The Theory of Witt Vectors

Joseph Rabinoff
Contents
1. Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2. Denition of the Witt Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3. Proof of the Existence of the Witt Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4. The Standard Topology on the Witt Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
5. The Frobenius and Verschiebung Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
6. Almost-Universal Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
7. The Artin-Hasse Exponential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
The theory of Witt vectors, while mostly elementary, manages to package such algebraic power into
the letter W that it turns up in many areas of mathematics. And whereas the Witt rings enjoy a great
number of well-known properties, it can be difcult to nd a written statement of these properties,
much less a proof. Difcult, that is, if one does not know that almost everything one could want
to know about the Witt vectors, including the vast majority
1
of the material below, is contained in
Hazewinkels book [Haz78]. As I was unaware of this reference until B. Conrad kindly pointed it out,
I undertook to write a reference manual for the theory of Witt vectors. The results Ive included are
simply those which Ive come across in my own research; I do not by any means claim that they are
comprehensive. My contribution to the material below mainly consists of the specic statements and
their proofs;
2
none of it can be considered original work. If the reader nds an omission or error, I
would be grateful if she would email me at rabinoff@post.harvard.edu.
As the theory of Witt vectors is quite elementary, the reader need only be familiar with the funda-
mental concepts and constructions from commutative algebra and category theory to understand the
vast majority of this paper. However, experience with p-adic rings and other linearly topological rings
would be helpful.
In this paper, all rings are commutative with identity element 1 (unless stated otherwise). For a ring
K, we let Alg
K
denote the category of K-algebras. Let Ab denote the category of abelian groups.
1 Motivation
The theory of Witt vectors arises from the study of the properties of certain p-adic rings, and indeed,
provides a construction of the unramied extensions of the p-adic integers, for example. These rings
are dened as follows:
Denition 1.1. Let p be a prime number. A ring R is called a strict p-ring provided that R is complete
and Hausdorff with respect to the p-adic topology, p is not a zero-divisor in R, and the residue ring
K = R/p is perfect (i.e., the map x x
p
is bijective on K).
For our purposes, the following are the important properties of strict p-rings (cf. [Ser79, Chap-
ter II]):
1
Basically, everything except Section 1.
2
Again, as I was unaware that everything is proved in [Haz78], many of the proofs are my own, and differ from those in
[Haz78]. The presentation is otherwise similar.
1
Theorem 1.2. Let K be a perfect ring of characteristic p.
1. There is a strict p-ring R with residue ring K, which is unique up to canonical isomorphism.
2. There exists a unique system of representatives : K R, called the Teichmller represen-
tatives, such that (xy) = (x)(y) for all x, y K.
3. Every element x of R can be written uniquely in the form x =

n=0
(x
n
) p
n
for x
n
K.
4. The formation of R and is functorial in K, in that if f : K K

is a homomorphism of
perfect rings of characteristic p, and R

is the strict p-ring with residue ring K

and section

,
then there is a unique homomorphism F : R R

making the following squares commute:

R
F
R

R
F
R

K
f
K

f
K

The map F is given by

F
_

n=0
(x
n
) p
n
_
=

n=0

(f(x
n
)) p
n
.
Example 1.3. Let R be an unramied extension of Z
p
, with residue eld K = R/p

= F
q
. Then
R is a strict p-ring, and is hence the unique strict p-ring with residue eld F
q
. The Teichmller
representatives are constructed as follows: we have F

= Z/(q1)Z, so that the nonzero elements

of F
q
are the roots of the polynomial X
q1
1. By Hensels Lemma, each element x of F

q
has a
unique lift (x) R also satisfying (x)
q1
1 = 0. Setting (0) = 0 completes the denition of
the map . In other words, the Teichmller representatives are exactly the (q 1)st roots of unity
in R, union {0}.
Theorem 1.2 is an abstract fact, which may be proved without yielding a useful construction of
the strict p-ring R with given residue ring K. But the strong unicity and functoriality of R indicates
that one should be able to construct it algebraically in terms of K. We can certainly reconstruct
the set underlying R as all sums of the form

n=0
(x
n
) p
n
, where we think of (x
n
) as a parameter
depending only on x
n
K. But in order to reconstruct the ring structure on R, we need to understand
the addition and multiplication laws in terms of the arithmetic of K. Put another way, if

(x
n
) p
n
+

(y
n
) p
n
=

(s
n
) p
n
, we need to write the s
n
in terms of the x
n
and y
n
, and similarly for
multiplication. By unicity, the answer should not depend on R, and by functoriality, the answer
should not even depend on K, in that the same addition and multiplication laws will have to work
for every K. This suggests that the s
n
will be given by polynomials in the x
n
and y
n
with p-integral
rational coefcients; that is, by polynomials whose coefcients are rational numbers with nonnegative
p-adic valuation. In fact the s
n
will be given by integer polynomials in the x
n
and y
n
.
The following lemma will fundamental in proving the p-integrality of polynomials:
Lemma 1.4. Let A be a ring, and let x, y A be such that x y (mod pA). Then for all i 0 we
have x
p
i
y
p
i
(mod p
i+1
A).
Proof.
We proceed by induction on i; the case i = 0 is clear. Let i 1, and write x
p
i1
= y
p
i1
+ p
i
z for
z A. Raising both sides side to the pth power, we obtain
x
p
i
= y
p
i
+
p1

n=1
_
p
n
_
y
p
i
(pn)
p
in
z
n
+p
ip
z
p
.
The lemma follows because p divides all of the binomial coefcients, and ip i + 1.
s
2
Let Rbe a strict p-ring with residue ring K, and suppose that

(x
n
) p
n
+

(y
n
) p
n
=

(s
n
) p
n
.
To calculate the s
n
, we proceed inductively. Looking mod p, we have
(x
0
) +(y
0
) (s
0
) (mod p),
so since (x) = x (mod p), we have x
0
+y
0
= s
0
. The nave second step is to write
(x
0
) +p(x
1
) +(y
0
) +p(y
1
) (s
0
) +p(s
1
) (x
0
+y
0
) +p(s
1
) (mod p
2
),
then rewrite to nd
p(s
1
) (x
0
) +(y
0
) (x
0
+y
0
) +p((x
1
) +(y
1
)) (mod p
2
).
But whereas we know that (x
0
) +(y
0
) (x
0
+y
0
) 0 (mod p), we have no idea what its residue
mod p
2
is. The trick to calculating s
1
is as follows. Since K is perfect, every x K has a unique pth
root, written x
1/p
. Since x
0
+y
0
= s
0
, we must have x
1/p
0
+y
1/p
0
= s
1/p
0
. By Lemma 1.4 above and the
fact that commutes with multiplication, we can write
(s
0
) = (s
1/p
0
)
p
= (x
1/p
0
+y
1/p
0
)
p
((x
1/p
0
) +(y
1/p
0
))
p
(mod p
2
).
Therefore,
p(s
1
) (x
1/p
0
)
p
+(y
1/p
0
)
p
((x
1/p
0
) +(y
1/p
0
))
p
+p((x
1
) +(y
1
)) (mod p
2
).
Expanding out the above equation and dividing by p, we obtain
(s
1
) (x
1
) +(y
1
)
p1

n=1
1
p
_
p
n
_
(x
n/p
0
)(y
(pn)/p
0
) (mod p),
and therefore,
s
1
= x
1
+y
1

p1

n=1
1
p
_
p
n
_
x
n/p
0
y
(pn)/p
0
.
The above bit of formal manipulation has nothing to do with K. Indeed, let X
0
, X
1
, Y
0
, Y
1
be
indeterminates, and dene polynomials w
1
(X
0
) = X
0
and w
p
(X
0
, X
1
) = X
p
0
+ pX
1
; then solve the
polynomial equations
S
0
= w
1
(S
0
) = w
1
(X
0
) +w
1
(Y
0
) = X
0
+Y
0
S
p
0
+pS
1
= w
p
(S
0
, S
1
) = w
p
(X
0
, X
1
) +w
p
(Y
0
, Y
1
) = X
p
0
+pX
1
+Y
p
0
+pY
1
for S
0
and S
1
. This is the same bit of algebra as above, except with X
0
replacing (x
1/p
0
), etc., so we
have
S
0
= X
0
+Y
0
S
1
= X
1
+Y
1

p1

n=1
1
p
_
p
n
_
X
n
0
Y
pn
0
.
In particular, S
0
Z[X
0
, Y
0
] and S
1
Z[X
0
, X
1
, Y
0
, Y
1
]. Substituting (x
1/p
0
) back in for X
0
, etc., we
see that s
0
= S
0
(x
0
, y
0
) and s
1
= S
1
(x
1/p
0
, y
1/p
0
, x
1
, y
1
).
Witt [Wit36] realized this, and also discovered the pattern. Dene
w
p
n(X
0
, X
1
, . . . , X
n
) =
n

i=0
p
i
X
p
ni
i
= X
p
n
0
+pX
p
n1
1
+ +p
n1
X
p
n1
+p
n
X
n
.
3
Letting X
1
, Y
1
, X
2
, Y
2
, . . . be indeterminates, inductively nd S
n
that solve the polynomial equations
w
p
n(S
0
, S
1
, . . . , S
n
) = w
p
n(X
0
, X
1
, . . . , X
n
) +w
p
n(Y
0
, Y
1
, . . . , Y
n
). (1.1)
As the only term of w
p
n(S
0
, S
1
, . . . , S
n
) involving S
n
is p
n
S
n
, it is clear that there are unique poly-
nomials S
n
with rational coefcients satisfying the above identity. What Witt showed is that in fact,
S
n
Z[X
0
, Y
0
, X
1
, Y
1
, . . . , X
n
, Y
n
]. Assuming this, we have:
Theorem 1.5. Let R be a strict p-ring, let K = R/p be its residue ring, and let : K R the system
of Teichmller representatives. Suppose that

n=0
(x
n
) p
n
+

n=0
(y
n
) p
n
=

n=0
(s
n
) p
n
.
Then with the S
n
as above, we have
s
n
= S
n
_
x
1/p
n
0
, y
1/p
n
0
, x
1/p
n1
1
, y
1/p
n1
1
, . . . , x
1/p
n1
, y
1/p
n1
, x
n
, y
n
_
.
Proof.
Let n 0, and dene s
i
to be
s
i
:= S
p
i (x
1/p
i
0
, y
1/p
i
0
, x
1/p
i1
1
, y
1/p
i1
1
, . . . , x
i
, y
i
)
for i n. This is a polynomial identity in K with integer coefcients, so we may take p
ni
th
roots to obtain
s
1/p
ni
i
= S
p
i (x
1/p
n
0
, y
1/p
n
0
, x
1/p
n1
1
, y
1/p
n1
1
, . . . , x
1/p
ni
i
, y
1/p
ni
i
).
Of course this is the same as saying that
S
p
i
_
(x
1/p
n
0
), (y
1/p
n
0
), (x
1/p
n1
1
), (y
1/p
n1
1
), . . . , (x
1/p
ni
i
), (y
1/p
ni
i
)
_
( s
1/p
ni
i
) (mod p),
so by Lemma 1.4,
S
p
i
_
(x
1/p
n
0
), (y
1/p
n
0
), (x
1/p
n1
1
), (y
1/p
n1
1
), . . . , (x
1/p
ni
i
), (y
1/p
ni
i
)
_
p
ni
( s
1/p
ni
i
)
p
ni
= ( s
i
) (mod p
ni+1
).
(1.2)
Substituting (x
1/p
ni
i
) for X
i
and (y
1/p
ni
i
) for Y
i
, into (1.1), we have the identity
(x
0
)+p(x
1
) + +p
n
(x
n
) +(y
0
) +p(y
1
) + +p
n
(y
n
)
= S
1
_
(x
1/p
n
0
), (y
1/p
n
0
)
_
p
n
+pS
p
_
(x
1/p
n
0
), (y
1/p
n
0
), (x
1/p
n1
1
), (y
1/p
n1
1
)
_
p
n1
+ +p
n
S
p
n
_
(x
1/p
n
0
), (y
1/p
n
0
), (x
1/p
n1
1
), (y
1/p
n1
1
), . . . , (x
n
), (y
n
)
_
in R. Using (1.2), we may rewrite the right-hand side of the above equation as
(x
1
)+p(x
1
) + +p
n
(x
n
) +(y
1
) +p(y
1
) + +p
n
(y
n
)
( s
1
) +p( s
1
) + +p
n
( s
n
) (mod p
n+1
).
This implies that s
i
= s
i
for i n, as desired.
s
4
Witt also showed the analogous result for multiplication: namely, if we solve the polynomial equa-
tions
w
p
n(Z
0
, Z
1
, . . . , Z
n
) = w
p
n(X
0
, X
1
, . . . , X
n
)w
p
n(Y
0
, Y
1
, . . . , Y
n
)
then Z
n
Z[X
0
, Y
0
, X
1
, Y
1
, . . . , X
n
, Y
n
]. Setting
z
n
= Z
n
_
x
1/p
n
0
, y
1/p
n
0
, x
1/p
n1
1
, y
1/p
n1
1
, . . . , x
1/p
n1
, y
1/p
n1
, x
n
, y
n
_
,
the reader can verify (following the proof of Theorem 1.5) that
_

(x
n
) p
n
_

(y
n
) p
n
_
=

(z
n
) p
n
.
Hence the ring laws on R are described entirely by the polynomials S
n
and Z
n
, which were obtained
algebraically. This motivates the denition of the Witt vectors: given any ring A, we will construct a
ring W
p
(A), whose addition and multiplication laws are somehow given by the polynomials S
n
and
Z
n
. Taking A = K, we will show (Theorem 2.13) that W
p
(K) is the strict p-ring with residue ring K.
2 Denition of the Witt Rings
There are several avors of Witt rings, so for the sake of uniformity in the statements of results, we
will dene Witt rings associated to the following subsets of the natural numbers:
Denition 2.1. A subset P N = {1, 2, 3, . . .} is a divisor-stable set provided that P = , and if
n P, then all proper divisors of n are also in P. If P is a divisor-stable set, we let (P) denote
the set of prime numbers contained in P.
Remark 2.2. Let P be a divisor-stable set. We make the following observations:
(i) Since P = , we automatically have 1 P.
(ii) If n P, then all prime factors of n are contained in (P).
(iii) The multiplicatively closed subset T generated by P is simply the set of products of primes in
(P).
Example 2.3. (i) The set N is divisor-stable, as are the nite sets {1, 2, . . . , n}.
(ii) Let p be a prime number. The set P
p
= {1, p, p
2
, p
3
, . . .} is divisor-stable, as are the nite sets
P
p(n)
= {1, p, p
2
, . . . , p
n
}.
The following is key to dening the Witt rings.
Denition 2.4. Let n N. Dene the nth Witt polynomial to be
w
n
=

d|n
dX
n/d
d
Z[{X
d
: d | n}].
For any divisor-stable set P and any ring A, dene the set
W
P
(A) :=

nP
A = A
P
;
and for x W
P
(A), we write x
n
for the nth coordinate, so x = (x
n
)
nP
. If P = N we write
W(A) for W
P
(A), and if P = P
p
= {1, p, p
2
, p
3
, . . .} for a prime p, we write W
p
(A) for W
P
(A). We
consider the Witt polynomials w
n
as set-theoretic maps w
n
: W
P
(A) A for n P, and we write
w

= (w
n
)
nP
: W
P
(A) A
P
.
For x W
P
(A), the values w
n
(x) for n P are called the ghost components of x, and the coordi-
nates x
n
are the Witt components.
5
The reason we do not write W
P
(A) for the codomain as well as the domain of w

above is because
they will soon have different ring structures.
Remark 2.5. Let P be a divisor-stable set, and let A be a ring such that all elements of P have
inverses in A. Then we can solve for the Witt components of x W
P
(A) in terms of its ghost
components w
n
(x) for n P, so w

: W
P
(A) A
P
is in fact a bijection. Similarly, if no element
of P is a zero-divisor in A, then w

is an injection.
Now we can state the main theorem of this section:
Theorem 2.6. Let P be a divisor-stable set. There is a unique covariant functor W
P
: Alg
Z
Alg
Z
,
such that for any ring A,
(i) W
P
(A) =

nP
A = A
P
as sets, and for a ring homomorphism f : A B,
W
P
(f)((a
n
)
nP
) = (f(a
n
))
nP
.
(ii) The maps w
n
: W
P
(A) A are homomorphisms of rings for all n P.
The zero element of W
P
(A) is (0, 0, . . .), and the unit element is (1, 0, 0, . . .).
We will give the proof of Theorem 2.6 in the next section. We will devote the rest of this section to
some of its consequences.
Denition 2.7. With the notation as in Theorem 2.6, we call the ring W
P
(A) the ring of P-Witt
vectors, or P-Witt ring, with coefcients in A. We call W(A) the big Witt ring with coefcients in
A, and for a prime p, W
p
(A) is the p-Witt ring with coefcients in A. When confusion about P is
impossible, will simply call W
P
(A) the Witt ring or ring of Witt vectors.
Remark 2.8 (Important!). We should emphasize that the p-Witt ring W
p
(A) for a prime p is by
far the most commonly used in practice (at least in number theory). In fact, in the presence of
a xed prime number p, authors will generally write W(A) for W
p
(A), and refer to W
p
(A) as
the ring of Witt vectors over A. In this case, people generally index the Witt components and
ghost components of a Witt vector x W
p
(A) by the exponent of p: in other words, people write
x = (x
0
, x
1
, x
2
, . . .) for (x
p
0 , x
p
1 , x
p
2 , . . .) and w
n
(x) for w
p
n(x), n 0. The notation we use in
this paper is therefore nonstandard for number-theoretic applications (although it is natural in our
more general context).
Remark 2.9. 1. If A is a K-algebra, it is not true in general that W
P
(A) is a K-algebra. For
example, if A = F
p
and P = {1, p, p
2
, p
3
, . . .} then W
P
(F
p
)

= Z
p
by Theorem 2.13, which is
not an F
p
-algebra. We may still consider W
P
as a functor from Alg
K
to Alg
Z
.
2. Let R = Z[{X
n
: n P}]. Then for any ring A, Hom(R, A) is naturally identied with
W
P
(A) as sets, and hence W
P
is representable. The ring structure on W
P
(A) makes R into a
ring object in Alg
Z
.
3. When all elements of P have an inverse in A, the condition that each w
n
be a ring homomor-
phism implies by Remark 2.5 that w

: W
P
(A)

A
P
is a ring isomorphism, where A
P
has
the product ring structure. Similarly, when the elements of P are not zero-divisors in A, the
map w

makes W
P
(A) into a subring of A
P
; however, w

(W
P
(A)) = A
P
in general.
4. Witt originally thought of the rings W
p
(A) as inverse limits of the rings W
P
p(n)
(A), where
P
p(n)
= {1, p, p
2
, . . . , p
n
}. Since the elements of W
P
p(n)
(A) have nitely many components,
Witt thought of them as vectors. This is the only sense in which rings of Witt vectors are
related to vectors; really they are rings, and in fact ring-valued functors.
At this point it is convenient to explain how Theorem 2.6 relates to Section 1. Let
R = Z[{X
n
, Y
n
: n P}],
6
and let X = (X
n
)
nP
, Y = (Y
n
)
nP
W
P
(R). Let S = X + Y be the sum in the ring W
P
(R). By
denition,
w
n
(S) = w
n
(X) +w
n
(Y )
for n P, so the solutions S
n
to the polynomial equations
w
n
((S
n
)
nP
) = w
n
((X
n
)
nP
) +w
n
((Y
n
)
nP
)
are contained in Z[{X
i
, Y
i
: i P}]. A similar result holds for multiplication: namely, if Z = X Y ,
then
w
n
(Z) = w
n
(X) w
n
(Y ),
and Z
n
Z[{X
i
, Y
i
: i P}]. These polynomials in fact give the ring laws for W
P
(A) for any ring A
(and any P, for that matter), as the following corollary shows:
Corollary 2.10. Let R, P, X, Y , Z, and S be as above. The polynomials S
n
and Z
n
do not depend
on the choice of P, and in addition,
S
n
, Z
n
Z[{X
i
, Y
i
: i | n}].
Let A be an arbitrary ring, and let x, y W
P
(A). Let s
n
= S
n
evaluated at the x
i
and y
i
, and
similarly for z
n
. Then
s = (s
n
)
nP
= x +y and z = (z
n
)
nP
= x y,
where the addition and multiplication takes place in W
P
(A).
Proof.
Solving the equation w
n
(S) = w
n
(X) + w
n
(Y ) explicitly for S
n
shows that S
n
only depends on
{X
i
, Y
i
: i | n}. As the equation w
n
(S) = w
n
(X) + w
n
(Y ) does not depend on P, neither does
S. The same statements hold with Z replacing S.
Dene a ring homomorphism f : R A by f(X
i
) = x
i
and f(Y
i
) = y
i
. Since W
P
(f) is a ring
homomorphism, we have
s = W
P
(f)(S) = W
P
(f)(X) +W
P
(f)(Y ) = x +y.
By the same argument, z = x y.
s
Remark 2.11. Corollary 2.10 essentially shows that the ring laws of W
P
(A) for an arbitrary ring
A can be calculated in W
P
(R) where R = Z[{X
n
, Y
n
: n P}]. Since R is a ring which
is torsion-free as a Z-module, one can often prove statements about W
P
(R) using the injection
w

: W
P
(R) R
P
of Remark 2.9(3), and then derive facts about W
P
(A). This often-used trick
is called reduction to the universal case, and it is extremely powerful one often cares about
Witt rings over rings A of characteristic p, but in order to prove theorems about these rings, one
reduces to the characteristic-0 case. We will make extensive use of this strategy; for instance, in
Section 7 we will dene a kind of characteristic-p exponential map.
Example 2.12. Let p be a prime number, and let P = P
p
= {1, p, p
2
, p
3
, . . .}. Note that
w
p
n =
n

i=0
p
i
X
p
ni
i
= X
p
n
1
+pX
p
n1
p
+ +p
n1
X
p
p
n1
+p
n
X
p
n,
which agrees with our previous denition of w
p
n, except that we have renamed the variable X
n
to
X
p
n (cf. Remark 2.8). By Corollary 2.10, if the S
p
n are dened such that
w
p
n(S
1
, S
p
, . . . , S
p
n) = w
p
n(X
1
, X
p
, . . . , X
p
n) +w
p
n(Y
1
, Y
p
, . . . , Y
p
n)
7
for all n, then S
p
n Z[X
p
i , Y
p
i ]
i=0,1,2,...,n
, and similarly for Z
p
n and multiplication, thus verifying
a claim that we made in Section 1. In this sense, for any ring R, the ring laws of W
p
(R) are given
by the same algebra as the ring laws of a strict p-ring.
To be explicit, we calculated in Section 1 that
S
1
= X
1
+Y
1
and S
p
= X
p
+Y
p

p1

n=1
1
p
_
p
n
_
X
n
1
Y
pn
1
.
As for multiplication, we have
Z
1
= w
1
(Z
1
) = w
1
(X
1
)w
1
(Y
1
) = X
1
Y
1
and
Z
p
1
+pZ
p
= w
p
(Z
1
, Z
p
) = w
p
(X
1
, X
p
)w
p
(Y
1
, Y
p
) = (X
p
1
+pX
p
)(Y
p
1
+pY
p
)
= (X
1
Y
1
)
p
+pX
p
1
Y
p
+pX
p
Y
p
1
+p
2
X
p
Y
p
= Z
p
= X
p
1
Y
p
+X
p
Y
p
1
+pX
p
Y
p
.
Given the above example, we can show how the Witt rings offer a construction of the strict p-ring
with a given perfect residue ring of characteristic p.
Theorem 2.13. Let K be a perfect ring of characteristic p, and let R be the strict p-ring with residue
ring K, with Teichmller reprsentatives : K R. Then the map f : W
p
(K) R given by
f(x
1
, x
p
, x
p
2 , . . .) =

n=0
(x
1/p
n
p
n ) p
n
is a ring isomorphism.
Proof.
It is clear that f is a well-dened bijection, and that f(1) = (1) = 1. Let x, y W
p
(K),
and let s = x + y. We will show that f(s) = f(x) + f(y), and leave the analogous proof that
f(xy) = f(x)f(y) to the reader. By Corollary 2.10,
s
p
i = S
p
i (x
1
, y
1
, x
p
, y
p
, . . . , x
p
i , y
p
i ),
and hence, since S
p
i is a polynomial with integer coefcients,
s
1/p
i
p
i
= S
p
i
_
x
1/p
i
1
, y
1/p
i
1
, x
1/p
i
p
, y
1/p
i
p
, . . . , x
1/p
i
p
i
, y
1/p
i
p
i
_
.
Let x
j
= x
1/p
j
p
j
, and similarly for y
j
and s
j
. Substituting into the above equation, we have
s
i
= S
p
i
_
x
1/p
i
0
, y
1/p
i
0
, x
1/p
i1
1
, y
1/p
i1
1
, . . . , x
p
i , y
p
i
_
,
so by Theorem 1.5,
n

i=0
p
i
(x
1/p
i
p
i
) +
n

i=0
p
i
(y
1/p
i
p
i
)
n

i=0
p
i
(s
1/p
i
p
i
) (mod p
n+1
).
Thus f(x +y) f(x) +f(y) (mod p
n+1
) for any n, completing the proof.
s
We will give another proof of Theorem 2.13 using Frobenius and Verschiebung maps in Section 5.
8
Remark 2.14. One amazing aspect of the universal construction of W
P
(A) is that, not only can we
recover the standard generalization of the ring Z
p
= W
p
(F
p
) to any perfect residue eld K of
characteristic p (namely, the unique unramied extension of Z
p
with residue eld K), but we can
in fact dene the ring W
p
(A) for any ring A, of arbitrary characteristic. In other words, the same
ring laws used to construct Z
p
can be used to dene its analogue for an arbitrary residue ring,
although now A = W
p
(A)/ ker(w
1
) instead of W
p
(A)/p. Perhaps even more amazingly, these Witt
rings will come equipped with Frobenius and Verschiebung maps associated to numbers n P
which are not necessarily prime; cf. Section 5.
To end this section, we give an immediate corollary of Corollary 2.10:
Corollary 2.15. Let P and P

be divisor-stable sets, with P

P. The quotient map

(x
n
)
nP
(x
n
)
nP
: W
P
(A) W
P
(A)
is a ring homomorphism for any ring A, and hence denes a natural transformation W
P
W
P

of ring-valued functors.
3 Proof of the Existence of the Witt Rings
The following argument is based on an exercise in Langs Algebra [Lan84]. The exercise number
varies by edition, but can be located by looking under the Witt vectors entry in the index. We highly
recommend that the reader do this exercise (with the caveat that that Lang denes the Frobenius
endomorphism incorrectly), but for completeness we include our solution here.
Lang states that Witt told him the following proof, but that it differs from his proof in [Wit36]. The
proof below uses the same ideas as in [Haz78].
Denition 3.1. For a ring A, we let (A) be the (multiplicative) abelian group
(A) = 1 +tAt.
Lemma 3.2. Let A be a ring. Every element f = 1 +

n=1
x
n
t
n
(A) can be written in the
form f =

n=1
(1 y
n
t
n
), for unique elements y
n
A. Furthermore, there are polynomials
Y
n
Z[X
1
, . . . , X
n
] and X

n
Z[Y

1
, . . . , Y

n
], independent of A, such that y
n
= Y
n
(x
1
, . . . , x
n
) and
x
n
= X

n
(y
1
, . . . , y
n
).
Proof.
First we will prove by induction on n that there are unique y
1
, . . . , y
n
such that
f(t)/
n

i=1
(1 y
i
t
i
) = 1 +O(t
n+1
).
The case n = 0 is clear. Supposing the claim to be true for n 1, write f(t)/

n1
i=1
(1 y
i
t
i
) =
1 +zt
n
+O(t
n+1
). We calculate that for y A,
(1 +yt
n
)
1
(1 +zt
n
+O(t
n+1
)) = 1 + (z y)t
n
+O(t
n+1
),
which has zero t
n
-coefcient if and only if y = z. Hence there is a unique value y
n
= z such that
f/

n
i=1
(1 y
i
t
i
) = 1 +O(t
n+1
).
The above proof shows that there are unique y
1
, y
2
, . . . A such that
f(t) =

n=1
(1 y
n
t
n
).
9
To show that y
n
Z[X
1
, . . . , X
n
], choose A = Z[X
1
, X
2
, . . .] and f = 1+

n1
X
n
t
n
. In this case
we necessarily have Y
n
A, and it is easy to see that Y
n
only depends on the rst n + 1 terms
of f, so in fact Y
n
Z[X
1
, X
2
, . . . , X
n
]. For an arbitrary ring B, dene a ring homomorphism
A B by X
n
x
n
for some choice of x
n
A, and extend to a map (A) (B). Taking the
image of the equality

(1Y
n
t
n
) = 1+

X
n
t
n
in (B), we nd that

(1y
n
t
n
) = 1+

x
n
t
n
,
where y
n
= Y
n
(x
1
, . . . , x
n
).
It is clear that the x
n
are integer polynomials in the y
n
.
s
Corollary 3.3. For any ring A, the map x f
x
: W(A) (A) dened by
f
x
(t) =

n=1
(1 x
n
t
n
) where x = (x
1
, x
2
, x
3
, . . .)
is a bijection.
Let A be a Q-algebra. The Mercator series denes a bijection log : (A)

tAt, whose inverse is
given by the standard exponential series exp : tAt

(A). Of course the log map takes products
to sums, as this is a formal property of the Mercator series, so log is an isomorphism of abelian groups.
It is clear that f tdf/dt : tAt

tAt is also an isomorphism of abelian groups, with inverse
_
t
1
() dt. Set
D = t
d
dt
log : (A)

tAt.
Lemma 3.4. Let A be a Q-algebra, and let x W(A). Then
D(f
x
(t)) =

n=1
w
n
(x) t
n
.
Proof.
The logarithmic derivative satises the standard identity
d
dt
log f(t) =
f

(t)
f(t)
.
It is not hard to see that
log
_

n=1
(1 x
n
t
n
)
_
=

n=1
log(1 x
n
t
n
),
so that
D(f
x
(t)) = t
d
dt
log(f
x
(t)) =

n=1
nx
n
t
n
1 x
n
t
n
=

n=1
(nx
n
t
n
+nx
2
n
t
2n
+nx
3
n
t
3n
+ ).
Hence the t
n
term of D(f
x
(t)) is exactly

d|n
dx
n/d
d
= w
n
(x), proving the lemma.
s
We need one last lemma before beginning the proof of Theorem 2.6.
Lemma 3.5. Let A be a Q-algebra, and let x, y W(A). Let
f(t) =

d,eN
_
1 x
m/d
d
y
m/e
e
t
m
_
de/m
, where m = lcm(d, e).
10
Then
D(f(t)) =

n=1
w
n
(x)w
n
(y) t
n
.
Proof.
As in the proof of Lemma 3.4, we have
D(f(t)) =

d,eN
de
m
D(1 x
m/d
d
y
m/e
e
t
m
)
=

d,eN
de
x
m/d
d
y
m/e
e
t
m
1 x
m/d
d
y
m/e
e
t
m
=

d,eN
de

nN
(x
m/d
d
y
m/e
e
t
m
)
n
.
The t
n
-term of the above series is

m|n

lcm(d,e)=m
(dx
n/d
d
)(ey
n/e
e
) =
_
_

d|n
dx
n/d
d
_
_
_
_

e|n
ey
n/e
e
_
_
= w
n
(x)w
n
(y).
s
In the proof of Theorem 2.6 we will argue by reduction to the universal case, as in Remark 2.11.
Proof of Theorem 2.6.
We will show that the big Witt functor W exists. Let A be a Q-algebra. By Remark 2.5, there is a
unique ring structure on W(A) making w

: W(A)

A
N
into a ring homomorphism, where the
codomain has the product ring structure. Since w

(0) = (0, 0, . . .) and w

(1, 0, 0, . . .) = (1, 1, . . .),

the zero element of W(A) is (0, 0, . . .) and the unit element is (1, 0, 0, . . .). As this construction is
obviously functorial in A, we have proved that W exists and is unique on the category Alg
Q

Alg
Z
. We must show that the ring laws are in fact dened over the integers.
Let R = Q[X
1
, Y
1
, X
2
, Y
2
, . . .], where the X
i
and Y
i
are indeterminates. Let
X = (X
1
, X
2
, . . .), Y = (Y
1
, Y
2
, . . .) W(R).
Let S = (S
1
, S
2
, . . .) W(R) be such that f
X
(t)f
Y
(t) = f
S
(t), i.e.,

n=1
(1 X
n
t
n
)

n=1
(1 Y
n
t
n
) =

n=1
(1 S
n
t
n
).
By Lemma 3.2 we have S
n
Z[X
1
, Y
1
, X
2
, Y
2
, . . .], and by Lemma 3.4,

n=1
w
n
(S) t
n
= D(f
S
(t)) = D(f
X
(t)f
Y
(t)) = D(f
X
(t)) +D(f
Y
(t))
=

n=1
(w
n
(X) +w
n
(Y )) t
n
.
Hence w

(S) = w

(X) + w

(Y ), so S = X + Y in W(R). Now let Z = (Z

1
, Z
2
, . . .) W(R) be
such that
f
Z
(t) =

d,eN
_
1 X
m/d
d
Y
m/e
e
t
m
_
de/m
, where m = lcm(d, e).
11
Then by Lemma 3.2 we have Z
n
Z[X
1
, Y
1
, X
2
, Y
2
, . . .], and by Lemma 3.5,

n=1
w
n
(Z) t
n
= D(f
Z
(t)) =

n1
w
n
(X)w
n
(Y ) t
n
,
so w

(Z) = w

(X)w

(Y ), and hence Z = X Y in W(R).

Let A be an arbitrary ring, and let x, y W(A). Dene s = x + y by s
n
= S
n
(x, y), and
z = x y by z
n
= Z
n
(x, y). Reasoning as in Corollary 2.10, it is clear that when A is a Q-algebra,
these recover the ring laws on W(A). In any case, we have constructed well-dened addition and
multiplication maps on W(A), which are functorial in A. We have not yet shown that W(A) is a
ring when equipped with these addition and multiplication laws.
Suppose that A embeds into a Q-algebra A

, which is to say, that A is torsionfree as a Z-

module. Then the inclusion W(A) W(A

) respects addition and multiplication, i.e., W(A) is a

subring of W(A

). Hence W(A) is a ring for such A. Now let B be an arbitrary ring, and choose
a set {x
i
}
iI
of generators of B as a Z-algebra. Set A = Z[{X
i
}
iI
], and let : A B be the
surjective ring homomorphism such that (X
i
) = x
i
. Then W() : W(A) W(B) also respects
the addition and multiplication laws, which is to say that W(B) is a quotient ring of W(A). As A
is a torsionfree Z-module, W(A) is a ring, so W(B) is a ring.
This completes the construction of a functor W satisfying the properties of Theorem 2.6. The
unicity of the ring structure on W(A) is proved in the same way as the previous paragraph:
namely, we know that W(A) has only one ring structure such that w

is a ring homomorphism
when A is a Q-algebra; hence it is determined when A embeds into a Q-algebra, and therefore,
when A is the quotient of a ring embedding into a Q-algebra.
We leave it to the reader to construct the functor W
P
for an arbitrary divisor-stable set P, using
the same addition and multiplication polynomials S
n
, Z
n
above.
s
4 The Standard Topology on the Witt Rings
There is a natural inverse limit topology on the Witt rings, which is an important piece of structure,
as almost all maps between Witt rings that we will see are continuous. This topology allows one to
make sense of innite sums of Witt vectors, which will be very useful in the sequel. We will assume
that the reader is familiar with the theory of linear topological rings; cf. [Gro60, 0.7].
Notation 4.1. Let P be a divisor-stable set. For n N write
P(n) = {m P : m n}.
It is clear that P(n) is a divisor-stable set. Let
n
=
P,n
: W
P
W
P(n)
be the projection.
It is obvious from the denitions that for any P,
W
P
(A) = lim

nN
W
P(n)
(A)
as rings, under the maps
n
.
Denition 4.2. Let P be a divisor-stable set, and let Abe a ring, equipped with the discrete topology.
The standard topology on W
P
(A) is by denition the inverse limit topology on lim

W
P(n)
(A), which
is the same as the product topology on W
P
(A) = A
P
.
The standard topology has the following properties, the proofs of which are obvious.
Proposition 4.3. Let P be a divisor-stable set, and let A be a ring.
12
1. The standard topology makes W
P
(A) into a topological ring, i.e., the ring laws on W
P
(A) are
continuous.
2. The ltered set of ideals {ker(
n
) : n N} forms a neighborhood base of the identity in
W
P
(A).
3. W
P
(A) is complete and Hausdorff with respect to the standard topology.
4. The sequence x
(n)
W
P
(A) is Cauchy if and only if, for all m P,
m
(x
(n)
) is constant for
n 0; the sequence converges to y W
P
(A) if and only if, for all m P,
m
(x
(n)
) =
m
(y)
for n 0.
5. The standard topology on W
P
(A) is discrete if and only if P is nite.
All maps between Witt rings that we have dened so far are continuous:
Proposition 4.4. Let P be a divisor-stable set and let A be a ring. The following maps are continu-
ous:
1. w
n
: W
P
(A) A for n P, where A has the discrete topology.
2. w

: W
P
(A) A
P
, where A
P
has the product topology induced by the discrete topology on
A.
3. The projection W
P
(A) W
P
(A) for P

P.
4. The homomorphism W
P
(f) : W
P
(A) W
P
(B) for a ring B and a ring homomorphism
f : A B.
Proof.
1.Since ker(
n
) ker(w
n
) we have that ker(w
n
) is open.
2.The product topology is dened to be the nest topology such that a product of continuous
maps is continuous.
3.Let
P,P
: W
P
(A) W
P
(A) be the projection. Then
1
P,P
(ker(
P

,n
)) ker(
P,n
).
4.This is clear because W
P
(f) = lim

W
P(n)
(f).
s
As stated above, one of the advantages of having a topology on W
P
(A) is the convergence of innite
sums. In order to take advantage of this property, we need to make a digression on simple arithmetic
in the Witt rings.
Proposition 4.5. Let P be a divisor-stable set, let A be a ring, and let x, y W
P
(A) be Witt vectors
such that for all n P, either x
n
= 0 or y
n
= 0. Let s = x +y. Then s
n
= x
n
+y
n
, i.e.,
s
n
=
_
x
n
if x
n
= 0
y
n
if y
n
= 0.
Proof.
As we are verifying the equality of polynomial equations, it sufces to prove the Proposition
universally, i.e., we may assume that A is a Q-algebra. In this case, w

: W
P
(A)

A
P
is an
isomorphism, so we may check that w

(s) = w

(x) + w

(y), when s
n
= x
n
+ y
n
. Indeed, since
for every n P either x
n
= 0 or y
n
= 0, we have
w
n
(s) =

d|n
d(x
d
+y
d
)
n/d
=

d|n
dx
n/d
d
+

d|n
dy
n/d
d
= w
n
(x) +w
n
(y).
s
13
Denition 4.6. Let P be a divisor-stable set, let A be a ring, and let a A. We write [a] W
P
(A)
for the Witt vector whose rst component is a, and whose other components are zero:
[a] = (a, 0, 0, 0, . . .).
We call [a] the Teichmller representative for a, as the following Proposition justies:
Proposition 4.7. Let P be a divisor-stable set, let A be a ring, and let a A. For x W
P
(A), we
have
[a]x = (a
n
x
n
)
nP
.
In particular, for a, b A, [ab] = [a][b], so the map a [a] : A W
P
(A) is multiplicative.
Proof.
Let y = (a
n
x
n
)
nP
W
P
(A). As in the proof of Proposition 4.5, we will check that w
n
(y) =
w
n
([a]x) for n P. Indeed,
w
n
(y) =

d|n
d(a
d
x
d
)
n/d
= a
n

d|n
dx
n/d
d
= w
n
([a])w
n
(x) = w
n
([a]x).
s
Let P be a divisor-stable set and let A be a ring. For n P and a A, we provisionally dene
V
n
[a] to be the Witt vector whose nth Witt component is a, and whose other components are zero. Let
x W
P
(A) be a Witt vector with only nitely many nonzero components x
n
. Then it is clear from
Proposition 4.5 that
x =

nP
V
n
[x
n
].
With topological considerations, the following stronger statement holds:
Proposition 4.8. Let P be a divisor-stable set and let A be a ring. Let x W
P
(A). Then
x =

nP
V
n
[x
n
] = lim
N

nP(N)
V
n
[x
n
].
The proof is immediate. See also the treatment in [Haz78].
5 The Frobenius and Verschiebung Maps
Let R be a nite unramied extension of Z
p
, with residue eld K = F
p
n and Teichmller repre-
sentatives : K R. It is a standard fact from the theory of local elds that the quotient map
R K induces an isomorphism from the automorphism group of R over Z
p
to Gal(K/F
p
) =
{1, Frob, Frob
2
, . . . , Frob
n1
}, where Frob(x) = x
p
is the Frobenius map. In other words, there is
a canonical lift F
p
of Frob to R = W
p
(K), which one can check is given by
F
p
_

n=0
(x
n
) p
n
_
=

n=0
(x
p
n
) p
n
. (5.1)
Whereas it is surprising that one can dene a p-Witt ring W
p
(A) for a ring A of arbitrary charac-
teristic, it is perhaps more surprising that W
p
(A) always carries a canonical lift F
p
of the Frobenius
map on W
p
(A)/pW
p
(A). In fact, the big Witt ring W(A) has a commuting family of Frobenius maps
F
n
for any natural number n. These maps, along with their cousins the Verchiebung maps, are very
important pieces of structure of the Witt rings, so we devote an entire section to them.
The following will be very useful when comparing homomorphisms from Witt rings to abelian
groups. Recall that we may consider W
P
as a functor Alg
K
Alg
Z
for any ring K; this added
exibility will be useful in the proof of Theorem 5.7.
14
Denition 5.1. For a divisor-stable set P, let W
+
P
: Alg
Z
Ab denote the functor that assigns to
each ring A the additive group underlying W
P
(A).
Lemma 5.2. Let K be a ring, and let G : Alg
K
Ab be a covariant abelian-group valued functor
on Alg
K
. We assume that G is representable, in the sense that there is a K-algebra R such that
G

= Hom
K
(R, ) as set-valued functors. Let P be a divisor-stable set, and let u, v : W
+
P
G be
two natural transformations. Let A
0
= K[x] and let x
0
= [x] W
P
(A
0
) (cf. Denition 4.6). If
u
A0
(x
0
) = v
A0
(x
0
), then u = v.
Proof.
Replacing u with u v, we must show that if u
A0
(x
0
) = 0 then u = 0. Precomposing u with the
natural quotient W W
P
, we may assume that P = N; hence we will consider u as a natural
transformation Ab, such that u
A0
(1 xt) = 0. By universality, we have that u
A
(1 at) = 0
for all A Alg
K
and all a A.
First we will show by induction on the degree of f 1 +tA[t] that u
A
(f) = 0. Let f 1 +tA[t]
have degree n, and let g(t) = t
n
f(1/t), so g is monic of degree n. If g has a root a A,
then since g is monic, we can write g(t) = (t a)g
1
(t), where g
1
is monic and deg(g
1
) < n.
Hence f(t) = t
n
g(1/t) = (1 at)f
1
(t), where f
1
1 + tA[t] and deg(f
1
) < n. Since u
A
is a
homomorphism, by induction we have u
A
(f) = 0.
Now suppose that g does not have a root in A. Let A

= A[X]/(g(X)). Then A A

, and
the residue a of X in A

is a root of g A

[t]. By the above argument, u

A
(f) = 0, so since
G(A) = Hom
K
(R, A) injects into G(A

) = Hom
K
(R, A

), we must have u
A
(f) = 0.
The above proves that for any ring A and any x W(A) such that x
n
= 0 for n 0, we have
u
A
(x) = 0; indeed, f
x
(t) is a polynomial (of nite degree).
Now let A = K[X
1
, X
2
, . . .], and let X = (X
1
, X
2
, . . .) W(A). By a universality argument,
it sufces to show that u
A
(X) = 0. Let = u
A
(X) Hom
K
(R, A), and let Hom
K
(R, A)
correspond to the zero element of the abelian group G(A). Let x R, so (x) and (x) only
involve nitely many X
i
. Let A
n
= A[X
1
, X
2
, . . . , X
n
] A, and suppose that (x), (x) A
n
.
Consider the commutative square
W(A)
uA
G(A)
W(A
n
)
uAn
G(A
n
)
where the vertical maps are induced by the homomorphisms
n
: A A
n
given by setting
X
i
= 0 for i > n. Then W(
n
)(X) is a Witt vector with nitely many nonzero components, so
u
An
(W(
n
)(X)) = 0. By commutivity of the square, we have

n
= G(
n
)() = 0 = u
An
(W(
n
)(X)) = G(
n
)(u
A
(X)) =
n
,
so
n
((x)) =
n
((x)). But (x), (x) A
n
A, so
(x) =
n
((x)) =
n
((x)) = (x).
As x was arbitrary, this proves that = , so u
A
(X) = 0.
s
Example 5.3. We will primarily apply Lemma 5.2 when G = W
+
P
, which is to say, R = K[{X
n
: n
P}]; cf. Remark 2.9.
Remark 5.4. The preceding lemma is stated without proof in [Car67]. Cartier claims it is true for
any functor G : Alg
K
Ab taking injections to injections, a claim that I cannot prove.
15
Now we can move on to dening the Frobenius and Verschiebung maps. As the Verschiebung is
easier to write down, we start with it.
Theorem 5.5. Let n N, and let P be a divisor-stable set. For any ring A dene V
n
: W
+
P
(A)
W
+
P
(A) by V
n
((x
m
)
mP
) = (y
m
)
mP
, where
y
m
=
_
0 if n m
x
m/n
if n | m.
Then V
n
is a continuous homomorphism of additive topological groups, and in fact denes a
natural transformation W
+
P
W
+
P
, having the following properties:
1.
w
m
(V
n
(x)) =
_
0 if n m
n w
m/n
(x) if n | m
2. If P = N, so W
P
(A) = W(A)

= (A), then V
n
: (A) (A) is given by V
n
(f(t)) = f(t
n
).
Proof.
Clearly V
n
is a natural transformation. It is obvious from the denition that when P = N we
have V
n
(f(t)) = f(t
n
). Hence V
n
is a homomorphism of additive groups when P = N, as
(fg)(t
n
) = f(t
n
)g(t
n
). For arbitrary P the square
W(A)
Vn
W(A)
W
P
(A)
Vn
W
P
(A)
commutes, so V
n
: W
+
P
(A) W
+
P
(A) is a homomorphism of additive groups. From the above it
is clear that V
n
: W
P
(A) W
P
(A) is the inverse limit of the maps V
n
: W
P(m)
(A) W
P(m)
(A)
for m N, so V
n
is continuous.
It remains to calculate w
m
(V
n
(x)). Let x W
P
(A) and y = V
n
(x). If n m then
w
m
(V
n
(x)) =

d|m
dy
m/d
d
= 0
since n d when d | m. If m = m

n then
w
m
(V
n
(x)) =

d|m
dy
m/d
d
=

d|m

(nd)y
m/(nd)
nd
= n

d|m

dx
m

/d
d
= nw
m
(x).
s
Note that, for x A, the above denition of V
n
[x] coincides with the provisional one given in
Section 4.
Example 5.6. When P = P
p
= {1, p, p
2
, . . .} we have
V
p
(x
1
, x
p
, x
p
2 , . . .) = (0, x
1
, x
p
, x
p
2 , . . .)
and
w

(V
p
(x)) = (0, pw
1
(x), pw
p
(x), pw
p
2 (x), . . .).
16
Theorem 5.7. Let n N, and let P be a divisor-stable set such that n P = {np : p P} P.
There is a unique natural transformation of ring-valued functors F
n
: W
P
W
P
such that for
every ring A and every x W
P
(A) and m P, we have
w
m
(F
n
(x)) = w
mn
(x).
The map F
n
is continuous. When P = N, so W
P
(A) = W(A)

= (A), the map F

n
: (A) (A)
is dened by
F
n
(f)(t
n
) =
_
N(f)
_
(t),
where N = N
At/At
n

: At At
n
is the norm map.
I believe it was Cartier in [Car67] who rst realized that the Frobenius and the norm map are
related.
Proof.
First suppose that P = N. Let A be a Q-algebra. The map
(x
1
, x
2
, x
3
, . . .) (x
n
, x
2n
, x
3n
, . . .) : A
N
A
N
is a ring homomorphism, so since w

: W(A)

A
N
is an isomorphism, it is clear that there
is a unique ring homomorphism F
n
: W(A) W(A) satisfying the desired property. We claim
that, on (A), we have F
n
(f)(t
n
) = N(f)(t). We use Lemma 5.2, as applied to Alg
Q
, to reduce
the claim to proving that F
n
(f)(t
n
) = N(f)(t) when f = 1 at note that F
n
and N are both
natural transformations of abelian-group valued functors (A) (A). By denition we have
w
mn
(f) = a
mn
, so w
m
(F
n
(f)) = a
mn
. As w
m
(1 a
n
t) = a
mn
, we must have F
n
(f)(t) = 1 a
n
t,
so F
n
(f)(t
n
) = 1 a
n
t
n
. To calculate N(f)(t), we choose the basis 1, t, t
2
, . . . , t
n1
of At over
At
n
. With respect to this basis, the matrix for multiplication by 1 at is

1at
=
_

_
1 a 0 0 0 0
0 1 a 0 0 0
0 0 1 a 0 0
.
.
.
.
.
.
at
n
0 0 0 0 1
_

_
Expanding about the rst column, we calculate
N(1 at) = det(
1at
) = 1 + (1)
n
at
n
(1)
n1
a
n1
= 1 a
n
t
n
,
which proves the claim.
At this point we can dene the natural transformation F
n
: W W on Alg
Z
by the formula
F
n
(f)(t
n
) = N(f)(t), as N(f)(t) is dened for any ring A. It remains to show that F
n
: W(A)
W(A) is a ring homomorphism for all A, and that F
n
is unique. Showing that F
n
is a ring
homomorphism is a standard universal argument: one just shows that the polynomials dening
F
n
on the ring Z[X
1
, X
2
, . . .] commute with the polynomials for addition and multiplication, by
reducing to the case of a Q-algebra. As usual, unicity follows by functoriality: clearly F
n
is
determined for any ring A embedding into a Q-algebra, and hence for any quotient of such a
ring, which is to say, any ring.
We must show that F
n
: W
P
W
P
exists and is unique for general P satisfying nP P.
Directly solving the polynomial equations w
mn
(x
1
, x
2
, . . .) = w
m
(y
1
, y
2
, . . .) = w
m
(F
n
(x)) shows
that y
m
only depends on the x
i
for i | nm. Therefore, the map F
n
: W(A) W(A) descends to
the quotient W
P
(A). Unicity follows from the same universality argument.
It remains to show that F
n
is continuous. As W
P
(A) is a rst-countable topological space,
it sufces to show that F
n
is sequentially continuous. Let x
(m)
W
P
(A) converge to a point
17
x W
P
(A). This means that for any N N, we have x
(m)
i
= x
i
for i N and m 0. Since
each Witt component of F
n
(y) only depends on nitely many components of y W
P
(A), it is
clear that F
n
(x
(m)
) F
n
(x).
s
Remark 5.8. The proofs of Theorems 5.5 and 5.7 show that F
n
and V
n
commute with the quotient
maps W
P
W
P
for appropriate P

P. Lemma 5.2 implies that F

n
and V
n
are determined by
the equations
F
n
(1 at) = 1 a
n
t and V
n
(1 at) = 1 at
n
on (A) (cf. [Car67]).
As formal consequences of Theorems 5.5 and 5.7, we have the following Propositions:
Proposition 5.9. Let n, m N, and let P be a divisor-stable set. We have
V
n
V
m
= V
nm
= V
m
V
n
.
When nP P and mP P we also have
F
n
F
m
= F
nm
= F
m
F
n
.
Proof.
The statement about the Verschiebung is obvious from the denition. One way to prove the claim
about the Frobenius is to argue as in the proof of Theorem 5.7: on (A), we have
F
n
F
m
(1 at) = F
n
(1 a
m
t) = 1 a
mn
t = F
mn
(1 at).
s
Proposition 5.10. Let n N and let P be a divisor-stable set with nP P. Let A be any ring, and
let x, y W
P
(A).
1. F
n
V
n
(x) = n x.
2. V
n
(F
n
(x)y) = xV
n
(y) (i.e., V
n
is F
1
n
-linear).
3. If m is prime to n then V
m
F
n
= F
n
V
m
.
4. For m N we have (V
n
x)
m
= n
m1
V
n
(x
m
).
Proof.
As all of the above are formal identities of polynomials, a standard universality argument allows
us to assume that A is a Q-algebra. In this case, w

: W
P
(A)

A
P
is an isomorphism, so we
need only check the identities on ghost components.
1.For m P we have
w
m
(F
n
(V
n
(x))) = w
mn
(V
n
(x)) = nw
m
(x).
Hence w

F
n
V
n
(x) = w

(nx), so F
n
V
n
(x) = nx.
2.For m P we have
w
nm
(V
n
(F
n
(x)y)) = nw
m
(F
n
(x)y) = nw
nm
(x)w
m
(y)
= w
nm
(x)w
nm
(V
n
y) = w
nm
(xV
n
y)
and when m P but n m,
w
m
(V
n
(F
n
(x)y)) = 0 = w
m
(x)w
m
(V
n
y) = w
m
(xV
n
y).
Thus w

(V
n
(F
n
(x)y)) = w

(xV
n
y).
18
3.By Proposition 5.9, we may assume that n and m are prime. For r P and x A we have
w
r
(V
m
(F
n
(x))) =
_
0 if m r
mw
r/m
(F
n
(x)) = mw
nr/m
(x) if m | r.
On the other hand,
w
r
(F
n
(V
m
(x))) = w
rn
(V
m
(x)) =
_
0 if m rn
mw
rn/m
(x) if m | rn.
Since m and n are distinct primes, m | rn if and only if m | r, so the assertion follows.
4.Using (2), we calculate
(V
n
x)
m
= (V
n
x)
m1
(V
n
x) = V
n
(F
n
((V
n
x)
m1
) x)
= V
n
(F
n
(V
n
x)
m1
x) = V
n
((nx)
m1
x) = n
m1
V
n
(x
m
).
s
Remark 5.11. It is not in general true that V
n
F
n
= n. However, it is clear from Proposition 5.12
that when n = p is prime and pA = 0, then F
p
and V
p
are commuting endomorphisms of W
P
(A),
so in this case we do have V
p
F
p
= F
p
V
p
= p.
The rst part of Proposition 5.10 is evidence that for a prime number p, F
p
and V
p
deserve to be
called Frobenius and Verschiebung maps, respectively. The following Proposition demonstrates that
fact beyond a doubt.
Proposition 5.12. Let p be a prime, and let P be a divisor-stable set with pP P. Let A be a ring,
and let x W
P
(A) and y = F
p
(x). Then
1. y
m
x
p
m
(mod pA) for m P, and
2. F
p
(x) x
p
(mod pW
P
(A)).
Proof.
We immediately reduce both assertions to the case when P = N. For the rst claim, we replace
A with A/pA; we must show that when pA = 0, we have F
p
(x
1
, x
2
, . . .) = (x
p
1
, x
p
2
, . . .). Let
A = F
p
[X
1
, X
2
, . . .], where the X
i
are indeterminates; by the usual arguments, it sufces to
show that F
p
(X
1
, X
2
, . . .) = (X
p
1
, X
p
2
, . . .). Replacing W(A) with (A), we want to show that
F
p
(

(1 X
n
t
n
)) =

(1 X
p
n
t
n
). Let f =

(1 X
n
t
n
). As F
p
(f)(t
p
) = N(f)(t), where N is the
norm from At to At
p
, it sufces to show that N(f) =

(1 X
p
n
t
pn
) = f
p
. Since f
p
At
p
,
we have
f
p
2
= N(f
p
) = N(f)
p
.
But At is a ring of characteristic p with no nilpotents, so the pth power map is injective, and
hence N(f) = f
p
, as desired.
The second assertion is more delicate. Let a A, and recall that [a] = (a, 0, 0, 0, . . .) denotes the
Teichmller representative of a. By Remark 5.8 and Proposition 4.7, we see that F
p
[a] = [a
p
] =
[a]
p
. Now let n > 1. If p | n, by Propositions 5.9 and 5.10 we have (V
n
[a])
p
= n
p1
V
n
([a]
p
) 0
(mod p), and
F
p
V
n
[a] = F
p
V
p
V
n/p
[a] = pV
n/p
[a] 0 (V
n
[a])
p
(mod p).
If p n,
F
p
V
n
[a] = V
n
F
p
[a] = V
n
[a
p
] = V
n
([a]
p
),
and by Fermats little theorem,
(V
n
[a])
p
= n
p1
V
n
([a]
p
) V
n
([a]
p
) = F
p
V
n
[a] (mod p).
19
This shows that F
p
V
n
[a] (V
n
[a])
p
(mod p) for all n N and a A, i.e., F
p
(x) x
p
(mod p)
for all Witt vectors x W(A) with only one nonzero component.
Let x W(A) be an arbitrary Witt vector, and let x
(m)
=

m
i=1
V
i
[x
i
], so x
(m)
has nitely many
nonzero components, and x
(m)
x as m . By the above, we have
F
p
(x
(m)
) =
m

i=1
F
p
V
i
[x
i
]
m

i=1
(V
i
[x
i
])
p

_
m

i=1
V
i
[x
i
]
_
p
= (x
(m)
)
p
(mod p).
By continuity of F
p
and of the ring laws,
F
p
(x
(m)
) (x
(m)
)
p
F
p
(x) x
p
as m .
Assume that A is a torsionfree Z-module; we may do this since any ring is a quotient of such a
ring. Since w

is an injection, W(A) is also a torsionfree Z-module, so there is a unique element

y
(m)
W(A) such that py
(m)
= F
p
(x
(m)
) (x
(m)
)
p
. Let W
(n)
(A) = W
N(n)
(A) be the ring of
length-n Witt vectors with coefcients in A, and let
n
: W(A) W
(n)
(A) be the projection.
Again since w

is injective, W
(n)
(A) is torsionfree as a Z-module. For all n we have
p
n
(y
(m)
) =
n
(F
p
(x
(m)
) (x
(m)
)
p
),
which is constant for m 0. Since p : W
(n)
(A) W
(n)
(A) is an injection, this shows that

n
(y
(m)
) is constant for m 0, so the y
(m)
form a Cauchy sequence, and hence converge to a
y W(A) such that py = F
p
(x) x
p
.
s
Remark 5.13. Neither conclusion of Proposition 5.12 is true in general if we replace p with an
integer n that is not a prime. For example, the rst Witt component of F
6
(x) is
w
6
(x) = x
6
1
+ 3x
2
3
+ 2x
3
2
+ 6x
6
x
6
1
(mod 6).
This implies that F
6
(x) x
6
(mod 6), since otherwise,
w
6
(x) = w
1
(F
6
(x)) w
1
(x)
6
= x
6
1
(mod 6),
which we just showed is false. Similarly, the rst component of F
4
(x) is not equivalent to x
4
(mod 4), so Proposition 5.12 is even false for nontrivial prime powers.
Note that when K is a ring of characteristic p, Proposition 5.12 shows that F
p
: W
P
(K) W
P
(K)
is given by F
p
((x
n
)
nP
) = (x
p
n
)
nP
, so when K is perfect, F
p
is an automorphism. Comparing with
(5.1), we see that when R is a strict p-ring with residue eld K, then F
p
: W
p
(K)

W
p
(K) agrees
with the natural lift of Frobenius on R under the isomorphism of Theorem 2.13.
We can also re-prove Theorem 2.13 quite easily using the Frobenius and Verschiebung maps. In
addition, we obtain that the isomorphism of Theorem 2.13 is a homeomorphism with respect to the
standard topology on W
p
(K) and the p-adic topology on R.
Theorem 5.14 (re-proof of Theorem 2.13). Let K be a perfect ring of characteristic p, and let R
be the strict p-ring with residue ring K, with Teichmller reprsentatives : K R. Then the map
f : W
p
(K) R given by
f(x
1
, x
p
, x
p
2 , . . .) =

n=0
(x
1/p
n
p
n ) p
n
is an isomorphism of topological rings.
20
Proof.
First we will prove that W
p
(K) is a strict p-ring with residue ring K. For x W
p
(K) we have
px = F
p
V
p
(x) = F
p
(0, x
p
0 , x
p
1 , x
p
2 , . . .) = (0, x
p
p
0
, x
p
p
1
, x
p
p
2
, . . .). (5.2)
Since K is perfect, pW
p
(K) = V
p
(W
p
(K)) = ker(w
1
). Hence
W
p
(K)/pW
p
(K)

= W
p
(K)/ ker(w
1
)

= K.
At this point it is also easy to see that the standard topology and the p-adic topology on W
p
(K)
coincide, so that W
p
(K) is p-adically Hausdorff and complete. It is clear by (5.2) that p is not
a zero-divisor in W
p
(K). Hence W
p
(K) is a strict p-ring with residue ring K, and Teichmller
representatives t : K W
p
(K) given by t(a) = [a], as in Denition 4.6.
By Theorem 1.2, the map f : W
p
(K) R given by
f
_

nN
t(x
p
n) p
n
_
=

(x
p
n) p
n
is an isomorphism of rings. Since t(x
p
n)p
n
= (F
p
V
p
)
n
[x
p
n] = V
p
n([x
p
n
p
n]), we have that
(x
p
0 , x
p
1 , x
p
2 , . . .) =

nN
t(x
1/p
n
p
n ) p
n
which completes the proof.
s
6 Almost-Universal Properties
I am not aware of any universal property that characterizes the Witt rings W
P
(A) for arbitrary A.
However, there are some conditions under which there exist canonical maps to and from Witt rings.
Zink [Zin02] attributes the maps dened in Theorem 6.1 and Corollary 6.3 to Cartier. A slightly more
general version can be found in [Haz78].
Recall that (P) denotes the set of prime numbers in P.
Theorem 6.1. Let P be a divisor-stable set, and let A be a ring such that no element of P is a zero-
divisor in A. Suppose that A is equipped with ring endomorphisms
p
: A A for all p (P),
such that:
1.
p
(x) x
p
(mod p) for all x A, and
2.
p

q
=
q

p
for all p, q (P).
Let n P, and let n = p
e1
1
p
er
r
be its prime factorization. Dene

n
=
e1
p1

er
pr
.
Then there is a unique ring homomorphism : A W
P
(A) such that w
n
=
n
for all n P.
Proof.
Let T Z be the multiplicative subset generated by P, and let A

= T
1
A, so A

contains
A, and w

denes an isomorphism W
P
(A

)

(A

)
P
. If f : A A is a ring endomorphism
then f(T) = T, so f extends uniquely to a ring endomorphism f

: A

. In particular, the
endomorphisms
n
extend to

n
: A

. Dene

= w
1

nP

n
: A

(A

)
P
W
P
(A

),
21
so

is the unique ring homomorphism such that w

n

n
for all n P. We need only show
that

(A) W
P
(A), as if this were true then =

|
A
would satisfy the required properties.
Let x A, and let

(x) = y = (y
n
)
nP
. We will show by induction on n that y
n
A. When
n = 1, we have y
1
= w
1
(y) =
1
(x) = x A. Now let n P be arbitrary, and let n = p
e1
1
p
er
r
be the prime factorization of n. Choose a prime factor p
i
of n, and set m = n/p
i
, so by induction,

m
(y) =

d|m
dy
m/d
d
with y
d
A. Therefore,

n
(y) =
pi
(y)
m
(y) =
pi
_
_

d|m
dy
m/d
d
_
_
=

d|m
d
pi
(y
d
)
m/d
.
Let d divide m, and let p
s
be the highest power of p dividing d, so that p
eis1
is the highest
power of p dividing m/d. Using Lemma 1.4 we have

pi
(y
d
)
m/d
=
_

pi
(y
d
)
p
e
i
s1
i
_
m/(dp
e
i
s1
i
)
y
n/d
d
(mod p
eis
i
A),
and thus d
pi
(y
d
)
m/d
dy
n/d
d
(mod p
ei
A). Hence

n
(y)

d|m
dy
n/d
d

d|n,d=n
dy
n/d
d
(mod p
ei
i
A).
Since this is true for all i, the Chinese remainder theorem yields

n
(y)

d|n,d=n
dy
n/d
d
(mod nA),
and therefore
y
n
=
1
n
_
_

n
(y)

d|n,d=n
dy
n/d
d
_
_
A.
s
The statement of Corollary 6.3 is easier to comprehend with the following bit of notation, suggested
by Zink [Zin02].
Notation 6.2. Let P be a divisor-stable set, and let A be a ring. For clarity, we write
w
n
: W
P
(W
P
(A)) W
P
(A)
for the Witt polynomials,

F
n
for the Frobenius on W
P
(W
P
(A)),

V
n
for the Verschiebung, etc.
Elements of W
P
(W
P
(A)) will be denoted with a hat as well, and for x W
P
(W
P
(A)), we write
x = ( x
n
)
nP
= ( x
n,m
)
n,mP
, where ( x
n,m
)
mP
are the Witt components of x
n
.
Corollary 6.3. Let P be a divisor-stable set such that nP P for all n P. There is a unique
natural transformation of ring-valued functors on Alg
Z
: W
P
W
P
W
P
satisfying w
n
= F
n
for all n P.
We also have the identity
W
P
(w
n
) = F
n
for all n P.
22
Proof.
For any ring A, we have a commuting family {F
n
: W
P
(A) W
P
(A) | n P} of ring en-
domorphisms of W
P
(A). By Proposition 5.12, for p (P), F
p
lifts the Frobenius map on
W
P
(A)/pW
P
(A). Let A be a ring that is torsionfree as a Z-module. Since w

: W
P
(A) A
P
is injective, we have in particular that no element of P is a zero-divisor in W
P
(A). Hence by
Theorem 6.1, there is a unique map
A
: W
P
(A) W
P
(W
P
(A)) such that w
n
= F
n
for all
n P. Since w

is an injection, it is easy to see that

A
is functorial in A. Hence exists and is
unique on rings that inject into Q-algebras.
Let R = Z[{X
n
: n P}], let X = (X
n
)
nP
W
P
(R), and let

Y = (

Y
n,m
)
n,mP
=
R
(X).
Then

Y
n,m
R, i.e.,

Y
n,m
is an integer polynomial in the X
n
. For an arbitrary ring A and x
W
P
(A), dene
A
(x) = y = ( y
n,m
)
n,mP
, where y
n,m
=

Y
n,m
(x). By functoriality, this recovers
the above denition of
A
when A is a torsionfree Z-module, so by the standard arguments,
A
is a ring homomorphism for arbitrary A. As
A
is certainly functorial in A, we see that exists
and is unique.
The identity W
P
(w
n
) = F
n
is a relation of integer polynomials, so by a universality argu-
ment, it sufces to check on ghost components. Let A be a ring, let x W
P
(A), and let y = (x).
Since w
m
(F
n
(x)) = w
mn
(x), we want to show that for all m P, w
m
(W
P
(w
n
)( y)) = w
mn
(x).
Indeed, W
P
(w
n
)( y) = (w
n
( y
m
))
mP
, so
w
m
(W
P
(w
n
)( y)) =

d|m
dw
n
( y
d
)
m/d
= w
n
_
_

d|m
d y
m/d
d
_
_
= w
n
( w
m
( y)) = w
n
(F
m
x) = w
nm
(x).
s
Remark 6.4. 1. In proof of Corollary 6.3, we calculated the useful relation
w
m
W
P
(w
n
) = w
n
w
m
.
2. Let P be as in Corollary 6.3, and let P

be a divisor-stable set containing P. Then by the same

proof as Corollary 6.3, there is a unique natural transformation : W
P
W
P
W
P
such
that w
n
= F
n
for n P.
There is also a bona de universal property of the Witt rings W
p
(K), where p is prime and K is
a perfect ring of characteristic p; however, I prefer to think of it as an almost-universal property
because it only works in such limited circumstances really it is a property of strict p-rings. The
following denition can be found in [Ser79, II.5].
Denition 6.5. A p-ring is a ring R that is Hausdorff and complete for the topology dened by a
decreasing sequence a
1
a
2
of ideals such that a
n
a
m
a
n+m
, and such that the residue
ring L = R/a
1
is perfect of characteristic p.
Theorem 6.6 is a characterization of W
p
(K) as the universally repelling object in the category of
p-rings whose residue ring is a K-algebra.
Theorem 6.6. Let p be prime, and let K be a perfect ring of characteristic p. Let R be a p-ring with
residue ring L, and let f : K L be a ring homomorphism. Then there is a unique continuous
homomorphism F : W
p
(K) R making the square
W
p
(K)
F
w1
R
K
f
L
(6.1)
23
commute.
Proof.
By [Ser79], Chapter II, Proposition 8, there is a unique multiplicative system of representatives
: L R. Recall from Theorem 5.14 and its proof that every element x W
p
(K) can be written
x =

n=0
[x
1/p
n
p
n ] p
n
.
Since power series in p converge in R, we may dene F : W
p
(K) R by
F(x) =

n=0
(f(x
1/p
n
p
n )) p
n
.
It is clear that the square (6.1) commutes. The proof that F is a ring homomorphism carries
over from the proof of Theorem 2.13 with little modication, replacing equivalences modulo p
n
with equivalences modulo a
n
. (Alternatively, cf. [Ser79], Chapter II, Proposition 9). Since the
standard topology on W
p
(K) coincides with the p-adic topology, and since p a
1
, we see that F
is continuous.
Uniqueness is proved as follows. Let F

: W
p
(K) R be another homomorphism satisfying
the conclusions of the theorem. By Proposition 8 in [Ser79], an element y R is in (L) if and
only if y is a p
n
th power for all n. Since K is perfect, [a] W
p
(K) is a p
n
th power for all n, so
F

([a]) = (f(a)). Hence p

n
F

([a]) = p
n
(f(a)) for all n, so by continuity, F = F

.
s
7 The Artin-Hasse Exponential
For any prime p and any ring A there is a natural quotient map W(A) W
p
(A). It is natural to ask if
that map has a section, i.e., if there is a natural inclusion of W
p
(A) into W(A). It is too much to expect
that such a section would be a ring homomorphism, but it turns out to be true that for Z
(p)
-algebras
A, there is a natural homomorphism of abelian groups
p
: W
p
(A) W(A) splitting W(A) W
p
(A).
Zink [Zin02] attributes the map
p
to Cartier. We will dene
p
, and show how it is a kind of p-adic
analogue of an exponential map, which is interesting since is difcult to make sense of the ordinary
exponential series exp(x) =

n=0
x
n
/n! over a ring in which there exist nonzero integers that are
zero divisors. Cartier uses
p
in his theory of modules classifying p-divisible groups, in which certain
modules over the rings W
p
(A) are a kind of linearization at the origin of a p-divisible group (like
a tangent space, or a Jet space); in this philosophy,
p
is in a sense the analogue of the exponential
map Lie(G) G, where G is a Lie group. Cartier theory, as well as a more thorough treatment of the
Artin-Hasse exponential, can be found in [Haz78].
Cartiers construction rests on the following amazing power series:
Denition 7.1. The Artin-Hasse exponential power series is dened by
hexp(x) = exp
_
x +
x
p
p
+
x
p
2
p
2
+
x
p
3
p
3
+
_
Qx.
Theorem 7.2. The Artin-Hasse exponential has p-integral coefcients, i.e.,
hexp(x) Z
(p)
x,
where Z
(p)
is the ring Z localized at the prime ideal (p) = pZ.
The above theorem can be proved in many ways, including:
24
1. Dworks criterion states that a power series f 1 +xQx has p-integral coefcients if and only
if f(x
p
)/f(x)
p
Z
(p)
x and f(x
p
)/f(x)
p
1 (mod p).
2. The coefcient of x
n
in n! hexp(x) is the number of elements of the symmetric group on n letters
whose order is a power of p; the result then follows froma general (rather difcult) group theory
fact.
See [Rob00] for details. We will prove Theorem 7.2 in a different way, suggested by Wikipedia.
Lemma 7.3. We have the following identity in Qx:
hexp(x) =

nN
pn
(1 x
n
)
(n)/n
,
where is the Mbius function.
It is worth clarifying that for f 1 +xQx, we dene a fractional power of f as
f
a/b
= exp(a/b log(f)).
Proof.
Taking the logarithm of both sides, we want to show that
x
p
p
+
x
p
2
p
2
+
x
p
3
p
3
+ =

pn
(n)
n
log(1 x
n
) =

pn
(n)
n

m=1
x
mn
m
.
The x
r
-coefcient of the right-hand side of the above equation is
1
r

n|r
pn
(n).
It is clear from the above equation that the x
p
n
coefcient is 1/p
n
. Let r = p
n
s, where p s. Then
the x
r
-coefcient is r
1

n|s
(n), which we claim is equal to zero. As (n) = 0 when n is not
squarefree, we may assume s = p
1
p
m
, where the p
i
are distinct primes. Then

n|s
(n) =

I{1,2,...,m}
(1)
#I
,
where the sum is taken over all distinct subsets I of {1, 2, . . . , m}. As there are
_
m
i
_
such subsets
of size i, we have

n|s
(n) =
m

i=0
_
m
i
_
(1)
i
= (1 1)
m
= 0.
s
Lemma 7.4. Let p be prime, let f(x) 1 + xZ
(p)
x have p-integral coefcients, and let g(x)
1 + xQx be such that g(x)
n
= f(x) for an integer n not divisible by p. Then g has p-integral
coefcients as well, i.e., g(x) 1 +xZ
(p)
x.
25
Proof.
Write
f(x) = 1 +

m1
a
m
x
m
and g(x) = 1 +

m1
b
m
x
m
,
and let a
0
= b
0
= 1. We will show inductively that b
m
Z
(p)
. Clearly b
1
= a
1
/n Z
(p)
. Suppose
that b
1
, . . . , b
m1
Z
(p)
. Comparing the x
m
-coefcients of the equality f(x) = g(x)
n
, we have
a
m
=

m1++mn=m
0mi
b
m1
b
mn
= nb
m
+

m1++mn=m
0mi<m
b
m1
b
mn
,
and therefore by induction,
b
m
=
a
m
n

1
n

m1++mn=m
0mi<m
b
m1
b
mn
Z
(p)
.
s
Using Lemma 7.3 and applying Lemma 7.4 to f(x) = 1 x
n
, we obtain the proof of Theorem 7.2.
Our goal is to nd a natural section
p
: W
+
p
(A) W
+
(A) of the quotient W
+
(A) W
+
p
(A)
for Z
(p)
-algebras A, and to show that
p
resembles an exponential map. The construction follows
the general strategy for Witt vectors: namely, we will make a universal construction for Q-algebras
A (where the logarithm and exponential do make sense), then show that the polynomials in our
construction have p-integral coefcients, so that it makes sense for Z
(p)
-algebras as well.
The rst thing one might try is to dene
p
(x) to be the Witt vector y such that y
p
r = x
p
r and y
n
= 0
when n is not a pth power. Whereas
p
is certainly a section of W(A) W
p
(A), it is unfortunately not
a homomorphism of additive groups. What one really wants is w
n
(
p
(x)) = 0 for n not a pth power,
and w
p
r (
p
(x)) = w
p
r (x).
Let A be a Q-algebra. Consider the isomorphism D : (A)

tAt of Section 3. Dene
p
:
tAt tAt by

p
_
_

n1
a
n
t
n
_
_
=

r0
a
p
r t
p
r
;
i.e.,
p
forgets the non-p-power coefcients of its input. Then
p
is an endomorphismof abelian groups,
so there is a unique endomorphism of (A), which we also denote by
p
, such that D
p
=
p
D.
Note that
p

p
=
p
.
Lemma 7.5. Let A be a Q-algebra, and let f =

n1
(1 x
n
t
n
) (A). Then

p
(f) =

r1
hexp
_
x
p
r t
p
r _
.
Proof.
We calculate
D
_
_

r1
hexp(x
p
r t
p
r
)
_
_
=

r1
t
d
dt

n1
x
p
n
p
r t
p
r+n
p
n
=

r1

n1
p
r
x
p
n
p
r t
p
r+n
=

m1
t
p
m

r+n=m
p
r
x
p
n
p
r =

m1
w
p
m(x) t
p
m
=
p
(D(f)).
The Lemma follows from the denition of
p
.
s
26
Using the canonical identication W(A)

= (A), we may think of

p
as an additive endomorphism
of W(A). Armed with Theorem 7.2 and Lemma 7.5, we are in a position to prove:
Theorem 7.6. Let p be a prime, let A be a Z
(p)
-algebra, and let : W(A) W
p
(A) be the projec-
tion. Dene
p
: W
+
(A) (A)

= W
+
(A) by

p
(x
1
, x
2
, x
3
, . . .) =

r1
hexp
_
x
p
r t
p
r _
.
Then
p
is an endomorphism of abelian groups, functorial in A, satisfying:
1.
p

p
=
p
.
2. If
p
(x) = y then x
p
r = y
p
r for all r 0.
3. (
p
(x)) = (x) for all x W(A).
4. w
p
r (
p
(x)) = w
p
r (x).
Furthermore, the restriction of the canonical projection W(A) W
p
(A) induces an isomorphism

p
W
+
(A)

W
+
p
(A) of abelian groups. Therefore, the group homomorphism

p
: W
+
p
(A)

=
p
W
+
(A) W
+
(A)
is a section of the projection W
+
(A) W
+
p
(A), satisfying w
p
r (
p
(x)) = w
p
r (x).
Proof.
By Theorem 7.2,
p
is well-dened, and by Lemma 7.5 and the standard universality arguments,
we see that
p
is an additive homomorphism such that
p

p
=
p
. Next we claim that, if
y =
p
(x), then y
p
r = x
p
r for r 0. It sufces to check the claim when A is a Q-algebra. By
denition, w
p
r (y) = w
p
r (x) for all r, so since the y
p
n are determined by the w
p
r (y) = w
p
r (x), we
must have y
p
n = x
p
n, as claimed. Thus (
p
(x)) = (x), so it is obvious that
p
W
+
(A) surjects
onto W
+
p
(A). Injectivity is clear because
p
(x) only depends on {x
p
r }
r0
.
s
What Theorem 7.6 says is that, given (x
p
r )
r0
W
p
(A), there is a canonical choice of the coordi-
nates x
n
where n is not a power of p, which respects the addition law. Explicitly, if we set y
p
r = x
p
r
and y
n
= 0 when n is not a power of p, then
p
(x) =
p
(y).
Now we will try to indicate in what sense
p
is a p-adic analogue of an exponential map. Let N be
a ring without unit, such that every element of N is nilpotent. Suppose that N has the structure of
Q-algebra. Then we can think of the exponential map as a group isomorphism exp : N

(1 + N),
where 1 +N is the abelian group with the law (1 +a)(1 +b) = 1 +(a +b +ab). Now suppose that N
is a Z
(p)
-algebra. The statement analogous to Lemma 3.2 says that every element of

(N) := 1 +N[t]
can be written uniquely as a nite product

(1 a
n
t
n
). This gives an identication of the additive
group

W
+
(N) of nite-length Witt vectors with entries in N, with the multiplicative group

(N),
analogous to the identication W
+
(A)

= (A) for a ring A. The nite version of Theorem 7.6 says
that
p

W
+
(N)

W
+
p
(N). Consider the composition
E
p
:

W
+
p
(N)

=
p

W
+
(N)

W
+
(N) = 1 +tN[t]
t1
(1 +N),
given explicitly by
E
p
(x
p
0 , x
p
1 , . . . , x
p
m, 0, 0, . . .) =
m

n=0
hexp(x
p
n).
It is clear that E
p
is a homomorphism of abelian groups, and one can show that its kernel is equal to
(Id V
p
)

W
+
p
(N). This exponential map is now an isomorphism
E
p
:

W
+
p
(N)/(Id V
p
)

(1 +N),
dened for any nilpotent Z
(p)
-algebra N.
The preceding discussion is a special case of an isomorphism from Zinks theory of displays [Zin02].
27
References
[Car67] Pierre Cartier, Groupes formels associs aux anneaux de Witt gnraliss, C. R. Acad. Sci. Paris
Sr. A-B 265 (1967), A49A52. MR MR0218361 (36 #1448)
[Gro60] A. Grothendieck, lments de gomtrie algbrique. I. Le langage des schmas, Inst. Hautes
tudes Sci. Publ. Math. (1960), no. 4, 228. MR MR0217083 (36 #177a)
[Haz78] Michiel Hazewinkel, Formal groups and applications, Pure and Applied Mathematics,
vol. 78, Academic Press Inc. [Harcourt Brace Jovanovich Publishers], New York, 1978. MR
MR506881 (82a:14020)
[Lan84] Serge Lang, Algebra, second ed., Addison-Wesley Publishing Company Advanced Book Pro-
gram, Reading, MA, 1984. MR MR783636 (86j:00003)
[Rob00] Alain M. Robert, A course in p-adic analysis, Graduate Texts in Mathematics, vol. 198,
Springer-Verlag, New York, 2000. MR MR1760253 (2001g:11182)
[Ser79] Jean-Pierre Serre, Local elds, Graduate Texts in Mathematics, vol. 67, Springer-Verlag,
New York, 1979, Translated from the French by Marvin Jay Greenberg. MR MR554237
(82e:12016)
[Wit36] Ernst Witt, Zyklische krper und algebren der characteristik p vom grad p
n
. struktur diskret
bewerteter perfekter krper mit vollkommenem restklassenkrper der charakteristik p, J. Reine
Angew. Math. (1936), no. 176, 126140.
[Zin02] Thomas Zink, The display of a formal p-divisible group, Astrisque (2002), no. 278, 127248,
Cohomologies p-adiques et applications arithmtiques, I. MR MR1922825 (2004b:14083)
Copyright 2007, Joseph Rabinoff.
28