betaFeedback

WorksMathematics

On Proximity Subgroups of Groups

Abstract

By a proximity subgroup\textit{proximity subgroup} of a group GG we mean a norm νν on GG valued in a complete lattice LL equipped with a binary sum (an additive counterpart of a multiplicative structure). The idea is that the smaller the value ν(y)ν(y) is, the closer y∈Gy\in G is to the subgroup of GG defined by the equation ν(x)=⊥ν(x)=\bot, where ⊥\bot is the bottom element of LL. We test whether this notion is a suitable substitute for the usual notion of a subgroup (recovered by taking L={⊥,⊤}L=\{\bot,\top\} with join as the sum), by showing that the form of proximity subgroups is noetherian. Intuitively, this means that proximity subgroups are as well suited for formulating and proving isomorphism theorems and homological diagram lemmas for groups as the usual subgroups are. We study several questions in relation to this, including what makes the noetherian form of usual subgroups special among all of these noetherian forms, and in fact, all noetherian forms of the category of groups.

Introduction

A noetherian form over a category is an additional structure on the category that includes a proper factorization system and enables one to formulate and prove isomorphism theorems and homological diagram lemmas [7, 3]. One of the motivating examples is the form of subgroups over the category of groups, where the isomorphism theorems and homological diagram lemmas are the usual ones. An abstract form replaces subgroups of a group GG with formal subgroups, these are elements of the fibre F−1(G)F^{-1}(G), where FF is a functor (the “form”) with codomain the category of groups.

In this paper we address the following question: are there other reasonable formal subgroups that are not the same as the usual subgroups, but which yield the same (or similar) isomorphism theorems and homological diagram lemmas for groups? We show that every non-trivial complete lattice LL equipped with a multiplicative structure (which we write in additive notation) provides such notion of a formal subgroup.

We call these formal subgroups proximity subgroups. More precisely, an LL-valued proximity subgroup of GG is a norm ν:G→L\nu: G \to L satisfying

ν(0)=⊥,ν(−g)=ν(g),ν(g+h)≤ν(g)+ν(h),\nu(0) = \bot,\qquad\nu(-g) = \nu(g),\qquad\nu(g + h) \le\nu(g) + \nu(h),

where the coefficient sum is associative, commutative and monotone, with identity ⊥\bot. The norm terminology agrees with that of Clementino, Hofmann and Tholen [2] in the quantale case. The bottom-level set of ν\nu is an ordinary subgroup; a proximity subgroup retains the additional information supplied by its other values. This includes ordinary extended real-valued norms, as well as the non-Archimedean case in which the coefficient sum is lattice join. Note that even though we write groups additively, they are not assumed abelian.

The resulting form of proximity subgroups is fibre-complete and noetherian (in particular, it is a topological functor [1]). Moreover, the construction is fully faithful: morphisms between these forms are precisely those induced by maps of coefficient lattices preserving arbitrary meets, finite joins and the coefficient sum. Here morphisms of noetherian forms preserve finite fibre operations and both direct and inverse image operations, without being required to preserve arbitrary fibre meets or joins. Since every complete lattice admits join as its sum, this also yields a full embedding of a category of non-trivial complete lattices into the category of noetherian forms over groups.

The usual subgroup form has a distinguished position not only among all proximity subgroup forms, but even among all noetherian forms over the category of groups. It was established in [10] that the form of subgroups is optimal (it is isomorphic to a subform of every other noetherian form over the same base category); we prove that, up to a unique isomorphism, the form of subgroups is in fact the only optimal noetherian form over groups.

Noetherian forms and their morphisms

We use the functorial definition of a noetherian form, together with its equivalent presentation by fibre lattices and image operations; see [7, 10].

Definition 1.1. A form over a category C\mathcal{C} is a faithful amnestic functor

F:A⟶C.F : \mathcal{A} \longrightarrow\mathcal{C}.

Here amnestic means that an isomorphism mapped to an identity is itself an identity. We assume that every fibre is small. For X∈CX \in\mathcal{C}, the fibre F−1(X)F^{-1}(X) is then a poset; its objects are called clusters in XX. We write S≤fTS \le_f T if there is a morphism S→TS \to T in A\mathcal{A} mapped by FF to f:X→Yf : X \to Y. Such a lift, when it exists, is unique.

The functor FF is a bifibration precisely when, for every f:X→Yf : X \to Y, there are functorial maps

f!:F−1(X)⇄F−1(Y):f∗f_! : F^{-1}(X) \rightleftarrows F^{-1}(Y) : f^*

such that

f!S≤T⟺S≤fT⟺S≤f∗T.(1)f_!S \le T \quad\Longleftrightarrow\quad S \le_f T \quad\Longleftrightarrow\quad S \le f^*T. \tag*{(1)}

These are the direct and inverse image operations. Their functoriality includes (gf)!=g!f!(gf)_! = g_!f_!, (gf)∗=f∗g∗(gf)^* = f^*g^*, and identity laws. Suppose also that each fibre is a bounded lattice, with bounds 0X,1X0_X,1_X. Put

Ker⁡Ff=f∗(0Y),Im⁡Ff=f!(1X).\operatorname{Ker}_F f=f^*(0_Y),\qquad\operatorname{Im}_F f=f_!(1_X).

A cluster is normal if it is a kernel, and conormal if it is an image. For a conormal SS in XX, an associated embedding is a morphism iS:S0→Xi_S:S_0\to X with image SS, universal among morphisms uu into XX satisfying Im⁡Fu≤S\operatorname{Im}_F u\leq S. Thus

Im⁡Fu≤S⟺u=iSv for a unique v.(2)\operatorname{Im}_F u\leq S\quad\Longleftrightarrow\quad u=i_Sv\text{ for a unique }v. \tag*{(2)}

Dually, an associated projection of a normal cluster NN in XX is a morphism pN:X→QNp_N:X\to Q_N with kernel NN such that

N≤Ker⁡Fu⟺u=vpN for a unique v.(3)N\leq\operatorname{Ker}_F u\quad\Longleftrightarrow\quad u=vp_N\text{ for a unique }v. \tag*{(3)}

These universal properties imply that associated embeddings are monomorphisms and associated projections are epimorphisms.

Definition 1.2. A form F:A→CF:\mathcal{A}\to\mathcal{C} is noetherian if it is a bifibration with bounded lattice fibres and satisfies the following conditions.

(N1) For every f:X→Yf:X\to Y and clusters SS in XX, TT in YY,

f∗f!S=S∨Ker⁡Ff,f!f∗T=T∧Im⁡Ff.(4)f^*f_!S=S\vee\operatorname{Ker}_F f,\qquad f_!f^*T=T\wedge\operatorname{Im}_F f. \tag*{(4)}

(N2) Associated embeddings and projections exist for conormal and normal clusters, respectively. Every ff factors as

f=iIm⁡Ff h pKer⁡Ff,f=i_{\operatorname{Im}_F f}\,h\,p_{\operatorname{Ker}_F f},

where hh is an isomorphism.

(N3) Binary joins of normal clusters are normal, and binary meets of conormal clusters are conormal.

A noetherian form is fibre-complete if all its fibres are complete lattices.

The term noetherian here refers to these categorical axioms, not to an ascending chain condition.

Definition 1.3. A morphism of noetherian forms F:A→CF:\mathcal{A}\to\mathcal{C} and F′:B→CF':\mathcal{B}\to\mathcal{C} is a functor T:A→BT:\mathcal{A}\to\mathcal{B} over C\mathcal{C} such that each fibre map preserves 0,1,∧,∨0,1,\wedge,\vee, and

TYf!F=f!F′TX,TXfF∗=fF′∗TY.(5)T_Yf_!^F=f_!^{F'}T_X,\qquad T_Xf_F^*=f_{F'}^*T_Y. \tag*{(5)}

No preservation of arbitrary fibre meets or joins is required. We denote this category by NF(C)\mathbf{NF}(\mathcal{C}), and its full subcategory of fibre-complete objects by NFc(C)\mathbf{NF}^{c}(\mathcal{C}).

Lemma 1.4. Morphisms of noetherian forms preserve kernels, images, and their associated projections and embeddings.

Proof. Preservation of bounds and (5) give TX(Ker⁡Ff)=Ker⁡F′fT_X(\operatorname{Ker}_F f)=\operatorname{Ker}_{F'}f and TY(Im⁡Ff)=Im⁡F′fT_Y(\operatorname{Im}_F f)=\operatorname{Im}_{F'}f. In a noetherian form a morphism is an associated embedding if and only if its kernel is bottom; dually, it is an associated projection if and only if its image is top. These criteria follow from the factorization axiom: a projection of bottom and an embedding of top are isomorphisms. Hence an associated embedding of SS remains an associated embedding of TXST_XS, and dually for projections. □

For optimality we also need a weaker notion. An operator of forms is any functor over the base. It is full when

S≤fT⟺αS≤fαT.S\leq_f T\quad\Longleftrightarrow\quad\alpha S\leq_f\alpha T.

A full operator identifies its source with a full subcategory over the base, called a subform. In particular its fibre maps are order embeddings. Full operators are not required to preserve fibre bounds, fibre operations, or either image operation.

Proposition 1.5. Every noetherian form over Grp\mathbf{Grp} has all injective homomorphisms as its associated embeddings and all surjective homomorphisms as its associated projections.

Proof. Associated embeddings are injective because they are monomorphisms. Associated projections are surjective because epimorphisms in Grp\mathbf{Grp} are surjective. If an injective homomorphism is factored by (N2), its projection factor is both injective and surjective, hence an isomorphism. Thus the original homomorphism is an associated embedding. The dual argument applies to a surjective homomorphism. □

The subgroup form is the functor

S:Grp2⟶Grp,(G,H)↦G,S : \mathbf{Grp}_2 \longrightarrow\mathbf{Grp}, \qquad(G,H) \mapsto G,

where H≤GH \le G and an arrow (G,H)→(K,J)(G,H) \to(K,J) is a homomorphism f:G→Kf : G \to K satisfying f(H)≤Jf(H) \le J. It is a fibre-complete noetherian form, with the usual direct and inverse images of subgroups.

Norms with values in multiplicative complete lattices

Definition 2.1. A multiplicative complete lattice in additive convention is a complete lattice LL, with least element 00 and greatest element ∞\infty, equipped with an associative, commutative, monotone operation

+:L×L⟶L+ : L \times L \longrightarrow L

whose identity is 00. We do not assume that ++ distributes over meets or joins.

This is the order-dual, additively written presentation of a complete lattice with monotone commutative monoid multiplication and top element as unit. Its order dual is a multiplicative lattice in the broad sense of [5, 4], with these additional monoid and monotonicity axioms. Some authors include distribution over arbitrary joins in the multiplicative definition; the difference in terminology is discussed in [6]. In additive convention that stronger requirement is

a+⋀ibi=⋀i(a+bi).(6)a + \bigwedge_i b_i = \bigwedge_i(a+b_i). \tag*{(6)}

It says precisely that the order dual is an integral commutative quantale. Neither (6) nor any other distributivity is used in our general construction.

Monotonicity and the unit law imply

a∨b≤a+b,a+∞=∞.(7)a \vee b \le a+b,\qquad a+\infty=\infty. \tag*{(7)}

Two examples will be used repeatedly:

([0,∞],≤,+,0),(L,≤,+,0) with a+b:=a∨b([0,\infty],\le,+,0),\qquad(L,\le,+,0)\ \text{with }a+b:=a\vee b

for an arbitrary complete lattice LL. In the first example addition is extended by a+∞=∞a+\infty=\infty. In the second, coefficient addition is idempotent. Conversely, if a+a=aa+a=a for every aa, then +=∨+=\vee: putting c=a∨bc=a\vee b gives c≤a+b≤c+c=cc\le a+b\le c+c=c.

Definition 2.2. Let LL be a multiplicative complete lattice in additive convention. An LL-valued norm on a group GG is a map ν:G→L\nu:G\to L such that

ν(0)=0,ν(−g)=ν(g),ν(g+h)≤ν(g)+ν(h).(8)\nu(0)=0,\qquad\nu(-g)=\nu(g),\qquad\nu(g+h)\leq\nu(g)+\nu(h). \tag*{(8)}

The pair (G,ν)(G,\nu) is called an LL-normed group. A homomorphism f:(G,ν)→(H,μ)f:(G,\nu)\to(H,\mu) is nonexpansive, or an LL-normed homomorphism, when

μ(f(g))≤ν(g)(g∈G).(9)\mu(f(g))\leq\nu(g)\qquad(g\in G). \tag*{(9)}

These objects and arrows form a category NGrpL\mathbf{NGrp}_{L}, with forgetful functor

UL:NGrpL⟶Grp.U_{L}:\mathbf{NGrp}_{L}\longrightarrow\mathbf{Grp}.

We write Norm⁡L(G)\operatorname{Norm}_{L}(G) for its fibre. Its order is reverse pointwise order:

ν⪯μ  ⟺  μ(g)≤ν(g)for every g∈G.(10)\nu\preceq\mu\iff\mu(g)\leq\nu(g)\quad\text{for every }g\in G. \tag*{(10)}

We use norm in the categorical sense of Clementino, Hofmann and Tholen [2 §5]: symmetry under inversion is required, but separation and finiteness are not. Thus ν(g)=0\nu(g)=0 need not imply g=0g=0, and the value ∞\infty is allowed. When (6) holds, V=L\mathcal{V}=L is an integral commutative quantale, and our LL-normed groups are precisely their V\mathcal{V}-normed groups. Our nonexpansive homomorphisms correspond to V\mathcal{V}-normed functors between the associated one-object categories. We retain these group-norm axioms for the more general coefficients of Definition 2.1, without assuming distributivity.

Coefficient comparisons use ≤\leq, whereas fibre comparisons use ⪯\preceq. We distinguish fibre operations by ∧F\wedge_{F}, ∨F\vee_{F} when needed. Ordinary extended real-valued group norms are obtained with L=[0,∞]L=[0,\infty] and ordinary addition; no homogeneity or conjugation invariance is imposed. For related real-valued length functions see [11]. Such norms correspond to maps S:G→[0,1]S:G\to[0,1] with

S(0)=1,S(−g)=S(g),S(g)S(h)≤S(g+h),S(0)=1,\qquad S(-g)=S(g),\qquad S(g)S(h)\leq S(g+h),

by S(g)=e−ν(g)S(g)=e^{-\nu(g)}, with e−∞=0e^{-\infty}=0 and −log⁡0=∞-\log0=\infty. With a+b=a∨ba+b=a\vee b, condition (8) instead becomes the non-Archimedean inequality. For real coefficients this uses maximum in place of addition. The lattice-valued fuzzy-subgroup literature, beginning with [8, 12], concerns the order-dual meet inequality.

Lemma 2.3. The functor ULU_{L} is a bifibration with complete lattice fibres. Inverse image is

f∗μ=μ∘f.f^{*}\mu=\mu\circ f.

For an arbitrary map u:G→Lu:G\to L, define

cG(u)(g)=⋀{η(g):η∈Norm⁡L(G), η(x)≤u(x) for all x∈G}.(11)c_{G}(u)(g)=\bigwedge\{\eta(g):\eta\in\operatorname{Norm}_{L}(G),\ \eta(x)\leq u(x)\text{ for all }x\in G\}. \tag*{(11)}

Then cG(u)c_{G}(u) is the greatest norm pointwise below uu, and direct image is

(f!ν)(h)=cH(rfν)(h),(rfν)(h)=⋀{g:f(g)=h}ν(g).(12)(f_{!}\nu)(h)=c_{H}(r_{f}\nu)(h),\qquad(r_{f}\nu)(h)=\bigwedge_{\{g:f(g)=h\}}\nu(g). \tag*{(12)}

The empty meet in (12) is ∞\infty.

Proof. Faithfulness and amnesticity are immediate. Arbitrary pointwise suprema of norms are norms. Indeed,

⋁iνi(g+h)≤⋁i(νi(g)+νi(h))≤(⋁iνi(g))+(⋁iνi(h)).\bigvee_{i}\nu_{i}(g+h)\leq\bigvee_{i}\bigl(\nu_{i}(g)+\nu_{i}(h)\bigr)\leq\left(\bigvee_{i}\nu_{i}(g)\right)+\left(\bigvee_{i}\nu_{i}(h)\right).

The empty supremum is the zero function. These suprema are the fibre meets:

(⋀F,iνi)(g)=⋁iνi(g).\left(\bigwedge_{F,i}\nu_i\right)(g)=\bigvee_i\nu_i(g).

Thus all meets exist and the fibres are complete. Their joins are

⋁F,iνi=cG(⋀iνi).\bigvee_{F,i}\nu_i=c_G\left(\bigwedge_i\nu_i\right).

The zero norm lies pointwise below every uu, and the preceding observation proves the assertion about cGc_G.

Composition with a homomorphism preserves norms. For μ∈Norm⁡L(H)\mu\in\operatorname{Norm}_L(H),

f!ν⪯μ⟺μ≤cH(rfν)pointwisef!\nu\preceq\mu\Longleftrightarrow\mu\leq c_H(r_f\nu)\quad\text{pointwise}
⟺μ≤rfν\Longleftrightarrow\mu\leq r_f\nu
⟺μ∘f≤ν\Longleftrightarrow\mu\circ f\leq\nu
⟺ν⪯f∗μ.\Longleftrightarrow\nu\preceq f^*\mu.

This is the adjunction describing lifts of ff. Inverse images are functorial; uniqueness of left adjoints gives functoriality of direct images. □

For a subgroup A≤GA\leq G, set

κA(g)={0,g∈A,∞,g∉A.(13)\kappa_A(g)= \begin{cases} 0, & g\in A,\\ \infty, & g\notin A. \end{cases} \tag*{(13)}

These are norms by (7). The fibre bottom and top are κ{0}\kappa_{\{0\}} and κG\kappa_G, respectively.

Lemma 2.4. Let p:G↠Hp:G\twoheadrightarrow H be surjective with kernel NN, and let i:A↪Gi:A\hookrightarrow G be an inclusion.

(i) Inverse image along pp identifies Norm⁡L(H)\operatorname{Norm}_L(H) with the upper fibre interval ↑κN\mathord{\uparrow}\kappa_N. Consequently

p∗p∗=id⁡,p∗p∗ν=ν∨FκN.p^*p_*=\operatorname{id},\qquad p_*p^*\nu=\nu\vee_F\kappa_N.

(ii) Direct image along ii extends a norm on AA by ∞\infty outside AA. It identifies Norm⁡L(A)\operatorname{Norm}_L(A) with the lower fibre interval ↓κA\mathord{\downarrow}\kappa_A. Consequently

i∗i∗=id⁡,i∗i∗ν=ν∧FκA.i^*i_*=\operatorname{id},\qquad i_*i^*\nu=\nu\wedge_F\kappa_A.

Proof. The condition κN⪯ν\kappa_N\preceq\nu says exactly that ν\nu vanishes on NN. For n∈Nn\in N the triangle inequality gives

ν(g+n)≤ν(g)+ν(n)=ν(g),ν(g)≤ν(g+n)+ν(−n)=ν(g+n).\nu(g+n)\leq\nu(g)+\nu(n)=\nu(g),\qquad\nu(g)\leq\nu(g+n)+\nu(-n)=\nu(g+n).

Hence ν\nu is constant on the cosets of NN, and descends uniquely to a norm on HH. Conversely every pullback vanishes on NN. The reflector onto the upper interval is fibre join with κN\kappa_N, proving (i).

Extension by ∞\infty is a norm: if both summands lie in AA, use the inequality in AA; otherwise its right-hand side is ∞\infty. A norm lies below κA\kappa_A in fibre order precisely when it is ∞\infty outside AA. Restriction and extension give the stated interval isomorphism, proving (ii). □

Theorem 2.5. For every nontrivial multiplicative complete lattice LL in additive convention, the functor ULU_L is a fibre-complete noetherian form. For every homomorphism f:G→Hf:G\to H,

Ker⁡ULf=κker⁡f,Im⁡ULf=κf(G).(14)\operatorname{Ker}_{U_L} f=\kappa_{\ker f},\qquad\operatorname{Im}_{U_L} f=\kappa_{f(G)}. \tag*{(14)}

Its associated embeddings and projections are precisely the injective and surjective group homomorphisms, respectively.

Proof. Factor ff as G→pf(G)→iHG \xrightarrow{p} f(G) \xrightarrow{i} H. Lemma 2.4 gives

f∗f!ν=p∗i∗i!p!ν=p∗p!ν=ν∨Fκker⁡f,f^{*}f_{!}\nu=p^{*}i^{*}i_{!}p_{!}\nu=p^{*}p_{!}\nu=\nu\vee_F\kappa_{\ker f},
f!f∗μ=i!p!p∗i∗μ=i!i∗μ=μ∧Fκf(G).f_{!}f^{*}\mu=i_{!}p_{!}p^{*}i^{*}\mu=i_{!}i^{*}\mu=\mu\wedge_F\kappa_{f(G)}.

The fibre bounds give (14), proving (N1).

Since 0≠∞0\ne\infty in LL,

κA⪯κB⟺A≤B.(15)\kappa_A\preceq\kappa_B\Longleftrightarrow A\le B. \tag*{(15)}

Thus A↪GA\hookrightarrow G has image κA\kappa_A and its ordinary subgroup universal property is exactly (2). For N⊴GN\trianglelefteq G, the quotient G↠G/NG\twoheadrightarrow G/N has kernel κN\kappa_N and satisfies (3). The usual group factorization establishes (N2). The assertion about all associated maps also follows from Proposition 1.5.

Finally, for normal subgroups NN, MM and arbitrary subgroups AA, BB,

κN∨FκM=κN+M,κA∧FκB=κA∩B.(16)\kappa_N\vee_F\kappa_M=\kappa_{N+M},\qquad\kappa_A\wedge_F\kappa_B=\kappa_{A\cap B}. \tag*{(16)}

The second equality is pointwise. For the first, the zero set of any norm is a subgroup. A fibre upper bound of κN,κM\kappa_N,\kappa_M therefore vanishes on the subgroup N+MN+M they generate, and hence lies above κN+M\kappa_{N+M} in fibre order. This proves (N3). □

Corollary 2.6. The normal clusters of ULU_L over GG are exactly κN\kappa_N with N⊴GN\trianglelefteq G, and the conormal clusters are exactly κA\kappa_A with A≤GA\le G. Every normal cluster is conormal.

Proof. Use the quotient G↠G/NG\twoheadrightarrow G/N, the inclusion A↪GA\hookrightarrow G, and (14). □

Proposition 2.7. The assignments A↦κAA\mapsto\kappa_A define a full operator S→ULS\to U_L that is also a morphism of noetherian forms. It preserves all fibre meets and joins.

Proof. For subgroups Ai≤GA_i\le G,

⋀F,iκAi=κ⋂iAi,⋁F,iκAi=κ⋃iAi.\bigwedge_{F,i}\kappa_{A_i}=\kappa_{\bigcap_i A_i},\qquad \bigvee_{F,i}\kappa_{A_i}=\kappa_{\bigcup_i A_i}.

The meet formula is pointwise. The join formula follows because a norm vanishing on all AiA_i vanishes on the subgroup they generate. These arguments include empty families. Moreover,

f∗κB=κf−1(B),f!κA=κf(A).f^{*}\kappa_B=\kappa_{f^{-1}(B)},\qquad f_{!}\kappa_A=\kappa_{f(A)}.

The first equality is immediate. The second follows from the adjunction and the fact that the zero set of a norm is a subgroup. Finally, κA≤fκB\kappa_A\le_f\kappa_B is equivalent to f(A)≤Bf(A)\le B, which proves fullness. □

Write 2={0,∞}2=\{0,\infty\} with addition given by maximum. Then U2≅SU_2\cong S. For the cyclic group C2={0,t}C_2=\{0,t\}, every a∈La\in L occurs uniquely as ν(t)\nu(t); hence

Norm⁡L(C2)≅Lop.(17)\operatorname{Norm}_L(C_2)\cong L^{\mathrm{op}}. \tag*{(17)}

If 0<a<∞0<a<\infty, this norm on C2C_2 is neither normal nor conormal. Nontriviality is essential: for the singleton coefficient lattice, every homomorphism has bottom kernel and top image, and the factorization axiom would force every homomorphism to be an isomorphism.

Proposition 2.8. If (6) holds, then direct image requires no closure:

(f!ν)(h)=⋀f(g)=hν(g).(f_!\nu)(h)=\bigwedge_{f(g)=h}\nu(g).

In particular, for ordinary extended norms,

(f!ν)(h)=inf⁡f(g)=hν(g).(f_!\nu)(h)=\inf_{f(g)=h}\nu(g).

Proof. Put u=rfνu=r_f\nu. It has u(0)=0u(0)=0 and u(−h)=u(h)u(-h)=u(h). Distributing addition over both infima gives

u(h)+u(k)=⋀f(g)=hf(g′)=k(ν(g)+ν(g′))u(h)+u(k)=\bigwedge_{\substack{f(g)=h\\f(g')=k}}\left(\nu(g)+\nu(g')\right)
≥⋀f(g)=hf(g′)=kν(g+g′)≥⋀f(z)=h+kν(z)=u(h+k).\geq\bigwedge_{\substack{f(g)=h\\f(g')=k}}\nu(g+g')\geq\bigwedge_{f(z)=h+k}\nu(z)=u(h+k).

Thus uu is already a norm. Extended real addition satisfies (6).

Example 2.9. The closure in (12) cannot be omitted in general. Let L=M3L=M_3 be the five-element lattice with distinct atoms aa, bb, cc, and use coefficient addition +=∨+=\vee. In G=(C2)4G=(C_2)^4, with basis e1,e2,e3,e4e_1,e_2,e_3,e_4, put

Aa=⟨e1⟩,Ab=⟨e2,e3⟩,Ac=⟨e4⟩.A_a=\langle e_1\rangle,\qquad A_b=\langle e_2,e_3\rangle,\qquad A_c=\langle e_4\rangle.

Define ν(0)=0\nu(0)=0, give the nonzero elements of AaA_a, AbA_b, AcA_c the values aa, bb, cc, respectively, and give every other element value ∞\infty. This is a norm: each set with a fixed atomic upper bound is a subgroup, and the join of distinct atoms is ∞\infty. Let q:G→(C2)2q:G\to(C_2)^2 send (u,v,w,z)(u,v,w,z) to (u+v,w+z)(u+v,w+z), and write x=(1,0)x=(1,0), y=(0,1)y=(0,1). Its raw fibre infimum satisfies

(rqν)(x)=0,(rqν)(y)=0,(rqν)(x+y)=b.(r_q\nu)(x)=0,\qquad(r_q\nu)(y)=0,\qquad(r_q\nu)(x+y)=b.

It fails the triangle inequality. Its norm closure is the zero function, which is the actual direct image.

Proposition 2.10. For a∈La\in L, the sublevel set

{g:ν(g)≤a}\{g:\nu(g)\leq a\}

is a subgroup for every LL-valued norm on every group if and only if a+a=aa+a=a. Consequently all sublevel sets of all such norms are subgroups if and only if coefficient addition is lattice join.

Proof. If a+a=aa+a=a, the identity, inverse and sum axioms establish the subgroup condition. Conversely, on C2⊕C2={0,x,y,x+y}C_2\oplus C_2=\{0,x,y,x+y\} define

ν(0)=0,ν(x)=ν(y)=a,ν(x+y)=a+a.\nu(0)=0,\qquad\nu(x)=\nu(y)=a,\qquad\nu(x+y)=a+a.

This is a norm; the remaining inequalities follow from monotonicity and 0≤a0\leq a. If a+a≠aa+a\ne a, its aa-sublevel set contains xx and yy but not x+yx+y. The final assertion follows from the idempotence characterization after (7).

For example, ν(n)=∣n∣\nu(n)=|n| on Z\mathbb{Z} has the sublevel set {−1,0,1}\{-1,0,1\} at a=1a=1, which is not a subgroup. This distinction does not affect Theorem 2.5, Corollary 2.6, or the characteristic embedding.

A full embedding of multiplicative complete lattices

Let MCL denote the category of nontrivial multiplicative complete lattices in additive convention. A morphism h:L→Mh : L \to M preserves arbitrary meets, finite joins, and addition:

h(⋀iai)=⋀ih(ai),h(a∨b)=h(a)∨h(b),h(0)=0,h(a+b)=h(a)+h(b).(18)\begin{aligned} h\left(\bigwedge_i a_i\right) &= \bigwedge_i h(a_i), \qquad h(a \vee b) = h(a) \vee h(b),\\ h(0) &= 0, \qquad h(a+b) = h(a) + h(b). \tag*{(18)} \end{aligned}

The empty meet gives h(∞)=∞h(\infty) = \infty. Thus bounds are preserved. In the order-dual multiplicative convention, these are the maps preserving arbitrary joins, finite meets, and multiplication. The choice of morphisms will be recovered from the noetherian forms themselves.

Proposition 3.1. For every morphism h:L→Mh : L \to M in MCL, postcomposition defines a morphism of noetherian forms

P(h):UL⟶UM,P(h)G(ν)=h∘ν.P(h) : U_L \longrightarrow U_M,\qquad P(h)_G(\nu) = h \circ\nu.

These assignments define a functor

P:MCL⟶NFc(Grp).P : \mathrm{MCL} \longrightarrow\mathrm{NF}^{\mathrm{c}}(\mathrm{Grp}).

Every P(h)GP(h)_G also preserves arbitrary fibre joins.

Proof. Preservation of 00 and addition, together with monotonicity, ensures that hνh\nu is a norm. Postcomposition preserves nonexpansive homomorphisms and commutes with inverse image.

Since hh preserves arbitrary meets, it has a left adjoint α:M→L\alpha: M \to L:

α(b)≤a⟺b≤h(a).\alpha(b) \leq a \quad\Longleftrightarrow\quad b \leq h(a).

The map α\alpha preserves 00 and is subadditive. Indeed, the unit inequalities give

b+c≤h(αb)+h(αc)=h(αb+αc),b+c \leq h(\alpha b)+h(\alpha c)=h(\alpha b+\alpha c),

and hence

α(b+c)≤αb+αc.\alpha(b+c)\leq\alpha b+\alpha c.

Consequently αμ\alpha\mu is a norm whenever μ\mu is one. Writing ΦG(ν)=hν\Phi_G(\nu)=h\nu and AG(μ)=αμA_G(\mu)=\alpha\mu, the pointwise adjunction becomes

ΦG(ν)⪯μ⟺μ≤hν⟺αμ≤ν⟺ν⪯AG(μ).\Phi_G(\nu)\preceq\mu\quad\Longleftrightarrow\quad\mu\leq h\nu\quad\Longleftrightarrow\quad\alpha\mu\leq\nu\quad\Longleftrightarrow\quad\nu\preceq A_G(\mu).

Thus ΦG⊣AG\Phi_G \dashv A_G in fibre order, so ΦG\Phi_G preserves arbitrary fibre joins. Finite fibre meets are pointwise finite coefficient joins, which hh preserves. Both fibre bounds are preserved because h(0)=0h(0)=0 and h(∞)=∞h(\infty)=\infty.

Postcomposition with α\alpha commutes with inverse image:

AGfM∗=fL∗AH.A_G f_M^* = f_L^* A_H.

Uniqueness of left adjoints therefore gives

ΦHfL!=fM!ΦG.\Phi_H f_L^! = f_M^! \Phi_G.

This establishes compatibility with direct image. Identities and composition are immediate. □

Lemma 3.2. Let T:UL→UMT:U_L\to U_M be a morphism of noetherian forms. There is a unique bounded lattice homomorphism h:L→Mh:L\to M such that

(TGν)(g)=h(ν(g))(19)(T_G\nu)(g)=h(\nu(g)) \tag*{(19)}

for every group GG, norm ν\nu, and element gg.

Proof. By (17), the fibre map TC2:Lop→MopT_{C_2}:L^{\mathrm{op}}\to M^{\mathrm{op}} determines a bounded lattice homomorphism h:L→Mh:L\to M. We prove that it determines every component.

Let η:Z→L\eta:\mathbb{Z}\to L be a norm. Write the infinite dihedral group additively as

D∞=Z⋊C2,(n,ε)+(m,δ)=(n+(−1)εm,ε+δ).D_\infty=\mathbb{Z}\rtimes C_2,\qquad(n,\varepsilon)+(m,\delta)=\left(n+(-1)^\varepsilon m,\varepsilon+\delta\right).

Here ε,δ∈{0,1}\varepsilon,\delta\in\{0,1\} and their sum is modulo 2. Put

r=(1,0),s=(0,1),λ(n,ε)=η(n).r=(1,0),\qquad s=(0,1),\qquad\lambda(n,\varepsilon)=\eta(n).

Symmetry and the triangle inequality for η\eta show that λ\lambda is a norm. Both ss and r+sr+s have order 2. Restricting to the two corresponding homomorphisms from C2C_2 gives

(TD∞λ)(s)=h(0)=0,(TD∞λ)(r+s)=h(η(1)).(T_{D_\infty}\lambda)(s)=h(0)=0,\qquad(T_{D_\infty}\lambda)(r+s)=h(\eta(1)).

For any norm ω\omega vanishing at ss, the triangle inequality and s+s=0s+s=0 imply

ω(r+s)≤ω(r),ω(r)≤ω(r+s).\omega(r+s)\leq\omega(r),\qquad\omega(r)\leq\omega(r+s).

Thus (TD∞λ)(r)=h(η(1))(T_{D_\infty}\lambda)(r)=h(\eta(1)). Pulling back along Z→D∞\mathbb{Z}\to D_\infty, n↦(n,0)n\mapsto(n,0), yields

(TZη)(1)=h(η(1)).(T_{\mathbb{Z}}\eta)(1)=h(\eta(1)).

Finally, pull back an arbitrary ν\nu along ug:Z→Gu_g:\mathbb{Z}\to G, 1↦g1\mapsto g. Naturality of inverse image gives

(TGν)(g)=(TZ(ug∗ν))(1)=h(ν(g)).(T_G\nu)(g)=(T_{\mathbb{Z}}(u_g^*\nu))(1)=h(\nu(g)).

Uniqueness already follows from the fibre over C2C_2.

Lemma 3.3. The map hh in Lemma 3.2 preserves coefficient addition.

Proof. Let V=C2⊕C2={0,x,y,x+y}V=C_2\oplus C_2=\{0,x,y,x+y\}. For a,b∈La,b\in L, list values at these four elements and consider the norms

A=(0,a,∞,∞),B=(0,∞,b,∞).A=(0,a,\infty,\infty),\qquad B=(0,\infty,b,\infty).

Their fibre join is

A∨FB=(0,a,b,a+b)(20)A\vee_F B=(0,a,b,a+b) \tag*{(20)}

The displayed tuple is a norm. Its triangle inequalities follow from commutativity, associativity and (7); for example, the only bound involving both independent generators is equality at x+yx+y, and a≤b+(a+b)a\leq b+(a+b) follows from monotonicity. Any norm pointwise below both AA and BB has values at most aa, bb, a+ba+b at xx, yy, x+yx+y, respectively. This proves (20).

Apply TVT_V, which preserves this fibre join, and use (19). Evaluation at x+yx+y gives

h(a+b)=h(a)+h(b).h(a+b)=h(a)+h(b).

Lemma 3.4. The map hh in Lemma 3.2 preserves arbitrary meets.

Proof. Let (ai)i∈I(a_i)_{i\in I} be a family in LL, and identify elements of EI=⨁i∈IC2E_I=\bigoplus_{i\in I} C_2 with finite subsets J⊆IJ\subseteq I, using symmetric difference as addition. Define

νI(J)=∑j∈Jaj,νI(∅)=0.\nu_I(J)=\sum_{j\in J}a_j,\qquad\nu_I(\varnothing)=0.

This is a norm. Passing from the concatenated sums for JJ and KK to the sum for J△KJ\mathbin{\triangle}K removes duplicate nonnegative summands, and hence cannot increase the value. Let εI:EI→C2\varepsilon_I:E_I\to C_2 be parity. Direct image into C2C_2 requires no correction of the raw fibre meet: every value at its nonzero element tt, together with value 00 at 00, defines a norm. Thus

(εI!νI)(t)=⋀J⊆I finite∣J∣ odd∑j∈Jaj=⋀i∈Iai.(21)(\varepsilon_I!\nu_I)(t)=\bigwedge_{\substack{J\subseteq I\text{ finite}\\ |J|\text{ odd}}}\sum_{j\in J}a_j=\bigwedge_{i\in I}a_i. \tag*{(21)}

Every nonempty sum is above each of its summands, while singleton subsets give the converse bound. If II is empty, both sides are ∞\infty.

By Lemma 3.3, hh preserves finite sums. Using (19), commutation with direct image, and (21), we obtain

h(⋀iai)=(TC2(εI!νI))(t)=(εI!(TEIνI))(t)=⋀J⊆I finite∣J∣ odd∑j∈Jh(aj)=⋀ih(ai).\begin{aligned} h\left(\bigwedge_i a_i\right) &=(T_{C_2}(\varepsilon_I!\nu_I))(t)\\ &=(\varepsilon_I!(T_{E_I}\nu_I))(t)\\ &=\bigwedge_{\substack{J\subseteq I\text{ finite}\\ |J|\text{ odd}}}\sum_{j\in J}h(a_j) =\bigwedge_i h(a_i). \end{aligned}

Theorem 3.5. The functor

P:MCL⟶NFc(Grp)P:\mathrm{MCL}\longrightarrow\mathrm{NF}^{c}(\mathrm{Grp})

is full and faithful. Explicitly,

Hom⁡MCL(L,M)≅Hom⁡NF(Grp)(UL,UM),h↦(ν↦hν).\operatorname{Hom}_{\mathrm{MCL}}(L,M)\cong\operatorname{Hom}_{\mathrm{NF}(\mathrm{Grp})}(U_L,U_M),\qquad h\mapsto(\nu\mapsto h\nu).

Proof. Proposition 3.1 constructs a morphism of noetherian forms from each coefficient morphism. Lemmas 3.2, 3.3 and 3.4 recover a unique coefficient morphism from every morphism of forms. □

Corollary 3.6. Let CLat∧,∨fin≠1\mathrm{CLat}^{\ne1}_{\wedge,\vee_{\mathrm{fin}}} be the category of nontrivial complete lattices and maps preserving arbitrary meets and finite joins. The assignment

L↦U(L,+=∨)L\mapsto U_{(L,+=\vee)}

is full and faithful. Equivalently, for the category CLat∨,∧fin≠1\mathrm{CLat}^{\ne1}_{\vee,\wedge_{\mathrm{fin}}} of nontrivial complete lattices and maps preserving arbitrary joins and finite meets, the assignment

A↦U(Aop,+=∨Aop))A\mapsto U_{(A^{\mathrm{op}},+=\vee_{A^{\mathrm{op}}}))}

is full and faithful.

Proof. The idempotent objects of MCL are exactly complete lattices with addition equal to join. Preservation of finite joins already ensures preservation of this addition, so these objects form the stated full subcategory. The second assertion follows by order duality. □

Corollary 3.7. The forms ULU_L and UMU_M are isomorphic over Grp if and only if LL and MM are isomorphic as multiplicative complete lattices. These forms have arbitrarily large fibres already over C2C_2.

Proof. Full faithfulness reflects and preserves isomorphisms, and (17) identifies the fibre over C2C_2. □\square

In particular the underlying complete lattice alone does not determine the form in the general construction. On [0,∞][0,\infty], ordinary addition and maximum yield nonisomorphic forms: the second addition is idempotent, while the first is not, so their coefficient structures cannot be isomorphic.

Remark 3.8. Although morphisms of noetherian forms need only preserve finite fibre operations, every morphism between the forms ULU_L preserves arbitrary fibre joins, by Theorem 3.5 and Proposition 3.1. It need not preserve arbitrary fibre meets. For example, give [0,∞][0,\infty] coefficient addition a+b=max⁡(a,b)a+b=\max(a,b), and let 2\mathbf{2} have the same addition. The map

h:[0,∞]⟶2,h(t)={0,t<∞,∞,t=∞,h : [0,\infty] \longrightarrow\mathbf{2}, \qquad h(t)=\begin{cases}0, & t<\infty,\\ \infty, & t=\infty,\end{cases}

preserves arbitrary meets, finite joins and addition. It fails to preserve the supremum of the sequence 1,2,3,…1,2,3,\ldots. Since fibre meets over C2C_2 are coefficient joins, the induced morphism fails to preserve arbitrary fibre meets.

The unique optimal form over groups

We now use full operators in the weaker sense introduced in Section 1. Following [10], a noetherian form is optimal if it is isomorphic to a subform of every noetherian form over the same base inducing the same associated projection–embedding factorization system. The subgroup form is optimal by [10]. By Proposition 1.5, all noetherian forms over Grp induce the same factorization system.

Lemma 4.1 (Rigidity of the subgroup form). Every full operator T:S→ST : \mathbf{S} \to\mathbf{S} over Grp is the identity.

Proof. Fullness says that, for every f:G→Kf : G \to K, H≤GH \leq G and J≤KJ \leq K,

f(H)≤J⟺f(TG(H))≤TK(J).(22)f(H) \leq J \Longleftrightarrow f(T_G(H)) \leq T_K(J). \tag*{(22)}

In particular each TGT_G is an order embedding. For every prime pp, the fibre over CpC_p is the two-element chain. Hence TCpT_{C_p} fixes both subgroups.

Write TZ(Z)=nZT_{\mathbb{Z}}(\mathbb{Z})=n\mathbb{Z} with n≥0n \geq0. For the quotient qp:Z→Cpq_p : \mathbb{Z} \to C_p, equation (22), applied to Z\mathbb{Z} and the trivial subgroup of CpC_p, implies

qp(nZ)≰0.q_p(n\mathbb{Z}) \nleq0.

Thus n≠0n \ne0 and no prime divides nn, so n=1n=1.

For any g∈Gg \in G, let ug:Z→Gu_g : \mathbb{Z} \to G send 11 to gg. For every H≤GH \leq G,

g∈H⟺ug(Z)≤Hg \in H \Longleftrightarrow u_g(\mathbb{Z}) \leq H
⟺ug(TZ(Z))≤TG(H)\Longleftrightarrow u_g(T_{\mathbb{Z}}(\mathbb{Z})) \leq T_G(H)
⟺g∈TG(H).\Longleftrightarrow g \in T_G(H).

Therefore TG(H)=HT_G(H)=H for every subgroup of every group. Since the subgroup form is faithful, TT is also the identity on arrows. □\square

Theorem 4.2. The subgroup form is the only optimal noetherian form over Grp, up to a unique isomorphism over Grp.

Proof. Let FF be optimal. Optimality of FF and of SS gives full operators

S→aF→bS.S \xrightarrow{a} F \xrightarrow{b} S.

Lemma 4.1 gives ba=id⁡Sba = \operatorname{id}_S. Since a full operator between forms is injective on each fibre, the equality b(ab)=bb(ab) = b implies ab=id⁡Fab = \operatorname{id}_F. Thus a,ba,b are inverse isomorphisms. Any two isomorphisms F→SF \to S differ by an automorphism of SS, which is the identity by the same lemma.

Corollary 4.3. For a noetherian form FF over Grp, the following conditions are equivalent:

  1. FF is optimal;

  2. FF is isomorphic to the subgroup form;

  3. every cluster of FF is conormal.

Proof. The first two conditions are equivalent by Theorem 4.2; the subgroup form is conormal. If every cluster of FF is conormal, associated embeddings identify its clusters with ordinary subgroups by Proposition 1.5. More explicitly, H≤GH \le G corresponds to Im⁡F(H↪G)\operatorname{Im}_F(H \hookrightarrow G). The universal property of associated embeddings identifies order and the relation along every homomorphism with those of the subgroup form. This is an isomorphism of forms.

Corollary 4.4. For a nontrivial multiplicative complete lattice LL in additive convention, the form ULU_L is optimal if and only if L≅2L \cong2 as a coefficient structure.

Proof. If ULU_L is optimal, then it is isomorphic to SS. Evaluating at C2C_2 gives Lop≅2L^{\mathrm{op}} \cong2. On a two-element coefficient lattice addition is forced to be join, so the converse follows from Proposition 2.7.

Remark 4.5. Theorem 4.2 is stronger than the existence of full operators in both directions: the rigidity lemma is what makes those operators inverse. It also implies that the subgroup form has no proper subform which is itself a noetherian form over groups.

Optimality does not assert initiality in NF(Grp). The operators appearing in its definition need not preserve bounds or image operations. For instance,

D(G)=Sub⁡(G)×2,D(G) = \operatorname{Sub}(G) \times2,

with both transports acting identically on the second coordinate, is a noetherian form. Its kernels are (ker⁡f,0)(\ker f, 0), its images are (f(G),∞)(f(G), \infty), and the usual group factorizations verify the axioms. There can be no morphism S→DS \to D in NF(Grp): at the trivial group such a morphism would have to be a bounded lattice homomorphism from the singleton lattice to 22.

Future directions

Tensor product of lattices. The full embedding includes both ordinary extended norms and the idempotent coefficient case. The latter retains a tensor description based solely on the subgroup form. Let Sup denote the category of complete lattices and arbitrary-join preserving maps. Its tensor product A⊗SupBA \otimes_{\mathbf{Sup}} B represents maps A×B→CA \times B \to C preserving arbitrary joins separately in each variable; see [9].

Suppose coefficient addition on LL is lattice join. The sublevel subgroups of ν∈Norm⁡L(G)\nu\in\operatorname{Norm}_{L}(G) define a map

Hν:L⟶Sub⁡(G),Hν(a)={g:ν(g)≤a},H_{\nu}: L \longrightarrow\operatorname{Sub}(G), \qquad H_{\nu}(a)=\{g:\nu(g)\leq a\},

with

Hν(⋀iai)=⋂iHν(ai).H_{\nu}\left(\bigwedge_i a_i\right)=\bigcap_i H_{\nu}(a_i).

Conversely a map preserving arbitrary meets in this way comes from a unique norm:

ν(g)=⋀{a:g∈H(a)}.\nu(g)=\bigwedge\{a:g\in H(a)\}.

Preservation of arbitrary meets ensures that the displayed infimum is an attained level; the subgroup condition yields the join triangle inequality. In the join-reversing description of the tensor product this gives the fibrewise lattice isomorphism

Norm⁡L(G)≅Lop⊗Sup⁡Sub⁡(G).(23)\operatorname{Norm}_{L}(G)\cong L^{\mathrm{op}}\otimes_{\operatorname{Sup}}\operatorname{Sub}(G). \tag*{(23)}

For a fibre-complete noetherian form FF over another base category, the corresponding candidate for tensoring with a complete lattice AA has fibres

(A⊙F)−1(X)=A⊗Sup⁡F−1(X),(A\mathbin{\odot}F)^{-1}(X)=A\otimes_{\operatorname{Sup}}F^{-1}(X),

with direct images induced by 1A⊗f!1_A\otimes f_!. A systematic study should determine the hypotheses under which these data satisfy all noetherian axioms, and the appropriate class of morphisms for which tensoring is functorial. These are separate questions: the tensor product is defined using arbitrary joins, whereas morphisms of noetherian forms need not preserve arbitrary joins. The full embedding established here does not, on its own, provide an action on the entire category of noetherian forms.

For general coefficient addition, Proposition 2.10 identifies the precise obstruction to the preceding level-set description. Only idempotent thresholds are guaranteed to define subgroups. Consequently (23) must be restricted to the idempotent case. A tensor construction for the full class of norms would have to retain coefficient addition as part of its data: the underlying complete lattice alone does not determine the form, as Corollary 3.7 shows. This issue is independent of the closure needed for direct images; Proposition 2.8 removes that closure for quantale coefficients, including ordinary extended real-valued norms, without making all sublevel sets subgroups.

Beyond groups. For other group-like structures, one may impose analogous inequalities involving their operations and coefficient addition. The questions are then whether the resulting forgetful functors are noetherian and whether the coefficient structures can be reconstructed from morphisms of forms. Our reconstruction uses C2C_2, the infinite dihedral group, C2⊕C2C_2\oplus C_2, and parity maps. Suitable replacements and the additional properties of the starting form must be identified for each proposed generalization. These extensions and a general tensor formulation are left for future work.

Arbitrary pseudometrics. A further direction is to replace norms by arbitrary LL-valued pseudometrics on the underlying group. These are functions d:G×G→Ld:G\times G\to L satisfying

d(x,x)=0,d(x,y)=d(y,x),d(x,z)≤d(x,y)+d(y,z).d(x,x)=0,\qquad d(x,y)=d(y,x),\qquad d(x,z)\leq d(x,y)+d(y,z).

No separation, finiteness or compatibility with the group operation is imposed. With nonexpansive group homomorphisms they form a category PMetGrpL\mathbf{PMetGrp}_{L}, whose forgetful functor we denote by VLV_L. Norms identify with the left-translation-invariant pseudometrics through

dν(x,y)=ν(−x+y),ν(g)=dν(0,g).d_{\nu}(x,y)=\nu(-x+y), \qquad\nu(g)=d_{\nu}(0,g).

When L=2L=2, arbitrary pseudometrics are equivalence relations: x∼yx\sim y precisely when d(x,y)=0d(x,y)=0. Thus V2V_2 is the pullback of the equivalence-relation form on pointed sets along Grp→Set∗\mathbf{Grp}\to\mathbf{Set}_*, as in van Niekerk’s construction [13]. It differs from U2U_2, the subgroup form.

The basic construction extends beyond the invariant case. Pointwise suprema of pseudometrics are pseudometrics, so the reverse pointwise fibres are complete. Let cGmet(u)c_G^{\mathrm{met}}(u) denote the greatest pseudometric pointwise below a function u:G×G→Lu:G\times G\to L. Then

f∗e(x,y)=e(fx,fy),f!d=cHmet(rfd),(rfd)(a,b)=⋀fx=afy=bd(x,y).f^*e(x,y)=e(fx,fy), \qquad f!d=c_H^{\mathrm{met}}(r_fd), \qquad(r_fd)(a,b)=\bigwedge_{\substack{fx=a\\fy=b}}d(x,y).

These operations make VLV_L noetherian. To see the first identity in (N1), a pseudometric zero on every fibre of ff is constant on pairs of fibres and hence descends to f(G)f(G). The second identity follows by extending a pseudometric on f(G)f(G) with distance ∞\infty between distinct points whenever at least one is outside f(G)f(G). Kernels and images are the 0,∞0,\infty-valued pseudometrics whose zero relations are, respectively,

Rf={(x,y):fx=fy},ΔH∪(f(G)×f(G)).R_f=\{(x,y):fx=fy\}, \qquad\Delta_H\cup\bigl(f(G)\times f(G)\bigr).

The usual quotient–subgroup factorization gives (N2). Joins of group congruences and intersections of subgroups give (N3). The same argument works over pointed sets, using pointed subsets in place of subgroups. None of these arguments requires distributivity of coefficient addition.

Change of coefficients also extends: the maps in MCL\mathbf{MCL} act by d↦h∘dd\mapsto h\circ d, giving a faithful functor

Q:MCL⟶NFc(Grp),Q(L)=VL.Q:\mathbf{MCL}\longrightarrow\mathbf{NF}^{c}(\mathbf{Grp}), \qquad Q(L)=V_L.

Indeed, the left adjoint of hh preserves 00 and is subadditive, so the proof of Proposition 3.1 applies to pseudometrics. Faithfulness follows from the fibre over C2C_2, which is LopL^{\mathrm{op}}. In the idempotent case, sublevel equivalence relations give the analogue of (23):

VL−1(G)≅Lop⊗SupEq⁡(∣G∣).V_L^{-1}(G)\cong L^{\mathrm{op}}\otimes_{\mathrm{Sup}}\operatorname{Eq}(|G|).

These observations suggest studying pseudometrics as a common extension of the norm and equivalence-relation constructions.

Several stronger conclusions do not extend. For C4→C2C_4\to C_2, the kernel has zero classes {0,2}\{0,2\} and {1,3}\{1,3\}, whereas a conormal cluster has at most one nonsingleton zero class. Thus normal clusters need not be conormal. Moreover, QQ is not full: inversion induces an automorphism of forms

JG(d)(x,y)=d(−x,−y).J_G(d)(x,y)=d(-x,-y).

It commutes with inverse images, and hence with direct images, and preserves all fibre operations. It fixes the fibre over C2C_2 pointwise, but on C3C_3 it exchanges the 0,∞0,\infty-valued pseudometrics with zero partitions {0,1}∣{2}\{0,1\}\mid\{2\} and {0,2}∣{1}\{0,2\}\mid\{1\}. Consequently it cannot be induced by a coefficient map. Full faithfulness may therefore require additional structure or a different choice of base category and morphisms.

Even for quantale coefficients, the closure in direct images generally remains necessary. For example, on C6C_6 take the 0,∞0,\infty-valued pseudometric with zero classes {0,1}\{0,1\}, {4,5}\{4,5\}, {2}\{2\} and {3}\{3\}. Along reduction modulo 33, its raw direct image has distances 00 between 0,10,1 and between 1,21,2, but distance ∞\infty between 0,20,2. It fails the triangle inequality. Thus Proposition 2.8 depends on the group-norm setting.

Adjunctions. Passing to arbitrary pseudometrics does not give QQ an adjoint in either direction with the morphisms of Definition 1.3. In fact, there are no morphisms of noetherian forms in either direction between SS and any VLV_L. A morphism S→VLS \to V_L would preserve the equality between the image of C2={0,2}↪C4C_2=\{0,2\}\hookrightarrow C_4 and the kernel of C4→C2C_4 \to C_2. In VLV_L their zero partitions are, respectively,

{0,2}∣{1}∣{3},{0,2}∣{1,3},\{0,2\}\mid\{1\}\mid\{3\},\qquad\{0,2\}\mid\{1,3\},

which are different. Conversely, the fibre of VLV_L over C3C_3 contains the bounded sublattice M3M_3 of 0,∞0,\infty-valued pseudometrics, while S(C3)=2S(C_3)=2. No bounded lattice homomorphism M3→2M_3\to2 exists: among the three atoms, two would have the same value, contradicting preservation of their meet or join. An adjunction A⊣QA\dashv Q would supply the impossible unit S→Q(AS)S\to Q(AS), and Q⊣BQ\dashv B would supply the impossible counit Q(BS)→SQ(BS)\to S. This obstruction persists even if one restricts to forms with singleton fibre over the trivial group. In particular, the forms VLV_L are not optimal, by § 5.

There is, however, a reflection for each fixed coefficient lattice:

RL:PMetGrpL⇄NGrpL:IL,RL⊣IL,R_L:\mathrm{PMetGrp}_L \rightleftarrows\mathrm{NGrp}_L:I_L,\qquad R_L\dashv I_L,

where IL(G,ν)=(G,dν)I_L(G,\nu)=(G,d_\nu) is fully faithful. Using the norm closure from (11), put

ud(g)=⋀x∈Gd(x,x+g),RL(G,d)=(G,cG(ud)).u_d(g)=\bigwedge_{x\in G}d(x,x+g),\qquad R_L(G,d)=(G,c_G(u_d)).

The corresponding invariant pseudometric is the greatest one pointwise below dd. For a homomorphism f:G→Hf:G\to H and an LL-norm μ\mu on HH, the condition that f:(G,d)→IL(H,μ)f:(G,d)\to I_L(H,\mu) be nonexpansive is equivalent to μ∘f≤ud\mu\circ f\le u_d, and hence to μ∘f≤cG(ud)\mu\circ f\le c_G(u_d), since μ∘f\mu\circ f is a norm. This proves the adjunction. It is an adjunction of the total categories over Grp\mathrm{Grp}, not an adjunction between coefficient lattices and noetherian forms. In particular, ILI_L is not a morphism of noetherian forms: for the above inclusion C2↪C4C_2\hookrightarrow C_4, it sends the norm-form image to the coset pseudometric, rather than the image in VLV_L. A systematic account of this reflection, and of coefficient reconstruction under suitable additional requirements, is left for future work.

(Kishan Dayaram) Department of Mathematical Sciences, Stellenbosch University, Private Bag X1, Matieland 7602, South Africa

(Ronald Mwita Kerario) Department of Mathematical Sciences, Stellenbosch University, Private Bag X1, Matieland 7602, South Africa, African Institute for Mathematical Sciences (AIMS), 6 Melrose Road, Muizenberg, Cape Town 7945, South Africa

(Zurab Janelidze) Department of Mathematical Sciences, Stellenbosch University, National Institute of Theoretical and Computational Science (NITheCS), Private Bag X1, Matieland 7602, South Africa

References

  1. [1]G. C. L. Brümmer, Topological functors and structure functors, in Categorical Topology, E. Binz and H. Herrlich (eds), Lecture Notes in Mathematics 540, Springer, Berlin–Heidelberg, 1976, 109–135. doi:10.1007/BFb0080855.DOI
  2. [2]M. M. Clementino, D. Hofmann and W. Tholen, Cauchy convergence in V-normed categories, Adv. Math. 470 (2025), article 110247. doi:10.1016/j.aim.2025.110247.DOI
  3. [3]K. K. Dayaram, A. Goswami, Z. Janelidze, D. Rodelo and T. Van der Linden, Homological lemmas for (non-abelian) group-like structures by diagram chasing in a self-dual context, Appl. Categ. Struct. 34 (2026), article 32. doi:10.1007/s10485-026-09862-2.DOI
  4. [4]A. Facchini, Algebraic structures from the point of view of complete multiplicative lattices, in Algebra and Coding Theory, Contemp. Math. 785 (2023), 113–131. Preprint: arXiv:2201.03295.arxiv.org/abs/2201.03295
  5. [5]A. Facchini, C. A. Finocchiaro and G. Janelidze, Abstractly constructed prime spectra, Algebra Universalis 83 (2022), article 8. doi:10.1007/s00012-021-00764-z.DOI
  6. [6]A. Goswami and T. Dube, On z-elements of multiplicative lattices, Algebra Universalis 86 (2025), article 4. doi:10.1007/s00012-024-00882-4.DOI
  7. [7]A. Goswami and Z. Janelidze, Duality in non-abelian algebra IV. Duality for groups and a universal isomorphism theorem, Adv. Math. 349 (2019), 781–812. doi:10.1016/j.aim.2019.04.019.DOI
  8. [8]J. A. Goguen, L-fuzzy sets, J. Math. Anal. Appl. 18 (1967), no. 1, 145–174. doi:10.1016/0022-247X(67)90189-8.DOI
  9. [9]J. Gutiérrez García, U. Höhle and T. Kubiak, Tensor products of complete lattices and their application in constructing quantales, Fuzzy Sets and Systems 313 (2017), 43–60. doi:10.1016/j.fss.2016.04.009.DOI
  10. [10]Z. Janelidze and F. van Niekerk, Toposes have an optimal noetherian form, J. Pure Appl. Algebra 230 (2026), no. 10, article 108340. doi:10.1016/j.jpaa.2026.108340. Preprint: arXiv:2304.03814.DOI
  11. [11]D. H. J. Polymath, On the construction of Noetherian forms on groups, Algebra Number Theory 12 (2018), no. 7, 1773–1786. doi:10.2140/ant.2018.12.1773.
  12. [12]A. Rosenfeld, Fuzzy groups, J. Math. Anal. Appl. 35 (1971), no. 3, 512–517. doi:10.1016/0022-247X(71)90199-5.DOI
  13. [13]F. K. van Niekerk, On the construction of Noetherian forms for algebraic structures, Theory Appl. Categ. 39 (2023), no. 8, 189–206. Available online.DOI

Paper details

vlnc:W_dvw7acgf
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents