Théorèmes de comparaison.

Comparison theorems are among the fundamental theorems for the study of numerical sequences/series:

Theorem: Let \(\sum_{n\in\mathbb{N}} u_n\) and \(\sum_{n\in\mathbb{N}} v_n\) be two series with not negative real terms (the theorem applies more generally to series with terms of constant sign from a certain rank).

i.

Assume: \[\exists n_0\in\mathbb{N}\quad \forall n\geqslant n_0,\quad u_n\leqslant v_n\] Then we have: \[\sum_{n\in\mathbb{N}} v_n \text{ converges } \Rightarrow \sum_{n\in\mathbb{N}} u_n \text{ converges}\]

ii.

Assume: \[u_n\underset{}{=}o(v_n)\] Then we have: \[\sum_{n\in\mathbb{N}} v_n \text{ converges } \Rightarrow \sum_{n\in\mathbb{N}} u_n \text{ converges}\]

iii.

Assume: \[u_n \sim v_n\] Then we have: \[\sum_{n\in\mathbb{N}} v_n \text{ converges } \Leftrightarrow \sum_{n\in\mathbb{N}} u_n \text{ converges}\]

Moreover in the convergence case we have equivalence of remainders: \[\sum_{k=n}^{+\infty}u_k \sim \sum_{k=n}^{+\infty}v_k\] In the divergence case, we have equivalence of partial sums: \[\sum_{k=0}^{n}u_k \sim \sum_{k=0}^{n}v_k\]

Definition - reminders

It is advisable to keep in mind all the formulations for the equivalence of two sequences: \(u_n\sim v_n\) iff \(u_n-v_n=o(v_n)\) iff for every \(\epsilon>0\) (however small), there exists an integer \(n_0\) such that \(\forall n\geqslant n_0\)\[ \begin{alignat*}{1} &\abs{u_n-v_n} \leqslant \epsilon \abs{v_n} \\ &\Leftrightarrow\quad (1-\epsilon) v_n \leqslant u_n \leqslant (1+\epsilon) v_n \qquad(\text{ si }v_n\geqslant0)\\ &\Leftrightarrow\quad 1-\epsilon \leqslant \frac{u_n}{v_n} \leqslant 1+\epsilon \\ &\Leftrightarrow\quad \abs{\frac{u_n}{v_n}-1} \leqslant \epsilon \\ \end{alignat*} \] the last two formulations being valid when \(v_n\) not 0 for \(n\) large, and equivalent to \(\frac{u_n}{v_n}\to 1\).

Of course, we can write the same inequalities by swapping the roles of \(u_n\) and \(v_n\).

Note

Since the properties in question are asymptotic, adapting the proof to the case of \(u_n,v_n\geqslant 0\) from a certain rank \(N\) is straightforward, simply by choosing indices \(n_0,n_1\) greater than \(N\).

The case \(u_n,v_n\leqslant 0\) is handled by considering \(\sum_{n\in\mathbb{N}}-u_n\) and replacing assumption i by \(v_n\leqslant u_n\).

Let \[ \begin{alignat*}{1} S_n &= \sum_{k=0}^n u_k \\ T_n &= \sum_{k=0}^n v_k \ \end{alignat*} \]
i.

Assume that the series \(\sum_{n\in\mathbb{N}}v_n\) converges. Let \(n\geqslant n_0\). \[ \begin{alignat*}{1} \sum_{k=0}^n u_k &= \sum_{k=0}^{n_0-1} u_k + \sum_{k=n_0}^n u_k \\ &\leqslant \sum_{k=0}^{n_0-1} u_k + \sum_{k=n_0}^n v_k \\ &\leqslant \sum_{k=0}^{n_0-1} u_k + \sum_{k=0}^n v_k \\ &\leqslant \sum_{k=0}^{n_0-1} u_k + \sum_{k=0}^{+\infty} v_k \\ \end{alignat*} \] The upperbound remains valid for \(n < n_0\).

This shows that the sequence of partial sums \((S_n)\) is bounded. As \(u_n\geqslant0\), the sequence \((S_n)\) is also increasing, so it converges, i.e. the series \(\sum_{n\in\mathbb{N}}u_n\) converges.

ii.

Let \(\epsilon=1\). Since \(u_n\underset{}{=}o(v_n)\), \[\exists n_0\in\mathbb{N}\quad \forall n\geqslant n_0,\quad u_n=\abs{u_n}\leqslant \epsilon\abs{v_n}=v_n\] which brings us back to the previous case.

iii.

Assume that the series \(\sum_{n\in\mathbb{N}}v_n\) converges.

Let \(\epsilon>0\). \[\exists n_0\in\mathbb{N}\quad\forall n\geqslant n_0,\quad \abs{u_n-v_n}\leqslant \epsilon v_n\] Thus, \(\forall n\geqslant n_0\), \[ \begin{alignat*}{1} \sum_{k=0}^n u_k &= \sum_{k=0}^{n_0-1} u_k + \sum_{k=n_0}^n u_k \\ &\leqslant \sum_{k=0}^{n_0-1} u_k + (1+\epsilon) \sum_{k=n_0}^n v_k\\ &\leqslant \sum_{k=0}^{n_0-1} u_k + (1+\epsilon) \sum_{k=0}^{+\infty} v_k \\ \end{alignat*} \] This remains valid for \(n < n_0\).

This shows that the sequence of partial sums \((S_n)\) is bounded from above. As \(u_n\geqslant0\), the sequence \((S_n)\) is also increasing, so it converges, i.e. the series \(\sum_{n\in\mathbb{N}}u_n\) converges.

By symmetry, we also have \[\sum_{n\in\mathbb{N}}u_n\text{ converges }\Rightarrow \sum_{n\in\mathbb{N}}v_n \text{ converges }\]

Suppose the series diverge. In this case the sequences of partial sums are not bounded and \[\lim_{n\to +\infty} S_n =\lim_{n\to +\infty} T_n = + \infty\]

When \(n\geqslant n_0\), \[ \begin{alignat*}{1} \sum_{k=0}^{n_0-1} u_k + (1-\epsilon) \sum_{k=n_0}^n v_k &\leqslant \sum_{k=0}^n u_k \leqslant \sum_{k=0}^{n_0-1} u_k + (1+\epsilon) \sum_{k=n_0}^n v_k\\ \Leftrightarrow\quad \underbrace{\sum_{k=0}^{n_0-1} u_k -(1-\epsilon) \sum_{k=0}^{n_0-1} v_k}_{=A} + (1-\epsilon) \sum_{k=0}^n v_k&\leqslant \sum_{k=0}^n u_k \leqslant \underbrace{\sum_{k=0}^{n_0-1} u_k - (1+\epsilon) \sum_{k=0}^{n_0-1} v_k}_{=B} + (1+\epsilon) \sum_{k=0}^n v_k\\ \end{alignat*} \]

The expressions \(A\) and \(B\) are constant (do not depend on \(n\)), so they are negligible compared with \(T_n\). We can therefore find an integer \(n_1\geqslant n_0\) such that \(\forall n\geqslant n_1\),\[ \begin{alignat*}{1} -\epsilon \sum_{k=0}^n v_k + (1-\epsilon) \sum_{k=0}^n v_k &\leqslant \sum_{k=0}^n u_k \leqslant \epsilon \sum_{k=0}^n v_k + (1+\epsilon) \sum_{k=0}^n v_k\\ \Leftrightarrow\quad (1-2\epsilon) \sum_{k=0}^n v_k &\leqslant \sum_{k=0}^n u_k \leqslant (1+2\epsilon) \sum_{k=0}^n v_k\\ \end{alignat*} \] Which proves \(S_n\sim T_n\).

If we now assume that the series converge, the previous reasoning is no longer valid, but for \(n\geqslant n_0\), by summing the relations between \(u_n\) and \(v_n\) up to infinity, we can write \[\begin{alignat*}{1} (1-\epsilon) \sum_{k=n}^{+\infty} v_k &\leqslant \sum_{k=n}^{+\infty} u_k \leqslant (1+\epsilon) \sum_{k=n}^{+\infty} v_k\ \end{alignat*} \] which proves that the remainders of the series are equivalent.

A numerical series is a sequence, the sequence of partial sums \((S_n)_{n\in\mathbb{N}}\).

Conversely, we tend to think less of it, but any sequence can always be written as a series: \[ u_n = \sum_{k=1}^n (u_k-u_{k-1}) + u_0 \] This series has the same nature as the sequence \((u_n)\). If \(u_n\to l\), we can also write the sequence as a remainder of a convergent series: \[ -u_n = \sum_{k=n}^{+\infty} (u_{k+1} - u_k) -l \]

In this way, we can often apply the partial sum/remainder equivalence theorem to a sequence when we're looking for an equivalent:

  1. If \(u_n\to l\), we'll try to apply the equivalence of remainders.
  2. If \(u_n \to \infty \), we'll try to apply the equivalence of partial sums


About

Welcome to Bitsflip v0.0! New contents and functionalities coming soon.

Archives

Somewhere else

  1. GitHub
  2. Twitter
  3. Superprof