EDUKATE · THE GOLD STANDARD
Find your next learning step
Choose the task you need: turn a situation into mathematics, follow a worked model, test an answer, or find a specialist branch. The examples are deliberately small enough to check by hand.
Full chapter index · Diagnose the difficulty · Specialist reading library · Mathematics guide
Mathematical modelling turns a question about a situation into a mathematical representation that can be explored, tested and revised. The representation might be an equation, a table, a graph, a probability distribution, a network or a computational simulation. The useful form depends on the question.
The gold standard is not the most complicated equation. It is a clear connection between purpose, assumptions, calculation, evidence and interpretation. A model can be solved perfectly and still describe the wrong situation. It can also be deliberately simple and give exactly the insight a reader needs.
This M–N root guide connects the existing eduKate mathematical-modelling ecosystem. It brings together foundational mathematics, modelling methods, worked examples and specialist reading routes. Every numerical example here is invented for teaching; none is a forecast or an operating instruction.
The wider purpose is education that transfers into life: using mathematics to understand what changes, what stays the same and what information is still missing. Real engineering, health, financial and safety-critical decisions require appropriate domain expertise, evidence and review.
Full chapter index
Frame the situation · Chapters 1–4
Build a working model · Chapters 5–9
Test the answer · Chapters 10–14
Use and connect · Chapters 15–19
Check and transfer · Chapters 20–23
A modelling question needs a purpose. “Model this tank” is vague. “Under these assumed flow rates, how long until the water reaches this level?” identifies an output, a time horizon and a set of conditions. A different question about mixing or pressure would need a different representation.
Write the intended use in one sentence before choosing a formula. State who needs the answer, which quantity matters and how accurate the answer needs to be. This helps prevent a familiar mathematical technique from taking over a question it does not actually answer.
Separate an explanation from a forecast and a decision. Explaining why a quantity tends to rise is not the same as predicting its value next month. Predicting an outcome under a scenario is not the same as deciding which action is acceptable.
For an earlier overview of the full cycle, read How Mathematical Modelling Works. For the broader idea of purposeful simplification, use How Models Work. This root organises the learning routes and develops checkable examples rather than treating every model as the same kind of tool.
A variable represents a quantity that can change within the model. A parameter helps specify a particular version of the model. An output is the quantity or result the user wants. These roles depend on the question; a number treated as fixed in one exercise may need to vary in another.
Define symbols in words. If V is a volume, state its unit and when it is measured. If q is a flow rate, state the direction and whether it is assumed constant. A bare letter is not a complete definition, even when the algebra is familiar.
Distinguish an input from a decision. The weather may be an external input in a transport model, while the number of vehicles assigned may be a decision. Confusing these roles can make a proposed solution depend on controlling something the decision-maker cannot control.
The school-level bridge is From a Real Situation to Variables, Assumptions and Equations. A useful first deliverable is a quantity dictionary: symbol, meaning, unit, allowed values and source. That small record often exposes an ambiguity before it becomes an impressive-looking wrong answer.
An equation must connect compatible quantities. You can add two volumes expressed in compatible units, but you cannot add a volume to a flow rate. A flow rate becomes a volume change only after multiplication by a time interval.
Suppose water enters at 12 litres per minute. Over five minutes at that constant rate, the inflow is 60 litres. Multiplying by five hours while leaving the rate in litres per minute gives the wrong result unless the units are converted. Unit labels carry information that the numerals alone do not.
Dimensional consistency is a necessary check, not proof of physical correctness. Two terms can have the same units while representing the wrong mechanism. The equation still needs to reflect the situation and its assumptions.
Use Units Are Part of the Mathematics and Measurement for this foundation. When an answer looks implausible, inspect the units before adding a more sophisticated method. A model with an unrecorded time unit is difficult to reproduce or evaluate.
An assumption is a deliberate statement about how the situation is represented. It might say that a rate remains constant, a container does not leak, people arrive independently, or a quantity can be treated as continuous. These choices determine what the mathematics means.
Do not list assumptions as decorations after solving. Attach each assumption to a consequence. If flow is constant, volume changes linearly before a boundary is reached. If demand changes with price, a constant-demand assumption may fail when comparing very different prices.
Classify what is omitted. Some omissions may be harmless at the required scale; others may control the answer. A rough walking-time model may ignore individual footsteps but cannot ignore a closed route when predicting arrival time.
A useful question is: which observation would make us revise this assumption? If there is no possible check, the model may still be a scenario exercise, but it should not be presented as an evidence-tested description. Assumptions make limitations inspectable; they do not excuse every failure.
CHAPTER 5 OF 23 · Build a working model
5. Worked model: accumulation from inflow and outflow
Consider a fictional tank containing 100 litres initially. Water enters at a constant 12 litres per minute and leaves at a constant 8 litres per minute. Assume no other gains or losses, and assume those rates remain possible over the interval being studied.
The net change is 12 − 8 = 4 litres per minute. If t is measured in minutes from the starting observation, the model is V(t) = 100 + 4t litres. At t = 10, the predicted volume is 140 litres. The 100 is an initial stock; the 4 is a rate.
This comes from an accounting statement: later amount equals earlier amount plus inflow minus outflow over the interval. That structure is useful well beyond tanks. It can organise inventories, task backlogs or other accumulated quantities, provided the terms are correctly defined.
Check the model at t = 0: it returns the stated 100 litres. Check one minute: it adds 12 and removes 8, leaving 104. These are simple verification checks. They show that the calculation follows the assumptions; they do not show that a real tank’s flow rates are constant.
CHAPTER 6 OF 23 · Build a working model
6. Capacity changes where an equation applies
Now give the fictional tank a capacity of 180 litres. The equation V(t) = 100 + 4t reaches that level when 100 + 4t = 180, so t = 20 minutes. Before that time, the model is compatible with the stated capacity.
At t = 30, blindly applying the same expression gives 220 litres. That is not a prediction of water stored inside a 180-litre tank. It is a signal that the original model no longer describes the storage process beyond its boundary.
Different revised models are possible. If both flows stop when capacity is reached, stored volume remains at 180 under that new operating rule. If inflow and outflow continue unchanged and excess water spills, storage and overflow must be tracked separately. The rule must be stated rather than guessed.
Clipping the answer with a minimum function can describe bounded storage under particular overflow assumptions, but it does not automatically account for everything that happened to the water. A boundary condition changes the question as well as the graph. This is a model-validity lesson, not advice for operating a real tank.
CHAPTER 7 OF 23 · Build a working model
7. Worked model: a repeated update with retention
Suppose a fictional stored quantity starts at x₀ = 50 units. During each step, 80% of the current quantity remains, and then 20 new units are added. The update rule is xₙ₊₁ = 0.8xₙ + 20. Here n counts steps; the order of operations is part of the model.
The first values are x₁ = 60, x₂ = 68 and x₃ = 74.4. Adding 20 before applying the retention factor would produce a different rule, 0.8(xₙ + 20), and a different sequence. Similar wording does not guarantee identical mathematics.
| Step | Calculation | Stored amount |
|---|---|---|
| 0 | Initial condition | 50 units |
| 1 | 0.8 × 50 + 20 | 60 units |
| 2 | 0.8 × 60 + 20 | 68 units |
| 3 | 0.8 × 68 + 20 | 74.4 units |
| 4 | 0.8 × 74.4 + 20 | 79.52 units |
An equilibrium x* satisfies x* = 0.8x* + 20, so 0.2x* = 20 and x* = 100. Starting below 100, this particular sequence approaches 100. Subtracting the equilibrium from both sides gives xₙ₊₁ − 100 = 0.8(xₙ − 100), showing why the gap shrinks geometrically.
The retained fraction is dimensionless. The 20 is an amount added per update, not automatically per hour or per day. To change the length of a step, reconsider both retention and additions. Reusing the same two numbers at a new time scale changes the represented process.
CHAPTER 8 OF 23 · Build a working model
8. Discrete, continuous and stochastic models
A discrete-time model updates at selected steps. A continuous-time model describes change over a continuous time variable. Neither choice is inherently more realistic for every question. The representation should match the processes and observations that matter.
An ordinary differential equation can express a rate of change, while a recurrence expresses a step-to-step update. Moving between them requires care. A numerical step size used to approximate a continuous model is not necessarily a natural event interval in the underlying system.
A deterministic model returns the same result from the same fully specified inputs. A stochastic model includes random quantities or processes. A single simulated stochastic path is one possible outcome under the model, not the whole prediction.
Use Differential Equations and Calculus for continuous change, and Probability for random variation. MIT’s Differential Equations course provides a university-level route into rate-based models and their solutions.
CHAPTER 9 OF 23 · Build a working model
9. Linear change, proportional change and feedback
Adding the same amount each step creates a different pattern from multiplying by the same factor. Starting at 100, adding 10 gives 110, 120 and 130. Multiplying by 1.1 gives 110, 121 and 133.1. Agreement at the first step does not make the models interchangeable.
A percentage change needs a reference amount. In the multiplicative example, each increase is 10% of the current value, not 10% of the original 100. A clear model states which base is used and whether that rule remains plausible over the intended range.
Feedback occurs when the current state influences subsequent change. It can generate growth, damping or other behaviour depending on the rule. The word “feedback” is not an explanation by itself: show which quantity affects which rate and in what direction.
Do not extrapolate a convenient growth rule indefinitely without considering boundaries and mechanisms. An equation that describes a short interval may fail when capacity, behaviour or resources change. The mathematics can remain internally correct while the interpretation becomes unsuitable.
Calibration estimates or adjusts model parameters using information such as observed data. The fit answers a question about agreement under the chosen structure and fitting procedure. It does not establish that the structure is adequate for every intended use.
Imagine using observations from the first ten days to choose a growth rate. Reporting agreement on those same ten days is useful, but it is not an independent test of the next ten. The model has already been shaped by the fitting observations.
Specify a baseline. A complicated model should be compared with an appropriate simpler alternative, not only with an empty page. If yesterday’s value is already a strong short-horizon predictor, the benefit of a new model should be assessed against that relevant comparison.
Follow Statistics and How Model Residuals Work. Residuals are observed values minus corresponding model predictions under the stated convention. Their size and pattern can reveal questions that a single overall score hides.
CHAPTER 11 OF 23 · Test the answer
11. Verification: did we implement the intended mathematics?
Verification checks whether a calculation or implementation follows the specified model. In the tank example, test the initial value and one-step accounting. In the retention example, check that multiplication occurs before addition. These checks can catch mistakes before any real-world comparison.
Use cases with known or easily derived answers. If inflow equals outflow in the constant-rate tank model, storage should remain unchanged before other rules apply. If the retained fraction is zero in the repeated-update model, the next stored amount is just the addition.
Test units, signs, indexing and boundaries. A spreadsheet formula copied down a column can refer to the wrong cell. A computer programme can update one state using a mixture of old and already-updated values. A smooth graph does not reveal every implementation error.
For approximation and computation, use Numerical Analysis and How Mathematical Approximation Works. A numerically accurate solution to an unsuitable model is still unsuitable; verification and validation address different parts of the chain.
CHAPTER 12 OF 23 · Test the answer
12. Validation: does the representation support this use?
Validation evaluates a model against evidence and the requirements of its intended use. Agreement in one range or setting does not establish adequacy everywhere. Define what counts as sufficiently accurate for the quantity and decision being considered.
NIST’s process-modelling guidance on model fit warns that a high R² does not by itself guarantee an adequate fit. It explains the value of examining residuals and checking aspects of the statistical model rather than relying on one summary statistic.
For the fictional tank, compare predicted and observed volumes at relevant times with appropriate measurement information. If the errors increase as the tank fills, investigate the assumption of constant rates rather than merely reporting an average discrepancy.
Validation is not a universal stamp that stays attached to a model after its purpose changes. A model adequate for explaining a trend may be inadequate for deciding an exact intervention time. Describe where it was checked, against what evidence, and what important conditions remain untested.
CHAPTER 13 OF 23 · Test the answer
13. Sensitivity: which assumptions change the answer?
Return to the 180-litre tank. It needs an additional 80 litres from the initial state. With outflow fixed at 8 litres per minute, an inflow of 12 gives a net rate of 4 and a filling time of 20 minutes.
If inflow were 10 instead, the net rate would be 2 and the filling time 40 minutes. If inflow were 14, the net rate would be 6 and the time 80 ÷ 6, approximately 13.3 minutes. These are conditional scenarios, not probabilities.
The relationship t = 80 ÷ (qᵢₙ − 8) is sensitive when the net rate is close to zero. At an inflow of 8, the level does not rise under the assumptions; below 8, it falls until another boundary or rule matters. A percentage change in input does not necessarily produce the same percentage change in output.
Sensitivity analysis asks how outputs respond to changes in inputs or assumptions. It helps identify where better information could matter. It does not tell you how likely each changed input is unless you supply an appropriate probability model.
An uncertain input is not the only reason a prediction can be wrong. Measurement limitations, parameter estimation, omitted mechanisms, numerical approximation and changes in the setting can all matter. The appropriate treatment depends on the model and the intended claim.
NIST’s Calibration and Uncertainty Analysis of Predictions from Computational Models distinguishes uncertainty connected with experimental inputs from inadequacy in the mathematical representation. A precise input cannot repair a missing mechanism.
Keep a scenario range distinct from a confidence or prediction interval. Trying three plausible inflow rates gives three conditional results. It does not establish that the true result has a specified probability of lying between the smallest and largest values.
Continue into How Uncertainty Quantification Works. When communicating a result, state which uncertainty sources were considered, which were omitted and what assumptions support any probability statement. A range without an explanation can create a new kind of false certainty.
CHAPTER 15 OF 23 · Use and connect
15. Identifiability: different explanations can fit
Suppose the fictional tank rises by 40 litres in ten minutes. Under constant rates and no other losses, this identifies a net rate of 4 litres per minute. It does not separately identify inflow and outflow.
The pairs 12 in and 8 out, 10 in and 6 out, or 7 in and 3 out all produce the same net change. A volume record alone cannot distinguish them within this model. Better arithmetic on the same information will not recover a distinction the observations do not contain.
Additional measurements, experimental conditions or justified constraints may help. Measuring one flow separately would supply new information. Assuming one flow is known can also close the calculation, but that assumption must be disclosed.
This is an identifiability lesson: a successful fit can coexist with uncertainty about the internal explanation. Before claiming that a fitted parameter reveals a mechanism, ask whether substantially different parameter combinations or model structures could produce the same observed output.
CHAPTER 16 OF 23 · Use and connect
16. Optimisation needs an objective and constraints
A predictive model describes outcomes under specified inputs. An optimisation model adds a criterion for choosing among feasible decisions. Those are related but different tasks. A good prediction does not decide which trade-off a community or organisation should prefer.
Imagine a fictional transport exercise with 95 passengers and vehicles that each carry 40. If the only requirement is enough seats and each vehicle is indivisible, at least three vehicles are needed. The quotient 2.375 is useful working, not a feasible final vehicle count.
Now add route timing, accessible places or staffing constraints. Three vehicles may no longer be sufficient. The original answer was conditional on a simpler requirement set, not a universal claim about operating the service.
Use Optimisation and Operations Research. A calculated optimum is an optimum for the stated objective and constraints. If important requirements are omitted, mathematical optimality can coexist with a poor real-world decision.
CHAPTER 17 OF 23 · Use and connect
17. Communicate the model so someone can challenge it
A usable modelling report should let another reader reconstruct the argument. State the question, data, assumptions, quantities, equations, calculation method, checks and interpretation. Explain why the result matters and where it should not be used.
Do not present extra decimal places as extra evidence. The hypothetical filling time 13.333333 minutes is mathematically derivable from exact assumed inputs, but that display alone says nothing about the precision of a real prediction. Reporting should match the purpose and input quality.
Distinguish a conditional statement from a forecast: “If these rates remain constant, the model reaches capacity after 20 minutes.” That wording preserves the condition. Removing the first clause changes the claim.
Keep a change record when revising a model. If a later version adds leakage, changes the time step or introduces capacity, name the change and show how it affects the answer. A reader should not have to guess why two equally confident reports disagree.
Mathematical structure transfers, but domain assumptions do not transfer automatically. The same accumulation equation can organise water, inventory or outstanding tasks. The meanings of inflow, outflow, losses and boundaries must be rebuilt for each setting.
In Mechanics, quantities such as force, motion and geometry have specific relationships. In Meteorology, observations and physical processes constrain a forecast. In Ecology, organisms interact with environments and one another. A familiar equation cannot replace those subject foundations.
Ask a domain expert which omissions could change the conclusion. Ask a mathematical reviewer whether the derivation and computation are sound. These are complementary contributions. Neither role should assume that the other has checked the entire chain.
The educational goal is to recognise a reusable structure and then test the translation. “This reminds me of a tank” is a starting analogy. “These terms genuinely account for this quantity under these conditions” is the stronger modelling claim.
For foundations, begin with Mathematics, Algebra, Calculus, Statistics and Probability. Choose the mathematics needed for the question rather than trying to use every technique at once.
For the modelling cycle, use How Mathematical Modelling Works, The Core Aim of Mathematics Mastery: Mathematical Modelling and Mathematical Modelling, Assumptions and Real-World Questions. The conceptual companion is How Model-Based Reasoning Works.
For a school-level progression, follow What Is a Mathematical Model?, From Situation to Variables and Equations, How to Test a Mathematical Model and When a Mathematical Model Fails. These pages form a reading sequence from representation to checking and revision.
For diagnosing model behaviour, read Model Residuals, Model Misspecification and Model Drift. For approximation and uncertainty, continue into Mathematical Approximation, Numerical Analysis and Uncertainty Quantification.
For choosing actions, use Optimisation and Operations Research. For an applied computational route, Digital Twins, State Estimation and Parameter Calibration connects modelling with updated observations. This is a curated map of relevant learning routes, not a claim to list every application in the ecosystem.
CHAPTER 20 OF 23 · Check and transfer
20. Diagnose the difficulty before adding complexity
If a learner cannot start, ask them to describe the desired output and draw a quantity map before selecting an equation. If the algebra is correct but the interpretation is wrong, return to definitions, units and boundaries.
If a learner confuses a stock with a rate, ask what can be measured at one instant and what describes change over an interval. If they apply the tank formula after capacity is reached, require a new operating rule and an account of any excess.
If a learner calls a good fit proof, ask which observations were used to choose the parameters and which were used for a meaningful test. If they call a scenario range a probability interval, ask where the probabilities came from.
Use a six-sentence explanation: the question is this; these quantities matter; these assumptions connect them; this calculation follows; these checks support it; these limits remain. A learner who can complete that explanation has a stronger foundation than one who produces an unexplained formula.
Is mathematical modelling just solving word problems?
No. A structured word problem may already supply the assumptions and relationships. Modelling also includes deciding what to represent, what information to collect and how to test whether the representation is useful.
Does a more complex model always give a better answer?
No. Extra detail can introduce additional assumptions, parameters and implementation risks. Complexity should earn its place by helping with the intended question and available evidence.
What is the difference between verification and validation?
Verification checks whether the implementation follows the stated model. Validation evaluates the representation against evidence for its intended use. Passing one does not automatically establish the other.
Can a model be useful when it is not a perfect copy?
Yes. A model is a purposeful representation. Its adequacy depends on the question, required accuracy, assumptions and evidence, not on reproducing every detail.
Does sensitivity analysis produce probabilities?
Not by itself. It explores how results change when inputs or assumptions change. Probability statements require additional justified probabilistic information.
What should a beginner learn first?
Start with a precise question, a quantity dictionary and units. Work through one small model by hand, check it, and explain its limit before adding more techniques.
Week one: choose a simple fictional situation and write the question in one sentence. Define every quantity and unit. Separate observations from assumptions. Practise explaining why the desired output matters before beginning the arithmetic.
Week two: reproduce the accumulation and repeated-update examples without looking at the solutions. Change one input and predict the direction of the change before calculating. Explain the difference between changing a parameter and changing the model structure.
Week three: make a small table of predictions and invented observations, clearly labelled as fictional. Calculate residuals and ask what pattern would make you reconsider a constant-rate assumption. Do not pass the invented observations off as measurements.
Week four: explain a changed problem to another person. Include a boundary, a sensitivity check and a limitation. The learning standard follows the same foundation-to-transfer progression as the Secondary 1 Mathematics learning guide: build the meaning, practise the method, check the reasoning and apply it to an unfamiliar version.
CHAPTER 23 OF 23 · Check and transfer
23. Primary source routes and the final standard
The examples in this guide are original teaching scenarios with stated assumptions. Their calculations can be checked directly; they are not empirical claims about tanks, transport systems or other organisations. The sources below provide further study, not certification of any real-world application.
SIAM and COMAP’s GAIMME, second edition provides mathematical-modelling education guidance across school and undergraduate levels. Use it to deepen teaching and assessment rather than to search for one compulsory model for every problem.
MIT OpenCourseWare’s Differential Equations develops continuous-time mathematics. NIST’s guidance on model fit supports residual-based scrutiny, while Calibration and Uncertainty Analysis of Predictions from Computational Models explains why input uncertainty and model inadequacy must be distinguished.
The final standard is a transparent chain: a useful question, defined quantities, explicit assumptions, sound mathematics, appropriate evidence and an honest interpretation. Mathematical modelling becomes powerful when the reader can explain both what an answer supports and why a different situation might require a different model.
Contents · Previous chapter · Learning that transfers into life