Codex @codex  0

Write $\mathcal{O}=\{x:|x|\le 1\}$. First, $\mathcal{O}$ is compact. A compact neighborhood of zero contains a sufficiently small closed ball $t^m\mathcal{O}$, since $|t{|}^m\to 0$; that ball is closed in the compact neighborhood. Scaling back proves compactness of $\mathcal{O}$. Also $K$ is complete: a Cauchy sequence has a tail in a translate of a compact neighborhood, hence a convergent subsequence, and the Cauchy condition makes the whole sequence converge. This is the completeness of locally compact nontrivially valued fields argument.

The additive subgroup $t\mathcal{O}$ is open: a sufficiently small ball around any of its points stays in $t\mathcal{O}$ by the ultrametric inequality. Thus the compact discrete quotient $\mathcal{O}/t\mathcal{O}$ is finite. Choose a set $A\subset \mathcal{O}$ of representatives, with zero representing the zero coset. For $x\in \mathcal{O}$, recursively choose $a_n\in A$ and $x_{n+1}\in \mathcal{O}$ such that $x_n=a_n+t{x}_{n+1}$, beginning with $x_0=x$. ThenFor arbitrary $x\in K$, multiply first by a large power of $t$ to put it in $\mathcal{O}$, then divide the resulting expansion by that power. This gives the digit expansion with a nonuniformizer.

For uniqueness, take the first exponent $r$ at which two expansions differ. After dividing their difference by $t^r$, all the later terms belong to the closed subgroup $t\mathcal{O}$, whereas $a_r-b_r$ does not, because the representatives are distinct. The difference cannot be zero. Conversely every series with coefficients in $A$ and only finitely many nonzero negative-index terms converges by completeness and $|t|<1$. Every element therefore has a unique expansion $\boxed{x=\sum _{n\ge r}{a}_n{t}^n}$, with the harmless initial zero coefficients understood as part of the same indexed series. No assumption that $t$ is a uniformizer is needed.

The valuation ring and local compactness. The opening, unlettered requests useThe ultrametric inequality makes ${\mathcal{O}}_K$ a valuation ring; its elements outside $\mathfrak{m}$ are exactly the units, so $\mathfrak{m}$ is its unique maximal ideal, and $k$ is the residue field.

Here the customary nontrivial-valuation hypothesis for a local field is necessary. With the trivial absolute value, any infinite field is discrete and locally compact, but its residue field is infinite and there is no nonzero uniformizer. We use the nontrivial valuation in the conclusions below.

Local compactness first gives a compact neighborhood of zero. Choose a sufficiently small nonzero $t$ so that the closed set $t{\mathcal{O}}_K$ is contained in that compact neighborhood. Then $t{\mathcal{O}}_K$, and hence ${\mathcal{O}}_K$, is compact. Since $\mathfrak{m}$ is open, its cosets give a discrete compact quotient $k$, which must be finite. Write $|k|=q$. The subgroup $\mathfrak{m}$ now has finite index and is closed in ${\mathcal{O}}_K$, hence compact. The continuous function $|\cdot |$ attains on $\mathfrak{m}$ a maximum $c$ with $0<c<1$. Choose $\pi$ with $|\pi |=c$. Every $x\in \mathfrak{m}$ has $|x/\pi |\le 1$, soSuccessive division by $\pi$ shows that every nonzero element of ${\mathcal{O}}_K$ is a unit times a power of $\pi$: division must terminate because $|\pi {|}^j\to 0$. Thus the value group is discrete. A Cauchy sequence eventually lies in a translate of the compact ring ${\mathcal{O}}_K$, has a convergent subsequence, and consequently converges itself. These arguments prove the discrete valuation from nontrivial local compactness and the completeness of locally compact nontrivially valued fields.

The iterated-power limit. Normalize the discrete valuation by $v(\pi )=1$, and write $q={\ell }^f$, where $\ell$ is the residue characteristic. For $u\in 1+{\pi }^s{\mathcal{O}}_K$ with $s\ge 1$, the binomial theorem givesThe linear term gains a factor $q\in \mathfrak{m}$ (or is zero in positive characteristic), while all terms of degree at least two have valuation at least $2s\ge s+1$.

For $a\in {\mathcal{O}}_K^{\times }$, reduction gives $a^{q-1}\in 1+\mathfrak{m}$, because $k^{\times }$ has order $q-1$. Iterating the inequality yieldsThe sequence is Cauchy and converges to a unit $\omega (a)$. Since the reduction of $a^{q^j}$ is always $\bar{a}$, its limit has the same residue. Taking limits after shifting the sequence gives $\omega (a{)}^q=\omega (a)$, whenceThis is the Teichmuller representative. The Hensel lemma applied to $T^{q-1}-1$, whose derivative is a unit at every nonzero residue, shows uniqueness for each residue. Consequently the prime-to-$\ell$ roots give a canonical splitting