Introduction
Scientific equations do not always originate from fundamental physical laws. Many are developed by examining experimental measurements, identifying regularities, and constructing mathematical relationships that reproduce observed behaviour. These equations are essential when a complete theoretical description is unavailable, excessively complicated, or unnecessary for the intended application. Empirical and phenomenological approaches allow scientists to quantify natural processes, predict outcomes, interpret experiments, and construct practical models.
Empirical Equations
Definition
An empirical equation is a mathematical relationship obtained primarily by fitting observations or experimental measurements. Its structure and numerical parameters are selected to represent the behaviour found in a particular dataset or experimental domain.
Development
Identify the scientific phenomenon and define the quantities that may influence it.
Design experiments or collect reliable observational measurements.
Determine the independent, dependent, and potentially influential variables.
Plot and examine the measurements to identify trends, patterns, correlations, and possible functional relationships.
Select a suitable mathematical structure, such as a linear, polynomial, exponential, logarithmic, or power-law relationship.
Estimate unknown coefficients using an appropriate fitting method.
Evaluate the fitted relationship using residuals, uncertainty estimates, and independent validation data.
Specify the range of conditions over which the equation is supported by evidence.
Characteristics
Its mathematical structure is strongly influenced by experimental evidence.
Its coefficients are commonly estimated from measured data.
It may describe a relationship accurately without explaining the underlying microscopic mechanism.
Its reliability depends on the quality, coverage, and representativeness of the observations.
Its predictive performance may decrease outside the experimental range.
Phenomenological Equations
Definition
A phenomenological equation represents an observed scientific process using a mathematical description informed by its characteristic behaviour and an approximate understanding of the mechanisms involved. It lies between a purely data-driven relationship and a complete first-principles derivation.
Characteristics
It incorporates experimentally observed behaviour and scientific interpretation.
Its mathematical structure may reflect known mechanisms, symmetries, constraints, or conservation requirements.
It often uses parameters that represent effective or averaged physical properties.
It can describe complicated systems without resolving every microscopic interaction.
Its assumptions and applicability must be stated explicitly.
Relationship To Empirical Equations
The two approaches overlap. A phenomenological equation may contain empirically determined coefficients, while an empirical relationship may be motivated by a physical interpretation. The distinction concerns the extent to which the mathematical structure is informed by an explanatory model rather than being selected mainly for its agreement with observations.
Mathematical Representation
General Form
y = f(x₁, x₂, ..., xₙ; a₁, a₂, ..., aₘ)
Explanation
A measured response y is represented as a function of independent variables x₁ through xₙ and model parameters a₁ through aₘ. The function may be selected from theoretical considerations, experimental patterns, or a combination of both.
Example Power Law
y = axᵇ
Interpretation
In a power-law relationship, a determines the scale of the response and b determines how the response changes with the independent variable. Both parameters may be estimated from experimental data, subject to the assumptions of the model.
Parameter Estimation
Definition
Parameter estimation is the process of determining numerical values for unknown coefficients so that a mathematical equation represents observed measurements as closely as justified by the data and its uncertainty.
Methods
| Name | Description |
|---|---|
Linear regression | Estimates coefficients in linear relationships by minimizing a specified measure of disagreement between observations and model predictions. |
Nonlinear least squares | Estimates parameters in models whose dependence on the parameters is nonlinear. |
Maximum likelihood estimation | Selects parameter values that make the observed data most plausible under an assumed statistical model. |
Bayesian parameter estimation | Combines observed data with prior information to obtain a probability distribution for uncertain parameters. |
Important Distinction
A good numerical fit does not necessarily establish that an equation represents the true mechanism of a phenomenon. Different mathematical models may fit the same dataset while producing different predictions outside the observed range.
Model Evaluation
Residuals
A residual is the difference between an observed value and the corresponding model prediction. Examining residual patterns can reveal systematic errors, missing variables, unsuitable functional forms, or changes in experimental conditions.
Goodness Of Fit
Measures such as the coefficient of determination, root-mean-square error, and likelihood-based criteria can help assess agreement between a model and observations. No single statistic establishes scientific validity.
Validation
A fitted equation should be evaluated using independent measurements or carefully separated validation data whenever possible.
Uncertainty
Measurement uncertainty, parameter uncertainty, model assumptions, and uncertainty in future input conditions all influence the reliability of predictions.
Extrapolation And Interpolation
Interpolation
Interpolation estimates behaviour within the range covered by available observations. It is generally more defensible when the data adequately represent the relevant conditions and the fitted relationship behaves appropriately.
Extrapolation
Extrapolation predicts behaviour beyond the observed range. It requires additional caution because the relationship may change when new mechanisms, physical limits, phase transitions, or environmental conditions become important.
Limitations
An empirical equation may reproduce observations without identifying their underlying causes.
A model fitted to a narrow range of conditions may not remain reliable under substantially different conditions.
Correlations between variables do not automatically establish causal relationships.
Overly flexible equations may fit random fluctuations rather than meaningful scientific patterns.
Unmeasured variables, systematic experimental errors, and changes in measurement procedures can bias the estimated relationship.
A phenomenological model may simplify important microscopic or spatial processes.
Parameter values may lose physical meaning if the equation is used outside the conditions under which it was developed.
Common Mistakes
Treating a strong statistical correlation as proof of a physical mechanism.
Assuming that every fitted coefficient represents a fundamental material or physical constant.
Using an equation beyond its validated temperature, pressure, concentration, spatial, or temporal range.
Ignoring measurement uncertainty when estimating parameters.
Selecting a complicated equation solely because it produces a smaller fitting error.
Confusing empirical equations with first-principles laws.
Applications
Estimating material properties from laboratory measurements.
Developing engineering correlations for heat and mass transfer.
Describing fluid behaviour in experimentally investigated operating ranges.
Representing biological growth and population patterns.
Constructing atmospheric and environmental prediction relationships.
Developing chemical reaction-rate correlations.
Representing complex systems when a complete microscopic theory is computationally impractical.
Worked Example
Problem
A laboratory experiment measures the displacement of an object at several times. The observations suggest that displacement increases approximately linearly with time over the measured interval.
Model
s = at + b
Interpretation
The coefficients a and b are estimated from the measured displacement-time data. If the experimental conditions support approximately constant velocity, a can be interpreted as an estimated velocity and b as an initial displacement.
Reasoning
The equation is empirical because its coefficients are determined from measurements. Its physical interpretation is strengthened if independent evidence supports approximately constant velocity.
Conclusion
The fitted equation may be useful within the measured interval, but its validity should not automatically be extended to periods when acceleration, frictional changes, or other influences become significant.
Selection Guidelines
Use an empirical equation when reliable observations are available but the underlying mechanism is insufficiently understood.
Use a phenomenological equation when observed behaviour can be represented more meaningfully by incorporating established physical principles or approximate mechanisms.
Prefer a first-principles model when the governing laws are sufficiently understood and the required information is available.
Compare alternative models using scientific plausibility, predictive performance, uncertainty, and applicability rather than fitting accuracy alone.