Göttingen, Math.-Phys. Kl. 1930, 58–76 (1930).
  40. [44]R. Landauer, Information is physical. Phys. Today. 44 (5), 23–29 (1991).
  41. [45]R. Landauer, The physical nature of information. Phys. Lett. A. 217 (4-5), 188–193 (1996).DOI
  42. [46]A. Kraskov, H. Stögbauer, P. Grassberger, Estimating mutual information. Phys. Rev. E. 69 (6), 066138 (2004).pubmed.ncbi.nlm.nih.gov/15244698
  43. [47]T. Suzuki, M. Yamada, T. Kanamori, H. Hachiya, M. Sugiyama, Relative density-ratio estimation for robust distribution comparison, in Proceedings of the 25th International Conference on Machine Learning (2008), pp. 1243–1250.
  44. [48]A. Singh, B. Póczos, L. Wasserman, Nonparametric estimation of conditional differential entropy, in Advances in Neural Information Processing Systems (2014), pp. 1370–1378.
  45. [49]W. Gao, S. Oh, P. Viswanath, Demystifying fixed k-nearest neighbor information estimators. IEEE Trans. Inf. Theory. 64 (9), 6119–6151 (2017).
  46. [50]M. I. Belghazi, et al., Mutual Information Neural Estimation, in Proceedings of the 35th International Conference on Machine Learning (2018), pp. 531–540.
  47. [51]W. H. Press, S. A. Teukolsky, W. T. Vetterling, B. P. Flannery, Numerical Recipes: The Art of Scientific Computing (Cambridge University Press) (2007).
  48. [52]T. M. Cover, P. E. Hart, Nearest Neighbor Pattern Classification. IEEE Trans. Inf. Theory. 13 (1), 21–27 (1967).DOI
  49. [55]S. A. Dixit, A. Gupta, H. Choudhary, T. Prabhakaran, Generalized scaling and model for friction in wall turbulence. Phys. Rev. Lett. 132 (1), 014001 (2024).DOI
  50. [56]S. A. Dixit, A. Gupta, H. Choudhary, T. Prabhakaran, Universal scaling of mean skin friction in turbulent boundary layers and fully developed pipe and channel flows. J. Fluid Mech. 943, A43 (2022).
  51. [57]M. Thring, Magnetohydrodynamic Power Generation. Nature 208 (5014), 966–967 (1965).
  52. [58]A. L. Hammond, Magnetohydrodynamic Power: More Efficient Use of Coal. Science 178 (4059), 386–387 (1972).pubmed.ncbi.nlm.nih.gov/17815358
  53. [59]A. Glaws, P. G. Constantine, J. N. Shadid, T. M. Wildey, Dimension reduction in magnetohydrodynamics power generation models: Dimensional analysis and active subspaces. Stat. Anal. Data Min. 10 (5), 312–325 (2017).DOI
  54. [60]D. D. Gu, W. Meiners, K. Wissenbach, R. Poprawe, Laser additive manufacturing of metallic components: materials, processes and mechanisms. International materials reviews 57 (3), 133–164 (2012).DOI
  55. [61]Y. Yuan, A. Lozano-Durán, Dimensionless learning based on information (2025), https://doi.org/10.5281/zenodo.17080657.

Paper details

vlnc:W_jv3tztcp
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents