PROCEEDINGS OF THE

AMERICAN MATHEMATICAL SOCIETY

Volume 140, Number 4, April 2012, Pages 12071219
S 0002-9939(2011)10989-4
Article electronically published on August 2, 2011
ON OPEN AND CLOSED MORPHISMS BETWEEN
SEMIALGEBRAIC SETS
JOS

E F. FERNANDO AND J. M. GAMBOA

(Communicated by Lev Borisov)
Abstract. In this work we study how open and closed semialgebraic maps
between two semialgebraic sets extend, via the corresponding spectral maps,
to the Zariski and maximal spectra of their respective rings of semialgebraic
and bounded semialgebraic functions.
1. Introduction
A subset M R
n
is said to be basic semialgebraic if it can be written as
M = x R
n
: f(x) = 0, g
1
(x) > 0, . . . , g
m
(x) > 0
for some polynomials f, g
1
, . . . , g
m
R[x
1
, . . . , x
n
]. The nite unions of basic semi-
algebraic sets are called semialgebraic sets. A continuous function f : M R is
said to be semialgebraic if its graph is a semialgebraic subset of R
n+1
. Usually,
semialgebraic function just means a function, not necessarily continuous, whose
graph is semialgebraic. However, since all semialgebraic functions occurring in this
article are continuous, we will omit for simplicity the continuity condition when we
refer to them.
The sum and product of functions, dened pointwise, endow the set o(M) of
semialgebraic functions on M with a natural structure of commutative ring whose
unity is the semialgebraic function with constant value 1. In fact o(M) is an R-
algebra if we identify each real number r with the constant function which just
attains this value. The most simple examples of semialgebraic functions on M are
the restrictions to M of polynomials in n variables. Other relevant ones are the
absolute value of a semialgebraic function, the maximum and the minimum of a
nite family of semialgebraic functions, the inverse and the k-root of a semialgebraic
function whenever these operations are well-dened.
It is obvious that the subset o

(M) of bounded semialgebraic functions on M

is a real subalgebra of o(M). In what follows, we denote by o

(M), indistinctly,
either o(M) or o

(M) in case the involved statements or arguments are valid for

both rings. Moreover, if p M, we will denote by m

p
the maximal ideal of all
functions in o

(M) vanishing at p.
Received by the editors July 26, 2010 and, in revised form, January 3, 2011.
2010 Mathematics Subject Classication. Primary 14P10, 54C30; Secondary 12D15, 13E99.
Key words and phrases. Semialgebraic function, semialgebraic set, Zariski spectrum, maximal
spectrum, open and closed maps, proper map, Bezoutian, quotient map.
The authors were supported by the Spanish GAAR MTM2008-00272, Proyecto Santander
Complutense PR34/07-15813 and GAAR Grupos UCM 910444.
c 2011 American Mathematical Society
1207
1208 JOS

E F. FERNANDO AND J. M. GAMBOA

Likewise, a semialgebraic map : N M between semialgebraic sets N R
n
and M R
m
is a continuous map whose coordinates are semialgebraic functions.
Given an open and closed map between two semialgebraic sets, our purpose
is to study the topological properties of its unique continuous extensions to the
Zariski and maximal spectra of their respective rings of semialgebraic and bounded
semialgebraic functions.
(1.1) Since the usual notation for spectral spaces becomes cumbersome, we replace
it by the following notation. Let M R
n
be a semialgebraic set. We denote
Spec
s
(M) = Spec(o(M));
Spec
*
s
(M) = Spec(o

(M));
Spec

s
(M) = Spec(o

(M)),

s
M = Spec
max
(o(M));

*
s
M = Spec
max
(o

(M));

s
M = Spec
max
(o

(M)).
All these spaces are endowed with the Zariski topology; see (2.1) below. Recall,
[FG4, 3.5], that the map
M
:
s
M
*
s
M, m m

, where m

is the unique
maximal ideal of o

(M) containing the prime ideal mo

(M), is a homeomorphism.
On the other hand, the map
: M Spec

s
(M), p m

p
embeds M (endowed with the Euclidean topology) into Spec

s
(M) as a dense sub-
space. In fact we identify M (M)

s
M and denote M =
*
s
M M.
(1.2) Each semialgebraic map : N M between semialgebraic sets N R
n
and M R
m
induces a ring homomorphism
,
: o

(M) o

(N), f f ,
and the spectral map associated to is Spec

s
() : Spec

s
(N) Spec

s
(M), p
(
,
)
1
(p). In fact, Spec

s
() is the unique continuous extension of with values
in Spec

s
(M), because N is dense in Spec

s
(N). Moreover, we proved in [FG2, 5.9]
that Spec
*
s
() : Spec
*
s
(N) Spec
*
s
(M) maps
*
s
N into
*
s
M; we denote
*
s
=
Spec
*
s
()[

*
s
N
:
*
s
N
*
s
M.
(1.3) On the other hand, we proved in [FG2, 4.8] the existence of semialgebraic
maps : N M and maximal ideals of o(N) whose image under the induced
map Spec
s
() : Spec
s
(N) Spec
s
(M) is not a maximal ideal of o(M). But,
o(M) being a Gelfand ring (see [FG1, 3.1(iii)] for an elementary proof), the map
s
M
: Spec
s
(M)
s
M, which maps each prime ideal of o(M) to the unique
maximal ideal of o(M) containing it, is, by [MO, 1.2], a (continuous) retraction.
We dene
s
= s
M
Spec
s
()[

s
N
:
s
N
s
M, which is a continuous map.
Note that N and M being dense in
s
N and
s
M respectively, the map
s
is the
unique continuous extension of : N M to
s
N taking values in
s
M.
(1.4) Again by [FG1, 3.1(iii)] and [MO, 1.2], there exists a retraction r
M
:
Spec
*
s
(M)
*
s
M which maps each prime ideal of o

(M) to the unique maximal

ideal of o

(M) containing it. Consider the inclusion maps i

M
:
s
M Spec
s
(M)
and j
M
:
*
s
M Spec
*
s
(M), and let k
M
: Spec
s
(M) Spec
*
s
(M), p p o

(M),
which is a homeomorphism onto its image; see [FG2, 3.2]. Moreover, s
M
=

1
M
r
M
k
M
and k
M
Spec
s
() = Spec
*
s
() k
N
. This, together with the
equality r
M
Spec
*
s
() j
N
= Spec
*
s
() j
N
, provides the following commutative
OPEN AND CLOSED MORPHISMS BETWEEN SEMIALGEBRAIC SETS 1209
diagram:
N


//

s
N

N
&&

 i
N
//
Spec
s
(N)
Spec
s
()

s
N
oo

 k
N
//
Spec
*
s
(N)
Spec
*
s
()

r
N
//

*
s
N

*
s

_
?
j
N
oo
M


//

s
M

M
88

 i
M
//
Spec
s
(M)
s
M
oo

 k
M
//
Spec
*
s
(M)
r
M
//

*
s
M
_
?
j
M
oo
Thus, via
N
and
M
, we can translate the properties of the operator
*
s
to prop-
erties of
s
. This is why we focus our attention on the study of the behaviour of

*
s
.
As one can imagine, to get relevant information about Spec
*
s
() and its restric-
tion
*
s
we must impose strong conditions to the map . Moreover, by the nature
of the used techniques, which come back to [P] (see also [Mu]), we restrict ourselves
to maps which are bounded over their bers. This is why in dealing with not
necessarily bounded semialgebraic functions, we impose to be a proper map. Our
main results in this direction are the following:
Theorem 1.5. Let : N M be an open, closed and surjective semialgebraic
map. Then both maps Spec
*
s
() : Spec
*
s
(N) Spec
*
s
(M) and
*
s
:
*
s
N
*
s
M
are open, proper and surjective.
Theorem 1.6. Let : N M be a semialgebraic map. Then the following
assertions are equivalent:
(i) The map : N M is open, proper and surjective.
(ii) The map Spec
*
s
() : Spec
*
s
(N) Spec
*
s
(M) is open, proper and surjective
and
*
s
(N) = M.
(iii) The map Spec
s
() : Spec
s
(N) Spec
s
(M) is open, proper and surjective
and the ber under of each isolated point of M is compact.
(iv) The map
*
s
:
*
s
N
*
s
M is open, proper and surjective and
*
s
(N) =
M.
A source of examples of maps to which Theorems 1.5 and 1.6 apply is the un-
ramied semialgebraic coverings with nite bers. However, other maps t such a
situation.
Examples 1.7. (i) Let N = (x
2
y
2
z
2
)(x
2
y
2
+ z
2
) = 0 R
3
, M = R
2
and be the restriction to N of the projection R
3
R
2
, (x, y, z) (x, y). One
can check that is open, proper and surjective.
(ii) Let N
1
, . . . , N
r
R
n
be a nite collection of compact semialgebraic sets and
let
i
: N
1
N
r
N
i
, (x
1
, . . . , x
r
) x
i
be the projection onto the ith-factor.
Then
i
is open, proper and surjective.
(iii) The symmetric group S
n
acts in a natural way on R
n
. The space of orbits
R
n
/S
n
admits a natural structure of ane semialgebraic space (see [B, 1.6]) and it is
homeomorphic to the semialgebraic subset M = x R
n
: Bez(x) is positive semi-
denite, where Bez denotes the quadratic form usually known as Bezoutian (see
[BCR, 6.2.7] and [PS, 0.1]). In fact, the map : R
n
R
n
, x (
1
(x), . . . ,
n
(x)),
1210 JOS

E F. FERNANDO AND J. M. GAMBOA

where
1
, . . . ,
n
are the elementary symmetric functions, induces a homeomor-
phism between R
n
/S
n
and M. In fact, the semialgebraic map : R
n
M is open,
proper and surjective.
(iv) In general, if a nite group of semialgebraic automorphisms acts on R
n
,
we get a semialgebraic quotient (see [B, 1.6]) and the canonical projection is a
semialgebraic map which is moreover open, proper and surjective.
The article is organized as follows. In Section 2 we present some preliminary
results concerning Zariski spectra of rings of semialgebraic and bounded semialge-
braic functions on a semialgebraic set that will be useful to prove Theorems 1.5 and
1.6 given in Section 3.
2. Preliminaries on Zariski and maximal spectra
We devote this section to recalling the main properties of the Zariski spectra of
rings of semialgebraic and bounded semialgebraic functions on a semialgebraic set
that we need in the sequel (see for instance [FG2, 3-6] for further details) and the
notation to be used.
(2.1) Zariski spectra of rings of semialgebraic functions. Recall that the
Zariski spectrum Spec

s
(M) of o

(M) is the collection of all prime ideals of o

(M).
This set Spec

s
(M) is endowed with the Zariski topology which has as a basis of open
sets the family of sets D
Spec

s
(M)
(f) = p Spec

s
(M) : f , p where f o

(M).
We will denote its complement by Z
Spec

s
(M)
(f) = Spec

s
(M) D
Spec

s
(M)
(f).
More generally, for each ideal a of a commutative ring with unity R, we denote
Z
Spec(R)
(a) = p Spec(R) : a p. If a = aR is a principal ideal, we write
Z
Spec(R)
(a) = Z
Spec(R)
(a). Next we recall some standard notation. If : A B is
a ring homomorphism and p Spec(A), we identify
Spec(B
p
) = q Spec(B) : q (p)B.
If is moreover injective, we write b A =
1
(b) for each ideal b in B, and aB
for the smallest ideal of B containing (a).
(2.1.1) It is well known (see [FG2, 3] for an elementary proof) that the Zariski and
the real spectrum of o

(M) coincide. Consequently, the subset

s
M of closed points
of Spec

s
(M) is, by [BCR, 7.1.25(ii)], a compact, Hausdor space which contains M
as a dense subspace; that is,

s
M is a Hausdor compactication of M. Observe
that if M is compact, then the embedding : M

s
M, p m

p
is in fact bijective
(because in this case M is dense and closed in

s
M) and so

s
M M. We denote
T

s
M
(f) = D
Spec

s
(M)
(f)

s
M
and
:

s
M
(f) =

s
M T

s
M
(f) = Z
Spec

s
(M)
(f)

s
M.
(2.1.2) As usual, given f, g o

(M), we say that f g if f(x) g(x) for all

x M; in the same vein, f is nonnegative if f 0. Moreover, the prime ideals of
the ring o

(M) satisfy a convexity condition which is ubiquitous in real geometry.

Namely, given f, g o

(M) such that g p and 0 f g, then also f p (see

[FG2, 3.1.2]).
(2.1.3) A useful consequence of the convexity is the following: The set of prime
ideals of the ring o

(M) containing a xed prime ideal p forms a chain.

OPEN AND CLOSED MORPHISMS BETWEEN SEMIALGEBRAIC SETS 1211
3. Open and closed semialgebraic morphisms
The purpose of this section is to prove Theorems 1.5 and 1.6. First, we proceed
to develop some auxiliary results to prove Theorem 1.5. To simplify the statements
we x a surjective closed and open semialgebraic map : N M and denote by

: o

(M) o

(N), f f the induced injective homomorphism. Notice

that, since is open and surjective, the image of

consists of those functions in

o

(N) which are constant on the bers of .

Since

is injective, we will write f instead of

(f) for each f o

(M),
and in this way we identify o

(M) with the subring of o

(N) consisting of those

f o

(N) which are constant on the bers of .

We consider in o

(N) the o

(M)-module structure induced by

, and for each

prime ideal p of o

(M) we denote by o

(N)
p
the localization of o

(N) at the
multiplicatively closed set o

(M) p. As usual, given functions f o

(M) and
g o

(N) we will write fg =

(f)g o

(N).
Next, we present an elementary but useful construction, which originates in [P].
Construction 3.1. Let : N M be an open, closed and surjective semialgebraic
map. Let h o

(N). Then, the functions

h
+
: M R, y suph(x) : (x) = y
and
h

: M R, y infh(x) : (x) = y
are bounded and semialgebraic. Indeed, since h

= (h)
+
, it is sucient to
study the function h
+
, which is bounded because h is. As to the continuity, and
since is open and surjective, the topology on M is the quotient topology for ,
and so it suces to prove the continuity of f = h
+
. Fix x
0
N and > 0.
Let x
1

1
((x
0
)) such that f(x
0
) /2 < h(x
1
) f(x
0
) and consider the open
intervals I = (, f(x
0
) + /2) and J = (f(x
0
) /2, +). Observe that the
open semialgebraic subsets U = h
1
(I) and V = h
1
(J) of N contain, respectively,

1
((x
0
)) and x
1
. Dene W
1
=
1
(M(NU)) and W
2
=
1
((V )). Clearly,
since is open and closed, both W
1
and W
2
are open semialgebraic subsets of N;
moreover, x
0
W = W
1
W
2
and in fact
1
((x)) U for all x W
1
. As
one can check straightforwardly, these sets fulll f(W
1
) (, f(x
0
) + /2] and
f(W
2
) (f(x
0
) /2, +). Thus, W is an open (semialgebraic) neighbourhood
of x
0
in N satisfying f(W) (f(x
0
) , f(x
0
) +), which proves the continuity of
f at x
0
.
Finally, notice that the graph of h
+
is a semialgebraic subset of M R because
both and f are semialgebraic and the supremum condition can be expressed in
the rst order language of the theory of ordered elds.
Remarks 3.2. (i) Given a function h o

(N), we have h

h h
+
.
Moreover, if g o

(N) is constant on the bers of , then g

= g = g
+
.
(ii) If the functions h, b o

(N) satisfy h b, then h

and h
+
b
+
.
(iii) For each f o

(M) we have

([f[) = [

(f)[, because both functions are

nonnegative and share the square:
(

([f[))
2
=

([f[
2
) =

(f
2
) = ([

(f)[)
2
.
Thus, the identication f =

(f) for f o

(M) is compatible with absolute

values.
1212 JOS

E F. FERNANDO AND J. M. GAMBOA

(iv) The continuity of the functions h

and h
+
fails to be true under milder
conditions on the map . Indeed, we can
(1) Consider the closed and surjective semialgebraic map : R R, t t
3
3t,
which is not open because it has a local maximum at t = 1. Dene h : R
R, x x/(1 + x
2
). The function h

: R R, y minh(x) : y = x
3
3x is
not continuous at y = 2.
Indeed, for each y R consider the polynomial P
y
(x) = x
3
3x y R[x]
whose discriminant
y
= 27(4 y
2
) vanishes at y = 2. In fact h

(2) = 1/2,
because P
2
(x) = (x + 1)
2
(x 2). On the other hand, for every > 0 one has

y
(2 + ) < 0, and so the polynomial P
2+
has a unique real root

. Thus,
h

(2 +) = h(

) =

/(1 +
2

). To estimate the value of

notice that
P
2+
(2) = < 0 while P
2+
(2 +) = (2 +)(4 +) > 0,
and so 2 <

< 2 + . Hence, lim

y2
+ h

(y) = 2/5 ,= 1/2, and the function h

is not continuous at y = 2.
(2) Consider the open and surjective semialgebraic map : R
2
R, (x, y) y
and the semialgebraic function h : R
2
R, (x, y) 1/(1 +(xy 1)
2
). Notice that
h
+
: R R, y
1
inf1 + (xy 1)
2
, x R
=

1 if y ,= 0,
1/2 if y = 0
is not continuous at y = 0.
Lemma 3.3. Let : N M be an open, closed and surjective semialgebraic map.
Let p be a prime ideal in o

(M) and let h o

(N) be a nonnegative function whose

image under the canonical homomorphism o

(N) o

(N)
p
is a unit. Then, also
the image of h

under the canonical homomorphism o

(M) o

(M)
p
is a unit.
Proof. We must prove that h

, p. By the hypothesis on h there exist g o

(N)
and f
1
, f
2
o

(M) p such that f

1
gh = f
2
. Since f
2
, p, the product f
1
g is not
identically zero. Hence, r = sup [(f
1
)(x)g(x)[ : x N > 0. The function
b = [f
1
g[/r satises 0 b(x) 1 for all x N and
H = bh =
[f
1
gh[
r
=
[f
2
[
r
o

(M).
This last equation means that H is constant on the bers of . Therefore from
the obvious inequalities 0 (H ) h on N, it follows, by Remark 3.2(ii), that
0 H

on M. Now, to prove that h

, p, it suces, by Remark (2.1.2), to

check that H

, p. But this is clear since, using 3.2(i), H

= H = [f
2
[/r , p, and
we are done.
Lemma 3.4 (Going-up). Let : N M be an open, closed and surjective semial-
gebraic map. Then, the homomorphism

: o

(M) o

(N), f f satises
the going-up property.
Proof. Let q be a prime ideal in o

(N), and let p be a prime ideal in o

(M)
containing q o

(M). Suppose, by way of contradiction, that no prime ideal of

o

(N) lying over p contains q. Via the canonical homomorphism : o

(N)
o

(N)
p
we can interpret Spec(o

(N)
p
) as a subset of Spec
*
s
(N), and the set of
prime ideals of o

(N) lying over p is Z

Spec
*
s
(N)
(po

(N)) Spec(o

(N)
p
). Hence,
OPEN AND CLOSED MORPHISMS BETWEEN SEMIALGEBRAIC SETS 1213
our assumption means that if a = q + po

(N), then
Z
Spec
*
s
(N)
(a) Spec(o

(N)
p
) = Z
Spec
*
s
(N)
(q) Z
Spec
*
s
(N)
(po

(N)) Spec(o

(N)
p
)
= .
Therefore, ao

(N)
p
= o

(N)
p
, and so there exist h q, f
1
, . . . , f
m
p, g
1
, . . . , g
m

o

(N) such that the image under of the function F = h + f

1
g
1
+ + f
m
g
m

o

(N) is a unit in o

(N)
p
. Hence, [F[ o

(N) is a nonnegative function whose

image in o

(N)
p
is a unit.
Let L be a common upper bound for all functions [g
i
[ and dene new functions

h = [h[ q and

f = L

m
i=1
[f
i
[ p. By (2.1.2), and since 0

h

h on N,
we get

h

q o

(M) p, and so

h

+

f p. Notice that,

f being constant on
the bers of , we have (

h +

f)

=

h

+

f p. On the other hand,
[F[ [h[ +
m

i=1
[g
i
[[f
i
[

h +L
m

i=1
[f
i
[ =

h +

f,
on N. This, together with the fact that [F[ is a unit in o

(N)
p
implies, by (2.1.2),
that the image of

h +

f in o

(N)
p
is a unit and, by Lemma 3.3, (

h +

f)

is also a
unit in o

(M)
p
. This contradicts the fact that (

h+

f)

p. Therefore,

satises
the going-up property.
Corollary 3.5. Let : N M be an open, closed and surjective semialgebraic
map. Let q Spec
*
s
(N) and p = q o

(M). Then, the ring homomorphism :

o

(M)
p
o

(N)
q
induced by satises the going-up property and it is injective.
Proof. First, let us check the going-up property. Indeed, let q
1
be a prime ideal in
o

(N) with q
1
q such that p
1
= q
1
o

(M) p. Consider a prime ideal p

2
in
o

(M) such that p

1
p
2
p. By Lemma 3.4, there exists q
2
Spec
*
s
(N) such that
q
1
q
2
and p
2
= q
2
o

(M). It only remains to check that q

2
Spec(o

(N)
q
),
that is, q
2
q. But since the set of prime ideals of o

(N) containing q
1
is, by
(2.1.3), a chain and q
2
o

(M) = p
2
p = q o

(M), we conclude q
2
q.
Next, let us show the injectivity of . Let F o

(M) and f o

(M) p be
such that F/f ker . Thus, gF = 0 for some g o

(N) q and, after changing g

by [g[ if necessary, we may assume that g is nonnegative. Clearly, 0 g g
+

on N; hence, by (2.1.2), g
+
, q, and so g
+
, p. Consequently, to prove the
injectivity of , it suces to check that g
+
F = 0. Otherwise, there would exist a
point y M such that g
+
(y)F(y) ,= 0; in particular, F(y) ,= 0. Therefore, for each
x
1
(y),
0 = (g(F ))(x) = g(x)F((x)) = g(x)F(y),
and so g(x) = 0. Thus,
1
(y) Z
N
(g), and this implies g
+
(y) = 0, a contradic-
tion.
Lemma 3.6 (Going-down). Let : N M be an open, closed and surjective
semialgebraic map. Then, the homomorphism

: o

(M) o

(N), f f
satises the going-down property.
Proof. Let q be a prime ideal in o

(N) and let p = q o

(M). We must prove

that the induced map : Spec(o

(N)
q
) Spec(o

(M)
p
) is surjective. It is a
closed map, by Corollary 3.5 and [AM, 5], and so its image im is a closed subset
of Spec(o

(M)
p
). Therefore, it is enough to see that im is a dense subspace of
Spec(o

(M)
p
).
1214 JOS

E F. FERNANDO AND J. M. GAMBOA

To that end we will use the injectivity of : o

(M)
p
o

(N)
q
proved in
Corollary 3.5, to show that im contains the set of minimal prime ideals of o

(M)
p
,
which is a dense subset of Spec(o

(M)
p
) because the ring o

(M)
p
has nite Krull
dimension; see [FG1, 4.1].
Indeed, to simplify notation write A = o

(M)
p
and B = o

(N)
q
. Given a
minimal prime ideal a in A, the induced homomorphism
a
: A
a
B
a
is also
injective. Whence B
a
is not zero and so it has a maximal ideal b. Thus,
1
a
(b)
is a prime ideal in A
a
which must be aA
a
(because it is the unique prime ideal of
A
a
); hence, (b) = a.
The next lemma, which will be used later, reduces the proof of Theorem 1.5
to studying the behaviour of the spectral map between the Zariski spectra. More
precisely,
Lemma 3.7. Let : N M be a semialgebraic map such that the induced map
Spec

s
() : Spec

s
(N) Spec

s
(M) is open, closed and surjective. Then,

s
:

s
N

s
M is an open, proper and surjective map.
Proof. Since Spec

s
() : Spec

s
(N) Spec

s
(M) is closed, it maps closed points to
closed points and therefore it restricts to

s
N as

s
:

s
N

s
M. Moreover,
since

s
is continuous,

s
N is a compact space and

s
M is Hausdor, it follows
that

s
is a proper map. To prove its surjectivity, let m

s
M. Since Spec

s
() :
Spec

s
(N) Spec

s
(M) is surjective, there exists a prime ideal p Spec

s
(N) with
Spec

s
()(p) = m

. Let n

be (the unique) maximal ideal of o

(N) containing p.
Then m

= Spec
*
s
()(p) Spec
*
s
()(n

) and, m

being a maximal ideal of o

(M)
and

s
= Spec

s
()[

s
N
, we get m

s
(n

).
As to the openness of

s
, let W be an open subset of

s
N, and denote by
M
:
Spec

s
(M)

s
M the retraction that maps each prime ideal of o

(M) to the unique

maximal ideal of o

(M) containing it. The equality

s
(W) = Spec

s
()(
1
M
(W))

s
M follows readily, and it proves that

s
(W) is an open subset of

s
M, because

M
is continuous and Spec

s
() is an open map.
Lemma 3.8. Let : N M be a semialgebraic map. Then, the bers of the
spectral map Spec

s
() : Spec

s
(N) Spec

s
(M) are compact.
Proof. Let p be a prime ideal of o

(M) and write K = Spec

s
()
1
(p). The closure
= Cl
Spec

s
(M)
(p)
of p is a nite and totally ordered set = p, p
1
, . . . , p
r
, with p p
1
. . . p
r
.
Now we distinguish two cases. If p is maximal, then K = Spec

s
()
1
() is a closed
subset of the compact space Spec

s
(N); hence it is compact.
If p is not maximal, let us choose a function f p
1
p. Then,
K = Spec

s
()
1
( D
Spec

s
(M)
(f)) = Spec

s
()
1
() D
Spec

s
(N)
(f )
is compact because it is a closed subset of the compact set D
Spec

s
(N)
(f ).
Now, we are ready to prove Theorem 1.5:
Proof of Theorem 1.5. First, observe that by Lemma 3.7, it is enough to prove
that Spec
*
s
() : Spec
*
s
(N) Spec
*
s
(M) is open, proper and surjective. Indeed, the
closedness of Spec
*
s
() is a straightforward consequence of the going-up property
(see [AM, 5]) proved in Lemma 3.4, while its surjectivity follows because the image
OPEN AND CLOSED MORPHISMS BETWEEN SEMIALGEBRAIC SETS 1215
of Spec
*
s
() is closed and it contains the dense subset (N) = M m

y
: y M
of Spec
*
s
(M). Moreover, by Lemma 3.8, the bers of Spec
*
s
() are compact. Thus,
it just remains to check that Spec
*
s
() is an open map. For that, it is enough to
check the equality
() D
Spec
*
s
(M)
(h
+
) = Spec
*
s
()(D
Spec
*
s
(N)
(h))
for every nonnegative function h o

(N), because D
Spec
*
s
(N)
([h[) = D
Spec
*
s
(N)
(h).
Notice that 0 h h
+
, which implies, by (2.1.2), that h
+
, Spec
*
s
()(q)
whenever h / q. Conversely, let p , Spec
*
s
()(D
Spec
*
s
(N)
(h)) be a prime ideal of
o

(M) and let us prove that h

+
p. Indeed, denote a = po

(N) and let us show

rst that h

a =

q
q where q Z
Spec
*
s
(N)
(a).
Otherwise, there would exist a prime ideal q
1
in o

(N) containing a such that

h , q
1
. The prime ideal p
1
= q
1
o

(M) contains p and, by the going-down

property of Lemma 3.6, there exists a prime ideal q in o

(N) lying over p and

contained in q
1
. Consequently, h , q (Spec
*
s
())
1
(p) Z
Spec
*
s
(N)
(h), because
p , Spec
*
s
()(D
Spec
*
s
(N)
(h)), a contradiction.
Hence, some power of h occurs in a; that is, h

d
i=1
f
i
g
i
for some 1,
f
i
p and g
i
o

(N). Let F
i
= [f
i
[ p and G
i
= [g
i
[ o

(N); hence,
h

d
i=1
(F
i
)G
i
on N and so 0 (h

)
+

d
i=1
(G
i
)
+
F
i
on M. By (2.1.2),
this implies, since

d
i=1
G
i,+
F
i
p, that also h

+
p, that is, h
+
p, as wanted.
Our next goal is to approach Theorem 1.6. Before that we need some prelimi-
naries:
Remarks 3.9. (i) To study the map Spec
s
() : Spec
s
(N) Spec
s
(M) induced by
a semialgebraic map : N M by using similar arguments to those used for the
spectra of rings of bounded semialgebraic functions, it is natural to impose to
be a proper and surjective map. Under this assumption its bers are nonempty
compact sets and so each function h o(N) induces two functions:
h
+
: M R, y maxh(x) : (x) = y
and
h

: M R, y minh(x) : (x) = y.
Moreover, an analogous proof to the one of Construction 3.1 shows that, if is also
an open map, then both h
+
and h

are semialgebraic functions; that is, they are

continuous with semialgebraic graphs.
(ii) On the other hand, as is well known, Spec
s
(N) and Spec
s
(M) are homeo-
morphic to the subsets S(N) and S(M) of Spec
*
s
(N) and Spec
*
s
(M), respectively,
dened by
S(N) = p Spec
*
s
(N) : p J(N) =
and
S(M) = q Spec
*
s
(M) : q J(M) = ,
1216 JOS

E F. FERNANDO AND J. M. GAMBOA

where J(N) = f o

(N) : Z
N
(f) = and J(M) = g o

(M) : Z
M
(g) =
(see for instance [FG2, 3.2]). After identifying these spaces, we get a commu-
tative diagram whose horizontal arrows are embeddings:
Spec
s
(N) S(N)


//
Spec
s
()

Spec
*
s
(N)
Spec
*
s
()

Spec
s
(M) S(M)


//
Spec
*
s
(M)
(iii) Again, the homomorphism

: o(M) o(N), f f endows o(N)

with a structure of o(M)-module, and in fact o(N) is the ring of fractions of o

(N)
with respect to J(M). To prove this, it suces to write each function f o(N)
as a quotient f = g/h where g = f/(1+f
2
+
) o

(N) and h = 1/(1+f

2
+
) J(M).
In other words, the canonical homomorphism o

(N)
W(M)
o

(N)
W(N)
= o(N)
is an isomorphism.
(iv) Hence, S(N) = p Spec o

(N) : p J(M) = , and so

Spec
*
s
()(Spec
*
s
(N) Spec
s
(N)) = Spec
*
s
(M) Spec
s
(M).
Therefore, Spec
s
(N) = Spec
*
s
()
1
(Spec
s
(M)).
For the sake of the reader we state next a useful auxiliary result whose proof can
be found in [Fe2, 3.9].
Lemma 3.10. Let : N M be a surjective semialgebraic map. Then, is
proper if and only if
*
s
(N) = M.
We are now in a position to prove Theorem 1.6. Namely,
Proof of Theorem 1.6. We will prove the chain of implications (i) = (ii) = (iii)
= (iv) = (i). Observe rst that (i) = (ii) follows from Theorem 1.5 and
Lemma 3.10.
(ii) = (iii) By Remark 3.9(iv), we have a commutative diagram
Spec
s
(N) = Spec
*
s
()
1
(Spec
s
(M))


//
Spec
s
()

Spec
*
s
(N)
Spec
*
s
()

Spec
s
(M)


//
Spec
*
s
(M)
whose horizontal arrows are embeddings. Thus, since Spec
*
s
() is open, closed and
surjective, so is Spec
s
(). Moreover, by Lemma 3.8, the bers of Spec
s
() are
compact, and so Spec
s
() : Spec
s
(N) Spec
s
(M) is also a proper map. Next, the
equality
*
s
(N) = M implies that
1
(p) = (
*
s
)
1
(m

p
), and this last set is a
compact subset of N for all p M, in particular, if p is an isolated point of M.
(iii) = (iv). It follows from Lemma 3.7 that
s
:
s
N
s
M is an open,
proper and surjective map. Thus, from (1.1)(1.4), the same holds true for the map

*
s
:
*
s
N
*
s
M, and it only remains to check that
*
s
(N) = M. This last task
will be done in several steps. In any case, we may assume from the beginning that
N R
n
is bounded, after changing N by its inverse image under the semialgebraic
homeomorphism
B
n
(0, 1) R
n
, x
x

1 |x|
2
OPEN AND CLOSED MORPHISMS BETWEEN SEMIALGEBRAIC SETS 1217
between the open ball B
n
(0, 1) R
n
of center the origin and radius 1 and R
n
.
(1.6.1) The ber
1
(p) is nowhere dense in N for each nonisolated point p of M.
Suppose that there exist a nonisolated point p M and a nonempty semialge-
braic subset W
1
(p) which is, moreover, open in N. Let us check rst that
V = WCl

*
s
N
(N) is a nonempty open subset of
*
s
N. Indeed, let N
lc
be the largest
locally compact and dense subset of N, and let
1
(N) = N N
lc
= Cl
R
n(Cl
R
n(N)
N) N (see [Fe1, 3.8]). By [Fe1, 6.8.1], Cl

*
s
N
(N) =
*
s
N N
lc
. Notice that by its
denition
1
(N) W =
1
(W); hence W
lc
= W
1
(W) = W
1
(N) ,= . Thus,
V = W (
*
s
N N
lc
) = W (N N
lc
) = W
1
(N) ,= .
Moreover, V N
lc
N, and so V is an open subset of N
lc
; hence V is an open
subset of
*
s
N, because the locally compact semialgebraic set N
lc
is open in its
Hausdor compactication
*
s
N. Now,
*
s
:
*
s
N
*
s
M being an open map,

*
s
(V ) = (V ) = p is an open subset of M, which contradicts the fact that p is
nonisolated in M.
(1.6.2) The ber
1
(p) is a compact set for all p M.
Fix a point p M; since by hypothesis
1
(p) is compact for each isolated
point p of M, we may assume that p is a nonisolated point. Suppose, by way of
contradiction, that C =
1
(p) is not compact and let q Cl
R
n(C) C. Since C
is, by 1.6.1, nowhere dense in N, we have q Cl
R
n(C) Cl
R
n(N) = Cl
R
n(N C).
By the Curve Selection Lemma [BCR, 2.5.5], there exists a semialgebraic path
: [0, 1] Cl
R
n(N) such that ((0, 1]) N C and (0) = q. Consider the
maximal ideal
n

= f o(N) : > 0 such that (f )[

(0, ]
= 0
of o(N) (see [Fe2, 3.4]). The image of n

under Spec
s
() is a maximal ideal, because
Spec
s
() maps closed points into closed points. Thus Spec
s
()(n

) = m
p
, by the
choice of . However, the semialgebraic function f = |x p| m
p
Spec
s
()(n

)
because the composition (f )(t) = |( )(t) p| does not vanish identically
on any interval of the type (0, ] with > 0, since ((0, 1]) N C = N
1
(p).
(1.6.3)
*
s
(N) M = , or equivalently,
*
s
(N) = M.
Suppose there exists a point p
*
s
(N) M. Then, there exists n N

*
s
N Spec
*
s
(N) such that Spec
*
s
()(n) =
*
s
(n) = m
p
. By [FG4, 5.4], we may
assume that
n = n

= f o(N) : > 0 such that (f )[

(0, ]
= 0,
for some semialgebraic path : [0, 1] Cl
R
n(N) such that ((0, 1]) N and
q = (0) Cl
R
n(N) N. Since the semialgebraic function f = |x p| m
p
=
Spec
*
s
()(n

), there exists > 0 such that (f )[

(0, ]
= |()(t)p|[
(0, ]
0.
We may assume that = 1, that is, ( )(t) = p for all t [0, 1]. Hence,
((0, 1])
1
(p), and since
1
(p) is compact, q = (0)
1
(p) N, a
contradiction. Thus,
*
s
(N) M = , as wanted.
(iv) = (i). Since
*
s
:
*
s
N
*
s
M is surjective and
*
s
(N) = M, the map
is surjective and, by Lemma 3.10, it is also proper. Finally, given an open subset
1218 JOS

E F. FERNANDO AND J. M. GAMBOA

W of
*
s
N, and since
*
s
(N) = M, we have

*
s
(W) M = (
*
s
(W N) M) (
*
s
(W N) M)
=
*
s
(W N) M = (W N),
which implies that is an open map, because
*
s
is.
Remarks 3.11. (i) Assertions (ii) and (iv) in Theorem 1.6 are not equivalent to
assertion (i) if we eliminate the hypothesis
*
s
(N) = M. Indeed, the inclusion
map : (0, 1) [0, 1] is neither proper nor closed nor surjective, but the induced
maps
Spec
*
s
() = id : Spec
*
s
((0, 1)) = Spec
*
s
([0, 1]) Spec
*
s
([0, 1]), and

*
s
= id :
*
s
(0, 1) =
*
s
[0, 1] = [0, 1] [0, 1]
are open, proper and surjective. Indeed, see [FG3, 3.9], the rings o

((0, 1)) and

o

([0, 1]) are isomorphic, and so Spec

*
s
((0, 1)) = Spec
*
s
([0, 1]). On the other hand,
for the equalities
*
s
(0, 1) =
*
s
[0, 1] = [0, 1] see [FG4, 4.9].
(ii) Assertion (iii) in Theorem 1.6 is not equivalent to assertion (i) if we eliminate
the hypothesis that the ber under of each isolated point of M is compact. Indeed,
if : N p is a constant semialgebraic map and N is not compact, the map
Spec
s
() : Spec
s
(N) Spec
s
(p) p is open, proper and surjective, while is
not proper.
References
[AM] M.F. Atiyah, I.G. Macdonald: Introduction to commutative algebra. Addison-Wesley Pub-
lishing Co., Reading, Mass.-London-Don Mills, Ontario: 1969. MR0242802 (39:4129)
[BCR] J. Bochnak, M. Coste, M.-F. Roy: Real algebraic geometry. Ergeb. Math. 36, Springer-
Verlag, Berlin: 1998. MR1659509 (2000a:14067)
[B] G.W. Brumel: Quotient spaces for semialgebraic equivalence relations. Math. Z. 195
(1987), no. 1, 6978. MR888127 (88i:14015)
[Fe1] J.F. Fernando: On chains of prime ideals in rings of semialgebraic functions. Preprint
RAAG (2010). http://www.mat.ucm.es/josefer/pdfs/preprint/chains.pdf
[Fe2] J.F. Fernando: On distinguished points of the remainder of the semialge-
braic Stone

Cech compactication of a semialgebraic set. Preprint RAAG (2010).

http://www.mat.ucm.es/josefer/pdfs/preprint/remainder.pdf
[FG1] J.F. Fernando, J.M. Gamboa: On the Krull dimension of rings of semialgebraic functions.
Preprint RAAG (2010). http://www.mat.ucm.es/josefer/pdfs/preprint/dim.pdf
[FG2] J.F. Fernando, J.M. Gamboa: On the spectra of rings of semialgebraic func-
tions. Collectanea Mathematica, to appear. http://www.mat.ucm.es/josefer/pdfs/
preprint/spectra.pdf
[FG3] J.F. Fernando, J.M. Gamboa: On Banach-Stone type theorems for semialgebraic sets.
Preprint RAAG (2010). http://www.mat.ucm.es/josefer/pdfs/preprint/homeo.pdf
[FG4] J.F. Fernando, J.M. Gamboa: On the semialgebraic Stone

Cech compactication of
a semialgebraic set. Trans. Amer. Math. Soc. (to appear). http://www.mat.ucm.es/
josefer/pdfs/preprint/mspectra.pdf
[MO] G. De Marco, A. Orsatti: Commutative rings in which every prime ideal is contained in
a unique maximal ideal. Proc. Amer. Math. Soc. 30 (1971), no. 3, 459-466. MR0282962
(44:196)
[Mu] M.-A. Mulero: Algebraic properties of rings of continuous functions. Fund. Math. 149
(1996), no. 1, 5566. MR1372357 (97c:16038)
OPEN AND CLOSED MORPHISMS BETWEEN SEMIALGEBRAIC SETS 1219
[P] V.I. Ponomarev: Open mappings of normal spaces. Dokl. Akad. Nauk SSSR 126 (1959),
716718. MR0107855 (21:6577)
[PS] C. Procesi, G. Schwarz: Inequalities dening orbit spaces. Invent. Math. 81 (1985), no. 3,
539554. MR807071 (87h:20078)
Departamento de

Algebra, Facultad de Ciencias Matem aticas, Universidad Com-
plutense de Madrid, 28040 Madrid, Spain
E-mail address: [email protected]
Departamento de

Algebra, Facultad de Ciencias Matem aticas, Universidad Com-
plutense de Madrid, 28040 Madrid, Spain
E-mail address: [email protected]