IB Math AI: Mathematical Modelling

Mathematical modelling is the engine of the IB Math Applications and Interpretation course, transforming abstract mathematics into a powerful tool for understanding and shaping the real world. It is the disciplined process of using mathematics to simulate, analyze, and predict the behavior of complex systems, from population dynamics to economic trends. Mastering this skill not only prepares you for exam success but also equips you with a versatile problem-solving mindset applicable to virtually any field.

The Modelling Cycle: A Structured Approach

Mathematical modelling is not a single calculation but an iterative process known as the modelling cycle. This framework provides a reliable roadmap for tackling open-ended, real-world problems. The cycle typically involves four key phases, though refinements often lead to multiple revolutions.

First, you must identify and specify the real-world problem. This means moving from a vague scenario—like "sales are decreasing"—to a precise question: "What will our quarterly revenue be six months from now?" Next, you make simplifying assumptions to make the problem mathematically tractable. You might assume marketing spend remains constant, ignore seasonal holidays, or consider a single product line. These assumptions are crucial; they define your model's scope and limitations.

With assumptions in place, you formulate the mathematical model. This is where you translate the problem into equations, functions, or diagrams. You choose a model type—linear, exponential, etc.—based on the behavior you expect. Then, you solve and interpret the model to generate a prediction or insight. Finally, and most importantly, you test the model against real data and evaluate its strengths and limitations. Does it match historical trends? If not, you return to an earlier stage, refine your assumptions, perhaps choose a different model type, and repeat the cycle. This iterative refinement is what makes modelling a scientific and robust practice.

Linear Models: The Foundation of Constant Change

Linear models are your first and most intuitive tool, representing situations where the rate of change is constant. They follow the general form $y=mx+c$, where $m$ is the constant rate of change (slope) and $c$ is the initial value (y-intercept). The core assumption here is that for every unit increase in the independent variable $x$, the dependent variable $y$ changes by a fixed amount $m$.

A classic IB example is modelling taxi fare: a fixed drop charge plus a constant cost per kilometer. If the drop charge is $€3$ and the cost per km is $€2$, the model is $F=2d+3$, where $F$ is fare and $d$ is distance in km. Linear regression on a scatter plot is the standard technique for deriving a line of best fit from real data. In an exam context, you’ll often be asked to interpret the slope and intercept in context. A common task is to use the model for extrapolation, like predicting cost for a future distance, while always noting the risk of predicting far beyond the data's range.

Exponential Models: Capturing Rapid Growth and Decay

When change occurs at a percentage rate proportional to the current value, you need an exponential model. Its general form is $y=k{a}^x$ or, more commonly in IB, $y=k{e}^{rx}$, where $k$ is the initial value, $a$ is the growth/decay factor, and $r$ is the continuous rate. The hallmark of this model is rapid, accelerating increase or decrease.

Exponential growth models phenomena like uncontrolled population growth or viral social media spread, while exponential decay models radioactive decay or depreciation of a car's value. For instance, a bacterial culture doubling every hour can be modelled by $P=P_0\cdot 2^t$, where $P_0$ is the initial population. A key skill is manipulating the model to find doubling or halving times. The critical limitation to remember is that pure exponential models are often unsustainable in the long term; real-world factors (like limited resources) eventually slow growth, necessitating a more sophisticated model like logistic growth.

Polynomial Models: Flexibility for Trends

Polynomial models, expressed as $y=a_n{x}^n+a_{n-1}{x}^{n-1}+...+a_{1}x+a_0$, offer great flexibility for modelling data with curves, turning points, and general trends. Quadratic models ($n=2$) are particularly common for projectile motion, where height over time follows a parabolic path. Higher-degree polynomials can fit more complex datasets but come with a significant risk.

The strength of a polynomial model is its ability to fit a dataset very closely. However, this is also its greatest weakness: overfitting. A high-degree polynomial may pass through every data point perfectly on your given set but produce wildly unrealistic predictions for values just outside it. The model loses its predictive power. In IB, you'll typically work with quadratic or cubic models. When using your GDC for polynomial regression, always question whether the number of turning points makes sense in the real-world context of the problem.

Sinusoidal Models: The Pattern of Periodicity

For phenomena that repeat in a regular cycle, you turn to sinusoidal models. These are based on the sine or cosine functions, with a general form like $y=a\sin (b(x-c))+d$ or $y=a\cos (b(x-c))+d$. Here, $a$ is the amplitude (half the range of oscillation), the period is given by $\frac{2\pi }{b}$, $c$ is the horizontal phase shift, and $d$ is the vertical shift or midline.

These models are essential for capturing periodic behavior. Classic exam examples include modelling the height of a tide over time, daily temperature fluctuations through the seasons, or the vertical motion of a Ferris wheel seat. The modelling challenge lies in determining the parameters from data. You might identify the maximum and minimum values to find the amplitude and midline, and measure the time between peaks to find the period. A common pitfall is confusing the period with the frequency; remember, period is the time for one full cycle, while frequency (often $b$ in a transformed function) is the number of cycles per unit time.

Common Pitfalls

1. Ignoring Your Own Assumptions: Every model is built on assumptions. The most frequent error is to present a model's prediction as absolute truth without stating its limitations. Correction: Always conclude your analysis by explicitly revisiting your initial assumptions. State, for example, "This linear revenue forecast assumes no changes in market competition, which may not hold true."

1. Misapplying a Model Type: Forcing a linear regression on data that clearly curves upwards is a recipe for inaccurate predictions. Correction: Always visualize your data with a scatter plot first. Examine the pattern: constant change suggests linear, a constant percentage change suggests exponential, and a repeating wave suggests sinusoidal.

1. Overfitting with Complex Polynomials: Using a 5th-degree polynomial to fit 6 data points will give a perfect $R^2$ value but a meaningless model. Correction: Prioritize simplicity and realism. Often, a simpler model that captures the general trend is more powerful for prediction than a complex one that fits historical noise.

1. Misinterpreting Parameters Out of Context: Stating "the slope is 2.5" is insufficient. Correction: Interpret every parameter within the scenario. You must say, "The slope of 2.5 indicates that for every additional hour of study, the predicted exam score increases by 2.5 points."

Summary

• Mathematical modelling is an iterative modelling cycle of problem identification, assumption-making, formulation, solution, testing, and refinement.
• Linear models ($y=mx+c$) describe constant-rate changes, exponential models ($y=k{a}^x$) capture proportional growth/decay, polynomial models offer flexible curves for trends, and sinusoidal models ($y=a\sin (b(x-c))+d$) map repeating, periodic behavior.
• The choice of model is guided by the real-world context and the observed pattern in the data, not just mathematical convenience.
• Every model has strengths and limitations defined by its underlying assumptions; a critical evaluation of these is a non-negotiable part of the process.
• Avoid common errors like overfitting, misinterpreting parameters, and presenting model outputs without their contextual caveats.