From “Getting Close” to a Precise Definition
A limit describes how the values of a function behave as its input approaches a target. Informally, we say that \(f(x)\) approaches \(L\) as \(x\) approaches \(a\) when values of \(f(x)\) can be made as close to \(L\) as we wish by taking \(x\) sufficiently close to \(a\).
The words “as close as we wish” and “sufficiently close” need precise meanings. The epsilon-delta definition supplies them. It separates two kinds of distance: the distance from an output to the proposed limit, and the distance from an input to the target.
The condition \(0<|x-a|<\delta\) says that \(x\) is within \(\delta\) of \(a\), but is not equal to \(a\). The conclusion \(|f(x)-L|<\varepsilon\) says that the corresponding output is within \(\varepsilon\) of \(L\). The definition therefore turns “input close” into a guarantee of “output close.”
What Epsilon and Delta Measure
The positive number \(\varepsilon\) is an allowed error in the output. It sets the required closeness to \(L\). The positive number \(\delta\) is an input distance that is sufficient to ensure that output accuracy. For a particular \(\varepsilon\), the definition requires that some suitable \(\delta\) can be found.
In the full definition, this implication must work for every positive \(\varepsilon\). A smaller \(\varepsilon\) asks for greater output accuracy and may require a smaller \(\delta\). There is no requirement that \(\delta\) equal \(\varepsilon\), or that one particular \(\delta\) work for every tolerance.
The symbols measure distances, not signed differences. For example, \(|f(x)-L|<\varepsilon\) means that the output lies between \(L-\varepsilon\) and \(L+\varepsilon\). Likewise, \(|x-a|<\delta\) means that the input lies between \(a-\delta\) and \(a+\delta\), with the target \(a\) itself excluded when considering a limit.
Why the Definition Captures a Limit
Suppose an output tolerance \(\varepsilon>0\) is given. The definition requires a positive input distance \(\delta\) that makes every permitted input \(x\) produce an output less than \(\varepsilon\) away from \(L\). Thus, however strict the requested output accuracy is, the input can be restricted enough to guarantee it.
Conversely, if some output tolerance cannot be guaranteed by any positive input distance, then the proposed value \(L\) is not the limit. There would always be inputs arbitrarily close to \(a\) whose function values fail to lie within that tolerance of \(L\). This explains the central logic: a limit is not established by one close input or a handful of samples. It is a guarantee about all domain inputs in a sufficiently small deleted interval around \(a\).
Worked Examples: Reading the Definition in Practice
Worked Example: A Tolerance for an Absolute-Value Function
Consider \(f(x)=|x|\) near \(a=0\), with proposed limit \(L=0\). Check whether \(\varepsilon=\frac{3}{10}\) and \(\delta=\frac{3}{10}\) give the required guarantee.
Take any \(x\) satisfying \(0<|x-0|<\frac{3}{10}\). Since \(f(x)=|x|\), the output error is
Thus every permitted input satisfies the output tolerance. The proposed \(\delta\) works for this particular \(\varepsilon\). The calculation checks the implication directly: the input-distance condition is the same as the required output-distance condition.
Check: The final bound is strict, as required: \(|f(x)-L|<\varepsilon\), not merely \(|f(x)-L|\le\varepsilon\).
Worked Example: A Reciprocal Function Near 2
Consider \(f(x)=\frac{1}{x}\) near \(a=2\), with proposed limit \(L=\frac{1}{2}\). Check whether \(\varepsilon=\frac{1}{10}\) and \(\delta=\frac{1}{5}\) work.
Assume \(x\) is in the domain and \(0<|x-2|<\frac{1}{5}\). The input condition gives
In particular, \(x>\frac{9}{5}>0\). Now compute the output error and use the bounds on \(x\):
Therefore the output error is less than \(\varepsilon\) whenever the input error is less than the proposed \(\delta\). The denominator is positive because the input restriction guarantees \(x>\frac{9}{5}\).
Check: The exact bound \(\frac{1}{18}\) is less than \(\frac{1}{10}\), since \(10<18\). So the chosen input restriction is sufficient for the requested tolerance.
Worked Example: A Changed Function Value at the Target
Define \(f:\mathbb{R}\to\mathbb{R}\) by \(f(x)=4\) when \(x\ne2\), and \(f(2)=99\). For the proposed limit \(L=4\) at \(a=2\), check \(\varepsilon=\frac{1}{4}\) using \(\delta=1\).
Suppose \(x\) satisfies \(0<|x-2|<1\). The first inequality gives \(x\ne2\), so the definition of the function gives \(f(x)=4\). Therefore
Thus \(\delta=1\) works for this tolerance. The value \(f(2)=99\) does not enter the check, because the definition tests inputs with \(x\ne2\).
Check: Every input covered by the implication has output exactly \(4\), so its distance from \(L=4\) is zero, which is below the positive tolerance.
The Target Value and the Function Value
The limit \(L\) describes nearby function values; it need not equal \(f(a)\). The condition \(0<|x-a|\) deliberately excludes the input \(x=a\), so the value assigned at that single input cannot affect whether the definition is satisfied. The third example illustrates this distinction.
Also, a limit statement concerns inputs in the domain \(D\). If a proposed interval around \(a\) includes numbers outside \(D\), those numbers are not tested. The implication applies to every \(x\in D\) meeting the input-distance condition.
Common Mistake: Including the Target Input
A tempting but incorrect version of the definition replaces \(0<|x-a|<\delta\) with \(|x-a|<\delta\), thereby including \(x=a\). For the function in the third example, that would require \(f(2)=99\) to be within \(\frac14\) of \(L=4\). But the calculation is
This would incorrectly reject the proposed limit. The correction is to use the definition as stated: the input must satisfy \(0<|x-a|<\delta\). This excludes \(a\) while still testing all nearby domain inputs distinct from \(a\).
Why Epsilon-Delta Language Matters
Graphs and tables can suggest what a limit might be, but they only display selected inputs or a visual scale. Epsilon-delta language states precisely what a proposed limit promises: any positive output tolerance can be met by choosing a sufficiently small input distance. It makes clear both what must be controlled and which inputs the guarantee covers.
When reading a limit claim, identify the target input \(a\), the proposed output \(L\), the output tolerance \(\varepsilon\), and the input restriction \(\delta\). Then ask whether the restriction actually guarantees the required output bound for every relevant input. This is the basic structure behind the formal definition.
Check Your Understanding
Use the definition to answer the following questions.
- In the statement \(\lim_{x\to a}f(x)=L\), what distance does \(\varepsilon\) measure, and what distance does \(\delta\) measure?
- Why does the condition on the input include \(0<|x-a|\), rather than only \(|x-a|<\delta\)?
- For \(f(x)=|x|\) near \(0\), does \(\delta=\frac{1}{5}\) guarantee \(|f(x)-0|<\frac{1}{4}\)? Explain using the inequalities.
- Can changing \(f(a)\) alone alter whether the epsilon-delta definition holds? Explain.
- In the reciprocal example, why was it necessary to establish that \(x>0\) before bounding the output error?
