Equivalent Definitions of Convergence

One Idea, Several Useful Tests

A sequence converges to $L$ when its terms eventually remain as close to $L$ as any prescribed positive tolerance requires. The epsilon definition expresses this with three nested quantifiers: for every $\varepsilon>0$, there is an index $N$ such that every $n\geq N$ satisfies $|a_n-L|<\varepsilon$. The order of these quantifiers matters. The index may depend on the tolerance, but once chosen it must work for every later term.

That definition can be restated in several equivalent ways. One version speaks of neighborhoods of $L$, another uses only a particular countable collection of tolerances, and a further version measures the largest error remaining in each tail. These formulations describe the same behavior, but each can make a different proof shorter or more transparent.

Definition: A sequence $(a_n)$ converges to $L\in\mathbb{R}$ if, for every $\varepsilon>0$, there exists $N\in\mathbb{N}_0$ such that $|a_n-L|<\varepsilon$ for every $n\geq N$.

Neighborhoods and Intervals

An open neighborhood of $L$ in $\mathbb{R}$ is an open set containing $L$. In particular, for every $\varepsilon>0$, the interval $(L-\varepsilon,L+\varepsilon)$ is a neighborhood of $L$. The epsilon definition says exactly that the sequence is eventually inside every such interval.

Theorem (Neighborhood Characterization of Convergence): A sequence $(a_n)$ converges to $L$ if and only if, for every open neighborhood $U$ of $L$, there exists $N\in\mathbb{N}_0$ such that $a_n\in U$ for every $n\geq N$. Equivalently, every open interval containing $L$ eventually contains all terms of the sequence.

Proof. Suppose first that $a_n\to L$, and let $U$ be an open neighborhood of $L$. Since $U$ is open and $L\in U$, there is a $\delta>0$ such that $(L-\delta,L+\delta)\subseteq U$. By convergence, there is an $N$ such that $|a_n-L|<\delta$ whenever $n\geq N$. The absolute-value inequality implies $L-\delta<a_n<L+\delta$, so $a_n\in U$ for every $n\geq N$.

Conversely, suppose every open neighborhood of $L$ eventually contains all terms. Given $\varepsilon>0$, take $U=(L-\varepsilon,L+\varepsilon)$. There is an $N$ such that $a_n\in U$ whenever $n\geq N$. Membership in this interval is equivalent to $|a_n-L|<\varepsilon$, which is the epsilon definition of convergence. The same reasoning shows that it is enough to test open intervals containing $L$, since every open neighborhood contains one of these intervals. $\square$

Worked Example: Convergence of a Rational Expression

Let $a_n=\frac{n}{n+1}$, where $n\in\mathbb{N}_0$. To test convergence to $1$, compute the error exactly: $$ |a_n-1|=\left|\frac{n}{n+1}-1\right|=\frac{1}{n+1}. $$ Given $\varepsilon>0$, choose $N\in\mathbb{N}_0$ such that $N+1>1/\varepsilon$. For $n\geq N$, we have $n+1\geq N+1>1/\varepsilon$, and hence $$ |a_n-1|=\frac{1}{n+1}\leq\frac{1}{N+1}<\varepsilon. $$ Thus $a_n\to1$.

The neighborhood formulation describes the same verification without referring first to an arbitrary error. For any open interval $(c,d)$ containing $1$, both $1-c$ and $d-1$ are positive. Set $\delta=\min\{1-c,d-1\}>0$. When $n$ is large enough that $1/(n+1)<\delta$, we have $1-\delta<a_n<1+\delta$, and therefore $c<a_n<d$. So every open interval containing $1$ eventually contains all terms.

A Countable Collection of Error Tests

The definition appears to require checking every positive real tolerance. In fact, it is enough to check the tolerances $1,1/2,1/3,\ldots$. These tolerances become arbitrarily small: given any positive $\varepsilon$, the Archimedean property provides a positive integer $k$ such that $1/k<\varepsilon$. An error smaller than $1/k$ is therefore also smaller than $\varepsilon$.

Theorem (Countable Error Test for Convergence): A sequence $(a_n)$ converges to $L$ if and only if, for every positive integer $k$, there exists $N_k\in\mathbb{N}_0$ such that $|a_n-L|<1/k$ for every $n\geq N_k$.

Proof. If $a_n\to L$, apply the definition with $\varepsilon=1/k$ for each positive integer $k$. This supplies the required $N_k$.

For the reverse implication, let $\varepsilon>0$. Choose a positive integer $k$ such that $1/k<\varepsilon$. By the assumed test, there is an $N_k$ such that, whenever $n\geq N_k$, $$ |a_n-L|<\frac{1}{k}<\varepsilon. $$ This proves the epsilon definition for the original arbitrary $\varepsilon$, and therefore $a_n\to L$. $\square$

Worked Example: A Shrinking Alternating Error

Consider $b_n=4+\frac{(-1)^n}{n+2}$ for $n\in\mathbb{N}_0$. Since $|(-1)^n|=1$, $$ |b_n-4|=\frac{1}{n+2}. $$ Fix a positive integer $k$. Choose $N_k\in\mathbb{N}_0$ with $N_k+2>k$. For $n\geq N_k$, we then have $$ |b_n-4|=\frac{1}{n+2}\leq\frac{1}{N_k+2}<\frac{1}{k}. $$ The countable error test proves $b_n\to4$. The alternating sign does not interfere with convergence because convergence depends on the magnitude of the error, not on whether the terms lie above or below the limit.

The Supremum of the Errors in a Tail

A tail of a sequence consists of all terms from some index onward. For a bounded sequence and a proposed limit $L$, define the tail error $$ E_N=\sup\{|a_n-L|:n\geq N\}. $$ Unlike the epsilon definition, which checks each term against a tolerance, $E_N$ records the largest error in the entire tail. As $N$ increases, the set of terms being considered gets smaller, so $E_N$ cannot increase.

Theorem (Tail-Supremum Characterization of Convergence): Suppose $(a_n)$ is a bounded real sequence and $L\in\mathbb{R}$. Then $a_n\to L$ if and only if $E_N\to0$, where $E_N=\sup\{|a_n-L|:n\geq N\}$.

Proof. Boundedness of $(a_n)$ ensures that each set of tail errors is bounded above, so $E_N$ is a finite nonnegative real number. Suppose first that $a_n\to L$. Given $\eta>0$, apply convergence with tolerance $\eta/2$. There is an $N$ such that $|a_n-L|<\eta/2$ for every $n\geq N$. Thus $\eta/2$ is an upper bound for the errors in that tail, and $$ 0\leq E_N\leq\frac{\eta}{2}<\eta. $$ For every $M\geq N$, the tail starting at $M$ is contained in the tail starting at $N$, so $0\leq E_M\leq E_N<\eta$. Therefore $E_N\to0$.

Conversely, suppose $E_N\to0$. Given $\varepsilon>0$, choose $N$ such that $E_N<\varepsilon$. For each $n\geq N$, the number $|a_n-L|$ belongs to the set whose supremum is $E_N$, and hence $|a_n-L|\leq E_N<\varepsilon$. This is the epsilon definition of $a_n\to L$. $\square$

Worked Example: Computing Tail Errors Exactly

For $b_n=4+\frac{(-1)^n}{n+2}$, the error at index $n$ is $1/(n+2)$. For every $n\geq N$, $$ |b_n-4|=\frac{1}{n+2}\leq\frac{1}{N+2}. $$ Equality occurs at $n=N$, so the supremum is attained and $$ E_N=\sup\left\{\frac{1}{n+2}:n\geq N\right\}=\frac{1}{N+2}. $$ Since $1/(N+2)\to0$, the tail-supremum characterization again gives $b_n\to4$. This formulation provides a uniform estimate: every term from index $N$ onward has error at most $1/(N+2)$.

Eventual Behavior and Finite Exceptions

All these criteria describe eventual behavior. The terms before the index $N$ may be far from $L$; convergence imposes no requirement that every term be close. This is why changing finitely many terms does not affect a sequence’s limit. It is also why a proof should not replace “for every $n\geq N$” with “for every $n$” unless the latter has actually been established.

Worked Example: A Large First Term Does Not Affect the Limit

Define $c_0=100$, and for $n\geq1$ define $c_n=2+1/n$. We claim $c_n\to2$. Given $\varepsilon>0$, choose an integer $N\geq1$ such that $1/N<\varepsilon$. For every $n\geq N$, $$ |c_n-2|=\frac{1}{n}\leq\frac{1}{N}<\varepsilon. $$ Thus the sequence converges to $2$, regardless of the value $c_0=100$. The interval criterion gives the same conclusion: every open interval containing $2$ contains all sufficiently late terms, although it need not contain the first term.

Choosing the Most Useful Formulation

The epsilon definition is the most direct choice when an explicit estimate for $|a_n-L|$ is available. The countable test is useful when estimates naturally involve $1/k$, or when one wants to replace an uncountable family of tolerances by a countable one. The neighborhood formulation is natural when the problem is stated in terms of open sets or intervals. The tail-supremum test is useful when a uniform bound on every error in a tail is available.

These are equivalent descriptions, not different notions of convergence. A proof may move between them, provided the quantifiers and inequalities are handled correctly. In particular, the index can depend on the tolerance, and a bound that holds for one tail does not automatically hold for earlier terms. Keeping those points visible makes convergence arguments both precise and easier to adapt.

Check Your Understanding

Use the equivalent formulations to answer the following questions.

Why does eventual membership in every interval containing $L$ imply the epsilon definition of convergence?
In the countable error test, how do you choose $k$ after an arbitrary tolerance $\varepsilon>0$ is given?
For a bounded sequence, why does convergence imply that the tail suprema $E_N$ tend to zero?
Why is it important that an epsilon estimate hold for every $n\geq N$, rather than only for $n=N$?
Can a sequence converge to $L$ if its first term is far from $L$? Explain using the eventual condition.