Past exam of the mathematics course of the University of Cambridge / 2010 / iii / Paper 24 / 3

Use the strong form of Hensel lemma: for $f\in {\mathbb{Z}}_p[T]$ and $a\in {\mathbb{Z}}_p$, ifthere is a unique polynomial root $b$ in the ball $v_p(b-a)>v_p(f^{\prime }(a))$. In particular, a simple polynomial root modulo $p$ lifts uniquely in its residue class. Newton iteration gives $v_p(b-a)=v_p(f(a))-v_p(f^{\prime }(a))$ when $f(a)\ne 0$.

For odd $p\notin \{5,7,13\}$, the setseach have $(p+1)/2$ elements, so they intersect. This gives a zero modulo $p$ with $z=1$. The derivative with respect to $z$ is $26$, nonzero modulo $p$, so Hensel's lemma lifts $z$ while holding $x,y$ fixed. This is the elementary proof of isotropy of nondegenerate ternary quadratic forms over finite fields followed by a simple-root lift.

At $p=5$, the residue vector $(0,1,1)$ is a zero, and the derivative $14y$ is nonzero. At $p=13$, use $(3,1,0)$, for which the value is 52 and the derivative $10x=30$ is nonzero modulo 13. Again a single-variable Hensel lift suffices.

At $p=2$, hold $x=2,z=1$ and use $f(Y)=7Y^2+33$. At $Y=1$, $f(1)=40$ has valuation 3, while $f^{\prime }(1)=14$ has valuation 1. Since $3>2$, the strong Hensel inequality gives a polynomial root $y\in {\mathbb{Z}}_2$, indeed $y\equiv 1(\mathrm{mod}4)$. This yields a nonzero solution.

At $p=7$, suppose a nonzero solution exists and scale it so that all coordinates are in ${\mathbb{Z}}_7$ and at least one is a unit. Reduction modulo 7 gives $z^2=5x^2$. Since 5 is not a square modulo 7, both $x,z$ are divisible by 7. In the original equation their terms are then divisible by 49, so $7y^2$ is divisible by 49 as well, forcing $y$ divisible by 7. This contradicts the normalization. Thus local isotropy of the five seven thirteen form gives

The intended statement uses a nontrivial Non-Archimedean absolute value. Under that convention choose $t$ with $0<|t|<1$. A compact neighborhood $C$ of zero contains an open ball of some radius $r>0$. For sufficiently large $N$, $t^N{\mathcal{O}}_K$ is contained in that ball. The valuation ring is closed, so $t^N{\mathcal{O}}_K$ is a closed subset of $C$ and is compact. Scaling shows that ${\mathcal{O}}_K$ itself is compact.

Its maximal ideal $\mathfrak{m}=\{x:|x|<1\}$ is open. The residue field $k={\mathcal{O}}_K/\mathfrak{m}$ is therefore discrete and compact, hence finite: the open cover by its singleton sets has a finite subcover. Since $\mathfrak{m}$ has finite index, its complement is a finite union of open cosets. Thus $\mathfrak{m}$ is also closed and compact.

The continuous absolute value on a field attains its maximum on $\mathfrak{m}$. Nontriviality supplies a nonzero element there, so this maximum is $\rho =|\pi |$ for some $\pi \ne 0$, with $0<\rho <1$. If $x\in \mathfrak{m}$, then $|x/\pi |\le 1$, so $\mathfrak{m}=\pi {\mathcal{O}}_K$.

For arbitrary $x\ne 0$, choose the integer $n$ for which ${\rho }^{n+1}<|x|\le {\rho }^n$. Then $\rho <|x/{\pi }^n|\le 1$. This element cannot lie in $\mathfrak{m}$, by maximality of $\rho$, so its absolute value on a field is one. Therefore $|x|={\rho }^n$. This proves discrete valuation from nontrivial local compactness:

The nontrivial valuation in the local compactness criterion is essential. If the trivial absolute value is admitted, the displayed source claim is false: take $\mathbb{Q}$ with that absolute value on a field. It is complete and discrete, hence locally compact, but its valuation ring and residue field are both $\mathbb{Q}$, which is infinite. The intended nontrivial case has been proved above.

A local field is nondiscrete and locally compact. In the non-Archimedean case its valuation ring is compact by the scaling argument in part (b). Any Cauchy sequence has a tail in a translate of a compact ball, so has a convergent subsequence; the Cauchy property forces the whole sequence to converge. Thus the field is complete. Part (b) gives a discrete valuation, a uniformizer $\pi$, and a finite residue field $k={\mathbb{F}}_q$, $q=p^f$.

We construct the coefficient field rather than presuming it exists. The polynomial $T^q-T$ has derivative $-1$ in characteristic $p$. Every residue element therefore has a unique polynomial root lift by Hensel's lemma. The set $S$ of these $q$ lifts is a field: qth powers preserve addition and multiplication, so sums and products remain polynomial root lifts; a nonzero lift has inverse $a^{q-2}$. Reduction identifies $S$ with ${\mathbb{F}}_q$. This is the finite coefficient field in positive-characteristic local fields.

For $x\in {\mathcal{O}}_K$, choose $a_0\in S$ with the same residue, write $x=a_0+\pi x_1$, and repeat with $x_1\in {\mathcal{O}}_K$. Completeness givesThe first nonzero coefficient determines the valuation, so the expansion is unique. Multiplying an arbitrary field element by a suitable power of $\pi$ reduces it to this case. Consequently substitution $T\mapsto \pi$ gives a valued-field isomorphismAddition and multiplication agree with those of Laurent series by convergence, and a nonzero series has a nonzero leading coefficient, establishing injectivity as well as surjectivity.

Conversely, ${\mathbb{F}}_q((T))$ is complete for its T-adic absolute value on a field. Its valuation ring ${\mathbb{F}}_q[[T]]$ is compact: fixing finitely many coefficients gives the finite partitions into balls, and the successive coefficient choices realize the inverse limit of the finite rings ${\mathbb{F}}_q[T]/(T^n)$. Equivalently it is the product of countably many finite discrete coefficient sets. Thus it is a non-Archimedean local field. Distinct $q$ give nonisomorphic valued fields because their residue fields have different sizes. Replacing $|T|$ by any number in $(0,1)$ gives an equivalent absolute value on a field.