Why the Difference of Squares Needs a Bound
To prove that \(\lim_{x\to a}x^2=a^2\) from the definition, we must control the size of \(|x^2-a^2|\) by controlling \(|x-a|\). The first step is algebraic: \(x^2-a^2=(x-a)(x+a)\). This identity splits the quantity we care about into the factor \(|x-a|\), which we can make as small as we like, and the factor \(|x+a|\), which is not small. The second factor depends on \(x\), so before it can be treated as a constant it must be bounded on the set of \(x\) values we are allowed to use. This tutorial derives that bound. The following tutorial uses it to choose \(\delta\) from \(\varepsilon\).
A bound is only meaningful on a restricted set. The factor \(|x+a|\) is unbounded as \(x\) ranges over all real numbers, since it grows without limit when \(|x|\) is large. But in a limit statement, only the values of \(x\) close to \(a\) matter, so we are free to restrict attention to a neighborhood of \(a\).
Proof. Write \(x+a=(x-a)+2a\). By the triangle inequality,
The last step uses the hypothesis \(|x-a|<1\). \(\square\)
Proof. By factoring and the lemma, \(|x^2-a^2|=|x-a|\,|x+a|<|x-a|\,(2|a|+1)\). The inequality is strict as long as \(x\ne a\), because then \(|x-a|>0\); if \(x=a\), both sides are zero and the bound holds with equality. \(\square\)
The constant \(2|a|+1\) depends only on the point \(a\), not on \(x\). That is what makes the bound useful: the quantity \(|x^2-a^2|\) is now below a fixed multiple of \(|x-a|\).
Worked Example: The Bound at a = 3
Find a bound of the form \(|x^2-9|<K|x-3|\) valid whenever \(|x-3|<1\).
Here \(a=3\), so \(K=2|3|+1=7\). To see it directly, \(|x-3|<1\) means \(2<x<4\), so \(5<x+3<7\) and \(|x+3|<7\). Hence \(|x^2-9|=|x-3||x+3|<7|x-3|\).
Check on sample values:
| x | |x² − 9| | 7|x − 3| | Bound holds? |
|---|---|---|---|
| 2.5 | 2.75 | 3.5 | yes |
| 3.5 | 3.25 | 3.5 | yes |
| 3.9 | 6.21 | 6.3 | yes |
| 2.1 | 4.59 | 6.3 | yes |
Worked Example: The Bound at a = −2
Bound \(|x^2-4|\) in terms of \(|x+2|\) when \(|x+2|<1\).
The point is \(a=-2\), since \(x^2-a^2=x^2-4\) and \(x-a=x+2\). The constant is \(2|{-2}|+1=5\). Directly: \(|x+2|<1\) means \(-3<x<-1\), so \(x-2\) lies in \((-5,-3)\) and \(|x-2|<5\). Thus \(x^2-4=(x+2)(x-2)\) satisfies \(|x^2-4|<5|x+2|\).
Check at \(x=-2.4\): \(|x^2-4|=|5.76-4|=1.76\), and \(5|x+2|=5(0.4)=2.0\). Since \(1.76<2.0\), the bound holds.
Worked Example: The Special Case a = 0
For \(a=0\), the theorem gives \(|x^2|<(0+1)|x|\) when \(|x|<1\), that is, \(x^2<|x|\).
This is the familiar fact that squaring a number of absolute value less than \(1\) makes it smaller. Directly, \(x^2=|x||x|<|x|\cdot1\).
Using a Different Radius
The number \(1\) in the lemma is a convenient choice, not a requirement. If instead \(|x-a|<r\) for some \(r>0\), the same proof gives \(|x+a|<2|a|+r\), and so \(|x^2-a^2|<(2|a|+r)|x-a|\). A smaller radius gives a smaller constant but forces a smaller allowed neighborhood. For \(a=3\) and \(r=\tfrac12\), the constant is \(6.5\) instead of \(7\). The choice of \(r=1\) is popular because it keeps the arithmetic simple.
Common Mistakes
Using \(2|a|\) without the extra \(1\). It is tempting to say \(|x+a|\le2|a|\) because \(x\) is \"close to \(a\)\". But take \(a=3\) and \(x=3.9\): then \(|x+a|=6.9>6=2|a|\). The inequality is false. The extra \(1\) in \(2|a|+1\) accounts for the fact that \(x\) may be a little farther from zero than \(a\) is.
Bounding without restricting \(x\). The statement \"\(|x+3|<7\)\" is true only for \(x\in(-10,4)\). If \(x=100\), then \(|x+3|=103\). Any proof that uses the bound must say that \(|x-a|<1\) is being assumed, and must make sure that the final choice of \(\delta\) respects that assumption.
Dividing by the wrong factor. The bound controls \(|x^2-a^2|\) by a multiple of \(|x-a|\), not the reverse. It does not say that \(|x-a|\) is small whenever \(|x^2-a^2|\) is small.
A Preview of the Epsilon-Delta Argument
The bound is the engine of the proof that \(\lim_{x\to a}x^2=a^2\). Suppose we are given \(\varepsilon>0\). If \(|x-a|<1\), the theorem gives \(|x^2-a^2|<(2|a|+1)|x-a|\). So if we also demand that \(|x-a|<\dfrac{\varepsilon}{2|a|+1}\), then \(|x^2-a^2|<\varepsilon\). Both demands, \(|x-a|<1\) and \(|x-a|<\varepsilon/(2|a|+1)\), are satisfied by taking \(\delta=\min\bigl(1,\varepsilon/(2|a|+1)\bigr)\). The next tutorial carries out this selection carefully; here the goal is only to see how the local bound feeds it.
Worked Example: Turning the Bound into a Number
For \(a=3\) and \(\varepsilon=0.35\), find a \(\delta\) that works, and test it.
The constant is \(K=7\). The second demand is \(|x-3|<0.35/7=0.05\). Since \(0.05<1\), take \(\delta=\min(1,0.05)=0.05\).
Test at the edge: \(x=3.049\) satisfies \(|x-3|=0.049<0.05\), and \(|x^2-9|=|9.296401-9|=0.296401<0.35\). On the other side, \(x=2.951\) gives \(|8.708401-9|=0.291599<0.35\). Both values stay inside the tolerance.
How the Constant Changes with a
The constant \(2|a|+1\) grows with \(|a|\), which reflects the geometry of the parabola: it is steeper far from the origin, so a given change in \(x\) produces a larger change in \(x^2\). The table lists the constants for several choices of \(a\).
| a | Constant 2|a| + 1 | Allowed range |x − a| < 1 | |x + a| really lies in |
|---|---|---|---|
| 0 | 1 | (−1, 1) | (−1, 1) |
| 1 | 3 | (0, 2) | (1, 3) |
| −2 | 5 | (−3, −1) | (−5, −3) |
| 3 | 7 | (2, 4) | (5, 7) |
| 10 | 21 | (9, 11) | (19, 21) |
In each row the true range of \(x+a\) is contained in \((-K,K)\), so \(K\) is a valid bound. For \(a=3\), the exact supremum of \(|x+3|\) is \(7\), so \(7\) cannot be improved. For \(a=0\) and \(a=1\) the value \(K\) matches the exact supremum as well. For \(a=-2\) the exact supremum is \(5\). In every case the bound \(2|a|+1\) is the best available when the radius is \(1\).
Why a Neighborhood Is Enough
A limit is a statement about what happens as \(x\) approaches \(a\). Values of \(x\) far from \(a\) are irrelevant to it. This is why we can restrict attention to \(|x-a|<1\): any proof only needs to work for sufficiently small \(\delta\), and shrinking \(\delta\) never hurts. The technique of first fixing an easy neighborhood, bounding all the awkward factors on it, and finally shrinking \(\delta\) to also satisfy the remaining condition appears again and again, for example when proving limits of \(1/x\), of \(\sqrt{x}\), and of products.
Worked Example: A Non-Integer Point
Find the constant in the bound at \(a=-\tfrac12\) and verify it at \(x=-0.9\) and \(x=-0.1\).
Here \(2|a|+1=2\) so \(K=2\). The assumption \(|x+\tfrac12|<1\) gives \(-1.5<x<0.5\), so \(x-\tfrac12\) lies in \((-2,0)\) and \(|x-\tfrac12|<2\), which confirms the constant.
At \(x=-0.9\): \(|x^2-\tfrac14|=|0.81-0.25|=0.56\), and \(2|x+\tfrac12|=2(0.4)=0.8\). At \(x=-0.1\): \(|0.01-0.25|=0.24\), and \(2(0.4)=0.8\). Both are below the bound.
Reading the Estimate Geometrically
The graph of \(y=x^2\) near the point \((a,a^2)\) lies between two lines through that point whose slopes are \(\pm(2|a|+1)\), as long as \(x\) stays within \(1\) of \(a\). That is a precise way to say that the parabola cannot rise or fall faster than a fixed rate near \(a\). The estimate in this tutorial is a first example of a Lipschitz bound: a function \(f\) satisfies one on an interval if \(|f(x)-f(y)|\le K|x-y|\) for a fixed constant \(K\). Here \(K=2|a|+1\) works on the interval \((a-1,a+1)\) when one of the two points is \(a\).
The two-sided structure of the final estimate is worth noticing. The bound does not say that \(x^2-a^2\) has a particular sign; it says only that its size is limited. When \(x>a\), the difference is positive, and when \(x<a\), it is negative. The absolute value handles both cases at once, which is why proofs about limits are always written with \(|\cdot|\).
Finally, remember that this estimate is only one of several that appear in the course. Similar bounds are derived for \(1/x\) near a nonzero point, for \(\sqrt{x}\), and for products of functions. Each has the same shape: factor the difference, bound the troublesome factor on a neighborhood, and keep the small factor \(|x-a|\).
Where This Fits in the Course
This tutorial belongs to the module “Limit Concepts and Estimates”. The tutorial just before it is “Proving a Quadratic Limit at an Arbitrary Input,” and the one that follows is “Choosing Delta From Epsilon.”
Practice Problems with Answers
Try each problem before reading the answer. Each answer below was verified by differentiation or by evaluating a difference quotient or integral numerically.
- Show that \(|x-5|<1\) implies \(|x^2-25|<11|x-5|\).
Answer. The constant is \(2(5)+1=11\), and \(|x+5|<11\) because \(4<x<6\) gives \(9<x+5<11\). - Find \(K\) so that \(|x^2-16|<K|x+4|\) whenever \(|x+4|<1\).
Answer. \(a=-4\), so \(K=2(4)+1=9\). - For \(a=2\) and \(|x-2|<\tfrac12\), find the best constant from the radius-\(r\) version.
Answer. \(|x+2|<2(2)+\tfrac12=4.5\), so \(|x^2-4|<4.5|x-2|\). - Use the bound at \(a=3\) to find a \(\delta\) that guarantees \(|x^2-9|<0.07\).
Answer. Take \(\delta=\min(1,0.07/7)=0.01\). Then \(|x-3|<0.01\) gives \(|x^2-9|<7(0.01)=0.07\).
Check Your Understanding
Work these without looking back.
- State and prove the lemma bounding \(|x+a|\) when \(|x-a|<1\).
- Find a constant \(K\) with \(|x^2-100|<K|x-10|\) for \(|x-10|<1\).
- Give a value of \(x\) near \(a=3\) showing that \(|x+a|\le2|a|\) is false in general.
- Why can the bound not hold for all real \(x\)? Give a specific counterexample.
- Where does the strict inequality in the theorem come from, and when does equality occur?