This deck reads the epsilon-N definition of convergence as a nested-quantifier game: given a tolerance, produce a threshold. It covers the uniqueness of limits, the fact that a convergent sequence is bounded, the algebra of limits, divergence to infinity, and the squeeze theorem. It targets the classic traps: fixing N before epsilon, thinking that boundedness forces convergence, dividing by a limit that is zero, and the belief that terms bunching up is by itself enough, which is the difference between Cauchy and convergent.
Subject: Foundations of Higher Mathematics · 105 slides · symbolic lesson
Open the interactive version of this deck · Homework for this lesson
Objectives
By the end of this deck you can:
1. State the epsilon-N definition of convergence and read it as a nested-quantifier statement.
2. Write a rigorous epsilon-N proof: given a tolerance, produce a working threshold.
3. Prove the core theorems: limits are unique, convergent sequences are bounded, and the algebra of limits.
4. Apply the squeeze theorem and diagnose the classic convergence traps.
Warm-up
Discussion prompt
Before we open Sequences & Convergence: without looking back, what was the main idea of Ordered Fields & the Completeness Axiom, and what could you do by the end of it that you could not do before?
Hint: One sentence for the idea, one for the skill. If the second one is blank, that is the part to revisit.
Answer:
The ordered-field axioms, why the rationals form an ordered field with gaps, supremum and infimum, and the Least Upper Bound axiom that pins down the real numbers as the unique complete ordered field. Derives the Archimedean property and density of the rationals from completeness and constructs the reals via Dedekind cuts.
Concept
A sequence of real numbers is not a bag of numbers. It is a rule that assigns a real number to every natural-number position.
\[ a : \mathbb{N} \to \mathbb{R}, \qquad n \mapsto a_n \]
sequence — A function from the natural numbers to the reals. The value at position n is called the n-th term and written a-sub-n. The whole object is written (a_n).
Counterexample
Discussion prompt
A sequence of real numbers is not a bag of numbers. It is a rule that assigns a real number to every natural-number position.
That is stated as though it always holds. Do one of two things: produce a case where it fails, or say precisely what rules such a case out. "It just does" is not on the menu.
Hint: Hunt at the extremes first — zero, one, negative, empty, equal. If every extreme survives, the reason they survive is the proof.
Intuition
Picture an infinite list, one entry per line, numbered by position. Line one holds the first term, line two the second, and it never stops.
Because the index runs through the naturals, there is always a next term. There is no last one to inspect.
Analogy
Discussion prompt
Explain An endless numbered list by analogy to something with no Foundations of Higher Mathematics in it at all — a queue, a recipe, a map, a bank balance, whatever fits. Then say where your analogy breaks.
Hint: An analogy that never breaks is not an analogy, it is the same idea wearing a hat. Find the seam — that is the part that is actually new.
Answer:
Picture an infinite list, one entry per line, numbered by position. Line one holds the first term, line two the second, and it never stops.
Concept
A sequence carries more information than the set of its values. Position matters, and a value may repeat.
The all-ones sequence and the alternating plus-and-minus-one sequence use the same two symbols, yet they are completely different sequences.
\[ (1,1,1,\dots) \ne (1,-1,1,-1,\dots) \]
Explain it
Discussion prompt
Explain Order and repetition matter to a student a year behind you. No notation, no jargon they have not met — and it still has to be true.
Hint: If your explanation needs a symbol they have never seen, you are describing the notation rather than the idea.
Answer:
A sequence carries more information than the set of its values. Position matters, and a value may repeat.
Concept
We want to capture the idea that the terms settle down toward a single number L as the position grows.
Not that the terms reach L, and not that they get closer every single step. The right idea is: past some point, every term is as close to L as you like.
Intuition
Draw a horizontal band of some chosen half-width around L. Convergence means: eventually the whole tail of the sequence lives inside that band, no matter how thin you made it.
Thinner band, later the tail has to enter, but it always does enter. That word eventually is the entire content of the definition.
Picture it
Figure (svg): A horizontal line at height L with a shaded band of half-width epsilon above and below it; scattered dots representing sequence terms start outside the band on the left and all fall inside the band on the right.
Discussion prompt
Read the picture before the words. What is this showing, and what is the one thing it is built to make obvious? Commit to an answer, then read on.
Hint: Name the parts, then say what changes between them — and if nothing changes, say what is being held still.
Answer:
Call the half-width of the band epsilon. The band is the set of points within epsilon of L.
Concept
Call the half-width of the band epsilon. The band is the set of points within epsilon of L.
Figure (svg): A horizontal line at height L with a shaded band of half-width epsilon above and below it; scattered dots representing sequence terms start outside the band on the left and all fall inside the band on the right.
Convergence says: for every band width epsilon, only finitely many terms are allowed to sit outside the band.
Concept
To say the tail is inside the band we name a threshold position N and require every term from position N onward to be inside.
\[ n \ge N \implies |a_n - L| < \varepsilon \]
The distance from the n-th term to L, written with absolute-value bars, is what must be smaller than epsilon.
Concept
Here is the definition in full. Read it slowly, left to right.
\[ \lim_{n\to\infty} a_n = L \;\iff\; \forall \varepsilon > 0 \; \exists N \in \mathbb{N} \; \forall n \ge N \; \big(|a_n - L| < \varepsilon\big) \]
converges to L — For every positive tolerance epsilon there is a threshold N so that every term from position N onward is within epsilon of L. When such an L exists the sequence is convergent.
Definition probe
Sort into buckets
Every line below is part of the definition of sequence or of converges to L — one or the other, never both. Put each where it belongs.
Intuition
Think of it as a game. An adversary hands you a tolerance epsilon, as small and mean as they like.
You must respond with a threshold N that works for that epsilon. If you can always answer, no matter what they pick, the sequence converges.
Crucially, you answer after seeing their epsilon. Your N is allowed to depend on it.
Concept
The definition is a for-all, then a there-exists, then a for-all. Order is everything, exactly as in the quantifier deck.
\[ \forall \varepsilon\; \exists N\; \forall n\ge N \]
Because the there-exists N sits to the right of the for-all epsilon, N may be chosen using epsilon. Swap them and you get a different, much stronger claim.
Concept
A cleaner slogan: a property holds eventually if it holds for all terms past some threshold, that is, for all but finitely many terms.
Then convergence to L reads: for every epsilon, the terms are eventually within epsilon of L. Finitely many early terms may misbehave freely.
Concept
To say a sequence does not converge to L, mechanically negate the definition, flipping each quantifier as you pass it.
\[ \exists \varepsilon > 0 \; \forall N \in \mathbb{N} \; \exists n \ge N \; \big(|a_n - L| \ge \varepsilon\big) \]
In words: some tolerance epsilon is bad, in that no matter how far out you set the threshold, some later term still escapes the band. A sequence diverges if this holds for every candidate L.
Ranking
Put in order
Put the moves of Warm-up: a constant sequence into the order they have to happen.
Why: These are the moves of the worked example in the order it makes them, and each one is set up by the one before it. Every epsilon-N proof begins by letting the adversary move first: take epsilon as given and positive.
Worked example
Claim: the constant sequence whose every term equals c converges to c.
\[ a_n = c \;\text{ for all } n, \qquad \lim_{n\to\infty} a_n = c \]
Fix an arbitrary tolerance epsilon greater than zero
Why: Every epsilon-N proof begins by letting the adversary move first: take epsilon as given and positive.
Compute the distance from a term to the target
Why: The distance is the absolute value of c minus c, which is zero.
\[ |a_n - c| = |c - c| = 0 \]
Choose the threshold N equal to 1
Why: Zero is smaller than every positive epsilon, so any threshold works. We may as well start at the first term.
Verify the choice
Why: For every n at least 1, the distance is 0, which is less than epsilon. The definition is satisfied for the given epsilon, and epsilon was arbitrary, so the limit is c.
\[ n \ge 1 \implies |a_n - c| = 0 < \varepsilon \]
Picture it
Animation
Shows: Each line of the worked example "Warm-up: a constant sequence", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: For every n at least 1, the distance is 0, which is less than epsilon. The definition is satisfied for the given epsilon, and epsilon was arbitrary, so the limit is c.
Concept
Almost every basic limit proof leans on one fact about the reals: the naturals are not bounded above.
\[ \forall x \in \mathbb{R}\; \exists N \in \mathbb{N}\; (N > x) \]
Equivalently, for any positive real there is a natural number whose reciprocal is smaller. This is what lets us turn a demand about epsilon into a concrete threshold N.
Step zero
Discussion prompt
Prove that one-over-n goes to zero — before any calculation: what is the plan? Name the moves in order, in plain English, without doing the arithmetic.
Hint: It starts with: Fix an arbitrary epsilon greater than zero
Answer:
Worked example
The signature first proof. Claim:
\[ \lim_{n\to\infty} \frac{1}{n} = 0 \]
Fix an arbitrary epsilon greater than zero
Why: Let the adversary pick the tolerance. Everything that follows may use this epsilon.
Do scratch work to find what N must beat
Why: We need the distance under epsilon. Solving the target inequality tells us how large n must be.
\[ \left|\tfrac{1}{n} - 0\right| = \tfrac{1}{n} < \varepsilon \;\Longleftrightarrow\; n > \tfrac{1}{\varepsilon} \]
Choose a natural threshold N larger than one over epsilon
Why: The Archimedean property guarantees such an N exists. This is the move that makes the proof concrete.
\[ \text{pick } N \in \mathbb{N} \text{ with } N > \tfrac{1}{\varepsilon} \]
Show the tail is inside the band
Why: For n at least N the reciprocal only shrinks, so it stays under one over N, which is under epsilon.
\[ n \ge N \implies \tfrac{1}{n} \le \tfrac{1}{N} < \varepsilon \]
Verify with a sample tolerance epsilon equal to one hundredth
Why: Then one over epsilon is 100, so N equal to 101 works: at n equal to 101 the term is about 0.0099, safely under 0.01. The construction really produces a working threshold.
\[ N = 101 > 100 = \tfrac{1}{0.01}, \qquad \tfrac{1}{101} \approx 0.0099 < 0.01 \]
Picture it
Animation
Shows: Each line of the worked example "Prove that one-over-n goes to zero", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: Then one over epsilon is 100, so N equal to 101 works: at n equal to 101 the term is about 0.0099, safely under 0.01. The construction really produces a working threshold.
Trap
A tempting shortcut: pick a threshold first, then wave at epsilon.
\[ \text{Let } N = 1.\; \text{ For } n \ge 1,\; \tfrac{1}{n} \le 1,\; \text{so } \tfrac{1}{n} < \varepsilon. \]
The last step is false. One is not less than every positive epsilon. Take epsilon equal to one half: the very first term, which is 1, is not within one half of zero.
\[ \varepsilon = \tfrac{1}{2}: \quad \tfrac{1}{1} = 1 \not< \tfrac{1}{2} \]
Let epsilon move first, then build N from it. The threshold is allowed to grow as epsilon shrinks.
\[ \text{Given } \varepsilon,\; \text{choose } N > \tfrac{1}{\varepsilon}. \]
Now for epsilon equal to one half the recipe returns a threshold past 2, and the tail from there really is inside the band. The quantifier order for-all-epsilon then there-exists-N is honored.
\[ \varepsilon = \tfrac{1}{2}: \; N = 3 > 2,\; n \ge 3 \implies \tfrac{1}{n} \le \tfrac{1}{3} < \tfrac{1}{2} \]
Estimation
Predict first
The absolute value bars in the definition let signs wash out. Claim:
Commit before you compute: what does A sign-alternating example come out to? A rough magnitude and the right form is enough — the point is to have something concrete to be wrong about.
Correct: Verify with epsilon equal to one tenth
Why: A prediction you can defend turns the computation into a check rather than a leap of faith — and an answer that contradicts it is caught on the spot. One over epsilon is 10, so N equal to 11 works: for every n at least 11 the distance one over n is at most one eleventh, which is under one tenth.
Worked example
The absolute value bars in the definition let signs wash out. Claim:
\[ \lim_{n\to\infty} \frac{(-1)^n}{n} = 0 \]
Fix an arbitrary epsilon greater than zero
Why: As always, the tolerance comes first.
Collapse the sign using the absolute value
Why: The numerator has size one regardless of parity, so the distance to zero is exactly one over n.
\[ \left|\tfrac{(-1)^n}{n} - 0\right| = \tfrac{|(-1)^n|}{n} = \tfrac{1}{n} \]
Choose N larger than one over epsilon
Why: The problem is now identical to the one-over-n proof, since the distance is the same quantity.
Verify with epsilon equal to one tenth
Why: One over epsilon is 10, so N equal to 11 works: for every n at least 11 the distance one over n is at most one eleventh, which is under one tenth. Alternating signs never mattered.
\[ n \ge 11 \implies \tfrac{1}{n} \le \tfrac{1}{11} < \tfrac{1}{10} \]
Picture it
Animation
Shows: Each line of the worked example "A sign-alternating example", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: One over epsilon is 10, so N equal to 11 works: for every n at least 11 the distance one over n is at most one eleventh, which is under one tenth. Alternating signs never mattered.
Concept
Finding N is exploratory: you run the target inequality backward to see how big n must be. That is scratch work.
The written proof runs forward: declare the N you found, then show the tail obeys the bound. Keep the search and the argument separate in your head, even if both use the same algebra.
Worked example
Claim, where the terms are a ratio of linear expressions:
\[ \lim_{n\to\infty} \frac{3n+1}{n+2} = 3 \]
Fix epsilon greater than zero and simplify the distance
Why: Combine over a common denominator so the size is transparent. The threes cancel, leaving a constant over n plus two.
\[ \left|\tfrac{3n+1}{n+2} - 3\right| = \left|\tfrac{3n+1 - 3(n+2)}{n+2}\right| = \tfrac{5}{n+2} \]
Scratch: solve the target inequality
Why: We need five over n-plus-two under epsilon, which rearranges to a lower bound on n.
\[ \tfrac{5}{n+2} < \varepsilon \iff n > \tfrac{5}{\varepsilon} - 2 \]
Choose N larger than five over epsilon
Why: This is even stronger than needed, and it is cleaner. Archimedes supplies such a natural number.
Show the tail obeys the bound
Why: For n at least N, since n-plus-two exceeds n, the distance is under five over N, which is under epsilon.
\[ n \ge N \implies \tfrac{5}{n+2} < \tfrac{5}{n} \le \tfrac{5}{N} < \varepsilon \]
Verify with epsilon equal to one hundredth
Why: Then five over epsilon is 500, so N equal to 500 works: at n equal to 500 the distance is five over 502, about 0.00996, under 0.01. The threshold really controls the tail.
\[ N = 500,\; \tfrac{5}{502} \approx 0.00996 < 0.01 \]
Picture it
Animation
Shows: Each line of the worked example "A rational sequence tending to three", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: Then five over epsilon is 500, so N equal to 500 works: at n equal to 500 the distance is five over 502, about 0.00996, under 0.01. The threshold really controls the tail.
Intuition
Before any epsilon, you should already smell the limit. For large n the plus-one and plus-two are dwarfed by the leading terms.
The ratio behaves like three-n over n, which is three. The epsilon-N proof is just the rigorous confirmation of this dominant-term instinct.
Step zero
Discussion prompt
A ratio tending to one — before any calculation: what is the plan? Name the moves in order, in plain English, without doing the arithmetic.
Hint: It starts with: Fix epsilon and simplify the distance
Answer:
Worked example
Claim:
\[ \lim_{n\to\infty} \frac{n+1}{n} = 1 \]
Fix epsilon and simplify the distance
Why: Split the fraction; the n-over-n is one and cancels, leaving one over n.
\[ \left|\tfrac{n+1}{n} - 1\right| = \left|\tfrac{1}{n}\right| = \tfrac{1}{n} \]
Choose N larger than one over epsilon
Why: Again the distance is one over n, so the same Archimedean choice finishes it.
Verify with epsilon equal to five thousandths
Why: One over epsilon is 200, so N equal to 201 works: for n at least 201 the distance is at most one over 201, under 0.005. Confirmed.
\[ n \ge 201 \implies \tfrac{1}{n} \le \tfrac{1}{201} < 0.005 \]
Picture it
Animation
Shows: Each line of the worked example "A ratio tending to one", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: One over epsilon is 200, so N equal to 201 works: for n at least 201 the distance is at most one over 201, under 0.005. Confirmed.
Concept
A convergent sequence has exactly one limit. This is what licenses us to write the limit and give it a name.
\[ a_n \to L \;\text{ and }\; a_n \to L' \implies L = L' \]
Intuition
Suppose L and L-prime were different limits. Put a thin band around each, thin enough that the two bands do not overlap.
The tail would have to live inside both bands at once. But they are disjoint, so no single term can be in both. That is the contradiction the proof makes precise.
Missing information
Discussion prompt
Suppose a sequence converges to both L and L-prime, and assume for contradiction they differ.
What do you need to know — or decide — before the first line can be written? List everything the problem has to hand you.
Hint: Anything you would have to invent to get started is a thing the problem must supply.
Answer:
Since the limits differ, this epsilon is strictly positive, so both convergence statements apply to it.
Worked example
Suppose a sequence converges to both L and L-prime, and assume for contradiction they differ.
Choose epsilon to be half the gap between the two limits
Why: Since the limits differ, this epsilon is strictly positive, so both convergence statements apply to it.
\[ \varepsilon = \tfrac{|L - L'|}{2} > 0 \]
Pull a threshold from each limit and take the larger
Why: Past the larger threshold both bounds hold simultaneously for the same term a-sub-n.
\[ n \ge \max(N_1, N_2): \quad |a_n - L| < \varepsilon,\; |a_n - L'| < \varepsilon \]
Bound the gap by the triangle inequality
Why: Route from L to L-prime through a-sub-n. Each leg is under epsilon, so the gap is under twice epsilon, which is the gap itself.
\[ |L - L'| \le |L - a_n| + |a_n - L'| < \varepsilon + \varepsilon = |L - L'| \]
Verify the contradiction forces the limits equal
Why: We derived that the gap is strictly less than itself, which is impossible. Hence the assumption fails and L equals L-prime.
\[ |L - L'| < |L - L'| \;\text{is absurd} \implies L = L' \]
Picture it
Animation
Shows: Each line of the worked example "Prove the limit is unique", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: We derived that the gap is strictly less than itself, which is impossible. Hence the assumption fails and L equals L-prime.
Concept
A sequence is bounded if a single number caps the size of every term at once.
\[ \exists M > 0 \; \forall n \; (|a_n| \le M) \]
bounded sequence — There is one constant M so that every term lies between negative M and positive M. The same M must work for all positions, not a different cap for each term.
Concept
Every convergent sequence is bounded. The converse fails, and that gap is a favorite exam trap.
\[ a_n \to L \implies (a_n) \text{ is bounded} \]
Intuition
Fix any one tolerance, say one. Past some threshold the whole tail is packed near L, hence bounded there.
The only terms left are the finitely many before the threshold. A finite set of numbers always has a largest size. Take the bigger of the two caps and you have bounded the whole sequence.
Estimation
Predict first
Suppose the sequence converges to L. We produce a single cap M.
Commit before you compute: what does Prove convergent implies bounded come out to? A rough magnitude and the right form is enough — the point is to have something concrete to be wrong about.
Correct: Verify M bounds every term
Why: A prediction you can defend turns the computation into a check rather than a leap of faith — and an answer that contradicts it is caught on the spot. For positions before N the size is at most M-zero, hence at most M; for positions from N on the size is under absolute-L-plus-one, hence at most M.
Worked example
Suppose the sequence converges to L. We produce a single cap M.
Apply convergence with the specific tolerance one
Why: The definition holds for every epsilon, so in particular for epsilon equal to one. This yields a threshold N.
\[ \exists N\; \forall n \ge N: \; |a_n - L| < 1 \implies |a_n| < |L| + 1 \]
Cap the finitely many early terms separately
Why: The terms before position N form a finite list, so their sizes have a maximum. This is where finiteness does the work.
\[ M_0 = \max\{|a_1|, |a_2|, \dots, |a_{N-1}|\} \]
Take M to be the larger of the two caps
Why: One cap controls the tail, the other the head. Their maximum controls every term.
\[ M = \max\big(M_0,\; |L| + 1\big) \]
Verify M bounds every term
Why: For positions before N the size is at most M-zero, hence at most M; for positions from N on the size is under absolute-L-plus-one, hence at most M. Both regimes obey the same cap, so the sequence is bounded.
\[ \forall n\; |a_n| \le M \]
Picture it
Animation
Shows: Each line of the worked example "Prove convergent implies bounded", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: For positions before N the size is at most M-zero, hence at most M; for positions from N on the size is under absolute-L-plus-one, hence at most M. Both regimes obey the same cap, so the sequence is bounded.
Trap
The false converse: since the sequence stays trapped between minus one and one, surely it converges.
\[ a_n = (-1)^n \in [-1, 1] \;\Rightarrow\; \text{converges?} \]
Boundedness is necessary for convergence, not sufficient. This sequence forever hops between two values and settles on neither.
It diverges. Suppose a limit L existed and use the divergence test with tolerance one.
\[ \text{if } a_n \to L: \; \text{eventually } |1 - L| < 1 \text{ and } |-1 - L| < 1 \]
Then the distance between the two hop values would be under two, yet it is exactly two. Contradiction, so no limit exists.
\[ 2 = |1 - (-1)| \le |1 - L| + |L - (-1)| < 1 + 1 = 2 \]
Concept
Once you know two sequences converge, you rarely need epsilon again. Limits pass through arithmetic. First, sums.
\[ a_n \to A,\; b_n \to B \implies a_n + b_n \to A + B \]
The limit of a sum is the sum of the limits. The same holds for differences.
Concept
Products and constant multiples behave too.
\[ a_n b_n \to AB, \qquad c \, a_n \to c\,A \]
So limits are linear and multiplicative. Building limits from known ones is pure bookkeeping once these rules are in hand.
Concept
Quotients work as well, but with one guard rail: the limit of the denominator must not be zero.
\[ a_n \to A,\; b_n \to B \ne 0 \implies \frac{a_n}{b_n} \to \frac{A}{B} \]
That nonzero condition is not decoration. Drop it and the rule collapses, as a later trap shows.
Intuition
If two quantities are each eventually pinned near their targets, their sum is pinned near the combined target: the errors add, and half of a small tolerance plus half of a small tolerance is still small.
That epsilon-over-two accounting is the whole proof of the sum rule, which we do next.
Ranking
Put in order
Put the moves of Compute a limit with the algebra rules into the order they have to happen.
Why: These are the moves of the worked example in the order it makes them, and each one is set up by the one before it. Dividing by n-squared exposes pieces we already know how to take limits of.
Worked example
Find the limit of a ratio of quadratics.
\[ \lim_{n\to\infty} \frac{2n^2 + 3n}{n^2 + 1} \]
Divide top and bottom by the highest power
Why: Dividing by n-squared exposes pieces we already know how to take limits of. It converts growing quantities into ones tending to constants.
\[ \frac{2n^2 + 3n}{n^2 + 1} = \frac{2 + \tfrac{3}{n}}{1 + \tfrac{1}{n^2}} \]
Take limits piece by piece
Why: The three-over-n and one-over-n-squared both tend to zero by our basic limits, and constants are constant.
\[ \tfrac{3}{n} \to 0, \qquad \tfrac{1}{n^2} \to 0 \]
Assemble with the sum and quotient rules
Why: Numerator tends to two, denominator to one, and one is nonzero, so the quotient rule applies and gives two.
\[ \frac{2 + 0}{1 + 0} = 2 \]
Verify numerically at a large index
Why: At n equal to 100 the ratio is 20300 over 10001, about 2.03, already hugging the predicted value two. The algebra checks out against the numbers.
\[ n=100: \; \frac{20300}{10001} \approx 2.03 \]
Picture it
Animation
Shows: Each line of the worked example "Compute a limit with the algebra rules", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: At n equal to 100 the ratio is 20300 over 10001, about 2.03, already hugging the predicted value two. The algebra checks out against the numbers.
Intuition
For a ratio of polynomials the trick is always the same: divide through by the highest power present.
Everything with a smaller power melts to zero, and you are left comparing the leading coefficients. Equal degrees give their ratio; a bigger bottom degree gives zero.
Hypothesis
Predict first
Prove the sum rule from epsilon-N is about to be worked. State your hypothesis first: which rule or definition decides this one, and what is the first move it forces? Then watch whether the example agrees with you.
Correct: Fix epsilon and split the budget in half
Why: We will control each sequence to within half the tolerance, so their combined error stays under the full tolerance.
A hypothesis you wrote down is falsifiable; a vague sense of how it will go is not. If the example opens somewhere else, that gap is the thing worth chasing.
Worked example
Suppose the first sequence tends to A and the second to B. Show their sum tends to A plus B.
Fix epsilon and split the budget in half
Why: We will control each sequence to within half the tolerance, so their combined error stays under the full tolerance.
\[ \text{target: } |(a_n + b_n) - (A + B)| < \varepsilon \]
Get a threshold for each sequence at tolerance epsilon-over-two
Why: Each convergence applies to the positive number epsilon-over-two, yielding two thresholds.
\[ n \ge N_1: |a_n - A| < \tfrac{\varepsilon}{2}, \quad n \ge N_2: |b_n - B| < \tfrac{\varepsilon}{2} \]
Take the larger threshold and apply the triangle inequality
Why: Past both thresholds, group the error into the two separate errors, each under half the tolerance.
\[ |(a_n{+}b_n){-}(A{+}B)| \le |a_n{-}A| + |b_n{-}B| < \tfrac{\varepsilon}{2} + \tfrac{\varepsilon}{2} \]
Verify the halves add to epsilon
Why: The two halves sum to exactly epsilon, so past the larger threshold the combined error is under epsilon. Since epsilon was arbitrary, the sum converges to A plus B.
\[ \tfrac{\varepsilon}{2} + \tfrac{\varepsilon}{2} = \varepsilon \]
Picture it
Animation
Shows: Each line of the worked example "Prove the sum rule from epsilon-N", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: The two halves sum to exactly epsilon, so past the larger threshold the combined error is under epsilon. Since epsilon was arbitrary, the sum converges to A plus B.
Trap
Applying the quotient rule blindly. Consider the ratio where both parts tend to zero.
\[ a_n = \tfrac{1}{n},\; b_n = \tfrac{1}{n^2}; \quad \frac{a_n}{b_n} \to \frac{0}{0}? \]
Writing zero over zero is meaningless, and pretending the limit is one, or zero, is worse. The quotient rule simply does not apply when the bottom limit is zero.
Simplify first, then take the limit of what remains.
\[ \frac{a_n}{b_n} = \frac{1/n}{1/n^2} = n \]
The simplified sequence is just n, which grows without bound. So the quotient diverges to infinity. The moral: check the denominator limit before invoking the rule.
\[ \frac{a_n}{b_n} = n \to \infty \]
Break the constraint
Discussion prompt
The rule this trap just fixed:
Simplify first, then take the limit of what remains.
Now break it on purpose. Build a case that violates it and follow the consequences until something visibly fails. Where does the failure first show up — and would you have noticed it if you had not been looking?
Hint: The dangerous rules are the ones whose violation still produces an answer. If yours fails loudly, try to find one that fails quietly.
Concept
Some sequences fail to converge because they run off to infinity. This is a specific kind of divergence with its own definition.
\[ a_n \to \infty \;\iff\; \forall M > 0 \; \exists N \; \forall n \ge N \; (a_n > M) \]
The tolerance epsilon is replaced by a challenge height M: eventually every term clears any bar you name.
Step zero
Discussion prompt
Prove that n tends to infinity — before any calculation: what is the plan? Name the moves in order, in plain English, without doing the arithmetic.
Hint: It starts with: Fix an arbitrary challenge height M greater than zero
Answer:
Worked example
The identity sequence, whose n-th term is n itself, diverges to infinity.
\[ a_n = n, \qquad a_n \to \infty \]
Fix an arbitrary challenge height M greater than zero
Why: The adversary now names a bar to clear rather than a tolerance to beat.
Choose a natural threshold N larger than M
Why: The Archimedean property supplies a natural number exceeding any real, here M.
\[ \text{pick } N \in \mathbb{N} \text{ with } N > M \]
Show the tail clears the bar
Why: For n at least N the term is n, which is at least N, which exceeds M.
\[ n \ge N \implies a_n = n \ge N > M \]
Verify with a sample height M equal to one thousand
Why: Then N equal to 1001 works: every term from position 1001 on exceeds 1000. The recipe clears any named bar, so the sequence diverges to infinity.
\[ M = 1000: \; N = 1001,\; n \ge 1001 \implies n > 1000 \]
Picture it
Animation
Shows: Each line of the worked example "Prove that n tends to infinity", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: Then N equal to 1001 works: every term from position 1001 on exceeds 1000. The recipe clears any named bar, so the sequence diverges to infinity.
Missing information
Discussion prompt
Show that the plus-and-minus-one sequence has no limit at all, using the negated definition.
What do you need to know — or decide — before the first line can be written? List everything the problem has to hand you.
Hint: Anything you would have to invent to get started is a thing the problem must supply.
Answer:
To prove non-existence of any limit, suppose one exists and derive an absurdity.
Worked example
Show that the plus-and-minus-one sequence has no limit at all, using the negated definition.
\[ a_n = (-1)^n \;\text{ diverges} \]
Assume toward contradiction it converges to some L
Why: To prove non-existence of any limit, suppose one exists and derive an absurdity.
Apply the definition with tolerance one
Why: Some threshold must place the whole tail within distance one of L. That tail contains both a plus-one term and a minus-one term.
\[ n \ge N: \; |a_n - L| < 1 \]
Measure the gap between the two hop values through L
Why: Both hop values lie within one of L, so by the triangle inequality their mutual distance is under two.
\[ |1 - (-1)| \le |1 - L| + |L - (-1)| < 1 + 1 = 2 \]
Verify the contradiction
Why: The gap between plus one and minus one is exactly two, but we bounded it strictly under two. Impossible, so no L works and the sequence diverges.
\[ 2 < 2 \;\text{is false} \implies \text{no limit exists} \]
Picture it
Animation
Shows: Each line of the worked example "Prove the alternating sign sequence diverges", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: The gap between plus one and minus one is exactly two, but we bounded it strictly under two. Impossible, so no L works and the sequence diverges.
Trap
Treating divergence to infinity as ordinary convergence, then plugging infinity into the algebra rules.
\[ n \to \infty,\; \tfrac{1}{n} \to 0 \;\Rightarrow\; n \cdot \tfrac{1}{n} \to \infty \cdot 0 = 0? \]
Infinity is not a real number, so the product rule does not apply, and infinity-times-zero is not a valid computation.
A sequence that diverges to infinity is not convergent. The algebra-of-limits rules require genuine real limits on both factors.
\[ n \cdot \tfrac{1}{n} = 1 \to 1 \]
Here the product simplifies to the constant one, whose limit is one, not zero. Simplify the actual expression rather than manipulating the symbol for infinity.
Concept
When a sequence is hard to attack directly, trap it between two friendlier sequences that share a limit.
\[ b_n \le a_n \le c_n \;\text{ and }\; b_n \to L,\; c_n \to L \implies a_n \to L \]
If the outer two both converge to L, the trapped middle sequence has nowhere to go but L.
Intuition
Picture two converging sequences closing in on L like a shrinking sandwich. The middle sequence is pinned between the slices.
As the gap between the slices vanishes, the filling is forced to the same limit. The inequalities only need to hold eventually, not from the start.
Estimation
Predict first
The numerator wobbles and has no limit of its own, so attack it with a squeeze.
Commit before you compute: what does Squeeze: cosine over n come out to? A rough magnitude and the right form is enough — the point is to have something concrete to be wrong about.
Correct: Verify the squeeze forces the middle to zero
Why: A prediction you can defend turns the computation into a check rather than a leap of faith — and an answer that contradicts it is caught on the spot. Both outer sequences share the limit zero, so the trapped middle must also converge to zero.
Worked example
The numerator wobbles and has no limit of its own, so attack it with a squeeze.
\[ \lim_{n\to\infty} \frac{\cos n}{n} = 0 \]
Bound the wobbling numerator
Why: Cosine of anything lives between negative one and one, a fact independent of n.
\[ -1 \le \cos n \le 1 \]
Divide the bounds by the positive quantity n
Why: Dividing by a positive number preserves the inequalities and produces the two slices of the sandwich.
\[ -\tfrac{1}{n} \le \frac{\cos n}{n} \le \tfrac{1}{n} \]
Send both slices to zero
Why: Both the negative and positive reciprocal sequences tend to zero, so their common limit is zero.
\[ -\tfrac{1}{n} \to 0, \qquad \tfrac{1}{n} \to 0 \]
Verify the squeeze forces the middle to zero
Why: Both outer sequences share the limit zero, so the trapped middle must also converge to zero. Check at n equal to 100: the term is at most one hundredth in size, already tiny.
\[ \left|\tfrac{\cos 100}{100}\right| \le \tfrac{1}{100} = 0.01 \]
Picture it
Animation
Shows: Each line of the worked example "Squeeze: cosine over n", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: Both outer sequences share the limit zero, so the trapped middle must also converge to zero. Check at n equal to 100: the term is at most one hundredth in size, already tiny.
Step zero
Discussion prompt
Squeeze: a geometric decay — before any calculation: what is the plan? Name the moves in order, in plain English, without doing the arithmetic.
Hint: It starts with: Bound two to the n below by n
Answer:
Worked example
Show the halving sequence tends to zero without touching logarithms.
\[ \lim_{n\to\infty} \frac{1}{2^n} = 0 \]
Bound two to the n below by n
Why: A quick induction shows two to the n is at least n for every natural n, so its reciprocal is at most one over n.
\[ 2^n \ge n \implies \tfrac{1}{2^n} \le \tfrac{1}{n} \]
Trap the term between zero and one over n
Why: The term is positive, giving a lower slice of zero, and bounded above by the reciprocal of n.
\[ 0 \le \tfrac{1}{2^n} \le \tfrac{1}{n} \]
Both slices converge to zero
Why: The constant zero and the sequence one over n both tend to zero, so the squeeze applies.
Verify at a sample index
Why: At n equal to 10 the term is one over 1024, comfortably below the slice one over 10. The middle is pinned to zero.
\[ n=10: \; 0 \le \tfrac{1}{1024} \le \tfrac{1}{10} \]
Picture it
Animation
Shows: Each line of the worked example "Squeeze: a geometric decay", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: At n equal to 10 the term is one over 1024, comfortably below the slice one over 10. The middle is pinned to zero.
Concept
Every hypothesis in these theorems only needs to hold eventually. The squeeze inequalities, boundedness, monotonicity: all may fail for a finite prefix.
Convergence is a statement about the infinite tail. The first million terms can be anything; they are a finite prefix and cannot change the limit.
Concept
Take a convergent sequence and overwrite any finite number of its terms with arbitrary values. The limit is unchanged.
Past the last altered position the two sequences agree exactly, so the same threshold N works for both. Limits ignore finite edits.
Explain it
Discussion prompt
Explain Changing finitely many terms changes nothing to a student a year behind you. No notation, no jargon they have not met — and it still has to be true.
Hint: If your explanation needs a symbol they have never seen, you are describing the notation rather than the idea.
Answer:
Take a convergent sequence and overwrite any finite number of its terms with arbitrary values. The limit is unchanged.
Trap
A seductive claim: if consecutive terms get arbitrarily close to each other, the sequence must converge.
\[ |a_{n+1} - a_n| \to 0 \;\Rightarrow\; (a_n) \text{ converges?} \]
False. The partial sums of the harmonic series have consecutive gaps one over n, tending to zero, yet the sequence marches off to infinity.
\[ H_n = \sum_{k=1}^{n} \tfrac{1}{k}: \quad |H_{n+1} - H_n| = \tfrac{1}{n+1} \to 0, \; H_n \to \infty \]
The correct closeness condition is Cauchy: all terms far out are close to each other, not merely neighbors. Even that only guarantees convergence because the reals are complete.
\[ \forall \varepsilon > 0\; \exists N\; \forall m,n \ge N\; (|a_m - a_n| < \varepsilon) \]
Over the rationals a Cauchy sequence can fail to converge, for instance decimal truncations of the square root of two. Completeness of the reals is exactly what rescues the implication, a theme the next deck develops.
Two truths and a lie
Sort into buckets
Some of these hold up and some are the exact mistakes this lesson is built to prevent. Sort them.
Concept
Iterative algorithms produce a sequence of guesses, and convergence is what it means for the algorithm to settle on an answer.
Each round updates the current guess by a rule. The epsilon-N definition is precisely the promise that the guesses eventually land within any required accuracy of the true answer.
Analogy
Discussion prompt
Explain CS tie: fixed-point iteration by analogy to something with no Foundations of Higher Mathematics in it at all — a queue, a recipe, a map, a bank balance, whatever fits. Then say where your analogy breaks.
Hint: An analogy that never breaks is not an analogy, it is the same idea wearing a hat. Find the seam — that is the part that is actually new.
Answer:
Iterative algorithms produce a sequence of guesses, and convergence is what it means for the algorithm to settle on an answer.
Intuition
A well-behaved iteration shrinks its error by a constant factor every step. Halving is the friendliest case: the distance to the target is cut in two each round.
After enough rounds the error is below any tolerance you set. Turning that into a threshold N is exactly an epsilon-N proof.
Counterexample
Discussion prompt
A well-behaved iteration shrinks its error by a constant factor every step. Halving is the friendliest case: the distance to the target is cut in two each round.
That is stated as though it always holds. Do one of two things: produce a case where it fails, or say precisely what rules such a case out. "It just does" is not on the menu.
Hint: Hunt at the extremes first — zero, one, negative, empty, equal. If every extreme survives, the reason they survive is the proof.
Ranking
Put in order
Put the moves of An iteration that halves its error into the order they have to happen.
Why: These are the moves of the worked example in the order it makes them, and each one is set up by the one before it. Because two to the power n-minus-one is at least n, the error is at most one over n, a sequence we have fully tamed.
Worked example
Start at one and halve each step. Show the guesses converge to zero.
\[ a_1 = 1,\quad a_{n+1} = \tfrac{1}{2} a_n \implies a_n = \tfrac{1}{2^{\,n-1}} \]
Fix epsilon and bound the term by one over n
Why: Because two to the power n-minus-one is at least n, the error is at most one over n, a sequence we have fully tamed.
\[ 2^{\,n-1} \ge n \implies a_n = \tfrac{1}{2^{\,n-1}} \le \tfrac{1}{n} \]
Choose N larger than one over epsilon
Why: The same Archimedean threshold that beats one over n also beats the smaller error a-sub-n.
Show the tail is under epsilon
Why: For n at least N the error is at most one over n, which is at most one over N, under epsilon.
\[ n \ge N \implies a_n \le \tfrac{1}{n} \le \tfrac{1}{N} < \varepsilon \]
Verify with epsilon equal to one hundredth
Why: N equal to 101 works via the one-over-n bound; in fact the true error at n equal to eight is already one over 128, under 0.01. The algorithm provably settles to zero.
\[ a_8 = \tfrac{1}{128} \approx 0.0078 < 0.01 \]
Picture it
Animation
Shows: Each line of the worked example "An iteration that halves its error", appearing one at a time.
The same working the example does, in the order a tutor would write it.
Takeaway: N equal to 101 works via the one-over-n bound; in fact the true error at n equal to eight is already one over 128, under 0.01. The algorithm provably settles to zero.
Pattern
1. Let epsilon be given, arbitrary and positive
Why: The adversary moves first. Nothing you write may secretly depend on a particular epsilon.
2. Simplify the distance from the term to L
Why: Combine fractions and use the absolute value to get a clean expression in n.
3. Scratch: solve the target inequality for n
Why: This reveals how large n must be. It is search, not proof, so keep it off to the side.
4. Declare a natural threshold N using Archimedes
Why: Choose N at least as large as the bound your scratch work produced.
5. Prove the tail obeys the bound and conclude
Why: For every n at least N, run the algebra forward to reach a distance under epsilon. Since epsilon was arbitrary, the limit is established.
Pattern
Reciprocal bound
Why: To beat a one-over-n type distance, take N past one over epsilon by the Archimedean property.
Over-estimate to simplify
Why: Replacing the distance by something larger but simpler, such as bounding n-plus-two below by n, gives a cleaner N and is still valid.
Split the tolerance
Why: For a sum of two errors, control each to within epsilon-over-two so the total stays under epsilon.
Take the maximum of thresholds
Why: When several conditions each need their own threshold, the largest one makes them hold simultaneously.
Real world
Discussion prompt
Outside this lesson: where does Sequences & Convergence actually turn up? Name one concrete situation — a job, a piece of software someone ships, a decision somebody has to make — and say which part of Toolkit for choosing N is doing the work in it.
Hint: Vague is the failure mode here. "Engineering" is not a situation; "deciding whether this build is fast enough to ship" is.
Answer:
The epsilon-N definition of convergence read as a nested-quantifier game: given a tolerance, produce a threshold. Covers uniqueness of limits, convergent implies bounded, the algebra of limits, divergence to infinity, and the squeeze theorem.
Elimination
Eliminate the wrong options
What is the smallest natural number N such that every term from position N onward is within 0.01 of 0?
3 of these 4 are wrong. Strike them one at a time, and say what rules each one out before you strike the next. The survivor is the answer.
Survives elimination: A
Why: We need one over n strictly less than 0.01, which holds exactly when n is greater than 100. The first natural number satisfying that is 101, so the smallest working threshold is N equal to 101.
Check
Work with the sequence one over n and the tolerance one hundredth.
\[ a_n = \tfrac{1}{n}, \quad \varepsilon = 0.01, \quad L = 0 \]
Check your understanding
What is the smallest natural number N such that every term from position N onward is within 0.01 of 0?
Answer: A
Why: We need one over n strictly less than 0.01, which holds exactly when n is greater than 100. The first natural number satisfying that is 101, so the smallest working threshold is N equal to 101.
Prediction
Predict first
Which of these sequences is bounded yet does NOT converge?
Answer it in your own words, now, with nothing to choose from. The options are on the next slide — and picking the right one off a list is an easier skill than producing it.
Correct: The sequence whose n-th term is negative-one to the n
Why: The alternating sequence stays inside the interval from negative one to one, so it is bounded, but it hops forever between 1 and negative 1 and settles on no single value, so it diverges.
Check
Recall that convergent forces bounded, but not the reverse.
Check your understanding
Which of these sequences is bounded yet does NOT converge?
Answer: A
Why: The alternating sequence stays inside the interval from negative one to one, so it is bounded, but it hops forever between 1 and negative 1 and settles on no single value, so it diverges.
Check
Evaluate the limit of a ratio of quadratics.
\[ \lim_{n\to\infty} \frac{2n^2 + 3n}{n^2 + 1} \]
Check your understanding
What is the value of this limit?
Answer: A
Why: Dividing numerator and denominator by n squared gives two plus three-over-n over one plus one-over-n-squared, which tends to two over one. With equal degrees the limit is the ratio of leading coefficients, namely 2.
Check
Consider the sequence whose n-th term is cosine of n divided by n.
Check your understanding
What is the limit of this sequence?
Answer: A
Why: Since cosine of n always lies between negative one and one, the term is squeezed between negative one over n and one over n. Both bounds tend to 0, so by the squeeze theorem the limit is 0.
Prediction
Predict first
Which statement correctly defines the sequence converging to L?
Answer it in your own words, now, with nothing to choose from. The options are on the next slide — and picking the right one off a list is an easier skill than producing it.
Correct: For every epsilon greater than 0 there exists a threshold N such that for all n at least N, the distance from a-sub-n to L is less than epsilon.
Why: Convergence requires the threshold N to be chosen after and depending on epsilon, with the bound holding for the whole tail. That is the order for-all-epsilon, then there-exists-N, then for-all-n at least N.
Check
The quantifier order in the definition of convergence is not negotiable. Pick the correct statement.
Check your understanding
Which statement correctly defines the sequence converging to L?
Answer: A
Why: Convergence requires the threshold N to be chosen after and depending on epsilon, with the bound holding for the whole tail. That is the order for-all-epsilon, then there-exists-N, then for-all-n at least N.
Elimination
Eliminate the wrong options
Which reasoning correctly proves this sequence diverges?
3 of these 4 are wrong. Strike them one at a time, and say what rules each one out before you strike the next. The survivor is the answer.
Survives elimination: A
Why: Using the negated definition with tolerance 1: past any threshold the tail contains both 1 and negative 1, whose separation is 2, so they cannot both sit within distance 1 of a common L. Hence no limit exists.
Check
Consider again the sequence whose n-th term is negative-one to the n.
Check your understanding
Which reasoning correctly proves this sequence diverges?
Answer: A
Why: Using the negated definition with tolerance 1: past any threshold the tail contains both 1 and negative 1, whose separation is 2, so they cannot both sit within distance 1 of a common L. Hence no limit exists.
Connect it up
Draw it
One page, no notation unless you need it: draw how these connect — The epsilon-N proof recipe · Toolkit for choosing N · A sequence is a function on the naturals · An endless numbered list · Order and repetition matter. Put an arrow wherever one of them is what makes another possible, and label the arrow with why.
Recap
Convergence is a game with a fixed quantifier order: for every tolerance epsilon, there is a threshold N, past which the whole tail sits within epsilon of the limit. You produce N from epsilon, never the other way around.
You can now run the epsilon-N recipe on reciprocal and rational sequences, and you have proved the pillars: limits are unique, convergent sequences are bounded, and limits respect sums, products, and quotients with a nonzero denominator.
You can bound hard sequences with the squeeze theorem, handle divergence to infinity, and diagnose the traps: fixing N first, mistaking bounded for convergent, dividing by a zero limit, and thinking neighbors bunching up is enough without completeness.
| Result | One-line takeaway |
|---|---|
| epsilon-N | given the tolerance, build the threshold |
| Uniqueness | two limits force a gap smaller than itself |
| Convergent implies bounded | tame the tail, cap the finite head |
| Squeeze | trap between two sequences with a shared limit |
Next comes the machinery that turns these hypotheses into guarantees: monotone convergence, Bolzano-Weierstrass, and Cauchy sequences, all resting on the completeness of the reals.
Want this taught 1-on-1? Alexander tutors Foundations of Higher Mathematics — $55/session, free consultation.