The Epsilon-N Definition of Convergence

From the Neighborhood Picture to a Precise Test

The previous tutorial described convergence as a statement about the tail: however small a neighborhood of a proposed limit is chosen, all sufficiently late terms lie inside it. The epsilon-N definition writes that idea with explicit quantifiers. It specifies what “however small” means, what “sufficiently late” means, and which indices must satisfy the required closeness.

Let \((a_n)\) be a real sequence and let \(L\in\mathbb{R}\). The central quantity is the distance from the \(n\)th term to \(L\), namely \(|a_n-L|\). A positive number \(\varepsilon\) specifies how small that distance must be. The index \(N\) is a threshold: once \(n\) reaches \(N\), every later term must meet the distance requirement.

Read the order of the quantifiers carefully: for every positive \(\varepsilon\), there exists a threshold \(N\), and for every index \(n\geq N\), the distance is less than \(\varepsilon\). The threshold is allowed to depend on the tolerance. In general, a smaller tolerance may require a larger threshold. But after a threshold has been chosen, it must work for every index from there onward, not just for a selection of terms.

The inequality is strict. A term whose distance is exactly \(\varepsilon\) does not satisfy the requirement for that tolerance. Also, \(N\) itself is included: the condition is for \(n\geq N\), not only for \(n>N\). These details matter when checking a proposed threshold at its first index.

Finding and Checking a Threshold

A proof using the definition usually begins by taking an arbitrary \(\varepsilon>0\). One then solves or estimates the inequality \(|a_n-L|<\varepsilon\) to find how large \(n\) needs to be. Finally, one chooses a nonnegative integer threshold \(N\) that guarantees the inequality for every \(n\geq N\). The threshold need not be the smallest possible one; it only needs to work.

A common proof gap is to show only that the distance becomes small “as \(n\) grows,” without identifying how a chosen \(\varepsilon\) leads to an \(N\). Another is to check the inequality at \(n=N\) but not explain why it holds for every larger index. The definition requires both the threshold and the whole tail estimate.

Two Useful Equivalent Tests

The definition can be reorganized in ways that are often easier to use. One test replaces all positive tolerances by a particular countable list. Another describes convergence in terms of indices where the desired inequality fails. Neither changes the meaning of convergence; each makes a different part of the quantifiers easier to handle.

Proof. First suppose \(a_n\to L\). Let \(k\) be any positive integer. Since \(\dfrac{1}{k}>0\), the definition applied with \(\varepsilon=\dfrac{1}{k}\) gives an \(N\in\mathbb{N}_0\) such that \(|a_n-L|<\dfrac{1}{k}\) for every \(n\geq N\). This proves the stated condition for every \(k\).

Conversely, suppose the stated condition holds for every positive integer \(k\), and let \(\varepsilon>0\). By the Archimedean property, choose a positive integer \(k\) such that \(\dfrac{1}{k}<\varepsilon\). The assumed condition for this \(k\) gives an \(N\) such that, for every \(n\geq N\),

Thus for this arbitrary \(\varepsilon>0\), a threshold works for every \(n\geq N\). By the epsilon-N definition, \(a_n\to L\). \(\square\)

This criterion is useful when a proof naturally establishes bounds such as \(1\), \(\dfrac12\), \(\dfrac13\), and so on. It is enough to establish eventual closeness for every reciprocal integer: any positive tolerance, however small, is larger than one of those reciprocal tolerances.

Proof. Suppose first that \(a_n\to L\), and fix \(\varepsilon>0\). By definition, there is an \(N\in\mathbb{N}_0\) such that \(|a_n-L|<\varepsilon\) whenever \(n\geq N\). No such \(n\) belongs to \(B_\varepsilon\). Hence \(B_\varepsilon\subseteq\{0,1,\ldots,N-1\}\), which is finite (and is empty if \(N=0\)). So \(B_\varepsilon\) is finite.

Conversely, suppose \(B_\varepsilon\) is finite for every \(\varepsilon>0\), and fix an arbitrary \(\varepsilon>0\). If \(B_\varepsilon\) is empty, take \(N=0\); then every index satisfies \(|a_n-L|<\varepsilon\). If \(B_\varepsilon\) is nonempty, it is a finite set of nonnegative integers and therefore has a largest element \(m\). Take \(N=m+1\). For every \(n\geq N\), we have \(n>m\), so \(n\notin B_\varepsilon\). By the definition of \(B_\varepsilon\), this means \(|a_n-L|<\varepsilon\). In either case there is a threshold for the chosen \(\varepsilon\). Since \(\varepsilon\) was arbitrary, \(a_n\to L\). \(\square\)

This formulation captures “eventually” as “all failures are confined to a finite set.” The set of exceptional indices may depend on \(\varepsilon\), and it can contain early terms far from the limit. What convergence rules out is infinitely many failures for even one fixed positive tolerance.

One Tolerance Can Disprove Convergence

The definition also gives a direct way to show that a sequence does not converge to a proposed value: find one positive tolerance for which no threshold works. To show that a sequence does not converge at all, the argument must rule out every proposed limit. The next example does this by showing that terms eventually move farther away from any fixed real number.

What the Quantifiers Do—and Do Not—Require

The epsilon-N definition does not say that every term is close to the limit. It permits any finite number of early terms to lie outside a chosen tolerance. Nor does it say that the sequence must become equal to its limit or move steadily toward it. It says that for each requested tolerance, one threshold places the entire remaining tail within that tolerance.

The order of the quantifiers cannot be reversed. The threshold may depend on \(\varepsilon\), but the index \(n\) ranges over every index at or beyond that threshold. A proof that finds a suitable index for each tolerance but does not control all later terms is incomplete. The intuitive neighborhood picture from the previous tutorial is precisely this uniform control of the tail: one \(N\) works for every \(n\geq N\) at the chosen \(\varepsilon\).

When reading or writing a convergence proof, identify four parts: the proposed limit \(L\), an arbitrary positive \(\varepsilon\), a threshold \(N\) chosen after \(\varepsilon\), and an estimate valid for every \(n\geq N\). Keeping those roles separate prevents the most common quantifier errors and turns the intuitive statement into a checkable argument.