arXiv is now an independent nonprofit! Learn more
License: CC BY-NC-SA 4.0
arXiv:2608.18828v1 [math.CO] 19 Aug 2026

Counterexample to the Bougard–Joret Conjecture

Joyentanuj Das Thanks: Department of Mathematics, College of Engineering and Technology, SRM Institute of Science and Technology, Kattankulathur, Chennai 603203, India. E-mail addresses: joyentanuj@gmail.com; joyentad@srmist.edu.in.    and    Sayan Gupta Note: School of Mathematical Sciences, NISER Bhubaneswar (An OCC of Homi Bhabha National Institute, Mumbai, 400094), India. Email: sayan.gupta@niser.ac.in
Abstract

For admissible integers n,α,kn,\alpha,k, let f⁡(n,α,k)f(n,\alpha,k) be the minimum number of edges in a kk-connected graph of order nn and independence number α\alpha. A conjecture of Bougard and Joret predicts that f⁡(n,α,k)=⌈n​k/2⌉f(n,\alpha,k)=\lceil nk/2\rceil when n≤k​αn\leq k\alpha, under the assumptions n≥2​αn\geq 2\alpha, n≥α+kn\geq\alpha+k, α≥2\alpha\geq 2, and k≥3k\geq 3. We disprove this prediction, determine f⁡(n,α,k)f(n,\alpha,k) throughout the boundary n=α+kn=\alpha+k, and characterize every extremal graph on that boundary. In particular, for every k≥4k\geq 4,

f⁡(2​k−1,k−1,k)=k2−1,f(2k-1,k-1,k)=k^{2}-1,

whereas the conjectured value is k2−⌊k/2⌋k^{2}-\lfloor k/2\rfloor. The extremal graphs in this family are precisely K¯k−1∨T\overline{K}_{k-1}\mathbin{\vee}T, where TT is an arbitrary tree of order kk. The smallest-order failure has parameters (n,α,k)=(7,3,4)(n,\alpha,k)=(7,3,4), and no admissible counterexample has smaller order.

Keywords. Vertex connectivity, independence number, extremal size, Turán theorem

MSC 2020. 05C35, 05C40.

1 Introduction and literature

All graphs in this paper are finite, simple, and undirected. The problem of minimizing the size of a graph subject to a prescribed independence number is a complementary formulation of the classical extremal problem of Turán [7]. If connectivity is not required, the unique minimizer of order nn and independence number α\alpha is the disjoint union of α\alpha cliques whose orders differ by at most one. Ore asked for the corresponding minimum under connectedness [6]. The connected problem was solved independently by Christophe et al. [2] and Gitler and Valencia [3]; see also the short arguments of Bougard and Joret [1] and Yuan [9].

For parameters for which the class is nonempty, let f⁡(n,α,k)f(n,\alpha,k) denote the minimum size of a kk-connected graph of order nn and independence number α\alpha. A graph attaining this minimum is called (n,α,k)(n,\alpha,k)-extremal, or simply extremal when the parameters are clear.

Bougard and Joret determined the 22-connected case and then considered higher vertex connectivity [1]. For integers satisfying

n≥2​α,n≥α+k,α≥2,k≥3,n\geq 2\alpha,\qquad n\geq\alpha+k,\qquad\alpha\geq 2,\qquad k\geq 3, (1.1)

they conjectured that

f⁡(n,α,k)={⌈n​k2⌉,n≤k​α,t⁡(n,α)+⌈k​α2⌉,n>k​α.f(n,\alpha,k)=\begin{cases}\left\lceil\dfrac{nk}{2}\right\rceil,&n\leq k\alpha,\\[6.0pt] t(n,\alpha)+\left\lceil\dfrac{k\alpha}{2}\right\rceil,&n>k\alpha.\end{cases} (1.2)

This is Conjecture 1 in Section 6 of [1, pp. 11–12]. Here t⁡(n,α)t(n,\alpha) is the size of the balanced union of α\alpha cliques such that the total order is nn. This notation must be distinguished from the convention in which a Turán graph is complete multipartite. The original paper established (1.2) on the boundary n=k​αn=k\alpha: for α=2\alpha=2 this covers all n≥2​kn\geq 2k, and for α≥3\alpha\geq 3 it covers the range

n≥⌈(k−2)​α2⌉​α+2​α,n\geq\left\lceil\frac{(k-2)\alpha}{2}\right\rceil\alpha+2\alpha,

a bound obtained from a clique-partition theorem [1, Section 6, p. 12]. Wang and Wu later recorded that the conjecture for k≥3k\geq 3 was “still unsettled” [8, Section 1]. Liu and Ning reproduced the statement as their Conjecture 3.1 and described it as “still wide open” [5, Section 3, p. 121]. We are not aware of an earlier counterexample.

The first line of (1.2) starts from the universal inequality δ⁡(G)≥k\delta(G)\geq k, which gives 2​e​(G)≥n​k2\mathrm{e}(G)\geq nk. A degree sum, however, does not detect every connectivity cost. On the boundary n=α+kn=\alpha+k, the admissibility condition n≥2​αn\geq 2\alpha is equivalent to k≥αk\geq\alpha; moreover, n≤k​αn\leq k\alpha holds automatically because α≥2\alpha\geq 2 and k≥3k\geq 3. Deleting a maximum independent set leaves exactly kk vertices. When α=k−1\alpha=k-1, those vertices must induce a connected graph. That residual connectivity costs k−1k-1 edges, whereas the degree-sum estimate for a one-connected graph accounts for only ⌈k/2⌉\lceil k/2\rceil edges. This is the source of the counterexamples.

Section 3 determines the exact boundary value and all graphs attaining it. It shows that (1.2) is correct on n=α+kn=\alpha+k except precisely when k−α=1k-\alpha=1 and k≥4k\geq 4. Section 4 proves that order seven is minimal and records the standard clique-partition estimate relevant to the second line of (1.2). The latter is included to distinguish the unresolved second regime from the boundary obstruction found here.

2 Notation and preliminary facts

For a graph GG, write V⁡(G)V(G) and E⁡(G)E(G) for its vertex and edge sets, e⁡(G)=|E⁡(G)|\mathrm{e}(G)=|E(G)| for its size, δ⁡(G)\delta(G) for its minimum degree, α⁡(G)\alpha(G) for its independence number, and κ⁡(G)\kappa(G) for its vertex connectivity. A graph is kk-connected if it has more than kk vertices and deleting any set of fewer than kk vertices leaves a connected graph. The join G∨HG\mathbin{\vee}H is obtained from disjoint copies of GG and HH by adding every edge with one endpoint in each copy. We write K¯s\overline{K}_{s} for the edgeless graph of order ss and PsP_{s} for the path of order ss.

If n=q​α+rn=q\alpha+r, where 0≤r<α0\leq r<\alpha, define

t⁡(n,α)=r​(q+12)+(α−r)​(q2).t(n,\alpha)=r\binom{q+1}{2}+(\alpha-r)\binom{q}{2}. (2.1)

Thus t⁡(n,α)t(n,\alpha) is the number of edges in r​Kq+1∪(α−r)​KqrK_{q+1}\cup(\alpha-r)K_{q}. Convexity of x↦(x2)x\mapsto\binom{x}{2} implies that among all partitions s1+⋯+sα=ns_{1}+\cdots+s_{\alpha}=n into nonnegative integers, the quantity ∑i(si2)\sum_{i}\binom{s_{i}}{2} is minimized when the sis_{i} differ by at most one; its minimum is (2.1).

For integers p≥2p\geq 2 and 0≤d≤p−10\leq d\leq p-1, let m⁡(p,d)m(p,d) be the minimum size of a graph of order pp that is dd-connected, with the convention that no condition is imposed when d=0d=0.

For d≥2d\geq 2, the next exact minimum follows from Harary’s determination of the maximum vertex connectivity at fixed order and size [4]; the cases d=0,1d=0,1 are elementary. We record the formula for later use, especially its exceptional line for d=1d=1.

Proposition 2.1.

For p≥2p\geq 2,

m⁡(p,d)={0,d=0,p−1,d=1,⌈p​d2⌉,2≤d≤p−1.m(p,d)=\begin{cases}0,&d=0,\\ p-1,&d=1,\\ \left\lceil\dfrac{pd}{2}\right\rceil,&2\leq d\leq p-1.\end{cases} (2.2)

The exceptional second line of (2.2) is essential below. For d=1d=1, the degree-sum quantity ⌈p/2⌉\lceil p/2\rceil equals p−1p-1 only when p∈{2,3}p\in\{2,3\} and is strictly smaller when p≥4p\geq 4.

3 The exact value at the order boundary

On n=α+kn=\alpha+k, the assumptions in (1.1) are equivalent to α≥2\alpha\geq 2, k≥3k\geq 3, and k≥αk\geq\alpha. Every such triple belongs to the first regime of (1.2).

Theorem 3.1.

Let α≥2\alpha\geq 2, k≥3k\geq 3, and k≥αk\geq\alpha. Then

f⁡(α+k,α,k)=α​k+m⁡(k,k−α).f(\alpha+k,\alpha,k)=\alpha k+m(k,k-\alpha). (3.1)

Equivalently,

f⁡(α+k,α,k)={k2,α=k,k2−1,α=k−1,⌈k⁡(α+k)2⌉,2≤α≤k−2.f(\alpha+k,\alpha,k)=\begin{cases}k^{2},&\alpha=k,\\ k^{2}-1,&\alpha=k-1,\\ \left\lceil\dfrac{k(\alpha+k)}{2}\right\rceil,&2\leq\alpha\leq k-2.\end{cases} (3.2)
Proof.

Set n=α+kn=\alpha+k and put

d=k−α.d=k-\alpha.

The assumption k≥αk\geq\alpha gives d≥0d\geq 0, while α≥2\alpha\geq 2 gives d<kd<k. We prove a lower bound valid for every admissible graph and then construct a graph that attains it.

Let GG be a kk-connected graph of order nn with α⁡(G)=α\alpha(G)=\alpha. Choose a maximum independent set S⊆V⁡(G)S\subseteq V(G) and write

T=V⁡(G)∖S.T=V(G)\setminus S.

Then |S|=α|S|=\alpha and

|T|=n−|S|=(α+k)−α=k.|T|=n-|S|=(\alpha+k)-\alpha=k.

Because GG is kk-connected, every vertex has degree at least kk. Now fix s∈Ss\in S. Since SS is independent, ss has no neighbor in SS. Thus all neighbors of ss must lie in the kk-element set TT. The two inequalities

k≤degG⁡(s)≤|T|=kk\leq\deg_{G}(s)\leq|T|=k

force degG⁡(s)=k\deg_{G}(s)=k and NG​(s)=TN_{G}(s)=T. Since this holds for every s∈Ss\in S, all possible edges between SS and TT occur. There are exactly |S|​|T|=α​k|S||T|=\alpha k such edges, while there are no edges inside SS. If

H=G⁡[T],H=G[T],

then the edge set of GG is consequently the disjoint union of the α​k\alpha k edges between SS and TT and the edges of HH. Therefore

e⁡(G)=α​k+e⁡(H).\mathrm{e}(G)=\alpha k+\mathrm{e}(H). (3.3)

We next determine the connectivity that GG forces on HH. If d=0d=0, our definition of m⁡(k,0)m(k,0) imposes no condition, so nothing is required. Suppose that d≥1d\geq 1, and let X⊆TX\subseteq T be arbitrary with |X|<d|X|<d. The sets SS and XX are disjoint, and hence

|S∪X|=α+|X|<α+d=α+(k−α)=k.|S\cup X|=\alpha+|X|<\alpha+d=\alpha+(k-\alpha)=k.

Deleting S∪XS\cup X from GG therefore leaves a connected graph. But the vertices that remain are precisely T∖XT\setminus X, and all remaining edges are the edges of H−XH-X; in other words,

G−(S∪X)=H−X.G-(S\cup X)=H-X.

Thus H−XH-X is connected for every XX with |X|<d|X|<d. Moreover, |V⁡(H)|=k>d|V(H)|=k>d, so the order requirement in the definition of dd-connectivity is satisfied. It follows that HH is dd-connected. By the definition of m⁡(k,d)m(k,d) and (3.3), every admissible GG satisfies

e⁡(G)=α​k+e⁡(H)≥α​k+m⁡(k,d).\mathrm{e}(G)=\alpha k+\mathrm{e}(H)\geq\alpha k+m(k,d). (3.4)

We now show that equality is attainable. Choose a graph HdH_{d} of order kk with

e⁡(Hd)=m⁡(k,d)\mathrm{e}(H_{d})=m(k,d)

that is dd-connected when d≥1d\geq 1. Such a graph exists by Proposition 2.1; more explicitly, we may take H0=K¯kH_{0}=\overline{K}_{k} when d=0d=0, H1=PkH_{1}=P_{k} when d=1d=1, and any graph whose existence is guaranteed by Proposition 2.1 when d≥2d\geq 2. On a disjoint set AA of α\alpha vertices, form

G0=K¯α∨Hd.G_{0}=\overline{K}_{\alpha}\mathbin{\vee}H_{d}. (3.5)

Thus AA is independent, every vertex of AA is adjacent to every vertex of HdH_{d}, and the only other edges are those of HdH_{d}. It follows immediately that

e⁡(G0)=α​k+e⁡(Hd)=α​k+m⁡(k,d).\mathrm{e}(G_{0})=\alpha k+\mathrm{e}(H_{d})=\alpha k+m(k,d). (3.6)

We must verify that G0G_{0} has exactly the prescribed independence number and is kk-connected.

No independent set of a join can contain a vertex from each side, because every vertex on one side is adjacent to every vertex on the other. Hence

α⁡(G0)=max⁡{α,α⁡(Hd)}.\alpha(G_{0})=\max\{\alpha,\alpha(H_{d})\}. (3.7)

We check α⁡(Hd)≤α\alpha(H_{d})\leq\alpha in each of the three possible cases.

If d=0d=0, then k=αk=\alpha and Hd=K¯kH_{d}=\overline{K}_{k}, so α⁡(Hd)=k=α\alpha(H_{d})=k=\alpha. In this case G0=Kk,kG_{0}=K_{k,k}.

If d=1d=1, then k−α=1k-\alpha=1, so α=k−1\alpha=k-1. We chose Hd=PkH_{d}=P_{k}. This path has an edge and therefore its entire kk-vertex set is not independent; consequently α⁡(Hd)≤k−1=α\alpha(H_{d})\leq k-1=\alpha.

Finally, suppose d≥2d\geq 2. Since HdH_{d} is dd-connected, δ⁡(Hd)≥d\delta(H_{d})\geq d. If HdH_{d} contained an independent set II with |I|≥α+1|I|\geq\alpha+1, choose any u∈Iu\in I. The vertex uu has no neighbor in II, so every neighbor of uu belongs to V⁡(Hd)∖IV(H_{d})\setminus I. There are at most

k−(α+1)=k−α−1=d−1k-(\alpha+1)=k-\alpha-1=d-1

vertices in that set, giving degHd⁡(u)≤d−1\deg_{H_{d}}(u)\leq d-1, a contradiction. Thus α⁡(Hd)≤α\alpha(H_{d})\leq\alpha. Equation (3.7) now gives

α⁡(G0)=α\alpha(G_{0})=\alpha

in all cases.

It remains to prove kk-connectivity. The graph G0G_{0} has α+k>k\alpha+k>k vertices. Let Y⊆V⁡(G0)Y\subseteq V(G_{0}) be any set with |Y|<k|Y|<k. Because HdH_{d} has exactly kk vertices, at least one vertex of HdH_{d} survives the deletion of YY.

Suppose first that at least one vertex of the independent side AA also survives. Then both sides of the join have a surviving vertex. Every surviving vertex of AA is adjacent to every surviving vertex of HdH_{d}. It follows that all surviving vertices lie in a single component: two vertices on different sides are adjacent, two vertices of AA have a common surviving neighbor in HdH_{d}, and two vertices of HdH_{d} have a common surviving neighbor in AA. Hence G0−YG_{0}-Y is connected, independently of the internal structure of Hd−YH_{d}-Y.

Suppose instead that no vertex of AA survives. Then A⊆YA\subseteq Y. If d=0d=0, then |A|=α=k|A|=\alpha=k, contradicting |Y|<k|Y|<k, so this case can occur only when d≥1d\geq 1. Put X=Y∩V⁡(Hd)X=Y\cap V(H_{d}). Because AA and V⁡(Hd)V(H_{d}) are disjoint and A⊆YA\subseteq Y,

|X|=|Y|−|A|<k−α=d.|X|=|Y|-|A|<k-\alpha=d.

All remaining vertices lie in HdH_{d}, and

G0−Y=Hd−X.G_{0}-Y=H_{d}-X.

The latter graph is connected because HdH_{d} is dd-connected. We have therefore shown that deletion of every set of fewer than kk vertices leaves G0G_{0} connected. Thus G0G_{0} is kk-connected.

The graph G0G_{0} is admissible, and (3.6) matches the universal lower bound (3.4). This proves (3.1).

It remains only to rewrite the answer. If α=k\alpha=k, then d=0d=0, so

α​k+m⁡(k,0)=k2.\alpha k+m(k,0)=k^{2}.

If α=k−1\alpha=k-1, then d=1d=1, and Proposition 2.1 gives

α​k+m⁡(k,1)=k⁡(k−1)+(k−1)=k2−1.\alpha k+m(k,1)=k(k-1)+(k-1)=k^{2}-1.

If 2≤α≤k−22\leq\alpha\leq k-2, then d=k−α≥2d=k-\alpha\geq 2, and

α​k+m⁡(k,d)\displaystyle\alpha k+m(k,d) =α​k+⌈k⁡(k−α)2⌉\displaystyle=\alpha k+\left\lceil\frac{k(k-\alpha)}{2}\right\rceil
=⌈α​k+k⁡(k−α)2⌉\displaystyle=\left\lceil\alpha k+\frac{k(k-\alpha)}{2}\right\rceil
=⌈k⁡(α+k)2⌉.\displaystyle=\left\lceil\frac{k(\alpha+k)}{2}\right\rceil.

In the second equality we used the elementary identity z+⌈x⌉=⌈z+x⌉z+\lceil x\rceil=\lceil z+x\rceil for an integer zz. These three cases give (3.2). ∎

Corollary 3.2.

Let α≥2\alpha\geq 2, k≥3k\geq 3, k≥αk\geq\alpha, and put d=k−αd=k-\alpha. A kk-connected graph GG of order α+k\alpha+k and independence number α\alpha is extremal if and only if

G≅K¯α∨H,G\cong\overline{K}_{\alpha}\mathbin{\vee}H,

where HH has order kk and satisfies one of the following conditions:

  1. (i)

    if d=0d=0, then H≅K¯kH\cong\overline{K}_{k};

  2. (ii)

    if d=1d=1, then HH is a tree;

  3. (iii)

    if d≥2d\geq 2, then HH is dd-connected and e⁡(H)=⌈k​d/2⌉\mathrm{e}(H)=\lceil kd/2\rceil.

Proof.

We first prove necessity. Let GG be an extremal graph for f⁡(α+k,α,k)f(\alpha+k,\alpha,k). Choose a maximum independent set S⊆V⁡(G)S\subseteq V(G), so |S|=α|S|=\alpha, and put

T=V⁡(G)∖S,H=G⁡[T].T=V(G)\setminus S,\qquad H=G[T].

Then |T|=k|T|=k. As shown in the lower-bound part of Theorem 3.1, δ⁡(G)≥k\delta(G)\geq k, while a vertex of SS can have neighbors only among the kk vertices of TT. Hence every vertex of SS is adjacent to every vertex of TT. There are no edges inside SS, and therefore

G≅K¯α∨H,e⁡(G)=α​k+e⁡(H).G\cong\overline{K}_{\alpha}\mathbin{\vee}H,\qquad\mathrm{e}(G)=\alpha k+\mathrm{e}(H). (3.8)

The same proof also shows that HH is dd-connected when d≥1d\geq 1. Indeed, if X⊆TX\subseteq T and |X|<d|X|<d, then

|S∪X|=α+|X|<α+d=k,|S\cup X|=\alpha+|X|<\alpha+d=k,

so kk-connectivity of GG implies that G−(S∪X)=H−XG-(S\cup X)=H-X is connected. When d=0d=0, no condition on HH is imposed.

Since GG is extremal, Theorem 3.1 and (3.8) give

α​k+e⁡(H)=e⁡(G)=f⁡(α+k,α,k)=α​k+m⁡(k,d).\alpha k+\mathrm{e}(H)=\mathrm{e}(G)=f(\alpha+k,\alpha,k)=\alpha k+m(k,d).

Cancelling α​k\alpha k yields

e⁡(H)=m⁡(k,d).\mathrm{e}(H)=m(k,d). (3.9)

We now read this equality in the three possible cases.

If d=0d=0, Proposition 2.1 gives m⁡(k,0)=0m(k,0)=0. Thus HH has no edges and, since it has order kk, H≅K¯kH\cong\overline{K}_{k}. If d=1d=1, then HH is connected and (3.9) gives e⁡(H)=m⁡(k,1)=k−1\mathrm{e}(H)=m(k,1)=k-1. A connected graph of order kk contains a spanning tree with k−1k-1 edges; as HH has no additional edges, it is itself a tree. Finally, if d≥2d\geq 2, then HH is dd-connected and Proposition 2.1 turns (3.9) into

e⁡(H)=⌈k​d2⌉.\mathrm{e}(H)=\left\lceil\frac{kd}{2}\right\rceil.

This proves that every extremal graph has one of the stated forms.

We next prove sufficiency. Let HH be a graph of order kk satisfying the condition corresponding to dd, let AA be an independent set of order α\alpha disjoint from HH, and set

G=K¯α∨H.G=\overline{K}_{\alpha}\mathbin{\vee}H.

Since the number of edges between K¯α\overline{K}_{\alpha} and HH is α​k\alpha k, the total number of edges e⁡(G)e(G) is clearly α​k+m⁡(k,d)\alpha k+m(k,d). It remains to show that α⁡(G)=α\alpha(G)=\alpha and κ⁡(G)=k\kappa(G)=k. However, the proof follows exactly the same way as Theorem 3.1 and hence we skip it here. Therefore, every graph of one of the stated forms is extremal, completing the proof of both directions. ∎

Corollary 3.3.

For every integer k≥4k\geq 4,

f⁡(2​k−1,k−1,k)=k2−1.f(2k-1,k-1,k)=k^{2}-1. (3.10)

The extremal graphs are precisely K¯k−1∨T\overline{K}_{k-1}\mathbin{\vee}T, where TT is a tree of order kk. In particular, (1.2) is false.

Proof.

Fix an integer k≥4k\geq 4 and put

α=k−1,n=2​k−1.\alpha=k-1,\qquad n=2k-1.

We first verify that all hypotheses of the extremal problem are satisfied. Certainly α≥3≥2\alpha\geq 3\geq 2 and k≥3k\geq 3. Moreover,

n=2​k−1=(k−1)+k=α+kn=2k-1=(k-1)+k=\alpha+k

and

n=2​k−1≥2​k−2=2​α.n=2k-1\geq 2k-2=2\alpha.

Thus the triple (n,α,k)(n,\alpha,k) satisfies (1.1). Since n=α+kn=\alpha+k and k≥αk\geq\alpha, Theorem 3.1 applies. Here

d=k−α=k−(k−1)=1,d=k-\alpha=k-(k-1)=1,

so either (3.1) and Proposition 2.1, or directly the middle line of (3.2), gives

f⁡(2​k−1,k−1,k)\displaystyle f(2k-1,k-1,k) =(k−1)​k+m⁡(k,1)\displaystyle=(k-1)k+m(k,1)
=k⁡(k−1)+(k−1)\displaystyle=k(k-1)+(k-1)
=k2−1.\displaystyle=k^{2}-1.

This proves (3.10).

We next determine all equality cases. In the present specialization d=1d=1. Corollary 3.2 says that an extremal graph is precisely a join

K¯k−1∨H,\overline{K}_{k-1}\mathbin{\vee}H,

where HH is a tree of order kk. Conversely, every such tree gives an extremal graph by the same corollary. Thus the displayed family is neither merely a collection of examples nor dependent on the choice of a particular tree: it is the complete extremal family.

It remains to compare the exact value with the proposed formula. The parameters lie in the first regime of (1.2), because

k​α−n\displaystyle k\alpha-n =k⁡(k−1)−(2​k−1)\displaystyle=k(k-1)-(2k-1)
=k2−3​k+1\displaystyle=k^{2}-3k+1
=(k−1)​(k−2)−1>0\displaystyle=(k-1)(k-2)-1>0

for every k≥4k\geq 4. Hence (1.2) predicts

⌈n​k2⌉\displaystyle\left\lceil\frac{nk}{2}\right\rceil =⌈k⁡(2​k−1)2⌉\displaystyle=\left\lceil\frac{k(2k-1)}{2}\right\rceil
=⌈k2−k2⌉\displaystyle=\left\lceil k^{2}-\frac{k}{2}\right\rceil
=k2−⌊k2⌋.\displaystyle=k^{2}-\left\lfloor\frac{k}{2}\right\rfloor.

The last equality uses ⌈N−x⌉=N−⌊x⌋\lceil N-x\rceil=N-\lfloor x\rfloor when NN is an integer. Since k≥4k\geq 4 implies ⌊k/2⌋≥2\lfloor k/2\rfloor\geq 2, the conjectured value is at most k2−2k^{2}-2, whereas the exact value is k2−1k^{2}-1. More precisely, the difference between the exact and conjectured values is

(k2−1)−(k2−⌊k2⌋)=⌊k2⌋−1>0.(k^{2}-1)-\left(k^{2}-\left\lfloor\frac{k}{2}\right\rfloor\right)=\left\lfloor\frac{k}{2}\right\rfloor-1>0.

Therefore (1.2) fails for every k≥4k\geq 4. ∎

4 Consequences and remaining directions

Proposition 4.1.

Among triples satisfying (1.1), the minimum possible order of a counterexample to (1.2) is seven. More precisely, f⁡(7,3,4)=15f(7,3,4)=15, while (1.2) predicts 1414.

Proof.

We first establish the failure at order seven. For (n,α,k)=(7,3,4)(n,\alpha,k)=(7,3,4) we have n=α+kn=\alpha+k and d=k−α=1d=k-\alpha=1. Theorem 3.1 therefore gives

f⁡(7,3,4)=3⋅4+m⁡(4,1)=12+3=15.f(7,3,4)=3\cdot 4+m(4,1)=12+3=15.

This value is attained, for example, by

G0=K¯3∨P4.G_{0}=\overline{K}_{3}\mathbin{\vee}P_{4}.

Indeed, G0G_{0} has the twelve edges between the two join sides and the three edges of P4P_{4}, for a total of fifteen. An independent set in a join lies in one side, so

α⁡(G0)=max⁡{3,α⁡(P4)}=3.\alpha(G_{0})=\max\{3,\alpha(P_{4})\}=3.

To check connectivity directly, delete at most three vertices. At least one of the four path vertices survives. If a vertex of the independent three-vertex side also survives, the surviving portions of the two sides are joined completely and hence form a connected graph. If the entire independent side is deleted, no path vertex was deleted, and the remaining graph is P4P_{4}, which is connected. Thus G0G_{0} is 44-connected. On the other hand, 7≤4⋅37\leq 4\cdot 3, so the first line of (1.2) predicts only

⌈7⋅42⌉=14.\left\lceil\frac{7\cdot 4}{2}\right\rceil=14.

Consequently a counterexample of order seven exists.

There is also a useful direct explanation of why the predicted value fourteen cannot occur. Suppose, to the contrary, that a 44-connected graph GG of order seven with α⁡(G)=3\alpha(G)=3 has fourteen edges. Since δ⁡(G)≥4\delta(G)\geq 4,

28=2​e​(G)=∑v∈V⁡(G)degG⁡(v)≥7⋅4=28.28=2\mathrm{e}(G)=\sum_{v\in V(G)}\deg_{G}(v)\geq 7\cdot 4=28.

Equality at the two ends forces every vertex to have degree four; hence GG is 44-regular. Each vertex consequently has degree 6−4=26-4=2 in the complement G¯\overline{G}, so G¯\overline{G} is 22-regular.

The equality α⁡(G)=3\alpha(G)=3 means that some three vertices are pairwise nonadjacent in GG, and these vertices induce a triangle in G¯\overline{G}. Each vertex of this triangle already has its two required neighbors within the triangle. Since G¯\overline{G} is 22-regular, no edge of G¯\overline{G} can join the triangle to any of the other four vertices. Thus the triangle is an entire component of G¯\overline{G}. The remaining four vertices also induce a simple 22-regular graph, and the only such graph on four vertices is a 44-cycle. Therefore

G¯=C3∪C4.\overline{G}=C_{3}\cup C_{4}.

Delete from GG the three vertices belonging to the C3C_{3} component of G¯\overline{G}. The graph left in GG is

C4¯=2​K2,\overline{C_{4}}=2K_{2},

which is disconnected. This contradicts 44-connectivity because only three vertices were deleted.

We now prove that no smaller admissible order can yield a counterexample. The inequalities in (1.1) include

α≥2,k≥3,n≥2​α,n≥α+k.\alpha\geq 2,\qquad k\geq 3,\qquad n\geq 2\alpha,\qquad n\geq\alpha+k.

In particular, n≥2+3=5n\geq 2+3=5, so there is no admissible triple with n≤4n\leq 4. If n=5n=5, the inequality 2​α≤52\alpha\leq 5 and the condition α≥2\alpha\geq 2 force α=2\alpha=2; then 3≤k≤n−α=33\leq k\leq n-\alpha=3, so k=3k=3. This gives only

(n,α,k)=(5,2,3).(n,\alpha,k)=(5,2,3).

If n=6n=6, then 2≤α≤32\leq\alpha\leq 3. For α=2\alpha=2, the inequalities 3≤k≤6−23\leq k\leq 6-2 give k∈{3,4}k\in\{3,4\}; for α=3\alpha=3, they give 3≤k≤6−33\leq k\leq 6-3, and hence k=3k=3. The complete list of admissible triples with n≤6n\leq 6 is therefore

(5,2,3),(6,2,3),(6,2,4),(6,3,3).(5,2,3),\qquad(6,2,3),\qquad(6,2,4),\qquad(6,3,3). (4.1)

Every triple in (4.1) satisfies n≤k​αn\leq k\alpha; this also follows at once from n≤6≤k​αn\leq 6\leq k\alpha. Hence the conjectured value in every case is ⌈n​k/2⌉\lceil nk/2\rceil. Any kk-connected graph has minimum degree at least kk, and the handshaking identity yields

e⁡(G)≥⌈n​k2⌉.\mathrm{e}(G)\geq\left\lceil\frac{nk}{2}\right\rceil. (4.2)

It remains to exhibit, for each triple, a graph attaining this lower bound with the required independence number and connectivity.

For (5,2,3)(5,2,3), let G1=K5−E⁡(M2),G_{1}=K_{5}-E(M_{2}), where M2M_{2} is a matching of two edges (for reference see Figure 1(a)). It has

e⁡(G1)=(52)−2=8=⌈5⋅32⌉.\mathrm{e}(G_{1})=\binom{5}{2}-2=8=\left\lceil\frac{5\cdot 3}{2}\right\rceil.
v1v_{1}v2v_{2}v3v_{3}v4v_{4}v5v_{5}
(a) K5−E⁡(M2)K_{5}-E(M_{2})
v1v_{1}v2v_{2}v3v_{3}v4v_{4}v5v_{5}v6v_{6}
(b) K6−E⁡(M3)K_{6}-E(M_{3})
Figure 1: Graphs G1G_{1} and G2G_{2}.

The endpoints of either deleted edge form an independent pair. No three vertices are independent, because the non-edges of G1G_{1} form a matching and therefore cannot contain all three pairs among a three-element set. Thus α⁡(G1)=2\alpha(G_{1})=2. After deleting fewer than three vertices, at least three vertices remain. The remaining graph is a complete graph with some pairwise disjoint edges removed. If two surviving vertices are nonadjacent, they are the endpoints of one removed matching edge, and every third surviving vertex is adjacent to both. Thus all surviving vertices lie in one component. Hence G1G_{1} is 33-connected.

For (6,2,4)(6,2,4), let G2=K6−E⁡(M3),G_{2}=K_{6}-E(M_{3}), where M3M_{3} is a perfect matching (for reference see Figure 1(b)). Then

e⁡(G2)=(62)−3=12=6⋅42.\mathrm{e}(G_{2})=\binom{6}{2}-3=12=\frac{6\cdot 4}{2}.

Exactly the same matching argument gives α⁡(G2)=2\alpha(G_{2})=2. Deleting fewer than four vertices leaves at least three vertices, and the same common-neighbor argument shows that the remaining graph is connected. Therefore G2G_{2} is 44-connected.

For (6,2,3)(6,2,3), we use the triangular prism G3G_{3} (for reference see Figure 2). Denote its two vertex-disjoint triangles by

A={a1,a2,a3},B={b1,b2,b3},A=\{a_{1},a_{2},a_{3}\},\qquad B=\{b_{1},b_{2},b_{3}\},

and let its remaining three edges be ai​bia_{i}b_{i} for 1≤i≤31\leq i\leq 3. The graph has three edges in each triangle and three matching edges, so it has nine edges. An independent set contains at most one vertex of each triangle, while a1a_{1} and b2b_{2} are nonadjacent; hence its independence number is exactly two. Delete at most two vertices. Each triangle retains at least one vertex, and two deleted vertices cannot meet all three pairwise vertex-disjoint matching edges. Thus some matching edge ai​bia_{i}b_{i} survives. Each nonempty remnant of a triangle is connected, and the surviving matching edge joins the two remnants. The resulting graph is connected, so the triangular prism is 33-connected.

a1a_{1}a2a_{2}a3a_{3}b1b_{1}b2b_{2}b3b_{3}
Figure 2: G3G_{3}

Finally, for (6,3,3)(6,3,3), consider the graph G4=K3,3G_{4}=K_{3,3}. It has nine edges. Each bipartition class is an independent set of order three, while an independent set cannot meet both classes because every possible cross-edge is present. Thus α⁡(G4)=3\alpha(G_{4})=3. Deleting fewer than three vertices cannot remove all three vertices of either bipartition class. At least one vertex therefore survives in each class, and the remaining graph is a complete bipartite graph with both parts nonempty, which is connected. Hence G4G_{4} is 33-connected.

All four graphs attain the corresponding lower bound (4.2). Thus (1.2) holds for every admissible triple of order at most six. Since it fails for (7,3,4)(7,3,4), seven is the minimum possible order of a counterexample. ∎

The following standard clique-partition estimate is the counting implication used by Bougard and Joret in their verification of the large-order range [1, Section 6, p. 12]. We record it for reference; it does not imply that every extremal graph has the required partition.

Proposition 4.2.

Let α≥2\alpha\geq 2, and let GG be a kk-connected graph of order nn. If V⁡(G)V(G) has a partition C1∪⋯∪CαC_{1}\cup\cdots\cup C_{\alpha} into α\alpha nonempty cliques, then

e⁡(G)≥t⁡(n,α)+⌈k​α2⌉.\mathrm{e}(G)\geq t(n,\alpha)+\left\lceil\frac{k\alpha}{2}\right\rceil. (4.3)

Bougard and Joret proved a clique-partition conclusion under an explicit large-order hypothesis and thereby verified the second line of (1.2) in that range [1]. The general second regime remains a separate question: a proof must either establish the partition or obtain (4.3) without it. The counterfamily above shows that the first regime requires revision before a unified statement can be true. The boundary n=α+kn=\alpha+k is completely determined by (3.2), and Corollary 3.2 determines every equality case there. Any corrected conjecture must therefore include the exceptional value k2−1k^{2}-1 and the extremal family K¯k−1∨T\overline{K}_{k-1}\mathbin{\vee}T when α=k−1\alpha=k-1.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process

During the preparation of this work, the authors used AI to discuss proof strategies, organize and check bibliographic information, check algebraic calculations and proof exposition, and improve language and readability. After using AI, the authors reviewed and edited the content as needed, independently verified the cited sources, calculations, statements, and proofs, and take full responsibility for the content of the article.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.

Data availability

Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.

References

  • [1] N. Bougard and G. Joret, Turán’s theorem and kk-connected graphs, J. Graph Theory 58 (2008), no. 1, 1–13, doi:10.1002/jgt.20289.
  • [2] J. Christophe, S. Dewez, J.-P. Doignon, G. Fasbender, P. Grégoire, D. Huygens, M. Labbé, S. Elloumi, H. Mélot, and H. Yaman, Linear inequalities among graph invariants: using GraPHedron to uncover optimal relationships, Networks 52 (2008), no. 4, 287–298, doi:10.1002/net.20250.
  • [3] I. Gitler and C. E. Valencia, On bounds for some graph invariants, Bol. Soc. Mat. Mexicana (3) 16 (2010), no. 2, 73–94.
  • [4] F. Harary, The maximum connectivity of a graph, Proc. Natl. Acad. Sci. USA 48 (1962), 1142–1146, doi:10.1073/pnas.48.7.1142.
  • [5] L. Liu and B. Ning, A spectral analogue of Ore’s problem on Turán theorem, Linear Algebra Appl. 735 (2026), 112–122, doi:10.1016/j.laa.2026.01.012.
  • [6] O. Ore, Theory of Graphs, American Mathematical Society Colloquium Publications, Vol. 38, American Mathematical Society, Providence, RI, 1962.
  • [7] P. Turán, Eine Extremalaufgabe aus der Graphentheorie, Mat. Fiz. Lapok 48 (1941), 436–452.
  • [8] T. Wang and B. Wu, The size of graphs with given feedback vertex number, Discrete Appl. Math. 314 (2022), 213–222, doi:10.1016/j.dam.2022.02.013.
  • [9] L.-T. Yuan, A note on Turán’s theorem, Appl. Math. Comput. 361 (2019), 13–14, doi:10.1016/j.amc.2019.05.020.