The Algebra of Limits

Suppose you know that f(x)→3f(x) \to 3 and g(x)→5g(x) \to 5 as x→ax \to a. Can you conclude, without any further estimation, that f(x)+g(x)→8f(x) + g(x) \to 8? Can you also conclude that f(x)g(x)→15f(x)g(x) \to 15, and that f(x)/g(x)→3/5f(x)/g(x) \to 3/5? The answer is yes. This article proves exactly why. These facts feel obvious once stated. But "obvious" is precisely the kind of claim analysis insists on justifying from the definitions. This is because limits are secretly universally-quantified statements about all tolerances ε>0\varepsilon > 0 at once. The importance of this result is structural rather than dramatic. Almost every limit computation a calculus student performs is really an appeal to these laws. Examples include finding lim⁡x→2(x3−5x+1)\lim_{x\to 2}(x^3 - 5x + 1), or lim⁡x→0sin⁡xx⋅x2\lim_{x \to 0} \frac{\sin x}{x} \cdot x^2, or the derivative quotient itself. These are really appeals to the algebra-of-limits laws, applied repeatedly to a handful of primitive limits (constants, the identity function). Without this theorem, every polynomial limit would require a fresh ε\varepsilon-δ\delta argument from scratch. With it, such computations become mechanical. The big picture is that limits are compatible with the field operations of R\mathbb{R}. These operations are addition, multiplication, scalar multiplication, and division (away from zero denominators). This compatibility is not automatic. It must be proved using the triangle inequality and careful bookkeeping of how errors combine. Once proved, though, it becomes an algebraic toolkit. This toolkit hides all ε\varepsilon-δ\delta machinery from routine computation. This article states the precise definition of a functional limit. It proves the four algebra-of-limits laws (sum, constant-multiple, product, quotient) in full rigorous detail. It illustrates the machinery on two genuinely asymmetric examples. It also shows via counterexample exactly which hypothesis cannot be dropped. It closes by connecting these laws to Continuity at a Point and The Algebra of Continuous Functions, where the same machinery reappears almost verbatim.

Key takeaways
  • The algebra of limits says limits respect +,−,×,÷+, -, \times, \div (with nonzero denominator limit), and scalar multiplication.
  • All four laws are proved directly from the ε\varepsilon-δ\delta definition of a functional limit, using the triangle inequality.
  • The product and quotient laws require an auxiliary boundedness step: convergence near aa forces local boundedness of ff.
  • The quotient law additionally needs the limit of the denominator to be nonzero, so it stays bounded away from 00 nearby.
  • The Sequential Characterization of Limits gives a one-line alternate proof via the analogous sequence laws.
  • Key idea: every algebra-of-limits law is proved by writing the target quantity as a sum of small pieces, each controlled by shrinking δ\delta.

Where this lives today

The algebra of limits is the computational engine behind differentiation itself: the derivative f′(a)=lim⁡h→0f(a+h)−f(a)hf'(a) = \lim_{h\to 0} \frac{f(a+h)-f(a)}{h} is manipulated using exactly these sum/product/quotient laws to derive the product rule, quotient rule, and polynomial differentiation formulas. In numerical analysis, error propagation estimates for compound formulas (sums, products, ratios of measured quantities) mirror the ε\varepsilon-splitting technique used here. In physics, limiting arguments for instantaneous velocity, rates of reaction, and continuum limits of discrete models all implicitly rely on the sum and product laws to combine multiple simultaneously-vanishing quantities. In probability and statistics, the algebra of limits underlies manipulations of moment generating functions and characteristic functions as parameters tend to limiting values. In computer science and numerical software, these laws justify why floating-point approximations of compound arithmetic expressions converge to the exact value as precision increases, provided no division by a vanishing quantity occurs — precisely the caveat isolated in the Quotient Law.

Ask the tutor: Why does the product law's proof break down if we only assume ff and gg are bounded near aa, without assuming either has a limit there?

DefinitionLimit of a function at a point

Let f:D→Rf : D \to \mathbb{R} where D⊆RD \subseteq \mathbb{R}, and let aa be a limit point of DD (so every neighborhood of aa contains points of DD other than aa). We say

lim⁡x→af(x)=L\lim_{x \to a} f(x) = L

if for every ε>0\varepsilon > 0 there exists δ>0\delta > 0 such that, for all x∈Dx \in D,

0<∣x−a∣<δ  ⟹  ∣f(x)−L∣<ε.0 < |x - a| < \delta \implies |f(x) - L| < \varepsilon.

By Uniqueness of Limits and Basic Examples, such an LL, if it exists, is unique, which justifies calling it the limit and writing L=lim⁡x→af(x)L = \lim_{x\to a} f(x).

Intuition

Think of ε\varepsilon as a tolerance imposed on the output and δ\delta as the resulting tolerance you are permitted on the input. To say f(x)→Lf(x) \to L is to say: however tightly you squeeze the output window around LL, you can find an input window around aa (punctured at aa itself) that gets mapped inside it. The algebra of limits is about combining two such guarantees. If ff's outputs are trapped near LL once xx is δ1\delta_1-close to aa, and gg's outputs are trapped near MM once xx is δ2\delta_2-close to aa, then taking δ=min⁡(δ1,δ2)\delta = \min(\delta_1, \delta_2) traps both simultaneously. Addition and scalar multiplication are then easy: small plus small is small, and a constant times small is small. Multiplication is subtler, because f(x)g(x)−LMf(x)g(x) - LM does not decompose so simply — it requires knowing that f(x)f(x) stays bounded nearby, not just that it converges, so that an error in gg doesn't get amplified without control. Division needs, on top of that, a guarantee that the denominator stays bounded away from zero.

Figure
Two error tolerances shrink to a common δ; both f and g land inside their ε–windows simultaneously
x a a−δ a+δ f L L−ε L+ε g M M−ε M+ε Choosing δ = min(δ₁, δ₂) traps f and g together
TheoremAlgebra of Limits

Let f,g:D→Rf, g : D \to \mathbb{R} with D⊆RD \subseteq \mathbb{R}, let aa be a limit point of DD, and suppose

lim⁡x→af(x)=L,lim⁡x→ag(x)=M,\lim_{x \to a} f(x) = L, \qquad \lim_{x \to a} g(x) = M,

with L,M∈RL, M \in \mathbb{R}. Let c∈Rc \in \mathbb{R} be a constant. Then: 1. (Sum Law) lim⁡x→a(f(x)+g(x))=L+M\displaystyle \lim_{x \to a} \big(f(x) + g(x)\big) = L + M. 2. (Constant-Multiple Law) lim⁡x→a(c⋅f(x))=cL\displaystyle \lim_{x \to a} \big(c \cdot f(x)\big) = cL. 3. (Product Law) lim⁡x→a(f(x)g(x))=LM\displaystyle \lim_{x \to a} \big(f(x) g(x)\big) = LM. 4. (Quotient Law) If in addition M≠0M \neq 0, then there is a punctured neighborhood of aa on which g(x)≠0g(x) \neq 0, and

lim⁡x→af(x)g(x)=LM.\lim_{x \to a} \frac{f(x)}{g(x)} = \frac{L}{M}.
LemmaLocal boundedness of a convergent function

If lim⁡x→af(x)=L\lim_{x\to a} f(x) = L, then there exist δ0>0\delta_0 > 0 and K>0K > 0 such that

x∈D, 0<∣x−a∣<δ0  ⟹  ∣f(x)∣≤K.x \in D,\ 0 < |x-a| < \delta_0 \implies |f(x)| \le K.
Proof

(Boundedness lemma.) Apply the definition of the limit with ε=1\varepsilon = 1. There exists δ0>0\delta_0 > 0 such that for all x∈Dx \in D with 0<∣x−a∣<δ00 < |x - a| < \delta_0,

∣f(x)−L∣<1.|f(x) - L| < 1.

By the triangle inequality,

∣f(x)∣=∣f(x)−L+L∣≤∣f(x)−L∣+∣L∣<1+∣L∣.|f(x)| = |f(x) - L + L| \le |f(x) - L| + |L| < 1 + |L|.

Set K=1+∣L∣K = 1 + |L|. This proves the lemma. ■\blacksquare

(1. Sum Law.) Technique: direct ε\varepsilon-δ\delta verification, splitting the error via the triangle inequality. Let ε>0\varepsilon > 0. Since lim⁡x→af(x)=L\lim_{x\to a} f(x) = L, applying the definition with tolerance ε/2>0\varepsilon/2 > 0 gives δ1>0\delta_1 > 0 such that

x∈D, 0<∣x−a∣<δ1  ⟹  ∣f(x)−L∣<ε/2.x \in D, \ 0 < |x-a| < \delta_1 \implies |f(x) - L| < \varepsilon/2.

Since lim⁡x→ag(x)=M\lim_{x\to a} g(x) = M, applying the definition with tolerance ε/2>0\varepsilon/2 > 0 gives δ2>0\delta_2 > 0 such that

x∈D, 0<∣x−a∣<δ2  ⟹  ∣g(x)−M∣<ε/2.x \in D, \ 0 < |x-a| < \delta_2 \implies |g(x) - M| < \varepsilon/2.

Let δ=min⁡(δ1,δ2)>0\delta = \min(\delta_1, \delta_2) > 0. If x∈Dx \in D and 0<∣x−a∣<δ0 < |x - a| < \delta, then both conditions above hold simultaneously, so by the triangle inequality,

∣(f(x)+g(x))−(L+M)∣=∣(f(x)−L)+(g(x)−M)∣≤∣f(x)−L∣+∣g(x)−M∣<ε2+ε2=ε.\big|(f(x)+g(x)) - (L+M)\big| = \big|(f(x)-L) + (g(x)-M)\big| \le |f(x)-L| + |g(x)-M| < \tfrac{\varepsilon}{2}+\tfrac{\varepsilon}{2} = \varepsilon.

Since ε>0\varepsilon > 0 was arbitrary, lim⁡x→a(f(x)+g(x))=L+M\lim_{x\to a}(f(x)+g(x)) = L+M. ■\blacksquare

(2. Constant-Multiple Law.) Technique: direct ε\varepsilon-δ\delta verification, treating c=0c=0 separately. If c=0c = 0, then cf(x)=0cf(x) = 0 for all xx, and the constant function 00 trivially has limit 0=cL0 = c L at aa (given any ε>0\varepsilon>0, any δ>0\delta>0 works, since ∣0−0∣=0<ε|0 - 0| = 0 < \varepsilon). Assume now c≠0c \neq 0. Let ε>0\varepsilon > 0. Then ε/∣c∣>0\varepsilon / |c| > 0, so applying the definition of lim⁡x→af(x)=L\lim_{x\to a} f(x) = L with tolerance ε/∣c∣\varepsilon/|c|, there is δ>0\delta > 0 such that

x∈D, 0<∣x−a∣<δ  ⟹  ∣f(x)−L∣<ε/∣c∣.x \in D, \ 0 < |x-a| < \delta \implies |f(x) - L| < \varepsilon / |c|.

For such xx,

∣cf(x)−cL∣=∣c∣ ∣f(x)−L∣<∣c∣⋅ε∣c∣=ε.|c f(x) - c L| = |c|\,|f(x) - L| < |c| \cdot \frac{\varepsilon}{|c|} = \varepsilon.

Since ε>0\varepsilon>0 was arbitrary, lim⁡x→acf(x)=cL\lim_{x\to a} c f(x) = cL. ■\blacksquare

(3. Product Law.) Technique: algebraic decomposition ("add and subtract a cross term") combined with the boundedness lemma. The key identity, valid for all x∈Dx \in D, is

f(x)g(x)−LM=f(x)(g(x)−M)+M(f(x)−L).(∗)f(x)g(x) - LM = f(x)\big(g(x) - M\big) + M\big(f(x) - L\big). \tag{$\ast$}

Indeed, expanding the right side:

f(x)g(x)−f(x)M+Mf(x)−ML=f(x)g(x)−ML,f(x)g(x) - f(x)M + Mf(x) - ML = f(x)g(x) - ML,

which equals the left side since ML=LMML = LM. This is a purely algebraic identity, true for every xx, with no limiting process invoked yet. By the boundedness lemma applied to ff, there exist δ0>0\delta_0 > 0 and K>0K > 0 such that

x∈D, 0<∣x−a∣<δ0  ⟹  ∣f(x)∣≤K.x \in D, \ 0 < |x-a| < \delta_0 \implies |f(x)| \le K.

Fix ε>0\varepsilon > 0. We will bound each term on the right of (∗)(\ast) by ε/2\varepsilon/2. For the first term: since K>0K > 0, choose (by the definition of lim⁡x→ag(x)=M\lim_{x\to a} g(x) = M, using tolerance ε2K>0\dfrac{\varepsilon}{2K}>0) a δ1>0\delta_1 > 0 such that

x∈D, 0<∣x−a∣<δ1  ⟹  ∣g(x)−M∣<ε2K.x \in D, \ 0 < |x-a| < \delta_1 \implies |g(x) - M| < \frac{\varepsilon}{2K}.

Then for 0<∣x−a∣<min⁡(δ0,δ1)0 < |x-a| < \min(\delta_0,\delta_1),

∣f(x)∣ ∣g(x)−M∣≤K⋅ε2K=ε2.|f(x)| \, |g(x)-M| \le K \cdot \frac{\varepsilon}{2K} = \frac{\varepsilon}{2}.

For the second term: if M=0M = 0 this term is identically 00 and trivially less than ε/2\varepsilon/2; if M≠0M \ne 0, choose (by the definition of lim⁡x→af(x)=L\lim_{x\to a} f(x)=L, using tolerance ε2∣M∣>0\dfrac{\varepsilon}{2|M|}>0) a δ2>0\delta_2 > 0 such that

x∈D, 0<∣x−a∣<δ2  ⟹  ∣f(x)−L∣<ε2∣M∣,x \in D, \ 0 < |x-a| < \delta_2 \implies |f(x)-L| < \frac{\varepsilon}{2|M|},

so that for such xx, ∣M∣ ∣f(x)−L∣<∣M∣⋅ε2∣M∣=ε2|M|\,|f(x)-L| < |M|\cdot \frac{\varepsilon}{2|M|} = \frac{\varepsilon}{2}. (In the case M=0M=0 we may simply set δ2=δ1\delta_2 = \delta_1, since the bound 0<ε/20 < \varepsilon/2 holds trivially.) Now let δ=min⁡(δ0,δ1,δ2)>0\delta = \min(\delta_0, \delta_1, \delta_2) > 0. For x∈Dx \in D with 0<∣x−a∣<δ0 < |x - a| < \delta, applying the triangle inequality to (∗)(\ast):

∣f(x)g(x)−LM∣≤∣f(x)∣ ∣g(x)−M∣+∣M∣ ∣f(x)−L∣<ε2+ε2=ε.|f(x)g(x) - LM| \le |f(x)|\,|g(x)-M| + |M|\,|f(x)-L| < \frac{\varepsilon}{2} + \frac{\varepsilon}{2} = \varepsilon.

Since ε>0\varepsilon > 0 was arbitrary, lim⁡x→af(x)g(x)=LM\lim_{x\to a} f(x)g(x) = LM. ■\blacksquare

(4. Quotient Law.) Technique: reduce to the special case lim⁡x→a1g(x)=1M\lim_{x\to a} \frac{1}{g(x)} = \frac{1}{M} via the product law, after first establishing that gg is bounded away from 00 near aa. Step A: gg is bounded away from 00 near aa, and this gives the "there is a punctured neighborhood on which g(x)≠0g(x)\ne 0" claim. Since M≠0M \ne 0, apply the definition of lim⁡x→ag(x)=M\lim_{x\to a} g(x) = M with tolerance ε=∣M∣/2>0\varepsilon = |M|/2 > 0: there is δ3>0\delta_3 > 0 such that

x∈D, 0<∣x−a∣<δ3  ⟹  ∣g(x)−M∣<∣M∣2.x \in D, \ 0 < |x-a| < \delta_3 \implies |g(x) - M| < \frac{|M|}{2}.

By the reverse triangle inequality, for such xx,

∣g(x)∣=∣M−(M−g(x))∣≥∣M∣−∣M−g(x)∣>∣M∣−∣M∣2=∣M∣2>0.|g(x)| = |M - (M - g(x))| \ge |M| - |M - g(x)| > |M| - \frac{|M|}{2} = \frac{|M|}{2} > 0.

In particular g(x)≠0g(x) \neq 0 for all x∈Dx \in D with 0<∣x−a∣<δ30 < |x-a| < \delta_3: this is exactly the punctured neighborhood claimed in the theorem, so f(x)/g(x)f(x)/g(x) is well-defined there. Moreover ∣g(x)∣>∣M∣/2|g(x)| > |M|/2 on this set. Step B: lim⁡x→a1g(x)=1M\lim_{x\to a} \dfrac{1}{g(x)} = \dfrac{1}{M}. Let ε>0\varepsilon > 0. For x∈Dx \in D with 0<∣x−a∣<δ30 < |x-a| < \delta_3,

∣1g(x)−1M∣=∣M−g(x)∣∣g(x)∣ ∣M∣<∣g(x)−M∣∣M∣2⋅∣M∣=2∣M∣2 ∣g(x)−M∣.\left|\frac{1}{g(x)} - \frac{1}{M}\right| = \frac{|M - g(x)|}{|g(x)|\,|M|} < \frac{|g(x)-M|}{\frac{|M|}{2}\cdot |M|} = \frac{2}{|M|^2}\,|g(x) - M|.

Since lim⁡x→ag(x)=M\lim_{x\to a} g(x) = M, apply the definition with tolerance ε∣M∣22>0\dfrac{\varepsilon |M|^2}{2} > 0 to get δ4>0\delta_4 > 0 such that

x∈D, 0<∣x−a∣<δ4  ⟹  ∣g(x)−M∣<ε∣M∣22.x \in D, \ 0 < |x-a| < \delta_4 \implies |g(x)-M| < \frac{\varepsilon |M|^2}{2}.

Let δ=min⁡(δ3,δ4)>0\delta = \min(\delta_3, \delta_4) > 0. For x∈Dx \in D with 0<∣x−a∣<δ0 < |x-a| < \delta, combining the two displayed bounds,

∣1g(x)−1M∣<2∣M∣2⋅ε∣M∣22=ε.\left|\frac{1}{g(x)} - \frac{1}{M}\right| < \frac{2}{|M|^2}\cdot \frac{\varepsilon |M|^2}{2} = \varepsilon.

Since ε>0\varepsilon>0 was arbitrary, lim⁡x→a1g(x)=1M\lim_{x\to a} \frac{1}{g(x)} = \frac{1}{M}. Step C: conclude via the Product Law. We have lim⁡x→af(x)=L\lim_{x\to a} f(x) = L and lim⁡x→a1g(x)=1M\lim_{x\to a}\frac{1}{g(x)} = \frac{1}{M}, both already established (the second in Step B, which used only Step A and the basic definition, not the Product Law itself, so there is no circularity). By part (3) of this very theorem (Product Law), applied to the functions ff and 1/g1/g on the common punctured neighborhood where both are defined,

lim⁡x→af(x)g(x)=lim⁡x→a(f(x)⋅1g(x))=L⋅1M=LM.\lim_{x\to a} \frac{f(x)}{g(x)} = \lim_{x\to a}\left(f(x)\cdot \frac{1}{g(x)}\right) = L \cdot \frac{1}{M} = \frac{L}{M}.

This completes the proof of the Quotient Law, and of the theorem. ■\blacksquare

∎
CorollaryDifference Law

Under the hypotheses of the theorem, lim⁡x→a(f(x)−g(x))=L−M\displaystyle \lim_{x\to a}\big(f(x) - g(x)\big) = L - M.

Proof

Write f(x)−g(x)=f(x)+(−1)g(x)f(x) - g(x) = f(x) + (-1)g(x). By the Constant-Multiple Law with c=−1c=-1, lim⁡x→a(−1)g(x)=−M\lim_{x\to a}(-1)g(x) = -M. By the Sum Law applied to ff and (−1)g(-1)g, lim⁡x→a(f(x)+(−1)g(x))=L+(−M)=L−M\lim_{x\to a}\big(f(x) + (-1)g(x)\big) = L + (-M) = L - M. ■\blacksquare

∎

Worked examples

Example 1 (a genuinely mixed rational computation). Let

f(x)=4x2−x+7,g(x)=x−32x+1,f(x) = 4x^2 - x + 7, \qquad g(x) = \frac{x-3}{2x+1},

and compute lim⁡x→2(f(x)⋅g(x)−5f(x))\lim_{x\to 2} \big(f(x)\cdot g(x) - 5f(x)\big). First isolate the "atomic" limits at a=2a=2. From the constant function laws and the identity limit lim⁡x→2x=2\lim_{x\to 2} x = 2 (an immediate ε\varepsilon-δ\delta check: take δ=ε\delta = \varepsilon), repeated use of the Product Law gives lim⁡x→2x2=2⋅2=4\lim_{x\to 2} x^2 = 2\cdot 2 = 4, and the Constant-Multiple and Sum Laws give

lim⁡x→2f(x)=lim⁡x→2(4x2−x+7)=4(4)−2+7=21.\lim_{x\to 2} f(x) = \lim_{x\to2}\big(4x^2 - x + 7\big) = 4(4) - 2 + 7 = 21.

For gg: the numerator x−3→2−3=−1x - 3 \to 2-3=-1 by the Sum Law (with the constant −3-3), and the denominator 2x+1→2(2)+1=52x+1 \to 2(2)+1 = 5 by the Constant-Multiple and Sum Laws. Since the denominator's limit 5≠05 \ne 0, the Quotient Law applies:

lim⁡x→2g(x)=−15.\lim_{x\to 2} g(x) = \frac{-1}{5}.

Now assemble the target expression. By the Product Law, lim⁡x→2f(x)g(x)=21⋅(−15)=−215\lim_{x\to 2} f(x)g(x) = 21 \cdot \left(-\tfrac15\right) = -\tfrac{21}{5}. By the Constant-Multiple Law, lim⁡x→25f(x)=5(21)=105\lim_{x\to 2} 5f(x) = 5(21) = 105. By the Difference Law,

lim⁡x→2(f(x)g(x)−5f(x))=−215−105=−215−5255=−5465.\lim_{x\to 2}\big(f(x)g(x) - 5f(x)\big) = -\frac{21}{5} - 105 = -\frac{21}{5} - \frac{525}{5} = -\frac{546}{5}.

Every step invoked only the theorem's laws and the elementary limits lim⁡x=a\lim x = a and lim⁡c=c\lim c = c — no fresh ε\varepsilon-δ\delta argument was needed for this degree-3 rational expression.

Example 2 (where the quotient law's nonzero-denominator hypothesis is load-bearing, resolved by cancellation). Let

h(x)=x2−x−6x−3,a=3.h(x) = \frac{x^2 - x - 6}{x - 3}, \qquad a = 3.

Here lim⁡x→3(x−3)=0\lim_{x\to 3}(x-3) = 0, so the Quotient Law's hypothesis M≠0M \ne 0 fails for the naive decomposition h=(x2−x−6)/(x−3)h = (x^2-x-6)/(x-3) — the theorem simply does not apply to f(x)=x2−x−6f(x)=x^2-x-6 and g(x)=x−3g(x) = x-3 directly. This is not a contradiction; it only means we must find another route. Since a=3a=3 is a limit point of the domain D=R∖{3}D = \mathbb{R}\setminus\{3\} and we only ever evaluate hh at points x≠3x \ne 3 (as required by the punctured condition 0<∣x−a∣0 < |x-a| in the very definition of the limit), we may factor for x≠3x \ne 3:

x2−x−6=(x−3)(x+2)  ⟹  h(x)=(x−3)(x+2)x−3=x+2for all x≠3.x^2 - x - 6 = (x-3)(x+2) \implies h(x) = \frac{(x-3)(x+2)}{x-3} = x + 2 \quad \text{for all } x \ne 3.

Two functions that agree everywhere on a punctured neighborhood of aa have identical limits at aa (immediate from the definition, since the defining implication only ever concerns xx with 0<∣x−a∣<δ0 < |x-a|<\delta). Hence

lim⁡x→3h(x)=lim⁡x→3(x+2)=3+2=5,\lim_{x\to 3} h(x) = \lim_{x\to 3} (x+2) = 3 + 2 = 5,

the last equality by the Sum Law applied to lim⁡x→3x=3\lim_{x\to3} x = 3 and the constant 22. This example shows precisely how the algebra of limits is used together with algebraic manipulation (factoring) to legally sidestep a 0/00/0 obstruction, rather than illegally applying the Quotient Law to a zero-limit denominator.

CounterexampleWhy the Quotient Law needs M≠0M \neq 0

Consider f(x)=1f(x) = 1 (constant) and g(x)=xg(x) = x, with a=0a = 0. Then

lim⁡x→0f(x)=1,lim⁡x→0g(x)=0=M.\lim_{x\to 0} f(x) = 1, \qquad \lim_{x\to 0} g(x) = 0 = M.

The hypothesis M≠0M \ne 0 of the Quotient Law fails. Examine f(x)g(x)=1x\displaystyle \frac{f(x)}{g(x)} = \frac{1}{x} for xx near 00, x≠0x\ne 0. No real number LL can serve as lim⁡x→01x\lim_{x\to 0} \frac{1}{x}. Indeed, suppose toward contradiction that lim⁡x→01x=L\lim_{x\to0}\frac1x = L for some L∈RL \in \mathbb{R}. Apply the definition with ε=1\varepsilon = 1: there is δ>0\delta>0 such that 0<∣x∣<δ  ⟹  ∣1/x−L∣<10<|x|<\delta \implies |1/x - L| < 1, forcing ∣1/x∣<∣L∣+1|1/x| < |L|+1 for all such xx. But taking x=1∣L∣+2x = \frac{1}{|L|+2} (which satisfies 0<x<δ0 < x < \delta once we also shrink it below δ\delta, e.g. replace it by x0=min⁡(δ/2,1∣L∣+2)>0x_0 = \min\left(\delta/2, \frac{1}{|L|+2}\right) > 0), we get 1/x0≥∣L∣+2>∣L∣+11/x_0 \ge |L|+2 > |L|+1, a contradiction. Hence no such LL exists, and lim⁡x→01x\lim_{x\to0} \frac1x does not exist as a real number. This shows the conclusion of the Quotient Law can fail completely — not merely give the wrong value, but fail to produce any real limit — once the hypothesis M≠0M \ne 0 is dropped. The other three laws (sum, constant-multiple, product) have no analogous nonzero requirement, which is exactly why division is singled out with an extra hypothesis in the theorem.

Scope, edge cases & common traps

SituationWhat happens
L,ML, M both finite, M≠0M \ne 0All four laws apply exactly as stated.
M=0M = 0, L≠0L \ne 0Quotient Law inapplicable; in fact f/gf/g typically has no finite limit (as in the counterexample).
M=0M = 0, L=0L = 0Quotient Law inapplicable; f/gf/g may or may not have a limit — must be analyzed directly (e.g. via factoring, as in Example 2).
f,gf, g defined only on a set DD with aa a limit pointTheorem holds verbatim; all δ\delta's are intersected with DD automatically since the definition already quantifies over x∈Dx \in D.
Infinite limits (f(x)→∞f(x)\to\infty)Not covered by this theorem as stated; requires separate definitions and separate (partially different) arithmetic rules.
Composing more than two functionsApply the theorem repeatedly (it is not "self-generalizing" to nn functions without induction, though the induction is routine).

The mathematician behind it

Origins · 1 minAugustin-Louis Cauchy and how this was discovered
AC
Augustin-Louis Cauchy
1789–1857 · Early 19th-century Paris, École Polytechnique

Cauchy was among the first to insist that calculus be rebuilt on precise inequality-based definitions of limit, rather than infinitesimals treated as vague "vanishing quantities." His 1821 Cours d'Analyse introduced the style of argument — fix an arbitrary tolerance, then produce a controlling bound — that underlies every ε\varepsilon-δ\delta proof in this article, including the algebra-of-limits laws proved here. The now-standard ε\varepsilon-δ\delta formalism was sharpened further by Karl Weierstrass a few decades later, but the strategy of bounding an error by splitting it into controllable pieces is already fully Cauchy's.

History
- 1821 — Cauchy's Cours d'Analyse gives rigorous, inequality-based treatment of limits and convergence.
- 1823 — Cauchy defines the derivative as a limit of difference quotients, relying implicitly on limit arithmetic.
- 1861 — Weierstrass, in Berlin lectures, formalizes the modern ε\varepsilon-δ\delta definition of limit used throughout this article.
- 1872 — Weierstrass's construction of a continuous, nowhere-differentiable function underscores why rigorous limit laws (not intuition) were needed.
- 1900s — The ε\varepsilon-δ\delta framework, including the algebra-of-limits theorem, becomes standard in analysis textbooks worldwide.
Why do we need the boundedness lemma for the product law but not for the sum law?

In the sum law, the error ∣(f(x)+g(x))−(L+M)∣|(f(x)+g(x))-(L+M)| splits exactly into ∣f(x)−L∣+∣g(x)−M∣|f(x)-L| + |g(x)-M|, two pieces each independently made small. In the product law, the natural decomposition f(x)g(x)−LM=f(x)(g(x)−M)+M(f(x)−L)f(x)g(x)-LM = f(x)(g(x)-M) + M(f(x)-L) has a product of an error term with f(x)f(x) itself. If f(x)f(x) were unbounded nearby, an arbitrarily small ∣g(x)−M∣|g(x)-M| could still be amplified into a large product. The boundedness lemma guarantees ∣f(x)∣≤K|f(x)|\le K near aa, so ∣f(x)∣∣g(x)−M∣≤K∣g(x)−M∣|f(x)||g(x)-M| \le K|g(x)-M|, which we can then make small by choosing g(x)−Mg(x)-M smaller than ε/(2K)\varepsilon/(2K).

Does the algebra of limits work for one-sided limits and limits at infinity?

Yes, with essentially identical proofs. One-sided limits replace the punctured condition 0<∣x−a∣<δ0<|x-a|<\delta with a<x<a+δa<x<a+\delta (right-hand) or a−δ<x<aa-\delta<x<a (left-hand), and every step of the sum/product/quotient proofs goes through verbatim since they never used symmetry around aa. Limits at infinity replace 0<∣x−a∣<δ0<|x-a|<\delta with x>Nx > N for some large NN, and the same ε\varepsilon-splitting arguments apply with NN playing the role of δ\delta. The core technique — decompose the error, bound each piece — is agnostic to which flavor of limit is used.

Can the Product Law be proved without the boundedness lemma, using a different decomposition?

Not without smuggling in some substitute for boundedness. Any decomposition of f(x)g(x)−LMf(x)g(x)-LM into pieces involving differences f(x)−Lf(x)-L and g(x)−Mg(x)-M will, by dimensional/algebraic necessity, multiply one of these differences by either f(x)f(x), g(x)g(x), LL, or MM. Since LL and MM are fixed constants, those terms are automatically bounded, but any term multiplied by the variable quantity f(x)f(x) or g(x)g(x) needs that variable quantity controlled — which is exactly what the boundedness lemma supplies for free once convergence is assumed.

Why does the Quotient Law prove g(x)≠0g(x)\neq 0 near aa instead of assuming it?

Because assuming it would be circular or at least redundant: the fact follows automatically from M≠0M \ne 0 together with convergence, via the reverse triangle inequality. Proving it as a step (rather than assuming it) is both more economical and more informative — it shows precisely how close xx must be to aa (namely, within the δ3\delta_3 from Step A) before division is even legitimate, which matters for constructing the final δ\delta in the proof.

Is the converse of the Sum Law true — if lim⁡(f+g)\lim(f+g) exists, must lim⁡f\lim f and lim⁡g\lim g individually exist?

No, and this is worth flagging even though it is not part of the theorem as stated. Take f(x)=1xf(x) = \frac{1}{x} and g(x)=−1xg(x) = -\frac{1}{x} at a=0a=0: neither lim⁡x→0f(x)\lim_{x\to0}f(x) nor lim⁡x→0g(x)\lim_{x\to0}g(x) exists (each blows up), yet f(x)+g(x)=0f(x)+g(x)=0 for all x≠0x\ne0, so lim⁡x→0(f+g)(x)=0\lim_{x\to0}(f+g)(x)=0 trivially. The Sum Law is a one-directional implication: existence of the individual limits guarantees existence and value of the limit of the sum, but not conversely.

How does this theorem foreshadow continuity?

A function ff is continuous at aa precisely when lim⁡x→af(x)=f(a)\lim_{x\to a} f(x) = f(a), folding the limit point and the function value together (see Continuity at a Point). Once continuity is defined this way, the algebra-of-limits theorem instantly implies that sums, scalar multiples, products, and quotients (with nonzero denominator) of continuous functions are continuous — this is The Algebra of Continuous Functions, and its proof is a near-immediate corollary of the theorem proved here, just substituting f(a)f(a) and g(a)g(a) for LL and MM.

Why does the proof rely on ε/2 splits and not, say, ε/3 or arbitrary weights?

Any positive weights summing to at most ε\varepsilon would work equally well — ε/2+ε/2\varepsilon/2 + \varepsilon/2, ε/3+2ε/3\varepsilon/3+2\varepsilon/3, or even 0.9ε+0.1ε0.9\varepsilon + 0.1\varepsilon all yield a bound <ε<\varepsilon (or ≤ε\le\varepsilon, adjusted to be strict). The choice of equal halves is purely a matter of convention and readability; the only mathematical requirement is that the chosen positive numbers used as intermediate tolerances sum to no more than the target ε\varepsilon, so that the final triangle-inequality bound comes out strictly below ε\varepsilon.

Before this

Where this leads