Maths > Algebraic geometry > Hensel's lemma
Hensel's lemma and algebraic functions
Posted by Martin Orr on Monday, 05 April 2010 at 22:34
An algebraic function is a function which we obtain by solving a polynomial in two variables
and
to write
as a function of
. In general polynomials have more than one root, so (informally) we get a multi-valued function. In this post I will restrict attention to regions of the plane in which we can unambiguously pick a single "branch" of the function. What happens where branches meet will be the subject of a later post.
I shall give an algorithm for expressing an algebraic function (in such a nicely behaved region) as a power series, thereby proving that a power series solution to the original polynomial exists. The generalisation of this result to a complete discrete valuation ring is Hensel's lemma, and is particularly important to number theorists in the case of p-adic integers (which were invented by Hensel). In this post I will focus on the case of algebraic functions, as it is easier to apply geometric intuition.
Algorithm for constructing power series
Suppose we have a polynomial
, which defines a curve in the affine plane.
We want to solve the equation
to obtain
as a power series in
.
(I am interested in formal solutions here, and don't care about convergence of the power series.)
Of course, in general there will be several values of
corresponding to a single
,
so there are several power series which solve the equation.
Consider a value
at which
has a simple zero (i.e.
).
Then as we move along the curve near
,
is a single-valued function of
,
so we can expect to look for a single power series solution with constant term
.
Of course it is not obvious that the function
can actually be expressed by a power series.
This will follow from giving an algorithm to construct a power series solution (ignoring convergence issues).


For example, take
.
Then
has a simple zero at
.
So there should be a unique power series
satisfying
.
To find it, we can solve for
one at a time:
-
Try
. Substitute this in
and we get a polynomial in
with no constant term. The
term is
so we set
to make this term vanish. -
Try
. Substituting this in
, we get a polynomial where the constant and
terms vanish. The
term is
so setting
, this term vanishes. -
Try
. Substituting this in
, we get a polynomial where the constant,
and
terms vanish. The
term is
so setting
, this term vanishes.

Continuing in this way, we build up the power series
The graph to the right shows that the first four terms of this power series approximate the curve
near
.
Proof that the algorithm works
Now we shall prove that the algorithm does construct a power series
satisfying
.
At each step we are setting
to be a polynomial of degree
, with unknown
coefficient and other coefficients known.
By induction, we have chosen
so that the terms of degree less than
in the expansion of
vanish.
The
term in
does not affect terms of lower degree in
, so terms of degree less than
in
vanish.
We need to show that there is a unique value for
such that the
term of
vanishes.
But this term has the form
for some constant
, so indeed the condition
ensures that there is a unique
making it vanish.
If we let
be the power series obtained as the limit of the
, then the terms of
of degree at most
are the same as the terms of
of degree at most
, so are all 0.
Since this holds for all
,
as formal power series.
(Note: it is perhaps wrong to call this an "algorithm" since it doesn't terminate. It can't terminate because it is constructing an infinite object, a power series. But in finite time it will give us any particular term of the power series.)
Hensel's lemma
The above algorithm gives a constructive proof of:
Hensel's lemma. Let
be a polynomial in
, and suppose that
is a simple root of
. Then there is a unique
(the ring of formal power series) such that
and
as formal power series.
Note that the lemma is entirely asymmetric between
and
.
We can look at it as solving a polynomial in one variable
, whose coefficients happen to be elements of the ring
.
Since the solutions end up in
, it is more natural to consider polynomials in
with coefficients in
(the resulting functions
are sometimes called algebroid functions).
Exactly the same algorithm still works.
If you know what a complete discrete valuation ring is, you can observe that the algorithm works for polynomials over any complete DVR (the function
, generalising
, being the reduction of
mod the maximal ideal).
Any DVR
is a 1-dimensional ring, so
is 2-dimensional.
You can look at Hensel's lemma as being about giving a local expression for
on the curve
, near a point
(where
is the closed point of
).
The condition that
is the condition that near this point, the curve has just one branch.