Why the Order of the Quantifiers Matters
A limit statement is a guarantee about how close function values become to a proposed limiting value when inputs are sufficiently close to a point. The epsilon-delta definition expresses that guarantee using two positive numbers: an output tolerance \(\varepsilon\) and an input radius \(\delta\).
The key is not merely that both numbers appear. Their order determines what the guarantee requires. The tolerance is chosen first; the radius may then be selected in response. Reversing that order asks for a radius that works before the tolerance is known, which is generally a much stronger demand.
The Quantified Limit Statement
Let \(f:D\to\mathbb{R}\), let \(a\) be a point approached by inputs from \(D\), and let \(L\in\mathbb{R}\). The statement \(\lim_{x\to a}f(x)=L\) means that for every positive output tolerance, there is a positive input radius such that every domain input within that radius—but not equal to \(a\)—has function value within the tolerance of \(L\).
The condition that \(a\) is approached by points of \(D\) rules out a vacuous situation in which there are no domain inputs near \(a\) other than \(a\) itself. The condition \(0<|x-a|\) excludes the point \(a\); a limit concerns nearby inputs, not necessarily the value \(f(a)\).
Read the quantifiers from left to right. First, someone specifies \(\varepsilon>0\), the required accuracy in the output. Next, a \(\delta>0\) is selected. Then the implication must hold for every \(x\in D\). In particular, \(\delta\) cannot be chosen separately for each \(x\): a single radius must control all eligible inputs.
An arbitrary \(\varepsilon>0\) specifies how close \(f(x)\) must be to \(L\).
Choose some \(\delta>0\), which is allowed to depend on the tolerance.
For every \(x\in D\) with \(0<|x-a|<\delta\), the output must satisfy \(|f(x)-L|<\varepsilon\).
What May Depend on What?
The symbols \(\varepsilon\) and \(\delta\) do different jobs. The tolerance \(\varepsilon\) is prescribed from outside the guarantee. The radius \(\delta\) is the response to that tolerance. It is therefore natural to think of the choice as \(\delta=\delta(\varepsilon)\), although the definition does not require one particular formula or even a unique choice.
The response may also depend on the function \(f\), the point \(a\), and the proposed limit \(L\), all of which are fixed in the statement. Once \(\varepsilon\) and \(\delta\) have been selected, however, the condition must hold for every eligible \(x\). A proposed choice that works only for one input point is not enough.
Worked Example: Reading the Choices in Order
Consider the general quantified statement that, for each positive tolerance, a positive radius can be chosen to make the required estimate hold for every nearby input. Suppose a particular tolerance is \(\varepsilon=0.2\), and a proposed response is \(\delta=0.05\).
The tolerance is chosen first: \(0.2>0\). The proposed radius is then checked: \(0.05>0\). The remaining obligation is not to check one selected input, but to verify the estimate for every \(x\in D\) satisfying \(0<|x-a|<0.05\). If even one such \(x\) violates \(|f(x)-L|<0.2\), this particular radius fails.
That failure would show only that \(\delta=0.05\) is not a successful response to \(\varepsilon=0.2\). It would not by itself disprove the limit statement: another positive radius might work. Conversely, checking a few nearby inputs does not establish that the radius works for all of them.
Check: The radius is chosen after the tolerance, and its estimate must hold for every eligible input. That is exactly the order of the quantifiers.
A Simple Model of the Quantifier Order
Before applying the order to a full limit statement, it helps to examine a simpler claim about two positive numbers. Consider \(\forall\varepsilon>0\;\exists\delta>0\) such that \(\delta<\varepsilon\). This statement is true: whatever positive tolerance is specified, one can choose a smaller positive number.
Worked Example: A Radius That Responds to a Tolerance
We want to verify \(\forall\varepsilon>0\;\exists\delta>0\) such that \(\delta<\varepsilon\).
Take an arbitrary \(\varepsilon>0\). Choose \(\delta=\varepsilon/2\). Since \(\varepsilon>0\), division by \(2\) gives \(\delta>0\). Also, \(1/2<1\), so \(\varepsilon/2<\varepsilon\). Thus this choice is positive and satisfies the required inequality.
The choice works for every positive \(\varepsilon\), because the construction begins with an arbitrary \(\varepsilon\). It is allowed to change when \(\varepsilon\) changes.
Check: Substituting the choice gives \(0<\delta=\varepsilon/2<\varepsilon\), as required.
Now reverse those two quantifiers: \(\exists\delta>0\;\forall\varepsilon>0\) such that \(\delta<\varepsilon\). This is false. A fixed positive \(\delta\) cannot be smaller than every positive \(\varepsilon\), because one may choose \(\varepsilon=\delta/2\), which is positive but smaller than \(\delta\).
Worked Example: Why Reversing the Quantifiers Fails
Suppose, for contradiction, that there were a fixed \(\delta_*>0\) such that \(\delta_*<\varepsilon\) for every \(\varepsilon>0\).
Choose \(\varepsilon=\delta_*/2\). This is an allowed positive value because \(\delta_*>0\). The claimed inequality would then require \(\delta_*<\delta_*/2\). But \(\delta_*/2<\delta_*\), so the required inequality is false.
Therefore no single positive \(\delta_*\) works for every positive \(\varepsilon\). The original order succeeds because the choice can respond to \(\varepsilon\); the reversed order fails because \(\delta\) must be fixed in advance.
Check: The contradiction uses a permitted tolerance, \(\varepsilon=\delta_*/2\), and shows that it defeats any proposed fixed positive radius.
What Reversing the Limit Quantifiers Would Claim
In the limit definition, changing the order of the first two quantifiers produces the form \(\exists\delta>0\;\forall\varepsilon>0\;\forall x\in D\), followed by the same implication. This says that one fixed radius works for every positive output tolerance and for every input in that punctured radius.
That requirement is far stronger than the usual limit condition. In fact, if a fixed \(x\) in the punctured radius had \(|f(x)-L|=c>0\), choosing \(\varepsilon=c/2\) would demand \(c<c/2\), which is impossible. Thus every such \(x\) would have to satisfy \(f(x)=L\) exactly. The usual definition asks for values to be as close as desired by making the input sufficiently close; the reversed version requires exact equality throughout one fixed punctured neighborhood.
Worked Example: Interpreting a Fixed-Radius Claim
Suppose someone proposes the reversed claim for a function \(f\), point \(a\), and value \(L\), with a fixed radius \(\delta_*>0\). Consider any \(x\in D\) satisfying \(0<|x-a|<\delta_*\).
If \(f(x)\ne L\), then \(c=|f(x)-L|>0\). Select \(\varepsilon=c/2>0\). Since \(x\) is inside the proposed fixed radius, the claim would require \(|f(x)-L|<\varepsilon\), or \(c<c/2\). This contradicts \(c>0\).
Consequently, the reversed claim can hold only if every domain input in that punctured radius has \(f(x)=L\). Conversely, if every such input has \(f(x)=L\), then its error is \(0<\varepsilon\) for every \(\varepsilon>0\), so the fixed-radius claim does hold.
Check: The reversed claim is equivalent to exact equality \(f(x)=L\) for all domain inputs in some punctured radius, not merely closeness to \(L\).
The Exact Negation
Understanding the order also makes it possible to negate a limit statement correctly. The negation must undo each quantifier in order: “for every” becomes “there exists,” and “there exists” becomes “for every.” Negating the implication also requires both its premise and the failure of its conclusion.
The negation says there is one positive tolerance \(\varepsilon_0\) that cannot be met, no matter how small a positive radius is proposed. For every proposed radius, there is a domain input inside it whose output error is at least \(\varepsilon_0\). The tolerance is fixed first and works against every radius; the violating input may change when the radius changes.
There is a particular \(\varepsilon_0>0\) for which the guarantee fails.
Every positive radius \(\delta\) fails for that same \(\varepsilon_0\).
For each radius, some eligible \(x\) lies inside it and has error at least \(\varepsilon_0\).
Worked Example: Translating a Failed Guarantee
Suppose a statement claims that a limit equals \(L\), and consider the logical form of its failure. The failure does not say that every tolerance is impossible to meet. It says that there is at least one particular tolerance, call it \(\varepsilon_0>0\), that cannot be met by any radius.
For a proposed radius \(\delta=0.1\), the failure condition requires some \(x\in D\) with \(0<|x-a|<0.1\) and \(|f(x)-L|\ge\varepsilon_0\). If the radius is changed to \(\delta=0.01\), it requires an input (possibly a different one) with \(0<|x-a|<0.01\) and the same error threshold \(\varepsilon_0\).
The value \(\varepsilon_0\) remains fixed across all these tests. The input \(x\) is allowed to vary with \(\delta\), because the existential quantifier for \(x\) comes after the universal quantifier for \(\delta\).
Check: A single tolerance defeats every radius, while a possibly different nearby input witnesses failure for each radius.
A Common Quantifier Mistake
A frequent error is to choose a radius that works for one tolerance and then silently use it for all tolerances. For example, in the simple inequality \(\delta<\varepsilon\), choosing \(\delta=0.005\) works when \(\varepsilon=0.01\), since \(0.005<0.01\). It does not work when \(\varepsilon=0.001\), since \(0.005\not<0.001\).
The correction is not to look for one positive \(\delta\) that handles every tolerance. Instead, start with an arbitrary positive \(\varepsilon\), then give a positive \(\delta\) in response and verify the required condition. A successful radius for one tolerance establishes only that one instance of the quantified demand has been met.
Why This Order Matters
The order turns an intuitive statement—nearby inputs produce outputs close to \(L\)—into a precise guarantee. A person may demand any positive output accuracy, however small. The limit claim promises an input radius adequate for that demand. Because the demanded accuracy can become more stringent, the suitable radius may have to become smaller too.
The same order distinguishes proof from a collection of individual checks. A calculation for one tolerance and one input is not the full claim. The quantifiers require a response for every tolerance and uniform control over all inputs within the corresponding radius. When the claim is false, the negated order identifies exactly what must happen: one tolerance remains unmet at every radius.
Check Your Understanding
Answer each question by attending carefully to which choice is made first.
- In the limit definition, which quantity is chosen first: \(\varepsilon\) or \(\delta\)?
- May the value of \(\delta\) depend on \(\varepsilon\)? May it depend on the input \(x\)? Explain the difference.
- What does the reversed order \(\exists\delta>0\;\forall\varepsilon>0\) require that the usual limit definition does not?
- Write the logical form of the failure of a limit statement using a fixed positive tolerance and an arbitrary positive radius.
- Why does finding one radius that works for one tolerance not prove the full limit statement?
