Processing math: 100%
Research article Special Issues

The ADMM algorithm for audio signal recovery and performance modification with the dual Douglas-Rachford dynamical system

  • Practitioners employ operator splitting methods—such as alternating direction method of multipliers (ADMM) and its "dual" Douglas-Rachford method (DR)—to solve many kinds of optimization problems. We provide a gentle introduction to these algorithms, and illustrations of their duality-like relationship in the context of solving basis pursuit problems for audio signal recovery. Recently, researchers have used the dynamical systems associated with the iterates of splitting methods to motivate the development of schemes to improve performance. These developments include a class of methods that act by iteratively minimizing surrogates for a Lyapunov function for the dynamical system. An exemplar of this class is currently state-of-the-art for the feasibility problem of finding wavelets with special structure. Early experimental evidence has also suggested that, when implemented in a primal-dual (ADMM and DR) framework, this exemplar may provide improved performance for basis pursuit problems. We provide a reasonable way to compute the updates for this exemplar, and we study the application of this method to the aforementioned basis pursuit audio problems. We provide experimental results and visualizations of the dynamical system for the dual DR sequence. We observe that for highly structured problems with real data, the algorithmic behavior is noticeably different than for randomly generated problems.

    Citation: Andrew Calcan, Scott B. Lindstrom. The ADMM algorithm for audio signal recovery and performance modification with the dual Douglas-Rachford dynamical system[J]. AIMS Mathematics, 2024, 9(6): 14640-14657. doi: 10.3934/math.2024712

    Related Papers:

    [1] Saad Ihsan Butt, Artion Kashuri, Muhammad Umar, Adnan Aslam, Wei Gao . Hermite-Jensen-Mercer type inequalities via Ψ-Riemann-Liouville k-fractional integrals. AIMS Mathematics, 2020, 5(5): 5193-5220. doi: 10.3934/math.2020334
    [2] Fangfang Ma . Fractional version of the Jensen-Mercer and Hermite-Jensen-Mercer type inequalities for strongly h-convex function. AIMS Mathematics, 2022, 7(1): 784-803. doi: 10.3934/math.2022047
    [3] Ghulam Farid, Hafsa Yasmeen, Hijaz Ahmad, Chahn Yong Jung . Riemann-Liouville Fractional integral operators with respect to increasing functions and strongly (α,m)-convex functions. AIMS Mathematics, 2021, 6(10): 11403-11424. doi: 10.3934/math.2021661
    [4] Muhammad Umar, Saad Ihsan Butt, Youngsoo Seol . Milne and Hermite-Hadamard's type inequalities for strongly multiplicative convex function via multiplicative calculus. AIMS Mathematics, 2024, 9(12): 34090-34108. doi: 10.3934/math.20241625
    [5] Muhammad Aslam Noor, Khalida Inayat Noor . Higher order strongly general convex functions and variational inequalities. AIMS Mathematics, 2020, 5(4): 3646-3663. doi: 10.3934/math.2020236
    [6] Alper Ekinci, Erhan Set, Thabet Abdeljawad, Nabil Mlaiki . Generalizations of Ostrowski type inequalities via F-convexity. AIMS Mathematics, 2022, 7(4): 7106-7116. doi: 10.3934/math.2022396
    [7] Muhammad Zakria Javed, Muhammad Uzair Awan, Loredana Ciurdariu, Omar Mutab Alsalami . Pseudo-ordering and δ1-level mappings: A study in fuzzy interval convex analysis. AIMS Mathematics, 2025, 10(3): 7154-7190. doi: 10.3934/math.2025327
    [8] Peiyu Yan, Qi Li, Yu Ming Chu, Sana Mukhtar, Shumaila Waheed . On some fractional integral inequalities for generalized strongly modified h-convex functions. AIMS Mathematics, 2020, 5(6): 6620-6638. doi: 10.3934/math.2020426
    [9] Jia-Bao Liu, Saad Ihsan Butt, Jamshed Nasir, Adnan Aslam, Asfand Fahad, Jarunee Soontharanon . Jensen-Mercer variant of Hermite-Hadamard type inequalities via Atangana-Baleanu fractional operator. AIMS Mathematics, 2022, 7(2): 2123-2141. doi: 10.3934/math.2022121
    [10] Imran Abbas Baloch, Aqeel Ahmad Mughal, Yu-Ming Chu, Absar Ul Haq, Manuel De La Sen . A variant of Jensen-type inequality and related results for harmonic convex functions. AIMS Mathematics, 2020, 5(6): 6404-6418. doi: 10.3934/math.2020412
  • Practitioners employ operator splitting methods—such as alternating direction method of multipliers (ADMM) and its "dual" Douglas-Rachford method (DR)—to solve many kinds of optimization problems. We provide a gentle introduction to these algorithms, and illustrations of their duality-like relationship in the context of solving basis pursuit problems for audio signal recovery. Recently, researchers have used the dynamical systems associated with the iterates of splitting methods to motivate the development of schemes to improve performance. These developments include a class of methods that act by iteratively minimizing surrogates for a Lyapunov function for the dynamical system. An exemplar of this class is currently state-of-the-art for the feasibility problem of finding wavelets with special structure. Early experimental evidence has also suggested that, when implemented in a primal-dual (ADMM and DR) framework, this exemplar may provide improved performance for basis pursuit problems. We provide a reasonable way to compute the updates for this exemplar, and we study the application of this method to the aforementioned basis pursuit audio problems. We provide experimental results and visualizations of the dynamical system for the dual DR sequence. We observe that for highly structured problems with real data, the algorithmic behavior is noticeably different than for randomly generated problems.



    The graphs considered in this paper are finite, undirected and simple. Let G be a graph with vertex set V(G) and edge set E(G). For any XV(G), denote by G[X] the subgraph of G induced by X. For any u,vV(G), denote by dG(u,v) the distance from u to v in G as well as dG(v), NG(v) and NG[v] the degree, open neighborhood and closed neighborhood in G, respectively. Furthermore, for any UV(G), the open and closed neighbourhood of U are defined as NG(U)=vUNG(v) and NG[U]=NG(U)U, respectively. Two graphs G and H are disjoint if they have no vertices in common and no vertex of G is adjacent to any vertex of H. The union of graphs G and H is the graph GH with V(GH)=V(G)V(H) and E(GH)=E(G)E(H).

    A set SV(G) is called a 2-packing of graph G if dG(x,y)>2 for every pair of distinct vertices x,yS. A set DV(G) is called a dominating set of G if every vertex of G is either in D or adjacent to a vertex of D. The domination number γ(G) is the cardinality of a minimum dominating set of G. We denote by MDS_(G) the set of all the minimum dominating sets. That is, MDS_(G)={DD is a minimum dominating set of G}.

    Remark. We use the symbol MDS_ but not MDS, because MDS(G) has been defined to be the set of all the minimal dominating set of a graph G in the textbook [14].

    Definition 1.1. A vertex vV(G) is called a critical vertex of G if γ(Gv)=γ(G)1.

    Observation 1.2. For any vV(G),

    γ(Gv)=γ(G)1γ(Gv)γ(G)1. (1.1)

    Definition 1.3. A graph G is called vertex-critical if every vertex of G is critical.

    The research on vertex-critical graphs was conducted in [4]. Afterwards, authors studied on its diameter [9], connectivity [1], existence of perfect matching [1,2,16] and factor critical property [2,22,23]. Interested reads could consider the existence of irregular dominating set on vertex-critical graphs [7,18].

    Furthermore, based on the right and the left of Formula (1.1), Brigham et al. [5] and Phillips et al. [21] extended the notion of vertex-critical graphs by introducing (γ,k)-critical graphs and (γ,t)-critical graphs, respectively.

    Definition 1.4. [5] A graph G is called (γ,k)-critical if γ(GS)<γ(G) for every SV(G) with |S|=k.

    Definition 1.5. [21] A graph G is called (γ,t)-critical if γ(GS)=γ(G)t for every 2-packing S of G with |S|=t.

    In Definition 1.4, if k=2, then G is called domination bicritical. For more information of (γ,k)-critical or domination bicritical graphs, readers are suggested to refer to [8,10,11,17,19,20].

    Now, again based on the left of Formula (1.1), we introduce the definition of strong critical vertex-set to extend the notion of critical vertex in the following Definition 1.1. (It is easy to get that a strong critical vertex-set of G is also a 2-packing of G.) To compare Observation 1.2 and Definition 1.3, we give Observation 1.2 and Definition 1.3, and then we introduce Definition 1.6 with a remark.

    Definition 1.1. [25] A set SV(G) is called a strong critical vertex-set (or just st-critical vertex-set for short) of G if γ(GS)=γ(G)|S|.

    Observation 1.2. For any SV(G), γ(GS)=γ(G)|S|γ(GS)γ(G)|S|.

    Definition 1.3. A graph G is called strong l-vertex-set-critical if V(G) can be partitioned into l non-empty strong critical vertex-sets of G.

    Definition 1.6. Let S1,S2,,Sl be non-empty strong critical vertex-sets of G. If {S1,S2,,Sl} is a partition of V(G), then we call {S1,S2,,Sl} or S1S2Sl as a strong critical vertex-set partition of G.

    Lemma 1.7. [25] A subset of an st-critical vertex-set of G is still an st-critical vertex-set of G.

    Remark. Let S1=S11S21 with S11,S21. According to Definition 1.6 and Lemma 1.7, if S1S2Sl is an st-critical vertex-set partition of G, then S11S21S2Sl is also an st-critical vertex-set partition of G. Thus, in general, if G is a strong l-vertex-set-critical graph, then it may also be a strong j-vertex-set-critical graph for any lj|V(G)|.

    When we talk about st-critical vertex-sets, we would like to mention another related notion---Two-colored γ-set. The present authors think that both of them are important on the problem of building family of graphs that make the equality hold in Vizing's Conjecture [15,25].

    Definition 1.8. [15] Let DMDS_(G). D is called a two-colored γ-set of G if D partitions as D=D1D2 such that V(G)NG[D1]=D2 and V(G)NG[D2]=D1.

    In Definition 1.8, since V(G)NG[D1]=D2, we can deduce that D1MDS_(GD2). So γ(GD2)=|D1|=|D||D2|=γ(G)|D2|, which implies that D2 is an st-critical vertex-set of G, and so is D1 symmetrically. Because two-colored γ-set is not the motif of this paper, we just introduce a proposition and a conjecture about it below.

    Proposition 1.9. [15] If G is a generalized comb and H has a two-colored γ-set, then γ(GH)=γ(G)γ(H).

    Conjecture 1.10. [13] If G is a connected bipartite graph such that V(G) can be partitioned into two-colored γ-sets, then G is the 4-cycle or G can be obtained from K2t,2t by removing the edges of t vertex-disjoint 4-cycles.

    In Proposition 1.9, "" represents the cartesian product and a nontrivial connected graph G is called a generalized comb if each vertex of degree greater than one is adjacent to exactly one 1-degree-vertex of G. Conjecture 1.10 tries to give a necessary condition for a connected bipartite graph whose vertex set can be partitioned into two-colored γ-sets. Note that if a graph can be partitioned into k two-colored γ-sets, then it can partitioned into 2k domination critical vertex-sets.

    At last, we sketch the coming two sections. To extend the previous concept---Vertex coalescence, we introduce the concept of vertex-set coalescence and give two theorems about it in Section 2, which are fundamental results of strong l-vertex-set-critical graphs. Let Cl={GG is a strong l-vertex-set-critical connected graph}. We will obtain that Cl if and only if l{2,3,5} in Section 3.

    Brigham et al. [4] studied on the vertex-critical graphs, and listed the following Theorems 2.2 and 2.3 without proofs because they thought the proofs are cumbersome but straightforward. In order to state these two theorems, we have to introduce the notion of vertex coalescence first. (Readers who want to know the concept of edge coalescence can refer to [12].)

    Definition 2.1. [4,15] Let G and H be two disjoint graphs with xV(G) and yV(H). The vertex coalescence GxyH (or just GH if x and y are arbitrary) of G and H via x and y, is the graph obtained from the union of G and H by identifying x with y. (Refer to Figure 1.)

    Figure 1.  The vertex coalescence of graphs H4,8 and C4.

    Agreement. In this section, when identifying x with y, we choose x but y to represent the identified vertex in the resulting graph.

    Theorem 2.2. [4] Let G and H be two disjoint graphs, and let GH be any vertex coalescence of G and H. Then γ(G)+γ(H)1γ(GH)γ(G)+γ(H). Furthermore, if both G and H are vertex-critical or GH is vertex-critical, then γ(GH)=γ(G)+γ(H)1.

    Theorem 2.3. [4] The graph GH is vertex-critical if and only if both G and H are vertex-critical.

    To compare Brigham's results, we give the corresponding results on strong l-vertex-set-critical graphs one to one (see Definition 2.1, Theorems 2.2 and 2.3). For the mathematical rigor, we are going to prove them without the supporting of Theorems 2.2 and 2.3, where in fact, our proofs include the derivation of Brigham's results. Before this, we need to introduce the definition of "compatible" and two lemmas.

    Definition 2.4. Let G be a graph with x,yV(G). x and y are called compatible in G if there exists D0MDS_(G) such that {x,y}D0, and incompatible in G if |{x,y}D|<2 for any DMDS_(G).

    Lemma 2.5. [6] Let G be a graph with x,yV(G), and G be the graph obtained from G by identifying the two vertices x and y. Then γ(G)<γ(G) if and only if x and y are compatible or at least one of x and y is critical in G.

    Lemma 2.6. Let J be a graph with x,yV(J), and J be the graph obtained from J by identifying the two vertices x and y. Then γ(J)1γ(J)γ(J) with the second equality holds if and only if x and y are incompatible and neither x nor y is critical in J.

    Proof. Let DMDS_(J). Then D{y} is a dominating set of J, and so γ(J)|D{y}|γ(J)+1, which implies that γ(J)1γ(J). Let DMDS_(J) and

    D0={D,ifyD,(D{y}){x},ifyD.

    Then D0 is a dominating set of J, and so γ(J)|D0||D|=γ(J).

    Now, since γ(J)γ(J), it follows from the contrapositive of Lemma 2.5 that γ(J)=γ(J) if and only if x and y are incompatible and neither x nor y is critical in J.

    Definition 2.1. Let G and H be two disjoint graphs with XV(G), YV(H) and |X|=|Y|. Let X={x1,x2,,xm} and Y={y1,y2,,ym}. The vertex-set coalescence GXYH of G and H via X and Y, is the graph obtained from the union of G and H by identifying xi with yi for every 1im. (Refer to Figure 2.)

    Figure 2.  Illustration for the vertex-set coalescence GXYH.

    Theorem 2.2. Let G and H be two disjoint graphs with XV(G), YV(H) and |X|=|Y|. Let G and H be the subgraphs of GXYH induced by V(G) and V(HY)X, respectively. Then

    (a) γ(G)+γ(H)|X|γ(GXYH)γ(G)+γ(H). Furthermore, the first equality holds if X is an st-critical vertex-set of G or there exists ˜DGMDS_(G) and ˜DHMDS_(H) such that X˜DG and Y˜DH; the second equality holds only if DGDH= for any DGMDS_(G) and DHMDS_(H).

    (b) X and Y are st-critical vertex-sets of G and H respectively if and only if X is an st-critical vertex-set of GXYH;

    (c) if X is an st-critical vertex-set of GXYH, then γ(GXYH)=γ(G)+γ(H)|X|.

    Proof. (a) Firstly, let DMDS_(GXYH), B1 and B2 be the subsets of X which can not be dominated by DV(G) and DV(H) in G and H, respectively. Then (DV(G))B1 and (DV(H))B2 are dominating sets of G and H, respectively. Since DMDS_(GXYH), it follows that the vertices of X not dominated by DV(G) in G must be dominated by DV(H) in H, and the converse is also true. Thus B1B2=. Now, we have γ(G)+γ(H)|(DV(G))B1|+|(DV(H))B2|=|D|+|DX|+|B1|+|B2|γ(GXYH)+|X|, which implies that γ(G)+γ(H)|X|γ(GXYH). Secondly, let DGMDS_(G) and DHMDS_(H). Since G and H are spanning subgraphs of G and H respectively, it follows that γ(GXYH)|DGDH|γ(G)+γ(H)γ(G)+γ(H).

    Furthermore, if X is an st-critical vertex-set of G, then γ(GXYH)γ(GX)+γ(H)γ(GX)+γ(H)=γ(G)|X|+γ(H), which implies that γ(GXYH)=γ(G)+γ(H)|X|; if there exists ˜DGMDS_(G) and ˜DHMDS_(H) such that X˜DG and Y˜DH, then γ(GXYH)|(˜DGX)˜DH|=|˜DG||X|+|˜DH|=γ(G)+γ(H)|X|, which also implies that γ(GXYH)=γ(G)+γ(H)|X|. Meanwhile, if γ(GXYH)=γ(G)+γ(H), then from γ(GXYH)|DGDH|=|DG|+|DH||DGDH|γ(G)+γ(H)0γ(G)+γ(H), we obtain that DGDH=.

    (b) () Let DGMDS_(GX) and DHMDS_(GY). Then DGDH is a dominating set of GXYHX. So γ(GXYHX)|DGDH|=γ(G)|X|+γ(H)|Y|=γ(G)+γ(H)2|X|. By Item (a), we have γ(G)+γ(H)2|X|γ(GXYH)|X|. By Observation 1.2, X is an st-critical vertex-set of GXYH.

    () We are going to prove the sufficiency by induction on |X|. When |X|=1, we let X={x}, Y={y} and J=GH. If γ(GxyH)=γ(G)+γ(H), then by Lemma 2.6, neither x nor y is critical in J. Thus γ(G)+γ(H)1=γ(GxyH)1=γ(GxyHx)=γ(Gx)+γ(Hy)γ(G)+γ(H), a contradiction. So we have γ(GxyH)=γ(G)+γ(H)1 by Item (a). Thus γ(G)1+γ(H)1=γ(GxyH)1=γ(GxyHx)=γ(Gx)+γ(Hy)γ(G)1+γ(H)1, from which we have γ(Gx)=γ(G)1 and γ(Hy)=γ(H)1, and so the sufficiency holds.

    Suppose that the sufficiency holds when |X|=n (n1). We consider the case when |X|=n+1 below. Let xX, yY, X0=X{x}, Y0=Y{y}, J=GX0Y0H and J=GXYH. Let D1MDS_(GX), D2MDS_(HY) and D=D1XD2. Since X is an st-critical vertex-set of J, it follows that DMDS_(J). Also, D is a dominating set of Jy. So γ(Jy)|D|=γ(J). By Definition 1.1 and Lemma 2.6, we have

    γ(Jy){=γ(J)1=γ(J),ify is a critical vertex of J,γ(J)γ(J),ify is not a critical vertex of J,

    from which we know γ(Jy)γ(J). Thus γ(Jy)=γ(J). Therefore γ((Jy)X)=γ(GX)+γ(HY)=γ(JX)=γ(J)|X|=γ(Jy)|X|, which implies that X is, and so X0 is, an st-critical vertex-set of Jy. Since γ((JX0)x)=γ(JX)=γ(J)|X|=γ((J)|X0|)|{x}|=γ(JX0)1, we have x is a critical vertex of JX0. Note that Jy=GX0Y0(Hy) and JX0=(GX0)xy(HY0). By the inductive hypothesis, we have that X0 is an st-critical vertex-set of G and x is a critical vertex of GX0. Hence γ(GX)=γ((GX0)x)=γ(GX0)1=γ(G)|X0|1=γ(G)|X|. That is to say, X is an st-critical vertex-set of G. Symmetrically, one can prove that Y is an st-critical vertex-set of H. Thus the result is true when |X|=n+1. Item (b) follows.

    (c) It is an immediate result of Items (b) and (a).

    Remark for Theorem 2.2(a). Let J=GXYH. In this item, we give a sufficient condition for γ(G)+γ(H)|X|=γ(J) and a necessary condition for γ(J)=γ(G)+γ(H). However, neither of them is sufficient and necessary condition. Here, we give the counter examples.

    (I) As shown in Figure 3(i-1) and (i-2), we have γ(G)+γ(H)|X|=γ(J), but X is not an st-critical vertex-set of G and XDG for any DGMDS_(G).

    Figure 3.  Counter examples mentioned in the remark for Theorem 2.2(a).

    (II) As shown in Figure 3(ii-1), we have {r,x4} and {s,x1} are unique minimum dominating sets of G and H respectively, and {r,x4}{s,x1}=, but γ(J)γ(G)+γ(H); and in (ii-2), we have DGDH= for any DGMDS_(G) and DHMDS_(H) (because XDG= for any DGMDS_(G)), but γ(J)γ(G)+γ(H).

    Theorem 2.3. Let G and H be two disjoint graphs. Let XiV(G) for i=1,2,,k and YjV(H) for j=1,2,,l with |X1|=|Y1|. Then {X1,X2,,Xk} and {Y1,Y2,,Yl} are st-critical vertex-set partitions of G and H respectively if and only if {X1,X2,,Xk,Y2,Y3,,Yl} is an st-critical vertex-set partition of G.X1Y1H.

    Proof. Let X={X1,X2,,Xk}, Y={Y1,Y2,,Yl} and X.Y={X1,X2,,Xk,Y2,Y3,,Yl}.

    () Clearly, X.Y is a partition of V(GX1Y1H). For any SX.Y, we have SX or SY. If SX, then by Theorem 2.2(c), we have γ(GX1Y1HS)γ(GS)+γ(HY1)=γ(G)|S|+γ(H)|X1|=γ(GX1Y1H)|S|. Similarly, we can also prove that γ(GX1Y1HS)γ(GX1Y1H)|S| if SY. The necessity follows.

    () Clearly, X and Y are partitions of V(G) and V(H), respectively. Firstly, by Theorem 2.2 (b), X1 and Y1 are st-critical vertex-sets of G and H, respectively. Secondly, for any SX{X1}, we let ˙DMDS_(GX1Y1HS), L=X1(X1˙D) and LG be the subset of L that can be dominated by ˙DV(G) in GX1Y1H. Let H be the subgraph of GX1Y1H induced by V(HY1)X1. Then ˙DV(G) and ˙DV(H) are dominating sets of (GS)(LLG) and HLG, respectively. So

    |˙D|=|˙DV(G)|+|˙DV(H)||˙DX1|γ((GS)(LLG))+γ(HLG)|X1˙D|γ(GS)|LLG|+γ(H)|LG||X1˙D|(by Observation 1.2)γ(G)|S|+γ(H)|X1|=γ(GX1Y1H)|S|(by Theorem 2.2 (c))=|˙D|.

    By the forth equality, we have γ(GS)=γ(G)|S|. Thirdly, for any SY{Y1}, we can similarly prove that γ(HS)=γ(H)|S|. From these three observations, the sufficiency follows.

    In this section, we write dG()=d(), NG()=N(), NG[]=N[] and DG=D, as well as C4C4=(C4)2, C4C4C4=(C4)3 and so on for belief.

    Lemma 3.1. [25] An st-critical vertex-set of a graph G is a 2-packing of G.

    Lemma 3.2. [24] If d(u)=1 and vN(u), then v is not a critical vertex of G. (This implies that a vertex-critical graph has no vertices of degree one.)

    Lemma 3.3. [25] Let S be an st-critical vertex-set of G. If DMDS_(GS), then |D|=γ(G)|S| and DN(S)=.

    Lemma 3.4. Let S be an st-critical vertex-set of G and wV(GS).

    (a) If zN(w)S, then there exists v0N(w){z} such that N(v0)S=.

    (b) Let uvwz be a path or a cycle in G(i.e. u=z is possible). If u,zS, then d(w)>2.

    (c) Let X=N(w). If 2|X|3 and N(x)S for every xX, then |N(X)S|=1.

    (d) Let uvwyz be a trail in G. If u,zS and d(w)=2, then u=z.

    Proof. (a) Suppose to the contrary that N(v)S for every vN(w){z}. Then N[w]{z}N(S). By Lemma 3.3, there exists DMDS_(GS) such that D(N[w]{z})=. However, we see that D can not dominate w in GS, a contradiction.

    (b) It is an immediate result of Item (a).

    (c) Suppose to the contrary that |N(X)S|1. By Lemma 3.1, we have |N(X)S||X|. This implies |N(X)S|=2 or 3. Let {r,s}N(X)S. Then N(r)N(s)=. So we must have that at least one of r and s, say r, is adjacent to only one element of X. Thus we may suppose that {r}=N(x)S, where xX. Note that N(x)S for every xX implies XN(S). By Lemma 3.3, there exists DMDS_(GS) such that DX= and |D|+|S|=γ(G). In order to dominate w in GS, we have wD. However, (D{w})(S{r}){x} is a dominating set of G with cardinality γ(G)1, a contradiction.

    (d) It is an immediate result of Item (c).

    Theorem 3.5. There exists a connected graph G such that V(G) can be partitioned into l strong critical vertex-sets if and only if l{2,3,5}.

    Proof. () Let kZ+ and H4,8 be the (Harary) graph as shown in Figure 1. Based on the fact that Z+{2,3,5}={1}{3kk2}{3k+1k1}{3k+2k2}, we let

    G={K1,ifl=1,(C4)k,ifl{3kk2}{3k+1k1},H4,8(C4)k2,ifl{3k+2k2}.

    Noting that V(C4) and V(H4,8) can be partitioned into 4 and 8 st-critical vertex-sets respectively, we can recursively deduce that V((C4)k) and V(H4,8(C4)k2) can be partitioned into 3k+1 (k1) and 3k+2 (k2) st-critical vertex-sets respectively by Theorem 2.3. Also, note that V((C4)2) can be partitioned into 6 st-critical vertex-sets. So V((C4)k) can be partitioned into 3k (k2) st-critical vertex-sets. The sufficiency follows.

    () Suppose to the contrary that l{2,3,5}. If l=2, then by Lemma 3.1, we get that d(h)=1 for every hV(G), which implies that GK2, contradicting the fact that K2 is not a vertex-critical graph. If l=3, then by Lemmas 3.2 and 3.1, we deduce that d(h)=2 for every hV(G), which implies that G is a cycle. However, one can check that this is impossible. (According to the two well-known facts that γ(Cn)=n3 and γ(Pn)=n3, we can easily deduce that a cycle of order at least 4 can not own an st-critical vertex-set of cardinality 2.)

    If l=5, then we let V(G)=S1S2S3S4S5 be an st-critical vertex-set partition of G. By Lemmas 3.2 and 3.1, we have that 2d(g)4 for every gV(G).

    Claim 1. Let {j,k,l,m,n}={1,2,3,4,5}. If N(sn)={sj,sk,sl}, where siSi for every i{j,k,l,n}, then |N({sj,sk,sl})Sm|=1.

    For convenience, suppose without loss of generality that (j,k,l,m,n)=(1,2,3,4,5). We use reduction to absurdity. Assume that |N({s1,s2,s3})S4|1. Then since

    N(s5)={s1,s2,s3},

    by the contrapositive of Lemma 3.4(c), at least one of s1, s2 and s3, say s1, satisfies N(s1)S4=. Thus N(s1){s5}S2S3. To combine this with Lemma 3.4(b), we must have d(s1)2, which implies that d(s1)=3. So N(s1)S2 and N(s1)S3.

    Since s1N(s5)S1, by Lemma 3.4(a), one of N(s2)S1 and N(s3)S1, say N(s2)S1, is empty set. By Lemma 3.1, we have (N(s2){s5})(S2S5)=. Since s3N(s5) and N(s1)S3, by Lemma 3.4(a), we obtain that N(s2)S3=. So by Lemma 3.2, we have N(s2)S4. Let N(s2)S4={s4}. Then

    N(s2)={s4,s5}.

    So we have N(s4)S5= by the contrapositive of Lemma 3.4(b). Thus N(s4){s2}S1S3. Since d(s2)=2, we have s4s3E(G) or s4s1E(G) by Lemma 3.4(d). However, we have supposed that N(s1)S4= in the third sentence of the first paragraph. Thus, only s4s3E(G) holds, which implies that

    N(s4)={s2,s3}.

    So by Lemma 3.4(b), we have

    N(s3)S2=.

    Now, if N(s3)S1=, then s1 is a cut-vertex of G (refer to Figure 4(i-a)). Thus by Theorem 2.3, G[{s1,s2,s3,s4,s5}] is vertex-critical. However, one can check that it is not true. If N(s3)S1, let N(s3)S1={r1}. (r1=s1 is possible.) Then {r1}{s1} is a vertex-cut of G (refer to Figure 4 (i-b)). By Lemmas 3.2 and 1.7 and Theorem 2.3, G[{s1,s2,s3,s4,s5,r1}] is vertex-critical, which is also not true.

    Figure 4.  Illustration for the proofs of Claim 1 and Claim 2-A.

    Claim 2. d(g)3 for every gV(G).

    Without loss of generality, suppose to the contrary that there exists s5S5 such that N(s5)={s1,s2,s3}, where siSi, i=1,2,3. By Claim 1, we can let

    N({s1,s2,s3})S4={s4}. (3.1)

    Case A. At least two of s1, s2 and s3, say s2 and s3, have degree 2 in G.

    Then, by Lemma 3.4(b), we obtain that N(s2)S1= and N(s3)S1=, as well as N(s2)S3= and N(s3)S2=. So we must have N(s2)S4 and N(s3)S4 because d(s2)=d(s3)=2. By (3.1), we have N(s2)S4={s4}=N(s3)S4. Again by Lemma 3.4(b), we have N(s4)S5=.

    If N(s4)S1, then by Lemma 3.4(d), we have N(s4)S1={s1}. From this, we see that either G=G[{s1,s2,s3,s4,s5}], or s1 is a cut-vertex of G (no matter N(s4)S1= or not). Altogether, we have G[{s1,s2,s3,s4,s5}] is vertex-critical by Theorem 2.3. This is not true (refer to Figure 4(ii-A).)

    Case B. At most one of s1, s2 and s3, say s1, has degree 2 in G.

    Then d(s2),d(s3)3. Since s1N(s5), by Lemma 3.4(a), at least one of N(s2)S1= and N(s3)S1=, say N(s2)S1=, holds. So N(s2)S3S4S5, and thus d(s2)=3. This implies that N(s2)S4 and N(s2)S3. From the former, we get N(s2)S4={s4} while by the latter we can let N(s2)S3={r3}. (r3=s3 is possible.) Since s3N(s5), we get

    N(s1)S3= (3.2)

    by Lemma 3.4(a). There are two subcases.

    When N(s1)S2=, we have N(s1)S4 since d(s1)2. By (3.1), we have N(s1)S4={s4}. So N(s1)={s4,s5}. Thus by Lemma 3.4(b), we have N(s4)S5=. Since r3N(s2), we get N(s4)S3= by Lemma 3.4(a), and so N(s4)={s1,s2}. Now, we see that {r3}{s3} is a vertex-cut of G. (Refer to Figure 5(ii-B1).) By Theorem 2.3, G[{s1,s2,s3,s4,s5,r3}] is vertex-critical, which is not true.

    Figure 5.  Illustration for the proofs of Claim 2-B and Claim 3.

    When N(s1)S2, by (3.2) and Lemma 3.4(b), we have N(s1)S4, which implies that N(s1)S4={s4}. Since s2N(s5), we have N(s3)S2= by Lemma 3.4(a). So d(s3)=3, and thus N(s3)S1 and N(s3)S4. By (3.1), we have N(s3)S4={s4}. (Refer to Figure 5(ii-B2).) Now, we have r3N(s2), N(s4)S3={s3} and N(s5)S3={s3}. However, according to Lemma 3.4(a), this is impossible.

    Claim 3. d(g)4 for every gV(G).

    Without loss of generality, suppose to the contrary that there exists some s5S5 such that N(s5)={s1,s2,s3,s4}, where siSi, i=1,2,3,4. For every 1i4, by Lemma 3.4(b) and Claim 2, we have d(si)2 and d(si)3, which implies that d(si)=4. (Refer to Figure 5(iii).) However, by Lemma 3.4(a), this is impossible.

    Now, by Claims 2 and 3, we get that dH(g)=2 for every gV(G), which implies that G is a cycle, a contradiction. The necessity follows.

    In [25], the authors got the following Proposition 4.1, which tells us thatC4={C4}, where C4 was defined in the last paragraph of Section 1. It is easy to see that the circulant graph C121,5, the vertex coalescence C4C4 and the Harary graph H4,6 (see Figure 6) belong to C6. Referring to Proposition 4.1, we want to know whether C6 is a finite set. So we present Problem 4.2.

    Figure 6.  Three elements of C6.

    Proposition 4.1. [25] Let H be a connected graph. Then V(H) can be partitioned into 4 strong critical vertex-sets if and only if HC4.

    Problem 4.2. Give a constructive characterization of the connected graphs G such that V(G) can be partitioned into 6 strong critical vertex-sets of G.

    It is known that each graph has a degree sequence, but a given sequence may not be a degree sequence of any simple graph. For instance, the sequence (7, 6, 5, 4, 3, 3, 2) cannot become a degree sequence of a simple graph (see [3], Ex. 1.5.6). If V(G)=S1S2Sl is a strong critical vertex-set partition of a graph G, then we call the sequence (|S1|,|S2|,,|Sl|) a strong critical vertex-set sequence of G. It is noteworthy that even a connected graph may own different strong critical vertex-set sequences. For example, both (3, 2, 2, 1, 1, 1, 1, 1, 1) and (2, 2, 2, 2, 1, 1, 1, 1, 1) are strong critical vertex-set sequences of the graph depicted in Figure 7. Also, for connected graphs, it follows from Theorem 3.5 that the strong critical vertex-set sequence (1, 1, 1, 1) exists but (1, 1, 1, 1, 2) does not exist.

    Figure 7.  A graph with more than one strong critical vertex-set sequences.

    Problem 4.3. What kinds of strong critical vertex-set sequences do exist? Or to be concrete about it, if (|S1|,|S2|,,|Sl|) is a strong critical vertex-set sequence of a connected graph G, then what are the relations of |S1|,|S2|,,|Sl|, l and γ(G)?

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    (1) Thank Professor D. F. Rall for his suggestion on the definition of strong critical vertex-set, and e-mailing the article (ref. [21]) to us!

    (2) Funding Information:

    (i) National Natural Science Foundation of China (No. 12061047);

    (ii) National Natural Science Foundation of Fujian Province (No. 2022J05199);

    (iii) Jiangxi Provincial Natural Science Foundation (No. 20212BAB201027);

    (iv) Scientific Research Project Funding Plan of Jianghan University (No. 2022XKZX11).

    The authors declare that they have no conflicts of interest in this work.



    [1] R. Arefidamghani, R. Behling, Y. Bello-Cruz, A. Iusem, L. Santos, The circumcentered-reflection method achieves better rates than alternating projections, Comput. Optim. Appl., 79 (2021), 507–530. http://dx.doi.org/10.1007/s10589-021-00275-6 doi: 10.1007/s10589-021-00275-6
    [2] H. Attouch, On the maximality of the sum of two maximal monotone operators, Nonlinear Anal.-Theor., 5 (1981), 143–147. http://dx.doi.org/10.1016/0362-546X(81)90039-0 doi: 10.1016/0362-546X(81)90039-0
    [3] H. Bauschke, P. Combettes, Convex analysis and monotone operator theory in Hilbert spaces, Cham: Springer, 2 Eds., 2017. http://dx.doi.org/10.1007/978-3-319-48311-5
    [4] H. Bauschke, H. Ouyang, X. Wang, On circumcenters of finite sets in Hilbert spaces, arXiv: 1807.02093.
    [5] R. Behling, Y. Bello-Cruz, L. Santos, Circumcentering the Douglas-Rachford method, Numer. Algor., 78 (2018), 759–776. http://dx.doi.org/10.1007/s11075-017-0399-5 doi: 10.1007/s11075-017-0399-5
    [6] R. Behling, Y. Bello-Cruz, L. Santos, On the linear convergence of the circumcentered-reflection method, Oper. Res. Lett., 46 (2018), 159–162. http://dx.doi.org/10.1016/j.orl.2017.11.018 doi: 10.1016/j.orl.2017.11.018
    [7] R. Behling, Y. Bello-Cruz, L. Santos, On the circumcentered-reflection method for the convex feasibility problem, Numer. Algor., 86 (2021), 1475–1494. http://dx.doi.org/10.1007/s11075-020-00941-6 doi: 10.1007/s11075-020-00941-6
    [8] J. Benoist, The Douglas-Rachford algorithm for the case of the sphere and the line, J. Glob. Optim., 63 (2015), 363–380. http://dx.doi.org/10.1007/s10898-015-0296-1 doi: 10.1007/s10898-015-0296-1
    [9] D. Bertsekas, Convex optimization theory, Melrose: Athena Scientific, 2009.
    [10] J. Borwein, A. Lewis, Convex analysis and nonlinear optimization: theory and examples, New York: Springer, 2 Eds., 2006. http://dx.doi.org/10.1007/978-0-387-31256-9
    [11] S. Boyd, L. Vandenberghe, Convex optimisation, Cambridge: Cambridge University Press, 2004. http://dx.doi.org/10.1017/CBO9780511804441
    [12] S. Boyd, N. Parikh, E. Chu, B. Peleato, J. Eckstein, Matlab scripts for alternating direction method of multipliers, 2011. Available from: https://web.stanford.edu/~boyd/papers/admm/.
    [13] S. Boyd, N. Parikh, E. Chu, B. Peleato, J. Eckstein, Distributed optimisation and statistical learning via the alternating direction method of multipliers, New York: Now Foundations and Trends, 2010. http://dx.doi.org/10.1561/2200000016
    [14] M. Dao, M. Tam, A Luapunov-type approach to convergence of the Douglas-Rachford algorithm, J. Glob. Optim., 73 (2019), 83–112. http://dx.doi.org/10.1007/s10898-018-0677-3 doi: 10.1007/s10898-018-0677-3
    [15] N. Dizon, J. Hogan, S. Lindstrom, Circumcentered reflections method for wavelet feasibility problems, ANZIAM J., 62 (2020), C98–C111. http://dx.doi.org/10.21914/anziamj.v62.16118 doi: 10.21914/anziamj.v62.16118
    [16] N. Dizon, J. Hogan, S. Lindstrom, Centering projection methods for wavelet feasibility problems, In Current trends in analysis, its applications and computation, Cham: Birkhäuser, 2022. http://dx.doi.org/10.1007/978-3-030-87502-2_66
    [17] Neil Dizon, J. A. Hogan, and Scott B. Lindstrom, Circumcentering reflection methods for nonconvex feasibility problems. Set-Valued Var. Anal., 30 (2022), 943–973. http://dx.doi.org/10.1007/s11228-021-00626-9
    [18] E. Dolan, J. Moré, Benchmarking optimisation software with performance profiles, Math. Program., 91 (2022), 201–213. http://dx.doi.org/10.1007/s101070100263 doi: 10.1007/s101070100263
    [19] J. Eckstein, W. Yao, Understanding the convergence of the alternating direction method of multipliers: theoretical and computational perspectives, Pac. J. Optim., 11 (2015), 619–644.
    [20] D. Gabay, Applications of the method of multipliers to variational inequalities, Studies in Mathematics and its Applications, 15 (1983), 299–331. http://dx.doi.org/10.1016/S0168-2024(08)70034-1 doi: 10.1016/S0168-2024(08)70034-1
    [21] O. Giladi, B. Rüffer, A Lyapunov function construction for a non-convex Douglas-Rachford iteration, J. Optim. Theory Appl., 180 (2019), 729–750. http://dx.doi.org/10.1007/s10957-018-1405-3 doi: 10.1007/s10957-018-1405-3
    [22] B. He, X. Yuan, On the O(1/n) convergence rate of the Douglas-Rachford alternating direction method, SIAM J. Numer. Anal., 50 (2012), 700–709. http://dx.doi.org/10.1137/110836936 doi: 10.1137/110836936
    [23] B. He, X. Yuan, On the convergence rate of Douglas-Rachford operator splitting method, Math. Program., 153 (2015), 715–722. http://dx.doi.org/10.1007/s10107-014-0805-x doi: 10.1007/s10107-014-0805-x
    [24] S. B. Lindstrom, Computable centering methods for spiralling algorithms and their duals with motivations from the theory of Lyapunov functions, Comput. Optim. Appl., 83 (2022), 999–1026. http://dx.doi.org/10.1007/s10589-022-00413-8 doi: 10.1007/s10589-022-00413-8
    [25] S. B. Lindstrom, B. Sims, Survey: sixty years of Douglas-Rachford, J. Aust. Math. Soc., 110 (2021), 333–370. http://dx.doi.org/10.1017/S1446788719000570 doi: 10.1017/S1446788719000570
    [26] P. Lions, B. Mercier, Splitting algorithms for the sum of two nonlinear operators, SIAM J. Numer. Anal., 16 (1979), 964–979. http://dx.doi.org/10.1137/0716071 doi: 10.1137/0716071
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1414) PDF downloads(84) Cited by(0)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog