Overview
A real scientific system may contain many interacting components that are difficult to investigate directly. Mathematical modelling simplifies the system into a structured representation while preserving the relationships important to the research question. The resulting model can then be converted into a computational procedure. For example, the movement of a planet can be represented using gravitational equations, the spread of heat using a differential equation, and population growth using a mathematical growth model. The computer uses numerical methods to investigate these mathematical descriptions.
Modelling Process
| Stage | Explanation |
|---|---|
Understand the real system | Study the phenomenon, relevant scientific laws, available observations, and the mechanisms believed to control the system. |
Define the modelling objective | Determine whether the model should explain behavior, estimate parameters, predict future states, compare scenarios, or support optimization. |
Identify important variables | Select the quantities needed to represent the system and distinguish changing state variables from fixed or estimated parameters. |
Establish assumptions | Simplify the real system by identifying processes that can reasonably be ignored or approximated for the chosen purpose. |
Select mathematical relationships | Use scientific laws, empirical relationships, conservation equations, probability models, or other mathematical structures to connect variables. |
Construct the mathematical model | Write the equations, functions, constraints, and conditions that describe the system. |
Analyze mathematical properties | Examine whether solutions exist, whether they are unique, and whether the model behaves consistently with scientific expectations. |
Choose a computational method | Select numerical algorithms and computational representations suitable for solving or simulating the model. |
Implement and simulate | Translate the mathematical model into software and execute it using appropriate parameters and input data. |
Calibrate and validate | Estimate uncertain parameters where appropriate and compare model behavior with independent observations or experimental evidence. |
Refine the model | Modify assumptions, parameters, equations, or model structure when evidence reveals important limitations. |
Types Of Mathematical Models
| Type | Explanation | Example |
|---|---|---|
Deterministic models | Produce the same output when supplied with the same inputs, parameters, and initial conditions. | An ideal projectile-motion model with fixed initial conditions and constant gravitational acceleration. |
Stochastic models | Include random variables or probabilistic processes, so repeated simulations may produce different outcomes. | A model of random population fluctuations or radioactive decay. |
Static models | Describe relationships at a particular state or time without explicitly modelling temporal evolution. | A linear model estimating structural deformation under a fixed load. |
Dynamic models | Describe how system variables change over time. | A differential-equation model of population growth. |
Linear models | Use linear relationships between relevant variables within the mathematical formulation. | A linear system of equations describing certain simplified electrical networks. |
Nonlinear models | Include nonlinear relationships that may produce complex responses, multiple equilibria, or sensitive dependence on initial conditions. | Many fluid-flow, chemical reaction, and ecological models. |
Continuous models | Represent variables as continuous functions of time, space, or other quantities. | A continuous heat-diffusion model. |
Discrete models | Represent states or changes using discrete values, events, or time steps. | A discrete population model in which population size changes once per generation. |
Mechanistic models | Represent behavior using understood scientific mechanisms or governing principles. | A physical model based on conservation of mass and momentum. |
Empirical models | Describe relationships inferred from observations or experimental measurements. | A fitted equation describing how a measured material property changes with temperature. |
Mathematical Representations
| Representation | Use | Example |
|---|---|---|
Algebraic equations | Represent relationships among variables that do not necessarily involve derivatives. | Ohm's law, V = IR. |
Ordinary differential equations | Describe systems whose states change with one independent variable, commonly time. | A simple population-growth equation. |
Partial differential equations | Describe systems whose variables depend on multiple independent variables such as space and time. | The heat equation and wave equation. |
Matrix equations | Represent systems of linear relationships and large computational problems. | Ax = b. |
Probability distributions | Represent uncertainty and random variation. | A normal distribution used in a suitable measurement-error model. |
Optimization formulations | Represent problems involving the minimization or maximization of a defined objective. | Finding a material design that minimizes weight while satisfying strength constraints. |
Fundamental Model Examples
| Model | Equation | Meaning | Variables | Assumptions | Limitation |
|---|---|---|---|---|---|
Exponential population growth | dN/dt = rN | Describes idealized population growth when the growth rate is proportional to the current population size. | N Population size, measured as a count of individuals. T Time, measured in an appropriate time unit. R Per-capita growth-rate parameter, with units of inverse time. |
| Real populations often experience resource limitations, competition, migration, disease, and environmental variation. |
Newton's second law | F = ma | Relates the net force acting on an object to its mass and acceleration in the classical regime. | F Net force, measured in newtons (N). M Mass, measured in kilograms (kg). A Acceleration, measured in metres per second squared (m/s²). |
| The simple equation does not by itself specify every force or interaction in a physical system. |
One-dimensional heat equation | ∂T/∂t = α ∂²T/∂x² | Describes transient heat diffusion in a homogeneous material under a simplified one-dimensional model. | T Temperature, measured in kelvin (K) or degrees Celsius when used consistently for temperature differences. T Time, measured in seconds (s). X Spatial coordinate, measured in metres (m). Α Thermal diffusivity, measured in square metres per second (m²/s). |
| A real system may require multidimensional heat transfer, variable material properties, internal heat sources, or more complex boundary conditions. |
Model Quality
Adequacy
A model is adequate when it represents the important mechanisms needed to answer the stated scientific question.
Simplicity
A useful model should avoid unnecessary complexity while preserving the behavior relevant to its objective.
Consistency
The model should respect relevant physical laws, dimensions, units, and known scientific constraints.
Identifiability
The available data should contain sufficient information to estimate the parameters that the model attempts to determine.
Predictive Performance
Predictions should be evaluated against independent observations or suitable reference evidence.
Robustness
The model should be examined to determine how sensitive its conclusions are to plausible changes in parameters and assumptions.
Calibration And Validation
Calibration
The process of estimating or adjusting model parameters so that model outputs agree appropriately with selected observations or reference data.
Validation
The process of evaluating whether the model is sufficiently representative of the real system for its intended application.
Distinction
A model can fit calibration data well but perform poorly on independent data. Agreement with the data used to adjust the model is not sufficient evidence of general predictive validity.
Worked Example
Problem
Construct a mathematical model for the cooling of hot water in a room.
Assumptions
The water is well mixed.
Room temperature remains approximately constant.
The effective heat-transfer coefficient is constant over the observation period.
The water's effective heat capacity is treated as constant.
Model
dT/dt = -k(T - T_room)
Analytical Solution
T(t) = T_room + (T_initial - T_room)e^(-kt)
Variables
T(t)
Water temperature at time t.
T room
Constant surrounding temperature.
T initial
Initial water temperature.
K
Positive cooling-rate constant, measured in inverse time.
T
Elapsed time.
Computational Investigation
Measure the initial water temperature and room temperature.
Record water temperature at regular time intervals.
Estimate the cooling parameter using suitable data-fitting methods.
Calculate predicted temperatures using the mathematical solution or a numerical differential-equation solver.
Compare predictions with measurements.
Examine residuals and investigate whether the model assumptions remain reasonable.
Interpretation
The model predicts an exponential approach toward room temperature. Deviations between predicted and observed values may indicate measurement error, changing environmental conditions, evaporation, or inadequacy of the simplified model.
Model Limitations
Models simplify reality and cannot represent every feature of a physical system.
Incorrect assumptions can create systematic errors even when the computational implementation is correct.
Some parameters cannot be uniquely estimated from the available observations.
A model may be valid under one range of conditions but unreliable outside that range.
Overfitting can make an empirical model describe existing observations well while producing poor predictions for new conditions.
Mathematical complexity can increase computational cost and make interpretation more difficult.
Practical Learning
Construct an exponential population-growth model and simulate population size over time.
Modify the growth-rate parameter and compare the resulting trajectories.
Develop a simple projectile-motion model and investigate the effect of air resistance as an additional physical process.
Create a cooling model and compare its predictions with measured temperature data.
Use numerical methods to solve a differential equation and compare the result with an analytical solution when available.
Perform a sensitivity analysis to determine which parameters have the greatest effect on model outputs.
Learning Objectives
Explain the role of mathematical models in scientific computing.
Distinguish deterministic, stochastic, static, dynamic, linear, and nonlinear models.
Identify suitable mathematical representations for scientific systems.
Construct a basic mathematical model from a scientific question.
Understand model assumptions, calibration, validation, and limitations.
Interpret how changes in parameters affect computational predictions.
Assessment Questions
What is mathematical modelling and why is it important in scientific computing?
How does a deterministic model differ from a stochastic model?
What is the difference between an ordinary and a partial differential equation?
Why are assumptions necessary when constructing a mathematical model?
How do calibration and validation differ?
Why can a model that fits existing data still produce unreliable predictions?