1 Introduction

Let \(G\) be a graph and \(H\) a component of \(G\). For any \(v\in {V(G)}\), we use \(N_G(v)\) to denote the set of neighbors of \(v\) in \(G\). For any \(e\in {E(G)}\), we use \(V_G(e)\) to denote the set of vertices of \(G\) incident with \(e\). (If understood, we omit the reference to G.) If \(G\ne \emptyset \) and \(G-v\) has a perfect matching for every \(v\in V(G)\), then \(G\) is called factor-critical. Factor-critical graphs have been extensively studied in the past [2, 4, 5, 7, 8]. We say that \(H\) is odd if \(|H|\) is odd and that \(H\) is a tight component (respectively, anti-tight component) if

  • \(H\) is factor-critical, and

  • \(|N(v)\cap V(e)|\ne 1\) (respectively, \(|N(v)\cap V(e)|\ne 2\)) for any \(v\in V(H)\), any perfect matching \(M\) in \(H-v\), and any \(e\in M\).

When \(H=G\), we simply say that \(G\) is tight (respectively, anti-tight).

Lee et al. [6] used tight components to solve a problem of Bollob\(\acute{\mathrm{a}}\)s and Scott [1, 9] about the dependence on minimum degree of bounds on judicious bisections. A block in a graph \(G\) is a maximal connected subgraph that contains no cut vertex. Hence, if a block of G is not 2-connected, then it must be induced by a cut edge of \(G\). A complete subgraph of a graph is usually called a clique. It is easy to see that odd cliques are tight. Lee et al. [3] observed that if every block of a connected graph \(G\) is an odd clique then \(G\) is tight, and mentioned that it is not clear if every tight component is of this form. In this note, we answer this question in the affirmative.

Theorem 1

A connected graph is tight iff every block of the graph is an odd clique.

It is apparent that if a graph \(G\) is factor-critical and triangle-free, then \(G\) is anti-tight. But it is not true that every anti-tight graph \(G\) is triangle-free (see Fig. 1). It seems difficult to characterize the anti-tight graphs. A connected graph \(G\) is minimally anti-tight if \(G\) is anti-tight but \(G-e\) is not anti-tight for every \(e\in E(G)\). Concerning the minimally anti-tight graphs, we have

Fig. 1
figure 1

An anti-tight graph with a triangle

Theorem 2

Every minimally anti-tight graph \(G\) is triangle-free.

Let \(G\) be a graph and \(M\) a matching in \(G\). A vertex \(v\) is called matched if it is incident with an edge in \(M\), and otherwise \(v\) is called unmatched. An M-alternating path is a path in which the edges belong alternatively to \(E(G)\setminus M\) and \(M\). An M-augmenting path is an alternating path that starts from and ends at unmatched vertices. Let \(H\) be a subgraph of \(G\), we use \(M\Delta H\) to denote the symmetric difference of \(M\) and \(E(H)\). In general, we follow the terminology in [3].

2 Proof of Theorem 1

Lemma 1

Let \(G_1\) and \(G_2\) be subgraphs of a connected graph \(G\) such that \(G= G_1\cup G_2\) and \(|V(G_1)\cap V(G_2)|=1\). Then \(G\) is tight iff both \(G_1\) and \(G_2\) are tight.

Proof

Let \(x\in V(G_1)\cap V(G_2)\). Suppose G is tight. Let \(v\in V(G_1)\). Then \(G-v\) has a perfect matching, say \(M\). Note that \(|V(G_1)|\) is odd and thus the edge in \(M\) incident with \(x\) belongs to \(G_1\) when \(x\ne v\); so \(M\cap E(G_1)\) is a perfect matching in \(G_1-v\). Now let \(M_1\) be a perfect matching in \(G_1-v\). Since \(G-x\) has a perfect matching, \(G_2-x\) has a perfect matching, say \(M_2\). Then \(M:=M_1\cup M_2\) is a perfect matching in \(G-v\). Thus, since \(G\) is tight, for any \(e\in M_1\), \(|N_G(v)\cap V(e)|\ne 1\); so \(|N_{G_1}(v)\cap V(e)| = |N_G(v)\cap V(e)|\ne 1\). Hence, \(G_1\) is tight. Similarly, we can show that \(G_2\) is tight.

Now assume both \(G_1\) and \(G_2\) are tight. Let \(v\in V(G)\). Then \(v\in V(G_i)\) for some \(i\in {1,2}\) and, without loss of generality, let \(v\in V(G_1)\). Let \(M_1\) be a perfect matching in \(G_1-v\), and \(M_2\) a perfect matching in \(G_2-x\); so \(M_1\cup M_2\) is a perfect matching in \(G-v\). Now for any perfect matching \(M\) in \(G-v\), \(M\cap E(G_1)\) is a perfect matching in \(G_1-v\) while \(M\cap E(G_2)\) is a perfect matching in \(G_2-x\). Moreover, for any \(e\in M\cap E(G_1)\), \(|N_G(v)\cap V(e)| =|N_{G_1}(v)\cap V(e)|\ne 1\) (as \(G_1\) is tight). Now suppose \(e\in M\cap E(G_2)\). If \(v\ne x\) then \(|N_G(v)\cap V(e)|=0\ne 1\). So assume \(v=x\). Then since \(G_2\) is tight, \(|N_G(v)\cap V(e)| =|N_{G_2}(v)\cap V(e)|\ne 1\). Hence, \(G\) is also tight. \(\square \)

Lemma 2

Let \(G\) be a \(2\)-connected tight graph. Then \(G\) is a complete graph with odd order.

Proof

Suppose to the contrary that \(G\) is not a complete graph. Then we can find three vertices \(x,y,z\) such that \(yx\in E(G)\), \(yz\in E(G)\) but \(xz\notin E(G)\). As \(G-y\) has a perfect matching, say \(M\), we let \(xx_1\in M\) and \(zz_1\in M\). Since \(|\{x,x_1\}\cap N_G(y)|\ne 1\) and \(|\{z,z_1\}\cap N_G(y)|\ne 1\), \(x_1y\in E(G)\) and \(z_1y\in E(G)\). Note that \(G':=G-\{x,y,z\}\) has no perfect matching, otherwise, \(G-\{x\}\) has a perfect matching containing \(yz\) and \(|N_G(x)\cap \{y,z\}|=1\), a contradiction. Then \(x_1z_1\notin E(G)\). If \(xz_1\in E(G)\), \(G-z\) has a perfect matching \(M'=(M-\{xx_1,zz_1\})\cup \{xz_1,yx_1\}\) and \(|N_G(z)\cap \{x,z_1\}|=1\), a contradiction. Hence, \(xz_1\notin E(G)\). Similarly, \(zx_1\notin E(G)\).

Suppose \(G-y\) has an \(M\)-alternating path \(P'=xy_2\cdots y_kz\). If \(xy_2\in M\) and \(y_kz\in M\), then \(y_2=x_1\) and \(y_k=z_1\), hence, \(M\Delta yxy_2\cdots y_k\) is a perfect matching in \(G-z\) and \(|N_G(z)\cap \{y,x\}|=1\), a contradiction. If \(xy_2\in M\) and \(y_kz\notin M\), then \(y_2=x_1\) and \(y_k\ne z_1\), clearly, \(M\Delta y_2\cdots y_kzz_1y\) is a perfect matching in \(G-x\) and \(|N_G(x)\cap \{z_1,y\}|=1\), a contradiction. If \(xy_2\notin M\) and \(y_kz\in M\), then we get a similar contradiction. Thus, \(xy_2\notin M\) and \(y_kz\notin M\). Then \(y_2\ne x_1\) and \(y_k\ne z_1\). So \(M\Delta xy_2\cdots y_kzz_1y\) is a perfect matching in \(G-x_1\) and \(|N_G(x_1)\cap \{z_1,y\}|=1\), a contradiction.

Thus, \(G-y\) does not contain any \(M\)-alternating path from \(x\) to \(z\). If \(x_1\) and \(z\) are linked by a path \(P_1\) in \(G-\{x,y\}\), let \(P=xx_1+P_1=xx_1v_3\cdots v_iv_{i+1}v_{i+2}\cdots z\) and assume that \(xx_1v_3\cdots v_iv_{i+1}\) is an \(M\)-alternating path, \(v_{i-1}v_i\in M\) but \(v_iv_{i+1}, v_{i+1}v_{i+2} \notin M\) for some \(i\). Choose \(P\) so that \(xx_1v_3\cdots v_iv_{i+1}\) is longest and \(v_{i+1}v_{i+2}\cdots z\) is shortest. Then the matched vertex \(v'_{i+1}\) with \(v_{i+1}\) is not in \(P\) as \(xx_1, v_3v_4, \cdots \), \(v_{i-1}v_i \in M\) and \(v_{i+1}v_{i+2}\cdots z\) is shortest. If \(v_iv'_{i+1}\notin E(G)\), then \(M\Delta yxx_1v_3\cdots v_i\) is a perfect matching in \(G-v_{i}\) and \(|N_G(v_i)\cap \{v_{i+1},v'_{i+1}\}|=1\), a contradiction. Thus \(v_iv'_{i+1}\in E(G)\). Now \(xx_1v_3\cdots v_iv'_{i+1}\) \(v_{i+1}v_{i+2}\cdots z\) is a path contradicts the choice of \(P\), as it has a longer \(M\)-alternating section starting from \(x\).

Hence \(x_1\) and \(z\) are not linked by any path in \(G-\{x,y\}\). Then since \(G\) is \(2\)-connected, there is a path \(P=xv_2\cdots z\) in \(G-y\) such that \(x_1\notin P\). Let \(P=xv_2\cdots v_iv_{i+1}v_{i+2}\cdots z\) such that \(xv_2v_3\cdots v_iv_{i+1}\) is an \(M\)-alternating path, \(v_{i-1}v_i\in M\) but \(v_iv_{i+1}, v_{i+1}v_{i+2} \notin M\). Choose \(P\) so that \(xv_2\cdots v_iv_{i+1}\) is longest and \(v_{i+1}v_{i+2}\cdots z\) is shortest. Then the matched vertex \(v'_{i+1}\) with \(v_{i+1}\) is not in \(P\) as \(v_{i+1}\ne x_1\) and \(v_2v_3, \cdots , v_{i-1}v_i \in M\) and \(v_{i+1}v_{i+2}\cdots z\) is shortest. If \(v_iv'_{i+1}\notin E(G)\), then \(M\Delta yx_1xv_2\cdots v_i\) is a perfect matching in \(G-v_{i}\) and \(|N_G(v_i)\cap \{v_{i+1},v'_{i+1}\}|=1\), a contradiction. Thus \(v_iv'_{i+1}\in E(G)\). So \(xv_2v_3\cdots v_iv'_{i+1}v_{i+1}v_{i+2}\cdots z\) is a path contradicts the choice of \(P\), as it has a longer \(M\)-alternating section starting from \(x\). \(\square \)

A simple corollary of Lemma 1 is the observation of Lee, Loh and Sudakov that if every block of a graph \(G\) is an odd clique then \(G\) is a tight component. Clearly, Lemma 1 and Lemma 2 imply Theorem 1 which says the converse.

3 Proof of Theorem 2

Assume to the contrary that \(G\) is minimally anti-tight and contains a triangle \(xyzx\). If \(G-xz-v\) has a perfect matching for every \(v\in V(G)\), every such matching \(M\) is also a perfect matching in \(G-v\). Since \(G\) is anti-tight, \(|N_{G-xz}(v)\cap V(e)|\le |N_{G}(v)\cap V(e)|\le 1\) for every \(e\in M\). Hence, \(G-xz\) is anti-tight, a contradiction.

Thus, there is a vertex \(u\in V(G)\) such that \(G-xz-u\) contains no perfect matching. \(G-y\) has a perfect matching \(M\) and \(xz\notin M\). This implies \(u\ne y\). Let \(\widetilde{M}\) be a perfect matching in \(G-u\). Since \(G-xz-u\) contains no perfect matching, \(xz\in \widetilde{M}\). Let \(G'\) be the multigraph with \(V(G')=V(G)\) and whose edges are edges in \(M\) and \(\widetilde{M}\). It is easy to see that \(d_{G'}(u)=d_{G'}(y)=1\) and for any \(v\in V(G)-\{u,y\}\), \(d_{G'}(v)=2\). So there is a path \(P=uv_1v_2\cdots v_ty \) in \(G'\) between \(u\) and \(y\), and the length of \(P\) is even. If \(xz\notin P\), then \(xz\) is an edge of an even cycle \(C\) which is a component of \(G'\), and \(\widetilde{M}\Delta C\) is a perfect matching of \(G-xz-u\), a contradiction. Hence, \(xz\in P\), and let \(\{v_i,v_{i+1}\}=\{x,z\}\). Since \(v_iv_{i+1}\notin M\), the length of \(uv_1\cdots v_i\) is odd and \(C_1:=v_{i}\cdots v_ty+yv_i\) is an even cycle in \(G\). Now \(\widetilde{M}\Delta C_1\) is a perfect matching of \(G-xz-u\), a contradiction. \(\square \)