Real analysis and proof habits

Real analysis rebuilds calculus on definitions that a skeptical reader can check. "Gets close" becomes "for every tolerance, there is a stage after which the terms stay inside it." The point is not pedantry. Some pictures survive the definition, and some pictures quietly lie. This pillar is a map of that first encounter: sequences, limits, series, and the habits that make a short proof hold. The intro to proofs pathway is where you write the arguments. This page says what the words are and in what order to meet them.

Who this pillar is for

Start here if a proof course is in front of you, or if your calculus so far was mostly computation. You might differentiate easily and freeze when a definition says "for every epsilon there exists a delta." That freeze is the audience. You might also be meeting series tests as a list of recipes and wanting the hypothesis each test actually uses.

If you still need the computational map of derivatives and integrals, read calculus foundations in parallel. If you want three written checkpoints with a suggested rewrite, follow intro to proofs after this page. Reading either page is not the same as producing a proof.

Prerequisites

You need algebra, function notation, and the willingness to define a word before you use it. The examples are limits, continuity, sequences, and series. You do not need a prior course in symbolic logic. The quantifiers you need are the ones a limit already uses, and the order is part of the meaning.

If limits are new, start with calculus foundations and the glossary entries for limit and continuity. If derivative language is rusty, repair it with what a derivative actually measures, then come back. A full development, including the least upper bound property with proofs, belongs to a book. Abbott's Understanding Analysis and MIT OpenCourseWare are named on the sources page. This pillar will not invent a construction of the real numbers.

What you should be able to do

When you finish the core, you should be able to do the following with a short statement.

• Unpack a definition into quantifiers before any algebra.
• Write an epsilon delta argument for a linear limit, in the forward direction, with the tolerance given first.
• Say why a single counterexample kills a universal claim, and why many examples never prove one.
• Tell a sequence from a series, and say what a partial sum is.
• Name the hypothesis of the comparison test, the ratio test, and the root test, including the case where a test stays silent.
• Separate scratch work from the write up you would hand to a classmate.

Sequences, limits, and what "eventually" means

A sequence $(a_n)$ converges to $L$ when the terms eventually stay inside every neighborhood of $L$, no matter how tight. Formally: for every $\varepsilon>0$ there exists an index $N$ such that $n>N$ implies $|a_n-L|<\varepsilon$. The index $N$ may depend on $\varepsilon$. A smaller tolerance is allowed to wait longer. That dependence is the whole idea of "eventually."

The sequence $a_n=1/n$ converges to $0$. Given $\varepsilon>0$, choose an integer $N>1/\varepsilon$. Then $n>N$ forces $1/n<\varepsilon$. The algebra is short. The order is the content: $\varepsilon$ is handed to you, and only then do you name $N$.

On the real numbers, a bounded monotone sequence converges. For an increasing sequence bounded above, the limit is the least upper bound of the set of terms. This pillar will not construct the reals. An outline is the right ambition. The full argument is in a book on the sources page, and intro to proofs asks you to write that outline.

Epsilon delta, and the order of the quantifiers

The epsilon delta definition of $\lim_{x\to a}f(x)=L$ says: for every $\varepsilon>0$ there exists $\delta>0$ such that $0<|x-a|<\delta$ implies $|f(x)-L|<\varepsilon$. The output tolerance comes first. The input window is yours to choose, and it may depend on $\varepsilon$. The point $x=a$ is excluded. A limit does not require $f(a)$ to exist. Continuity at $a$ adds that $f(a)$ exists and equals the limit.

Epsilon delta definition and delta finder is the long version of that paragraph. Here is a short argument you can try before that page, or after it as a second function. Claim: $\lim_{x\to 1}(2x+3)=5$. Let $\varepsilon>0$. Then $|(2x+3)-5|=2|x-1|$. Choose $\delta=\varepsilon/2$. If $0<|x-1|<\delta$, then $2|x-1|<\varepsilon$. The write up runs in that order. Scratch work may start from the goal and solve for $\delta$. The polished paragraph starts from $\varepsilon$.

The numerical limit estimator can sample a hole, where nearby values settle and the function is undefined at the point. That picture matches the definition, because the definition never consults the point. The epsilon delta finder takes the next step. You propose $L$ and $\varepsilon$, and it searches for a large $\delta$ that keeps sampled outputs inside the band. Compare that window with a $\delta$ you can prove, such as $\varepsilon/3$ for a line of slope $3$. The samples still do not replace the forward write up. The FAQ on calculators versus understanding says the same thing.

Series as sequences of partial sums

A series $\sum a_n$ means the sequence of partial sums, and convergence of the series means those partial sums settle. You do not add infinitely many numbers in one gesture. You ask whether the finite totals approach a limit. Read sequences, series, and when infinite sums make sense until that sentence is automatic.

Terms going to zero are required, because late kicks would keep the total from settling. The converse is false. The harmonic series $\sum 1/n$ has terms that vanish, and the partial sums still wander off. One counterexample kills "terms go to zero, so the series converges."

The comparison test needs nonnegative terms and an inequality you prove. The ratio test and the root test conclude convergence when their limit is less than $1$, divergence when it is greater than $1$, and silence when it equals $1$. Convergence tests for series without fog spells out the hypotheses. The sequence calculator adds finitely many terms. A finite total is not a convergence verdict. A Taylor polynomial is a different finite object. It matches a function and its derivatives at one center, and a remainder bound is not a convergence test. The Taylor polynomial calculator builds that polynomial for the standard functions.

Proof habits that transfer

These habits keep a short argument checkable. Intro to proofs turns them into three checkpoints: a linear limit, the sum of two continuous functions, and an outline of monotone convergence.

• Unpack definitions. Write the epsilon delta or epsilon $N$ claim before any algebra.
• Keep scratch and the write up apart. Scratch may run backward. The paragraph you keep runs forward.
• Name quantifiers. "For all" and "there exists" are different, and order matters.
• Use counterexamples on purpose. One legal counterexample kills a universal claim. A pile of examples does not prove one.
• End by naming the result. The last sentence should say what was established.

Suggested learning order

Work in this order. One definition per sitting is enough. The pathway repeats this arc and then asks you to write.

1. Glossary: limit, epsilon delta, continuity, sequence, partial sum, series. Say which word is a list and which is a sum.
2. Epsilon delta intuition. Write the limit of $2x+3$ at $x=1$ in the forward direction.
3. The limit estimator only as a picture, and the epsilon delta finder when you already have an $\varepsilon$. Then write a sentence a table cannot prove.
4. Intro to proofs for the three checkpoints, including the rewrite.
5. Sequences and series, with the sequence calculator used only for a finite sum.
6. Convergence tests with the comparison, ratio, and root tests. Write the silent case for each.
7. Theorems this site only outlines: the sources page.

Checkpoints and practice habits

Keep a polished column and a counterexample column. Scratch can be messy. The polished column cannot.

Quantifier check. For $a_n=1/n$ and $\varepsilon=0.01$, an index $N=100$ works, and so does any larger index. An index that depends on a specific decimal is still an existence proof for that one tolerance. The definition asks you to be ready for every tolerance.

Limit check. Finish $\lim_{x\to 1}(2x+3)=5$ without looking. If your first sentence is "let $\delta$ equal," you started in the middle. Start with "let $\varepsilon>0$."

Converse check. "Convergent series have terms going to zero" is true. "Terms going to zero implies the series converges" is false. The harmonic series is a counterexample. Write it in the counterexample column so the two directions stay distinct.

Test check. For $\sum n^{2}/3^{n}$, the ratio of successive absolute terms tends to $1/3$, which is less than $1$, so the ratio test says the series converges absolutely. State the limit before you open a tool. The sequence calculator will not take this limit for you.

Common pitfalls

In a universal claim, examples are scratch. One legal counterexample is the decisive object. If you cannot say what would count as a counterexample, you do not yet know what the claim says.

Related reading

Write next on intro to proofs. Computation lives on calculus foundations and the AP Calculus AB pathway. The same habits apply to a short linear dependence argument on the linear algebra pillar. Routes are compared on the pathways hub.