Introduction
A sequence is an ordered list of terms, indexed by the positive integers. Its first few terms can look irregular, large, or unrelated to its later behavior. Convergence is concerned with what happens after those initial terms: do all sufficiently late terms lie as close as desired to one fixed real number?
This focus on the late terms is essential. A limit is not determined by a finite collection of values, nor can a finite change at the start of a sequence destroy convergence. The precise definition makes this idea clear.
Neighborhoods of a Real Number
A neighborhood of a real number \(L\) is a set containing an open interval centered at \(L\). The intervals we will use are of the form \((L-\varepsilon,L+\varepsilon)\), where \(\varepsilon>0\). A term \(a_n\) lies in this interval exactly when its distance from \(L\) is less than \(\varepsilon\).
The same definition can be stated in neighborhood language: every neighborhood of \(L\) contains all sufficiently late terms of the sequence. The phrase all sufficiently late terms matters. It does not say that all terms must lie in the neighborhood. It says that there is a point in the sequence after which every term does.
The quantifiers also specify how the choice works. First, a positive tolerance \(\varepsilon\) is given, no matter how small. Then an index \(N\) must be found so that the inequality holds for every \(n\geq N\). The index may depend on \(\varepsilon\). The definition does not require one fixed \(N\) to work for every tolerance.
Worked Example: A Sequence That Is Eventually Constant
Worked Example: A Sequence That Is Eventually Constant
Define \(a_1=-20\), \(a_2=7\), and \(a_n=4\) for every \(n\geq 3\). Show that \(a_n\to 4\).
Let \(\varepsilon>0\) be arbitrary. Choose \(N=3\). If \(n\geq 3\), the definition of the sequence gives \(a_n=4\), so $$ |a_n-4|=|4-4|=0<\varepsilon. $$ This works for every positive \(\varepsilon\), so the sequence converges to \(4\).
Check: For example, if \(\varepsilon=0.01\), then \(n\geq 3\) gives \(|a_n-4|=0<0.01\). The first two terms do not satisfy this particular closeness condition, but the definition only requires it for every term from \(N\) onward.
Why the Beginning Does Not Determine the Limit
Suppose two sequences agree from some index onward, even if their first terms are very different. For a proposed limit \(L\), the definition only tests terms with \(n\geq N\). If the sequences have already become identical by that point, they give the same answer to every such test.
Here is a direct proof. Let the sequences be \((a_n)\) and \((b_n)\), and suppose they differ only at indices in a finite set \(F\). If \(F\) is empty, the sequences are identical. Otherwise, because \(F\) is finite, it has a largest index \(M\). After index \(M\), the terms agree.
Suppose \(a_n\to L\). For any \(\varepsilon>0\), the definition supplies an index \(N\) such that \(|a_n-L|<\varepsilon\) for all \(n\geq N\).
Choose an index \(N'\) at least as large as both \(N\) and \(M+1\). Then, whenever \(n\geq N'\), the terms agree: \(b_n=a_n\).
For every \(n\geq N'\), we have \(|b_n-L|=|a_n-L|<\varepsilon\). Since this holds for every \(\varepsilon>0\), \(b_n\to L\).
The same reasoning with the roles of \(a_n\) and \(b_n\) reversed proves the converse. Thus finite changes preserve convergence in both directions, and whenever the sequences converge, they have the same limit. The proof works because every finite set of changed indices lies entirely before some later index.
Worked Example: Different Beginnings, the Same Tail
Worked Example: Different Beginnings, the Same Tail
Consider $$ a_1=100,\quad a_2=-3,\quad a_3=0,\quad a_n=6\text{ for }n\geq 4, $$ and $$ b_1=-1000,\quad b_2=2,\quad b_3=6,\quad b_n=6\text{ for }n\geq 4. $$ Show that both sequences converge to \(6\).
Let \(\varepsilon>0\) be arbitrary and choose \(N=4\). For every \(n\geq 4\), both sequences have value \(6\). Therefore $$ |a_n-6|=|6-6|=0<\varepsilon \quad\text{and}\quad |b_n-6|=|6-6|=0<\varepsilon. $$ So both sequences converge to \(6\), regardless of their different first three terms.
Check: At the specific tolerance \(\varepsilon=0.5\), the term \(a_1=100\) is not within \(0.5\) of \(6\), since \(|100-6|=94\). But every term with \(n\geq 4\) is within \(0.5\), since its distance is \(0\). This is exactly what convergence requires.
Convergence Requires the Whole Tail
Finite beginnings do not control convergence, but that does not mean a few late terms are enough. The condition must hold for every term after a suitable index. If terms keep returning far from a proposed limit, the sequence cannot converge to that value.
Worked Example: Persistent Oscillation
Worked Example: Persistent Oscillation
Consider \(a_n=(-1)^n\), so the sequence is \(-1,1,-1,1,\ldots\). Show that it does not converge to any real number.
Suppose, for contradiction, that it converges to some \(L\in\mathbb{R}\). Apply the definition with \(\varepsilon=\frac12\). There would have to be an index \(N\) such that every \(n\geq N\) satisfies $$ |a_n-L|<\frac12. $$ There are even indices larger than \(N\), and their terms equal \(1\). There are also odd indices larger than \(N\), and their terms equal \(-1\). The condition would therefore require both $$ |1-L|<\frac12 \quad\text{and}\quad |-1-L|<\frac12. $$ The first inequality says \(\frac12<L<\frac32\). The second says \(-\frac32<L<-\frac12\). These intervals do not overlap, so no real number \(L\) can satisfy both requirements. This contradicts the assumed convergence.
Check: No matter how large \(N\) is, both an even and an odd index occur after it. Thus the two values \(1\) and \(-1\) keep appearing in the tail; the oscillation does not disappear.
What the Definition Does—and Does Not—Say
The definition makes several distinctions worth keeping clear.
- Terms need not equal the limit. They need to become arbitrarily close to it in the stated sense.
- Early terms may be exceptions. The index \(N\) can be chosen after any finite collection of irregular terms.
- One successful tolerance is not enough. The condition must hold for every \(\varepsilon>0\), including arbitrarily small tolerances.
- One late term is not enough. Once \(N\) is chosen, the inequality must hold for every \(n\geq N\), not just for selected indices.
A common mistake is to check only a few terms and conclude that the sequence converges because those terms appear close to a number. For instance, the terms of \((-1)^n\) include values equal to \(1\) repeatedly. Seeing a term equal to \(1\) does not establish convergence to \(1\). With \(\varepsilon=\frac12\), every term equal to \(-1\) violates the required inequality \(|a_n-1|<\frac12\), and such terms occur arbitrarily far along the sequence.
Why Tail Behavior Matters
Many sequence questions involve formulas that are complicated or exceptional for a small number of indices but have a simpler pattern later. The convergence definition gives a precise reason that such finite exceptions can be set aside: every neighborhood test begins at some index \(N\), and the terms before \(N\) are not part of that test.
At the same time, the definition prevents us from ignoring behavior that continues indefinitely. A sequence can have a long stretch of terms near a candidate limit and still fail to converge if later terms keep leaving a fixed neighborhood. The distinction is not between “early” and “late” in an informal sense; it is between finitely many exceptions and a condition that holds for every term in the tail.
Check Your Understanding
Use the definition and the idea of a tail to answer these questions.
- State the epsilon-N definition of \(a_n\to L\) using quantifiers.
- A sequence has \(a_n=2\) for every \(n\geq 10\). What is its limit, and what index works for every \(\varepsilon>0\)?
- Why can changing the first five terms of a convergent sequence not change its limit?
- What part of the convergence definition fails for the sequence \(a_n=(-1)^n\) with candidate limit \(1\) and \(\varepsilon=\frac12\)?
- Does convergence require every term to lie in every neighborhood of the limit, or only every term after some index? Explain.