Reading an Epsilon-N Definition

Read the Quantifiers Before the Inequality

The previous tutorial stated the epsilon-N definition of convergence and gave equivalent tests. This tutorial focuses on how to read that definition accurately. A common difficulty is not the distance estimate itself, but keeping track of which quantities are chosen, which ones are universally quantified, and how many indices a chosen threshold must control.

For a real sequence \((a_n)\) and a proposed limit \(L\), the definition says: for every \(\varepsilon>0\), there exists an \(N\in\mathbb{N}_0\) such that for every \(n\in\mathbb{N}_0\), if \(n\geq N\), then \(|a_n-L|<\varepsilon\). Each part has a distinct role. The tolerance \(\varepsilon\) is chosen first. The threshold \(N\) may depend on that tolerance. After \(N\) is chosen, the inequality must hold for every index \(n\) at or beyond it.

Reading the Epsilon-N Definition: For each requested positive tolerance \(\varepsilon\), convergence requires one threshold \(N\) that controls the entire tail: \(|a_n-L|<\varepsilon\) for every \(n\geq N\). The threshold may depend on \(\varepsilon\), but it cannot depend on which later index \(n\) is being checked.

The phrase “there exists an \(N\)” does not mean that there is one threshold that must work for every possible tolerance. It means that after a particular \(\varepsilon\) is specified, a suitable threshold can be selected. A smaller tolerance may require a different, possibly larger, threshold. In contrast, once that threshold has been selected, it must work for all later indices at once—not just for one index or for a chosen subsequence.

The condition applies at \(n=N\), because the comparison is \(n\geq N\). It also requires a strict inequality: a distance equal to \(\varepsilon\) does not meet the condition. These details make a proposed threshold directly testable.

What a Threshold Does—and Does Not—Say

A threshold is a starting point for a tail, not a claim about every term in the sequence. Terms with indices less than \(N\) are not required to satisfy the tolerance estimate. They may be far from \(L\), and their number may depend on the chosen tolerance. The definition makes a claim about all terms from \(N\) onward, not about the sequence’s entire history.

Theorem (Larger Thresholds Also Work): Suppose that for some \(\varepsilon>0\) and \(N\in\mathbb{N}_0\), \(|a_n-L|<\varepsilon\) for every \(n\geq N\). Then the same inequality holds for every \(n\geq M\), for every \(M\in\mathbb{N}_0\) with \(M\geq N\).

Proof. Let \(M\geq N\), and take any \(n\in\mathbb{N}_0\) such that \(n\geq M\). Then \(n\geq M\geq N\), so the assumed property of \(N\) gives \(|a_n-L|<\varepsilon\). Since this holds for every \(n\geq M\), the threshold \(M\) also works for \(\varepsilon\). \(\square\)

This result explains why a threshold need not be the smallest possible one. Once a threshold works, moving it farther out cannot introduce a failing index into the tail. The theorem does not say that smaller thresholds work; they include additional terms that may fail the estimate.

Worked Example: Reading a Threshold for a Constant Sequence

Let \(a_n=4\) for every \(n\in\mathbb{N}_0\), and consider the proposed limit \(L=4\). For an arbitrary \(\varepsilon>0\), the distance at every index is

|a_n-L|=|4-4|=0<\varepsilon.

Thus \(N=0\) works: every index \(n\in\mathbb{N}_0\) satisfies \(n\geq0\), and the inequality holds at every such index. By the larger-threshold theorem, \(N=7\), or any other nonnegative integer, also works for this same \(\varepsilon\). The definition requires the existence of a working threshold; it does not require that the threshold be unique or minimal.

Checking What a Proposed Statement Really Claims

When reading a convergence claim, it helps to rewrite it in ordinary language without changing its quantifiers. For example, “for every \(\varepsilon>0\), there exists \(N\)” means that the threshold can be selected after the tolerance is known. “For every \(n\geq N\)” means that the estimate must hold at every index in the tail. Replacing that last phrase with “for some \(n\geq N\)” would make a much weaker claim: it would only require one successful term in the tail.

Likewise, one threshold that works for all \(\varepsilon>0\) would be a stronger requirement than the definition. Convergence allows the threshold to change with the tolerance. These distinctions are especially important when inspecting a proof: ask whether the proposed \(N\) was chosen for the tolerance at hand, and whether the subsequent estimate applies to an arbitrary index beyond \(N\).

Worked Example: Verifying a Particular Tail Claim

Define \(b_n=5+\dfrac{1}{n+1}\) for \(n\in\mathbb{N}_0\). Consider the specific claim that for \(\varepsilon=\dfrac{1}{10}\), the threshold \(N=10\) works for \(L=5\). For any \(n\geq10\), we have \(n+1\geq11\), and therefore

|b_n-5| =\left|5+\frac{1}{n+1}-5\right| =\frac{1}{n+1} \leq\frac{1}{11} <\frac{1}{10}.

This verifies the claim for every \(n\geq10\), including \(n=10\). At that first index, the distance is \(1/11\), which is strictly less than \(1/10\). The calculation checks a particular tolerance and a particular proposed threshold; by itself, it does not verify the definition for every positive tolerance. The definition asks for a suitable threshold separately for each \(\varepsilon>0\).

The distinction between “one tail index” and “every tail index” is not cosmetic. A single term might happen to be close to \(L\), even when later terms are not. A convergence threshold must rule out every failure after its starting index. The same care applies to the inequality sign: if the definition requires a distance strictly less than \(\varepsilon\), proving only that the distance is at most \(\varepsilon\) may not be enough.

Worked Example: A Failed Estimate at the Chosen Tolerance

Let \(c_n=(-1)^n\), and test the proposed limit \(L=0\) using \(\varepsilon=\dfrac12\). At every index \(n\in\mathbb{N}_0\), the value \(c_n\) is either \(1\) or \(-1\), so

|c_n-0|=|(-1)^n|=1\geq\frac12.

For any proposed threshold \(N\), the index \(n=N\) satisfies \(n\geq N\) but fails \(|c_n-0|<\dfrac12\). Thus no threshold works for this tolerance, and the definition does not hold for the proposed limit \(0\). This argument rules out \(0\) as a limit; to show that a sequence has no limit, one must rule out every proposed real value.

Reading the Negation: What Failure Requires

The quantifiers also give an exact way to describe failure. To disprove convergence to a particular \(L\), it is not enough to say that the terms do not seem to settle down. One must identify a positive tolerance for which every proposed threshold has a failing index beyond it. The failing index may depend on the proposed threshold.

Theorem (Failure-Witness Criterion): A real sequence \((a_n)\) does not converge to \(L\in\mathbb{R}\) if and only if there exists an \(\varepsilon_0>0\) such that for every \(N\in\mathbb{N}_0\), there is an \(n\in\mathbb{N}_0\) with \(n\geq N\) and \(|a_n-L|\geq\varepsilon_0\).

Proof. The definition of convergence to \(L\) has the logical form: for every \(\varepsilon>0\), there exists \(N\in\mathbb{N}_0\) such that for every \(n\in\mathbb{N}_0\), if \(n\geq N\), then \(|a_n-L|<\varepsilon\). Negating the first quantifier gives the existence of some \(\varepsilon_0>0\). Negating the existence of \(N\) says that for every \(N\in\mathbb{N}_0\), the required tail condition fails. That failure means there is an \(n\in\mathbb{N}_0\) with \(n\geq N\) for which \(|a_n-L|<\varepsilon_0\) is false. Since distances are real numbers, this is equivalent to \(|a_n-L|\geq\varepsilon_0\). These are exactly the conditions in the theorem. Each step reverses a quantifier or negates the stated inequality, so the implications work in both directions. \(\square\)

The order of the quantifiers in this criterion matters. A single tolerance \(\varepsilon_0\) must work against every proposed threshold, but the bad index \(n\) may change when \(N\) changes. This describes failures occurring arbitrarily far out in the sequence. A failing term among the initial indices alone is not enough: a later threshold could simply start after it.

Worked Example: The Alternating Sequence Has No Limit

Let \(c_n=(-1)^n\), and let \(L\in\mathbb{R}\) be any proposed limit. We show that the failure-witness criterion applies with \(\varepsilon_0=1\). Given any \(N\in\mathbb{N}_0\), the consecutive indices \(N\) and \(N+1\) have opposite parity. Thus the corresponding terms are \(1\) and \(-1\), in some order. The triangle inequality gives

2=|1-(-1)| \leq |1-L|+|L-(-1)|.

If both distances on the right were less than \(1\), their sum would be less than \(2\), contradicting this inequality. Therefore at least one of \(|1-L|\) and \(|-1-L|\) is at least \(1\). The index \(N\) gives one of the two values and the index \(N+1\) gives the other; both indices are at least \(N\). Hence there is an \(n\geq N\) with \(|c_n-L|\geq1\). Since \(N\) was arbitrary, the criterion shows that \(c_n\) does not converge to this \(L\). Because \(L\) was arbitrary, the sequence has no real limit.

A Practical Reading Checklist

When you encounter an epsilon-N statement, separate its parts before evaluating the argument:

Proposed limit: Which real number \(L\) is being tested?
Tolerance: Is \(\varepsilon\) arbitrary and positive, or has a particular value been fixed?
Threshold: Is \(N\) chosen after the tolerance, and is it a nonnegative integer?
Tail: Does the estimate hold for every \(n\geq N\), including \(n=N\)?
Failure: If convergence is being disproved, is one fixed positive tolerance paired with a failing index beyond every proposed threshold?

These checks prevent several common misreadings. The definition does not demand that all terms be close, only all terms in a suitable tail. It does not require one threshold to work for all tolerances. And it does not allow a proof to verify the inequality at just a few later indices. Convergence is a statement of uniform control over each tail once its threshold is chosen.

Check Your Understanding

Use the quantifiers and results in this tutorial to interpret the claims precisely.

In the epsilon-N definition, which choice may depend on \(\varepsilon\), and which part must then hold for every index in a tail?
If \(N\) works for a tolerance, why does every \(M\geq N\) also work for that same tolerance?
Does a threshold have to be the smallest possible one? Explain.
State the failure-witness criterion in words, identifying which quantity is fixed and which may vary with the threshold.
Why does finding one index that fails the estimate not, by itself, prove failure of convergence?