4 Hensel’s Lemma

Theorem 4.1 (Hensel’s Lemma version 1). Assuming that:

Then there exists a unique xOK such that f(x)=0 and |xa|<|f(a)|.

Proof. Let πOK be a uniformiser and let r=v(f(a)), with v the normalised valuation (v(π)=1). We construct a sequence (xn)n=1 in OK such that:

  • (i) f(xn)0(modπn+2r)
  • (ii) xn+1xn(modπn+r)

Take x1=a: then f(x1)0(modπ1+2r).

Now we suppose we have constructed x1,,xn satisfying (i) and (ii). Define

xn+1=xnf(xn)f(xn).

Since xnx1(modπr+1), we have

v(f(xn))=v(f(xi))=r,

and hence

f(xn)f(xn)0(modπn+r)

by (i).

It follows that xn+1xn(modπn+r), so (ii) holds. Note that letting X,Y be indeterminates, we have

f(X+Y)=f0(X)+f1(X)Y+f2(X)Y2+,

where fi(X)OK[X] and f0(X)=f(X), f1(X)=f(X). Thus

f(xn+1)=f(xn)+cf(xn)+c2f2(xN)+c2f2(xn)+πn+2r+1

where c=f(xn)f(xn).

Since c0(modπn+r) and v(fi(xn))0 we have

f(xn+1)f(xn)+f(xn)c0(modπn+2r+1),

so (i) holds.

Property (ii) implies that (xn)n=1 is Cauchy, so let xOK such that xnx. Then f(x)=limnf(xn)=0 by (i).

Moreover, (ii) impies that

a=x1xn(modπr+1)nax(modπr+1)|xa|<|f(a)|

This proves existence.

Uniqueness: suppose x also satisfies f(x)=0, |xa|<|f(a)|. Set δ=xx0. Then

|xa|<|f(a)||xa|<|f(a)|,

and the ultrametric inequality implies

|δ|=|xx|<|f(a)|=|f(x)|.

But

0=f(x)=f(x+δ)=f(x)=0+f(x)δ+|||δ|2.

Hence |f(x)δ||δ|2, so |f(x)|<|δ|, a contradiction.

Corollary 4.2. Let (K,||) be a complete discretely valued field. Let f(X)OK[X] and c¯k:=OKm a simple root of f¯(X):=f(X)(modm)k[X]. Then there exists a unique xOK such that f(x)=0, xc¯(modm).

Proof. Apply Theorem 4.1 to a lift cOK of c¯. Then |f(c)|<1=|f(c)|2 since c¯ is a simple root.

Example. f(X)=X22 has a simple root modulo 7. Thus 277.

Corollary 4.3.

p×(p×)2{(2)2if p>2(2)3if p=2

Proof. Case p>2: Let bp×. Applying to f(X)=X2b, we find that b(p×)2 if and only if b(𝔽p×)2. Thus p×(p×)2𝔽p×(𝔽p×)22 (𝔽p×(p1)).

We have an isomorphism

p××(,+)p×

given by (u,n)upn. Thus

p×(p×)2(2)2.

Case p=2: Let b2×. Consider f(X)=X2b. Note f(X)=2X0(mod2). Let b1(mod8). Then

|f(1)|=23<22=|f(1)|2.

Hensel’s Lemma version 1 gives

b(2×)2b1(mod8).

Then

2×(2×)2(8)×(2)j.

Again using 2×2××, we find that 2×(2)3.

Remark. Proof uses the iteration

xn+1=xnf(xn)f(xn),

which is the non-archimedean analogue of the unewton Raphson method.

PIC

Theorem 4.4 (Hensel’s Lemma version 2). Assuming that:

  • (K,||) is a complete discretely valued field

  • f(X)OK[X]

  • f¯(X):=f(X)(modm)k[X] factorises as f¯(X)=g¯(X)h¯(X) in k[X]

  • g¯(X) and h¯(X) coprime.

Then there is a factorisation
f(X)=g(X)h(X)

in OK[X], with g¯(X)g(X)(modm), h¯(X)h(X)(modm) and degg¯=degg.

Proof. Example Sheet 1.

Corollary 4.5. Let (K,||) be a complete discretely valued field. Let

f(X)=anXn++anK[X]

with a0,an0. If f(X) is irreducible, then |ai|max(|a0|,|an|) for all i.

Proof. Upon scaling, we may assume f(X)OK[X] with maxi(|ai|)=1. Thus we need to show that max(|a0|,|an|)=1. If not, let r minimal such that |ar|=1, then 0<r<n. Thus we have

f¯(X)=Xr(ar++anXnr)(modm).

Then Theorem 4.4 implies f(X)=g(X)h(X) with 0<deg<n.