Minimum distance signless Laplacian spread among bipartite graphs

Let GG be a bipartite graph with nn vertices. The minimum distance signless Laplacian spread conjecture.

SQ(G)SQ(Kn2,n2).S_{\mathcal{Q}}(G)\geq S_{\mathcal{Q}}\left(K_{\left\lfloor\frac{n}{2}\right\rfloor,\left\lceil\frac{n}{2}\right\rceil}\right).

Equality holds if and only if

GKn2,n2.G\cong K_{\left\lfloor\frac{n}{2}\right\rfloor,\left\lceil\frac{n}{2}\right\rceil}.

This proposes that the balanced complete bipartite graph minimizes the distance signless Laplacian spread among all bipartite graphs on nn vertices. The claim is motivated by computations and comparisons of small examples in the paper; its general validity is left as a problem for further research.

References

Primary source

Lihua You, Liyong Ren and Guanglong Yu, “Distance and distance signless Laplacian spread of connected graphs”, arXiv:1607.00473 (2016).

Progress summary

Refreshed
Claimed progress

The conjecture remains unproved publicly, but an unverified submission dated August 26, 2026 argues that the proposed graph is indeed the unique minimizer.

The 2016 paper formulates the claim that the balanced complete bipartite graph minimizes the distance signless Laplacian spread among connected bipartite graphs of order nn, with equality only for that graph.

Known results

  • The claim was checked computationally for all connected bipartite graphs on 44 and 55 vertices, with further computations supporting it.
  • Among complete bipartite graphs Ka,naK_{a,n-a}, the balanced graph minimizes the spread, with equality only in the balanced case.
  • Extending this comparison to all connected bipartite graphs was left open in 2016.

Community submission (unverified), August 26, 2026

A submitted proof argues the full inequality using distance-sum lower bounds for the two parts, Rayleigh-quotient estimates for the distance signless Laplacian, and separate treatment of complete and non-complete bipartite graphs. It claims the equality case is uniquely the balanced complete bipartite graph, but no independent verification is recorded.

Current status (as of August 2026): the published literature still treats the conjecture as open, while a community submission claims a proof dated August 26, 2026 that remains unverified.

Sources

Solutions 1

Balanced complete bipartite graphs uniquely minimize the spread

Statement and scope

As in the source's global convention, all graphs below are finite, simple, connected, and undirected. Connectedness is necessary for the distance matrix to be defined. Let

Q(G)=Tr(G)+D(G)\mathcal Q(G)=\operatorname{Tr}(G)+\mathcal D(G)

be the distance signless Laplacian, and write q1(G)qn(G)q_1(G)\geq\cdots\geq q_n(G) for its eigenvalues. Its spread is SQ(G)=q1(G)qn(G)S_{\mathcal Q}(G)=q_1(G)-q_n(G).

We prove that every connected bipartite graph GG of order nn satisfies

SQ(G)SQ ⁣(Kn/2,n/2),S_{\mathcal Q}(G)\geq S_{\mathcal Q}\!\left( K_{\lfloor n/2\rfloor,\lceil n/2\rceil} \right),

with equality exactly for the balanced complete bipartite graph.

The cases n3n\leq3 are immediate: up to isomorphism there is only one connected bipartite graph at each such order. Hence assume n4n\geq4.

Three distance sums

Let ABA\cup B be the bipartition, with

a=Ab=B,n=a+b,d=ab.a=|A|\geq b=|B|,\qquad n=a+b,\qquad d=a-b.

Set

X={u,v}AdG(u,v),Y={u,v}BdG(u,v),Z=uA, vBdG(u,v).\begin{aligned} X&=\sum_{\{u,v\}\subseteq A}d_G(u,v),\\ Y&=\sum_{\{u,v\}\subseteq B}d_G(u,v),\\ Z&=\sum_{u\in A,\ v\in B}d_G(u,v). \end{aligned}

The Wiener index is W=X+Y+ZW=X+Y+Z. Bipartite parity gives

Xa(a1),Yb(b1),Zab.(1)X\geq a(a-1),\qquad Y\geq b(b-1),\qquad Z\geq ab. \tag{1}

If GG is not complete bipartite, some cross-part pair is a nonedge. Its odd distance is at least 33, so in that case

Zab+2.(2)Z\geq ab+2. \tag{2}

Rayleigh bounds from the two parts

The all-ones vector gives

q1(G)1TQ(G)1n=4Wn.(3)q_1(G)\geq\frac{\mathbf1^{T}\mathcal Q(G)\mathbf1}{n} =\frac{4W}{n}. \tag{3}

Suppose first that a,b2a,b\geq2. Consider the (a1)(a-1)-dimensional space of vectors supported on AA whose coordinates sum to zero. If PA=IAJA/aP_A=I_A-J_A/a is its orthogonal projection, the average Rayleigh quotient of Q(G)\mathcal Q(G) on an orthonormal basis of this space is

tr(PAQAA)a1.\frac{\operatorname{tr}(P_A\mathcal Q_{AA})}{a-1}.

Now

trQAA=2X+Z,1ATQAA1A=4X+Z.\operatorname{tr}\mathcal Q_{AA}=2X+Z, \qquad \mathbf1_A^T\mathcal Q_{AA}\mathbf1_A=4X+Z.

Consequently

qn(G)rA:=2(a2)X+(a1)Za(a1).(4)q_n(G)\leq r_A:=\frac{2(a-2)X+(a-1)Z}{a(a-1)}. \tag{4}

The same calculation on BB gives

qn(G)rB:=2(b2)Y+(b1)Zb(b1).(5)q_n(G)\leq r_B:=\frac{2(b-2)Y+(b-1)Z}{b(b-1)}. \tag{5}

For every 0θ10\leq\theta\leq1, equations (3)--(5) imply

SQ(G)4(X+Y+Z)nθrA(1θ)rB=cXX+cYY+cZZ,(6)S_{\mathcal Q}(G) \geq \frac{4(X+Y+Z)}n-\theta r_A-(1-\theta)r_B =c_XX+c_YY+c_ZZ, \tag{6}

where

cX=4nθ2(a2)a(a1),cY=4n(1θ)2(b2)b(b1),cZ=4nθa1θb.(7)\begin{aligned} c_X&=\frac4n-\theta\frac{2(a-2)}{a(a-1)},\\ c_Y&=\frac4n-(1-\theta)\frac{2(b-2)}{b(b-1)},\\ c_Z&=\frac4n-\frac{\theta}{a}-\frac{1-\theta}{b}. \end{aligned} \tag{7}

Choosing the convex weight

We choose θ\theta so that all three coefficients in (7) are nonnegative. Define

θY=a(b2)b2n(b2),θZ=a(a3b)nd.\theta_Y= \frac{a(b-2)-b^2}{n(b-2)}, \qquad \theta_Z= \frac{a(a-3b)}{nd}.

Only the rows in which their numerators are positive use these quantities. Substitution in (6) at the lower endpoints in (1) gives the following table.

Part sizesθ\thetaBase lower bound BθB_\theta
b4, a(b2)b2b\geq4,\ a(b-2)\leq b^2003n2+d2+d2n\frac{3n}{2}+\frac d2+\frac{d^2}{n}
b4, a(b2)>b2b\geq4,\ a(b-2)>b^2θY\theta_Y3n2+d(bn2d)2n(b2)\frac{3n}{2}+\frac{d(bn-2d)}{2n(b-2)}
b=3, a9b=3,\ a\leq9003n2+d2+d2n\frac{3n}{2}+\frac d2+\frac{d^2}{n}
b=3, a>9b=3,\ a>9θZ\theta_Z2n2n
b=2, a6b=2,\ a\leq6003n2+d2+d2n\frac{3n}{2}+\frac d2+\frac{d^2}{n}
b=2, a>6b=2,\ a>6θZ\theta_Z2n2n

Here are the coefficient checks. First, for every listed weight,

cX4n2(a2)a(a1)=2{a(ab)+2b}na(a1)>0.(8)c_X\geq \frac4n-\frac{2(a-2)}{a(a-1)} =\frac{2\{a(a-b)+2b\}}{na(a-1)}>0. \tag{8}

For the zero-weight rows, their defining inequalities give cY0c_Y\geq0 and imply cZ0c_Z\geq0. In the θY\theta_Y row, cY=0c_Y=0. If a3ba\leq3b, then already cZ0c_Z\geq0 at θ=0\theta=0; if a>3ba>3b, then

θYθZ=b{ab4a+b2}(ab)(a+b)(b2)0\theta_Y-\theta_Z =\frac{b\{ab-4a+b^2\}}{(a-b)(a+b)(b-2)}\geq0

because b4b\geq4. Thus cZ0c_Z\geq0 there as well. In each θZ\theta_Z row, cZ=0c_Z=0. When b=3b=3, θZ(a9)/(a+3)\theta_Z\geq(a-9)/(a+3), which is the threshold for cY0c_Y\geq0; when b=2b=2, the term involving b2b-2 vanishes. Direct numerator comparisons also give 0θY,θZ10\leq\theta_Y,\theta_Z\leq1.

It follows from (1), (6), and (7) that

SQ(G)Bθ.(9)S_{\mathcal Q}(G)\geq B_\theta. \tag{9}

Comparison with the balanced target

A block calculation for Ka,bK_{a,b} gives the eigenvalues

n+a4[a1 times],n+b4[b1 times],5n8±9n232ab2.(10)\begin{gathered} n+a-4\quad [a-1\text{ times}],\qquad n+b-4\quad [b-1\text{ times}],\\ \frac{5n-8\pm\sqrt{9n^2-32ab}}2. \end{gathered} \tag{10}

Therefore, for n4n\geq4, the balanced target is

Tn={3n2,n even,2n+1+n2+82,n odd.(11)T_n= \begin{cases} \dfrac{3n}{2},&n\ \text{even},\\[4pt] \dfrac{2n+1+\sqrt{n^2+8}}2,&n\ \text{odd}. \end{cases} \tag{11}

For odd n5n\geq5,

Tn3n2=12+4n2+8+n<12+2n.(12)T_n-\frac{3n}{2} =\frac12+\frac4{\sqrt{n^2+8}+n} <\frac12+\frac2n. \tag{12}

Now suppose the parts are unbalanced. In a zero-weight row, the surplus over 3n/23n/2 is d/2+d2/nd/2+d^2/n. This is positive for even nn; for odd nn, unbalancedness gives d3d\geq3, and (12) makes the inequality strict. In the θY\theta_Y row,

BθY3n2=d(bn2d)2n(b2)d2,B_{\theta_Y}-\frac{3n}{2} =\frac{d(bn-2d)}{2n(b-2)} \geq\frac d2,

because dnd\leq n. This again beats (11)--(12), strictly. Finally, BθZ=2n>TnB_{\theta_Z}=2n>T_n.

It remains to handle balanced parts. If nn is even, take θ=0\theta=0. The base bound is exactly TnT_n, while cZ=2/n>0c_Z=2/n>0. By (2), every noncomplete graph has spread strictly larger than TnT_n.

If n5n\geq5 is odd and d=1d=1, again take θ=0\theta=0. Then

B0=3n2+12+1n,cZ=2(n2)n(n1).B_0=\frac{3n}{2}+\frac12+\frac1n, \qquad c_Z=\frac{2(n-2)}{n(n-1)}.

The shortfall of B0B_0 from TnT_n is

4n2+8+n1n<1n,\frac4{\sqrt{n^2+8}+n}-\frac1n<\frac1n,

whereas a missing cross edge adds at least

2cZ=4(n2)n(n1)>1n2c_Z=\frac{4(n-2)}{n(n-1)}>\frac1n

to the right side of (6). Thus every noncomplete graph is again strictly above TnT_n. Formula (10) shows that the complete balanced graph attains TnT_n.

Finally, if b=1b=1, connectedness forces G=K1,n1G=K_{1,n-1}. For n4n\geq4, (10) gives

SQ(K1,n1)=9n232n+32.S_{\mathcal Q}(K_{1,n-1}) =\sqrt{9n^2-32n+32}.

For even nn, its square exceeds (3n/2)2(3n/2)^2, since the difference has numerator 27n2128n+128>027n^2-128n+128>0. For odd n5n\geq5,

9n232n+32>2n1>Tn;\sqrt{9n^2-32n+32}>2n-1>T_n;

the two squared comparisons reduce respectively to 5n228n+31>05n^2-28n+31>0 and 3n212n+1>03n^2-12n+1>0.

All unbalanced cases are therefore strict, and every balanced noncomplete case is strict. Equality occurs exactly for

GKn/2,n/2.\boxed{G\cong K_{\lfloor n/2\rfloor,\lceil n/2\rceil}}.

Lean: https://github.com/antoshashakov/Principia-Math-In-Progress/blob/main/mathdb-open-problems/problems/333521/Problem333521.lean

Solved by the Principia Math harness. Check out our work at principia-math.com

Models used: GPT 5.6 Sol, Fable