This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

\catchline

TWO-SIDED IDEALS IN LEAVITT PATH ALGEBRAS

PINAR COLAK ppekcagl@sfu.ca Department Of Mathematics, Simon Fraser University
8888 University Drive, Burnaby, BC V5A 1S6, Canada
((Day Month Year); (Day Month Year); (Day Month Year))
Abstract

We explicitly describe two-sided ideals in Leavitt path algebras associated with a row-finite graph. Our main result is that any two-sided ideal II of a Leavitt path algebra associated with a row-finite graph is generated by elements of the form v+i=1nλigiv+\sum_{i=1}^{n}\lambda_{i}g^{i}, where gg is a cycle based at vertex vv. We use this result to show that a Leavitt path algebra is two-sided Noetherian if and only if the ascending chain condition holds for hereditary and saturated closures of the subsets of the vertices of the row-finite graph EE.

keywords:
Leavitt path algebra; two-sided Noetherian; two-sided ideal
{history}\comby

(xxxxxxxxx)

\ccode

2000 Mathematics Subject Classification: 16D70


Throughout this paper KK denotes a field. For a directed graph EE, the Leavitt path algebra LK(E)L_{K}(E) of EE with coefficients in KK has received much recent attention, see e.g. [1], [5], [6]. The two-sided ideal structure of LK(E)L_{K}(E) has been an important focus of much of this work. In this paper we provide an explicit description of a generating set for any two-sided ideal of LK(E)L_{K}(E), where EE is any row-finite graph. We then use this description to identify those row-finite graphs EE for which LK(E)L_{K}(E) is two-sided Noetherian.

We briefly recall the basic definitions.

A directed graph E=(E0,E1,r,s)E=(E^{0},E^{1},r,s) consists of two sets E0E^{0}, E1E^{1} and functions r,s:E1E0.r,s:E^{1}\rightarrow E^{0}. The elements of E0E^{0} are called vertices and elements of E1E^{1} are called edges. For each eE1e\in E^{1}, r(e)r(e) is the range of ee and s(e)s(e) is the source of ee. If r(e)=vr(e)=v and s(e)=ws(e)=w, then we say that vv emits ee and that ww receives ee. A vertex which emits no edges is called a sink. A graph is called row-finite if every vertex is the source of at most finitely many edges.

A path μ\mu in a graph EE is a sequence of edges μ=e1en\mu=e{{}_{1}}\cdots e{{}_{n}} such that r(e)i=s(e)i+1r(e{{}_{i}})=s(e{{}_{i+1}}) for i=1,,n1i=1,\dots,n-1. We define the source of μ\mu by s(μ):=s(e)1s(\mu):=s(e{{}_{1}}) and the range of μ\mu by r(μ):=r(e)nr(\mu):=r(e{{}_{n}}). If we have r(μ)=s(μ)r(\mu)=s(\mu) and s(e)is(ei+1)s(e{{}_{i}})\neq s(e_{i+1}) for every iji\neq j, then μ\mu is called a cycle. A closed path based at vv is a path μ=e1en\mu=e_{1}\cdots e_{n}, with ejE1e_{j}\in E^{1}, n1n\geq 1 and such that s(μ)=r(μ)=vs(\mu)=r(\mu)=v. We denote the set of all such paths by CP(v)CP(v). A closed simple path based at vv is a closed path based at vv, μ=e1en\mu=e_{1}\dots e_{n}, such that s(ej)vs(e_{j})\neq v for j>1j>1. We denote the set of all such paths by CSP(v)CSP(v). Note that a cycle is a closed simple path based at any of its vertices. However the converse may not be true, as a closed simple path based at vv may visit some of its vertices (but not vv) more than once.

Let vv be a vertex in E0E^{0}. If there is no cycle based at vv, then we let g=vg=v and call it the trivial cycle. If gg is a cycle based at vv of length at least 1, then gg is called a non-trivial cycle.

Let E=(E0,E1)E=(E^{0},E^{1}) be any directed graph, and let KK be a field. We define the Leavitt path KK-algebra LK(E)L_{K}(E) associated with EE as the KK-algebra generated by a set vE0{v\in E^{0}} together with a set {e,e|eE1}\{e,e^{*}|e\in E^{1}\}, which satisfy the following relations:

  1. 1.

    vv=δv,vvvv^{\prime}=\delta_{v,v^{\prime}}v for all v,vE0v,v^{\prime}\in E^{0}.

  2. 2.

    s(e)e=er(e)=es(e)e=er(e)=e for all eE1e\in E^{1}.

  3. 3.

    r(e)e=es(e)=er(e)e^{*}=e^{*}s(e)=e^{*} for all eE1e\in E^{1}.

  4. 4.

    ef=δe,fr(e)e^{*}f=\delta_{e,f}r(e) for all e,fE1e,f\in E^{1}.

  5. 5.

    v={eE1|s(e)=v}eev=\sum_{\{e\in E^{1}|s(e)=v\}}ee^{*} for every vE0v\in E^{0} such that 0<|s1(v)|<0<|s^{-1}(v)|<\infty.

The elements of E1E^{1} are called real edges, while for eE1e\in E^{1} we call ee^{*} a ghost edge. The set {e|eE1}\{e^{*}|e\in E^{1}\} is denoted by (E1)(E^{1})^{*}. We let r(e)r(e^{*}) denote s(e)s(e), and we let s(e)s(e^{*}) denote r(e)r(e). We say that a path in LK(E)L_{K}(E) is a real path (resp., a ghost path) if it contains no terms of the form eie^{*}_{i} (resp., eie_{i}). We say that pLK(E)p\in L_{K}(E) is a polynomial in only real edges (resp., in only ghost edges) if it is a sum of real paths (resp., ghost edges). The length of a real path (resp., ghost path) μ\mu, denoted by |μ||\mu|, is the number of edges it contains. The length of vE0v\in E^{0} is 0. Let xx be a polynomial in only real edges (resp., in only ghost edges) in LK(E)L_{K}(E). If x=μ1++μnx=\mu_{1}+\cdots+\mu_{n}, where the μi\mu_{i}’s are real paths (resp., ghost paths), then the length of xx, denoted by |x||x|, is defined as maxi=1,,n{|μi|}\mathrm{max}_{i=1,\dots,n}\{|\mu_{i}|\}. An edge ee is called an exit to the path μ=e1en\mu=e_{1}\cdots e_{n} if there exists ii such that s(e)=s(ei)s(e)=s(e_{i}) and eeie\neq e_{i}.

The proofs of the following can be found in [1].

Lemma 1

LK(E)L_{K}(E) is spanned as a KK-vector space by monomials

  1. 1.

    kvikv_{i} with kKk\in K and viE0v_{i}\in E^{0}, or

  2. 2.

    ke1eaf1fbke_{1}\dots e_{a}f^{*}_{1}\dots f^{*}_{b} where kKk\in K; a,b0a,b\geq 0, a+b>0a+b>0, e1,,ea,f1,,fbE1e_{1},\dots,e_{a},f_{1},\dots,f_{b}\in E^{1}.

Lemma 2

If μ,νCSP(v)\mu,\nu\in CSP(v), then μν=δμ,νv\mu^{*}\nu=\delta_{\mu,\nu}v. For every μCP(v)\mu\in CP(v) there exist unique μ1,,μmCSP(v)\mu_{1},\dots,\mu_{m}\in CSP(v) such that μ=μ1μm\mu=\mu_{1}\cdots\mu_{m}.

For a given graph EE we define a preorder \geq on the vertex set E0E^{0} by: vwv\geq w if and only if v=wv=w or there is a path μ\mu such that s(μ)=vs(\mu)=v and r(μ)=wr(\mu)=w. We say that a subset HE0H\subseteq E^{0} is hereditary if wHw\in H and wvw\geq v imply vHv\in H. We say HH is saturated if whenever 0<|s1(v)|<0<|s^{-1}(v)|<\infty and {r(e):s(e)=v}H\{r(e):s(e)=v\}\subseteq H, then vHv\in H. The hereditary saturated closure of a set XE0X\subset E^{0} is defined as the smallest hereditary and saturated subset of E0E^{0} containing XX. For the hereditary saturated closure of XX we use the notation given in [3]: X¯=n=0Λn(X)\overline{X}=\bigcup^{\infty}_{n=0}\Lambda_{n}(X), where

Λ0(X):={vE0|xvforsomexX},andforn1,\Lambda_{0}(X):=\{v\in E^{0}~|~x\geq v~\mathrm{for~some}~x\in X\},{\rm~and~for}~n\geq 1,
Λn(X):={yE0|0<|s1(y)|<andr(s1(y))Λn1(X)}Λn1(X).\Lambda_{n}(X):=\{y\in E^{0}~|~0<|s^{-1}(y)|<\infty~\mathrm{and}~r(s^{-1}(y))\subseteq\Lambda_{n-1}(X)\}\cup\Lambda_{n-1}(X).
Example 3

Let E=(E0,E1,r,s)E=(E^{0},E^{1},r,s) be the directed graph where E0={v,w}E^{0}=\{v,w\} and E1={e1,e2,e3}E^{1}=\{e_{1},e_{2},e_{3}\} such that r(e1)=s(e1)=vr(e_{1})=s(e_{1})=v and r(e2)=r(e3)=s(e3)=wr(e_{2})=r(e_{3})=s(e_{3})=w.

Then {v1}¯={v1,v2}\overline{\{v_{1}\}}=\{v_{1},v_{2}\}, whereas {v2}¯={v2}\overline{\{v_{2}\}}=\{v_{2}\}.

Example 4

Let E=(E0,E1,r,s)E=(E^{0},E^{1},r,s) be a directed graph where E0={vi|i}E^{0}=\{v_{i}~|~i\in\mathbb{Z}\} and E1={ei|i}E^{1}=\{e_{i}~|~i\in\mathbb{Z}\} such that r(ei)=vir(e_{i})=v_{i} and s(ei)=vi1s(e_{i})=v_{i-1}.

Let X={v0}X=\{v_{0}\}. Then we get Λ0(X)={v0,v1,}\Lambda_{0}(X)=\{v_{0},v_{1},\dots\}. Furthermore,

Λ1(X)\displaystyle\Lambda_{1}(X) =Λ0{v0}{yE0|0<|s1(y)|<andr(s1(y))Λ0(X)}\displaystyle=\Lambda_{0}\{v_{0}\}\cup\{y\in E^{0}~|~0<|s^{-1}(y)|<\infty~\mathrm{and}~r(s^{-1}(y))\subseteq\Lambda_{0}(X)\}
={v0,v1,}{v1}\displaystyle=\{v_{0},v_{1},\dots\}\cup\{v_{-1}\}
={v1,v0,v1,}.\displaystyle=\{v_{-1},v_{0},v_{1},\dots\}.

Similarly, Λk(X)={vk,vk+1,}\Lambda_{k}(X)=\{v_{-k},v_{-k+1},\dots\}, and hence X¯=n=0Λn(X)=E0\overline{X}=\bigcup^{\infty}_{n=0}\Lambda_{n}(X)=E^{0}.

With the introductory remarks now complete, we begin our discussion of the main result with the following important observation.

Remark 5

If II is a two-sided ideal of LK(E)L_{K}(E) and μ=μ1++μn\mu=\mu_{1}+\cdots+\mu_{n} is in II, where μ1,,μn\mu_{1},\dots,\mu_{n} are monomials in LK(E)L_{K}(E), then γi=s(μi)μr(μi)\gamma_{i}=s(\mu_{i})\mu r(\mu_{i}) is the sum of those μj\mu_{j} whose sources are all the same and whose ranges are all the same; specifically, the sum of those μj\mu_{j} for which s(μj)=s(μi)s(\mu_{j})=s(\mu_{i}) and r(μj)=r(μi)r(\mu_{j})=r(\mu_{i}). Moreover, γiI\gamma_{i}\in I. Thus we may write μ=γ1++γm\mu=\gamma_{1}+\cdots+\gamma_{m}, with each γi\gamma_{i} with the above properties.

Notation. Let LK(E)RL_{K}(E)_{\rm{R}} (resp., LK(E)GL_{K}(E)_{\rm{G}}) be the subring of elements in LK(E)L_{K}(E) whose terms involve only real edges (resp., only ghost edges).

Lemma 6

Let II be a two-sided ideal of LK(E)L_{K}(E) and Ireal=ILK(E)RI_{\rm{real}}=I\cap L_{K}(E)_{\rm{R}}. Then IrealI_{\rm real} is the two-sided ideal of LK(E)RL_{K}(E)_{\rm R} generated by elements of IrealI_{\rm{real}} having the form v+i=1nλigiv+\sum^{n}_{i=1}\lambda_{i}g^{i}, where vE0v\in E^{0}, gg is a cycle based at vv and λiK\lambda_{i}\in K for 1in1\leq i\leq n.

Proof 0.1.

Let JJ be the ideal of LK(E)RL_{K}(E)_{\rm R} generated by elements in IrealI_{\rm{real}} of the indicated form. Our claim is J=IrealJ=I_{\rm{real}}. Towards a contradiction, suppose IrealJI_{\rm{real}}\setminus J\neq\emptyset; choose μIrealJ\mu\in I_{\rm{real}}\setminus J of minimal length. By Remark 5, we can write μ=τ1++τm\mu=\tau_{1}+\cdots+\tau_{m} with each τi\tau_{i} is in IrealI_{\rm{real}} and is the sum of those paths whose sources are all the same and whose ranges are all the same. Since μJ\mu\not\in J, one of the τiJ\tau_{i}\not\in J. Replacing μ\mu by τi\tau_{i}, we may assume that μ=λ1μ1++λnμn\mu=\lambda_{1}\mu_{1}+\cdots+\lambda_{n}\mu_{n} where all the μi\mu_{i} have the same source and the same range. First we claim that one of the μi\mu_{i} must have length 0, i.e. μi=kv\mu_{i}=kv for some vertex vE0v\in E^{0} and kKk\in K. Suppose not. Then for each ii we can write μi=eiνi\mu_{i}=e_{i}\nu_{i} where eiE1e_{i}\in E^{1}. So μ=i=1neiνi\mu=\sum^{n}_{i=1}e_{i}\nu_{i}. Now

eiμ={j|ej=ei}λjνjIreale_{i}^{*}\mu=\sum_{\{j|e_{j}=e_{i}\}}\lambda_{j}\nu_{j}\in I_{\rm{real}}

and has smaller length than μ\mu. So eiμJe_{i}^{*}\mu\in J and hence clearly eieiμJe_{i}e_{i}^{*}\mu\in J. Then

μ=distincteieieiμJ,\mu=\sum_{{\rm distinct}~e_{i}}e_{i}e_{i}^{*}\mu\in J,

a contradiction. So we can assume without loss of generality that μ1=kv\mu_{1}=kv, with vv a vertex. Since all the terms in μ\mu have the same source and the same range, each μi\mu_{i} is a closed path based at vv. Multiplying by a scalar if necessary we can write μ=v+λ2μ2++λnμn\mu=v+\lambda_{2}\mu_{2}+\cdots+\lambda_{n}\mu_{n}, μi\mu_{i} closed paths at vv.

Case I: There exists no, or exactly one, closed simple path at vv. If there are no closed simple paths at vv then we get μJ\mu\in J, a contradiction. If there is exactly one closed simple path gg based at vv then necessarily gg must be a cycle. Furthermore, the only paths in EE which have source and range equal to vv are powers of gg. Then μ=v+i=2nλigmiJ\mu=v+\sum_{i=2}^{n}\lambda_{i}g^{m_{i}}\in J, a contradiction.

Case II: There exist at least two distinct closed simple paths g1g_{1} and g2g_{2} based at vv. As g1g2g_{1}\neq g_{2} and neither is a subpath of the other, g2g1=0=g1g2g_{2}^{*}g_{1}=0=g_{1}^{*}g_{2}. Without loss of generality assume |μ2||μn|1|\mu_{2}|\geq\cdots\geq|\mu_{n}|\geq 1. Then for some kk\in\mathbb{N} |g1k|>|μ2||g_{1}^{k}|>|\mu_{2}|. Multiplying by (g1)k(g_{1}^{*})^{k} on the left and g1kg_{1}^{k} on the right, we get

μ=(g1)kμ(g1)k=v+i=2nλi(g1)kμi(g1)k.\mu^{\prime}=(g^{*}_{1})^{k}\mu(g_{1})^{k}=v+\sum_{i=2}^{n}\lambda_{i}(g_{1}^{*})^{k}\mu_{i}(g_{1})^{k}.

Note that if 0(g1)kμi(g1)k0\neq(g_{1}^{*})^{k}\mu_{i}(g_{1})^{k}, then (g1)kμi0(g_{1}^{*})^{k}\mu_{i}\neq 0. Since |g1k|>|μi||g_{1}^{k}|>|\mu_{i}|, we get g1k=μiμig_{1}^{k}=\mu_{i}\mu_{i}^{\prime} for some path μi\mu_{i}^{\prime}. Since the μi\mu_{i} are closed paths based at the vertex vv, one gets from the equation (g1)k=μiμi(g_{1})^{k}=\mu_{i}\mu_{i}^{\prime} that μi=(g1)r\mu_{i}=(g_{1})^{r} for some integer rkr\leq k. So μi\mu_{i} commutes with (g1)k(g_{1})^{k} and thus each non-zero term (g1)kμi(g1)k=μi(g_{1}^{*})^{k}\mu_{i}(g_{1})^{k}=\mu_{i}.

Since g2g1=0g_{2}^{*}g_{1}=0, g2μi=0g_{2}^{*}\mu_{i}=0 for every i{2,,n}i\in\{2,\dots,n\} and so we get g2μg2=g2vg2=vILK(E)R=Irealg_{2}^{*}\mu^{\prime}g_{2}=g_{2}^{*}vg_{2}=v\in I\cap L_{K}(E)_{\rm R}=I_{\rm real}, which implies that vv is in JJ. Then μ=μvJ\mu=\mu v\in J, a contradiction.

It can be easily shown that the analogue of Lemma 6 is true for IghostI_{\mathrm{ghost}}. We state this for the sake of completeness.

Lemma 7.

Let II be a two-sided ideal of LK(E)L_{K}(E). Then IghostI_{\mathrm{ghost}} is generated by elements of the form v+k=1mλk(g)kv+\sum^{m}_{k=1}\lambda_{k}(g^{*})^{k}, where vE0v\in E^{0}, gg is a cycle at vv and λiK\lambda_{i}\in K for 1in1\leq i\leq n.

Theorem 8.

Let EE be a row-finite graph. Let II be any two-sided ideal of LK(E)L_{K}(E). Then II is generated by elements of the form v+k=1mλkgkv+\sum^{m}_{k=1}\lambda_{k}g^{k}, where vE0v\in E^{0}, gg is a cycle at vv and λ1,,λmK\lambda_{1},\dots,\lambda_{m}\in K.

Proof 0.2.

Let JJ be the two-sided ideal of LK(E)L_{K}(E) generated by IrealI_{\rm{real}}. By Lemma 6, it is enough if we show that I=JI=J. Suppose not. Choose x=i=1dλiμiνix=\sum^{d}_{i=1}\lambda_{i}\mu_{i}\nu_{i}^{*} in IJI\setminus J, where dd is minimal and μ1,,μd,ν1,,νd\mu_{1},\dots,\mu_{d},\nu_{1},\dots,\nu_{d} are real paths in LK(E)RL_{K}(E)_{R}. By Remark 5, x=α1++αmx=\alpha_{1}+\cdots+\alpha_{m}, where each αiI\alpha_{i}\in I and is a sum of those monomials all having the same source and same range. Since xJx\not\in J, αjJ\alpha_{j}\not\in J for some jj. By the minimality of dd, we can replace xx by αi\alpha_{i}. Thus we we can assume that x=i=1dλiμiνix=\sum^{d}_{i=1}\lambda_{i}\mu_{i}\nu_{i}^{*}, where for all ii, jj, s(μiνi)=s(μjνj)s(\mu_{i}\nu_{i}^{*})=s(\mu_{j}\nu_{j}^{*}) and r(μiνi)=r(μjνj)r(\mu_{i}\nu_{i}^{*})=r(\mu_{j}\nu_{j}^{*}). Among all such x=i=1dλiμiνiIJx=\sum^{d}_{i=1}\lambda_{i}\mu_{i}\nu_{i}^{*}\in I\setminus J with minimal dd, select one for which (|ν1|,,|νd|)(|\nu_{1}|,\dots,|\nu_{d}|) is the smallest in the lexicographic order of (+)d(\mathbb{Z}^{+})^{d}. First note that we have |νi|>0|\nu_{i}|>0 for some ii, otherwise xx is in IrealJ.I_{\rm real}\subset J. Let ee be in E1E^{1}. Then note that

xe=i=1dλiμiνie=i=1dλiμi(νi)xe=\sum^{d}_{i=1}\lambda_{i}\mu_{i}\nu_{i}^{*}e=\sum^{d^{\prime}}_{i=1}\lambda_{i}\mu_{i}(\nu_{i}^{\prime})^{*}

either has fewer terms (d<d)(d^{\prime}<d), or d=dd=d^{\prime} and (|νi|,,|νd|)(|\nu_{i}^{\prime}|,\dots,|\nu_{d}^{\prime}|) is smaller than (|ν1|,,|νd|)(|\nu_{1}|,\dots,|\nu_{d}|). Then by minimality, we get xexe is in JJ for every eE1e\in E^{1}. Since |νi|>0|\nu_{i}|>0 for some ii, ww is not a sink and emits finitely many edges. Hence we have

x=xw=x{ejE1:s(ej)=v}ejej={ejE1:s(ej)=v}(xej)ejJ.x=xw=x\sum_{\{e_{j}\in E^{1}:s(e_{j})=v\}}e_{j}e_{j}^{*}=\sum_{\{e_{j}\in E^{1}:s(e_{j})=v\}}(xe_{j})e_{j}^{*}\in J.

We get a contradiction, so the result follows.

Remark 9.

We note that the Theorem does not hold for arbitrary graphs. An example is the “infinite clock”: Let E0={v,w1,w2,}E^{0}=\{v,w_{1},w_{2},\dots\} and E1={e1,e2,}E^{1}=\{e_{1},e_{2},\dots\} with r(ei)=wir(e_{i})=w_{i} and s(ei)=vs(e_{i})=v. Then the two-sided ideal generated by ve1e1v-e_{1}e_{1}^{*} is not generated by the elements of the desired form.

Notation. The element v+i=1nλigiv+\sum_{i=1}^{n}\lambda_{i}g^{i} is denoted by p(g)p(g), where p(x)=1+λ2x++λnxnK[x]p(x)=1+\lambda_{2}x+\cdots+\lambda_{n}x^{n}\in K[x].

Now the Theorem is in hand, we are going to put the pieces together to get the Noetherian result.

Remark 10.

Let gg be a cycle based at vE0v\in E^{0} and let p1(x),p2(x)K[x]p_{1}(x),p_{2}(x)\in K[x] be such that p1(g),p2(g)Ip_{1}(g),p_{2}(g)\in I. If we let q(x)=gcd(p1(x),p2(x))K[x]q(x)={\rm gcd}(p_{1}(x),p_{2}(x))\in K[x], then q(g)<p1(g),p2(g)>q(g)\in<p_{1}(g),p_{2}(g)>.

Lemma 11.

Let II be a two-sided ideal of LK(E)L_{K}(E), where EE is an arbitrary graph. Suppose gg, hh are two non-trivial cycles based at distinct vertices uu, vv respectively. Suppose u+argr=p(g)u+\sum a_{r}g^{r}=p(g) and v+bshs=q(h)v+\sum b_{s}h^{s}=q(h) both belong to II, where p(x)p(x) and q(x)q(x) are polynomials of smallest positive degree in K[x]K[x] with p(0)=1=q(0)p(0)=1=q(0) such that p(g)Ip(g)\in I and q(h)Iq(h)\in I. If uvu\geq v, then q(h)<p(g)>.q(h)\in<p(g)>.

Proof 0.3.

Let μ\mu be a path from uu to vv. We claim vv must lie on the cycle gg. Because, otherwise, μg=0\mu^{*}g=0 and so μp(g)μ=μuμ+arμgrμ=vI\mu^{*}p(g)\mu=\mu^{*}u\mu+\sum a_{r}\mu^{*}g^{r}\mu=v\in I. This contradicts the fact that degq(x)>0{\rm deg}q(x)>0. So we can write g=μνg=\mu\nu and h=νμh=\nu\mu where ν\nu is the part of gg from vv to uu. Since μgμ=h\mu^{*}g\mu=h, we get μp(g)μ=p(h)I\mu^{*}p(g)\mu=p(h)\in I. By the minimality of q(x)q(x), q(x)q(x) is a divisor of p(x)p(x) in K[x]K[x]. Similarly, since νq(h)ν=q(g)I\nu^{*}q(h)\nu=q(g)\in I, we conclude that p(x)p(x) is a divisor of q(x)q(x). Thus q(x)=kp(x)q(x)=kp(x) for some kKk\in K. Since p(0)=1=q(0)p(0)=1=q(0), q(x)=p(x)q(x)=p(x). Hence q(h)=p(h)=μp(g)μ<p(g)>q(h)=p(h)=\mu^{*}p(g)\mu\in<p(g)>.

The next Lemma and its proof is implicit in the proof of Lemma 7 in [2].

Lemma 12.

Let EE be an arbitrary graph and SE0S\subset E^{0}. If vS¯v\in\overline{S}, and there is a non-trivial cycle based at vv, then uvu\geq v for some uSu\in S.

Proof 0.4.

We recall that S¯=n0Λn(S)\overline{S}=\cup_{n\geq 0}\Lambda_{n}(S). Let kk be the smallest non-negative integer such that vΛk(S)v\in\Lambda_{k}(S). We prove the Lemma by the induction on kk, the Lemma being true by definition when k=0k=0. Assume k>0k>0 and that the Lemma holds when k=n1k=n-1. Let k=nk=n. Since vΛn(S)Λn1(S)v\in\Lambda_{n}(S)\setminus\Lambda_{n-1}(S), 0<|s1(v)|<0<|s^{-1}(v)|<\infty and {w1,,wm}=r(s1(v))Λn1(S)\{w_{1},\dots,w_{m}\}=r(s^{-1}(v))\subset\Lambda_{n-1}(S). Since vv is the base of a non-trivial cycle gg, one of the vertices, say, wjw_{j} lies on the cycle gg and so wjvw_{j}\geq v. Since wjΛn1(S)w_{j}\in\Lambda_{n-1}(S) and is the base of a cycle, by induction there is a uSu\in S such that uwju\geq w_{j}. Then uvu\geq v, as desired.

We also need the following Lemma, whose proof is given in the first paragraph of the proof of Theorem 5.7 in [8].

Lemma 13.

Let EE be an arbitrary graph and let HH be a hereditary and saturated subset of vertices in EE. If II is the two-sided ideal generated by HH, then IE0=HI\cap E^{0}=H.

Theorem 14.

Let EE be a row-finite graph. Then the following are equivalent:

{arabiclist}

LK(E)L_{K}(E) has a.c.c. on two-sided ideals,

LK(E)L_{K}(E) has a.c.c. on two-sided graded ideals,

The hereditary saturated closures of the subsets of the vertices in E0E^{0} satisfy a.c.c..

Proof 0.5.

(3)(1)(3)\Rightarrow(1) Suppose the ascending chain condition holds on the hereditary saturated closures of the subsets of E0E^{0}. Let II be a two-sided ideal of LK(E)L_{K}(E). By Theorem 8 and by Remark 10, II is generated by the set

T=\displaystyle T= {v+rλrgr=p(g)I|vE0,gisacycle(maybetrivial)basedatvand\displaystyle\{v+\sum_{r}\lambda_{r}g^{r}=p(g)\in I~|~v\in E^{0},g{\rm~is~a~cycle~(may~be~trivial)~based~at~}v~{\rm and}
p(x)K[x]isapolynomialofsmallestdegreesuchthatp(g)Iandp(0)=1}.\displaystyle~p(x)\in K[x]{\rm~is~a~polynomial~of~smallest~degree~such~that~}p(g)\in I{\rm~and~}p(0)=1\}.

It is well known that two-sided Noetherian is equivalent to every two-sided ideal being finitely generated, so we wish to show that II is generated by a finite subset of TT.

Suppose, towards a contradiction, there are infinitely many pi(gi)=vi+λrgirTp_{i}(g_{i})=v_{i}+\sum\lambda_{r}g_{i}^{r}\in T with iHi\in H, an infinite set and for each ii, gig_{i} is a non-trivial cycle based at viv_{i} and that degpi(x)>0{\rm deg}p_{i}(x)>0. By Lemma 11, we may assume that for any two ii, jj with iji\neq j, vivjv_{i}\ngeq v_{j}. Well-order the set HH and consider it as the set of all ordinals less than an infinite ordinal κ\kappa. Define S1=v1S_{1}={v_{1}} and for any α<κ\alpha<\kappa, define Sα=β<αSβS_{\alpha}=\cup_{\beta<\alpha}S_{\beta} if α\alpha is a limit ordinal, and define Sα=Sβ{vβ+1}S_{\alpha}=S_{\beta}\cup\{v_{\beta+1}\} if α\alpha is a non-limit ordinal of the form β+1\beta+1. By the hypothesis the ascending chain of hereditary saturated closures of subsets S¯1S¯2S¯α\overline{S}_{1}\subset\overline{S}_{2}\subset\cdots\overline{S}_{\alpha}\subset\cdots becomes stationary after a finite number of steps. So there is an integer nn such that S¯n=S¯n+1=\overline{S}_{n}=\overline{S}_{n+1}=\cdots. Now vn+1S¯n+1=S¯nv_{n+1}\in\overline{S}_{n+1}=\overline{S}_{n} and by Lemma 12, there is a viSnv_{i}\in S_{n} such that vivn+1v_{i}\geq v_{n+1}. This is a contradiction. Hence the set W={pi(gi)T|degpi(x)>0}W=\{p_{i}(g_{i})\in T~|~{\rm deg}p_{i}(x)>0\} is finite.

So by the previous paragraph, if there are only finitely many pi(gi)p_{i}(g_{i}) in TT with degpi(x)=0{\rm deg}p_{i}(x)=0, that is, only finitely many vertices in TT, then we are done. We index the vertices vαv_{\alpha} in TT by ordinals α<κ\alpha<\kappa, an infinite ordinal, then as before, we get a well-ordered ascending chain of hereditary saturated closure of subsets S¯1S¯2S¯α\overline{S}_{1}\subset\overline{S}_{2}\subset\cdots\subset\overline{S}_{\alpha}\subset\cdots (α<κ)(\alpha<\kappa) where S1={v1}S_{1}=\{v_{1}\} and the SαS_{\alpha} are inductively defined as before. Since, by hypothesis, this chain becomes stationary after a finite number of steps, there is an integer nn such that S¯α=S¯n\overline{S}_{\alpha}=\overline{S}_{n} for all α>n\alpha>n. Thus {vα|α<κ}S¯n\{v_{\alpha}~|~\alpha<\kappa\}\subset\overline{S}_{n}. Since the ideal generated by the finite set Sn={v1,,vn}S_{n}=\{v_{1},\dots,v_{n}\} contains S¯n\overline{S}_{n}, we conclude that the ideal II is generated by the finite set WSnW\cup S_{n}. Thus the Leavitt path algebra is two-sided Noetherian.

(1)(3)(1)\Rightarrow(3) Conversely, suppose LK(E)L_{K}(E) is two-sided Noetherian. Consider an ascending chain of hereditary saturated closures of subsets of vertices S¯1S¯2\overline{S}_{1}\subset\overline{S}_{2}\subset\cdots in E0E^{0}. Consider the corresponding ascending chain of two-sided ideals I1I2,I_{1}\subset I_{2}\subset\cdots, where for each integer ii, IiI_{i} is the two-sided ideal generated by S¯i\overline{S}_{i}. By hypothesis, there is an integer nn such that In=IiI_{n}=I_{i} for all i>ni>n. We claim that S¯i=S¯n\overline{S}_{i}=\overline{S}_{n} for all i>ni>n. Otherwise, we can find a vertex wS¯iS¯nw\in\overline{S}_{i}\setminus\overline{S}_{n} and since wIi=Inw\in I_{i}=I_{n}, wInE0=S¯nw\in I_{n}\cap E^{0}=\overline{S}_{n} by Lemma 13 and this is a contradiction.

(1)(2)(1)\Leftrightarrow(2) It is well-known (see [8]) that if I(H)I(H) is a two-sided ideal of LK(E)L_{K}(E) generated by a hereditary and saturated subset HH of E0E^{0}, then I(H)I(H) is a graded ideal of LK(E)L_{K}(E). If we call LK(E)L_{K}(E) graded two-sided Noetherian if graded two-sided ideals of LK(E)L_{K}(E) satisfy the ascending chain condition, then Theorem 14 states that for any graph EE, LK(E)L_{K}(E) is two-sided Noetherian if and only if it is graded two-sided Noetherian.

Remark 15.

We note that this result only shows that the a.c.c. on graded ideals is sufficient to get a.c.c. on all ideals, and that we are not proving that every ideal in a Noetherian Leavitt path algebra is graded. As an example we can consider K[x,x1]K[x,x^{-1}], which is the Leavitt path algebra of the graph with one vertex and one loop. Note that although this Leavitt path algebra has infinitely many ideals, it is nonetheless Noetherian, but has only the trivial graded ideals.

Now we easily see

Corollary 16.

Every Leavitt path algebra with a finite graph is two-sided Noetherian.

We conclude by presenting another example of a non-Noetherian Leavitt path algebra.

Example 17

Let E=(E0,E1,r,s)E=(E^{0},E^{1},r,s) be the directed graph where E0={v,w1,w2,w3,}E^{0}=\{v,w_{1},w_{2},w_{3},\dots\} and E1={e1,e2,}{f1,f2,}E^{1}=\{e_{1},e_{2},\dots\}\cup\{f_{1},f_{2},\dots\} is such that r(ei)=vr(e_{i})=v and s(ei)=r(fi)=s(fi)=wis(e_{i})=r(f_{i})=s(f_{i})=w_{i}. The graph of this Leavitt path algebra is given in Figure 1.

\psfigfile=Figure2a.pdf

Figure 1: Leavitt path algebra defined in Example 17.

Note that if we let Si={w1,,wi}S_{i}=\{w_{1},\dots,w_{i}\}, then S1¯S2¯\overline{S_{1}}\subset\overline{S_{2}}\subset\cdots is a non-terminating ascending chain of hereditary saturated closures of sets in E0E^{0}. Hence by Theorem 8, LK(E)L_{K}(E) is not two-sided Noetherian. Indeed, <w1><w1,w2><w_{1}>\subset<w_{1},w_{2}>\subset\cdots is a non-terminating ascending chain of ideals in LK(E)L_{K}(E).

In [4] we present some additional consequences of Theorem 8, including a description of the two-sided artinian Leavitt path algebras.

Acknowledgments

The author is indebted to Dr. Jason P. Bell for the guidance and the technical discussions since the inception of this work, as well as to Dr. Gene Abrams and to Dr. Kulumani M. Rangaswamy for their thorough reviews and support. The author also thanks Dr. Gonzalo Aranda Pino and Dr. Kathi Crow for their valuable feedback on the first draft of this paper, and Azhvan Sheikh Ahmady for reading and improving this manuscript.

References

  • [1] G. Abrams and G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra 293 (2005) 319–334.
  • [2] G. Abrams and G. Aranda Pino, Purely infinite simple Leavitt path algebras, J. Pure and Applied Algebra 207 (2006) 553–563.
  • [3] G. Abrams and G. Aranda Pino, The Leavitt path algebras of arbitrary graphs, Houston J. Math. 34 (2006) 423–443.
  • [4] G. Abrams, J. P. Bell, P. Colak and K. M. Rangaswamy, Two-sided ideals in Leavitt path algebras: Applications, in preparation.
  • [5] P. Ara, M. A. Moreno and E. Pardo, Nonstable KK-theory for graph algebras, Algebras Represent. Theory 10 (2007) 157–178.
  • [6] K. R. Goodearl, Leavitt path algebras and direct limits, Rings, Modules and Representations. Contemporary Mathematics series, (2009) 165–188.
  • [7] I. Raeburn, Graph algebras. CBMS Regional Conference Series in Mathematics, vol 103, American Mathematical Society, Providence, 2005.
  • [8] M. Tomforde, Uniqueness theorems and ideal structure for Leavitt path algebras, J. of Algebra 318 (2007) 270–299.