2.AEF,A'DA=A'EY
A.4. Circular Filters
3. Locally Constant Functions on Zp 1. Review of General Properties1.Review of General Properties
coo
chi
`ti
Take now x _ -m in this equality:
Sf(0) - Sf(-m) = f(-m)+... + f(-I)
= f(a(m - 1))+...+f(a(0))
= S(f o a)(m ).
Since Sf (0) = 0, the result follows.
Example. Let a = 1+ t E l+ M C Cp and take
f(x)=ax =(I+t)x=Y,
k>O
Then we have
x k
Sf(x)
tk(k +
(1 +ttx - 1 (t 54 0),k>0
ax- I
Sf(x)= a-1 (a01).
3. Locally Constant Functions on Zp
A
A
3. Locally Constant Functions on Zp 179
pxooF. Give E the discrete metric. Since X is compact, f : X -> E is uniformly continuous. Hence there is a S > 0 such that
d(x, y) < S = d(f (x), f (y)) < 1 f (x) = f (y), and the conclusion follows.
The set of functions X -* E is denoted by J'(X; E), and when E = K is a field, .1(X; K) = J7(X) is a vector space over K (omitted from the notation if this field is implicit from the context). When X is a compact ultrametric space, the locally constant functions X -> K form a K-vector subspace .1'c(X) of .1(X).
The characteristic functions of clopen balls of X form a system of generators of
Pcm-
3.2.
Characteristic Functions of Balls of Zp
We are interested in locally constant functions on X = Zp taking values in any abelian group M (this abelian group will typically be an extension K of Qp). Let us start by the study of the (uniformly) locally constant functions f E .Ik(Z p; M)
satisfying
Ix-YI <IP'I=Pj
I17' f(x)=f(y)
for some fixed integer j > 0. These are the functions that are constant on all closed balls of radiusr j = 1/pd. Since the balls in question are the cosets of pj Z P in Zp, these functions are the elements of the vector space
Fj = -7:7(Z/p3Z) = J"(Zp/P'Zp) C .1t°(Zp; K).
In fact,we have a partition
Zp=
1_L(i+p3Zp)
O<i <p,
into balls of radiusrj, and
i + p-'Zp = B<p-;(i)
(0 < i < pi)
is anenumeration of these balls in Zp. For fixed j the characteristic functions lpi+j = characteristic function of the ball B<1 p, (i) (0 < i < pj )
makeup a basis of the finite-dimensional space Fj. When we let j increase, the subspacesFj also increase, and
Fk(Zp; K) = U Fj-
j>O
Unfortunately,the previously given basis of Fj has no element in common with the basisconstructed similarly in Fj_1. A clever way of constructing coherent bases
.may OOHOOHOOH
"i3 ...
of the spaces Fj - where the basis of Fj extends the basis of Fj_t - has been devised by M. van der Put. Let
cPo.o = 1 characteristic function of ZP (i = 0)
1/ri = coil characteristic function of i + pZp
(1 < i < p)
cp;,2 characteristic function of i + p2Zp
(p < i < p2),
etc.Generally,
1lri = cpi, j characteristic function of i + pJZp if pj-1 < i < pJ.
Since absolute values of elements of Zp can only be powers of p, we have
Ixl<
1Ixl< lj
i
p
(Pf-r < i < pi),
and i/ii = cpi, j is also the characteristic function of the ball
Bi={xEZp:Ix-iJ <1/i)
(with the convention B0 = ZPfor i = 0).
On the other hand, the indices i in the range pJ-1 < i < pi are precisely those that admit an expansion of length j in base p, namely an expansion of the form
i=io+i1p+...+ij-1Pt-1
(0 ieP-1,ij-1o0).
Definition. The length of an integer i > 1 is the integer v = v(i) > 1 such that the expansion of i in base p has digits ie = O for £ > v, while iv-1 ¢ 0.
With this definition, the van der Put sequence is defined by
*j = Wi,v(i) : characteristic function of i + p'(i)Zp Here are the first few functions:
1
V0,r
...
cpp-1,1VO,2
...
epp-1,2 Vp,2...
4'p2-1 2c°o,3
...
Vp-1,3 ...... ...
(Pp2,3...
Wp3-1,3cp0,j
The sequence (* )i<p; appears at the top of this triangular table of characteristic functions.
Proposition. The sequence (*j )o<i <p, is a basis of F1 (j > 0).
3. Locally Constant Functions on Zp 181
PROOF For fixed j > 0,the components of any f E Fj in the known basis ii,j (i < pj) of Fj are the (constant) values off on the balls i + piZp:
f = > f(i)cpi,j-
i<p,
In particular, for f = Ire, the characteristic function of Be = .£ + p°Zp, we have a sum of the form
where the indices that occur are the same as those occurring in the partition
BE = U (i + p3Zp)
They are the indices i such that 0 < i < pi and i - .£ (mod p)" (in order to have i E Be). These indices can be listed:
i =2, £+p", £+2p", ....
The first one is f itself, and they are all greater than or equal to £. The matrix of the components of the 1,Ire in the basis Vi,j is lower triangular with l's on its diagonal (all its entries are 0's and l's). This matrix U has determinant 1 and hence is invertible: The *t (0 < £ < pi) form a basis of Fj. If we write U
= I + N,
the matrix N is lower triangular with 0's on its diagonal and hence is nilpotent: A power of N vanishes. This proves that
U-t = I - N +
N2_ ... + (-1)mNm if
Nm+t = 0.In particular, the inverse U-1 of U has integral entries: The components of the'p,, j in the basis (ilre)are also integers.
Here is aneven more precise result.
Proposition. If f = E ai i,ri E Fj, the coefficients are given by ao = f(0) and a» = f(n) - f(n-) (n > 1),
where n_
= n -
pv-' denotes the integer of length strictly smaller than n obtained by deleting itstop digit in base p.PROOF We have already observed that f (0) = ao. Fix a positive integer n and
considerthe sum f(n) _ >i<p, a i/it (n) in which di(n) = 0 or 1. More precisely,
i/ri(n)=1nEBi
min - i mod p"(`)
the digits of n and i are the same up to v(i)
t i is an initial partial sum of n.
This shows that
f (n) = ao + (*) + an, whereas
f(n-) = ao + (*)- Hence f(n) - f(n_) = an as claimed in the proposition.
Corollary. When f = ai>Jri E Fj takes its values in an ultrametric field, we have
Il f 11 =max Jail .
PROOF For each x E ZP we have *j (x) = 0 or 1: 1*i(x)I < 1 and If (x) J = I E ai ii (x)I << max Jai 1.
This proves
Il f IJ = sup If (x) J < max Jai 1.
Conversely, ao = f (0) = Iaol
11f II, and for n > 1,Ianl = I f(n) - f(n-)I < max(If(n)I, If(n-)U < IJf1I, hence maxIan I < IIfJI-
Since (iJri)i>o is a basis of Y"(Zp, K) = Uj>o Fr, it is easy to generalize the preceding results to all locally constant functions (taking their values in an extension K of Qp).
Theorem. Let f : ZP -+ K be a locally constant function. Define
ao = f(0), a n = f(n) - f(n-) (n> 1).
Then f =
aiifi is a finite sum and 11 f II = supi Iai I.3.3. The van der Put Theorem
We are now able to give the main result, namely the representation of any contin- uous f : ZP -* K where K is a complete extension of Qp.
Theorem. Let f : ZP -+ K be a continuous function. Define 40 = f (0), an = f (n) - f (n-) (n > I)-
88w
4. Ultrametnc Banach Spaces 183 Then la, I 0, and > ai Ij converges uniformly to f. Moreover,
If II = sup Jai I = max Jai I.
i i
PROOF Since In - n_ 1 0 (n -* oo) and f is uniformly continuous, we have Ian I = If(n) - f (n _ )I - 0 (n -* oo), and the series converges uniformly. The sum of this series is a continuous function,
g=Eai'`i.
We still have to prove f = g. Since these functions are continuous, it is enough to show that their restrictions to the dense subset N are the same. The obvious equality f (0) = ao = g(0) can be used as the first step in an induction on n. Let 91 = Fi<p; aiY'i. Forn < pi we have
f(n) - f(n_) = a
(by definition)= coefficient of gj (since n < pi)
= gi(n) - gi(n-), f(n) - gi(n) _ .f(n-) - gi(n-)-
This shows that if f and g, agree on (0, 1, 2,