alphaFeedback

WorksMathematics

Counterexamples to a conjecture and improved bounds for bipanchromatic colorings

AI contributions
Not recorded for this paper
View details

Abstract

We study the bipanchromatic number χBχ_B, introduced by Lalou et al. (Discrete Appl. Math. 391 (2026), 511-522) as the maximum number of colors in a panchromatic hypergraph coloring where every color appears at least twice. They show that merging uniquely occurring colors in a maximum panchromatic coloring gives the lower bound β=χP−⌈α/2⌉≤χBβ=χ_P-\lceilα/2\rceil \leq χ_B, where χPχ_P is the panchromatic number and αα is the minimum number of such colors among maximum panchromatic colorings. They conjectured that χB=βχ_B=β and subsequently disproved it (arXiv:2610.08042). We identify the smallest counterexample, with six vertices and three hyperedges, both minimum possible. We show that the ratio bound χB/β<2χ_B/β<2 is asymptotically sharp and prove minimality results for the order and hyperedge count. For 3≤α<χP3\leα<χ_P we show that every integer from ββ to χP−1χ_P-1 occurs as χBχ_B while χPχ_P and αα are held fixed. Furthermore, we discover that the values ∣V∣,χP,α|V|,χ_P,α directly decide whether the conjectured equality holds except when α=4α=4, where a matching reduction resolves the ambiguity. We derive the optimal universal lower bound LL depending only on ∣V∣,χP,α|V|,χ_P,α in the nontrivial regime 0<2α<∣V∣0<2α<|V|. Finally, we prove χB/L≤5/4χ_B/L\le5/4 and χB/L≤6/5+1/(5L)χ_B/L\le6/5+1/(5L), with 5/45/4 sharp and 6/56/5 asymptotically sharp as L→∞L\to\infty.

Introduction

A panchromatic coloring [1, 2, 3, 4] uses every color in every hyperedge, with maximum χP\chi_P colors. Lalou et al. [1] introduced bipanchromatic colorings to study completely independent spanning trees in split graphs. These colorings use each color at least twice overall, with maximum χB\chi_B. Let α\alpha be the minimum number of colors occurring once in a χP\chi_P-coloring. Their conjecture [1 Conjecture 1] predicts χB=χP−⌈α/2⌉\chi_B = \chi_P - \lceil\alpha/2\rceil by merging those colors.

The same authors subsequently disproved this conjecture [5]. They construct counterexamples as three-hyperedge sunflowers with a common set of 6d6d vertices and three disjoint petals of size qq, where d≥1d \ge1 and q≥2dq \ge2d. Their parameters are χP=6d+q\chi_P = 6d + q, α=6d\alpha= 6d, and χB=4d+q\chi_B = 4d + q. Thus the conjectured value is exceeded by exactly dd, on 6d+3q≥12d6d + 3q \ge12d vertices. Our three-hyperedge construction uses 12d−612d - 6 vertices for the same excess (Proposition 2).

Without restricting the number of hyperedges, we prove that the minimum order is 4d+24d + 2 and classify all examples attaining it (Theorem 6). Their family has actual-to-conjectured ratio at most 6/56/5. We determine the sharp worst-case ratio over all hypergraphs, proving that it is strictly below two and can approach two (Theorem 5). We develop the coloring bounds and equality criteria suggested in their concluding remarks [5 Section 5].

Sections 2–3 introduce the colorings and counterexamples. Section 4 establishes sharp bounds and characterizes minimum-size counterexamples. Section 5 determines, with minimum order, all possible χB\chi_B values with fixed χP\chi_P and α\alpha, showing that no formula depending only on χP\chi_P and α\alpha can determine χB\chi_B for every hypergraph. Section 6 shows that the hypergraph order, χP\chi_P and α\alpha determine if the conjectured equality holds, except when α=4\alpha= 4 which is resolved by maximum matching. In Section 7, the optimal lower bound LL for χB\chi_B is determined from ∣V∣|V|, χP\chi_P and α\alpha, with χB/L≤5/4\chi_B/L \le5/4 and χB/L≤6/5+1/(5L)\chi_B/L \le6/5 + 1/(5L), where 5/45/4 is sharp and 6/56/5 is asymptotically sharp as L→∞L \to\infty.

Preliminaries

A hypergraph H=(V,E)H=(V,\mathcal{E}) consists of a non-empty set V=V(H)V=V(H) (the vertices of HH) and a hyperedge set E(H)=E⊂P(V)E(H)=\mathcal{E}\subset\mathcal{P}(V). If every hyperedge of HH contains exactly rr vertices, HH is rr-uniform. A proper coloring of a hypergraph assigns colors to its vertices so that no hyperedge of size at least two is monochromatic.

Definition 1 (Panchromatic colors). A panchromatic kk-coloring of HH assigns kk colors to its vertices so that each hyperedge contains at least one vertex of each color. The maximum such kk is the panchromatic number χP(H)\chi_P(H).

Definition 2 (Minimum unique-color count). A color is unique in a panchromatic coloring of HH if it is used on exactly one vertex. For every integer k∈[χP(H)]k\in[\chi_P(H)], let αk(H)\alpha_k(H) denote the minimum number of vertices with unique colors among all panchromatic kk-colorings of HH. By convention, α0(H)=0\alpha_0(H)=0.

Definition 3 (Bipanchromatic colors). A panchromatic kk-coloring is bipanchromatic [1] if no color is unique. The bipanchromatic number χB(H)\chi_B(H) of HH is the maximum kk such that HH admits a bipanchromatic coloring.

Observe that χB(H)=max⁡{k∈N:αk=0}\chi_B(H)=\max\{k\in\mathbb{N}:\alpha_k=0\}. Since every color appears on at least two vertices, it follows that χB(H)≤⌊∣V∣/2⌋\chi_B(H)\le\lfloor|V|/2\rfloor.

When the hypergraph HH is implied, we omit the argument HH and write χP\chi_P, χB\chi_B, and αk\alpha_k. Write α=αχP\alpha=\alpha_{\chi_P}. Lalou et al. [1 Conjecture 1] conjectured that

χB=χP−⌈α2⌉.(1)\chi_B=\chi_P-\left\lceil\frac{\alpha}{2}\right\rceil. \tag*{(1)}

This equality was disproved in [5 Proposition 1]. For ∣V∣≥2|V|\ge2, choose a panchromatic χP\chi_P-coloring with α\alpha unique colors. Merge the unique colors in pairs. If one unique color remains, merge it with any other color. This results in a bipanchromatic coloring with χP−⌈α/2⌉\chi_P-\lceil\alpha/2\rceil colors. Thus, χP−⌈α/2⌉\chi_P-\lceil\alpha/2\rceil is a lower bound for χB\chi_B. The equality in [1 Conjecture 1] would mean that this merging procedure is optimal. The examples in § 3 have bipanchromatic colorings with more colors than this procedure produces. We define excess ϵ=χB−(χP−⌈α/2⌉)\epsilon=\chi_B-(\chi_P-\lceil\alpha/2\rceil) which measures the difference between χB\chi_B and the lower bound in (1).

Counterexamples

The sunflower Sn=(V,E)S_n=(V,\mathcal{E}) is an (n+1)(n+1)-uniform hypergraph with 2n2n vertices and nn hyperedges. The set of vertices is V=S⊔PV=S\sqcup P, where S={s1,…,sn}S=\{s_1,\ldots,s_n\} and P={p1,…,pn}P=\{p_1,\ldots,p_n\} are disjoint, and its hyperedges are ei=S∪{pi}e_i=S\cup\{p_i\} for all i∈[n]i\in[n].

Theorem 1 (Sunflower counterexamples). For SnS_n with n≥3n\ge3,

χP=n+1,α=n,χB=n.(2)\chi_P=n+1,\qquad\alpha=n,\qquad\chi_B=n. \tag*{(2)}

Consequently, ϵ=⌈n/2⌉−1≥1\epsilon=\lceil n/2\rceil-1\ge1, so every SnS_n is a counterexample to [1 Conjecture 1]. In particular, for every n∈Nn\in\mathbb{N}, S2n+1S_{2n+1} has 4n+24n+2 vertices, 2n+12n+1 hyperedges, and ϵ(S2n+1)=n\epsilon(S_{2n+1})=n.

Proof. We can color each sis_i by ii, and all pip_i by n+1n+1. This is a panchromatic coloring (Figure 1(a)), and χP=n+1\chi_P=n+1. Any panchromatic (n+1)(n+1)-coloring of SnS_n uses each color only once per hyperedge S⊔{pi}S\sqcup\{p_i\}. The colors on SS have to be unique, thus α=∣S∣=n\alpha=|S|=n and every pip_i has the same color. Merging the unique colors gives n+1−⌈n2⌉n+1-\left\lceil\frac{n}{2}\right\rceil colors (Figure 1(b)). However, we can color each {si,pi}\{s_i,p_i\} distinctly (Figure 1(c)). Each hyperedge then contains nn colors, none of which is unique. Notice that χB≤∣V∣/2=n\chi_B\leq|V|/2=n, thus χB=n\chi_B=n and ϵ=⌊n2⌋−1\epsilon=\left\lfloor\frac{n}{2}\right\rfloor-1.

$\chi_P$-coloring of $S_n$

(a)

Merging-based coloring

(b)

Bipanchromatic $\chi_B$-coloring

(c)

Figure 1. SnS_n colorings. Hyperedges are ei=S∪{pi}e_i = S \cup\{p_i\}. (a) Unique colors on SS and one color on PP give n+1n+1 colors. (b) Merging on SS gives 1+⌊n2⌋1+\left\lfloor\frac{n}{2}\right\rfloor colors. (c) Each {si,pi}\{s_i,p_i\} has its own color, giving nn colors.

Thus, the gap between χB\chi_B and the quantity in [1 Conjecture 1] is unbounded. In particular, adding any fixed constant to the right-hand side cannot make the conjecture valid in general.

Figure 1(a) shows a χP\chi_P-coloring of SnS_n attaining α\alpha. Figure 1(b) shows a bipanchromatic coloring obtained by merging unique colors. Figure 1(c) shows a bipanchromatic coloring with more colors.

For excess nn, the three-hyperedge construction in [5 Proposition 1] uses at least 12n12n vertices. The following construction uses 12n−612n-6. It is obtained by enlarging S3S_3. Replace each vertex by a group of 2n−12n-1 vertices and each hyperedge by the union of the groups corresponding to its vertices.

Proposition 2 (Three-hyperedge counterexamples). For n∈Nn\in\mathbb{N}, let SS, P1P_1, P2P_2, P3P_3 be pairwise disjoint sets with ∣S∣=6n−3|S|=6n-3 and ∣Pi∣=2n−1|P_i|=2n-1. Let V=S⊔P1⊔P2⊔P3V=S\sqcup P_1\sqcup P_2\sqcup P_3, ei=S⊔Pie_i=S\sqcup P_i, and Δn=(V,{e1,e2,e3})\Delta_n=(V,\{e_1,e_2,e_3\}). The excess ϵ(Δn)\epsilon(\Delta_n) is equal to nn.

Proof. Color the vertices of SS with distinct colors. Using ∣Pi∣|P_i| additional distinct colors, color the vertices of each PiP_i with distinct colors. This is panchromatic, so χP=∣ei∣=8n−4\chi_P=|e_i|=8n-4. Any χP\chi_P-coloring of Δn\Delta_n uses each color only once per hyperedge S⊔PiS\sqcup P_i, so only the colors on SS are unique, thus αχP=∣S∣\alpha_{\chi_P}=|S|. Now color the vertices of SS with distinct colors and repeat each such color on exactly one vertex in P1∪P2∪P3P_1\cup P_2\cup P_3. This gives χB=∣S∣\chi_B=|S|. Thus, ϵ(Δn)=(6n−3)−(8n−4−⌈6n−32⌉)=n\epsilon(\Delta_n)=(6n-3)-\left(8n-4-\left\lceil\frac{6n-3}{2}\right\rceil\right)=n.

Bounds and minimality

Consider an arbitrary hypergraph HH with panchromatic number χP\chi_P. For any integer k≤χPk\leq\chi_P, let βk=k−⌈αk2⌉\beta_k=k-\left\lceil\frac{\alpha_k}{2}\right\rceil. In particular, let β=βχP\beta=\beta_{\chi_P}. [1 Conjecture 1] asserts that χB=β\chi_B=\beta, but this does not hold in general. However, for any kk-panchromatic coloring of HH attaining αk\alpha_k, the pairing argument used in [1] produces a bipanchromatic coloring with βk\beta_k colors. Thus, for every k≤χPk\leq\chi_P we obtain a lower bound for χB\chi_B, and the largest of these bounds equals χB\chi_B.

Corollary 3 (Merging-based lower bound). For an arbitrary hypergraph HH and for each integer k≤χPk \le\chi_{P} we obtain that χB≥βk\chi_{B} \ge\beta_{k}. Furthermore, we have that χB=max⁡k{βk}\chi_{B} = \max_{k}\{\beta_{k}\}.

Lemma 4 (Equality cases and counterexample bounds). (i) If α∈{0,1,2,χP}\alpha\in\{0,1,2,\chi_{P}\}, then χB=β\chi_{B} = \beta. In particular, equality holds if χP≤3\chi_{P} \le3. (ii) If χB≠β\chi_{B} \ne\beta, then 3≤α≤χP−13 \le\alpha\le\chi_{P} - 1 and ⌈χP/2⌉≤β<χB≤χP−1\lceil\chi_{P}/2 \rceil\le\beta< \chi_{B} \le\chi_{P} - 1.

Proof. (i) Corollary 3 gives χB≥β\chi_{B} \ge\beta. If α=0\alpha= 0, a χP\chi_{P}-coloring is already bipanchromatic, so χB=χP=β\chi_{B} = \chi_{P} = \beta. If α>0\alpha> 0, no χP\chi_{P}-coloring is bipanchromatic, so χB≤χP−1\chi_{B} \le\chi_{P} - 1. For α∈{1,2}\alpha\in\{1,2\}, χB=β\chi_{B} = \beta. If α=χP\alpha= \chi_{P}, then ∣V∣=χP|V| = \chi_{P}, so χB≤⌊∣V∣/2⌋=⌊χP/2⌋=β\chi_{B} \le\lfloor|V|/2 \rfloor= \lfloor\chi_{P}/2 \rfloor= \beta, so χB=β\chi_{B} = \beta.

(ii) By (i), 3≤α≤χP−13 \le\alpha\le\chi_{P} - 1. Thus β≥χP−⌈χP−12⌉=⌊χP2⌋\beta\ge\chi_{P} - \left\lceil\frac{\chi_{P}-1}{2} \right\rceil= \left\lfloor\frac{\chi_{P}}{2} \right\rfloor. Corollary 3 gives β<χB\beta< \chi_{B}, and α>0\alpha> 0 gives χB≤χP−1\chi_{B} \le\chi_{P} - 1 as above. □\square

Let ρ=χB/β\rho= \chi_{B}/\beta denote the actual-to-conjectured ratio.

Theorem 5 (Sharp ratio bounds). If ∣V∣≥2|V| \ge2, then 1≤ρ<21 \le\rho< 2, and the upper bound is asymptotically sharp. For χP≥2\chi_{P} \ge2, the sharper bound ρ≤χP−1⌈χP/2⌉\rho\le\frac{\chi_{P}-1}{\lceil\chi_{P}/2 \rceil} is attained for every fixed value of χP\chi_{P}.

Proof. By Lemma 4 it holds that ρ≥1\rho\ge1, with equality precisely when (1) is true. Suppose that χP≥2\chi_{P} \ge2. If ρ≠1\rho\ne1, then Lemma 4(ii) gives the upper bound. For n≥3n \ge3, Theorem 1 gives ρ(Sn)=χP−1⌈χP/2⌉\rho(S_{n}) = \frac{\chi_{P}-1}{\lceil\chi_{P}/2 \rceil}. The sharp bounds are strictly below 2 and tend to 2 as χP→∞\chi_{P} \to\infty. □\square

The order of a hypergraph HH is the number of its vertices ∣V(H)∣|V(H)|. We show S2n+1S_{2n+1} has minimum order among hypergraphs satisfying ϵ≥n\epsilon\ge n.

Theorem 6 (Excess-based minimum order). If a hypergraph H=(V,E)H = (V,\mathcal{E}) satisfies ϵ(H)≥n\epsilon(H) \ge n, then ∣V∣≥∣V(S2n+1)∣=4n+2|V| \ge|V(S_{2n+1})| = 4n+2. Moreover, if ϵ(H)=n\epsilon(H) = n and ∣V∣=4n+2|V| = 4n+2, then, after removing hyperedges that properly contain other hyperedges, H≅S2n+1H \cong S_{2n+1}.

Proof. By Lemma 4(ii), β≥⌈χP/2⌉\beta\ge\lceil\chi_{P}/2 \rceil and χB≤χP−1\chi_{B} \le\chi_{P} - 1. Hence

n≤ϵ=χB−β≤χB−⌈χP/2⌉≤⌊χB−12⌋.(3)n \le\epsilon= \chi_{B} - \beta\le\chi_{B} - \lceil\chi_{P}/2 \rceil\le\left\lfloor\frac{\chi_{B}-1}{2} \right\rfloor. \tag*{(3)}

Thus, χB≥2n+1\chi_{B} \ge2n+1, and consequently ∣V∣≥2χB≥4n+2=∣V(S2n+1)∣|V| \ge2\chi_{B} \ge4n+2 = |V(S_{2n+1})|.

Suppose that ϵ(H)=n\epsilon(H) = n and ∣V∣=4n+2|V| = 4n+2. It follows from (3) that χB=2n+1\chi_{B} = 2n+1 and β=⌈χP/2⌉=n+1\beta= \lceil\chi_{P}/2 \rceil= n+1. Since χP≠χB\chi_{P} \ne\chi_{B} (ϵ>0\epsilon> 0), we obtain χP=2n+2\chi_{P} = 2n+2. Moreover, ⌈α2⌉=χP−β=n+1\left\lceil\frac{\alpha}{2} \right\rceil= \chi_{P} - \beta= n+1. Since α≠χP\alpha\ne\chi_{P} it follows that α=2n+1\alpha= 2n+1.

Consider a panchromatic (χP,α)(\chi_{P},\alpha)-coloring of HH, and let SS be the set of its α\alpha uniquely colored vertices. All remaining ∣V∣−α|V|-\alpha vertices have a common color CC. For every vertex vv of color CC, there is a hyperedge S⊔{v}S \sqcup\{v\}. Otherwise, if every hyperedge containing vv also contained another vertex of vv’s color, recoloring vv with an existing unique color preserves panchromaticity and contradicts the supposed α\alpha. Removing hyperedges that properly contain other hyperedges leaves S2ϵ+1S_{2\epsilon+1}.

Every counterexample has at least three hyperedges. Choose a χP\chi_{P}-coloring attaining α\alpha. If every non-unique color occurred twice, then ∣V∣=2χP−α|V| = 2\chi_{P} - \alpha, so χB≤⌊∣V∣/2⌋=β\chi_{B} \le\lfloor|V|/2 \rfloor= \beta, a contradiction. Thus some color CC occurs on at least three vertices. Since α>0\alpha> 0, the same recoloring argument shows that each such vertex is the only CC-colored vertex in some hyperedge. These hyperedges are distinct.

Proposition 2 attains this three-hyperedge minimum with ϵ(Δn)=n\epsilon(\Delta_{n}) = n for every n≥1n \ge1. However, Theorem 6 implies that every hypergraph of minimum order satisfying ϵ(H)≥n\epsilon(H) \ge n has at least 2n+12n+1 hyperedges. Thus, for n≥2n \ge2, the minimum numbers of vertices and hyperedges cannot be attained simultaneously.

Nondetermination

By Corollary 3 and Lemma 4, if χP>α≥3\chi_P > \alpha\ge3, then χP−⌈α/2⌉≤χB<χP\chi_P - \lceil\alpha/2\rceil\le\chi_B < \chi_P. We will show that every integer value of χB\chi_B in this range can be realized for the given pair (χP,α)(\chi_P,\alpha). Thus, χB\chi_B cannot be expressed as a function of χP\chi_P and α\alpha alone.

Proposition 7 (Realization of prescribed coloring parameters). Let pp, aa, bb be integers satisfying

3≤a<pandp−⌈a/2⌉≤b<p.(4)3 \le a < p \quad\text{and}\quad p - \lceil a/2\rceil\le b < p. \tag*{(4)}

Then there exists a hypergraph satisfying χP=p\chi_P = p, α=a\alpha= a, χB=b\chi_B = b.

Proof. Start with two hyperedges e0e_0 and en+1e_{n+1}, each of size pp, whose intersection is a set SS of size aa, as shown in Figure 2(a). Choose vertices x0∈e0∖Sx_0 \in e_0 \setminus S and xn+1∈en+1∖Sx_{n+1} \in e_{n+1} \setminus S, and write

e0=S⊔U⊔{x0},en+1=S⊔W⊔{xn+1},e_0 = S \mathbin{\sqcup} U \mathbin{\sqcup} \{x_0\}, \qquad e_{n+1} = S \mathbin{\sqcup} W \mathbin{\sqcup} \{x_{n+1}\},

where ∣U∣=∣W∣=p−a−1|U| = |W| = p-a-1.

Since any hypergraph admitting a bipanchromatic bb-coloring has at least 2b2b vertices, we set

n=max⁡{0,2b−(2p−a)}(5)n = \max\{0, 2b-(2p-a)\} \tag*{(5)}

and add nn new vertices x1,…,xnx_1,\ldots,x_n. Let X={x0,x1,…,xn,xn+1}X = \{x_0,x_1,\ldots,x_n,x_{n+1}\}. We then add nn new hyperedges

ei=(e0∖{x0})∪{xi},for all i∈[n],e_i = (e_0 \setminus\{x_0\}) \mathbin{\cup} \{x_i\}, \quad\text{for all } i \in[n],

see in Figure 2(b). This gives a pp-uniform hypergraph H=(V,E)H=(V,\mathcal{E}), where

V=S⊔U⊔W⊔X,E={e0,…,en+1}.(6)V = S \mathbin{\sqcup} U \mathbin{\sqcup} W \mathbin{\sqcup} X, \qquad\mathcal{E} = \{e_0,\ldots,e_{n+1}\}. \tag*{(6)}

By (4) and the definition of nn, ∣V∣∈{2b,2b+1}|V| \in\{2b,2b+1\}. We define a pp-panchromatic coloring of HH as follows:

  1. The vertices of SS receive pairwise distinct colors.

  1. The vertices of UU and WW are paired, with each pair receiving its own color. This introduces p−a−1p-a-1 additional colors.

  1. Introduce one new color. Assign that same color to every vertex of XX.

See Figure 2(b) for details. Every hyperedge contains all pp colors, so χP=p\chi_P=p. Moreover, exactly the colors assigned to the vertices of SS are unique in any (χP,α)(\chi_P,\alpha)-coloring, and hence α=a\alpha=a.

To obtain a bipanchromatic bb-coloring of HH (Figure 2(c)), we retain the colors assigned to the vertices of U∪WU \cup W. Since ∣X∣≤∣S∣|X| \le|S|, we can pair the vertices of XX with a subset of SS. We then merge the colors of the remaining vertices of SS in pairs, as indicated in the figure. We recolor any leftover vertex with an existing pair’s color. The resulting coloring is bipanchromatic and uses bb colors. Thus, χB=b\chi_B=b.

Proposition 7 construction and colorings

Figure 2. Proposition 7 at (p,a,b)=(8,6,6)(p,a,b)=(8,6,6) and n=2n=2. S={s1,…,s6}S=\{s_1,\ldots,s_6\}, U={u}U=\{u\} and W={w}W=\{w\}. The contours represent the hyperedges ei=S⊔{u,xi}e_i=S\sqcup\{u,x_i\}, e3=S⊔{w,x3}e_3=S\sqcup\{w,x_3\}. (a) The initial construction. (b) Adding new vertices and hyperedges, and 8-coloring. (c) Bipanchromatic coloring by pairing.

The construction realizes the prescribed parameters using at most α\alpha hyperedges, each of size χP\chi_P. Together with Lemma 4(i), this shows that χP\chi_P and α\alpha uniquely determine χB\chi_B only when α∈{0,1,2,χP}\alpha\in\{0,1,2,\chi_P\}.

The minimum number of vertices in a hypergraph HH with prescribed feasible parameters (χP,α,χB)(\chi_P,\alpha,\chi_B) is max⁡{2χP−α,2χB}\max\{2\chi_P-\alpha,2\chi_B\}. Indeed, a χP\chi_P-coloring attaining α\alpha gives ∣V∣≥α+2(χP−α)=2χP−α|V| \ge\alpha+ 2(\chi_P-\alpha) = 2\chi_P-\alpha. A bipanchromatic χB\chi_B-coloring also gives ∣V∣≥2χB|V| \ge2\chi_B.

For 3≤α<χP3\leq\alpha<\chi_P, the construction in Proposition 7 has exactly max⁡{2χP−α,2χB}\max\{2\chi_P-\alpha,2\chi_B\} vertices.

In the remaining cases, Lemma 4(i) gives χB=β\chi_B=\beta. Take two sets of size χP\chi_P intersecting in α\alpha vertices, use them as hyperedges, and take their union as the vertex set. When α=χP\alpha=\chi_P, this gives just one hyperedge. The panchromatic number is χP\chi_P, and every maximum panchromatic coloring has unique colors precisely on the α\alpha common vertices. The same lemma therefore gives bipanchromatic number β\beta. This construction has 2χP−α=max⁡{2χP−α,2β}2\chi_P-\alpha=\max\{2\chi_P-\alpha,2\beta\} vertices, attaining the lower bound.

Characterization of equality

Here we discuss conditions characterizing χB=β\chi_B=\beta (equivalently ϵ=0\epsilon=0) for an arbitrary hypergraph.

Proposition 8 (Zero excess characterization). Assume that HH is a hypergraph with α∈N∖{1,2,4}\alpha\in\mathbb{N}\setminus\{1,2,4\}. Then χB=β\chi_B=\beta if and only if β=⌊∣V∣2⌋\beta=\left\lfloor\frac{|V|}{2}\right\rfloor.

Proof. Assume that χB=β\chi_B=\beta and consider a (χP,α)(\chi_P,\alpha)-coloring of HH. Let n=3n=3 if α\alpha is odd and n=4n=4 otherwise. If a color CC occurs at least nn times, recolor all vertices of color CC with nn existing unique colors, using each at least once. Merge the remaining α−n\alpha-n uniquely colored vertices in pairs. This yields a β+1\beta+1 bipanchromatic coloring of HH, contradicting χB=β\chi_B=\beta. Therefore, if α\alpha is odd, then every color occurs at most twice, and if α\alpha is even, every color occurs at most thrice. If two distinct colors each occur three times, recolor their six vertices with six colors that are used only once (since α≥6\alpha\geq6), and then merge the remaining uniquely colored vertices in pairs, again obtaining a coloring with β+1\beta+1 colors. Thus, at most one color occurs thrice, so ∣V∣∈{2χP−α,2χP−α+1}|V|\in\{2\chi_P-\alpha,2\chi_P-\alpha+1\}, and ⌊∣V∣/2⌋=β\lfloor|V|/2\rfloor=\beta.

The converse follows from the definition.

For a hypergraph HH with α=4\alpha=4, the equality χB=β\chi_B=\beta requires further consideration. Consider mutually disjoint sets ∣S∣=4|S|=4, {v,a,b,x0,x1,x2}\{v,a,b,x_0,x_1,x_2\} and ∣Pi∣=2|P_i|=2 for i=1,2,3i=1,2,3. We define two hypergraphs H−H_- and H+H_+ whose hyperedge sets are, respectively

E−={S⊔Pi:i∈[3]}andE+={S⊔{v,x0},S⊔{v,x1},S⊔{v,x2},S⊔{a,b}}.\mathcal{E}_-=\{S\sqcup P_i:i\in[3]\}\quad\text{and}\quad\mathcal{E}_+=\{S\sqcup\{v,x_0\},S\sqcup\{v,x_1\},S\sqcup\{v,x_2\},S\sqcup\{a,b\}\}.

Notice that ∣V(H−)∣=∣V(H+)∣=10|V(H_-)|=|V(H_+)|=10, χP(H−)=χP(H+)=6\chi_P(H_-)=\chi_P(H_+)=6, and α(H−)=α(H+)=4\alpha(H_-)=\alpha(H_+)=4. However, β=χB(H−)=4<∣V∣/2\beta=\chi_B(H_-)=4<|V|/2 and β<χB(H+)=5=∣V∣/2\beta<\chi_B(H_+)=5=|V|/2.

Therefore, if α(H)=4\alpha(H)=4, we need an additional analysis. For a given hypergraph H=(V,E)H=(V,\mathcal{E}), define a simple graph GH=(V,E)G_H=(V,E), where

E={uv:u,v∈V,{u,v}∩e≠∅ for all e∈E}.E=\{uv:u,v\in V,\{u,v\}\cap e\ne\varnothing\text{ for all }e\in\mathcal{E}\}.

Proposition 9 (Matching reduction). Let HH be a hypergraph with α(H)=4\alpha(H)=4 and let ν\nu be the size of a maximum matching of GHG_H. Then χB=β\chi_B=\beta if and only if ∣V∣+ν≤3χP−4|V|+\nu\le3\chi_P-4.

Proof. Since α=4\alpha=4 and β=χP−2\beta=\chi_P-2, then χB∈{χP−2,χP−1}\chi_B\in\{\chi_P-2,\chi_P-1\}. Consider any bipanchromatic kk-coloring. Each color occurring exactly twice determines an edge of GHG_H. These edges are vertex-disjoint, so they form a matching in GHG_H. Therefore, we have at most ν\nu colors used exactly two times. Consequently, ∣V∣≥2ν+3(k−ν)|V|\ge2\nu+3(k-\nu), so k≤χB≤⌊(∣V∣+ν)/3⌋k\le\chi_B\le\lfloor(|V|+\nu)/3\rfloor.

If ∣V∣+ν≤3χP−4|V|+\nu\le3\chi_P-4, then χB≤χP−2=β\chi_B\le\chi_P-2=\beta, and therefore ϵ=0\epsilon=0.

For the converse, assume ϵ=0\epsilon=0 and choose a panchromatic χP\chi_P-coloring attaining α=4\alpha=4. If some color CC occurs at least four times, we can recolor all vertices of color CC with the four colors that are used only once. This produces a bipanchromatic coloring with χP−1\chi_P-1 colors, contradicting ϵ=0\epsilon=0.

Therefore, every color is used at most three times. Let CkC_k denote the number of colors used exactly kk times. Then, χP=4+C2+C3\chi_P=4+C_2+C_3, and ∣V∣=4+2C2+3C3|V|=4+2C_2+3C_3.

The pairs corresponding to the C2C_2 color classes must form a maximum matching in GH[V∖S]G_H[V\setminus S], where SS is the set of vertices receiving the four unique colors. Otherwise, an augmenting path would yield a larger matching, whose pairs could be assigned distinct colors in a panchromatic coloring of HH (the pairs intersect every hyperedge). The endpoints uu and vv of such an augmenting path belong to three-vertex color classes C3C_3. Reassign the colors so that each matching pair forms its own color class. If uu and vv belong to the same three-vertex color class, assign the color of one of the four uniquely colored vertices to the third vertex of that class. This produces a χP\chi_P-coloring with three uniquely colored vertices, a contradiction. If uu and vv belong to different three-vertex color classes, assign the four unique colors to their other four vertices. This produces a bipanchromatic coloring with χP−1\chi_P-1 colors, again a contradiction, since χB=β=χP−2\chi_B=\beta=\chi_P-2. Thus, ν≤C2+4\nu\le C_2+4, and thus

∣V∣+ν≤(4+2C2+3C3)+(C2+4)=3χP−4.|V|+\nu\le(4+2C_2+3C_3)+(C_2+4)=3\chi_P-4.

Given HH and promises χP\chi_P and α\alpha, ϵ=0\epsilon=0 (or χB=β\chi_B=\beta) is decidable in polynomial time.

Order-aware improved bounds

In this section, we introduce another lower bound LL for the bipanchromatic number, based on our construction, and show that β≤L≤χB\beta\le L\le\chi_B. Thus, LL improves the lower bound β\beta.

Consider a hypergraph H=(V,E)H=(V,\mathcal{E}) with ∣V∣≥2|V|\ge2 and α>0\alpha>0. If α≥∣V∣/2\alpha\ge|V|/2, then χB=⌊∣V∣/2⌋\chi_B=\lfloor|V|/2\rfloor. Indeed, the unique vertices can be paired with the non-unique vertices, and any remaining vertices can be assigned colors so that every color used occurs at least twice. For the sunflower SnS_n (see Theorem 1), where n≥3n\ge3, we have α=∣V∣/2\alpha=|V|/2 and χB=∣V∣/2\chi_B=|V|/2, while χB/β→2\chi_B/\beta\to2 as n→∞n\to\infty.

We now consider the case 0<α<∣V∣/20<\alpha<|V|/2 and choose a panchromatic χP\chi_P-coloring of HH with α\alpha unique colors. Let m=χP−αm=\chi_P-\alpha denote the number of colors used more than once. Suppose that color ii is used on cic_i vertices, where

c1≥c2≥⋯≥cm>1.c_1\ge c_2\ge\cdots\ge c_m>1.

Observe that c1+c2+⋯+cm=∣V∣−αc_1+c_2+\cdots+c_m=|V|-\alpha.

Let d∈[m]d\in[m] be the smallest index such that c1+c2+⋯+cd≥αc_1+c_2+\cdots+c_d\ge\alpha. Such an integer exists because ∣V∣−α>α|V|-\alpha>\alpha. Recolor all vertices in the first dd color classes with the α\alpha unique colors so that each of these colors is used at least twice. This yields a bipanchromatic coloring of HH with Ld=χP−dL_d = \chi_P - d colors.

Since ci≥2c_i \ge2 for every ii, we have d≤⌈α/2⌉d \le\lceil\alpha/2 \rceil. Consequently, χB≥Ld=χP−d≥β\chi_B \ge L_d = \chi_P - d \ge\beta.

To obtain a bound for dd, write Euclidean division of ∣V∣−α|V|-\alpha by m=χP−αm=\chi_P-\alpha as:

∣V∣−α=qm+r,|V|-\alpha=qm+r,
q=⌊∣V∣−αχP−α⌋≥2,0≤r<m.q=\left\lfloor\frac{|V|-\alpha}{\chi_P-\alpha}\right\rfloor\ge2,\qquad0\le r<m.

Define

D=max⁡{⌈αq+1⌉,⌈α−rq⌉},L=χP−D.(7)D=\max\left\{\left\lceil\frac{\alpha}{q+1}\right\rceil,\left\lceil\frac{\alpha-r}{q}\right\rceil\right\},\qquad L=\chi_P-D. \tag*{(7)}

Unlike LdL_d, which uses the chosen coloring, LL depends only on ∣V∣|V|, χP\chi_P, and α\alpha.

Proposition 10 (Least class recoloring). Assume that the parameters of a hypergraph are defined as above. If 0<α<∣V∣/20<\alpha<|V|/2, then

d≤D≤⌈α2⌉,β≤L≤Ld≤χB.d\le D\le\left\lceil\frac{\alpha}{2}\right\rceil,\qquad\beta\le L\le L_d\le\chi_B.

Proof. For every j∈[m]j\in[m], if cj≥q+1c_j\ge q+1, the first jj classes contain at least j(q+1)j(q+1) vertices. Otherwise, the last m−jm-j classes contain at most (m−j)q(m-j)q vertices, so the first jj contain at least jq+rjq+r. Thus

∑i=1jci≥jq+min⁡{j,r}.\sum_{i=1}^{j}c_i\ge jq+\min\{j,r\}.

The balanced sequence

c^1=⋯=c^r=q+1,c^r+1=⋯=c^m=q\widehat{c}_1=\cdots=\widehat{c}_r=q+1,\qquad\widehat{c}_{r+1}=\cdots=\widehat{c}_m=q

attains equality for every jj. Its first jj terms sum to at least α\alpha exactly when both j(q+1)≥αj(q+1)\ge\alpha and jq+r≥αjq+r\ge\alpha, so DD is its least such index. Therefore d≤Dd\le D. Since q≥2q\ge2 and r≥0r\ge0, the formula for DD gives D≤⌈α/2⌉D\le\lceil\alpha/2\rceil, proving the stated bounds.

Remark. If the three-hyperedge family described in [5 Proposition 1] satisfies 0<α<∣V∣/20<\alpha<|V|/2, then its parameters considered in Proposition 10 are q=3q=3, r=0r=0, and D=α/3D=\alpha/3. Thus L=χP−α/3=χBL=\chi_P-\alpha/3=\chi_B. The boundary case 2α=∣V∣2\alpha=|V| is covered by the observation at the start of this section. So, our order-aware bound gives the exact bipanchromatic number throughout that family.

Let U=min⁡{χP−1,⌊∣V∣/2⌋}U=\min\{\chi_P-1,\lfloor|V|/2\rfloor\}. If 0<α<∣V∣/20<\alpha<|V|/2, then β≤L≤Ld≤χB≤U\beta\le L\le L_d\le\chi_B\le U, and UU is an upper bound for χB\chi_B.

Now, we examine the relations between UU, χB\chi_B, and LL.

Theorem 11 (Improved bound ratio). The bounds LL and UU always satisfy

5U≤6L+1,1≤χBL≤UL≤54.(8)5U\le6L+1,\qquad1\le\frac{\chi_B}{L}\le\frac{U}{L}\le\frac{5}{4}. \tag*{(8)}

In particular, L≤3L\le3 implies U=LU=L. For all feasible inputs,

χBL≤65+15L.(9)\frac{\chi_B}{L}\le\frac{6}{5}+\frac{1}{5L}. \tag*{(9)}

Proof. Use the balanced sequence c^1,…,c^m\hat{c}_1,\ldots,\hat{c}_m from Proposition 10. These are not necessarily the actual color-class sizes of HH. Their sum is ∣V∣−α|V|-\alpha, and DD is the least index for which the first DD terms sum to at least α\alpha.

Set g=U−L≥0g=U-L\ge0. We prove that L≥5g−1L\ge5g-1. Since U≤χP−1=L+D−1U\le\chi_P-1=L+D-1, we obtain g≤D−1g\le D-1. By minimality of DD, it follows that

c^D(D−1)≤c^1+⋯+c^D−1≤α−1.\hat{c}_D(D-1)\le\hat{c}_1+\cdots+\hat{c}_{D-1}\le\alpha-1.

Consequently,

α≥c^Dg+1(10)\alpha\ge\hat{c}_Dg+1 \tag*{(10)}

Since ∑i=1mc^i=∣V∣−α\sum_{i=1}^{m}\hat{c}_i=|V|-\alpha, and c^i≤c^D\hat{c}_i\le\hat{c}_D for i≥Di\ge D,

∣V∣=α+∑i=1D−1c^i+∑i=Dmc^i≤α+(α−1)+c^D(m−D+1).(11)|V|=\alpha+\sum_{i=1}^{D-1}\hat{c}_i+\sum_{i=D}^{m}\hat{c}_i\le\alpha+(\alpha-1)+\hat{c}_D(m-D+1). \tag*{(11)}

Since U=L+g=α+m−D+gU=L+g=\alpha+m-D+g and U≤∣V∣/2U\le|V|/2 (by definition), inequality (11) yields

(c^D−2)(m−D+1)≥2g−1.(12)(\hat{c}_D-2)(m-D+1)\ge2g-1. \tag*{(12)}

We use L=α+m−DL=\alpha+m-D and m−D≥0m-D\ge0 to consider the following four cases:

c^D\hat{c}_DFrom (12)From (10)Conclusion
2g=0g=0–L≥5g−1L\ge5g-1
3m−D≥2g−2m-D\ge2g-2α≥3g+1\alpha\ge3g+1L≥5g−1L\ge5g-1
4m−D≥g−1m-D\ge g-1α≥4g+1\alpha\ge4g+1L≥5gL\ge5g
≥5\ge5–α≥5g+1\alpha\ge5g+1L≥5g+1L\ge5g+1

Table 1.

Therefore, 5g≤L+15g\le L+1, which gives 5U≤6L+15U\le6L+1. For L≤3L\le3 we obtain U=LU=L. For L≥4L\ge4,

UL≤65+15L≤54.\frac{U}{L}\le\frac{6}{5}+\frac{1}{5L}\le\frac{5}{4}.

□

Remark. The conclusions of Theorem 11 also hold with LdL_d in place of LL. Indeed, L≤Ld≤χBL\le L_d\le\chi_B, so 5U≤6L+1≤6Ld+15U\le6L+1\le6L_d+1 and χB/Ld≤χB/L≤5/4\chi_B/L_d\le\chi_B/L\le5/4.

We prove that for fixed parameters (∣V∣,χP,α)(|V|,\chi_P,\alpha), without more information, the bound LL is the best possible.

Proposition 12 (Optimality). For any valid parameters (∣V∣,χP,α)(|V|,\chi_P,\alpha), with 0<α<∣V∣20<\alpha<\frac{|V|}{2}, there exists a hypergraph with these prescribed parameters that satisfies χB=L\chi_B=L.

Proof. Compute qq and rr as before. Set m=χP−αm=\chi_P-\alpha and consider pairwise disjoint sets S,P1,…,Pq,ZS,P_1,\ldots,P_q,Z with ∣S∣=α|S|=\alpha, ∣Pi∣=m|P_i|=m, and ∣Z∣=r|Z|=r. Choose R⊆P1R\subseteq P_1 with ∣R∣=r|R|=r. We define the hypergraph H=(V,E)H=(V,\mathcal{E}), where

V=S\mathbin{\mathaccent\cdot\cup}\bigcup_{i=1}^{q}P_i\mathbin{\mathaccent\cdot\cup}Z,\qquad\mathcal{E}=\{S\mathbin{\mathaccent\cdot\cup}P_i:i\in[q]\}\cup\{S\mathbin{\mathaccent\cdot\cup}(P_1\setminus R)\mathbin{\mathaccent\cdot\cup}Z\}.

Thus ∣V∣=α+qm+r|V|=\alpha+qm+r is attained, and every hyperedge has size χP\chi_P. Now, we define a coloring of HH such that α\alpha vertices in SS receive distinct colors, each PiP_i uses all mm remaining colors once, and, if r>0r > 0, each color from RR is used exactly once on ZZ. This gives a panchromatic coloring of HH with χP\chi_P colors. In any panchromatic χP\chi_P-coloring of HH, precisely the vertices in SS have unique colors, so α\alpha is attained.

Now, we consider a bipanchromatic kk-coloring of HH. Assume that precisely s≤αs \le\alpha colors occur on SS. Each of the other k−sk-s colors occurs in every PiP_i, so at least qq times. Those also absent from P1∖RP_1 \setminus R must occur on ZZ, so they occur at least q+1q+1 times. At most m−rm-r of the k−sk-s colors occur on P1∖RP_1 \setminus R. Therefore

∣V∣≥2s+q(k−s),∣V∣≥2s+(q+1)(k−s)−(m−r).|V| \ge2s + q(k-s), \qquad|V| \ge2s + (q+1)(k-s) - (m-r).

Substituting s≤αs \le\alpha, ∣V∣=α+qm+r|V| = \alpha+ qm + r, and χP=α+m\chi_P = \alpha+ m gives

q(χP−k)≥α−r,(q+1)(χP−k)≥α.q(\chi_P-k) \ge\alpha-r, \qquad(q+1)(\chi_P-k) \ge\alpha.

By the definition of DD, it follows that

χP−k≥max⁡{αq+1,α−rq}=D.\chi_P-k \ge\max\left\{\frac{\alpha}{q+1}, \frac{\alpha-r}{q}\right\} = D.

Thus every bipanchromatic coloring has at most χP−D=L\chi_P-D=L colors. Since also L≤χBL \le\chi_B by Proposition 10, we obtain χB=L\chi_B=L.

Therefore, at every feasible triple in this range, any universally valid lower bound using only (∣V∣,χP,α)(|V|,\chi_P,\alpha) is at most LL. Since LL is itself valid, it is the strongest such lower bound.

Corollary 13 (Optimal ratio guarantees). For LL, the factor 5/45/4 in (8) is attained. The constant 6/56/5 in (9) is asymptotically sharp as L→∞L \to\infty. These are optimal constants from ∣V∣|V|, χP\chi_P, α\alpha.

Proof. For any n∈Nn \in\mathbb{N}, Proposition 7 describes a hypergraph whose parameters (∣V∣,χP,α,χB)(|V|,\chi_P,\alpha,\chi_B) are (12n−2,6n,3n+1,6n−1)(12n-2,6n,3n+1,6n-1). Here we obtain q=3q=3, r=0r=0, D=n+1D=n+1, and L=5n−1L=5n-1, so

χBL=6n−15n−1=65+15(5n−1).\frac{\chi_B}{L} = \frac{6n-1}{5n-1} = \frac{6}{5} + \frac{1}{5(5n-1)}.

This is 5/45/4 at n=1n=1 and tends to 6/56/5 as n→∞n \to\infty. Proposition 12 confirms optimality.

AI Assistance

ChatGPT was used to assist with proof development. All AI-assisted arguments were reviewed and revised by the authors, who take full responsibility for the final manuscript.

References

  1. [1]M. Lalou, N. Mbarek, A. Skender, O. Togni, Completely independent spanning trees in split graphs: Structural properties and complexity, Discrete Appl. Math. 391 (2026), 511–522.
  2. [2]L. Lovász, A generalization of König’s theorem, Acta Math. Acad. Sci. Hungar. 21 (1970), 443–446.DOI
  3. [3]B. Bollobás, D. Pritchard, T. Rothvoß, A. Scott, Cover-decomposition and polychromatic numbers, SIAM J. Discrete Math. 27 (2013), 240–256.
  4. [4]D. Cherkashin, A note on panchromatic colorings, Discrete Math. 341 (2018), 652–657.arxiv.org/abs/1705.03797
  5. [5]M. Lalou, N. Mbarek, A. Skender, O. Togni, On hypergraph colorings and completely independent spanning trees in chordal graphs, arXiv:2610.08042v1 (2026).arxiv.org/abs/2610.08042

Paper details

vlnc:W_uxckru8d
v1 · latestHistory (1)

published · Fresh — added to the Hub within the last 30 days

added ·

Usage

views · 8

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

CC BY 4.0

Original source

  • Mathematics

Not recorded for this paper · View statement →

Looking for related papers…
Contents