View source for Math for Non-Geeks/Limit theorems

{{#invoke:Math for Non-Geeks/Seite|oben}}

Limit proofs using the <math>\epsilon</math>-definition are quite laborious. In this chapter we will study some limit theorems that simplifies matters.

== Limit theorems ==
The limit theorems are the following:

{{:Math for Non-Geeks: Template:Satz
 |titel=Limit theorems
 |satz=
Let <math>\left(a_n\right)_{n \in \N}</math> and <math>\left(b_n\right)_{n \in \N}</math> be two convergent sequences with <math>\lim_{n\rightarrow\infty} a_n = a</math> and <math>\lim_{n\rightarrow\infty} b_n = b</math>. Let also <math>k\in\N</math> be arbitrary. Then we have

{{List
 |abstand=1em
 |liste=
* <math>\lim_{n\rightarrow\infty} |a_n| = |a|</math>
* <math>\lim_{n\rightarrow\infty} a_n + b_n = a + b</math>
* <math>\lim_{n\rightarrow\infty} \lambda \cdot a_n = \lambda \cdot a</math> for all <math>\lambda \in \R</math>
* <math>\lim_{n\rightarrow\infty} a_n \cdot b_n = a \cdot b</math>
* <math>\lim_{n\rightarrow\infty} a_n^k = a^k</math>
}}

If furthermore <math>b \neq 0</math> and <math>b_n \neq 0</math> for all <math>n \in \N</math>, then we also have

* <math>\lim_{n\rightarrow\infty} \frac{1}{b_n} = \frac{1}{b}</math>
* <math>\lim_{n\rightarrow\infty} \frac{a_n}{b_n} = \frac{a}{b}</math>

For <math>k \in \N</math> and <math>a_n \ge 0</math> for all <math>n \in \N</math>:

{{Math|<math>\lim_{n\rightarrow\infty} \sqrt[k]{a_n} = \sqrt[k]{a}</math>}}
}}

{{:Math for Non-Geeks: Template:Warnung|1=These rules can only be applied if the respective sequences converge. As soon as one of the involved sequences diverge, we cannot use the rules.

Also note that <math>\infty</math> and <math>-\infty</math> are not real numbers and therefore no valid limits. If, for instance, <math>\lim_{n\rightarrow\infty} a_n = \infty</math>, then <math>\left(a_n\right)_{n \in \N}</math> is divergent and we cannot use the limit theorems.}}

== Monotonicity rule: limit estimates==
We also have the following monotonicity rule, which can be used to estimate limits:

{{:Math for Non-Geeks: Template:Satz
 |titel=Monotonicity rule
 |satz=
Le <math>\left(a_n\right)_{n \in \N}</math> and <math>\left(b_n\right)_{n \in \N}</math> be two convergent sequences. If <math>a_n \le b_n</math> for almost all <math>n \in \N</math>, then the limits satisfy the inequality

{{Math|<math>\lim_{n\rightarrow\infty} a_n \le \lim_{n\rightarrow\infty} b_n</math>}}
}}

== Example: Computing the limit of a sequence ==
Consider the following sequence

{{Math|<math>x_n = \frac{5 - \sqrt[n]{23}}{\sqrt[n]{n} + \tfrac{1}{n^2}}</math>}}

This sequence is convergent. A proof using the <math>\epsilon</math>-definition would be rather complicated. Fortunately we can break the whole sequence apart into individual sequences where the convergence and limits are known. For example <math>\lim_{n\rightarrow\infty} \sqrt[n]{23} = 1</math>. Using the limit theorems we can compute the limit step by step:

{{Math|<math>\begin{align}
\lim_{n\rightarrow\infty} x_n  & = \lim_{n\rightarrow\infty} \frac{5 - \sqrt[n]{23}}{\sqrt[n]{n} + \tfrac{1}{n^2}} \\[0.3em]
& \quad{\color{OliveGreen} \left\downarrow\ \text{quotient rule} \right.} \\[0.3em]
&= \frac{\lim_{n\rightarrow\infty} 5 - \sqrt[n]{23}}{\lim_{n\rightarrow\infty} \sqrt[n]{n} + \tfrac{1}{n^2}} \\[0.3em]
& \quad{\color{OliveGreen} \left\downarrow\ \text{sum rule} \right.} \\[0.3em]
&= \frac{ \left(\lim_{n\rightarrow\infty} 5\right) + \left(\lim_{n\rightarrow\infty} - \sqrt[n]{23}\right)}{\left(\lim_{n\rightarrow\infty} \sqrt[n]{n} \right) + \left(\lim_{n\rightarrow\infty} \tfrac{1}{n^2}\right)} \\[0.3em]
& \quad{\color{OliveGreen} \left\downarrow\ \text{factor rule} \right.} \\[0.3em]
&= \frac{ \left(\lim_{n\rightarrow\infty} 5\right) - \left(\lim_{n\rightarrow\infty} \sqrt[n]{23}\right)}{\left(\lim_{n\rightarrow\infty} \sqrt[n]{n} \right) + \left(\lim_{n\rightarrow\infty} \tfrac{1}{n^2}\right)} \\[0.3em]
& \quad{\color{OliveGreen} \left\downarrow\ \text{calculate known limit} \right.} \\[0.3em]
&= \frac{ 5 - 1}{1 + 0}= 4
\end{align}</math>}}

In this manner we can show that <math>(x_n)_{n\in\N}</math> is convergent and the limit is <math>4</math>. Unfortunately this derivation is flawed: We are applying the limit theorems ''before'' we showed that the individual sequences are convergent. That those sequences do indeed converges becomes clear only after we have already employed the limit theorems. Therefore the above is not a valid proof. Instead we could proceed as follows:

{{Math|<math>\begin{array}{ll}
& \begin{align}
   \lim_{n\rightarrow\infty} 5 = 5 & \land \lim_{n\rightarrow\infty} \sqrt[n]{23} = 1 \land \lim_{n\rightarrow\infty} \sqrt[n]{n} = 1 \\[0.3em]
                                   & \land \lim_{n\rightarrow\infty} \tfrac{1}{n^2} = 0 
  \end{align} \\[0.7em]
& \quad{\color{OliveGreen} \left\downarrow\ \text{factor rule} \right.} \\[0.7em]
\implies & 
  \begin{align}
    \lim_{n\rightarrow\infty} 5 = 5 & \land \lim_{n\rightarrow\infty} -\sqrt[n]{23} = -1 \land \lim_{n\rightarrow\infty} \sqrt[n]{n} = 1 \\[0.3em]
                                    & \land \lim_{n\rightarrow\infty} \tfrac{1}{n^2} = 0
  \end{align} \\[0.7em]
& \quad{\color{OliveGreen} \left\downarrow\ \text{sum rule} \right.} \\[0.7em]
\implies & \lim_{n\rightarrow\infty} 5 -\sqrt[n]{23} = 5-1=4 \land \lim_{n\rightarrow\infty} \sqrt[n]{n} + \tfrac{1}{n^2} = 1 + 0 = 1 \\[0.7em]
& \quad{\color{OliveGreen} \left\downarrow\ \text{quotient rule} \right.} \\[0.7em]
\implies & \lim_{n\rightarrow\infty} \frac{5 -\sqrt[n]{23}}{\sqrt[n]{n} + \tfrac{1}{n^2}} = \frac 41 = 4
\end{array}</math>}}

We start with the sequences of which we know that they converge. By applying the limit theorems in reverse order we can derive the convergence and limit of the original sequence <math>(x_n)_{n\in\N}</math>. The symbol <math>\land</math> is  the logical [[Math for Non-Geeks: Junktor#Anchor:Konjunktion|conjunction]], which should be read as "and".

Writing the proof in the above way is time-consuming and no fun. Most of the time we prefer writing down the first version. We use the limit theorems even though we don't know if the sequences converge. We must argue afterwards, that it was okay to use the limit theorems in the first place. But this is the case, because at the end, everything converges. Since the last steps worked, we were allowed to do the steps before. So if we want to write the proof like in the first version, we need to make sure at the end to give a justification why applying the limit theorems was a valid thing to do.

== Problems with divergent sequences ==
As stated many times, we cannot use the limit theorems if one of the partial sequences diverges. If we forget this, we can quickly get nonsensical results:

{{Math|<math>\begin{align}
1 & = \lim_{n\rightarrow\infty} 1 \\
& = \lim_{n\rightarrow\infty} n\cdot \tfrac 1n \\
& = \left(\lim_{n\rightarrow\infty} n\right) \cdot \left(\lim_{n\rightarrow\infty} \tfrac 1n\right) \\
& = \infty \cdot 0 \\
& = 0
\end{align}</math>}}

{{:Math for Non-Geeks: Template:Frage
 |frage=What is wrong in the above argumentation?
 |antwort=
<math>\infty</math> is not a real number and thus <math>(n)_{n\in\N}</math> is a divergent sequence with <math>\lim_{n\rightarrow\infty} n = \infty</math>. A sequence is convergent only if the limit is a real number. The product rule <math>\lim_{n\rightarrow\infty} a_n \cdot b_n = \left(\lim_{n\rightarrow\infty} a_n\right) \cdot \left(\lim_{n\rightarrow\infty} b_n\right)</math> cannot be applied.
}}

== Proof of limit theorems ==
=== Absolute value rule ===
{{:Math for Non-Geeks: Template:Satz
 |titel=Absolute value limit rule
 |satz=
Let <math>(a_n)_{n\in\N}</math> be a convergent sequence with limit <math>a</math>. Then <math>\lim_{n\rightarrow\infty} |a_n|=|a|</math>.

 |lösungsweg-noprint=ja
 |lösungsweg=

If <math>\lim_{n\to\infty} a_n = a</math> then the distance <math>|a_n-a|</math> becomes arbitrary small. We need to show that <math>||a_n|-|a||</math> also becomes arbitrary small. In the chapter [[Math for Non-Geeks: Betrag, Maximum und Minimum#Anchor:Betrag|Absolute value]] we proved the following inequality:

{{Math|<math>||x|-|y|| \le |x-y|</math>}}

Thus we have

{{Math|<math>||a_n|-|a|| \le |a_n-a|</math>}}

This means that if <math>|a_n-a|</math> is smaller than <math>\epsilon</math>, by transitivity this is true also for <math>||a_n|-|a||</math>. We can use this in our convergence proof. Let <math>\epsilon > 0</math>. We need to find <math>N\in\N</math>, so that <math>||a_n|-|a|| < \epsilon</math> for all <math>n\ge N</math>. Because of <math>\lim_{n\to\infty} a_n = a</math> we know that there is <math>\tilde N\in\N</math>, so that <math>|a_n-a|<\epsilon</math> for all <math>n\ge \tilde N</math>.

But as we have seen, if <math>|a_n-a|<\epsilon</math> then also <math>||a_n|-|a|| < \epsilon</math>. So we can set <math>N=\tilde{N}</math>. Since <math>|a_n-a|<\epsilon</math> is true for all <math>n \ge N</math>, it follows that <math>||a_n|-|a|| < \epsilon</math> is true for all <math>n\ge N</math>.

 |beweis=

Let <math>\epsilon > 0</math> be arbitrary. Because <math>(a_n)_{n\in\N}</math> converges to <math>a</math>, there is  <math>N \in \N</math> with <math>|a_n-a| < \epsilon</math> for all <math>n \ge N</math>. From the inverted triangle inequality <math>||x|-|y|| \le |x-y|</math> it follows

{{Math|<math>||a_n|-|a|| \le |a_n-a| < \epsilon</math>}}

for all <math>n \ge N</math>.
}}



=== Inversion of the absolute value rule ===
If <math>(|a_n|)_{n\in \N}</math> is a null sequence, then the inversion of the absolute value rule holds, as well. From <math>\lim_{n\to \infty} |a_n|=0</math> , we have <math>\lim_{n\to \infty} a_n=0</math>:

{{:Math for Non-Geeks: Template:Satz
 |satz=
Let <math>(a_n)_{n\in\N}</math> be a sequence. If <math>\lim_{n\to \infty} |a_n|=0</math> , then <math>(a_n)_{n\in\N}</math> converges to zero, i.e. <math>\lim_{n\to \infty} a_n=0</math>.

 |beweis=
Since <math>\lim_{n\to \infty} |a_n|=0</math> , we have that

{{-|For every <math>\epsilon>0</math> there is an <math>N\in \N</math> with <math>||a_n|-0|<\epsilon</math> for all <math>n \ge N</math>.}}

Now, <math>||a_n|-0|=||a_n||=|a_n|=|a_n-0|</math>. Therefore, we also have that

{{-|For every <math>\epsilon>0</math> there is an <math>N\in \N</math> with <math>|a_n-0|<\epsilon</math>  for all <math>n \ge N</math>.}}

Or in other words, <math>\lim_{n\to \infty} a_n=0</math>.
}}

This inversion only holds for null sequences. For a general sequence, it may not hold. An example is the divergent sequence <math>a_n=(-1)^n</math>. The absolute value sequence <math>\lim_{n\to \infty} |a_n|=1</math> converges. So <math>\lim_{n\to \infty} |a_n|=1</math> but <math>\lim_{n\to \infty} a_n \neq 1</math>. The latter limit even does not exist.

=== The sum rule ===
{{:Math for Non-Geeks: Template:Satz
 |titel=Limit theorems for sums
 |satz=
Let <math>(a_n)_{n\in\N}</math> be a sequence converging to <math>a</math> and <math>(b_n)_{n\in\N}</math> a sequence converging to <math>b</math>. Then, the sequence <math>(a_n+b_n)_{n\in\N}</math> also converges with <math>\lim_{n\rightarrow\infty} a_n + b_n = a+b</math>.

 |lösungsweg=

We need to prove that <math>|(a_n+b_n)-(a+b)|</math> gets arbitrarily small. What we can use is that <math>|a_n-a|</math> and <math>|b_n-b|</math> get arbitrarily small. That means, we need to construct an upper bound for <math>|(a_n+b_n)-(a+b)|</math> which uses <math>|a_n-a|</math> and/ or <math>|b_n-b|</math> . The trick is to use the triangle inequality in <math>(a_n+b_n)-(a+b)</math>:

{{Math|<math>\begin{align}
|(a_n+b_n)-(a+b)| & = |(a_n-a) + (b_n - b)| \\
& \le |a_n-a| + |b_n-b|
\end{align}</math>}}

Since <math>|a_n-a|</math> and <math>|b_n-b|</math> get arbitrarily small, this estimate should suffice. It remains to establish the bounds in terms of <math> \epsilon </math>. This is done by bounding each of the two summands against <math>\tfrac \epsilon2</math> : When we get <math>|a_n-a| < \tfrac \epsilon2</math> and <math>|b_n-b| < \tfrac \epsilon2</math> , then

{{Math|<math>|a_n-a| + |b_n-b| < \tfrac\epsilon2 + \tfrac \epsilon2 = \epsilon</math>}}

We know that for some <math>N_1</math> there is <math>|a_n-a| < \tfrac\epsilon2</math> for all <math>n \ge N_1</math> . Analogously, there is some <math>N_2</math> with <math>|b_n-b| < \tfrac\epsilon2</math> for all <math>n \ge N_2</math>. For the proof, we need <math>|a_n-a| < \tfrac \epsilon2</math> and <math>|b_n-b| < \tfrac \epsilon2</math> simultaneously. So both <math>n \ge N_1</math> and <math>n \ge N_2</math> should hold for all <math>n\ge N</math> with some suitable <math>N\in\N</math> . The smallest suitable <math>N</math> is then given by <math>N = \max\{N_1,\,N_2\}</math> : From <math>n \ge \max\{N_1,\,N_2\}</math> , we get <math>n \ge N_1</math> and <math>n \ge N_2</math>.

 |beweis=

Let <math>\epsilon > 0</math> be arbitrary. Then, there is an <math>N_1</math> with <math>|a_n-a| < \tfrac{\epsilon}{2}</math> for all <math>n \ge N_1</math>, since <math>\lim_{n\rightarrow\infty} a_n = a</math> . Analogously, since <math>\lim_{n\to\infty} b_n = b</math> there is an <math>N_2</math> with <math>|b_n-b| < \tfrac \epsilon2</math> for all <math>n \ge N_2</math>. We choose <math>N = \max\{N_1,\,N_2\}</math>. Let <math>n \ge N</math>  be arbitrary. Then,

{{Math|<math>\begin{align}
|(a_n+b_n) - (a+b) | & = |(a_n-a) + (b_n - b)| \\[0.5em]
& \quad {\color{OliveGreen} \left\downarrow\ \text{triangle inequality}\right.}\\[0.5em]
& \le |a_n-a| + |b_n-b| \\
& < \tfrac \epsilon2 + \tfrac \epsilon2 = \epsilon
\end{align}</math>}}
}}

=== The factor rule ===
{{:Math for Non-Geeks: Template:Satz
 |titel=Factor rule for limits
 |satz=
Let <math>\lambda \in \R</math> be arbitrary and <math>(a_n)_{n\in\N}</math> a converging sequence with limit <math>a</math>. Then, the sequence <math>(\lambda a_n)_{n\in\N}</math> converges, as well with <math>\lim_{n\rightarrow\infty} \lambda a_n = \lambda a</math>.

 |lösungsweg=

In order to prove <math>\lim_{n\rightarrow\infty} \lambda a_n = \lambda a</math> , we need to establish a bound <math>|\lambda a_n - \lambda a| < \epsilon</math> for almost all <math>n \in \N</math> . This "to-be-achieved"-inequality can be equivalently reformulated:

{{Math|<math>\begin{array}{lrl}
& |\lambda a_n - \lambda a| & < \epsilon \\
\iff{} & |\lambda \cdot (a_n - a)| & < \epsilon \\
\iff{} & |\lambda| \cdot |a_n - a| & < \epsilon
\end{array}</math>}}

It is not allowed to directly divide by <math>|\lambda|</math> , since there might be <math>\lambda = 0</math> . However, we can easily treat the case <math>\lambda = 0</math> : there, we need to show <math>\lim_{n\to\infty} \lambda a_n = \lambda a = 0\cdot a = 0</math> . And since <math>\lambda a_n = 0 \cdot a_n = 0</math> , we trivially have <math>\lim_{n\to\infty} \lambda a_n = \lim_{n\to\infty} 0 = 0</math>. So the assumption is established within case <math>\lambda = 0</math> . In the case <math>\lambda \neq 0</math> , we can divide by <math>|\lambda|</math>:

{{Math|<math>\begin{array}{lrl}
& |\lambda| \cdot |a_n - a| & < \epsilon \\
\iff{} &  |a_n - a| & < \frac{\epsilon}{|\lambda|}
\end{array}</math>}}

Since <math>|a_n-a|</math> converges to <math>0</math> , there is an <math>N\in \N</math>, such that <math>|a_n-a|< \tfrac \epsilon{|\lambda|}</math> for all <math>n \ge N</math>.
}}
{{noprint|
{{:Math for Non-Geeks: Template:Beweis
 |titel=Factor rule for limits
 |beweis=

Let <math>\epsilon > 0</math> be arbitrary. in addition, let <math>\lambda \in \R</math> and <math>(a_n)_{n\in\N}</math> a sequence converging to <math>a</math>.

{{:Math for Non-Geeks: Template:Fallunterscheidung
 |fall1=<math>\lambda = 0</math>
 |beweis1=
There is <math>\lambda a_n = 0 \cdot a_n = 0</math> so we have convergence

{{Math|<math>\lim_{n\rightarrow\infty} \lambda a_n = \lim_{n\rightarrow\infty} 0 = 0 = 0\cdot a = \lambda \cdot a </math>}}

 |fall2=<math>\lambda \neq 0</math>
 |beweis2=
Choose <math>N</math> such that <math>|a_n-a|< \tfrac \epsilon{|\lambda|}</math> for all <math>n \ge N</math> . Such an <math>N</math> can be found, since <math>(a_n)_{n\in\N}</math> converges to <math>a</math> . Then,

{{Math|<math>\begin{array}{lrl}
&  |a_n - a| & < \frac{\epsilon}{|\lambda|} \\
\implies{} & |\lambda| \cdot |a_n - a| & < \epsilon \\
\implies{} & |\lambda \cdot (a_n - a)| & < \epsilon \\
\implies{} & |\lambda a_n - \lambda a| & < \epsilon
\end{array}</math>}}

this shown the assumption <math>\lim_{n\to\infty} \lambda a_n = \lambda a</math> .
}}}}
}}

=== The product rule ===
{{:Math for Non-Geeks: Template:Satz
 |titel=Product rule for limits
 |satz=
Let <math>(a_n)_{n\in\N}</math> be a sequence converging to <math>a</math> and <math>(b_n)_{n\in\N}</math> a sequence converging to <math>b</math>. Then, the product sequence <math>(a_n\cdot b_n)_{n\in\N}</math> converges as well with <math>\lim_{n\rightarrow\infty} a_n \cdot b_n = a\cdot b</math>.

 |beweis=

Let <math>\epsilon > 0</math> be arbitrary.

We need to show that <math>|a_nb_n-ab|<\epsilon</math> for all <math>n\ge N</math> , where we need to choose <math>N</math> depending on <math>\epsilon</math> . What we can use is that <math>|a_n-a|</math> and <math>|b_n-b|</math> get arbitrarily small, since <math>(a_n)_{n\in\N}</math> converges to <math>a</math> and <math>(b_n)_{n\in\N}</math> converges to <math>b</math> .This requires a deliberate reformulation of <math>|a_nb_n-ab|</math> , such that we get an upper bound including <math>|a_n-a|</math> and <math>|b_n-b|</math> . The trick is to add a "smart zero" in the form of <math>a b_n-a b_n</math>:

{{Math|<math>\begin{align}
& |a_nb_n-ab| \\[0.3em]
& \quad {\color{OliveGreen}\left\downarrow\ a b_n-a b_n=0 \right.} \\[0.3em]
=\ & |a_nb_n-ab+ab_n-ab_n| \\[0.3em]
& \quad {\color{OliveGreen}\left\downarrow\ \text{reformulate} \right.} \\[0.3em]
=\ & |a_nb_n-ab_n+ab_n-ab| \\[0.3em]
& \quad {\color{OliveGreen}\left\downarrow\ \text{factor out} \right.} \\[0.3em]
=\ & |b_n(a_n-a)+a(b_n-b)| \\[0.3em]
& \quad {\color{OliveGreen}\left\downarrow\ \text{triangle inequality} \right.} \\[0.3em]
\le\ & |b_n(a_n-a)|+|a(b_n-b)| \\[0.3em]
\le\ & |b_n|\cdot |a_n-a|+|a|\cdot |b_n-b|
\end{align}</math>}}

If we can get the two summands below <math>\tfrac{\epsilon}{2}</math> for all<math>n\ge N</math> , then we are done.

'''Bounding the second summand'''

The sequence <math>(b_n)_{n\in\N}</math> converges to <math>b</math> and by the factor rule with <math>\lambda = a</math> we have <math>\lim_{n\to\infty} a\cdot b_n=a\cdot b</math>. Hence, <math>\lim_{n\to\infty} a\cdot(b_n-b)=0</math> by the sum rule, so there is an <math>N_1\in\N</math> such that for all <math>n\ge N_1</math> we have <math>|a|\cdot|b_n-b|=|a(b_n-b)|<\tfrac{\epsilon}{2}</math>.

'''Bounding the first summand'''

This term is a bit more complicated, since the factor <math>|b_n|</math> is not constant. However, we can bound it from above by a constant: 

In the previous chapter, we have seen that convergent sequences are bounded. So there is an <math>M\in\R_0^+</math> with <math>|b_n|\le M</math> for all <math>n\in\N</math>. 

Now we can replace <math>|b_n|</math> within the bound by the constant <math>M</math> and we have <math>|b_n|\cdot |a_n-a|\le M\cdot |a_n - a|</math> . Now, we proceed as for the second summand. We use the factor rule with <math>\lambda =M</math> and get some <math>N_2\in\N</math> with <math>M\cdot |a_n-a|<\tfrac{\epsilon}{2}</math> for all <math>n\ge N_2</math>, which in turn implies <math>|b_n|\cdot |a_n-a|<\tfrac{\epsilon}{2}</math>. 

'''Conclusion'''

It remains to choose a suitable <math>N</math>. For any <math>n\ge N</math> we need to have <math>n\ge N_1</math> and <math>n\ge N_2</math> . This can be done by choosing <math>N:=\max\{N_1,N_2\}</math>, which only depends on <math>\epsilon</math>, as <math>N_1</math> and <math>N_2</math> also only depend on <math>\epsilon</math>. 

Now, for all <math>n\ge N</math> there is
{{Math|<math>|a_nb_n-ab|\le\  |b_n|\cdot |a_n-a|+|a|\cdot |b_n-b|<\tfrac{\epsilon}{2}+ \tfrac{\epsilon}{2}=\epsilon</math>}}
}}

=== The power rule ===
{{:Math for Non-Geeks: Template:Satz
 |titel=Limit theorems for powers
 |satz=
Let <math>(a_n)_{n\in\N}</math> be a sequence converging to <math>a</math>. Let <math>k \in \N</math> be any natural number . Then, the sequence <math>\left(a_n^k\right)_{n\in\N}</math> also converges with <math>\lim_{n\to\infty} a_n^k = a^k</math>.

 |lösungsweg-noprint=ja
 |lösungsweg=
The power rule is a consequence of the product rule- For the sequence <math>\left(a_n^2\right)_{n\in\N}</math> we have:

{{Math|<math>\begin{align}
\lim_{n\to \infty} a_n^2 & = \lim_{n \to \infty} a_n \cdot a_n \\[0.5em]
& \ {\color{OliveGreen} \left\downarrow\ \text{product rule} \right.} \\[0.5em]
& = \lim_{n \to \infty} a_n \cdot \lim_{n \to \infty} a_n \\
& = a\cdot a = a^2
\end{align}</math>}}

Repeating this step <math>k</math> times yields:

{{Math|<math>\begin{align}
\lim_{n\to \infty} a_n^k & = \lim_{n \to \infty} \underbrace{a_n \cdot a_n \cdot \ldots \cdot a_n}_{k\text{ times}} \\[0.5em]
& \ {\color{OliveGreen} \left\downarrow\ \text{product rule} \right.} \\[0.5em]
&= \underbrace{\lim_{n \to \infty} a_n \cdot \lim_{n \to \infty} a_n \cdot \ldots \cdot \lim_{n \to \infty} a_n}_{k\text{ times}} \\[0.5em]
&= \underbrace{a\cdot a \cdot \ldots \cdot a}_{k \text{ times}} = a^k
\end{align}</math>}}

Now, the above argument with dots is not accepted as a proper proof in mathematics. The "clean" formulation is via induction in <math>k</math>.

 |beweis=
The proof is by induction, using the product rule.
{{:Math for Non-Geeks: Template:Vollständige Induktion
 |erfuellungsmenge=<math>k\in\N</math>

 |aussageform=
{{Math|<math>\lim_{n\to\infty} a_n^k = a^k</math>}}

 |induktionsanfang=
{{Math|<math>\lim_{n\to\infty} a_n^1 = \lim_{n\to\infty} a_n = a = a^1</math>}}

 |induktionsvoraussetzung=
{{Math|<math>\lim_{n\to\infty} a_n^k = a^k</math>}}

 |induktionsbehauptung=
{{Math|<math>\lim_{n\to\infty} a_n^{k+1} = a^{k+1}</math>}}

 |beweis_induktionsschritt=
{{Math|<math>\begin{align}
\lim_{n \to \infty} a_n^{k+1} & = \lim_{n \to \infty} a_n^{k}\cdot a_n \\[0.5em]
& \ {\color{OliveGreen} \left\downarrow\ \text{product rule} \right.} \\[0.5em]
& = \lim_{n \to \infty} a_n^{k}\cdot \lim_{n \to \infty}a_n \\[0.5em]
& \ {\color{OliveGreen} \left\downarrow\ \text{induction assumption} \right.} \\[0.5em]
& = a^k \cdot a = a^{k+1}
\end{align}</math>}}

}}
}}
{{noprint|{{:Math for Non-Geeks: Template:Beispiel
 |titel=An example for the power rule
 |beispiel=
We can show for all <math>k \in \N</math> that the sequence <math>\left(\sqrt[n]{n^k}\right)_{n\in\N}</math> converges to <math>1</math>:

{{Math|<math>\begin{align}
\lim_{n \to \infty} \sqrt[n]{n^k} & = \lim_{n \to \infty} (\sqrt[n]{n})^k \\
& = \left(\lim_{n \to \infty} \sqrt[n]{n}\right)^k \\
& = 1^k = 1
\end{align}</math>}}
}}}}

=== The quotient rule {{Anchor|quotient rule}} ===
{{:Math for Non-Geeks: Template:Satz
 |titel=Quotient rule for limits
 |satz=
Let <math>(a_n)_{n\in\N}</math> be a sequence converging to <math>a</math> and <math>(b_n)_{n\in\N}</math> a sequence converging to <math>b\neq 0</math> where <math>b_n \neq 0</math> for all <math>n \in \N</math>. Then, the quotient sequence <math>(\tfrac{a_n}{b_n})_{n\in\N}</math> converges with <math>\lim_{n\rightarrow\infty} \tfrac{a_n}{b_n} = \tfrac{a}{b}</math>.

 |lösungsweg=
It suffices to show <math>\lim_{n \to \infty} \tfrac{1}{b_n} = \tfrac 1b</math> . Then, by the product rule, we get

{{Math|<math>\lim_{n \to \infty} \tfrac{a_n}{b_n} = \lim_{n \to \infty} a_n \cdot \tfrac{1}{b_n} = \lim_{n \to \infty} a_n \cdot \lim_{n \to \infty} \tfrac{1}{b_n} = a \cdot \tfrac 1b = \tfrac ab</math>}}

In order to show <math>\lim_{n \to \infty} \tfrac{1}{b_n} = \tfrac 1b</math>, we need to show that <math>|\tfrac{1}{b_n}-\tfrac{1}{b}|</math> gets arbitrarily small. We may use that <math>|b_n-b|</math> gets arbitrarily small, as <math>(b_n)_{n\in\N}</math> converges to <math>b</math> . So we need to re-formulate <math>|\tfrac{1}{b_n}-\tfrac{1}{b}|</math> in a suitable way:

{{Math|<math>\begin{align}
\left| \frac{1}{b_n}-\frac{1}{b}\right| & = \left|\frac{b}{b_nb}-\frac{b_n}{bb_n}\right| \\[0.3em]
& =  \left|\frac{b-b_n}{b_nb}\right| \\[0.3em]
& = \frac{|b-b_n|}{|b_n||b|}
\end{align}</math>}}

The term <math>|b_n-b|</math> can be controlled, meaning we can make it arbitrarily small. The <math>|b|</math> in the denominator is just a constant and will not affect the convergence. The term <math>|b_n|</math> needs some more work, since it depends on <math> n </math>. More precisely, we need to bound <math>\tfrac{1}{|b_n|}</math> from above by a constant. That means, we require a lower bound for <math>|b_n|</math>.

By assumption, <math>\lim_{n \to \infty} b_n = b \neq 0</math>, so <math> b </math> keeps a positive distance from 0. Due to convergence, we have an <math>N_1 \in \N</math>, such that all sequence elements <math>(b_n)_{n\in\N}</math> with <math> n \ge N_1 </math> satisfy <math>|b_n-b|<\tfrac{|b|}2</math> , i.e. they are not further away from <math> b </math> than half the distance from <math> b </math> to 0. Hence, <math>|b_n|\ge|b|-|b_n-b|\ge|b|-\tfrac{|b|}2=\tfrac{|b|}2</math> for all <math>n \ge N_1</math>. For the entire expression, that means

{{Math|<math>\begin{align}
\left| \frac{1}{b_n}-\frac{1}{b}\right| & = \frac{|b-b_n|}{|b_n||b|} \\[0.5em]
& \le \frac{|b-b_n|}{\tfrac{|b|}{2}|b|} \\[0.5em]
& = \frac{2}{|b|^2} |b-b_n|
\end{align}</math>}}

This expression can be made arbitrarily small, as <math>|b-b_n|</math> can be chosen arbitrarily small. For any given <math>\epsilon >0</math> we choose an <math>N_2 \in \N</math> such that for all <math>n \ge N_2</math> there is

{{Math|<math>\frac{2}{|b|^2} |b-b_n| < \epsilon \iff |b_n-b| < \frac{\epsilon |b|^2}{2}</math>}}

For all <math>n \ge \max \{N_1,N_2\}</math> it then follows that:

{{Math|<math>\begin{align}
\left| \frac{1}{b_n}-\frac{1}{b}\right| & \le \frac{2}{|b|^2} |b-b_n| \\[0.5em]
& < \frac{2}{|b|^2} \cdot \frac{\epsilon |b|^2}{2} \\[0.5em]
& = \epsilon
\end{align}</math>}}

and we will be done with the proof. Now, let's write down the proof for <math>\lim_{n \to \infty} \tfrac{1}{b_n} = \tfrac 1b</math> in a formal way.

 |beweis-noprint=ja
 |beweis=

Let <math>\epsilon > 0</math> be given. Since <math>\lim_{n \to \infty} b_n = b \neq 0</math> there is an <math>N_1\in\N</math>, such that <math>|b_n| \ge \tfrac{|b|}{2}</math> for all <math>n \ge N_1</math> . In addition, there is an <math>N_2 \in \N</math> with <math>|b_n-b| < \frac{\epsilon |b|^2}{2}</math> for all <math>n \ge N_2</math>. Then, for <math>n \ge \max \{ N_1,N_2\}</math>, we have:

{{Math|<math>\begin{align}
\left| \frac{1}{b_n}-\frac{1}{b}\right| & = \left|\frac{b}{b_nb}-\frac{b_n}{bb_n}\right| \\[0.3em]
& =  \left|\frac{b-b_n}{b_nb}\right| \\[0.3em]
& = \frac{|b-b_n|}{|b_n||b|} \\[0.5em]
& \le \frac{|b-b_n|}{\tfrac{|b|}{2}|b|} \\[0.5em]
& = \frac{2}{|b|^2} |b-b_n| \\[0.5em]
& < \frac{2}{|b|^2} \cdot \frac{\epsilon |b|^2}{2} \\[0.5em]
& = \epsilon
\end{align}</math>}}

Therefore, <math>\lim_{n \to \infty} \tfrac1b_n = \tfrac 1b</math>. The product rule now implies

{{Math|<math>\lim_{n \to \infty} \tfrac{a_n}{b_n} = \lim_{n \to \infty} a_n \cdot \tfrac{1}{b_n} = \lim_{n \to \infty} a_n \cdot \lim_{n \to \infty} \tfrac{1}{b_n} = a \cdot \tfrac 1b = \tfrac ab</math>}}
}}

=== The root rule ===

The following rule is in fact a generalization to the power rule above. It extends its validity from integer powers (like <math> a^m, m \in \N </math>) to powers of the form <math> a^{1/k} = \sqrt[k]{a}</math>. Combining both rules, we get a limit theorem for all positive rational powers (like <math> a^{m/k}</math>).

{{:Math for Non-Geeks: Template:Satz
 |titel=Limit theorems for roots
 |satz=
Let <math>(a_n)_{n\in\N}</math> be a non-negative sequence converging to <math>a</math>. In addition, let <math>k \in \N</math>. Then, the sequence <math>(\sqrt[k]{a_n})_{n\in\N}</math> converges to <math>\lim_{n\to\infty} \sqrt[k]{a_n} = \sqrt[k]{a}</math>.

 |lösungsweg-noprint=ja
 |lösungsweg=

We need to find an upper bound for the absolute value <math>|\sqrt[k]{a_n}-\sqrt[k]{a}|</math> . What we can use is that <math>|a_n-a|</math> can be made arbitrarily small. Both expressions can be related by an auxiliary formula: For <math>x\ge y\ge0</math> and <math>k \in \N</math> there is

{{Math|<math>0<\sqrt[k]{x}-\sqrt[k]{y} \le \sqrt[k]{x-y}</math>}}

The reason why this holds is that the root function <math>f(x) = \sqrt[k]{x} </math> is concave (meaning it curves down). You may draw this function yourself and verify why this holds.
Taking the absolute of this equation, we get:

{{Math|<math>|\sqrt[k]{x}-\sqrt[k]{y}| = \sqrt[k]{x}-\sqrt[k]{y} \le \sqrt[k]{x-y} = \sqrt[k]{|x-y|}</math>}}

For <math>y\ge x\ge 0</math>, we get 

{{Math|<math>|\sqrt[k]{x}-\sqrt[k]{y}| = \sqrt[k]{y}-\sqrt[k]{x} \le \sqrt[k]{y-x} = \sqrt[k]{|x-y|}</math>}}

Hence, <math>|\sqrt[k]{x}-\sqrt[k]{y}| \le \sqrt[k]{|x-y|}</math>. We apply this auxiliary formula with <math>x=a_n</math> and <math>y=a</math>:

{{Math|<math>|\sqrt[k]{a_n}-\sqrt[k]{a}| \le \sqrt[k]{|a_n-a|}</math>}}

The expression <math>\sqrt[k]{|a_n-a|}</math> can be made arbitrarily small by setting <math>|a_n-a|</math> "small enough":

{{Math|<math>\sqrt[k]{|a_n-a|} < \epsilon \iff |a_n-a| < \epsilon^k</math>}}

So <math>|a_n-a| < \epsilon^k</math> is "small enough" and we have proven the desired inequality <math>|\sqrt[k]{a_n}-\sqrt[k]{a}| < \epsilon</math>.

 |beweis=

Let <math>k \in \N</math> and <math>(a_n)_{n\in\N}</math> be a non-negative sequence converging to <math>a</math>. In addition, let <math>\epsilon > 0</math> be given. Since <math>\lim_{n\rightarrow\infty} a_n = a</math> there is an <math>N \in \N</math> with <math>|a_n-a| < \epsilon^k</math> for all <math>n \ge N</math>. Hence, for <math>n \ge N</math> there is

{{Math|<math>\begin{align}
|\sqrt[k]{a_n}-\sqrt[k]{a}| & \underset{\text{formula}}{\overset{\text{auxiliary-}}{\le}} \sqrt[k]{|a_n-a|} \\[0.3em]
& \quad{\color{OliveGreen} \left\downarrow\ |a_n-a| < \epsilon \right.}\\[0.3em]
& < \sqrt[k]{\epsilon^k} = \epsilon
\end{align}</math>}}
}}

=== The monotonicity rule ===
{{:Math for Non-Geeks: Template:Satz
 |titel=Monotonicity rule for limits
 |satz=
Let <math>(a_n)_{n\in\N}</math> and <math>(b_n)_{n\in\N}</math> be sequences converging to <math>a</math> and <math>b</math>. In addition, let <math>a_n \le b_n</math> for almost all <math>n \in \N</math>. Then, we also have <math>a \le b</math>.

 |zusammenfassung=
The proof is by contradiction: We assume <math>a > b</math> and show that this cannot hold true.

 |beweis=
Assume <math>a>b</math>. Since <math>\lim_{n\to \infty} a_n=a</math> and <math>\lim_{n\to \infty} b_n=b</math>, for <math>\epsilon = \tfrac{a-b}{2}>0</math> (an interval width, which is half of the spacing between <math>a</math> and <math>b</math>), there are thresholds <math>N_1,N_2\in \N</math> such that <math>|a_n-a|<\tfrac{a-b}{2}</math> for all <math>n \ge N_1</math> and <math>|b_n-b|<\tfrac{a-b}{2}</math> for all <math>n \ge N_2</math>. Hence for <math>n \ge \max \{ N_1,N_2 \}</math> (beyond both thresholds):

{{Math|<math>\begin{align}
b_n-a_n & = b_n-a_n+0+0\\
& = b_n-a_n+b-a+a-b\\
& = b_n-b+(b-a)+a-a_n\\
& \le |b_n-b|+(b-a)+|a_n-a|\\
& < \tfrac{a-b}{2}+(b-a)+\tfrac{a-b}{2}\\
& = a-b+b-a = 0
\end{align}</math>}}

Therefore, <math>b_n<a_n</math> for all <math>n \ge \max \{ N_1,N_2 \}</math>. This contradicts <math>a_n \le b_n</math> which must hold for almost all <math>n</math>. So the assumption <math>a > b</math> was wrong and <math>a\le b</math> must hold.
}}

=== Remarks concerning the monotonicity rule ===

There is a special case: we consider a constant <math>b_n=c</math> :

{{-|Let <math>(a_n)_{n\in\N}</math> be a sequence covering to  <math>a</math> and <math>a_n \le c</math> (or analogously, <math>a_n \ge c</math>) for almost all <math>n \in \N</math>. Then <math>a \le c</math> (or analogously, <math>a \ge c</math>).}}

The above proposition implies:

{{-|Let <math>(a_n)_{n\in\N}</math> be a convergent sequence and almost all elements are within the interval <math>I=[a,b]</math>. Then also the limit must be within the interval <math>I</math>.}}

Both cases can be connected: „<math>a_n \le b_n</math>“ and „<math>a_n \ge b_n</math>“:

{{-|Let <math>(a_n)_{n\in\N}</math> and <math>(b_n)_{n\in\N}</math> be sequences converging to <math>a</math> and <math>b</math>, and let <math>a_n = b_n</math> for almost all <math>n \in \N</math>. Then, both limits also coincide: <math>a = b</math>.}}

{{:Math for Non-Geeks: Template:Warnung|
The monotonicity rule does not hold for "<math> < </math> and <math> > </math>", since within the limit, strict inequalities "become <math> \le </math> and <math> \ge </math>". An example are the sequences <math>(a_n)_{n\in\N}=(-\tfrac 1n)_{n\in\N}</math> and <math>(b_n)_{n\in\N}=(\tfrac 1n)_{n\in\N}</math>. In this case <math>a_n < 0</math> for all <math>n \in \N</math>, but in the limit, there is <math>\lim_{n \to \infty} a_n = 0</math> . So the limit of <math>(a_n)_{n\in\N}</math> is not strictly smaller than <math>0</math>. Analogously, for all <math>n\in\N</math> there is <math>b_n > 0</math> but in the limit,<math>\lim_{n\to \infty} b_n = 0 \ngtr 0</math>.
}}

{{#invoke:Math for Non-Geeks/Seite|unten}}

{{BookCat}}