Algebraic Limit Theorems

Suppose you know two sequences converge: an→3a_n \to 3 and bn→5b_n \to 5. Does it follow, without any further checking, that an+bn→8a_n + b_n \to 8? Does it follow that anbn→15a_n b_n \to 15, or that an/bn→3/5a_n/b_n \to 3/5? The answer is yes. But the word "follow" is doing real work here. It demands a proof built directly from the Epsilon-N Definition of a Limit. It does not permit an appeal to intuition about what limits "should" do. This matters because almost every limit computation a student ever performs silently invokes one fact. That fact is that limits are compatible with the arithmetic operations. Examples include evaluating lim⁡(3n2+1)/(n2+7)\lim (3n^2+1)/(n^2+7), differentiating via difference quotients, and summing infinite series. Without a theorem justifying this, each such computation would require a fresh, from-scratch ε\varepsilon-NN argument. The Algebraic Limit Theorem is the machine that converts hard analytic estimates into an algebra of limits. These estimates are done once and for all. The resulting algebra can be used mechanically ever after. The big picture is that convergence is a local and quantitative notion. A sequence converges to LL if its terms eventually lie in any prescribed neighborhood of LL. This quantitative control is exactly strong enough to survive several operations. These are addition, scalar multiplication, multiplication, and (with one necessary exception) division. The proofs are elementary triangle-inequality arguments. But they are the prototype for essentially all "limits interact well with continuous operations" results in analysis. This includes continuity of functions built from sums and products. This article states the precise definitions. It proves the sum, scalar multiple, product, and quotient laws in full. It discusses exactly why the quotient rule needs a nonvanishing-limit hypothesis. It also shows two nontrivial worked computations. It closes by pointing toward the Order Limit Theorem and Squeeze Theorem. These extend this algebraic toolkit to inequalities.

Key takeaways
  • If an→La_n \to L and bn→Mb_n \to M, then an+bn→L+Ma_n + b_n \to L+M, can→cLca_n \to cL, and anbn→LMa_n b_n \to LM.
  • If additionally M≠0M \neq 0, then an/bn→L/Ma_n/b_n \to L/M (eventually well-defined, since bn≠0b_n \neq 0 eventually).
  • All four laws are proved from the raw ε\varepsilon-NN definition using the triangle inequality and algebraic manipulation ("add and subtract a cross term").
  • A convergent sequence is automatically bounded; this boundedness is the key extra fact needed to control the product rule's error term.
  • The quotient law genuinely requires M≠0M \neq 0 — dropping it produces sequences with no limit at all, not just a wrong answer.
  • Key idea: convergence is preserved by +, −, ×, ÷+,\ -,\ \times,\ \div (away from zero denominators) because the ε\varepsilon-N estimates for sums and products can be built, term by term, out of the estimates for the pieces.

Where this lives today

The Algebraic Limit Theorem is invisible infrastructure: every time a calculus student computes lim⁡x→cf(x)g(x)\lim_{x\to c} \frac{f(x)}{g(x)} by splitting into pieces, or a numerical analyst argues that a scheme's error terms add up correctly, this theorem (or its continuous-variable analogue) is being used. In real analysis it is the base case for proving that sums, products, and quotients of continuous functions are continuous, since continuity is defined via sequential or ε\varepsilon-δ\delta limits. In the theory of infinite series, term-by-term algebraic manipulation of convergent series (e.g., combining ∑an\sum a_n and ∑bn\sum b_n into ∑(an+bn)\sum(a_n+b_n)) rests on exactly the Sum Rule applied to partial sums. In numerical linear algebra and scientific computing, convergence proofs for iterative methods (Newton's method, gradient descent) routinely decompose error sequences into sums and products of simpler convergent sequences, invoking these laws at each step. In probability theory, the Algebraic Limit Theorem for sequences underlies the algebra of limits in probability (e.g., Slutsky's theorem, which extends these ideas to convergence in distribution/probability). In cryptography and complexity theory, asymptotic analysis of algorithms' running times uses limit arithmetic to combine big-O estimates derived from convergent ratio tests.

Ask the tutor: How does the Algebraic Limit Theorem generalize to prove that sums and products of continuous functions are continuous?

DefinitionConvergence of a sequence

Let (an)n=1∞(a_n)_{n=1}^\infty be a sequence of real numbers and let L∈RL \in \mathbb{R}. We say (an)(a_n) converges to LL, written

lim⁡n→∞an=Loran→L,\lim_{n\to\infty} a_n = L \quad\text{or}\quad a_n \to L,

if for every ε>0\varepsilon > 0 there exists N∈NN \in \mathbb{N} such that

n≥N  ⟹  ∣an−L∣<ε.n \ge N \implies |a_n - L| < \varepsilon.

A sequence is convergent if such an LL exists; by Uniqueness of Limits and Boundedness, this LL is unique when it exists, and every convergent sequence is bounded (there is B>0B>0 with ∣an∣≤B|a_n| \le B for all nn).

Intuition

Think of convergence as a promise: no matter how small a tolerance ε\varepsilon you demand, the sequence keeps its promise to stay within ε\varepsilon of LL from some point onward. The Algebraic Limit Theorem says that promises made by (an)(a_n) and (bn)(b_n) separately can be combined into a promise for (an+bn)(a_n+b_n), (anbn)(a_n b_n), and (carefully) (an/bn)(a_n/b_n). The addition law is the easiest to picture: if ana_n is eventually within ε/2\varepsilon/2 of LL and bnb_n is eventually within ε/2\varepsilon/2 of MM, then an+bna_n + b_n is eventually within ε\varepsilon of L+ML + M, by the triangle inequality. Splitting the tolerance ε\varepsilon in half between the two sequences is the whole trick. The product law is subtler, because anbn−LMa_nb_n - LM does not split as a sum of two small pieces automatically — you must insert and subtract a cross term, anbn−anM+anM−LMa_nb_n - a_nM + a_nM - LM, and then use boundedness of (an)(a_n) (guaranteed because it converges) to control the first piece, since bn−Mb_n - M alone being small is not enough — it needs to be multiplied by something that doesn't blow up.

Figure
Splitting the tolerance ε for the sum law
L=3 L−ε/2 L+ε/2 M=5 M−ε/2 M+ε/2 a_n within ε/2 of L, b_n within ε/2 of M ⇒ (a_n+b_n) within ε of L+M=8

Central result

TheoremAlgebraic Limit Theorem

Let (an)(a_n) and (bn)(b_n) be sequences of real numbers with

lim⁡n→∞an=L,lim⁡n→∞bn=M,\lim_{n\to\infty} a_n = L, \qquad \lim_{n\to\infty} b_n = M,

and let c∈Rc \in \mathbb{R} be a constant. Then: 1. (Sum Rule) lim⁡n→∞(an+bn)=L+M\displaystyle \lim_{n\to\infty}(a_n + b_n) = L + M. 2. (Scalar Rule) lim⁡n→∞(c an)=cL\displaystyle \lim_{n\to\infty}(c\, a_n) = cL. 3. (Product Rule) lim⁡n→∞(anbn)=LM\displaystyle \lim_{n\to\infty}(a_n b_n) = LM. 4. (Quotient Rule) If, in addition, M≠0M \neq 0, then there exists N0N_0 such that bn≠0b_n \neq 0 for all n≥N0n \ge N_0, and

lim⁡n→∞anbn=LM,\lim_{n\to\infty} \frac{a_n}{b_n} = \frac{L}{M},

where the sequence an/bna_n/b_n is understood to be defined for n≥N0n \ge N_0.

LemmaConvergent sequences are bounded

If an→La_n \to L, then there exists B>0B > 0 such that ∣an∣≤B|a_n| \le B for all n∈Nn \in \mathbb{N}.

Proof

This is Uniqueness of Limits and Boundedness, reproduced for completeness. Apply the definition of convergence with ε=1\varepsilon = 1: there is N1N_1 such that n≥N1  ⟹  ∣an−L∣<1n \ge N_1 \implies |a_n - L| < 1, hence ∣an∣≤∣L∣+1|a_n| \le |L| + 1 for all n≥N1n \ge N_1 by the triangle inequality. For the finitely many indices n=1,…,N1−1n = 1, \dots, N_1 - 1, set B0=max⁡{∣a1∣,…,∣aN1−1∣}B_0 = \max\{|a_1|, \dots, |a_{N_1-1}|\}. Taking

B=max⁡{B0,∣L∣+1}B = \max\{B_0, |L|+1\}

gives ∣an∣≤B|a_n| \le B for every n∈Nn \in \mathbb{N}.

∎
Proof

We prove each part in turn using only the ε\varepsilon-NN definition, the triangle inequality, and the boundedness lemma. This is the standard direct estimation technique of real analysis: express the quantity to be bounded, insert/subtract convenient cross terms, and bound each piece by choosing NN large enough.

(1) Sum Rule. Let ε>0\varepsilon > 0. Since an→La_n \to L, applied to ε/2>0\varepsilon/2 > 0, there is N1N_1 such that

n≥N1  ⟹  ∣an−L∣<ε/2.n \ge N_1 \implies |a_n - L| < \varepsilon/2.

Since bn→Mb_n \to M, applied to ε/2>0\varepsilon/2 > 0, there is N2N_2 such that

n≥N2  ⟹  ∣bn−M∣<ε/2.n \ge N_2 \implies |b_n - M| < \varepsilon/2.

Let N=max⁡{N1,N2}N = \max\{N_1, N_2\}. For n≥Nn \ge N, both inequalities hold, so by the triangle inequality,

∣(an+bn)−(L+M)∣=∣(an−L)+(bn−M)∣≤∣an−L∣+∣bn−M∣<ε2+ε2=ε.|(a_n+b_n) - (L+M)| = |(a_n - L) + (b_n - M)| \le |a_n - L| + |b_n - M| < \frac{\varepsilon}{2} + \frac{\varepsilon}{2} = \varepsilon.

Since ε>0\varepsilon > 0 was arbitrary, an+bn→L+Ma_n + b_n \to L + M.

(2) Scalar Rule. If c=0c = 0 the sequence canc a_n is identically 0=cL0 = cL and the claim is trivial. Suppose c≠0c \neq 0. Let ε>0\varepsilon > 0. Applying the definition of an→La_n \to L to ε/∣c∣>0\varepsilon/|c| > 0, there is NN such that

n≥N  ⟹  ∣an−L∣<ε∣c∣.n \ge N \implies |a_n - L| < \frac{\varepsilon}{|c|}.

Then for n≥Nn \ge N,

∣can−cL∣=∣c∣∣an−L∣<∣c∣⋅ε∣c∣=ε.|c a_n - cL| = |c||a_n - L| < |c|\cdot\frac{\varepsilon}{|c|} = \varepsilon.

Hence can→cLc a_n \to cL.

(3) Product Rule. By the Lemma, since (an)(a_n) converges it is bounded: fix B>0B > 0 with ∣an∣≤B|a_n| \le B for all nn. We may also assume B≥∣M∣B \ge |M| by enlarging BB if necessary (this simplifies the final bookkeeping but is not essential). Write the key algebraic identity, inserting and subtracting anMa_n M:

anbn−LM=anbn−anM+anM−LM=an(bn−M)+M(an−L).a_n b_n - LM = a_n b_n - a_n M + a_n M - LM = a_n(b_n - M) + M(a_n - L).

By the triangle inequality,

∣anbn−LM∣≤∣an∣∣bn−M∣+∣M∣∣an−L∣≤B∣bn−M∣+∣M∣∣an−L∣.|a_n b_n - LM| \le |a_n||b_n - M| + |M||a_n - L| \le B|b_n - M| + |M||a_n - L|.

Let ε>0\varepsilon > 0. Choose N1N_1 so that n≥N1  ⟹  ∣bn−M∣<ε2Bn \ge N_1 \implies |b_n - M| < \dfrac{\varepsilon}{2B} (possible since B>0B > 0; here we use B≥∣an∣>0B\ge |a_n|>0 eventually, but even if some sequence is eventually 00 the same B>0B>0 bound works since BB was fixed positive). Choose N2N_2 so that n≥N2  ⟹  ∣an−L∣<ε2(∣M∣+1)n \ge N_2 \implies |a_n - L| < \dfrac{\varepsilon}{2(|M|+1)} (the +1+1 avoids dividing by zero if M=0M = 0). Let N=max⁡{N1,N2}N = \max\{N_1, N_2\}. For n≥Nn \ge N,

∣anbn−LM∣≤B⋅ε2B+∣M∣⋅ε2(∣M∣+1)<ε2+ε2=ε.|a_n b_n - LM| \le B \cdot \frac{\varepsilon}{2B} + |M|\cdot \frac{\varepsilon}{2(|M|+1)} < \frac{\varepsilon}{2} + \frac{\varepsilon}{2} = \varepsilon.

Hence anbn→LMa_n b_n \to LM.

(4) Quotient Rule. First we show bn≠0b_n \neq 0 eventually. Since M≠0M \neq 0, apply the definition of bn→Mb_n \to M with ε0=∣M∣/2>0\varepsilon_0 = |M|/2 > 0: there is N0N_0 such that

n≥N0  ⟹  ∣bn−M∣<∣M∣2.n \ge N_0 \implies |b_n - M| < \frac{|M|}{2}.

By the reverse triangle inequality, for n≥N0n \ge N_0,

∣bn∣≥∣M∣−∣bn−M∣>∣M∣−∣M∣2=∣M∣2>0,|b_n| \ge |M| - |b_n - M| > |M| - \frac{|M|}{2} = \frac{|M|}{2} > 0,

so bn≠0b_n \neq 0 for all n≥N0n \ge N_0, and moreover ∣bn∣>∣M∣/2|b_n| > |M|/2 for such nn. This proves the well-definedness claim and gives us a uniform lower bound to control division. It suffices to prove 1/bn→1/M1/b_n \to 1/M (for n≥N0n \ge N_0), since then an/bn=an⋅(1/bn)→L⋅(1/M)=L/Ma_n / b_n = a_n \cdot (1/b_n) \to L \cdot (1/M) = L/M by the already-established Product Rule (applied to the sequences (an)(a_n) and (1/bn)n≥N0(1/b_n)_{n\ge N_0}, which is legitimate since discarding finitely many initial terms does not affect convergence or its limit — a direct consequence of the definition, as the same NN works after shifting). To show 1/bn→1/M1/b_n \to 1/M: for n≥N0n \ge N_0,

∣1bn−1M∣=∣M−bn∣∣bn∣∣M∣<∣bn−M∣∣M∣2⋅∣M∣=2∣bn−M∣M2.\left|\frac{1}{b_n} - \frac{1}{M}\right| = \frac{|M - b_n|}{|b_n||M|} < \frac{|b_n - M|}{\frac{|M|}{2}\cdot |M|} = \frac{2|b_n - M|}{M^2}.

Let ε>0\varepsilon > 0. Since bn→Mb_n \to M, there is N1≥N0N_1 \ge N_0 such that

n≥N1  ⟹  ∣bn−M∣<εM22.n \ge N_1 \implies |b_n - M| < \frac{\varepsilon M^2}{2}.

Then for n≥N1n \ge N_1,

∣1bn−1M∣<2M2⋅εM22=ε.\left|\frac{1}{b_n} - \frac{1}{M}\right| < \frac{2}{M^2}\cdot \frac{\varepsilon M^2}{2} = \varepsilon.

Hence 1/bn→1/M1/b_n \to 1/M, and by the Product Rule, an/bn→L/Ma_n/b_n \to L/M. ■\blacksquare

∎

Worked examples

Example 1 (a genuine rational sequence, all rules used). Let

an=7n2−2n+13n2+5n−4,n≥2 (to avoid a small denominator zero, if any).a_n = \frac{7n^2 - 2n + 1}{3n^2 + 5n - 4}, \qquad n \ge 2 \text{ (to avoid a small denominator zero, if any)}.

We compute lim⁡n→∞an\lim_{n\to\infty} a_n using the Algebraic Limit Theorem, not by asserting the answer. Divide numerator and denominator by n2n^2 (valid for n≥1n \ge 1, since n2≠0n^2 \neq 0):

an=7−2n+1n23+5n−4n2.a_n = \frac{7 - \dfrac{2}{n} + \dfrac{1}{n^2}}{3 + \dfrac{5}{n} - \dfrac{4}{n^2}}.

We take as previously established (or provable by the same ε\varepsilon-NN method) that 1/n→01/n \to 0 and hence, by the Product Rule applied to 1/n1/n with itself, 1/n2=(1/n)(1/n)→0⋅0=01/n^2 = (1/n)(1/n) \to 0 \cdot 0 = 0. Now apply the Scalar Rule to get 2/n=2⋅(1/n)→2⋅0=02/n = 2\cdot(1/n) \to 2\cdot 0 = 0, 5/n→5⋅0=05/n \to 5\cdot 0 = 0, and 4/n2→4⋅0=04/n^2 \to 4\cdot 0 = 0. Apply the Sum Rule (extended to finitely many terms by induction, itself an easy consequence of applying the two-term Sum Rule repeatedly) to the numerator:

7−2n+1n2⟶7−0+0=7,7 - \frac{2}{n} + \frac{1}{n^2} \longrightarrow 7 - 0 + 0 = 7,

and to the denominator:

3+5n−4n2⟶3+0−0=3.3 + \frac{5}{n} - \frac{4}{n^2} \longrightarrow 3 + 0 - 0 = 3.

Since the denominator's limit 3≠03 \neq 0, the Quotient Rule applies and gives

lim⁡n→∞an=73.\lim_{n\to\infty} a_n = \frac{7}{3}.

Note that we never needed to guess the answer: the algebra of limits produced it from the elementary limit 1/n→01/n \to 0.

Example 2 (a product/quotient combination with a subtlety). Let

bn=(1+1n)⋅4n−32n+1⋅1n+2sin⁡(n).b_n = \left(1 + \frac{1}{n}\right)\cdot \frac{4n - 3}{2n + 1} \cdot \frac{1}{n+2}\sin(n).

This looks intimidating, but let's break it into pieces that the theorem can handle. Write bn=cndnenb_n = c_n d_n e_n where

cn=1+1n,dn=4n−32n+1,en=sin⁡nn+2.c_n = 1 + \frac{1}{n}, \qquad d_n = \frac{4n-3}{2n+1}, \qquad e_n = \frac{\sin n}{n+2}.

For cnc_n: by the Sum Rule, cn=1+(1/n)→1+0=1c_n = 1 + (1/n) \to 1 + 0 = 1. For dnd_n: divide numerator and denominator by nn: dn=4−3/n2+1/nd_n = \dfrac{4 - 3/n}{2+1/n}. By the Sum and Scalar Rules the numerator →4−0=4\to 4 - 0 = 4 and denominator →2+0=2≠0\to 2+0=2 \neq 0, so by the Quotient Rule dn→4/2=2d_n \to 4/2 = 2. For ene_n: here sin⁡n\sin n does not converge (it oscillates), so the Algebraic Limit Theorem's hypotheses do not directly apply to sin⁡n\sin n and n+2n+2 separately as a quotient. Instead, note ∣sin⁡n∣≤1|\sin n| \le 1 for all nn, so

0≤∣sin⁡nn+2∣≤1n+2→0.0 \le \left|\frac{\sin n}{n+2}\right| \le \frac{1}{n+2} \to 0.

This last step is not the Algebraic Limit Theorem — it is a preview of the Order Limit Theorem and Squeeze Theorem, needed precisely because sin⁡n\sin n has no limit and hence the algebraic rules for quotients cannot be invoked term-by-term. Granting that squeeze argument, en→0e_n \to 0. Now apply the Product Rule twice (first to cndnc_n d_n, then to (cndn)en(c_nd_n) e_n):

cndn→1⋅2=2,bn=(cndn)en→2⋅0=0.c_n d_n \to 1\cdot 2 = 2, \qquad b_n = (c_n d_n)e_n \to 2 \cdot 0 = 0.

This example shows both the reach of the theorem (handling three-way products) and its limits: one factor required an outside tool because it was not itself convergent.

Counterexamplewhy M≠0M \neq 0 is necessary in the Quotient Rule

Drop the hypothesis M≠0M \neq 0 and consider an=1a_n = 1 for all nn (so an→L=1a_n \to L = 1) and bn=1/nb_n = 1/n (so bn→M=0b_n \to M = 0). The quotient sequence is

anbn=11/n=n,\frac{a_n}{b_n} = \frac{1}{1/n} = n,

which diverges to +∞+\infty: for any proposed limit ℓ∈R\ell \in \mathbb{R}, taking ε=1\varepsilon = 1 and any NN, we can pick n≥max⁡{N,ℓ+2}n \ge \max\{N, \ell+2\} so that ∣n−ℓ∣≥2>1|n - \ell| \ge 2 > 1, so n↛ℓn \not\to \ell. Hence an/bna_n/b_n has no limit at all — the conclusion of the Quotient Rule fails completely, not merely by giving the "wrong" numeral. This shows M≠0M \neq 0 cannot be weakened or removed: it is not a technical convenience but a load-bearing hypothesis, since the very act of dividing by numbers shrinking to 00 can force unboundedness. A second cautionary case: let an=(−1)n/n→0=La_n = (-1)^n/n \to 0 = L and again bn=1/n→0=Mb_n = 1/n \to 0 = M. Now

anbn=(−1)n,\frac{a_n}{b_n} = (-1)^n,

which does not converge at all (it has two distinct subsequential limits 11 and −1-1), even though here the numerator's limit is also 00. This shows that when M=0M = 0, the quotient's behavior depends delicately on the relationship between ana_n and bnb_n and no single formula like "L/ML/M" can be assigned — reinforcing that M≠0M \neq 0 is exactly the right dividing line.

Scope, edge cases & common traps

SituationWhat happensWhy
M≠0M \neq 0 in Quotient Rulean/bn→L/Ma_n/b_n \to L/M, well-defined eventuallyproved above via 1/bn→1/M1/b_n \to 1/M
M=0M = 0, L≠0L \neq 0an/bna_n/b_n is unbounded near infinity, cannot convergedenominator shrinks to 00 faster than any bound allows
M=0=LM = 0 = Lanything can happen: limit may exist (any value), or not existindeterminate 0/00/0 form; needs extra tools (L'Hôpital-type or direct estimation)
One sequence divergent (e.g. sin⁡n\sin n)Sum/Product Rules simply do not apply; must isolate the divergent factor or use Squeezehypotheses of the theorem require both sequences to converge
Only finitely many terms defined (e.g. bn=0b_n=0 for small nn)Still fine: convergence only concerns tail behaviorchanging/discarding finitely many terms never changes a limit

The mathematician behind it

Origins · 1 minAugustin-Louis Cauchy and how this was discovered
AC
Augustin-Louis Cauchy
1789–1857 · Early 19th-century French analysis

Cauchy, working at the École Polytechnique in Paris, was among the first mathematicians to insist that calculus be built on precise inequalities rather than infinitesimal intuition. In his Cours d'analyse (1821), he formulated convergence in terms recognizably close to the modern ε\varepsilon-NN definition, and it is in exactly this spirit — reducing "limits behave well under arithmetic" to explicit, checkable inequality chases — that the Algebraic Limit Theorem is proved. The rigorous reformulation completed later by Weierstrass turned Cauchy's inequalities into the fully quantified ε\varepsilon-NN language used throughout this article.

History
- 1821 — Cauchy's Cours d'analyse gives an early rigorous treatment of limits and series convergence.
- 1861 — Weierstrass, lecturing in Berlin, formalizes the ε\varepsilon-δ\delta/ε\varepsilon-NN definitions in the form used today.
- 1872 — Dedekind and Cantor construct the real numbers rigorously, securing the ambient field in which these limit laws hold.
- 1976 — Rudin's Principles of Mathematical Analysis (3rd ed.) codifies the algebraic limit theorem as a standard early result in real analysis courses.
- present — the theorem is a universal first step in every rigorous calculus and analysis curriculum, underlying continuity, differentiation, and series manipulations.
Does the Algebraic Limit Theorem also work for sequences of complex numbers or vectors?

Yes, with essentially the same proofs. The triangle inequality ∣z+w∣≤∣z∣+∣w∣|z+w|\le|z|+|w| holds in C\mathbb{C} and in any normed vector space, and boundedness of convergent sequences also holds there, so the Sum, Scalar, and Product Rules (for complex multiplication or for a scalar times a vector) go through verbatim. The Quotient Rule needs a notion of division, so it applies directly to C\mathbb{C} (a field) but not to general vector spaces, where "dividing by a vector" is not defined; there one instead studies quotients by scalars or invertible linear maps.

Can the Sum Rule be extended to infinitely many sequences at once?

Not directly — the theorem as stated combines finitely many sequences. Extending the Sum Rule to a countably infinite family, i.e. asking when lim⁡n∑k=1∞an(k)=∑k=1∞lim⁡nan(k)\lim_n \sum_{k=1}^\infty a_n(k) = \sum_{k=1}^\infty \lim_n a_n(k), is a fundamentally different and harder question involving uniform convergence or dominated convergence type hypotheses, because interchanging two limiting processes is not automatic. The Algebraic Limit Theorem only licenses interchanging a limit with a finite sum, product, or quotient.

What if both ana_n and bnb_n diverge — can an+bna_n + b_n still converge?

Yes, absolutely, and this is precisely why the theorem is stated only as a one-directional implication. For example, an=na_n = n and bn=−nb_n = -n both diverge (to +∞+\infty and −∞-\infty respectively), yet an+bn=0a_n + b_n = 0 for every nn, so the sum sequence converges trivially to 00. The Algebraic Limit Theorem never claims the converse; it only guarantees that if both pieces converge, then the combination converges to the expected value.

Why does the proof of the Product Rule bother with the identity anbn−LM=an(bn−M)+M(an−L)a_nb_n - LM = a_n(b_n-M)+M(a_n-L) instead of a more direct estimate?

Because anbn−LMa_n b_n - LM does not obviously decompose into small pieces on its own; the identity is an algebraic trick (add and subtract anMa_nM) that isolates exactly the two error terms bn−Mb_n - M and an−La_n - L that we know are small, multiplied by quantities (ana_n and MM) that we know are bounded. Without this specific rewriting, there is no way to directly relate ∣anbn−LM∣|a_nb_n - LM| to the two known small quantities ∣an−L∣|a_n-L| and ∣bn−M∣|b_n-M|.

Is it true that if an/bn→L/Ma_n/b_n \to L/M and bn→M≠0b_n \to M \neq 0, then automatically an→La_n \to L?

Yes, and this is a genuine (provable) partial converse: if bn→M≠0b_n \to M \neq 0 and an/bn→ca_n/b_n \to c for some real cc, then since an=(an/bn)⋅bna_n = (a_n/b_n)\cdot b_n, the Product Rule gives an→c⋅Ma_n \to c \cdot M. So ana_n does converge, to cMcM. This is a legitimate converse-style deduction, but note it is really just the forward Product Rule applied to the sequences (an/bn)(a_n/b_n) and (bn)(b_n), not a new independent theorem.

Does the theorem say anything about lim⁡(an)bn\lim (a_n)^{b_n}, i.e., exponent sequences?

No — exponentiation is not one of the four operations covered (sum, scalar multiple, product, quotient), and (an)bn(a_n)^{b_n} can behave very differently, producing genuinely indeterminate forms like 1∞1^\infty or 000^0 even when an→1a_n \to 1 and bn→∞b_n \to \infty. Handling such limits typically requires taking logarithms to convert the exponential expression into a product/quotient (to which the Algebraic Limit Theorem does apply) combined with continuity of exp⁡\exp and log⁡\log, or specialized techniques beyond this theorem's direct scope.

Why is it important that NN in the proofs is allowed to depend on ε\varepsilon?

This dependence is the entire content of the ε\varepsilon-NN definition: the promise of convergence is that for every tolerance ε\varepsilon, some threshold NN (which may need to be larger for smaller ε\varepsilon) works. In every step of the Algebraic Limit Theorem's proof, we choose thresholds N1,N2N_1, N_2 tailored to a given ε\varepsilon (often ε/2\varepsilon/2 or ε/(2B)\varepsilon/(2B)), then take N=max⁡{N1,N2}N = \max\{N_1,N_2\}. If NN could not depend on ε\varepsilon, no nontrivial sequence would converge to anything, since a single fixed NN could never handle arbitrarily small tolerances.

Before this

Where this leads