• Keine Ergebnisse gefunden

Dagger completions and bornological torsion-freeness

2. Basic notions

In this section, we recall some basic notions on bornological modules and bounded homomorphisms. See [11] for more details. We also study the inheritance properties of separatedness and completeness for submodules, quotients and extensions.

Abornology on a setX is a collectionBX of subsets of X, calledbounded sets, such that all finite subsets are bounded and subsets and finite unions of bounded subsets are bounded. LetV be a complete discrete valuation ring. Abornological V-module is aV-moduleM with a bornology such that every bounded subset is

2. BASIC NOTIONS 23

contained in a boundedV-submodule. In particular, theV-submodule generated by a bounded subset is again bounded. We always writeBM for the bornology onM. Let MM be a V-submodule. The subspace bornology on M consists of all subsets of M that are bounded in M. The quotient bornology on M/M consists of all subsets of the form q(S) with S ∈ BM, where qMM/M is the canonical projection. We always equip submodules and quotients with these canonical bornologies.

Let M andN be two bornological V-modules. A V-module map fMN is bounded if f(S) ∈ BN for all S ∈ BM. Bornological V-modules and bounded V-module maps form an additive category. The isomorphisms in this category are called bornological isomorphisms. A bounded V-module map fMN is a bornological embeddingif the induced mapMf(M)is a bornological isomorphism, wheref(M) ⊆N carries the subspace bornology. It is a bornological quotient map if the induced mapM/kerfN is a bornological isomorphism. Equivalently, for eachT∈ BN there is S∈ BM withf(S) =T.

Anextension of bornologicalV-modules is a diagram of V-modules MÐ→f M Ð→g M′′

that is algebraically exact and such that f is a bornological embedding and g a bornological quotient map. Equivalently,g is a cokernel off andf a kernel ofg in the additive category of bornologicalV-modules. Asplit extension is an extension with a boundedV-linear mapsM′′M such thatgs=idM′′.

Let M be a bornological V-module. A sequence (xn)n∈N in M converges towardsxM if there are S∈ BM and a sequence (δn)n∈N inV with lim∣δn∣ =0 andxnxδnS for alln∈N. It is aCauchysequence if there areS∈ BM and a sequence(δn)n∈NinV with lim∣δn∣ =0 andxnxmδjS for all n, m, j∈Nwith n, mj. Since any bounded subset is contained in a bounded V-submodule, a sequence inM converges or is Cauchy if and only if it converges or is Cauchy in the π-adic topology on some boundedV-submodule of M.

We call a subsetS of M closed if xS for any sequence inS that converges inM toxM. These are the closed subsets of a topology onM. Bounded maps preserve convergent sequences and Cauchy sequences. Thus pre-images of closed subsets under bounded maps remain closed. That is, bounded maps are continuous for these canonical topologies.

2.1. Separated bornological modules. We call M separated if limits of convergent sequences inM are unique. If M is not separated, then the constant sequence 0 has a non-zero limit. Therefore,M is separated if and only if{0} ⊆M is closed. AndM is separated if and only if anyS∈ BM is contained in aπ-adically separated boundedV-submodule.

Lemma 2.1. Let MÐ→f MÐ→g M′′ be an extension of bornologicalV-modules.

(1) If M is separated, so is M.

(2) The quotientM′′ is separated if and only if f(M)is closed inM. (3) If Mand M′′ are separated andM′′ is torsion-free, thenM is separated.

Proof. Assertion (1) is trivial.

IfM′′ is separated, then{0} ⊆M′′ is closed. Henceg1({0}) =f(M)is closed inM. If M′′ is not separated, then the constant sequence 0 inM′′ converges to some non-zerox′′M′′. That is, there are a bounded subsetS′′M′′ and a null

sequence(δn)n∈NinV withx′′−0∈δnS′′ for alln∈N. Sinceg is a bornological quotient map, there arexM andS∈ BM withg(x) =x′′ andg(S) =S′′. We may choose yn′′S′′ withx′′=δnyn′′ and ynS withg(yn) =yn′′. So g(xδnyn) =0.

Thus the sequence(xδnyn)lies inf(M). It converges tox, which does not belong tof(M)becausex′′≠0. Sof(M)is not closed. This finishes the proof of (2).

We prove (3). Let xM belong to the closure of {0} inM. That is, there are S ∈ BM and a null sequence (δn)n∈N inV withxδnS for all n∈ N. Then g(x) ∈δng(S)for alln∈N. This impliesg(x) =0 becauseM′′is separated. So there is yM withf(y) =x. Andf(y) =xδnS. ChoosexnS withf(y) =δnxn. We may assume δn ≠0 for all n∈ Nbecause otherwise xδnS is 0. Since M′′

is torsion-free,δnxnf(M)impliesg(xn) =0. So we may writexn=f(yn)for some ynM. Sincef is a bornological embedding, the set {ynn∈N} inM is bounded. SinceMis separated and y=δnyn for alln∈N, we gety=0. Hence

x=0. So{0}is closed inM.

The quotientM/{0}of a bornologicalV-moduleM by the closure of 0 is called the separated quotient of M. It is separated by Lemma 2.1, and it is the largest separated quotient ofM. Even more, the quotient mapMM/{0}is the universal arrow to a separated bornologicalV-module, that is, any bounded V-linear map fromM to a separated bornological V-module factors uniquely throughM/{0}.

The following example shows that Lemma 2.1.(3) fails without the torsion-freeness assumption.

Example2.2. LetM=V and letM =V[x]/S, whereS is theV-submodule of V[x]generated by 1−πnxn for all n∈N. We embedM =V as multiples of 1=x0. Then

M/M=⊕

n=1

V/(πn),

We endow M,Mand M/Mwith the bornologies where all subsets are bounded.

We get an extension of bornologicalV-modulesVM ↠ ⊕n=1V/(πn). HereV and⊕n=1V/(πn)areπ-adically separated, butM is not: the constant sequence 1 inM converges to 0 because 1=1−πnxn+πnxnπnxn inM.

2.2. Completeness. We call a bornological V-module M complete if it is separated and for any S ∈ BM there is T ∈ BM so that all S-Cauchy sequences areT-convergent. Equivalently, anyS∈ BM is contained in aπ-adically complete bounded V-submodule (see [11, Proposition 2.8]). By definition, any Cauchy sequence in a complete bornologicalV-module has a unique limit.

Theorem 2.3. LetMÐ→f M Ð→g M′′ be an extension of bornologicalV-modules.

(1) If M is complete and f(M)is closed inM, thenM is complete.

(2) IfMis complete,M separated, andM′′torsion-free, thenf(M)is closed inM.

(3) LetM be complete. ThenM′′ is complete if and only if f(M)is closed inM.

(4) If M andM′′ are complete and M is separated, thenM is complete. If M andM′′ are complete andM′′ is torsion-free, then M is complete.

Proof. Statement (1) is [11, Lemma 2.13], and there is no need to repeat the proof here. It is somewhat similar to the proof of (4). Next we prove (2). Assume

2. BASIC NOTIONS 25

thatM is complete, thatM′′ is torsion-free, and thatf(M)is not closed in M. We are going to prove thatM is not separated. There is a sequence(xn)n∈Nin M for whichf(xn)n∈N converges inM towards somexf(M). So there is a bounded setSM and a sequence(δk)k∈NinV with lim∣δk∣ =0 andf(xn) −xδnSfor all n∈N. We may assume without loss of generality thatS is a boundedV-submodule and that the sequence of norms ∣δn∣ is decreasing: let δn be the δm for mn with maximal norm. Then f(xn) −xδnSδnS and still lim∣δn∣ = 0. We may write f(xn) −x=δnyn with ynS. Let m< n. Then δmg(ym) = −g(x) = δng(yn)andδn/δmV; this implies firstδm⋅ (g(ym) −g(yn) ⋅δn/δm) =0 and then g(ym) = g(yn) ⋅δn/δm because M′′ is torsion-free. So there is zm,nM with ym+f(zm,n) =ynδn/δm. We even havezm,nf1(S)becauseS is aV-submodule.

The subset f1(S) ⊆Mis bounded becausef is a bornological embedding. We get f(xn) −f(xm) =δnynδmym=f(δmzm,n)and hencexnxm=δmzm,nfor n>m.

This witnesses that the sequence(xn)n∈Nis Cauchy in M. SinceMis complete, it converges towards someyM. Thenf(xn)converges both towardsf(y) ∈f(M) and towardsxf(M). SoM is not separated. This finishes the proof of (2).

Next we prove (3). Iff(M)is not closed, then Lemma 2.1 shows thatM′′ is not separated and hence not complete. Conversely, we claim thatM′′is complete iff(M)is closed. Lemma 2.1 shows that M′′ is separated. LetS′′∈ BM′′. There is S ∈ BM with g(S) =S′′ because g is a bornological quotient map. And there is T ∈ BM so that any S-Cauchy sequence is T-convergent. We claim that any S′′-Cauchy sequence isg(T)-convergent. So let(x′′n)n∈N be anS′′-Cauchy sequence.

Thus there is a null sequence(δn)n∈NinV with x′′nx′′mδjS′′ for alln, m, j∈N withn, mj. As above, we may assume without loss of generality that the sequence of norms ∣δn∣ is decreasing. Choose anyx0M withg(x0) =x′′0. For each n∈N, chooseynS withx′′n+1x′′n=δng(yn). Let

xn∶=x0+δ0y0+ ⋯ +δn1yn1.

Then g(xn) = x′′n. And xn+1xn = δnynδnS. Sinceδn∣ is decreasing, this impliesxmxnδnS for allmn. So the sequence(xn)n∈N isS-Cauchy. Hence it isT-convergent. Thus g(xn) =x′′n isg(T)-convergent as asserted. This finishes the proof of (3).

Finally, we prove (4). So we assume M andM′′ to be complete. If M′′ is torsion-free, thenM is separated by Lemma 2.1. Hence the second statement in (4) is a special case of the first one. Let S ∈ BM. We must find T ∈ BM so that everyS-Cauchy sequence isT-convergent. SinceM is separated, this says that it is complete. Since M′′ is complete, there is aπ-adically complete V-submodule T0 ∈ BM′′ that contains g(S). Since g is a bornological quotient map, there is T1∈ BM withg(T1) =T0. Replacing it byT1+S, we may arrange, in addition, that ST1. Sincef is a bornological embedding,T2∶=f1(T1)is bounded inM. AsM is complete, there isT3∈ BM so that everyT2-Cauchy sequence isT3-convergent.

We claim that anyS-Cauchy sequence isT1+f(T3)-convergent. The proof of this claim will finish the proof of the theorem.

Let (xn)n∈N be an S-Cauchy sequence. So there are δnV andynS with lim∣δn∣ =0 andxn+1xn=δnyn. As above, we may assume that∣δn∣is decreasing and that δ0 = 1. Since g(yn+k) ∈ g(S) ⊆ T0 and T0 is π-adically complete, the

following series converges inT0:

(2.4) w˜n∶= −∑

k=0

δn+k

δn g(yn+k). Since ˜wnT0, there is wnT1 withg(wn) =w˜n. So

δng(wn) = lim

N→∞g(∑N

k=0

δn+kyn+k)

= lim

N→∞g(xnxN+n+1) =g(xn) − lim

N→∞g(xN). In particular,g(w0) =g(x0) −limN→∞g(xN). Now let

˜

xk ∶=xkδkwk+w0x0. Then

g(x˜k) =g(xk) −g(xk) + lim

N→∞g(xN) +g(x0) − lim

N→∞g(xN) −g(x0) =0.

So ˜xkf(M)for allk∈N. And

(2.5) x˜n+1x˜n=xn+1xnδn+1wn+1+δnwn

=δnyn+δnwnδn+1wn+1=δn⋅ (yn+wnδn+1

δn

wn+1). Letzn∶=yn+wnδn+1δn wn+1. A telescoping sum argument shows that

(2.6) g(zn) =g(yn) +w˜nδn+1 δn

˜ wn+1=0.

So znf(M). And znS+T1+T1 = T1. Thus there is ˆznf1(T1) = T2 with zn =f(zˆn). Equation (2.5) means that the sequence f1(x˜n) is T2-Cauchy.

Hence it isT3-convergent. So(x˜n)isf(T3)-convergent. Then(xn)isT1+f(T3)

-convergent.

The following examples show that the technical extra assumptions in (2) and (4) in Theorem 2.3 are necessary. They only involve extensions ofV-modules with the bornology where all subsets are bounded. For this bornology, bornological com-pleteness and separatedness are the same asπ-adic completeness and separatedness, respectively, and any extension ofV-modules is a bornological extension.

Example 2.7. Let M∶= {0} andM ∶=F with the bornology of all subsets.

Then M is bornologically complete, but not closed in M, and M/M = M is torsion-free. So Theorem 2.3.(2) needs the assumption thatM be separated.

Example 2.8. Let M be the V-module of all power series ∑n=0cnxn with lim∣cn∣ =0 and with the bornology where all subsets are bounded; this is theπ-adic completion of the polynomial algebraV[x]. Let M=M and define fMM, f(∑n=0cnxn) ∶= ∑n=0cnπnxn. This is a bornological embedding simply because all subsets inM =M are bounded. Letpn∶= ∑nj=0xj. This sequence in M=M does not converge. Nevertheless, the sequencef(pn) = ∑nj=0πjxj converges inM to

j=0πjxj. Thusf(M)is not closed inM, althoughM andMare complete andf is a bornological embedding. So Theorem 2.3.(2) needs the assumption thatM′′ be torsion-free.

2. BASIC NOTIONS 27

Example2.9. We modify Example 2.2 to produce an extension of V-modules NNN′′ whereN andN′′are π-adically complete, butN is notπ-adically separated and hence notπ-adically complete. We let N∶=V/(π) =F. We letN′′

be theπ-adic completion of theV-moduleM′′of Example 2.2. That is, N′′∶= {(cn)n∈N∈ ∏

n=0

V/(πn) ∶lim∣cn∣ =0}. This is indeed π-adically complete. So is

N1∶= {(cn)n∈N∈∏

n=0

V/(πn+1) ∶lim∣cn∣ =0}.

The kernel of the quotient map qN1N′′ is isomorphic to ∏n=0V/(π) = ∏NF. This is aF-vector space, and it contains theF-vector space∑n=0F. Since anyF-vector space has a basis, we may extend the linear functional ∑n=0F → F, (cn)n∈N

n=0cn, to a F-linear functional σ∶ ∏NF → F. Let L ∶= kerσ ⊆ kerq and let N∶=N1/L. The mapqdescends to a surjectiveπ-linear mapNN′′. Its kernel is isomorphic to∏NF/kerσ≅F=N. The functionalσ∶ ∏NF→F vanishes onδ0δk

for allk∈N, but not onδ0. When we identify∏NF≅kerq, we mapδk toπkδkN1. Soδ0 andπkδk get identified inN, butδ0does not become 0: it is the generator of N=V/(π)insideN. Since[δ0] =πk[δk]inN, theV-moduleN is notπ-adically separated.

Thecompletion M of a bornological V-moduleM is a complete bornological V-module with a boundedV-linear mapMM that is universal in the sense that any boundedV-linear map fromM to a complete bornologicalV-moduleX factors uniquely throughM. Such a completion exists and is unique up to isomorphism (see [11, Proposition 2.15]). We shall describe it more concretely later when we

need the details of its construction.

2.3. Vector spaces over the fraction field. Recall that F denotes the quotient field ofV. AnyV-linear map between twoF-vector spaces is alsoF-linear.

SoF-vector spaces withF-linear maps form a full subcategory in the category of V-modules. AV-moduleM comes from aF-vector space if and only if the map

(2.10) πMMM, mπm,

is invertible. We could define bornologicalF-vector spaces without reference toV. Instead, we realise them as bornologicalV-modules with an extra property:

Definition2.11. A bornologicalV-moduleM is a bornological F-vector space if the mapπM in (2.10) is a bornological isomorphism, that is, an invertible map with bounded inverse.

Given a bornologicalV-moduleM, the tensor productFM∶=FV M with the tensor product bornology (see [11, Lemma 2.18]) is a bornological F-vector space because multiplication byπis a bornological isomorphism onF.

Lemma 2.12. The canonical bounded V-linear map ιMMFM, m↦1⊗m,

is the universal arrow fromM to a bornologicalF-vector space, that is, any bounded V-linear mapfMN to a bornologicalF-vector spaceN factors uniquely through a boundedV-linear mapf#FMN, and this map is alsoF-linear.

Proof. A V-linear map f#FMN must be F-linear. Hence the only possible candidate is theF-linear map defined byf#(xm) ∶=xf(m)formM, xF. Any bounded submodule ofFM is contained inπkVSfor some bounded submoduleSM and somek∈N, andf#(πkVS) =πNk(f(S))is bounded inN becauseπN is a bornological isomorphism. Thusf#is bounded.