Adjacency-Like Conditions and Induced Ideal Graphs

Assuming condition (1), if $x\in\mathfrak{M}$ then $\ xJ\subseteq I$ and hence $\mathfrak{M}J\subseteq I$.

On the other hand, if $J\mathfrak{M}\subseteq I$ then for all $x\in\mathfrak{M},\ Jx\subseteq I$ and hence $x\in(I:_{R}J)$. Since we now have that $\mathfrak{M}\subseteq(I:_{R}J)$ and that $\mathfrak{M}$ is maximal and $J\not\subseteq I$, then $\mathfrak{M}=(I:_{R}J)$. ∎

Note that $\mathfrak{M}J\subseteq I$, and hence $\mathfrak{M}\subseteq(I:_{R}J)$. Since $J\not\subseteq I$, we must have equality.

To see that the converse does not hold in general, consider the domain $K[x,y]$, where $K$ is any field, and the ideals $I=(x,xy,y^{2})$ and $J=(x,y)$. Note that $(I:_{R}J)=J$ but $J^{2}\subsetneq I$. ∎

Let $I\subsetneq J$ be adjacent and let $\mathcal{C}=(I:_{R}J)$; we will show that $\mathcal{C}$ is maximal. Since $I$ is strictly contained in $J$, we can find an $x\in J\setminus I$; additionally, we note that $J=(I,x)$ because of the adjacency of $I$ and $J$.

We now choose an arbitrary $z\notin\mathcal{C}$ and note that $zx\notin I$ (indeed, if $zx\in I$ then the fact that $J=(I,x)$ would show that $z\in\mathcal{C}$ which is a contradition). Hence, it must be the case that $J=(I,zx)$, and since $x\in J$, we obtain

Since $(1-rz)$ conducts $x$ to $I$ and $J=(I,x)$, $(1-rz)\in\mathcal{C}$. Therefore $(\mathcal{C},z)=R$ for all $z\notin\mathcal{C}$ and so $\mathcal{C}$ is maximal.

To see that the converse does not hold, we revisit the example of the previous result. Letting $R=K[x,y]$, $A=(x^{2},xy,y^{2}),I=(x,xy,y^{2}),J=(x,y)$. Note that $(A:_{R}J)=J$ but $A$ and $J$ are not adjacent as $A\subsetneq I\subsetneq J$. ∎

Let $x,y\in J\setminus I$. Certainly $xy\in J$. By way of contradiction, we assume that $xy\in I$. Since $I$ and $J$ are adjacent and $x\notin I$, $(x,I)=J$.

Now let $j\in J$ be arbitrary. By the previous remark, we can find $r\in R$ and $i\in I$ such that $j=rx+i$. Multiplying this by $y$ we obtain that $yj=rxy+iy\in I$. Hence $y\in\mathcal{C}\bigcap J=I$ which is the desired contradiction. ∎

We will assume that $I\bigcap A$ and $J\bigcap A$ are distinct and suppose that $x\in(J\bigcap A)\setminus(I\bigcap A)$. Since $x\notin I$, it must be the case that $J=(I,x)$.

Let $j\in J\bigcap A$ be arbitrary. Since $J=(I,x)$, we have that

Suppose that $A$ is an ideal with $(I,x)\subsetneq A\subseteq(J,x)$ and let $a\in A\setminus(I,x)$. We write $a=j+rx$ with $j\in J\setminus I$. Since $I$ and $J$ are adjacent and $j\notin I$, $(I,j)=J$. As $A$ contains $x$, $A$ contains all of $J$; we conclude that $A=(J,x)$ and the proof is complete. ∎

Since $I\subsetneq J$ are adjacent, there is an $x\in J\setminus I$ such that $J=(I,x)$ (in fact, any $x\in J\setminus I$ will do). Suppose that $I_{S}\subsetneq J_{S}$ and let $\frac{b}{s}\in J_{S}\setminus I_{S}$. We can assume that $b\in J\setminus I$ and since $I\subsetneq J$ are adjacent, we can find an $r\in R$ and $\alpha\in I$ such that $b=rx+\alpha$. In $R_{S}$ we have the equation

Suppose first that $R$ is Noetherian. If $I\subseteq R$ is a nonzero ideal, then $R/I$ is Noetherian of dimension $0$ and hence is Artinian.

Now suppose that $R/I$ is Artinian for each nonzero ideal $I\subset R$. It suffices to show that every ideal of $R$ is finitely generated. Let $J\subset R$ be an arbitrary nonzero ideal. Let $0\neq x\in J$ and note that by hypothesis, $R/(x)$ is Artinian and so $J/(x)$ is finitely generated. From Lemma 3.9 it follows that $J$ is finitely generated and hence $R$ is Noetherian. ∎

The case for $GA_{0}(-)$ follows immediately from Proposition 3.7 and Proposition 3.8.

For $GA_{1}(-)$ we suppose that $A,B\subset R$ are distinct and have an edge between them. Then $A=B\mathfrak{M}$, with $\mathfrak{M}$ maximal. Note that if $\mathfrak{M}\bigcap S\neq\emptyset$ then $\mathfrak{M}_{S}=R_{S}$ and if not then $\mathfrak{M}_{S}\subset R_{S}$ is maximal. As $A_{S}=B_{S}\mathfrak{M}_{S}$, we have that $A_{S}$ and $B_{S}$ are either equal or have an edge between them. Since all ideals of $R_{S}$ are of this form, if $GA_{1}(R)$ is either connected or complete, then so if $GA_{1}(R_{S})$. Additionally, if $I\subseteq A,B$ then $A/I=(B/I)(\mathfrak{M}/I)$ and from this the result follows for $GA_{1}(R/I)$.

In a similar fashion, if we consider $GA_{2}(-)$ with the ideals/notation in the previous paragraph, we assume that $(A:_{R}B)=\mathfrak{M}$. In this case, it is easy to verify that $(A/I:_{R}B/I)=\mathfrak{M}/I$ and $\mathfrak{M}_{S}\subseteq(A_{S}:_{R}B_{S})$. Hence, again, the result follows for $GA_{2}(-)$. ∎

Using the notation above, we say that the ideal $I_{k}$ is a hinge ideal if $I_{k}$ either properly contains both $I_{k-1}$ and $I_{k+1}$ or is properly contained in them both. We proceed by induction on $m$, the number of hinge ideals between $I$ and $J$.

At the outset, we simplify matters by assuming that all of the ideals $\{I_{s}\}$ are contained in $J$ by intersecting the collection with $J$ and applying Lemma 3.5; we also note that if this chain is increasing, then the conclusion holds. We now describe a reduction process that will greatly simplify the inductive argument.

We first suppose that the first hinge ideal is contained in $I$; that is, we have the decreasing chain of ideals

Because of the adjacency of successive elements of the chain, we can find $x_{s}\in I_{s}\setminus I_{s+1},\ 0\leqslant s\leqslant t-1$ such that

We now make a preliminary refinement of the collection $\{I_{s}\}_{s=0}^{N}$ by letting $I_{s}^{\prime}=(I_{s},x_{0},x_{1},\cdots,x_{t-1})$ for $0\leqslant s\leqslant N$. Applying Lemma 3.6 $t$ times to the entire collection of ideals shows that the new collection $\{I_{s}^{\prime}\}$ begins as

With this in hand, we proceed to argue inductively on the number, $m$, of hinge ideals appearing between $I$ and $J$. If $m=0$ then the chain is increasing and the conclusion is immediate, and this, coupled with the above argument, gives the case $m=1$. We now suppose that the conclusion holds for $m\geqslant 0$ and consider the case of $m+1$ hinge ideals.

In the first case, we assume that the first hinge ideal $I_{t}$ is contained in $I$. The previous argument shows that we can refine so that $I=I_{t}^{\prime}$, and as above, the number of ideals in the new collection (after equal ideals are identified) is nonincreasing and the number of hinge ideals in the new collection is less than or equal to $m$. By induction, we are done in this case.

In the second case, we suppose that the first hinge ideal $I_{t}$ contains $I$. In this case, we consider the collection $\{I_{s}\}_{s=t}^{N}$. Since this collection has $m$ hinge ideals with the first being contained in $I_{t}$, we apply the first case and extract the increasing chain $\{I_{s}^{\prime}\}_{s=0}^{k}$ with $I_{0}^{\prime}=I_{t}$ with $k\leqslant N-t$. Combining this increasing chain with $\{I_{r}\}_{r=0}^{t}$ (and subtracting $1$ for the repetition of $I_{t}$), we obtain our increasing chain with length $t+k+1\leqslant N-t+1+t=N+1$. ∎

If $R$ is Artinian, then every finitely generated $R-$module (in particular, an ideal of $R$) has a composition series. Since every ideal of $R$ is connected to $(0)$, we have that $GA_{0}(R)$ is connected.

On the other hand, if $GA_{0}(R)$ is connected, then there is a (finite) path between $(0)$ and any ideal of $R$. Proposition 3.12 allows this to be refined to a composition series, and hence $R$ is Artinian. ∎

Suppose that $R$ is $1-$dimensional and Noetherian and let $I,J\subseteq R$ be nonzero ideals. To show that $GA^{*}_{0}(R)$ is connected, it suffices to show that $I$ and $I\bigcap J$ can be connected in a finite sequence of steps. To this end, we note that since $I\bigcap J$ is nonzero, the ring $R/(I\bigcap J)$ is $0-$dimensional and Noetherian, and hence Artinian. By Theorem 3.13 there is a finite sequence of adjacent ideals (of the displayed form)

Now we suppose that $GA^{*}_{0}(R)$ is connected. Let $I\subset R$ be an arbitrary nonzero ideal. Proposition 3.8 assures us that adjacency is preserved modulo $I$ and so we obtain that $GA_{0}(R/I)$ is connected. Hence Theorem 3.13 gives us that $R/I$ is Artinian (for any nonzero ideal $I$). From Proposition 3.10 we obtain that $R$ is Noetherian.

To see that $R$ is $1-$dimensional, we suppose that there is a chain of primes $(0)\subsetneq\mathfrak{P}\subsetneq\mathfrak{M}$. If we choose the ideal $I=\mathfrak{P}$, we would have that $\text{dim}(R/\mathfrak{P})\geqslant 1$ and hence $R/\mathfrak{P}$ is not Artinan. We conclude that $GA_{0}(R/\mathfrak{P})$ is not connected and again Proposition 3.8 gives us that $GA^{*}_{0}(R)$ cannot be connected and so $\text{dim}(R)\leqslant 1$. Because of the fact that $R$ is a domain that is not a field, then the dimension of $R$ is precisely $1$. ∎

Since the vertex sets are the same, this is immediate from Lemma 3.2 and Lemma 3.3. ∎

By Proposition 3.15, we have the implication $(1)\Longrightarrow(2)$. For the implication $(2)\Longrightarrow(3)$, we suppose that $GA_{2}(R)$ is connected and that $\mathfrak{M}$ is a maximal ideal of $R$. By assumption, there is a finite path from $\mathfrak{M}$ to the ideal $(0)$:

Since this is a path in the graph $GA_{2}(R)$, then successive ideals must have maximal conductor. We will say that $M_{k}=(I_{k+1}:I_{k})$ if $s_{2a}\leqslant k\leqslant s_{2a+1}$ for $0\leqslant a\leqslant t$.

We first note that

In a similar fashion, we note that

Inductively we obtain

For the implication $(3)\Longrightarrow(1)$, we will assume that there is a collection of maximal ideals $\mathfrak{M}_{i},\ 1\leqslant i\leqslant n$ such that $\mathfrak{M}_{1}\mathfrak{M}_{2}\cdots\mathfrak{M}_{n}=0$. To show that $GA_{1}(R)$ is connected, it suffices to show that if $I\subset R$ is an arbitrary ideal, then there is a finite path to the zero ideal. But note that

The fact that $\text{diam}(GA_{1}(R))\leqslant 2n$ is immediate from the proof of the implication $(3)\Longrightarrow(1)$ in Theorem 3.16. The fact that $\text{diam}(GA_{2}(R))\leqslant 2n$ follows from the fact that $GA_{1}(R)$ is a subgraph of $GA_{2}(R)$ (see Proposition 3.15). ∎

By Theorem 3.13, $GA_{0}(R)$ is connected if and only if $R$ is Artinian, and hence $R$ is $0-$dimensional, and, in this case, semilocal. If $GA_{1}(R)$ or $GA_{2}(R)$ is connected then Theorem 3.16 gives that $\mathfrak{M}_{1}\mathfrak{M}_{2}\cdots\mathfrak{M}_{n}=0$ for (not necessarily distinct) ideals $\{\mathfrak{M}_{1},\mathfrak{M}_{2},\cdots,\mathfrak{M}_{n}\}$. If $\mathfrak{M}$ is an arbitrary maximal ideal, then $\mathfrak{M}\supseteq\mathfrak{M}_{1}\mathfrak{M}_{2}\cdots\mathfrak{M}_{n}$ and hence $\mathfrak{M}=\mathfrak{M}_{k}$ for some $1\leqslant k\leqslant n$ which shows that the list of ideals $\{\mathfrak{M}_{1},\mathfrak{M}_{2},\cdots,\mathfrak{M}_{n}\}$ contains $\text{MaxSpec}(R)$. Hence $R$ is semiquasilocal.

To see that $R$ is $0-$dimensional, we appeal once again to the fact that

Recalling that this collection of maximal ideals contains $\text{MaxSpec}(R)$, we suppose that we can find a prime ideal $\mathfrak{P}$ such that $\mathfrak{M}_{k}\supsetneq\mathfrak{P}$. Since $\mathfrak{P}\supseteq\mathfrak{M}_{1}\mathfrak{M}_{2}\cdots\mathfrak{M}_{n}=0$, we must have that $\mathfrak{P}\supseteq\mathfrak{M}_{i}$ for some $1\leqslant i\leqslant n$. Hence $\mathfrak{M}_{i}\subsetneq\mathfrak{M}_{k}$ which is our desired contradiction. ∎

Combine the results from Proposition 3.15 and Theorem 3.16. ∎

We know that (2) and (3) are equivalent under any conditions and that condition (1) implies (2) and (3). Suppose that $R$ is Noetherian and any one of the conditions hold. By Corollary 3.18, $R$ is $0-$dimensional. So if we add “Noetherian” as a hypothesis, then $R$ is Artinian and all three conditions hold. ∎

For this proof, we discard the case where $R$ is a field as all of the conditions are satisfied vacuously. Since any complete graph is connected, $(2)\Longrightarrow(1)$ is immediate. For the implication $(1)\Longrightarrow(3)$, we let $I\subset R$ be an arbitrary ideal and $R$ the unit ideal. Since $GCl(R)$ is connected, there is a sequence of ideals connecting $R$ and $I$:

Since the edges above are in the graph $GCl(R)$, we must have, for all $1\leqslant i\leqslant n$, nonzero $k_{i}\in K$ such that $J_{i-1}=k_{i}J_{i}$. Note that $R=k_{1}k_{2}\cdots k_{n}I$ and hence, $I$ is principal.

Finally for the implication $(3)\Longrightarrow(2)$, we let $I=aR$ and $J=bR$ be two arbitrary ideals of $R$ (with $a,b\neq 0$). Note that $I=\frac{a}{b}J$ and hence $GCl(R)$ is complete. ∎

In this proof, we will ignore the simple case where $R$ is a field. For the forward implication, we will assume that $GCl_{int}(R)$ is complete. As $GCl_{int}(R)$ is a subgraph of $GCl(R)$, $GCl(R)$ must also be complete. By Theorem 4.1, $R$ must be a PID. It now suffices to show that $R$ is local. To this end, suppose that $\mathfrak{M}_{1}\neq\mathfrak{M}_{2}$ are maximal ideals. Without loss of generality, there is a nonzero $a\in R$ such that $\mathfrak{M}_{1}=a\mathfrak{M}_{2}$. If $a$ is a unit then $\mathfrak{M}_{1}=\mathfrak{M}_{2}$ and if $a$ is a nonunit then $\mathfrak{M}_{1}=a\mathfrak{M}_{2}\subsetneq\mathfrak{M}_{2}$; either gives a contradiction. Hence $R$ is a local PID and hence a Noetherian valuation domain.

On the other hand, if $R$ is a Noetherian valuation domain then any two nonzero proper ideals are of the form $(\pi^{n})$ and $(\pi^{m})$ where $\pi$ is a generator of the maximal ideal and $n,m\geqslant 1$. If we say (without loss of generality) that $n\leqslant m$ then $\pi^{m-n}(\pi^{n})=(\pi^{m})$ and hence $GCl_{int}(R)$ is complete. ∎

As $GCl_{int}(R)$ is a subgraph of $GCl(R)$, the fact that $GCl_{int}(R)$ is connected implies that $GCl(R)$ is connected. Hence, by Theorem 4.1, $R$ must be a PID.

On the other hand, if $R$ is a PID and $I=aR$ and $J=bR$ ($a,b\neq 0$) are arbitrary ideals, then we can connect $I$ and $J$ as follows:

The above demonstrates the veracity of the remark that $\text{diam}(GCl_{int}(R))\leqslant 2$. The fact that $\text{diam}(GCl_{int}(R))=1$ precisely when $R$ is local follows from Theorem 4.2. ∎

If $R$ is a Dedekind domain with quotient field $K$, then two ideals, $I$ and $J$, are in the same class of $\text{Cl}(R)$ if and only if $I=Jk$ for some nonzero $k\in K$. Hence each connected component of $GCl(R)$ is complete and these components are in one to one correspondence with the elements of $\text{Cl}(R)$.

For the $GCl_{int}(R)$ case, we first note that if there is a path from $I$ to $J$ then there must be some nonzero $k\in K$ such that $I=Jk$, so it remains to show that if $I$ and $J$ are in the same ideal class, then there is a path connecting them. To this end, we note that if $I=Jk$ for some nonzero $k\in K$, we can write $I=\frac{a}{b}J$ and in a similar fashion as before, we can connect $I$ and $J$ via

For the final statement, we first suppose that $R$ is a local PID (and hence a Noetherian valuation domain) with uniformizer $\pi$. If $I$ and $J$ are in the same class of $\text{Cl}(R)$ then $I=kJ$ for some nonzero $k\in K$. Since $K$ is the quotient field of a Noetherian valuation domain, we can write $k=\pi^{n}$ with $n\in\mathbb{Z}$ up to a unit in $R$. If $n\geqslant 0$ then $I=J$ or there is an edge between them. If $n<0$ then $J=\pi^{-n}I$ and again there is an edge between $I$ and $J$.