Applications and Interpretation · SL and HL

IB Maths AI course map: six threads across the five topics

Applications and Interpretation is usually taught chapter by chapter, but its questions rarely stay in one chapter. A regression line turns into a revenue function, a savings plan turns into a logarithm, a map of hospitals turns into straight-line equations. Below, the course is regrouped into six threads, the ideas that keep coming back, so you can practise the joins that papers test.

By , former Head of Maths with IB examining experience (independent) · Published 15 November 2026

Go to the six threads Talk prompts for teachers

What each AI topic is really about

AI shares its five topic names with AA, but not its focus. Our one-line summary of each is below (our words, not the guide's), with the larger pieces HL adds.

TopicIn one lineLarger HL pieces
1. Number and algebraWorking with real quantities: their size, their error, and money that grows period by period.Laws of logarithms, complex numbers, matrices, eigenvalues and eigenvectors
2. FunctionsPicking the shape of function that fits a situation, then judging whether it still fits.Composite and inverse functions, transformations, logistic and other models, straightening data with logs
3. Geometry and trigonometryMeasuring real objects and spaces, and working out which place is nearest to which.Radians, vectors, matrix transformations, graph theory (trees, routes and tours)
4. Statistics and probabilityCollecting data fairly, describing it, and testing a claim against it.Confidence intervals, Poisson, combining random variables, the central limit theorem, Markov chains
5. CalculusHow fast something changes, how much has built up, and where the best value is.Related rates, volumes of revolution, kinematics, differential equations, slope fields, Euler's method, phase portraits

Point-by-point lists: the AI SL and AI HL syllabus checklists. Taking Analysis and Approaches? It has its own AA course map.

Where each thread shows up

Filled dot: the thread already runs through that topic at SL. Open dot: it reaches that topic only in HL content.

The six threads

For each thread: the route through our topic pages (dashed stops are HL), one question that needs more than one topic, and one open task. Have a go first, then open the working. All answers were checked by hand.

1. The modelling loop

Most of AI follows one loop: choose a type of function, fit it to data, use it to answer a question, then ask where it stops being believable. The loop starts with straight lines and regression at SL and ends, at HL, with logistic models and differential equations.

Mixed question. From prices between $2 and $4, a café finds the regression line n=400−75p, where n is cups sold per day at a price of $p. (a) Write the daily revenue R as a function of p. (b) Find the price that gives the most revenue, and that revenue. (c) What does the model predict at p=6, and should you trust it?

Working

(a) R=p(400−75p)=400p−75p2.

(b) R′(p)=400−150p=0 gives p=83≈2.67. Then n=400−200=200 cups and R=83×200≈$533.33. R″(p)=−150<0, so this is a maximum.

(c) n=400−450=−50 cups, which is impossible. The line was fitted for 2≤p≤4; using it at p=6 is extrapolation.

Investigate this. Each cup costs the café $c to make, so the daily profit is (p−c)(400−75p). Try c=0.80, then other costs. How does the best price move when the cost goes up?

What you should find

The best price is 83+c2: for c=0.80 it is about $3.07, with 170 cups sold. It always sits halfway between the cost and 163, the price where sales fall to zero, so every extra dollar of cost raises the best price by only 50 cents.

2. Growth, decay and money

A geometric sequence, compound interest and an exponential model are one idea written three ways: multiply by the same factor each step. Logarithms answer "how long until…?". At HL, letting the steps shrink to nothing gives continuous growth and the equation dAdt=rA.

Mixed question. $5000 is invested at 3.6% a year, compounded monthly. (a) Find its value after 4 years. (b) After how many complete months is it first worth more than $6000? (c) HL: compare part (a) with 3.6% compounded continuously.

Working

(a) The monthly rate is 0.036÷12=0.003, so the value is 5000×1.00348≈$5773.18.

(b) Solve 1.003n>1.2: n>ln⁡1.2ln⁡1.003≈60.9, so 61 months. Check: after 60 months it is about $5984.47, after 61 about $6002.43.

(c) 5000e0.036×4=5000e0.144≈$5774.42, only about $1.24 more. Monthly compounding is already very close to continuous.

Investigate this. A common shortcut says money growing at r% a year doubles in about 72r years. Compare it with the exact doubling time for r=2,6,12 and 24. Where is the shortcut best, and why does it fail for large r?

What you should find

The exact time is ln⁡2ln⁡(1+r/100): about 35.0, 11.9, 6.1 and 3.2 years, against 36, 12, 6 and 3. The shortcut is best near 8% (exact 9.01, shortcut 9). It relies on ln⁡(1+x)≈x, which gets worse as x grows; 72 rather than 69.3 (from ln⁡2≈0.693) is a nudge that suits rates near 8%.

3. Data to decisions

Statistics in AI is a chain: how the data was collected, how it is summarised, which test fits it, and what the result does and does not allow you to say. HL adds confidence intervals, more tests and ways to straighten curved data with logarithms before fitting a line.

Mixed question. 200 students chose how to revise for a test. Of the 100 who used past papers, 70 passed; of the 100 who reread notes, 55 passed. (a) Test at the 5% level whether passing is independent of revision method. (b) A classmate says this proves past papers cause passes. Give one reason the data cannot show that.

Working

(a) H0: passing is independent of method. Expected counts are 62.5 passes and 37.5 fails in each group (125 and 75 shared equally). χ2=2×7.5262.5+2×7.5237.5=1.8+3.0=4.8, with 1 degree of freedom and p≈0.028. Since 0.028<0.05 (or 4.8>3.841), reject H0: there is evidence of an association.

(b) The students chose their own method, so something else, such as how much time each student put in, could explain both the choice and the result. A test of independence finds association, not cause.

Investigate this. Keep the pass rates at 70% and 55% but change the group size to 50, then 200. What happens to χ2? Roughly how big must each group be before the same gap counts as significant?

What you should find

With the rates fixed, χ2=0.048g for groups of size g: 2.4 for 50 (not significant), 9.6 for 200. You need 0.048g>3.841, so groups of a little over 80. Whether a gap is "significant" depends on the sample size as well as on the gap.

4. Estimate, then check with technology

The GDC solves equations, fits curves and integrates, but it cannot tell you whether an answer is sensible. AI keeps asking for an estimate, a percentage error or a numerical method alongside the exact value. At HL the same habit runs through Euler's method for differential equations.

Mixed question. A river is 12 m wide. Its depth, measured every 2 m from one bank, is 0, 1.2, 2.0, 2.4, 2.1, 1.3 and 0 m. (a) Use the trapezoidal rule to estimate the area of the cross-section. (b) A student models the depth as d(x)=x(12−x)15. Find the area under the model exactly. (c) Taking the model's value as exact, find the percentage error in (a).

Working

(a) h=2: area ≈22[0+0+2(1.2+2.0+2.4+2.1+1.3)]=18.0 m2.

(b) ∫01212x−x215dx=115[6x2−x33]012=864−57615=19.2 m2.

(c) |18.0−19.2|19.2×100%=6.25%.

Investigate this. Use the trapezoidal rule on the model d(x) itself with 2, 3, 6 and 12 strips. Is it always too small? What happens to the error each time the strip width halves?

What you should find

The estimates are 14.4, about 17.07, about 18.67 and about 19.07: always too small, because the curve bends downwards and every straight top cuts under it. The error is 2h215 for strip width h, so halving the width divides the error by 4.

5. Nearest, networks and routes

Voronoi diagrams split a map by "which site is nearest?", using perpendicular bisectors from coordinate geometry. HL graph theory asks the next questions: how to connect places as cheaply as possible, and how to visit them all. Adjacency matrices bring in topic 1.

Mixed question. Ambulance stations stand at A(0,0), B(8,0) and C(0,6), in km. (a) Show that the edges of the Voronoi diagram meet at (4,3), and that this point is 5 km from every station. (b) Find the equation of the edge between the cells of B and C. (c) A depot must go inside 0≤x≤8, 0≤y≤6, as far as possible from its nearest station. Show that (6.25,6) beats both (4,3) and the corner (8,6).

Working

(a) The bisector of AB is x=4 and the bisector of AC is y=3; they meet at (4,3). Its distance to each station is 42+32=5 km.

(b) BC has midpoint (4,3) and gradient −34, so the bisector has gradient 43: y=43x−73, or 4x−3y=7.

(c) (4,3) is 5 km from its nearest station. (8,6) is 6 km from B (8 from C, 10 from A). (6.25,6) lies on the edge in (b), so it is equally far from B and C: 1.752+62=6.25 km, and about 8.66 km from A. So 6.25 km beats 6 and 5. The best point is always a Voronoi vertex, a place where an edge meets the boundary, or a corner; checking all of them shows (6.25,6) is the best.

Investigate this. In a network where every pair of n towns is joined by a road, count the roads and the different round trips that visit every town once. Try n=5, 10 and 20. What does this say about solving a delivery route by trying every option?

What you should find

There are n(n−1)2 roads and (n−1)!2 round trips: 10 and 12 for 5 towns, 45 and 181 440 for 10, and 190 and about 6.1×1016 for 20. Checking every trip is hopeless, which is why HL uses methods that give an upper and a lower bound instead.

6. States and transitions

A tree diagram follows one step at a time. When the same step repeats (weather, brand loyalty, a board game), HL puts the probabilities in a transition matrix, raises it to a power for later steps, and uses eigenvalues for the long run. The same matrix ideas solve coupled differential equations.

Mixed question. If today is sunny, tomorrow is sunny with probability 0.8. If today is rainy, tomorrow is sunny with probability 0.4. (a) Today is sunny. Use a tree diagram to find the probability of rain in two days. (b) HL: write the transition matrix and check (a) with its square. (c) HL: find the long-run proportion of sunny days.

Working

(a) Sunny then rain: 0.8×0.2=0.16. Rain then rain: 0.2×0.6=0.12. Total 0.28.

(b) With columns "from" and rows "to" (sunny first), T=(0.80.40.20.6). The bottom-left entry of T2 is 0.2×0.8+0.6×0.2=0.28, matching (a).

(c) In the long run s=0.8s+0.4(1−s), so 0.6s=0.4 and s=23: two days in three are sunny.

Investigate this. Replace 0.8 and 0.4 with p and q. Find the long-run share of sunny days in terms of p and q, and the eigenvalues of the matrix. How quickly does the chain "forget" today's weather?

What you should find

The share is q1−p+q (which gives 23 above). The eigenvalues are 1 and p−q. The gap between tomorrow's probability and the long-run value shrinks by the factor p−q each day: here 0.4, so from a sunny start the chance of sun goes 0.8, 0.72, 0.688 and so on towards 23. If p=q, today tells you nothing about tomorrow.

Using the map to revise

For teachers: talk prompts

Six short discussion starters, one per thread, each in a different talk format. Give pairs a minute or two, then take two or three answers before showing anything. Project them one at a time with the timer, or print them as cards.

  1. Estimate first · 1 min

    $1000 grows at 9% a year. Without a calculator, will it double in about 5, 8 or 12 years? Commit to one before anyone works it out.

    Start with: "I picked … because each year it grows by …"

    What to listen for

    About 8 years: 1.098≈1.99. Listen for students who notice the growth itself grows; the 72 shortcut gives 72÷9=8.

    From thread 2

  2. Which one doesn't belong? · 2 min

    Four models for cups sold at price p: (A) n=400−75p, (B) n=520×0.7p, (C) n=900p, (D) n=400−75p2. Find a reason why each one could be the odd one out.

    Start with: "… is the only one where …"

    What to listen for

    A: the only constant drop per dollar. B: the only constant percentage drop. C: the only one where revenue pn never changes (always 900). D: the only one that falls faster as the price rises. B and C never reach zero.

    From thread 1

  3. Same and different · 2 min

    Two surveys both find that 70% of the past-papers group passed, against 55% of the notes group. One has 100 students per group and χ2=4.8; the other has 50 per group and χ2=2.4. What is the same? What is different?

    Start with: "Both surveys … but only one …"

    What to listen for

    Same: the percentages and the size of the gap. Different: the sample size, and so the conclusion. Only the larger survey is significant at 5% (critical value 3.841).

    From thread 3

  4. Pairs talk · 1–2 min

    The trapezoidal rule gave 18.0 m2 for the river; the model gave 19.2 m2. Before any calculation, could you have predicted which would be bigger? Explain it to your partner using a sketch.

    Start with: "The straight tops of the trapeziums …"

    What to listen for

    The curve bends downwards, so each straight top sits under it and the rule underestimates. Push for the sketch rather than the word "concave".

    From thread 4

  5. Stand and talk · 2 min

    Stand up and find someone you haven't worked with today. A van must visit 10 shops and return. Would you try every possible route? Estimate how many there are.

    Start with: "From the first shop there are … choices, then …"

    What to listen for

    9!2=181440 round trips. Listen for the multiplying choices (9, 8, 7…) and for why we halve: each tour can be driven both ways.

    From thread 5

  6. Estimate first · 1 min

    After a sunny day, 80% of days are sunny; after a rainy day, 40% are. Over a whole year, roughly what share of days are sunny: a half, two thirds or four fifths?

    Start with: "I think it's more than … because …"

    What to listen for

    Two thirds. Good reasons: it must lie between 40% and 80%, and sunny spells last longer than rainy ones.

    From thread 6

Where to go next

Common questions

What are the five topics in IB Maths AI?

The same five names as AA: number and algebra, functions, geometry and trigonometry, statistics and probability, and calculus. The content inside them is different: AI spends more time on modelling, data and technology, and HL adds matrices, graph theory, further statistics and differential equations.

Is AI just AA with a calculator?

No. AI has whole areas AA does not, such as Voronoi diagrams, financial maths, chi-squared tests and, at HL, networks and Markov chains. The GDC does the heavy arithmetic, so the marks go to choosing a method, setting it up and saying what the answer means.

Is this map the official AI syllabus?

No. The five topic names are the IB's; the six threads, the questions and the talk prompts are our own way of connecting the course. For every syllabus point, use our AI SL and AI HL syllabus checklists.

IB Math Revision is independent and is not affiliated with, endorsed by or connected to the International Baccalaureate Organization. This work/product/service has been developed independently from and is not endorsed by the International Baccalaureate Organization. International Baccalaureate, Baccalauréat International, Bachillerato Internacional and IB are registered trademarks owned by the International Baccalaureate Organization.