When I started studying (integral and differential) calculus I came across the phrase "in the limit". I realised this was an important idea and technique to justify the results I was using in calculus. I did study some introductory texts in real analysis but quickly I got bogged down in technical detail about limits and continuity in the result of Bolzano, Cauchy and Weierstrass.
Two "strands" that have started to help me clear the fog of confusion is a concerted effort to understand mathematical proof (see my app on that). Alongside proof I realised I needed to understand the idea of infinity, tackled in the wonderful book, Infinite Powers by Steven Strogatz, that I urge you to buy if you have any interest in calculus and its foundations. The list of topics in this app and their order in this app is my responsibility. However, a couple of examples are taken from Strogatz's book and I hope I have credited them fully. Infinite Powers is broader, deeper and richer than the brief introduction to the topic of the infinite given in this app. Any errors in explanation in this app are all mine.
Although not specifically about the infinite, A Very Short Introduction to Mathematics by Tim Gowers. The book gives short (of course) explanations about the idea of proof and the idea that calculating a limit is a "competition". Gowers is a brilliant mathematician who has managed to produce a work that is accessible to those with basic algebra skills can understand.
After that the field is wide-open for a more intense study of real analysis and maybe even complex analysis.
This app aims to make you aware of the role the infinite plays in large areas of mathematics. Hopefully you will be tempted to spend time investigating these ideas further.
Recommended reading
Infinite Powers — Steven Strogatz
A Very Short Introduction to Mathematics — Tim Gowers
Module 1
The Endless Number Line
Core idea: Integers never end. No matter how large a number you name, you can always add one more.
Infinity is not a number — it describes an unending process.
→ Drag the number line. Press Start Counting and watch what happens.
Drag ↑
n = 1
Will this process ever stop?
A usable definition of mathematically finite is an object that has a limited or countable size. So the positive whole numbers up to six, starting at one: \(\{1, 2, 3, 4, 5, 6\}\) is finite, the weight of a particular vehicle is finite — they can be measured or counted and some number associated with them. An infinite mathematical object is something that can be counted but the counting never ends, or its dimensions continue without limit.
The challenge in this app is to answer the question:
Are there any processes that never stop, that can be associated with an infinite number of steps, that can somehow generate a value that is finite?
Think about it
What is the largest integer?
There is no largest integer. For any integer \(N\), the integer \(N+1\) is larger. The process never terminates — this is what "infinite" means.
Is infinity a number?
No. \(\infty\) is not a real number. Expressions like \(\infty - \infty\) or \(\frac{\infty}{\infty}\) are indeterminate — they have no single value. Infinity describes a process or a limit, not a quantity you can substitute into arithmetic.
Are there different sizes of infinity?
Yes — Cantor showed this in the 19th century. The integers and rationals are "countably infinite" (\(\aleph_0\)), but the real numbers are a strictly larger "uncountably infinite" set. See Module 18 for the full argument.
Module 3
Early Encounters with Infinity
Core idea: Long before formal analysis, mathematicians bumped into infinity in concrete sums.
Some sums grow without bound; others, surprisingly, settle to a finite value.
The same question — "does adding more and more terms eventually produce a finite total?" — lies at the heart of everything that follows.
→ Use the tabs to explore four famous encounters with the infinite.
Add the positive integers one by one: the running total is
\(1, 3, 6, 10, 15, \ldots\) — the triangular numbers.
Each time we include the next integer the sum jumps by a larger amount, so the total grows without bound.
No matter how large a number \(M\) you choose, adding enough integers will eventually surpass \(M\).
S0 = 0
Growing without bound — this sum diverges.
Think about it
Is there a closed formula for the sum 1 + 2 + … + n?
Yes: \(S_n = \frac{n(n+1)}{2}\). Gauss reportedly discovered this as a child by pairing the first and last terms: \(1+n, 2+(n-1), \ldots\) each pair sums to \(n+1\), and there are \(n/2\) such pairs.
Why does this sum diverge?
Because \(S_n = \frac{n(n+1)}{2} \to \infty\) as \(n \to \infty\). The terms themselves \((a_n = n)\) grow, not shrink — and even if they shrank towards zero it wouldn't guarantee convergence (see the Harmonic tab). Here there is no doubt: the partial sums increase without limit.
Now add powers of \(\tfrac{1}{2}\): each new term is half the previous one.
The running total is \(1, 1.5, 1.75, 1.875, \ldots\) — always less than 2,
yet creeping ever closer.
Unlike the triangular sums, this process does not grow without bound:
the total settles towards the finite value 2.
Key formula: For any ratio \(|r| < 1\),
\[\sum_{n=0}^{\infty} r^n = \frac{1}{1-r}\]
Here \(r = \tfrac{1}{2}\), giving \(\dfrac{1}{1-\frac{1}{2}} = 2\).
The condition \(|r| < 1\) is essential — if \(|r| \ge 1\) the terms do not shrink and the sum diverges.
→ Module 6 (Infinite Series) lets you explore the geometric series interactively with adjustable \(a\) and \(r\), step through partial sums, and see what happens at the boundary \(|r|=1\).
Think about it
Why does the sum not grow without bound?
Each successive term is halved. The gap between the partial sum and 2 is \(2 - S_n = \left(\tfrac{1}{2}\right)^n\), which shrinks geometrically to zero. No matter how many terms you add, the sum is always less than 2 — and the remainder after \(n\) terms is exactly \(2-S_n=2\cdot(1/2)^n\).
What if the ratio were greater than 1?
Then the terms themselves would grow, and the sum would diverge to infinity. For example \(1 + 2 + 4 + 8 + \cdots\) has no finite sum. The condition \(|r| < 1\) is essential for the geometric series to converge.
The reciprocals \(1, \tfrac{1}{2}, \tfrac{1}{3}, \tfrac{1}{4},\ldots\) decrease to zero — yet their sum diverges.
What about summing only the odd reciprocals \(1, \tfrac{1}{3}, \tfrac{1}{5}, \tfrac{1}{7},\ldots\)?
Both sums exceed any bound you care to name — but they do so at very different speeds.
Odd harmonic series: also diverges — even more slowly.
Notice how much slower this grows than the full harmonic series: we are only using half the terms. But removing every other term cannot save a divergent series — it still diverges.
Both series plotted together. The harmonic series (navy) grows faster than the odd harmonic (red), but both grow without bound.
50
\(\sum 1/n\) \(\sum 1/(2n-1)\)
Think about it
Why does the harmonic series diverge if 1/n → 0?
Having terms that shrink to zero is necessary for convergence, but not sufficient. Oresme's 14th-century proof groups the terms cleverly: \(1/3 + 1/4 > 1/2\); \(1/5+1/6+1/7+1/8 > 1/2\); and so on. Each block contributes at least \(\tfrac{1}{2}\), so the sum grows by \(\tfrac{1}{2}\) for every doubling of terms — without limit.
Does the odd harmonic series diverge for the same reason?
Yes, by comparison: every odd-reciprocal term \(\tfrac{1}{2n-1}\) satisfies \(\tfrac{1}{2n-1} > \tfrac{1}{2n}\), so the odd harmonic sum exceeds \(\tfrac{1}{2}\sum\tfrac{1}{n}\). Since \(\sum\tfrac{1}{n}\) diverges and the odd sum is larger (term for term), it too must diverge by the Comparison Test.
A circle has radius \(r\) and circumference \(C\).
By slicing it into thin sectors and rearranging them, we can see why the area must equal \(\tfrac{1}{2}Cr\).
Since \(C = 2\pi r\), the area is \(\pi r^2\).
\(\text{Sectors arranged}\Rightarrow\text{rectangle}:\quad \text{Area} = \tfrac{C}{2}\times r = \pi r^2\)
As the number of sectors increases the rearranged shape approaches a perfect rectangle of width \(\tfrac{C}{2}\) and height \(r\). The area of this rectangle — and hence the original circle — is \(\tfrac{C}{2} \cdot r = \pi r \cdot r = \pi r^2\).
Think about it
Why does slicing into more sectors make the shape more rectangular?
Each sector has a curved arc on its outer edge. When we interleave sectors apex-up and apex-down, those arcs form the top and bottom of the rearranged shape. As the number of sectors \(n \to \infty\), each arc becomes shorter and straighter, so the top and bottom edges approach straight lines of total length \(C/2\), and the sides approach vertical lines of height \(r\). In the limit the shape is exactly a rectangle.
Where does π come from?
\(\pi\) is defined as the ratio of the circumference to the diameter: \(\pi = C/(2r)\), so \(C = 2\pi r\). Substituting into the area formula gives \(\text{Area} = \tfrac{C}{2} \cdot r = \pi r \cdot r = \pi r^2\). The argument above shows why the area involves \(C\); the definition of \(\pi\) then brings the familiar formula.
Module 2
Basic Numbers — Calculating with Infinity
Shift from discrete to continuous real values
Up to this point examples have used whole numbers also known as integers. To go further we need to understand the meaning of fractions (rational numbers) and then real numbers. The "Number systems" section of the Proof app, https://maths-apps.crlong.uk/Pure/Proof/Proof.html, goes into a bit more detail about these types of number — but not to the level a professional mathematician would think precise. In this whole area there is a trade off between precision and (lack of) understandability by even a determined reader.
Note: The content of this module is based on an example from the excellent Infinite Powers by Steven Strogatz. This book is highly recommended — it gives a historical view of the development of the idea of infinity. Over several hundred years, the gradually deepening understanding of what infinity means allowed the Integral and Differential Calculus to be placed on a firm mathematical foundation. The scope of this app is simply to introduce the basic ideas and applications of infinity in mathematics.
First let us identify the types of numbers most people are familiar with:
Integers (whole numbers): positive and negative and zero — \(\ldots, -3, -2, -1, 0, 1, 2, 3, \ldots\)
Fractions: also known as rational numbers — e.g. \(\tfrac{1}{2},\ \tfrac{3}{4},\ -\tfrac{7}{3}\)
Irrational numbers: which represent points on the number line that cannot be accurately expressed using fractions of integers — e.g. \(\sqrt{2},\ \pi,\ e\)
This app assumes you know what division of whole numbers is and how to divide by a fraction.
So onto the problem of treating infinity as a number.
Let's take a line 6 cm long and divide it into 10 equal parts. Each part is \(\tfrac{6}{10} = \tfrac{3}{5}\) cm or 0.6 cm long.
So we have 10 parts each of 0.6 cm making up 6 cm, or in pure number terms:
\[10 \times 0.6 = 6\]
Now if we divide the 6 cm into 100 equal length parts we end up with, using pure numbers:
\[100 \times 0.06 = 6\]
So the more parts we have the shorter each part becomes, but the total is a fixed finite number.
Suppose we are allowed to do this without end, so we have an "infinite" number, \(\infty\), of parts each of zero length. The calculation would be:
\[\infty \times 0 = 6\]
If this argument is valid we could start with any length, say 10 cm, and we would end up with:
\[\infty \times 0 = 10\]
The contradiction: Since both expressions equal \(\infty \times 0\), this would imply \(6 = 10\) — which is clearly nonsense!
This example seems rather silly, but similar calculations of multiplying ever larger numbers by ever smaller numbers come up a lot in Calculus. We need to find a way to determine whether we can make sense of the finite calculations when the use of infinity in calculations produces nonsense.
Think about it
Why does ∞ × 0 lead to a contradiction here?
The symbol \(\infty\) is not a fixed number — it describes an unending process. When we write \(\infty \times 0\) we are combining two unending processes (the number of parts growing without bound, and each part shrinking to zero) and naively assuming their product is fixed. The contradiction \(6 = 10\) shows this naive approach fails. The correct tool is the mathematical limit, which carefully analyses how the two processes interact.
Is ∞ × 0 always indeterminate?
Yes — \(\infty \times 0\) is one of the seven classical "indeterminate forms". Its value depends entirely on the rates at which the two quantities grow and shrink. For example \(n \times \tfrac{1}{n} = 1\) for all finite \(n\), so the limit as \(n\to\infty\) is 1. But \(n^2 \times \tfrac{1}{n} = n \to \infty\), while \(n \times \tfrac{1}{n^2} = \tfrac{1}{n} \to 0\). Same "form", three different answers.
Module 4
Approaching Zero
Another approach to the idea of infinity is to ask whether, by taking a succession or sequence of numbers without ever ending, we can get closer and closer to a particular number — the simplest example being zero.
Core idea: A sequence can get arbitrarily close to zero without ever equalling zero.
Different sequences approach zero at very different speeds — which matters enormously for series.
→ Use the tabs below to explore different sequences, or compare them all in the Race tab.
\(a_n = \dfrac{1}{n}\)
1
\(a_n\) = 1.000000
Getting smaller, but never reaching zero.
\(a_n = \dfrac{1}{n^2}\)
1
\(a_n\) = 1.000000
Decreasing faster than 1/n.
\(a_n = r^n\), \(|r|<1\)
0.50
1
\(a_n\) = 0.500000
Geometric decay toward zero.
All three sequences plotted together. Notice how \(1/n^2\) and \(r^n\) pull away from \(1/n\) — yet all three approach zero.
0.70
1/n 1/n² rⁿ
Think about it
Does 1/n ever equal zero?
No. For any finite positive integer \(n\), \(\frac{1}{n}>0\). The sequence approaches zero as \(n\to\infty\) but no term equals zero.
Why does the speed of decay matter?
Because "terms → 0" is not enough to guarantee a series converges. \(\sum 1/n\) diverges even though \(1/n\to0\), while \(\sum 1/n^2\) converges. The rate at which terms shrink determines whether the sum stays finite. See Module 6.
Common mistake
⚠ "Since 1/n gets smaller and smaller, it must eventually reach zero."
This confuses a limit with a value attained. \(\frac{1}{n}\to0\) means we can make \(\frac{1}{n}\) as small as we wish — but for any finite \(n\), \(\frac{1}{n}\) is strictly positive. Only the limit, not any term, equals zero.
Module 8
The 0/0 Problem
Core idea: \(\frac{0}{0}\) is indeterminate. The limit of a ratio of two functions: \(\frac{g(x)}{h(x)}\) as both \(g,h\to0\) can be anything — it depends entirely on how fast each approaches zero.
This is an important topic because in differential calculus you create expressions of the form \(\frac{g(x)}{h(x)}\) when trying to determine the derivative of a function \(f(x)\). Knowing that \(\frac{g(x)}{h(x)}\) approaches a finite value — and knowing what that value is — is often essential in determining the derivative.
\(\displaystyle\lim_{x\to0}\frac{\sin x}{x} = 1\)
x → 0 x = —
ratio = —
Both numerator and denominator → 0, yet the ratio → 1.
Both numerator and denominator → 0, yet the ratio → 0.
Why this matters for calculus: This limit appears directly when differentiating \(\cos x\) from first principles. The derivative is
\[\frac{d}{dx}\cos x = \lim_{h\to0}\frac{\cos(x+h)-\cos x}{h}\]
Expanding \(\cos(x+h)\) and simplifying produces the factor \(\dfrac{1-\cos h}{h}\), which must equal 0 for the result \(-\sin x\) to follow. Without knowing this limit the derivation cannot be completed.
Think about it
Why is 0/0 "indeterminate"?
Because \(\frac{0}{0}\) alone carries no information. As we saw above, limits of the form \(\frac{f}{g}\) with \(f,g\to0\) can equal any value — or be infinite. The symbol \(\frac{0}{0}\) is meaningless without knowing the specific functions.
What other indeterminate forms exist?
The seven classical forms: \(\frac{0}{0},\frac{\infty}{\infty},0\cdot\infty,\infty-\infty,0^0,1^\infty,\infty^0\). Each requires careful analysis — algebraic manipulation or L'Hôpital's rule — to evaluate.
Common mistake
⚠ "Both top and bottom go to zero, so the limit is zero."
Completely wrong. \(\lim_{x\to0}\frac{\sin x}{x}=1\), not zero. You cannot substitute the limits of numerator and denominator separately when both are zero — the answer depends on how they compare. Always check for a factorisation or use L'Hôpital.
Module 5
More Sequences
Notation / definitions: A sequence is a list of numbers whose members usually follow some pattern or can be generated by a "rule".
\(a_n\) is any term in the sequence, where \(a_1\) (or sometimes \(a_0\)) is the first term, \(a_2\) is the next term, then \(a_3\), and so on.
\(a_1, a_2, \ldots, a_7\) is the finite list of terms from \(a_1\) to \(a_7\), and no others.
\(a_3, a_4, a_5, \ldots\) is the list of terms starting at \(a_3\) and continuing for ever.
Core idea: \((a_n)\) is the sequence of numbers \(a_1, a_2, a_3, \ldots\) We can ask 'What are successive values of this sequence \((a_n)\) and are they getting closer to a particular value?' A sequence \(a_1,a_2,a_3,\ldots\) converges if its terms settle towards a finite limit. Otherwise it diverges — either by growing without bound, or by oscillating.
\(a_n = a+(n-1)d\)
2
3
d ≠ 0: diverges.
\(a_n = a\cdot r^{n-1}\)
10
0.5
|r| < 1: converges to 0.
\(a_n = A\cdot(-1)^n\)
8
Oscillates — no limit exists.
\(a_n = \dfrac{\sin n}{n}\)
Oscillates but converges to 0 (squeeze theorem: \(-\frac{1}{n}\le\frac{\sin n}{n}\le\frac{1}{n}\)).
Think about it
Does a bounded sequence always converge?
No. \(a_n=(-1)^n\) is bounded but diverges. However, a bounded monotone sequence always converges — this is the Monotone Convergence Theorem, a cornerstone of analysis.
When does a geometric sequence converge?
\(a_n=ar^{n-1}\) converges iff \(|r|<1\) (then \(a_n\to0\)) or \(a=0\). If \(|r|>1\) and \(a\ne0\) it diverges to \(\pm\infty\); if \(r=-1\) and \(a\ne0\) it oscillates.
Common mistake
⚠ "The sequence (−1)ⁿ is bounded, so it must converge."
Being bounded is necessary for convergence but not sufficient. \((-1)^n\) is bounded by \(\pm1\) yet oscillates forever. You also need the sequence to be eventually monotone (or use some other test) to conclude convergence.
Module 6
Infinite Series
This module assumes you understand sigma notation to define a sum of terms in a sequence, usually called a series.
Core idea: An infinite series \(\sum_{n=1}^\infty a_n\) is defined via partial sums \(S_N=\sum_{n=1}^N a_n\).
The series converges if \(S_N\to L\) as \(N\to\infty\).
→ Use Step Through to add one term at a time and watch the sum build up.
The harmonic series diverges — even though each term \(1/n \to 0\).
It grows without bound, but extraordinarily slowly: it takes over 10,000 terms to exceed 10.
This is the single most important counterexample in series analysis.
→ Module 3 (Early Encounters) has full interactive graphs of both the harmonic series \(\sum 1/n\)
and the odd harmonic series \(\sum 1/(2n-1)\), with sliders and comparison. Module 11 (Series Tests)
shows why the divergence test alone is inconclusive for \(\sum 1/n\).
Think about it
If terms → 0, why does the harmonic series diverge?
Terms going to zero is necessary for convergence but not sufficient. Oresme's 14th-century proof groups the terms: \(\tfrac{1}{3}+\tfrac{1}{4} > \tfrac{1}{2}\); \(\tfrac{1}{5}+\tfrac{1}{6}+\tfrac{1}{7}+\tfrac{1}{8} > \tfrac{1}{2}\); each block of doubling length contributes at least \(\tfrac{1}{2}\), so the sum grows without bound.
Alternating harmonic series CONVERGES (by the Alternating Series Test).
Think about it
If the terms → 0, does the series converge?
Not necessarily. The harmonic series \(\sum 1/n\) is the classic counterexample. Terms → 0 is only a necessary condition; the harmonic series fails the comparison test against \(\sum 1/2, \sum 1/4, \ldots\)
What is the sum \(\sum_{n=0}^\infty(1/2)^n\)?
Using \(\frac{a}{1-r}\) with \(a=1, r=1/2\): the sum is \(\frac{1}{1-1/2}=2\). Infinitely many positive terms summing to a finite value — this is what convergence really means.
Common mistake
⚠ "The terms of the harmonic series go to zero, so it must converge."
This is the single most common error in series. Terms going to zero is necessary but NOT sufficient. The harmonic series diverges despite \(1/n\to0\). Always apply a proper convergence test — never rely on this alone.
Module 7
What is a Limit?
Core idea: We say \(\lim_{n\to\infty}a_n = L\) if we can make \(a_n\) as close to \(L\) as we like by taking \(n\) large enough. The green band below shows the tolerance \(\varepsilon\) around \(L=0\).
→ Drag the \(\varepsilon\) slider. Watch how many red points fall outside the band — and find the threshold \(N\).
0.30
0.30
0.30
Formal definition of a limit
\(\displaystyle\lim_{n\to\infty}a_n = L \iff \forall\,\varepsilon>0,\ \exists\,N:\ n>N\Rightarrow|a_n-L|<\varepsilon\)
Symbol
Meaning
\(a_n\)
An arbitrary term in the sequence \((a_n)\).
\(\lim_{n\to\infty}a_n = L\)
The successive terms in the sequence \((a_n)\) approach \(L\).
\(\iff\)
Is logically the same as.
\(\forall\,\varepsilon>0\)
We take any value that is positive and small, i.e. close to zero — say \(0.01\) or \(0.001\) or \(0.000\ldots01\).
\(\exists\,N:\)
There will be some (probably "large") positive integer \(N\).
\(n>N\)
Every integer bigger than \(N\) — and it is vital that this is true for every integer \(n\).
\(\Rightarrow\)
Results in.
\(|a_n-L|\)
The difference between any and every term in the sequence beyond \(a_N\).
\({}<\varepsilon\)
Is smaller than our \(\varepsilon\).
So in this way we can get our terms as close as we like to the value \(L\) (the limit) even if we cannot actually "reach" the limit.
Another way of thinking about this — a competition:
I have a sequence of numbers I claim has a limit \(L\). You choose a small positive number \(\varepsilon > 0\). If somehow I can show that all values of the sequence \((a_n)\) after a particular value \(a_N\) are closer than your \(\varepsilon\) to my claimed limit \(L\) — i.e. \(|a_n - L| < \varepsilon\) — then \(L\) is the limit of the sequence.
Think about it
Does the limit have to be reached?
No. \(a_n=\frac{1}{n}\) has limit 0 but never equals 0. The limit is the value the sequence approaches, not a value it necessarily attains.
What happens if ε is halved?
For \(a_n=1/n\), halving \(\varepsilon\) doubles \(N\) (since \(N\approx 1/\varepsilon\)). For \(a_n=1/n^2\), \(N\) only grows like \(\sqrt{1/\varepsilon}\) — a much more modest increase. Faster-decaying sequences need smaller \(N\) for the same tolerance.
Common mistake
⚠ "The sequence equals its limit eventually."
Not in general. For \(a_n=1/n\), no term ever equals 0. The definition says the terms get arbitrarily close — not that they arrive. (Some sequences do eventually reach their limit, e.g. a constant sequence, but this is the exception.)
Module 9
Why Convergence Matters
Core idea: Convergent series let us compute exact values of \(\pi\), \(e\), \(\ln 2\) and the standard functions. They underpin calculus, Fourier analysis, and much of modern mathematics.
Taylor/Maclaurin series: If a function is smooth enough, it equals its Taylor series within its radius of convergence. Calculators use finite partial sums to evaluate \(\sin\), \(\cos\), \(e^x\) — see Module 16 for a live demonstration.
Think about it
What is the Basel problem?
Euler (1734) proved \(\sum_{n=1}^\infty\frac{1}{n^2}=\frac{\pi^2}{6}\approx1.6449\). The appearance of \(\pi\) in a sum over integers was astonishing. The proof uses the fact that \(\sin x\) can be written as an infinite product over its zeros.
Why does e appear everywhere?
\(e=\sum_{n=0}^\infty\frac{1}{n!}\approx2.718\). It arises as the base of natural logarithms, in compound interest, differential equations, and probability — because \(e^x\) is the unique function equal to its own derivative.
Module 10
Sequence Convergence Tests
Core idea: There are several important tests and criteria for determining whether a sequence converges. For a sequence \((a_n)\), convergence means there exists a number \(L\) such that
\[a_n \to L \quad \text{as } n\to\infty\]
Unlike series, where you add infinitely many terms, sequences are just lists of numbers, so the tests are somewhat different.
A sequence is monotone if it is always increasing (\(a_{n+1}\ge a_n\)) or always decreasing (\(a_{n+1}\le a_n\)). It is bounded if there exist real numbers \(m\) and \(M\) with \(m\le a_n\le M\) for all \(n\). Both conditions together guarantee convergence — even when the limit cannot be found explicitly.
Note: the theorem guarantees a limit exists; it does not always give its value. The precise theory of bounds belongs to real analysis.
If you can trap a sequence between two simpler sequences that share the same limit, the middle sequence is forced to that limit too. Most useful when \(a_n\) involves \(\sin\), \(\cos\), or oscillating factors multiplied by a term known to vanish.
A sequence is a Cauchy sequence if its terms eventually cluster arbitrarily tightly together — without needing to know the limit in advance. In the real numbers, Cauchy sequences and convergent sequences are the same thing. This is fundamental to real analysis and the construction of \(\mathbb{R}\).
A-level scope: the Cauchy criterion is more of an analysis concept than an A-level tool. It appears here because it underpins why limits work — it defines convergence without reference to a limit value.
Related to the Ratio Test for series, but applied directly to the terms of the sequence. Particularly effective for sequences involving factorials or exponentials.
Sequence \(a_n\)
\(a_{n+1}/a_n\)
\(r\)
Conclusion
\(\tfrac{1}{2^n}\)
\(\tfrac{1}{2}\)
\(\tfrac{1}{2}\)
\(\to 0\)
\(\tfrac{n!}{3^n}\)
\(\tfrac{n+1}{3}\)
\(\infty\)
Diverges
\(\tfrac{3^n}{n!}\)
\(\tfrac{3}{n+1}\)
\(0\)
\(\to 0\)
\(\tfrac{1}{n}\)
\(\tfrac{n}{n+1}\)
\(1\)
Inconclusive
Subsequences
\[(a_n)\to L \;\Longrightarrow\; \text{every subsequence also}\to L\]
The contrapositive is most useful: if you can find two subsequences that approach different limits, the original sequence diverges. A subsequence is formed by selecting an infinite subset of terms in order: \(a_{n_1}, a_{n_2}, a_{n_3},\ldots\)
Sequence
Subsequences
Conclusion
\((-1)^n\)
Even terms \(\to 1\), odd terms \(\to -1\)
Diverges
\(\sin\!\left(\tfrac{n\pi}{2}\right)\)
Takes values \(0,1,0,-1,\ldots\) repeatedly
Diverges
\(\tfrac{1}{n}\)
Every subsequence \(\to 0\)
Converges to \(0\)
Bolzano–Weierstrass: every bounded sequence has at least one convergent subsequence. This guarantees a convergent subsequence exists — but not that the full sequence converges.
A distinction worth keeping clear:
Sequence convergence asks whether the terms approach a fixed number.
Series convergence asks whether the running totals (partial sums) \(\displaystyle S_n = \sum_{k=1}^n a_k\) approach a fixed number.
A series is studied through the sequence of its partial sums — so sequence convergence is foundational.
Bounds
The understanding of bounds, upper bounds, and least upper bounds is fundamental to the Monotone Convergence Theorem, but firmly in the area of real analysis — beyond the scope of what is introduced in this app.
Module 11
Series Convergence Tests
Core idea: A toolkit of tests determines convergence for most series you'll meet. The various tests for the convergence of a series are based on tests performed on one or more terms in the sequence that have been added to make the series.
Divergence Test (nth-term test)
If \(\displaystyle\lim_{n\to\infty}a_n\) ≠ 0, then \(\displaystyle\sum a_n\) diverges.
Warning: If \(a_n\to0\) the test is inconclusive — the series might still diverge (harmonic). This test only proves divergence.
\(\displaystyle\sum_{n=1}^\infty\frac{1}{n^p}\) converges if \(p>1\), diverges if \(p\le1\)
The boundary at p = 1 is sharp. The harmonic series (\(p=1\)) diverges; the Basel series (\(p=2\)) converges. Drag the slider to see the partial sum behaviour change dramatically near \(p=1\).
1.00
S500 ≈ —
Common mistake
⚠ "The ratio test always works."
When \(L=1\) the ratio test is inconclusive — this happens for \(p\)-series, for example. When the ratio test fails, try comparison, the \(p\)-series test, or the alternating series test instead.
Module 13
Limits Applied to Functions
Applying the limit definition to simple examples of functions we are starting to scratch at real analysis.
The three examples below compare a function that is continuous at \(x = 1\) with two
different kinds of discontinuity there — one where the function is not defined at the
point, and one where it is defined but with a value that disagrees with the
limit. In all three cases the limit itself is the same.
Example 1: \(f(x) = x + 1\)
\(f\) is a straight line. It is defined and continuous everywhere.
At \(x = 1\): \(f(1) = 2\). The limit from both sides is equally straightforward:
\[\lim_{x \to 1} f(x) = \lim_{x \to 1}(x+1) = 2\]
f(1) = 2 | limit from left = 2 | limit from right = 2
Example 2: \(g(x) = \dfrac{x^2 - 1}{x - 1}\)
This looks more complicated, but we can factorise the numerator:
So \(g(x) = x + 1\) everywhere except \(x = 1\), where the denominator is zero
and \(g\) is undefined — shown as an open circle on the graph.
Despite this, the limit as \(x \to 1\) exists perfectly well, because we only ask what value
\(g(x)\) approaches, never what it equals at \(x = 1\) itself:
\[\lim_{x \to 1} g(x) = \lim_{x \to 1}(x+1) = 2\]
Notice that \(f(x) = x+1\) and \(g(x) = x+1\) (for \(x \ne 1\)) are identical in value
for every \(x \ne 1\), and both have the same limit \(2\) as \(x \to 1\).
The sole difference is that \(f(1) = 2\) is defined, while \(g(1)\) is not.
This makes \(g\) a textbook example of a removable discontinuity at \(x = 1\).
g(1) undefined | limit from left = 2 | limit from right = 2
Example 3: \(h(x) = \begin{cases}\dfrac{x^2-1}{x-1} & x \ne 1 \\ 1 & x = 1\end{cases}\)
Now we do assign a value at \(x = 1\), but we choose a value that disagrees with
the limit. For \(x \ne 1\) the formula simplifies exactly as in Example 2:
But at \(x = 1\) itself, the piecewise definition gives \(h(1) = 1\). The function is
defined at \(x = 1\) — yet it is still discontinuous there, because
\(h(1) = 1 \ne 2 = \lim_{x\to 1} h(x)\). The graph shows the characteristic picture: a
filled dot at \((1,\,1)\) (the actual value) and an open circle at
\((1,\,2)\) (the limit value the curve approaches).
This is still called a removable discontinuity: the limit exists,
so the discontinuity could be repaired simply by redefining \(h(1) = 2\) — that is, by
replacing the wrong value with the correct (limit) value.
h(1) = 1 | limit from left = 2 | limit from right = 2 | limit ≠ h(1)
Approaching \(x = 1\) from both sides
Move the slider to approach \(x = 1\) from either side. All three functions approach \(2\)
as \(x \to 1\). Slide to the centre (\(x = 1\) exactly) to see the key contrasts:
\(f(1) = 2\) (defined, equals limit); \(g(1)\) undefined; \(h(1) = 1\) (defined but not equal to limit).
0.500
f(x) = x + 1 → limit 2
f(x) = 1.500
g(x) = (x²−1)/(x−1) → limit 2
g(x) = 1.500
h(x): piecewise → limit 2
h(x) = 1.500
Slide toward x = 1 to see both limits.
Property
\(f(x) = x+1\)
\(g(x) = \dfrac{x^2-1}{x-1}\)
\(h(x)\) piecewise
Value at \(x = 1\)
\(f(1) = 2\)
Undefined
\(h(1) = 1\)
\(\lim_{x\to 1^-}\)
\(2\)
\(2\)
\(2\)
\(\lim_{x\to 1^+}\)
\(2\)
\(2\)
\(2\)
Limit equals value?
Yes — continuous at \(x=1\)
No — not defined at \(x=1\)
No — \(h(1)=1\ne 2\)
Same limit as \(f\)?
—
Yes: both limits equal \(2\)
Yes: all three limits equal \(2\)
Defined everywhere?
Yes
No — undefined at \(x = 1\)
Yes — but value disagrees with limit
Discontinuity type
None (continuous)
Removable (point missing)
Removable (wrong value assigned)
The key takeaway: All three functions produce exactly the same output for every
\(x \ne 1\), and all share the same limit \(2\) as \(x \to 1\). Yet at \(x = 1\) they differ completely:
\(f(1) = 2\) (defined and equals the limit — continuous);
\(g(1)\) is undefined (removable discontinuity — missing point);
\(h(1) = 1\) (defined but wrong value — also a removable discontinuity).
The limit describes what a function approaches; it is independent of
what the function does at the point — whether that is "nothing" (as in \(g\)) or
"something incorrect" (as in \(h\)). Both \(g\) and \(h\) have removable discontinuities at
\(x = 1\): in each case the repair is simply to set the value equal to \(2\).
Think about it
Why does cancelling (x − 1) give the limit but not the value at x = 1?
Cancellation is valid for \(x \ne 1\). When we write \(\frac{(x+1)(x-1)}{x-1} = x+1\), this equality holds only when \(x - 1 \ne 0\). The limit process takes \(x\) close to \(1\) but never equal to \(1\), so the cancellation is always legitimate in the limit calculation. The value \(g(1)\) itself remains undefined because the original expression requires dividing by zero there.
What is a removable discontinuity?
A removable discontinuity at a point \(a\) means the limit \(\lim_{x\to a}\) exists, but either the function is undefined at \(a\) or its value there does not equal the limit. It is called "removable" because you can repair the discontinuity simply by (re)defining the function value at \(a\) to equal the limit. For \(g\), define \(g(1) = 2\); for \(h\), redefine \(h(1) = 2\) — in both cases the function becomes continuous at \(x = 1\).
What makes f continuous at x = 1 but not g or h?
Three conditions must hold for continuity at a point \(a\): (1) the function is defined there, (2) the limit exists, and (3) the limit equals the function value. \(f\) satisfies all three: \(f(1)=2\), \(\lim f = 2\), they agree. \(g\) fails condition (1): \(g(1)\) is not defined at all. \(h\) satisfies (1) and (2) but fails (3): \(h(1)=1\) while \(\lim h = 2\), so the value and the limit disagree.
How do g and h differ if both are removable discontinuities?
Both \(g\) and \(h\) fail the continuity test at \(x=1\), but for different reasons. \(g\) has a missing point: the function is simply not defined at \(x=1\), so there is a "hole" in its graph at \((1,2)\). \(h\) has a mislabelled point: the function is defined, but the assigned value \(h(1)=1\) sits in the wrong place. On the graph this appears as a filled dot at \((1,1)\) together with an open circle at \((1,2)\). In both cases the repair is the same: set the value at \(x=1\) equal to \(2\).
Module 14
Bouncing Ball
Core idea: A ball dropped from height \(h\) bouncing to fraction \(r\) of its height each time travels
a total distance that is a geometric series. Even though it bounces infinitely many times, the total distance is finite!
\(\text{Total distance} = h + 2h r + 2h r^2 + \cdots = h\cdot\dfrac{1+r}{1-r}\)
20
0.70
Total distance = — m
Infinitely many bounces, finite total distance.
The ball bounces infinitely many times yet comes to rest in a finite time — because each bounce takes proportionally less time, just like Zeno's paradox.
Think about it
Why does the series converge even though there are infinitely many bounces?
Each successive bounce travels \(r\) times the distance of the previous one, with \(r<1\). So the distances form a geometric series \(h, 2hr, 2hr^2,\ldots\) which converges to \(h\cdot\frac{1+r}{1-r}\) when \(|r|<1\). Infinitely many positive terms can still sum to a finite value.
What happens as r → 1?
As \(r\to1^-\), the total distance \(h\cdot\frac{1+r}{1-r}\to\infty\). A "perfectly elastic" ball (\(r=1\)) bounces forever and travels infinite distance. This is why \(|r|<1\) is essential for convergence.
Module 15
The Koch Snowflake
Core idea: A fractal where the perimeter grows without bound at each step — yet the area it encloses converges to a finite value.
Two series, one divergent and one convergent, living in the same picture.
→ Drag the iteration slider to build the snowflake step by step.
0
Perimeter (side = 1)
P = 3
Grows without bound: P → ∞
Area (initial triangle = 1)
A = 1
Converges to 8/5 of original
How the perimeter diverges: At each step, every edge is replaced by 4 edges each \(\frac{1}{3}\) as long. So the total length is multiplied by \(\frac{4}{3}\) each time. Since \(\frac{4}{3}>1\), the perimeter grows as \(3\cdot\left(\frac{4}{3}\right)^n\to\infty\).
How the area converges: The extra area added at step \(n\) is \(3\cdot4^{n-1}\) triangles each with area \(\frac{1}{9^n}\), giving a geometric series that converges.
Think about it
Can a finite area have an infinite perimeter?
Yes — the Koch snowflake is the classic example. The enclosed area converges to \(\frac{8}{5}\) of the original triangle's area, while the perimeter grows without bound. This shows that area and perimeter can behave completely independently.
Module 16
Density of Fractions
Core idea: Between any two different fractions (rational numbers) there is always another fraction. This property is called density. Yet the rationals, despite being everywhere, are in some sense "smaller" than the real numbers.
→ Click "Find Middle" to zoom in between two fractions. You can keep going forever.
The mediant: Given fractions \(\frac{p}{q}\) and \(\frac{r}{s}\), their mediant \(\frac{p+r}{q+s}\) always lies strictly between them. This gives a constructive way to find a rational between any two rationals.
If the rationals are dense, how can there be "gaps"?
The rationals are dense on the real line — between any two rationals there is another. Yet \(\sqrt{2}\), \(\pi\), and \(e\) are irrational; they are not rational numbers. The irrationals fill the "gaps" between rationals and, surprisingly, there are far more irrationals than rationals (both are infinite, but different sizes of infinity — see Module 18).
Can we list all rational numbers?
Yes — Cantor showed the rationals are "countably infinite": you can pair them one-to-one with the natural numbers. The trick is to list them diagonally across a grid of \(\frac{p}{q}\) values. The real numbers cannot be listed this way — they are uncountably infinite.
Module 17
Calculator Approximation
Core idea: Your calculator doesn't "know" \(\sin x\). It approximates it using a finite partial sum of the Taylor series. Add more terms and watch the approximation improve — this is why convergence matters in practice.
Select a function button then the first few terms of the infinite series that is equivalent to the function will be displayed. The pattern of successive terms should be clear from the 4 displayed terms. As more terms of the series are taken the calculated value is closer to the actual value of the function at that point.
\(\sin x = x - \dfrac{x^3}{3!} + \dfrac{x^5}{5!} - \dfrac{x^7}{7!} + \cdots\)
Number of terms is how many terms of the infinite series, starting from the left and the variable x has increasing integer powers, are taken resulting in an approximation to the exact value of the function for a particular value of x.
Number of terms is how many terms of the infinite series, starting from the left and the variable x has increasing integer powers, are taken resulting in an approximation to the exact value of the function for a particular value of x.
Number of terms is how many terms of the infinite series, starting from the left and the variable x has increasing integer powers, are taken resulting in an approximation to the exact value of the function for a particular value of x.
1
Difference from exact value at \(x=3\): —
Why does it work? The error after \(k\) terms of the Taylor series for \(\sin x\) is no larger than \(\frac{|x|^{2k+1}}{(2k+1)!}\), which \(\to0\) as \(k\to\infty\) for any fixed \(x\). Convergence guarantees the approximation can be made as accurate as needed.
Think about it
Why does a single-term approximation fail far from zero?
Near \(x=0\), the Taylor series converges quickly because \(x^n\to0\) fast. Far from zero (large \(x\)), the early terms \(x^n/n!\) can be large before the factorial eventually dominates — you need more terms to overcome this initial growth. This is why calculators use range reduction: they first map any \(x\) to a small interval near 0, then apply the series.
Module 19
Infinity Paradoxes
Core idea: Infinity leads to results that feel paradoxical but are mathematically precise. Understanding these paradoxes deepens intuition for limits, series, and the real number system.
To cross a room you must first cross half, then half of what remains, then half again…
Infinitely many steps — yet you always arrive.
The geometric series converges because each step is \(\frac{1}{2}\) the previous. Zeno's error was assuming infinitely many steps must take infinite time. In mathematics, infinitely many terms can sum to a finite value — and in physics, each step takes half the time of the previous, so the total time is also finite.
Imagine a hotel with infinitely many rooms, all occupied. Can it take a new guest?
Yes! Move the guest in room \(n\) to room \(n+1\). Room 1 is now free. In fact, the hotel can accommodate infinitely many new guests by moving room \(n\) to room \(2n\), freeing all odd-numbered rooms. This shows \(\infty+1=\infty\) and \(\infty+\infty=\infty\) for countable sets.
But: Cantor proved you cannot accommodate uncountably many guests — the real numbers cannot be paired one-to-one with the rooms. There is a strictly larger infinity \(\aleph_1=2^{\aleph_0}\).
\(S=1-1+1-1+1-\cdots\) — what is the sum?
The series diverges by the standard definition: partial sums oscillate between 0 and 1, never converging. Some summation methods (Cesàro, Abel) assign the value \(\frac{1}{2}\), but these are extensions of ordinary summation, not the definition taught at A-level.
Is \(0.\overline{9}=0.9999\ldots\) exactly equal to 1, or just very close?
They are the same number. \(0.\overline{9}\) and \(1\) are two different decimal representations of the same real number, just as \(\frac{1}{2}=\frac{2}{4}\). The apparent paradox comes from treating an infinite decimal as if it "never quite reaches" its limit — but the limit is the number.
Algebraic proof: Let \(x=0.\overline{9}\). Then \(10x=9.\overline{9}=9+x\), so \(9x=9\), giving \(x=1\).
Think about it
Is there a number between 0.999… and 1?
No — because they are the same number! If two real numbers are different, you can always find a number between them (the rationals are dense). The fact that you cannot find anything between \(0.\overline{9}\) and \(1\) is precisely why they must be equal.
Summary
Summary & Revision
A concise reference for all the key results and tests covered in this app. Use this as a revision checklist.
\(\sin x = \sum (-1)^n x^{2n+1}/(2n+1)!\), all \(x\)
\(\cos x = \sum (-1)^n x^{2n}/(2n)!\), all \(x\)
\(\ln(1+x) = \sum (-1)^{n+1} x^n/n\), \(|x|\le1\)
Indeterminate forms
\(0/0,\ \infty/\infty,\ 0\cdot\infty\)
\(\infty-\infty,\ 0^0,\ 1^\infty,\ \infty^0\)
Use factorisation or L'Hôpital
\(\lim_{x\to0}\sin x/x = 1\)
Which convergence test to use?
1
Check \(a_n\to0\). If not → diverges immediately (divergence test). If yes → continue.
2
Is it a geometric series \(\sum ar^n\)? If yes → converges iff \(|r|<1\).
3
Is it a \(p\)-series \(\sum 1/n^p\)? If yes → converges iff \(p>1\).
4
Does it contain factorials or exponentials? Try the ratio test. If \(L=1\), try another test.
5
Does it look like a known series? Try comparison or limit comparison.
6
Does it alternate? Try the alternating series test (check terms decrease to 0).
Top misconceptions
\(a_n\to0\) does NOT mean \(\sum a_n\) converges
Bounded does NOT mean convergent
Ratio test with \(L=1\) is inconclusive
\(\infty\) is not a real number
\(0.\overline{9}=1\) exactly
Real-world connections
Bouncing ball → geometric series
Koch snowflake → fractal geometry
Calculator functions → Taylor series
Zeno's paradox → limits
Signal processing → Fourier series
Module 20
Fibonacci & the Golden Ratio
Core idea: The Fibonacci sequence \(1, 1, 2, 3, 5, 8, 13, 21, \ldots\) is defined by \(F_1=F_2=1\) and \(F_{n+1}=F_n+F_{n-1}\). Remarkably, the ratio of consecutive terms \(F_{n+1}/F_n\) converges to the golden ratio \(\varphi = \frac{1+\sqrt{5}}{2} \approx 1.618\ldots\)
→ Step through the sequence and watch the ratio converge.
The Fibonacci spiral approximates the golden spiral. Each square has side length equal to a Fibonacci number, and the quarter-circle arcs trace a spiral that appears throughout nature.
Fibonacci numbers appear in sunflower seed patterns, pine cones, and leaf arrangements (phyllotaxis) — because the golden angle \(360°/\varphi^2 \approx 137.5°\) is the "most irrational" angle, packing structures most efficiently.
The golden ratio has the remarkable property that it equals a continued fraction of all 1s — making it the "hardest number to approximate" by rationals.
If the ratio \(r = F_{n+1}/F_n\) converges to a limit \(L\), then dividing \(F_{n+1}=F_n+F_{n-1}\) by \(F_n\) gives \(L = 1 + 1/L\), so \(L^2 = L + 1\). The positive root of this is \(L = \frac{1+\sqrt{5}}{2} = \varphi\).
Is φ rational or irrational?
\(\varphi\) is irrational — it cannot be written as a fraction \(p/q\). Its continued fraction expansion \([1;1,1,1,\ldots]\) never terminates (rational numbers always have terminating continued fractions). In fact, among all irrationals, \(\varphi\) is the "hardest" to approximate by rationals — its rational approximants (the Fibonacci ratios) converge most slowly.
Common mistake
⚠ "The Fibonacci sequence and the golden ratio are just aesthetic — not rigorous maths."
They are completely rigorous. The convergence of \(F_{n+1}/F_n\to\varphi\) follows from the closed-form Binet formula \(F_n=(\varphi^n-\psi^n)/\sqrt5\) where \(|\psi|<1\), so \(\psi^n\to0\) and the ratio approaches \(\varphi\). The maths is as precise as any other limit proof.
Module 21
π from Infinite Series
Core idea: The number \(\pi\) can be computed using infinite series. Some converge remarkably slowly; others blazingly fast. The rate of convergence is the difference between a curiosity and a practical tool.
→ Step through each series and compare how quickly they reach π.
All three approximations plotted together on the same scale. Notice how Nilakantha (orange) converges far faster than Leibniz (navy) and Wallis (teal).
Leibniz Wallis Nilakantha
Key lesson: Leibniz's formula needs 1,000,000 terms to get 6 correct decimal places. Nilakantha's needs only about 50. Convergence speed is everything in practice.
Think about it
Why does the Leibniz series converge to π/4 and not something else?
It follows from the Taylor series for \(\arctan x\): \(\arctan x = x - \frac{x^3}{3} + \frac{x^5}{5} - \cdots\) for \(|x|\le1\). Substituting \(x=1\) gives \(\arctan 1 = \pi/4 = 1 - \frac{1}{3} + \frac{1}{5} - \cdots\). The convergence at \(x=1\) requires a separate argument (Abel's theorem), but it works.
What is the fastest known series for π?
The Chudnovsky algorithm, discovered in 1988, produces roughly 14 correct decimal digits per term. It has been used to compute π to over 100 trillion decimal places. It is based on Ramanujan's earlier series and involves deep results in complex analysis — far beyond A-level, but the key idea (faster convergence = fewer terms needed) is the same principle you see here.
Module 12
Telescoping Series
Core idea: A telescoping series is one where consecutive terms cancel in pairs, like the sections of a collapsible telescope closing up. What looks like an infinite sum collapses to a simple expression — often leaving just two terms.
→ Press "Add term" to see each cancellation happen step by step.
Each term is a difference of two fractions. When we sum N terms, nearly everything cancels:
\(\displaystyle\sum_{n=1}^N\frac{1}{n(n+1)} = \left(1-\frac{1}{2}\right)+\left(\frac{1}{2}-\frac{1}{3}\right)+\cdots+\left(\frac{1}{N}-\frac{1}{N+1}\right) = 1 - \frac{1}{N+1} \to 1\)
Using \(\ln\frac{n+1}{n} = \ln(n+1) - \ln(n)\), the partial sum telescopes to \(\ln(N+1) - \ln 1 = \ln(N+1) \to \infty\). A telescoping series can diverge!
SN = ln(1) = 0
This telescoping series diverges — partial sums grow without bound.
Another divergent telescoping series. The first term \(\sqrt{2}-1\) and last surviving term \(\sqrt{N+1}\) are left after cancellations — and \(\sqrt{N+1}\to\infty\).
SN = √2 − 1 = —
Diverges — the surviving term √(N+1) grows to ∞.
Think about it
How do you spot a telescoping series?
Look for terms that can be written as \(f(n) - f(n+1)\) (or \(f(n+1)-f(n)\)) for some function \(f\). The key technique is partial fractions: e.g. \(\frac{1}{n(n+1)}=\frac{1}{n}-\frac{1}{n+1}\). Once you spot the pattern, you can write out a few terms and see what survives after cancellation.
Can all series be made to telescope?
No — but many rational-function series can be decomposed using partial fractions into telescoping form. For example \(\sum\frac{1}{n(n+2)}\), \(\sum\frac{1}{n(n+1)(n+2)}\), and similar families all telescope. Series involving exponentials or trigonometric functions usually require different approaches.
Common mistake
⚠ "Telescoping always means the series converges."
Not at all. As Modules 17b and 17c show, telescoping just means terms cancel in sequence — but if the surviving terms still grow to infinity, the series diverges. Always check what's left after cancellation. The series \(\sum(\sqrt{n+1}-\sqrt{n})\) telescopes perfectly yet diverges.
Module 18
Cantor's Infinities — Counting the Uncountable
Core idea: Not all infinite sets are the same size. Georg Cantor (1874) proved that the integers and the rational numbers are the same "size" of infinity (\(leph_0\), countably infinite), but the real numbers are strictly larger — an uncountably infinite set that cannot be put into a one-to-one correspondence with the integers, no matter how cleverly you try.
What does "same size" mean for infinite sets?
Two finite sets have the same size if you can pair their elements one-to-one with nothing left over — like matching socks to feet.
Cantor extended this idea to infinite sets: two sets are the same size (have the same cardinality) if there exists a perfect one-to-one correspondence between them.
\(f : A o B ext{ is a bijection} \implies |A| = |B|\)
This gives a surprising result. The even integers seem to be "half" the integers — yet they can be paired perfectly:
Every integer \(n\) maps to \(2n\), every even integer \(2n\) maps back to \(n\) — a perfect bijection.
So \(|\mathbb{N}| = | ext{even integers}|\). A proper subset of an infinite set can have the same cardinality as the whole set — this is one hallmark of infinity.
Countably infinite (\(leph_0\)): A set is countably infinite if its elements can be listed in a sequence \(a_1, a_2, a_3, \ldots\) — that is, paired one-to-one with the natural numbers.
The integers \(\mathbb{Z}\), the rationals \(\mathbb{Q}\), and even the algebraic numbers are all countably infinite.
Pairs shown: 0
Think about it
Are the rational numbers countable?
Yes. Cantor's diagonal enumeration of fractions lists them as a 2D grid (row \(p\), column \(q\) = fraction \(p/q\)) and traverses it diagonally. Every fraction \(p/q\) appears somewhere in the list, so the rationals can be paired one-to-one with the natural numbers. This is deeply counterintuitive — the rationals are dense on the number line (between any two there is another), yet they are no more numerous than the integers.
What does Hilbert's Hotel have to do with this?
Hilbert's Hotel (see Module 19) is a physical metaphor for countable infinity: a hotel with \(leph_0\) rooms can always accommodate more guests by shifting room assignments — because \(leph_0 + 1 = leph_0\) and \(leph_0 + leph_0 = leph_0\). But it cannot accommodate uncountably many guests — there is no way to assign a room to every real number.
Cantor's Diagonal Argument (1891)
Suppose, for contradiction, that the real numbers between 0 and 1 could be listed: \(r_1, r_2, r_3, \ldots\)
Cantor showed that however long this list is, you can always construct a real number not on it — so no such complete list can exist.
The construction:
Assume we have a list \(r_1, r_2, r_3, \ldots\) of all reals in \([0,1]\), each written as an infinite decimal.
Build a new number \(d\) by going down the diagonal: take the \(n\)-th decimal digit of \(r_n\) and change it — say, replace any digit \(d\) by \((d+5) mod 10\).
The new number \(d\) differs from \(r_1\) in the 1st decimal place, from \(r_2\) in the 2nd, and so on.
So \(d\) is not equal to any \(r_n\) — it is not on the list. Contradiction.
The conclusion: The real numbers \(\mathbb{R}\) are uncountably infinite. They cannot be listed in a sequence. Their cardinality \(|\mathbb{R}| = 2^{leph_0}\) is strictly greater than \(leph_0\).
Think about it
Couldn't you just add the new number d to the list?
You could — but then Cantor's construction immediately produces another number not on the new list. The argument works against any purported complete list, no matter how it is extended. There is no way to "plug the gap" — every list of reals is incomplete.
Why does the argument fail for the rationals?
Cantor's diagonal construction produces a new real number \(d\) — but \(d\) is not guaranteed to be rational. The argument only shows that \(d
otin \{r_1, r_2, \ldots\}\); since the rationals are a proper subset of the reals, \(d\) might simply be irrational and not even a candidate for the rational list. The diagonal argument applies to the reals specifically because every decimal expansion (rational or not) is a valid real number.
A hierarchy of infinities
Cantor didn't stop at two sizes. He proved that for any set \(A\), its power set \(\mathcal{P}(A)\) — the set of all subsets of \(A\) — is always strictly larger. This gives an endless tower of infinities.
The Continuum Hypothesis asks: is there a set whose cardinality is strictly between \(leph_0\) and \(2^{leph_0}\)?
Gödel (1940) and Cohen (1963) together proved this question is independent of the standard axioms of set theory — it can be neither proved nor disproved from the axioms we normally use. This was one of the most striking results in 20th-century mathematics.
Think about it
Are there more reals than rationals, even though both are infinite?
Yes, in Cantor's precise sense: there is no bijection between \(\mathbb{Q}\) and \(\mathbb{R}\). The rationals are countably infinite (\(leph_0\)) while the reals are uncountably infinite (\(2^{leph_0}\)). In some intuitive sense, the irrationals "fill in" the number line while the rationals leave gaps — and there are far more irrationals than rationals.
Is \(2^{leph_0}\) equal to \(leph_1\)?
This is precisely the Continuum Hypothesis. \(leph_1\) is defined as the next cardinal after \(leph_0\) in the cardinal hierarchy. Whether \(2^{leph_0} = leph_1\) or \(2^{leph_0} > leph_1\) cannot be settled from the standard axioms of mathematics (ZFC). Both answers are consistent with those axioms.
Why Cantor's work matters
Cantor's theory provoked fierce opposition in his lifetime — Kronecker called it "a disease", Poincaré called it "a grave illness from which mathematics would recover." Today it is the foundation of modern set theory and underpins almost all of higher mathematics.
For analysis
The completeness of \(\mathbb{R}\) — the property that guarantees limits exist — is tied to its uncountability.
A countable set of points can have measure zero even if it is dense — the rationals in \([0,1]\) occupy "no length".
Most real numbers are transcendental (like \(\pi\) and \(e\)) — the algebraic numbers are only countably many.
For computation
There are only countably many computer programs, but uncountably many real-valued functions — so most functions are uncomputable.
Turing (1936) used a diagonal argument almost identical to Cantor's to prove the Halting Problem is undecidable.
Information theory: a countably infinite alphabet can represent any computable object.
Connection to this app: The density of the rationals (Module 16) makes more sense in light of Cantor's result — the rationals are everywhere on the number line yet constitute a set of measure zero. Hilbert's Hotel (Module 19) is a vivid illustration of \(leph_0\)-arithmetic. And the real number completeness that makes sequences and series converge to real limits ultimately rests on the uncountability shown here.
Common mistake
⚠ "Infinity is just infinity — you can't have different sizes."
This intuition fails. Cantor proved rigorously that no bijection can exist between \(\mathbb{N}\) and \(\mathbb{R}\). The proof is constructive: given any purported list of reals, you can exhibit a real number not on it. The existence of different sizes of infinity is not a philosophical position but a mathematical theorem.